projects/linux·blob v2.6.24-rc6 · refs

1820 lines · 42.7 KiB

1/*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/module.h>
8#include <linux/mm.h>
9#include <linux/utsname.h>
10#include <linux/mman.h>
11#include <linux/smp_lock.h>
12#include <linux/notifier.h>
13#include <linux/reboot.h>
14#include <linux/prctl.h>
15#include <linux/highuid.h>
16#include <linux/fs.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
37#include <linux/compat.h>
38#include <linux/syscalls.h>
39#include <linux/kprobes.h>
40#include <linux/user_namespace.h>
41
42#include <asm/uaccess.h>
43#include <asm/io.h>
44#include <asm/unistd.h>
45
46#ifndef SET_UNALIGN_CTL
47# define SET_UNALIGN_CTL(a,b) (-EINVAL)
48#endif
49#ifndef GET_UNALIGN_CTL
50# define GET_UNALIGN_CTL(a,b) (-EINVAL)
51#endif
52#ifndef SET_FPEMU_CTL
53# define SET_FPEMU_CTL(a,b) (-EINVAL)
54#endif
55#ifndef GET_FPEMU_CTL
56# define GET_FPEMU_CTL(a,b) (-EINVAL)
57#endif
58#ifndef SET_FPEXC_CTL
59# define SET_FPEXC_CTL(a,b) (-EINVAL)
60#endif
61#ifndef GET_FPEXC_CTL
62# define GET_FPEXC_CTL(a,b) (-EINVAL)
63#endif
64#ifndef GET_ENDIAN
65# define GET_ENDIAN(a,b) (-EINVAL)
66#endif
67#ifndef SET_ENDIAN
68# define SET_ENDIAN(a,b) (-EINVAL)
69#endif
70
71/*
72 * this is where the system-wide overflow UID and GID are defined, for
73 * architectures that now have 32-bit UID/GID but didn't in the past
74 */
75
76int overflowuid = DEFAULT_OVERFLOWUID;
77int overflowgid = DEFAULT_OVERFLOWGID;
78
79#ifdef CONFIG_UID16
80EXPORT_SYMBOL(overflowuid);
81EXPORT_SYMBOL(overflowgid);
82#endif
83
84/*
85 * the same as above, but for filesystems which can only store a 16-bit
86 * UID and GID. as such, this is needed on all architectures
87 */
88
89int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
90int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
91
92EXPORT_SYMBOL(fs_overflowuid);
93EXPORT_SYMBOL(fs_overflowgid);
94
95/*
96 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
97 */
98
99int C_A_D = 1;
100struct pid *cad_pid;
101EXPORT_SYMBOL(cad_pid);
102
103/*
104 * If set, this is used for preparing the system to power off.
105 */
106
107void (*pm_power_off_prepare)(void);
108
109static int set_one_prio(struct task_struct *p, int niceval, int error)
110{
111 int no_nice;
112
113 if (p->uid != current->euid &&
114 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
115 error = -EPERM;
116 goto out;
117 }
118 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
119 error = -EACCES;
120 goto out;
121 }
122 no_nice = security_task_setnice(p, niceval);
123 if (no_nice) {
124 error = no_nice;
125 goto out;
126 }
127 if (error == -ESRCH)
128 error = 0;
129 set_user_nice(p, niceval);
130out:
131 return error;
132}
133
134asmlinkage long sys_setpriority(int which, int who, int niceval)
135{
136 struct task_struct *g, *p;
137 struct user_struct *user;
138 int error = -EINVAL;
139 struct pid *pgrp;
140
141 if (which > PRIO_USER || which < PRIO_PROCESS)
142 goto out;
143
144 /* normalize: avoid signed division (rounding problems) */
145 error = -ESRCH;
146 if (niceval < -20)
147 niceval = -20;
148 if (niceval > 19)
149 niceval = 19;
150
151 read_lock(&tasklist_lock);
152 switch (which) {
153 case PRIO_PROCESS:
154 if (who)
155 p = find_task_by_vpid(who);
156 else
157 p = current;
158 if (p)
159 error = set_one_prio(p, niceval, error);
160 break;
161 case PRIO_PGRP:
162 if (who)
163 pgrp = find_vpid(who);
164 else
165 pgrp = task_pgrp(current);
166 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
