?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
megaraid: Don't use create_proc_read_entry()
[linux-2.6.git] / kernel / sys.c
blob39c9c4a2949f3166b114f11534957ff2f6060565
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 #include <linux/binfmts.h>
51
52 #include <linux/kmsg_dump.h>
53 /* Move somewhere else to avoid recompiling? */
54 #include <generated/utsrelease.h>
55
56 #include <asm/uaccess.h>
57 #include <asm/io.h>
58 #include <asm/unistd.h>
59
60 #ifndef SET_UNALIGN_CTL
61 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
62 #endif
63 #ifndef GET_UNALIGN_CTL
64 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
65 #endif
66 #ifndef SET_FPEMU_CTL
67 # define SET_FPEMU_CTL(a,b) (-EINVAL)
68 #endif
69 #ifndef GET_FPEMU_CTL
70 # define GET_FPEMU_CTL(a,b) (-EINVAL)
71 #endif
72 #ifndef SET_FPEXC_CTL
73 # define SET_FPEXC_CTL(a,b) (-EINVAL)
74 #endif
75 #ifndef GET_FPEXC_CTL
76 # define GET_FPEXC_CTL(a,b) (-EINVAL)
77 #endif
78 #ifndef GET_ENDIAN
79 # define GET_ENDIAN(a,b) (-EINVAL)
80 #endif
81 #ifndef SET_ENDIAN
82 # define SET_ENDIAN(a,b) (-EINVAL)
83 #endif
84 #ifndef GET_TSC_CTL
85 # define GET_TSC_CTL(a) (-EINVAL)
86 #endif
87 #ifndef SET_TSC_CTL
88 # define SET_TSC_CTL(a) (-EINVAL)
89 #endif
90
91 /*
92 * this is where the system-wide overflow UID and GID are defined, for
93 * architectures that now have 32-bit UID/GID but didn't in the past
94 */
95
96 int overflowuid = DEFAULT_OVERFLOWUID;
97 int overflowgid = DEFAULT_OVERFLOWGID;
98
99 EXPORT_SYMBOL(overflowuid);
100 EXPORT_SYMBOL(overflowgid);
101
102 /*
103 * the same as above, but for filesystems which can only store a 16-bit
104 * UID and GID. as such, this is needed on all architectures
105 */
106
107 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
108 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
109
110 EXPORT_SYMBOL(fs_overflowuid);
111 EXPORT_SYMBOL(fs_overflowgid);
112
113 /*
114 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
115 */
116
117 int C_A_D = 1;
118 struct pid *cad_pid;
119 EXPORT_SYMBOL(cad_pid);
120
121 /*
122 * If set, this is used for preparing the system to power off.
123 */
124
125 void (*pm_power_off_prepare)(void);
126
127 /*
128 * Returns true if current's euid is same as p's uid or euid,
129 * or has CAP_SYS_NICE to p's user_ns.
130 *
131 * Called with rcu_read_lock, creds are safe
132 */
133 static bool set_one_prio_perm(struct task_struct *p)
134 {
135 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
136
137 if (uid_eq(pcred->uid, cred->euid) ||
138 uid_eq(pcred->euid, cred->euid))
139 return true;
140 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
141 return true;
142 return false;
143 }
144
145 /*
146 * set the priority of a task
147 * - the caller must hold the RCU read lock
148 */
149 static int set_one_prio(struct task_struct *p, int niceval, int error)
150 {
151 int no_nice;
152
153 if (!set_one_prio_perm(p)) {
154 error = -EPERM;
155 goto out;
156 }
157 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
158 error = -EACCES;
159 goto out;
160 }
161 no_nice = security_task_setnice(p, niceval);
162 if (no_nice) {
163 error = no_nice;
164 goto out;
165 }
166 if (error == -ESRCH)
167 error = 0;
168 set_user_nice(p, niceval);
169 out:
170 return error;
171 }
172
173 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
174 {
175 struct task_struct *g, *p;
176 struct user_struct *user;
177 const struct cred *cred = current_cred();
178 int error = -EINVAL;
179 struct pid *pgrp;
180 kuid_t uid;
181
182 if (which > PRIO_USER || which < PRIO_PROCESS)
183 goto out;
184
185 /* normalize: avoid signed division (rounding problems) */
186 error = -ESRCH;
187 if (niceval < -20)
188 niceval = -20;
189 if (niceval > 19)
190 niceval = 19;
191
192 rcu_read_lock();
193 read_lock(&tasklist_lock);
194 switch (which) {
195 case PRIO_PROCESS:
196 if (who)
197 p = find_task_by_vpid(who);
198 else
199 p = current;
200 if (p)
201 error = set_one_prio(p, niceval, error);
202 break;
203 case PRIO_PGRP:
204 if (who)
205 pgrp = find_vpid(who);
206 else
207 pgrp = task_pgrp(current);
208 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
209 error = set_one_prio(p, niceval, error);
210 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
211 break;
212 case PRIO_USER:
213 uid = make_kuid(cred->user_ns, who);
214 user = cred->user;
215 if (!who)
216 uid = cred->uid;
217 else if (!uid_eq(uid, cred->uid) &&
218 !(user = find_user(uid)))
219 goto out_unlock; /* No processes for this user */
220
221 do_each_thread(g, p) {
222 if (uid_eq(task_uid(p), uid))
223 error = set_one_prio(p, niceval, error);
224 } while_each_thread(g, p);
225 if (!uid_eq(uid, cred->uid))
226 free_uid(user); /* For find_user() */
227 break;
228 }
229 out_unlock:
230 read_unlock(&tasklist_lock);
231 rcu_read_unlock();
232 out:
233 return error;
234 }
235
236 /*
237 * Ugh. To avoid negative return values, "getpriority()" will
238 * not return the normal nice-value, but a negated value that
239 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
240 * to stay compatible.
241 */
242 SYSCALL_DEFINE2(getpriority, int, which, int, who)
243 {
244 struct task_struct *g, *p;
245 struct user_struct *user;
246 const struct cred *cred = current_cred();
247 long niceval, retval = -ESRCH;
248 struct pid *pgrp;
249 kuid_t uid;
250
251 if (which > PRIO_USER || which < PRIO_PROCESS)
252 return -EINVAL;
253
254 rcu_read_lock();
255 read_lock(&tasklist_lock);
256 switch (which) {
257 case PRIO_PROCESS:
258 if (who)
259 p = find_task_by_vpid(who);
260 else
261 p = current;
262 if (p) {
263 niceval = 20 - task_nice(p);
264 if (niceval > retval)
265 retval = niceval;
266 }
267 break;
268 case PRIO_PGRP:
269 if (who)
270 pgrp = find_vpid(who);
271 else
272 pgrp = task_pgrp(current);
273 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
274 niceval = 20 - task_nice(p);
275 if (niceval > retval)
276 retval = niceval;
277 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
278 break;
279 case PRIO_USER:
280 uid = make_kuid(cred->user_ns, who);
281 user = cred->user;
282 if (!who)
283 uid = cred->uid;
284 else if (!uid_eq(uid, cred->uid) &&
285 !(user = find_user(uid)))
286 goto out_unlock; /* No processes for this user */
287
288 do_each_thread(g, p) {
289 if (uid_eq(task_uid(p), uid)) {
290 niceval = 20 - task_nice(p);
291 if (niceval > retval)
292 retval = niceval;
293 }
294 } while_each_thread(g, p);
295 if (!uid_eq(uid, cred->uid))
296 free_uid(user); /* for find_user() */
297 break;
298 }
299 out_unlock:
300 read_unlock(&tasklist_lock);
301 rcu_read_unlock();
302
303 return retval;
304 }
305
306 /**
307 * emergency_restart - reboot the system
308 *
309 * Without shutting down any hardware or taking any locks
310 * reboot the system. This is called when we know we are in
311 * trouble so this is our best effort to reboot. This is
312 * safe to call in interrupt context.
313 */
314 void emergency_restart(void)
315 {
316 kmsg_dump(KMSG_DUMP_EMERG);
317 machine_emergency_restart();
318 }
319 EXPORT_SYMBOL_GPL(emergency_restart);
320
321 void kernel_restart_prepare(char *cmd)
322 {
323 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
324 system_state = SYSTEM_RESTART;
325 usermodehelper_disable();
326 device_shutdown();
327 syscore_shutdown();
328 }
329
330 /**
331 * register_reboot_notifier - Register function to be called at reboot time
332 * @nb: Info about notifier function to be called
333 *
334 * Registers a function with the list of functions
335 * to be called at reboot time.
336 *
337 * Currently always returns zero, as blocking_notifier_chain_register()
338 * always returns zero.
339 */
340 int register_reboot_notifier(struct notifier_block *nb)
341 {
342 return blocking_notifier_chain_register(&reboot_notifier_list, nb);
343 }
344 EXPORT_SYMBOL(register_reboot_notifier);
345
346 /**
347 * unregister_reboot_notifier - Unregister previously registered reboot notifier
348 * @nb: Hook to be unregistered
349 *
350 * Unregisters a previously registered reboot
351 * notifier function.
352 *
353 * Returns zero on success, or %-ENOENT on failure.
354 */
355 int unregister_reboot_notifier(struct notifier_block *nb)
356 {
357 return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
358 }
359 EXPORT_SYMBOL(unregister_reboot_notifier);
360
361 /**
362 * kernel_restart - reboot the system
363 * @cmd: pointer to buffer containing command to execute for restart
364 * or %NULL
365 *
366 * Shutdown everything and perform a clean reboot.
367 * This is not safe to call in interrupt context.
368 */
369 void kernel_restart(char *cmd)
370 {
371 kernel_restart_prepare(cmd);
372 disable_nonboot_cpus();
373 if (!cmd)
374 printk(KERN_EMERG "Restarting system.\n");
375 else
376 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
377 kmsg_dump(KMSG_DUMP_RESTART);
378 machine_restart(cmd);
379 }
380 EXPORT_SYMBOL_GPL(kernel_restart);
381
382 static void kernel_shutdown_prepare(enum system_states state)
383 {
384 blocking_notifier_call_chain(&reboot_notifier_list,
385 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
386 system_state = state;
387 usermodehelper_disable();
388 device_shutdown();
389 }
390 /**
391 * kernel_halt - halt the system
392 *
393 * Shutdown everything and perform a clean system halt.
394 */
395 void kernel_halt(void)
396 {
397 kernel_shutdown_prepare(SYSTEM_HALT);
398 syscore_shutdown();
399 printk(KERN_EMERG "System halted.\n");
400 kmsg_dump(KMSG_DUMP_HALT);
401 machine_halt();
402 }
403
404 EXPORT_SYMBOL_GPL(kernel_halt);
405
406 /**
407 * kernel_power_off - power_off the system
408 *
409 * Shutdown everything and perform a clean system power_off.
410 */
411 void kernel_power_off(void)
412 {
413 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
414 if (pm_power_off_prepare)
415 pm_power_off_prepare();
416 disable_nonboot_cpus();
417 syscore_shutdown();
418 printk(KERN_EMERG "Power down.\n");
419 kmsg_dump(KMSG_DUMP_POWEROFF);
420 machine_power_off();
421 }
422 EXPORT_SYMBOL_GPL(kernel_power_off);
423
424 static DEFINE_MUTEX(reboot_mutex);
425
426 /*
427 * Reboot system call: for obvious reasons only root may call it,
428 * and even root needs to set up some magic numbers in the registers
429 * so that some mistake won't make this reboot the whole machine.
430 * You can also set the meaning of the ctrl-alt-del-key here.
431 *
432 * reboot doesn't sync: do that yourself before calling this.
433 */
434 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
435 void __user *, arg)
436 {
437 struct pid_namespace *pid_ns = task_active_pid_ns(current);
438 char buffer[256];
439 int ret = 0;
440
441 /* We only trust the superuser with rebooting the system. */
442 if (!ns_capable(pid_ns->user_ns, CAP_SYS_BOOT))
443 return -EPERM;
444
445 /* For safety, we require "magic" arguments. */
446 if (magic1 != LINUX_REBOOT_MAGIC1 ||
447 (magic2 != LINUX_REBOOT_MAGIC2 &&
448 magic2 != LINUX_REBOOT_MAGIC2A &&
449 magic2 != LINUX_REBOOT_MAGIC2B &&
450 magic2 != LINUX_REBOOT_MAGIC2C))
451 return -EINVAL;
452
453 /*
454 * If pid namespaces are enabled and the current task is in a child
455 * pid_namespace, the command is handled by reboot_pid_ns() which will
456 * call do_exit().
457 */
458 ret = reboot_pid_ns(pid_ns, cmd);
459 if (ret)
460 return ret;
461
462 /* Instead of trying to make the power_off code look like
463 * halt when pm_power_off is not set do it the easy way.
464 */
465 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
466 cmd = LINUX_REBOOT_CMD_HALT;
467
468 mutex_lock(&reboot_mutex);
469 switch (cmd) {
470 case LINUX_REBOOT_CMD_RESTART:
471 kernel_restart(NULL);
472 break;
473
474 case LINUX_REBOOT_CMD_CAD_ON:
475 C_A_D = 1;
476 break;
477
478 case LINUX_REBOOT_CMD_CAD_OFF:
479 C_A_D = 0;
480 break;
481
482 case LINUX_REBOOT_CMD_HALT:
483 kernel_halt();
484 do_exit(0);
485 panic("cannot halt");
486
487 case LINUX_REBOOT_CMD_POWER_OFF:
488 kernel_power_off();
489 do_exit(0);
490 break;
491
492 case LINUX_REBOOT_CMD_RESTART2:
493 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
494 ret = -EFAULT;
495 break;
496 }
497 buffer[sizeof(buffer) - 1] = '\0';
498
499 kernel_restart(buffer);
500 break;
501
502 #ifdef CONFIG_KEXEC
503 case LINUX_REBOOT_CMD_KEXEC:
504 ret = kernel_kexec();
505 break;
506 #endif
507
508 #ifdef CONFIG_HIBERNATION
509 case LINUX_REBOOT_CMD_SW_SUSPEND:
510 ret = hibernate();
511 break;
512 #endif
513
514 default:
515 ret = -EINVAL;
516 break;
517 }
518 mutex_unlock(&reboot_mutex);
519 return ret;
520 }
521
522 static void deferred_cad(struct work_struct *dummy)
523 {
524 kernel_restart(NULL);
525 }
526
527 /*
528 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
529 * As it's called within an interrupt, it may NOT sync: the only choice
530 * is whether to reboot at once, or just ignore the ctrl-alt-del.
531 */
532 void ctrl_alt_del(void)
533 {
534 static DECLARE_WORK(cad_work, deferred_cad);
535
536 if (C_A_D)
537 schedule_work(&cad_work);
538 else
539 kill_cad_pid(SIGINT, 1);
540 }
541
542 /*
543 * Unprivileged users may change the real gid to the effective gid
544 * or vice versa. (BSD-style)
545 *
546 * If you set the real gid at all, or set the effective gid to a value not
547 * equal to the real gid, then the saved gid is set to the new effective gid.
548 *
549 * This makes it possible for a setgid program to completely drop its
550 * privileges, which is often a useful assertion to make when you are doing
551 * a security audit over a program.
552 *
553 * The general idea is that a program which uses just setregid() will be
554 * 100% compatible with BSD. A program which uses just setgid() will be
555 * 100% compatible with POSIX with saved IDs.
556 *
557 * SMP: There are not races, the GIDs are checked only by filesystem
558 * operations (as far as semantic preservation is concerned).