167 error = set_one_prio(p, niceval, error);
168 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
169 break;
170 case PRIO_USER:
171 user = current->user;
172 if (!who)
173 who = current->uid;
174 else
175 if ((who != current->uid) && !(user = find_user(who)))
176 goto out_unlock; /* No processes for this user */
177
178 do_each_thread(g, p)
179 if (p->uid == who)
180 error = set_one_prio(p, niceval, error);
181 while_each_thread(g, p);
182 if (who != current->uid)
183 free_uid(user); /* For find_user() */
184 break;
185 }
186out_unlock:
187 read_unlock(&tasklist_lock);
188out:
189 return error;
190}
191
192/*
193 * Ugh. To avoid negative return values, "getpriority()" will
194 * not return the normal nice-value, but a negated value that
195 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
196 * to stay compatible.
197 */
198asmlinkage long sys_getpriority(int which, int who)
199{
200 struct task_struct *g, *p;
201 struct user_struct *user;
202 long niceval, retval = -ESRCH;
203 struct pid *pgrp;
204
205 if (which > PRIO_USER || which < PRIO_PROCESS)
206 return -EINVAL;
207
208 read_lock(&tasklist_lock);
209 switch (which) {
210 case PRIO_PROCESS:
211 if (who)
212 p = find_task_by_vpid(who);
213 else
214 p = current;
215 if (p) {
216 niceval = 20 - task_nice(p);
217 if (niceval > retval)
218 retval = niceval;
219 }
220 break;
221 case PRIO_PGRP:
222 if (who)
223 pgrp = find_vpid(who);
224 else
225 pgrp = task_pgrp(current);
226 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
227 niceval = 20 - task_nice(p);
228 if (niceval > retval)
229 retval = niceval;
230 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
231 break;
232 case PRIO_USER:
233 user = current->user;
234 if (!who)
235 who = current->uid;
236 else
237 if ((who != current->uid) && !(user = find_user(who)))
238 goto out_unlock; /* No processes for this user */
239
240 do_each_thread(g, p)
241 if (p->uid == who) {
242 niceval = 20 - task_nice(p);
243 if (niceval > retval)
244 retval = niceval;
245 }
246 while_each_thread(g, p);
247 if (who != current->uid)
248 free_uid(user); /* for find_user() */
249 break;
250 }
251out_unlock:
252 read_unlock(&tasklist_lock);
253
254 return retval;
255}
256
257/**
258 * emergency_restart - reboot the system
259 *
260 * Without shutting down any hardware or taking any locks
261 * reboot the system. This is called when we know we are in
262 * trouble so this is our best effort to reboot. This is
263 * safe to call in interrupt context.
264 */
265void emergency_restart(void)
266{
267 machine_emergency_restart();
268}
269EXPORT_SYMBOL_GPL(emergency_restart);
270
271static void kernel_restart_prepare(char *cmd)
272{
273 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
274 system_state = SYSTEM_RESTART;
275 device_shutdown();
276 sysdev_shutdown();
277}
278
279/**
280 * kernel_restart - reboot the system
281 * @cmd: pointer to buffer containing command to execute for restart
282 * or %NULL
283 *
284 * Shutdown everything and perform a clean reboot.
285 * This is not safe to call in interrupt context.
286 */
287void kernel_restart(char *cmd)
288{
289 kernel_restart_prepare(cmd);
290 if (!cmd)
291 printk(KERN_EMERG "Restarting system.\n");
292 else
293 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
294 machine_restart(cmd);
295}
296EXPORT_SYMBOL_GPL(kernel_restart);
297
298/**
299 * kernel_kexec - reboot the system
300 *
301 * Move into place and start executing a preloaded standalone
302 * executable. If nothing was preloaded return an error.
303 */
304static void kernel_kexec(void)
305{
306#ifdef CONFIG_KEXEC
307 struct kimage *image;
308 image = xchg(&kexec_image, NULL);
309 if (!image)
310 return;
311 kernel_restart_prepare(NULL);
312 printk(KERN_EMERG "Starting new kernel\n");
313 machine_shutdown();
314 machine_kexec(image);
315#endif
316}
317
318void kernel_shutdown_prepare(enum system_states state)
319{
320 blocking_notifier_call_chain(&reboot_notifier_list,
321 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
322 system_state = state;
323 device_shutdown();
324}
325/**
326 * kernel_halt - halt the system
327 *
328 * Shutdown everything and perform a clean system halt.
329 */
330void kernel_halt(void)
331{
332 kernel_shutdown_prepare(SYSTEM_HALT);
333 sysdev_shutdown();
334 printk(KERN_EMERG "System halted.\n");
335 machine_halt();
336}
337
338EXPORT_SYMBOL_GPL(kernel_halt);
339
340/**
341 * kernel_power_off - power_off the system
342 *
343 * Shutdown everything and perform a clean system power_off.
344 */
345void kernel_power_off(void)
346{
347 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
348 if (pm_power_off_prepare)
349 pm_power_off_prepare();
350 disable_nonboot_cpus();
351 sysdev_shutdown();
352 printk(KERN_EMERG "Power down.\n");
353 machine_power_off();
354}
355EXPORT_SYMBOL_GPL(kernel_power_off);
356/*
357 * Reboot system call: for obvious reasons only root may call it,
358 * and even root needs to set up some magic numbers in the registers
359 * so that some mistake won't make this reboot the whole machine.
360 * You can also set the meaning of the ctrl-alt-del-key here.
361 *
362 * reboot doesn't sync: do that yourself before calling this.
363 */
364asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
365{
366 char buffer[256];
367
368 /* We only trust the superuser with rebooting the system. */
369 if (!capable(CAP_SYS_BOOT))
370 return -EPERM;
371
372 /* For safety, we require "magic" arguments. */
373 if (magic1 != LINUX_REBOOT_MAGIC1 ||
374 (magic2 != LINUX_REBOOT_MAGIC2 &&
375 magic2 != LINUX_REBOOT_MAGIC2A &&
376 magic2 != LINUX_REBOOT_MAGIC2B &&
377 magic2 != LINUX_REBOOT_MAGIC2C))
378 return -EINVAL;
379
380 /* Instead of trying to make the power_off code look like
381 * halt when pm_power_off is not set do it the easy way.
382 */
383 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
384 cmd = LINUX_REBOOT_CMD_HALT;
385
386 lock_kernel();
387 switch (cmd) {
388 case LINUX_REBOOT_CMD_RESTART:
389 kernel_restart(NULL);
390 break;
391
392 case LINUX_REBOOT_CMD_CAD_ON:
393 C_A_D = 1;
394 break;
395
396 case LINUX_REBOOT_CMD_CAD_OFF:
397 C_A_D = 0;
398 break;
399
400 case LINUX_REBOOT_CMD_HALT:
401 kernel_halt();
402 unlock_kernel();
403 do_exit(0);
404 break;
405
406 case LINUX_REBOOT_CMD_POWER_OFF:
407 kernel_power_off();
408 unlock_kernel();
409 do_exit(0);
410 break;
411
412 case LINUX_REBOOT_CMD_RESTART2:
413 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
414 unlock_kernel();
415 return -EFAULT;
416 }
417 buffer[sizeof(buffer) - 1] = '\0';
418
419 kernel_restart(buffer);
420 break;
421
422 case LINUX_REBOOT_CMD_KEXEC:
423 kernel_kexec();
424 unlock_kernel();
425 return -EINVAL;
426
427#ifdef CONFIG_HIBERNATION
428 case LINUX_REBOOT_CMD_SW_SUSPEND:
429 {
430 int ret = hibernate();
431 unlock_kernel();
432 return ret;
433 }
434#endif
435
436 default:
437 unlock_kernel();
438 return -EINVAL;
439 }
440 unlock_kernel();
441 return 0;
442}
443
444static void deferred_cad(struct work_struct *dummy)
445{
446 kernel_restart(NULL);
447}
448
449/*
450 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
451 * As it's called within an interrupt, it may NOT sync: the only choice
452 * is whether to reboot at once, or just ignore the ctrl-alt-del.