559 */
560 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
561 {
562 struct user_namespace *ns = current_user_ns();
563 const struct cred *old;
564 struct cred *new;
565 int retval;
566 kgid_t krgid, kegid;
567
568 krgid = make_kgid(ns, rgid);
569 kegid = make_kgid(ns, egid);
570
571 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
572 return -EINVAL;
573 if ((egid != (gid_t) -1) && !gid_valid(kegid))
574 return -EINVAL;
575
576 new = prepare_creds();
577 if (!new)
578 return -ENOMEM;
579 old = current_cred();
580
581 retval = -EPERM;
582 if (rgid != (gid_t) -1) {
583 if (gid_eq(old->gid, krgid) ||
584 gid_eq(old->egid, krgid) ||
585 nsown_capable(CAP_SETGID))
586 new->gid = krgid;
587 else
588 goto error;
589 }
590 if (egid != (gid_t) -1) {
591 if (gid_eq(old->gid, kegid) ||
592 gid_eq(old->egid, kegid) ||
593 gid_eq(old->sgid, kegid) ||
594 nsown_capable(CAP_SETGID))
595 new->egid = kegid;
596 else
597 goto error;
598 }
599
600 if (rgid != (gid_t) -1 ||
601 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
602 new->sgid = new->egid;
603 new->fsgid = new->egid;
604
605 return commit_creds(new);
606
607 error:
608 abort_creds(new);
609 return retval;
610 }
611
612 /*
613 * setgid() is implemented like SysV w/ SAVED_IDS
614 *
615 * SMP: Same implicit races as above.
616 */
617 SYSCALL_DEFINE1(setgid, gid_t, gid)
618 {
619 struct user_namespace *ns = current_user_ns();
620 const struct cred *old;
621 struct cred *new;
622 int retval;
623 kgid_t kgid;
624
625 kgid = make_kgid(ns, gid);
626 if (!gid_valid(kgid))
627 return -EINVAL;
628
629 new = prepare_creds();
630 if (!new)
631 return -ENOMEM;
632 old = current_cred();
633
634 retval = -EPERM;
635 if (nsown_capable(CAP_SETGID))
636 new->gid = new->egid = new->sgid = new->fsgid = kgid;
637 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
638 new->egid = new->fsgid = kgid;
639 else
640 goto error;
641
642 return commit_creds(new);
643
644 error:
645 abort_creds(new);
646 return retval;
647 }
648
649 /*
650 * change the user struct in a credentials set to match the new UID
651 */
652 static int set_user(struct cred *new)
653 {
654 struct user_struct *new_user;
655
656 new_user = alloc_uid(new->uid);
657 if (!new_user)
658 return -EAGAIN;
659
660 /*
661 * We don't fail in case of NPROC limit excess here because too many
662 * poorly written programs don't check set*uid() return code, assuming
663 * it never fails if called by root. We may still enforce NPROC limit
664 * for programs doing set*uid()+execve() by harmlessly deferring the
665 * failure to the execve() stage.
666 */
667 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
668 new_user != INIT_USER)
669 current->flags |= PF_NPROC_EXCEEDED;
670 else
671 current->flags &= ~PF_NPROC_EXCEEDED;
672
673 free_uid(new->user);
674 new->user = new_user;
675 return 0;
676 }
677
678 /*
679 * Unprivileged users may change the real uid to the effective uid
680 * or vice versa. (BSD-style)
681 *
682 * If you set the real uid at all, or set the effective uid to a value not
683 * equal to the real uid, then the saved uid is set to the new effective uid.
684 *
685 * This makes it possible for a setuid program to completely drop its
686 * privileges, which is often a useful assertion to make when you are doing
687 * a security audit over a program.
688 *
689 * The general idea is that a program which uses just setreuid() will be
690 * 100% compatible with BSD. A program which uses just setuid() will be
691 * 100% compatible with POSIX with saved IDs.
692 */
693 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
694 {
695 struct user_namespace *ns = current_user_ns();
696 const struct cred *old;
697 struct cred *new;
698 int retval;
699 kuid_t kruid, keuid;
700
701 kruid = make_kuid(ns, ruid);
702 keuid = make_kuid(ns, euid);
703
704 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
705 return -EINVAL;
706 if ((euid != (uid_t) -1) && !uid_valid(keuid))
707 return -EINVAL;
708
709 new = prepare_creds();
710 if (!new)
711 return -ENOMEM;
712 old = current_cred();
713
714 retval = -EPERM;
715 if (ruid != (uid_t) -1) {
716 new->uid = kruid;
717 if (!uid_eq(old->uid, kruid) &&
718 !uid_eq(old->euid, kruid) &&
719 !nsown_capable(CAP_SETUID))
720 goto error;
721 }
722
723 if (euid != (uid_t) -1) {
724 new->euid = keuid;
725 if (!uid_eq(old->uid, keuid) &&
726 !uid_eq(old->euid, keuid) &&
727 !uid_eq(old->suid, keuid) &&
728 !nsown_capable(CAP_SETUID))
729 goto error;
730 }
731
732 if (!uid_eq(new->uid, old->uid)) {
733 retval = set_user(new);
734 if (retval < 0)
735 goto error;
736 }
737 if (ruid != (uid_t) -1 ||
738 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
739 new->suid = new->euid;
740 new->fsuid = new->euid;
741
742 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
743 if (retval < 0)
744 goto error;
745
746 return commit_creds(new);
747
748 error:
749 abort_creds(new);
750 return retval;
751 }
752
753 /*
754 * setuid() is implemented like SysV with SAVED_IDS
755 *
756 * Note that SAVED_ID's is deficient in that a setuid root program
757 * like sendmail, for example, cannot set its uid to be a normal
758 * user and then switch back, because if you're root, setuid() sets
759 * the saved uid too. If you don't like this, blame the bright people
760 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
761 * will allow a root program to temporarily drop privileges and be able to
762 * regain them by swapping the real and effective uid.
763 */
764 SYSCALL_DEFINE1(setuid, uid_t, uid)
765 {
766 struct user_namespace *ns = current_user_ns();
767 const struct cred *old;
768 struct cred *new;
769 int retval;
770 kuid_t kuid;
771
772 kuid = make_kuid(ns, uid);
773 if (!uid_valid(kuid))
774 return -EINVAL;
775
776 new = prepare_creds();
777 if (!new)
778 return -ENOMEM;
779 old = current_cred();
780
781 retval = -EPERM;
782 if (nsown_capable(CAP_SETUID)) {
783 new->suid = new->uid = kuid;
784 if (!uid_eq(kuid, old->uid)) {
785 retval = set_user(new);
786 if (retval < 0)
787 goto error;
788 }
789 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
790 goto error;
791 }
792
793 new->fsuid = new->euid = kuid;
794
795 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
796 if (retval < 0)
797 goto error;
798
799 return commit_creds(new);
800
801 error:
802 abort_creds(new);
803 return retval;
804 }
805
806
807 /*
808 * This function implements a generic ability to update ruid, euid,
809 * and suid. This allows you to implement the 4.4 compatible seteuid().
810 */
811 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
812 {
813 struct user_namespace *ns = current_user_ns();
814 const struct cred *old;
815 struct cred *new;
816 int retval;
817 kuid_t kruid, keuid, ksuid;
818
819 kruid = make_kuid(ns, ruid);
820 keuid = make_kuid(ns, euid);
821 ksuid = make_kuid(ns, suid);
822
823 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
824 return -EINVAL;
825
826 if ((euid != (uid_t) -1) && !uid_valid(keuid))
827 return -EINVAL;
828
829 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
830 return -EINVAL;
831
832 new = prepare_creds();
833 if (!new)
834 return -ENOMEM;
835
836 old = current_cred();
837
838 retval = -EPERM;
839 if (!nsown_capable(CAP_SETUID)) {
840 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
841 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
842 goto error;
843 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
844 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
845 goto error;
846 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
847 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
848 goto error;
849 }
850
851 if (ruid != (uid_t) -1) {
852 new->uid = kruid;
853 if (!uid_eq(kruid, old->uid)) {
854 retval = set_user(new);
855 if (retval < 0)
856 goto error;
857 }
858 }
859 if (euid != (uid_t) -1)
860 new->euid = keuid;
861 if (suid != (uid_t) -1)
862 new->suid = ksuid;
863 new->fsuid = new->euid;
864
865 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
866 if (retval < 0)
867 goto error;
868
869 return commit_creds(new);
870
871 error:
872 abort_creds(new);
873 return retval;
874 }
875
876 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
877 {
878 const struct cred *cred = current_cred();
879 int retval;
880 uid_t ruid, euid, suid;
881
882 ruid = from_kuid_munged(cred->user_ns, cred->uid);
883 euid = from_kuid_munged(cred->user_ns, cred->euid);
884 suid = from_kuid_munged(cred->user_ns, cred->suid);
885
886 if (!(retval = put_user(ruid, ruidp)) &&
887 !(retval = put_user(euid, euidp)))
888 retval = put_user(suid, suidp);
889
890 return retval;
891 }
892
893 /*
894 * Same as above, but for rgid, egid, sgid.
895 */
896 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
897 {
898 struct user_namespace *ns = current_user_ns();
899 const struct cred *old;
900 struct cred *new;
901 int retval;
902 kgid_t krgid, kegid, ksgid;
903
904 krgid = make_kgid(ns, rgid);
905 kegid = make_kgid(ns, egid);
906 ksgid = make_kgid(ns, sgid);
907
908 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
909 return -EINVAL;
910 if ((egid != (gid_t) -1) && !gid_valid(kegid))
911 return -EINVAL;
912 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
913 return -EINVAL;
914
915 new = prepare_creds();
916 if (!new)
917 return -ENOMEM;
918 old = current_cred();
919
920 retval = -EPERM;
921 if (!nsown_capable(CAP_SETGID)) {
922 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
923 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
924 goto error;
925 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
926 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
927 goto error;
928 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
929 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
930 goto error;
931 }
932
933 if (rgid != (gid_t) -1)
934 new->gid = krgid;
935 if (egid != (gid_t) -1)
936 new->egid = kegid;
937 if (sgid != (gid_t) -1)
938 new->sgid = ksgid;
939 new->fsgid = new->egid;
940
941 return commit_creds(new);
942
943 error:
944 abort_creds(new);
945 return retval;
946 }
947
948 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
949 {
950 const struct cred *cred = current_cred();
951 int retval;
952 gid_t rgid, egid, sgid;
953
954 rgid = from_kgid_munged(cred->user_ns, cred->gid);
955 egid = from_kgid_munged(cred->user_ns, cred->egid);
956 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
957
958 if (!(retval = put_user(rgid, rgidp)) &&
959 !(retval = put_user(egid, egidp)))
960 retval = put_user(sgid, sgidp);
961
962 return retval;
963 }
964
965
966 /*
967 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
968 * is used for "access()" and for the NFS daemon (letting nfsd stay at
969 * whatever uid it wants to). It normally shadows "euid", except when
970 * explicitly set by setfsuid() or for access..
971 */
972 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
973 {
974 const struct cred *old;
975 struct cred *new;
976 uid_t old_fsuid;
977 kuid_t kuid;
978
979 old = current_cred();
980 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
981
982 kuid = make_kuid(old->user_ns, uid);
983 if (!uid_valid(kuid))
984 return old_fsuid;
985
986 new = prepare_creds();
987 if (!new)
988 return old_fsuid;
989
990 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
991 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
992 nsown_capable(CAP_SETUID)) {
993 if (!uid_eq(kuid, old->fsuid)) {
994 new->fsuid = kuid;
995 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
996 goto change_okay;
997 }
998 }
999
1000 abort_creds(new);
1001 return old_fsuid;
1002
1003 change_okay:
1004 commit_creds(new);
1005 return old_fsuid;
1006 }
1007
1008 /*
1009 * Samma på svenska..
1010 */
1011 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
1012 {
1013 const struct cred *old;
1014 struct cred *new;
1015 gid_t old_fsgid;
1016 kgid_t kgid;
1017
1018 old = current_cred();
1019 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
1020
1021 kgid = make_kgid(old->user_ns, gid);
1022 if (!gid_valid(kgid))
1023 return old_fsgid;
1024
1025 new = prepare_creds();
1026 if (!new)
1027 return old_fsgid;
1028
1029 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
1030 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
1031 nsown_capable(CAP_SETGID)) {
1032 if (!gid_eq(kgid, old->fsgid)) {
1033 new->fsgid = kgid;
1034 goto change_okay;
1035 }
1036 }
1037
1038 abort_creds(new);
1039 return old_fsgid;
1040
1041 change_okay:
1042 commit_creds(new);
1043 return old_fsgid;
1044 }
1045
1046 void do_sys_times(struct tms *tms)
1047 {
1048 cputime_t tgutime, tgstime, cutime, cstime;
1049
1050 spin_lock_irq(&current->sighand->siglock);
1051 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1052 cutime = current->signal->cutime;
1053 cstime = current->signal->cstime;
1054 spin_unlock_irq(&current->sighand->siglock);
1055 tms->tms_utime = cputime_to_clock_t(tgutime);
1056 tms->tms_stime = cputime_to_clock_t(tgstime);
1057 tms->tms_cutime = cputime_to_clock_t(cutime);
1058 tms->tms_cstime = cputime_to_clock_t(cstime);
1059 }
1060
1061 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1062 {
1063 if (tbuf) {
1064 struct tms tmp;
1065
1066 do_sys_times(&tmp);
1067 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1068 return -EFAULT;
1069 }
1070 force_successful_syscall_return();
1071 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1072 }
1073
1074 /*
1075 * This needs some heavy checking ...
1076 * I just haven't the stomach for it. I also don't fully
1077 * understand sessions/pgrp etc. Let somebody who does explain it.
1078 *
1079 * OK, I think I have the protection semantics right.... this is really
1080 * only important on a multi-user system anyway, to make sure one user
1081 * can't send a signal to a process owned by another. -TYT, 12/12/91
1082 *
1083 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1084 * LBT 04.03.94
1085 */
1086 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1087 {
1088 struct task_struct *p;
1089 struct task_struct *group_leader = current->group_leader;
1090 struct pid *pgrp;
1091 int err;
1092
1093 if (!pid)
1094 pid = task_pid_vnr(group_leader);
1095 if (!pgid)
1096 pgid = pid;
1097 if (pgid < 0)
1098 return -EINVAL;
1099 rcu_read_lock();
1100
1101 /* From this point forward we keep holding onto the tasklist lock
1102 * so that our parent does not change from under us. -DaveM
1103 */
1104 write_lock_irq(&tasklist_lock);
1105
1106 err = -ESRCH;
1107 p = find_task_by_vpid(pid);
1108 if (!p)
1109 goto out;
1110
1111 err = -EINVAL;
1112 if (!thread_group_leader(p))
1113 goto out;
1114
1115 if (same_thread_group(p->real_parent, group_leader)) {
1116 err = -EPERM;
1117 if (task_session(p) != task_session(group_leader))
1118 goto out;
1119 err = -EACCES;
1120 if (p->did_exec)
1121 goto out;
1122 } else {
1123 err = -ESRCH;
1124 if (p != group_leader)
1125 goto out;
1126 }
1127
1128 err = -EPERM;
1129 if (p->signal->leader)
1130 goto out;
1131
1132 pgrp = task_pid(p);
1133 if (pgid != pid) {
1134 struct task_struct *g;
1135
1136 pgrp = find_vpid(pgid);
1137 g = pid_task(pgrp, PIDTYPE_PGID);
1138 if (!g || task_session(g) != task_session(group_leader))
1139 goto out;
1140 }
1141
1142 err = security_task_setpgid(p, pgid);
1143 if (err)
1144 goto out;
1145
1146 if (task_pgrp(p) != pgrp)
1147 change_pid(p, PIDTYPE_PGID, pgrp);
1148
1149 err = 0;
1150 out:
1151 /* All paths lead to here, thus we are safe. -DaveM */
1152 write_unlock_irq(&tasklist_lock);
1153 rcu_read_unlock();
1154 return err;
1155 }
1156
1157 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1158 {
1159 struct task_struct *p;
1160 struct pid *grp;
1161 int retval;
1162
1163 rcu_read_lock();
1164 if (!pid)
1165 grp = task_pgrp(current);
1166 else {
1167 retval = -ESRCH;
1168 p = find_task_by_vpid(pid);
1169 if (!p)
1170 goto out;
1171 grp = task_pgrp(p);
1172 if (!grp)
1173 goto out;
1174
1175 retval = security_task_getpgid(p);
1176 if (retval)
1177 goto out;
1178 }
1179 retval = pid_vnr(grp);
1180 out:
1181 rcu_read_unlock();
1182 return retval;
1183 }
1184
1185 #ifdef __ARCH_WANT_SYS_GETPGRP
1186
1187 SYSCALL_DEFINE0(getpgrp)
1188 {
1189 return sys_getpgid(0);
1190 }
1191
1192 #endif
1193
1194 SYSCALL_DEFINE1(getsid, pid_t, pid)
1195 {
1196 struct task_struct *p;
1197 struct pid *sid;
1198 int retval;
1199
1200 rcu_read_lock();
1201 if (!pid)
1202 sid = task_session(current);
1203 else {
1204 retval = -ESRCH;
1205 p = find_task_by_vpid(pid);
1206 if (!p)
1207 goto out;
1208 sid = task_session(p);
1209 if (!sid)
1210 goto out;
1211
1212 retval = security_task_getsid(p);
1213 if (retval)
1214 goto out;
1215 }
1216 retval = pid_vnr(sid);
1217 out:
1218 rcu_read_unlock();
1219 return retval;
1220 }
1221
1222 SYSCALL_DEFINE0(setsid)
1223 {
1224 struct task_struct *group_leader = current->group_leader;
1225 struct pid *sid = task_pid(group_leader);
1226 pid_t session = pid_vnr(sid);
1227 int err = -EPERM;
1228
1229 write_lock_irq(&tasklist_lock);
1230 /* Fail if I am already a session leader */
1231 if (group_leader->signal->leader)
1232 goto out;
1233
1234 /* Fail if a process group id already exists that equals the
1235 * proposed session id.