453 */
454void ctrl_alt_del(void)
455{
456 static DECLARE_WORK(cad_work, deferred_cad);
457
458 if (C_A_D)
459 schedule_work(&cad_work);
460 else
461 kill_cad_pid(SIGINT, 1);
462}
463
464/*
465 * Unprivileged users may change the real gid to the effective gid
466 * or vice versa. (BSD-style)
467 *
468 * If you set the real gid at all, or set the effective gid to a value not
469 * equal to the real gid, then the saved gid is set to the new effective gid.
470 *
471 * This makes it possible for a setgid program to completely drop its
472 * privileges, which is often a useful assertion to make when you are doing
473 * a security audit over a program.
474 *
475 * The general idea is that a program which uses just setregid() will be
476 * 100% compatible with BSD. A program which uses just setgid() will be
477 * 100% compatible with POSIX with saved IDs.
478 *
479 * SMP: There are not races, the GIDs are checked only by filesystem
480 * operations (as far as semantic preservation is concerned).
481 */
482asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
483{
484 int old_rgid = current->gid;
485 int old_egid = current->egid;
486 int new_rgid = old_rgid;
487 int new_egid = old_egid;
488 int retval;
489
490 retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
491 if (retval)
492 return retval;
493
494 if (rgid != (gid_t) -1) {
495 if ((old_rgid == rgid) ||
496 (current->egid==rgid) ||
497 capable(CAP_SETGID))
498 new_rgid = rgid;
499 else
500 return -EPERM;
501 }
502 if (egid != (gid_t) -1) {
503 if ((old_rgid == egid) ||
504 (current->egid == egid) ||
505 (current->sgid == egid) ||
506 capable(CAP_SETGID))
507 new_egid = egid;
508 else
509 return -EPERM;
510 }
511 if (new_egid != old_egid) {
512 set_dumpable(current->mm, suid_dumpable);
513 smp_wmb();
514 }
515 if (rgid != (gid_t) -1 ||
516 (egid != (gid_t) -1 && egid != old_rgid))
517 current->sgid = new_egid;
518 current->fsgid = new_egid;
519 current->egid = new_egid;
520 current->gid = new_rgid;
521 key_fsgid_changed(current);
522 proc_id_connector(current, PROC_EVENT_GID);
523 return 0;
524}
525
526/*
527 * setgid() is implemented like SysV w/ SAVED_IDS
528 *
529 * SMP: Same implicit races as above.