1236 */
1237 if (pid_task(sid, PIDTYPE_PGID))
1238 goto out;
1239
1240 group_leader->signal->leader = 1;
1241 __set_special_pids(sid);
1242
1243 proc_clear_tty(group_leader);
1244
1245 err = session;
1246 out:
1247 write_unlock_irq(&tasklist_lock);
1248 if (err > 0) {
1249 proc_sid_connector(group_leader);
1250 sched_autogroup_create_attach(group_leader);
1251 }
1252 return err;
1253 }
1254
1255 DECLARE_RWSEM(uts_sem);
1256
1257 #ifdef COMPAT_UTS_MACHINE
1258 #define override_architecture(name) \
1259 (personality(current->personality) == PER_LINUX32 && \
1260 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1261 sizeof(COMPAT_UTS_MACHINE)))
1262 #else
1263 #define override_architecture(name) 0
1264 #endif
1265
1266 /*
1267 * Work around broken programs that cannot handle "Linux 3.0".
1268 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1269 */
1270 static int override_release(char __user *release, size_t len)
1271 {
1272 int ret = 0;
1273
1274 if (current->personality & UNAME26) {
1275 const char *rest = UTS_RELEASE;
1276 char buf[65] = { 0 };
1277 int ndots = 0;
1278 unsigned v;
1279 size_t copy;
1280
1281 while (*rest) {
1282 if (*rest == '.' && ++ndots >= 3)
1283 break;
1284 if (!isdigit(*rest) && *rest != '.')
1285 break;
1286 rest++;
1287 }
1288 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1289 copy = clamp_t(size_t, len, 1, sizeof(buf));
1290 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1291 ret = copy_to_user(release, buf, copy + 1);
1292 }
1293 return ret;
1294 }
1295
1296 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1297 {
1298 int errno = 0;
1299
1300 down_read(&uts_sem);
1301 if (copy_to_user(name, utsname(), sizeof *name))
1302 errno = -EFAULT;
1303 up_read(&uts_sem);
1304
1305 if (!errno && override_release(name->release, sizeof(name->release)))
1306 errno = -EFAULT;
1307 if (!errno && override_architecture(name))
1308 errno = -EFAULT;
1309 return errno;
1310 }
1311
1312 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1313 /*
1314 * Old cruft
1315 */
1316 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1317 {
1318 int error = 0;
1319
1320 if (!name)
1321 return -EFAULT;
1322
1323 down_read(&uts_sem);
1324 if (copy_to_user(name, utsname(), sizeof(*name)))
1325 error = -EFAULT;
1326 up_read(&uts_sem);
1327
1328 if (!error && override_release(name->release, sizeof(name->release)))
1329 error = -EFAULT;
1330 if (!error && override_architecture(name))
1331 error = -EFAULT;
1332 return error;
1333 }
1334
1335 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1336 {
1337 int error;
1338
1339 if (!name)
1340 return -EFAULT;
1341 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1342 return -EFAULT;
1343
1344 down_read(&uts_sem);
1345 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1346 __OLD_UTS_LEN);
1347 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1348 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1349 __OLD_UTS_LEN);
1350 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1351 error |= __copy_to_user(&name->release, &utsname()->release,
1352 __OLD_UTS_LEN);
1353 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1354 error |= __copy_to_user(&name->version, &utsname()->version,
1355 __OLD_UTS_LEN);
1356 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1357 error |= __copy_to_user(&name->machine, &utsname()->machine,
1358 __OLD_UTS_LEN);
1359 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1360 up_read(&uts_sem);
1361
1362 if (!error && override_architecture(name))
1363 error = -EFAULT;
1364 if (!error && override_release(name->release, sizeof(name->release)))
1365 error = -EFAULT;
1366 return error ? -EFAULT : 0;
1367 }
1368 #endif
1369
1370 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1371 {
1372 int errno;
1373 char tmp[__NEW_UTS_LEN];
1374
1375 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1376 return -EPERM;
1377
1378 if (len < 0 || len > __NEW_UTS_LEN)
1379 return -EINVAL;
1380 down_write(&uts_sem);
1381 errno = -EFAULT;
1382 if (!copy_from_user(tmp, name, len)) {
1383 struct new_utsname *u = utsname();
1384
1385 memcpy(u->nodename, tmp, len);
1386 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1387 errno = 0;
1388 uts_proc_notify(UTS_PROC_HOSTNAME);
1389 }
1390 up_write(&uts_sem);
1391 return errno;
1392 }
1393
1394 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1395
1396 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1397 {
1398 int i, errno;
1399 struct new_utsname *u;
1400
1401 if (len < 0)
1402 return -EINVAL;
1403 down_read(&uts_sem);
1404 u = utsname();
1405 i = 1 + strlen(u->nodename);
1406 if (i > len)
1407 i = len;
1408 errno = 0;
1409 if (copy_to_user(name, u->nodename, i))
1410 errno = -EFAULT;
1411 up_read(&uts_sem);
1412 return errno;
1413 }
1414
1415 #endif
1416
1417 /*
1418 * Only setdomainname; getdomainname can be implemented by calling
1419 * uname()
1420 */
1421 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1422 {
1423 int errno;
1424 char tmp[__NEW_UTS_LEN];
1425
1426 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1427 return -EPERM;
1428 if (len < 0 || len > __NEW_UTS_LEN)
1429 return -EINVAL;
1430
1431 down_write(&uts_sem);
1432 errno = -EFAULT;
1433 if (!copy_from_user(tmp, name, len)) {
1434 struct new_utsname *u = utsname();
1435
1436 memcpy(u->domainname, tmp, len);
1437 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1438 errno = 0;
1439 uts_proc_notify(UTS_PROC_DOMAINNAME);
1440 }
1441 up_write(&uts_sem);
1442 return errno;
1443 }
1444
1445 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1446 {
1447 struct rlimit value;
1448 int ret;
1449
1450 ret = do_prlimit(current, resource, NULL, &value);
1451 if (!ret)
1452 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1453
1454 return ret;
1455 }
1456
1457 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1458
1459 /*
1460 * Back compatibility for getrlimit. Needed for some apps.
1461 */
1462
1463 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1464 struct rlimit __user *, rlim)
1465 {
1466 struct rlimit x;
1467 if (resource >= RLIM_NLIMITS)
1468 return -EINVAL;
1469
1470 task_lock(current->group_leader);
1471 x = current->signal->rlim[resource];
1472 task_unlock(current->group_leader);
1473 if (x.rlim_cur > 0x7FFFFFFF)
1474 x.rlim_cur = 0x7FFFFFFF;
1475 if (x.rlim_max > 0x7FFFFFFF)
1476 x.rlim_max = 0x7FFFFFFF;
1477 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1478 }
1479
1480 #endif
1481
1482 static inline bool rlim64_is_infinity(__u64 rlim64)
1483 {
1484 #if BITS_PER_LONG < 64
1485 return rlim64 >= ULONG_MAX;
1486 #else
1487 return rlim64 == RLIM64_INFINITY;
1488 #endif
1489 }
1490
1491 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1492 {
1493 if (rlim->rlim_cur == RLIM_INFINITY)
1494 rlim64->rlim_cur = RLIM64_INFINITY;
1495 else
1496 rlim64->rlim_cur = rlim->rlim_cur;
1497 if (rlim->rlim_max == RLIM_INFINITY)
1498 rlim64->rlim_max = RLIM64_INFINITY;
1499 else
1500 rlim64->rlim_max = rlim->rlim_max;
1501 }
1502
1503 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1504 {
1505 if (rlim64_is_infinity(rlim64->rlim_cur))
1506 rlim->rlim_cur = RLIM_INFINITY;
1507 else
1508 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1509 if (rlim64_is_infinity(rlim64->rlim_max))
1510 rlim->rlim_max = RLIM_INFINITY;
1511 else
1512 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1513 }
1514
1515 /* make sure you are allowed to change @tsk limits before calling this */
1516 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1517 struct rlimit *new_rlim, struct rlimit *old_rlim)
1518 {
1519 struct rlimit *rlim;
1520 int retval = 0;
1521
1522 if (resource >= RLIM_NLIMITS)
1523 return -EINVAL;
1524 if (new_rlim) {
1525 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1526 return -EINVAL;
1527 if (resource == RLIMIT_NOFILE &&
1528 new_rlim->rlim_max > sysctl_nr_open)
1529 return -EPERM;
1530 }
1531
1532 /* protect tsk->signal and tsk->sighand from disappearing */
1533 read_lock(&tasklist_lock);
1534 if (!tsk->sighand) {
1535 retval = -ESRCH;
1536 goto out;
1537 }
1538
1539 rlim = tsk->signal->rlim + resource;
1540 task_lock(tsk->group_leader);
1541 if (new_rlim) {
1542 /* Keep the capable check against init_user_ns until
1543 cgroups can contain all limits */
1544 if (new_rlim->rlim_max > rlim->rlim_max &&
1545 !capable(CAP_SYS_RESOURCE))
1546 retval = -EPERM;
1547 if (!retval)
1548 retval = security_task_setrlimit(tsk->group_leader,
1549 resource, new_rlim);
1550 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1551 /*
1552 * The caller is asking for an immediate RLIMIT_CPU
1553 * expiry. But we use the zero value to mean "it was
1554 * never set". So let's cheat and make it one second
1555 * instead
1556 */
1557 new_rlim->rlim_cur = 1;
1558 }
1559 }
1560 if (!retval) {
1561 if (old_rlim)
1562 *old_rlim = *rlim;
1563 if (new_rlim)
1564 *rlim = *new_rlim;
1565 }
1566 task_unlock(tsk->group_leader);
1567
1568 /*
1569 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1570 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1571 * very long-standing error, and fixing it now risks breakage of
1572 * applications, so we live with it
1573 */
1574 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1575 new_rlim->rlim_cur != RLIM_INFINITY)
1576 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1577 out:
1578 read_unlock(&tasklist_lock);
1579 return retval;
1580 }
1581
1582 /* rcu lock must be held */
1583 static int check_prlimit_permission(struct task_struct *task)
1584 {
1585 const struct cred *cred = current_cred(), *tcred;
1586
1587 if (current == task)
1588 return 0;
1589
1590 tcred = __task_cred(task);
1591 if (uid_eq(cred->uid, tcred->euid) &&
1592 uid_eq(cred->uid, tcred->suid) &&
1593 uid_eq(cred->uid, tcred->uid) &&
1594 gid_eq(cred->gid, tcred->egid) &&
1595 gid_eq(cred->gid, tcred->sgid) &&
1596 gid_eq(cred->gid, tcred->gid))
1597 return 0;
1598 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1599 return 0;
1600
1601 return -EPERM;
1602 }
1603
1604 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1605 const struct rlimit64 __user *, new_rlim,
1606 struct rlimit64 __user *, old_rlim)
1607 {
1608 struct rlimit64 old64, new64;
1609 struct rlimit old, new;
1610 struct task_struct *tsk;
1611 int ret;
1612
1613 if (new_rlim) {
1614 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1615 return -EFAULT;
1616 rlim64_to_rlim(&new64, &new);
1617 }
1618
1619 rcu_read_lock();
1620 tsk = pid ? find_task_by_vpid(pid) : current;
1621 if (!tsk) {
1622 rcu_read_unlock();
1623 return -ESRCH;
1624 }
1625 ret = check_prlimit_permission(tsk);
1626 if (ret) {
1627 rcu_read_unlock();
1628 return ret;
1629 }
1630 get_task_struct(tsk);
1631 rcu_read_unlock();
1632
1633 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1634 old_rlim ? &old : NULL);
1635
1636 if (!ret && old_rlim) {
1637 rlim_to_rlim64(&old, &old64);
1638 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1639 ret = -EFAULT;
1640 }
1641
1642 put_task_struct(tsk);
1643 return ret;
1644 }
1645
1646 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1647 {
1648 struct rlimit new_rlim;
1649
1650 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1651 return -EFAULT;
1652 return do_prlimit(current, resource, &new_rlim, NULL);
1653 }
1654
1655 /*
1656 * It would make sense to put struct rusage in the task_struct,
1657 * except that would make the task_struct be *really big*. After
1658 * task_struct gets moved into malloc'ed memory, it would
1659 * make sense to do this. It will make moving the rest of the information
1660 * a lot simpler! (Which we're not doing right now because we're not
1661 * measuring them yet).
1662 *
1663 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1664 * races with threads incrementing their own counters. But since word
1665 * reads are atomic, we either get new values or old values and we don't
1666 * care which for the sums. We always take the siglock to protect reading
1667 * the c* fields from p->signal from races with exit.c updating those
1668 * fields when reaping, so a sample either gets all the additions of a
1669 * given child after it's reaped, or none so this sample is before reaping.
1670 *
1671 * Locking:
1672 * We need to take the siglock for CHILDEREN, SELF and BOTH
1673 * for the cases current multithreaded, non-current single threaded
1674 * non-current multithreaded. Thread traversal is now safe with
1675 * the siglock held.
1676 * Strictly speaking, we donot need to take the siglock if we are current and
1677 * single threaded, as no one else can take our signal_struct away, no one
1678 * else can reap the children to update signal->c* counters, and no one else
1679 * can race with the signal-> fields. If we do not take any lock, the
1680 * signal-> fields could be read out of order while another thread was just
1681 * exiting. So we should place a read memory barrier when we avoid the lock.
1682 * On the writer side, write memory barrier is implied in __exit_signal
1683 * as __exit_signal releases the siglock spinlock after updating the signal->
1684 * fields. But we don't do this yet to keep things simple.