530 */
531asmlinkage long sys_setgid(gid_t gid)
532{
533 int old_egid = current->egid;
534 int retval;
535
536 retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
537 if (retval)
538 return retval;
539
540 if (capable(CAP_SETGID)) {
541 if (old_egid != gid) {
542 set_dumpable(current->mm, suid_dumpable);
543 smp_wmb();
544 }
545 current->gid = current->egid = current->sgid = current->fsgid = gid;
546 } else if ((gid == current->gid) || (gid == current->sgid)) {
547 if (old_egid != gid) {
548 set_dumpable(current->mm, suid_dumpable);
549 smp_wmb();
550 }
551 current->egid = current->fsgid = gid;
552 }
553 else
554 return -EPERM;
555
556 key_fsgid_changed(current);
557 proc_id_connector(current, PROC_EVENT_GID);
558 return 0;
559}
560
561static int set_user(uid_t new_ruid, int dumpclear)
562{
563 struct user_struct *new_user;
564
565 new_user = alloc_uid(current->nsproxy->user_ns, new_ruid);
566 if (!new_user)
567 return -EAGAIN;
568
569 if (atomic_read(&new_user->processes) >=
570 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
571 new_user != current->nsproxy->user_ns->root_user) {
572 free_uid(new_user);
573 return -EAGAIN;
574 }
575
576 switch_uid(new_user);
577
578 if (dumpclear) {
579 set_dumpable(current->mm, suid_dumpable);
580 smp_wmb();
581 }
582 current->uid = new_ruid;
583 return 0;
584}
585
586/*
587 * Unprivileged users may change the real uid to the effective uid
588 * or vice versa. (BSD-style)
589 *
590 * If you set the real uid at all, or set the effective uid to a value not
591 * equal to the real uid, then the saved uid is set to the new effective uid.
592 *
593 * This makes it possible for a setuid program to completely drop its
594 * privileges, which is often a useful assertion to make when you are doing
595 * a security audit over a program.
596 *
597 * The general idea is that a program which uses just setreuid() will be
598 * 100% compatible with BSD. A program which uses just setuid() will be
599 * 100% compatible with POSIX with saved IDs.
600 */
601asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
602{
603 int old_ruid, old_euid, old_suid, new_ruid, new_euid;
604 int retval;
605
606 retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
607 if (retval)
608 return retval;
609
610 new_ruid = old_ruid = current->uid;
611 new_euid = old_euid = current->euid;
612 old_suid = current->suid;
613
614 if (ruid != (uid_t) -1) {
615 new_ruid = ruid;
616 if ((old_ruid != ruid) &&
617 (current->euid != ruid) &&
618 !capable(CAP_SETUID))
619 return -EPERM;
620 }
621
622 if (euid != (uid_t) -1) {
623 new_euid = euid;
624 if ((old_ruid != euid) &&
625 (current->euid != euid) &&
626 (current->suid != euid) &&
627 !capable(CAP_SETUID))
628 return -EPERM;
629 }
630
631 if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
632 return -EAGAIN;
633
634 if (new_euid != old_euid) {
635 set_dumpable(current->mm, suid_dumpable);
636 smp_wmb();
637 }
638 current->fsuid = current->euid = new_euid;
639 if (ruid != (uid_t) -1 ||
640 (euid != (uid_t) -1 && euid != old_ruid))
641 current->suid = current->euid;
642 current->fsuid = current->euid;
643
644 key_fsuid_changed(current);
645 proc_id_connector(current, PROC_EVENT_UID);
646
647 return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
648}
649
650
651
652/*
653 * setuid() is implemented like SysV with SAVED_IDS
654 *
655 * Note that SAVED_ID's is deficient in that a setuid root program
656 * like sendmail, for example, cannot set its uid to be a normal
657 * user and then switch back, because if you're root, setuid() sets
658 * the saved uid too. If you don't like this, blame the bright people
659 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
660 * will allow a root program to temporarily drop privileges and be able to
661 * regain them by swapping the real and effective uid.
662 */
663asmlinkage long sys_setuid(uid_t uid)
664{
665 int old_euid = current->euid;
666 int old_ruid, old_suid, new_suid;
667 int retval;
668
669 retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
670 if (retval)
671 return retval;
672
673 old_ruid = current->uid;
674 old_suid = current->suid;
675 new_suid = old_suid;
676
677 if (capable(CAP_SETUID)) {
678 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
679 return -EAGAIN;
680 new_suid = uid;
681 } else if ((uid != current->uid) && (uid != new_suid))
682 return -EPERM;
683
684 if (old_euid != uid) {
685 set_dumpable(current->mm, suid_dumpable);
686 smp_wmb();
687 }
688 current->fsuid = current->euid = uid;
689 current->suid = new_suid;
690
691 key_fsuid_changed(current);
692 proc_id_connector(current, PROC_EVENT_UID);
693
694 return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
695}
696
697
698/*
699 * This function implements a generic ability to update ruid, euid,
700 * and suid. This allows you to implement the 4.4 compatible seteuid().