1685 *
1686 */
1687
1688 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1689 {
1690 r->ru_nvcsw += t->nvcsw;
1691 r->ru_nivcsw += t->nivcsw;
1692 r->ru_minflt += t->min_flt;
1693 r->ru_majflt += t->maj_flt;
1694 r->ru_inblock += task_io_get_inblock(t);
1695 r->ru_oublock += task_io_get_oublock(t);
1696 }
1697
1698 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1699 {
1700 struct task_struct *t;
1701 unsigned long flags;
1702 cputime_t tgutime, tgstime, utime, stime;
1703 unsigned long maxrss = 0;
1704
1705 memset((char *) r, 0, sizeof *r);
1706 utime = stime = 0;
1707
1708 if (who == RUSAGE_THREAD) {
1709 task_cputime_adjusted(current, &utime, &stime);
1710 accumulate_thread_rusage(p, r);
1711 maxrss = p->signal->maxrss;
1712 goto out;
1713 }
1714
1715 if (!lock_task_sighand(p, &flags))
1716 return;
1717
1718 switch (who) {
1719 case RUSAGE_BOTH:
1720 case RUSAGE_CHILDREN:
1721 utime = p->signal->cutime;
1722 stime = p->signal->cstime;
1723 r->ru_nvcsw = p->signal->cnvcsw;
1724 r->ru_nivcsw = p->signal->cnivcsw;
1725 r->ru_minflt = p->signal->cmin_flt;
1726 r->ru_majflt = p->signal->cmaj_flt;
1727 r->ru_inblock = p->signal->cinblock;
1728 r->ru_oublock = p->signal->coublock;
1729 maxrss = p->signal->cmaxrss;
1730
1731 if (who == RUSAGE_CHILDREN)
1732 break;
1733
1734 case RUSAGE_SELF:
1735 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1736 utime += tgutime;
1737 stime += tgstime;
1738 r->ru_nvcsw += p->signal->nvcsw;
1739 r->ru_nivcsw += p->signal->nivcsw;
1740 r->ru_minflt += p->signal->min_flt;
1741 r->ru_majflt += p->signal->maj_flt;
1742 r->ru_inblock += p->signal->inblock;
1743 r->ru_oublock += p->signal->oublock;
1744 if (maxrss < p->signal->maxrss)
1745 maxrss = p->signal->maxrss;
1746 t = p;
1747 do {
1748 accumulate_thread_rusage(t, r);
1749 t = next_thread(t);
1750 } while (t != p);
1751 break;
1752
1753 default:
1754 BUG();
1755 }
1756 unlock_task_sighand(p, &flags);
1757
1758 out:
1759 cputime_to_timeval(utime, &r->ru_utime);
1760 cputime_to_timeval(stime, &r->ru_stime);
1761
1762 if (who != RUSAGE_CHILDREN) {
1763 struct mm_struct *mm = get_task_mm(p);
1764 if (mm) {
1765 setmax_mm_hiwater_rss(&maxrss, mm);
1766 mmput(mm);
1767 }
1768 }
1769 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1770 }
1771
1772 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1773 {
1774 struct rusage r;
1775 k_getrusage(p, who, &r);
1776 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1777 }
1778
1779 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1780 {
1781 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1782 who != RUSAGE_THREAD)
1783 return -EINVAL;
1784 return getrusage(current, who, ru);
1785 }
1786
1787 SYSCALL_DEFINE1(umask, int, mask)
1788 {
1789 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1790 return mask;
1791 }
1792
1793 #ifdef CONFIG_CHECKPOINT_RESTORE
1794 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1795 {
1796 struct fd exe;
1797 struct inode *inode;
1798 int err;
1799
1800 exe = fdget(fd);
1801 if (!exe.file)
1802 return -EBADF;
1803
1804 inode = file_inode(exe.file);
1805
1806 /*
1807 * Because the original mm->exe_file points to executable file, make
1808 * sure that this one is executable as well, to avoid breaking an
1809 * overall picture.
1810 */
1811 err = -EACCES;
1812 if (!S_ISREG(inode->i_mode) ||
1813 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1814 goto exit;
1815
1816 err = inode_permission(inode, MAY_EXEC);
1817 if (err)
1818 goto exit;
1819
1820 down_write(&mm->mmap_sem);
1821
1822 /*
1823 * Forbid mm->exe_file change if old file still mapped.
1824 */
1825 err = -EBUSY;
1826 if (mm->exe_file) {
1827 struct vm_area_struct *vma;
1828
1829 for (vma = mm->mmap; vma; vma = vma->vm_next)
1830 if (vma->vm_file &&
1831 path_equal(&vma->vm_file->f_path,
1832 &mm->exe_file->f_path))
1833 goto exit_unlock;
1834 }
1835
1836 /*
1837 * The symlink can be changed only once, just to disallow arbitrary
1838 * transitions malicious software might bring in. This means one
1839 * could make a snapshot over all processes running and monitor
1840 * /proc/pid/exe changes to notice unusual activity if needed.
1841 */
1842 err = -EPERM;
1843 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1844 goto exit_unlock;
1845
1846 err = 0;
1847 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1848 exit_unlock:
1849 up_write(&mm->mmap_sem);
1850
1851 exit:
1852 fdput(exe);
1853 return err;
1854 }
1855
1856 static int prctl_set_mm(int opt, unsigned long addr,
1857 unsigned long arg4, unsigned long arg5)
1858 {
1859 unsigned long rlim = rlimit(RLIMIT_DATA);
1860 struct mm_struct *mm = current->mm;
1861 struct vm_area_struct *vma;
1862 int error;
1863
1864 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1865 return -EINVAL;
1866
1867 if (!capable(CAP_SYS_RESOURCE))
1868 return -EPERM;
1869
1870 if (opt == PR_SET_MM_EXE_FILE)
1871 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1872
1873 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1874 return -EINVAL;
1875
1876 error = -EINVAL;
1877
1878 down_read(&mm->mmap_sem);
1879 vma = find_vma(mm, addr);
1880
1881 switch (opt) {
1882 case PR_SET_MM_START_CODE:
1883 mm->start_code = addr;
1884 break;
1885 case PR_SET_MM_END_CODE:
1886 mm->end_code = addr;
1887 break;
1888 case PR_SET_MM_START_DATA:
1889 mm->start_data = addr;
1890 break;
1891 case PR_SET_MM_END_DATA:
1892 mm->end_data = addr;
1893 break;
1894
1895 case PR_SET_MM_START_BRK:
1896 if (addr <= mm->end_data)
1897 goto out;
1898
1899 if (rlim < RLIM_INFINITY &&
1900 (mm->brk - addr) +
1901 (mm->end_data - mm->start_data) > rlim)
1902 goto out;
1903
1904 mm->start_brk = addr;
1905 break;
1906
1907 case PR_SET_MM_BRK:
1908 if (addr <= mm->end_data)
1909 goto out;
1910
1911 if (rlim < RLIM_INFINITY &&
1912 (addr - mm->start_brk) +
1913 (mm->end_data - mm->start_data) > rlim)
1914 goto out;
1915
1916 mm->brk = addr;
1917 break;
1918
1919 /*
1920 * If command line arguments and environment
1921 * are placed somewhere else on stack, we can
1922 * set them up here, ARG_START/END to setup
1923 * command line argumets and ENV_START/END
1924 * for environment.
1925 */
1926 case PR_SET_MM_START_STACK:
1927 case PR_SET_MM_ARG_START:
1928 case PR_SET_MM_ARG_END:
1929 case PR_SET_MM_ENV_START:
1930 case PR_SET_MM_ENV_END:
1931 if (!vma) {
1932 error = -EFAULT;
1933 goto out;
1934 }
1935 if (opt == PR_SET_MM_START_STACK)
1936 mm->start_stack = addr;
1937 else if (opt == PR_SET_MM_ARG_START)
1938 mm->arg_start = addr;
1939 else if (opt == PR_SET_MM_ARG_END)
1940 mm->arg_end = addr;
1941 else if (opt == PR_SET_MM_ENV_START)
1942 mm->env_start = addr;
1943 else if (opt == PR_SET_MM_ENV_END)
1944 mm->env_end = addr;
1945 break;
1946
1947 /*
1948 * This doesn't move auxiliary vector itself
1949 * since it's pinned to mm_struct, but allow
1950 * to fill vector with new values. It's up
1951 * to a caller to provide sane values here
1952 * otherwise user space tools which use this
1953 * vector might be unhappy.