701 */
702asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
703{
704 int old_ruid = current->uid;
705 int old_euid = current->euid;
706 int old_suid = current->suid;
707 int retval;
708
709 retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
710 if (retval)
711 return retval;
712
713 if (!capable(CAP_SETUID)) {
714 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
715 (ruid != current->euid) && (ruid != current->suid))
716 return -EPERM;
717 if ((euid != (uid_t) -1) && (euid != current->uid) &&
718 (euid != current->euid) && (euid != current->suid))
719 return -EPERM;
720 if ((suid != (uid_t) -1) && (suid != current->uid) &&
721 (suid != current->euid) && (suid != current->suid))
722 return -EPERM;
723 }
724 if (ruid != (uid_t) -1) {
725 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
726 return -EAGAIN;
727 }
728 if (euid != (uid_t) -1) {
729 if (euid != current->euid) {
730 set_dumpable(current->mm, suid_dumpable);
731 smp_wmb();
732 }
733 current->euid = euid;
734 }
735 current->fsuid = current->euid;
736 if (suid != (uid_t) -1)
737 current->suid = suid;
738
739 key_fsuid_changed(current);
740 proc_id_connector(current, PROC_EVENT_UID);
741
742 return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
743}
744
745asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
746{
747 int retval;
748
749 if (!(retval = put_user(current->uid, ruid)) &&
750 !(retval = put_user(current->euid, euid)))
751 retval = put_user(current->suid, suid);
752
753 return retval;
754}
755
756/*
757 * Same as above, but for rgid, egid, sgid.
758 */
759asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
760{
761 int retval;
762
763 retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
764 if (retval)
765 return retval;
766
767 if (!capable(CAP_SETGID)) {
768 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
769 (rgid != current->egid) && (rgid != current->sgid))
770 return -EPERM;
771 if ((egid != (gid_t) -1) && (egid != current->gid) &&
772 (egid != current->egid) && (egid != current->sgid))
773 return -EPERM;
774 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
775 (sgid != current->egid) && (sgid != current->sgid))
776 return -EPERM;
777 }
778 if (egid != (gid_t) -1) {
779 if (egid != current->egid) {
780 set_dumpable(current->mm, suid_dumpable);
781 smp_wmb();
782 }
783 current->egid = egid;
784 }
785 current->fsgid = current->egid;
786 if (rgid != (gid_t) -1)
787 current->gid = rgid;
788 if (sgid != (gid_t) -1)
789 current->sgid = sgid;
790
791 key_fsgid_changed(current);
792 proc_id_connector(current, PROC_EVENT_GID);
793 return 0;
794}
795
796asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
797{
798 int retval;
799
800 if (!(retval = put_user(current->gid, rgid)) &&
801 !(retval = put_user(current->egid, egid)))
802 retval = put_user(current->sgid, sgid);
803
804 return retval;
805}
806
807
808/*
809 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
810 * is used for "access()" and for the NFS daemon (letting nfsd stay at
811 * whatever uid it wants to). It normally shadows "euid", except when
812 * explicitly set by setfsuid() or for access..
813 */
814asmlinkage long sys_setfsuid(uid_t uid)
815{
816 int old_fsuid;
817
818 old_fsuid = current->fsuid;
819 if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
820 return old_fsuid;
821
822 if (uid == current->uid || uid == current->euid ||
823 uid == current->suid || uid == current->fsuid ||
824 capable(CAP_SETUID)) {
825 if (uid != old_fsuid) {
826 set_dumpable(current->mm, suid_dumpable);
827 smp_wmb();
828 }
829 current->fsuid = uid;
830 }
831
832 key_fsuid_changed(current);
833 proc_id_connector(current, PROC_EVENT_UID);
834
835 security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
836
837 return old_fsuid;
838}
839
840/*
841 * Samma på svenska..
842 */
843asmlinkage long sys_setfsgid(gid_t gid)
844{
845 int old_fsgid;
846
847 old_fsgid = current->fsgid;
848 if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
849 return old_fsgid;
850
851 if (gid == current->gid || gid == current->egid ||
852 gid == current->sgid || gid == current->fsgid ||
853 capable(CAP_SETGID)) {
854 if (gid != old_fsgid) {
855 set_dumpable(current->mm, suid_dumpable);
856 smp_wmb();
857 }
858 current->fsgid = gid;
859 key_fsgid_changed(current);
860 proc_id_connector(current, PROC_EVENT_GID);
861 }
862 return old_fsgid;
863}
864
865asmlinkage long sys_times(struct tms __user * tbuf)
866{
867 /*
868 * In the SMP world we might just be unlucky and have one of
869 * the times increment as we use it. Since the value is an
870 * atomically safe type this is just fine. Conceptually its
871 * as if the syscall took an instant longer to occur.
872 */
873 if (tbuf) {
874 struct tms tmp;
875 struct task_struct *tsk = current;
876 struct task_struct *t;
877 cputime_t utime, stime, cutime, cstime;
878
879 spin_lock_irq(&tsk->sighand->siglock);
880 utime = tsk->signal->utime;
881 stime = tsk->signal->stime;
882 t = tsk;
883 do {
884 utime = cputime_add(utime, t->utime);
885 stime = cputime_add(stime, t->stime);
886 t = next_thread(t);
887 } while (t != tsk);
888
889 cutime = tsk->signal->cutime;
890 cstime = tsk->signal->cstime;
891 spin_unlock_irq(&tsk->sighand->siglock);
892
893 tmp.tms_utime = cputime_to_clock_t(utime);
894 tmp.tms_stime = cputime_to_clock_t(stime);
895 tmp.tms_cutime = cputime_to_clock_t(cutime);
896 tmp.tms_cstime = cputime_to_clock_t(cstime);
897 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
898 return -EFAULT;
899 }
900 return (long) jiffies_64_to_clock_t(get_jiffies_64());
901}
902
903/*
904 * This needs some heavy checking ...
905 * I just haven't the stomach for it. I also don't fully
906 * understand sessions/pgrp etc. Let somebody who does explain it.
907 *
908 * OK, I think I have the protection semantics right.... this is really
909 * only important on a multi-user system anyway, to make sure one user
910 * can't send a signal to a process owned by another. -TYT, 12/12/91
911 *
912 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
913 * LBT 04.03.94
914 */
915asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
916{
917 struct task_struct *p;
918 struct task_struct *group_leader = current->group_leader;
919 int err = -EINVAL;
920 struct pid_namespace *ns;
921
922 if (!pid)
923 pid = task_pid_vnr(group_leader);
924 if (!pgid)
925 pgid = pid;
926 if (pgid < 0)
927 return -EINVAL;
928
929 /* From this point forward we keep holding onto the tasklist lock
930 * so that our parent does not change from under us. -DaveM
931 */
932 ns = current->nsproxy->pid_ns;
933
934 write_lock_irq(&tasklist_lock);
935
936 err = -ESRCH;
937 p = find_task_by_pid_ns(pid, ns);
938 if (!p)
939 goto out;
940
941 err = -EINVAL;
942 if (!thread_group_leader(p))
943 goto out;
944
945 if (p->real_parent->tgid == group_leader->tgid) {
946 err = -EPERM;
947 if (task_session(p) != task_session(group_leader))