1954 */
1955 case PR_SET_MM_AUXV: {
1956 unsigned long user_auxv[AT_VECTOR_SIZE];
1957
1958 if (arg4 > sizeof(user_auxv))
1959 goto out;
1960 up_read(&mm->mmap_sem);
1961
1962 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
1963 return -EFAULT;
1964
1965 /* Make sure the last entry is always AT_NULL */
1966 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1967 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1968
1969 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1970
1971 task_lock(current);
1972 memcpy(mm->saved_auxv, user_auxv, arg4);
1973 task_unlock(current);
1974
1975 return 0;
1976 }
1977 default:
1978 goto out;
1979 }
1980
1981 error = 0;
1982 out:
1983 up_read(&mm->mmap_sem);
1984 return error;
1985 }
1986
1987 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1988 {
1989 return put_user(me->clear_child_tid, tid_addr);
1990 }
1991
1992 #else /* CONFIG_CHECKPOINT_RESTORE */
1993 static int prctl_set_mm(int opt, unsigned long addr,
1994 unsigned long arg4, unsigned long arg5)
1995 {
1996 return -EINVAL;
1997 }
1998 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
1999 {
2000 return -EINVAL;
2001 }
2002 #endif
2003
2004 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2005 unsigned long, arg4, unsigned long, arg5)
2006 {
2007 struct task_struct *me = current;
2008 unsigned char comm[sizeof(me->comm)];
2009 long error;
2010
2011 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2012 if (error != -ENOSYS)
2013 return error;
2014
2015 error = 0;
2016 switch (option) {
2017 case PR_SET_PDEATHSIG:
2018 if (!valid_signal(arg2)) {
2019 error = -EINVAL;
2020 break;
2021 }
2022 me->pdeath_signal = arg2;
2023 break;
2024 case PR_GET_PDEATHSIG:
2025 error = put_user(me->pdeath_signal, (int __user *)arg2);
2026 break;
2027 case PR_GET_DUMPABLE:
2028 error = get_dumpable(me->mm);
2029 break;
2030 case PR_SET_DUMPABLE:
2031 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2032 error = -EINVAL;
2033 break;
2034 }
2035 set_dumpable(me->mm, arg2);
2036 break;
2037
2038 case PR_SET_UNALIGN:
2039 error = SET_UNALIGN_CTL(me, arg2);
2040 break;
2041 case PR_GET_UNALIGN:
2042 error = GET_UNALIGN_CTL(me, arg2);
2043 break;
2044 case PR_SET_FPEMU:
2045 error = SET_FPEMU_CTL(me, arg2);
2046 break;
2047 case PR_GET_FPEMU:
2048 error = GET_FPEMU_CTL(me, arg2);
2049 break;
2050 case PR_SET_FPEXC:
2051 error = SET_FPEXC_CTL(me, arg2);
2052 break;
2053 case PR_GET_FPEXC:
2054 error = GET_FPEXC_CTL(me, arg2);
2055 break;
2056 case PR_GET_TIMING:
2057 error = PR_TIMING_STATISTICAL;
2058 break;
2059 case PR_SET_TIMING:
2060 if (arg2 != PR_TIMING_STATISTICAL)
2061 error = -EINVAL;
2062 break;
2063 case PR_SET_NAME:
2064 comm[sizeof(me->comm) - 1] = 0;
2065 if (strncpy_from_user(comm, (char __user *)arg2,
2066 sizeof(me->comm) - 1) < 0)
2067 return -EFAULT;
2068 set_task_comm(me, comm);
2069 proc_comm_connector(me);
2070 break;
2071 case PR_GET_NAME:
2072 get_task_comm(comm, me);
2073 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2074 return -EFAULT;
2075 break;
2076 case PR_GET_ENDIAN:
2077 error = GET_ENDIAN(me, arg2);
2078 break;
2079 case PR_SET_ENDIAN:
2080 error = SET_ENDIAN(me, arg2);
2081 break;
2082 case PR_GET_SECCOMP:
2083 error = prctl_get_seccomp();
2084 break;
2085 case PR_SET_SECCOMP:
2086 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2087 break;
2088 case PR_GET_TSC:
2089 error = GET_TSC_CTL(arg2);
2090 break;
2091 case PR_SET_TSC:
2092 error = SET_TSC_CTL(arg2);
2093 break;
2094 case PR_TASK_PERF_EVENTS_DISABLE:
2095 error = perf_event_task_disable();
2096 break;
2097 case PR_TASK_PERF_EVENTS_ENABLE:
2098 error = perf_event_task_enable();
2099 break;
2100 case PR_GET_TIMERSLACK:
2101 error = current->timer_slack_ns;
2102 break;
2103 case PR_SET_TIMERSLACK:
2104 if (arg2 <= 0)
2105 current->timer_slack_ns =
2106 current->default_timer_slack_ns;
2107 else
2108 current->timer_slack_ns = arg2;
2109 break;
2110 case PR_MCE_KILL:
2111 if (arg4 | arg5)
2112 return -EINVAL;
2113 switch (arg2) {
2114 case PR_MCE_KILL_CLEAR:
2115 if (arg3 != 0)
2116 return -EINVAL;
2117 current->flags &= ~PF_MCE_PROCESS;
2118 break;
2119 case PR_MCE_KILL_SET:
2120 current->flags |= PF_MCE_PROCESS;
2121 if (arg3 == PR_MCE_KILL_EARLY)
2122 current->flags |= PF_MCE_EARLY;
2123 else if (arg3 == PR_MCE_KILL_LATE)
2124 current->flags &= ~PF_MCE_EARLY;
2125 else if (arg3 == PR_MCE_KILL_DEFAULT)
2126 current->flags &=
2127 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2128 else
2129 return -EINVAL;
2130 break;
2131 default:
2132 return -EINVAL;
2133 }
2134 break;
2135 case PR_MCE_KILL_GET:
2136 if (arg2 | arg3 | arg4 | arg5)
2137 return -EINVAL;
2138 if (current->flags & PF_MCE_PROCESS)
2139 error = (current->flags & PF_MCE_EARLY) ?
2140 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2141 else
2142 error = PR_MCE_KILL_DEFAULT;
2143 break;
2144 case PR_SET_MM:
2145 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2146 break;
2147 case PR_GET_TID_ADDRESS:
2148 error = prctl_get_tid_address(me, (int __user **)arg2);
2149 break;
2150 case PR_SET_CHILD_SUBREAPER:
2151 me->signal->is_child_subreaper = !!arg2;
2152 break;
2153 case PR_GET_CHILD_SUBREAPER:
2154 error = put_user(me->signal->is_child_subreaper,
2155 (int __user *)arg2);
2156 break;
2157 case PR_SET_NO_NEW_PRIVS:
2158 if (arg2 != 1 || arg3 || arg4 || arg5)
2159 return -EINVAL;
2160
2161 current->no_new_privs = 1;
2162 break;
2163 case PR_GET_NO_NEW_PRIVS:
2164 if (arg2 || arg3 || arg4 || arg5)
2165 return -EINVAL;
2166 return current->no_new_privs ? 1 : 0;
2167 default:
2168 error = -EINVAL;
2169 break;
2170 }
2171 return error;
2172 }
2173
2174 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2175 struct getcpu_cache __user *, unused)
2176 {
2177 int err = 0;
2178 int cpu = raw_smp_processor_id();
2179 if (cpup)
2180 err |= put_user(cpu, cpup);
2181 if (nodep)
2182 err |= put_user(cpu_to_node(cpu), nodep);
2183 return err ? -EFAULT : 0;
2184 }
2185
2186 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2187
2188 static int __orderly_poweroff(bool force)
2189 {
2190 char **argv;
2191 static char *envp[] = {
2192 "HOME=/",
2193 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2194 NULL
2195 };
2196 int ret;
2197
2198 argv = argv_split(GFP_KERNEL, poweroff_cmd, NULL);
2199 if (argv) {
2200 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
2201 argv_free(argv);
2202 } else {
2203 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2204 __func__, poweroff_cmd);
2205 ret = -ENOMEM;
2206 }
2207
2208 if (ret && force) {
2209 printk(KERN_WARNING "Failed to start orderly shutdown: "
2210 "forcing the issue\n");
2211 /*
2212 * I guess this should try to kick off some daemon to sync and
2213 * poweroff asap. Or not even bother syncing if we're doing an
2214 * emergency shutdown?
2215 */
2216 emergency_sync();
2217 kernel_power_off();
2218 }
2219
2220 return ret;
2221 }
2222
2223 static bool poweroff_force;
2224
2225 static void poweroff_work_func(struct work_struct *work)
2226 {
2227 __orderly_poweroff(poweroff_force);
2228 }
2229
2230 static DECLARE_WORK(poweroff_work, poweroff_work_func);
2231
2232 /**
2233 * orderly_poweroff - Trigger an orderly system poweroff
2234 * @force: force poweroff if command execution fails
2235 *
2236 * This may be called from any context to trigger a system shutdown.
2237 * If the orderly shutdown fails, it will force an immediate shutdown.
2238 */
2239 int orderly_poweroff(bool force)
2240 {
2241 if (force) /* do not override the pending "true" */
2242 poweroff_force = true;
2243 schedule_work(&poweroff_work);
2244 return 0;
2245 }
2246 EXPORT_SYMBOL_GPL(orderly_poweroff);
Linus' kernel tree