948 goto out;
949 err = -EACCES;
950 if (p->did_exec)
951 goto out;
952 } else {
953 err = -ESRCH;
954 if (p != group_leader)
955 goto out;
956 }
957
958 err = -EPERM;
959 if (p->signal->leader)
960 goto out;
961
962 if (pgid != pid) {
963 struct task_struct *g;
964
965 g = find_task_by_pid_type_ns(PIDTYPE_PGID, pgid, ns);
966 if (!g || task_session(g) != task_session(group_leader))
967 goto out;
968 }
969
970 err = security_task_setpgid(p, pgid);
971 if (err)
972 goto out;
973
974 if (task_pgrp_nr_ns(p, ns) != pgid) {
975 struct pid *pid;
976
977 detach_pid(p, PIDTYPE_PGID);
978 pid = find_vpid(pgid);
979 attach_pid(p, PIDTYPE_PGID, pid);
980 set_task_pgrp(p, pid_nr(pid));
981 }
982
983 err = 0;
984out:
985 /* All paths lead to here, thus we are safe. -DaveM */
986 write_unlock_irq(&tasklist_lock);
987 return err;
988}
989
990asmlinkage long sys_getpgid(pid_t pid)
991{
992 if (!pid)
993 return task_pgrp_vnr(current);
994 else {
995 int retval;
996 struct task_struct *p;
997 struct pid_namespace *ns;
998
999 ns = current->nsproxy->pid_ns;
1000
1001 read_lock(&tasklist_lock);
1002 p = find_task_by_pid_ns(pid, ns);
1003 retval = -ESRCH;
1004 if (p) {
1005 retval = security_task_getpgid(p);
1006 if (!retval)
1007 retval = task_pgrp_nr_ns(p, ns);
1008 }
1009 read_unlock(&tasklist_lock);
1010 return retval;
1011 }
1012}
1013
1014#ifdef __ARCH_WANT_SYS_GETPGRP
1015
1016asmlinkage long sys_getpgrp(void)
1017{
1018 /* SMP - assuming writes are word atomic this is fine */
1019 return task_pgrp_vnr(current);
1020}
1021
1022#endif
1023
1024asmlinkage long sys_getsid(pid_t pid)
1025{
1026 if (!pid)
1027 return task_session_vnr(current);
1028 else {
1029 int retval;
1030 struct task_struct *p;
1031 struct pid_namespace *ns;
1032
1033 ns = current->nsproxy->pid_ns;
1034
1035 read_lock(&tasklist_lock);
1036 p = find_task_by_pid_ns(pid, ns);
1037 retval = -ESRCH;
1038 if (p) {
1039 retval = security_task_getsid(p);
1040 if (!retval)
1041 retval = task_session_nr_ns(p, ns);
1042 }
1043 read_unlock(&tasklist_lock);
1044 return retval;
1045 }
1046}
1047
1048asmlinkage long sys_setsid(void)
1049{
1050 struct task_struct *group_leader = current->group_leader;
1051 pid_t session;
1052 int err = -EPERM;
1053
1054 write_lock_irq(&tasklist_lock);
1055
1056 /* Fail if I am already a session leader */
1057 if (group_leader->signal->leader)
1058 goto out;
1059
1060 session = group_leader->pid;
1061 /* Fail if a process group id already exists that equals the
1062 * proposed session id.
1063 *
1064 * Don't check if session id == 1 because kernel threads use this
1065 * session id and so the check will always fail and make it so
1066 * init cannot successfully call setsid.
1067 */
1068 if (session > 1 && find_task_by_pid_type_ns(PIDTYPE_PGID,
1069 session, &init_pid_ns))
1070 goto out;
1071
1072 group_leader->signal->leader = 1;
1073 __set_special_pids(session, session);
1074
1075 spin_lock(&group_leader->sighand->siglock);
1076 group_leader->signal->tty = NULL;
1077 spin_unlock(&group_leader->sighand->siglock);
1078
1079 err = task_pgrp_vnr(group_leader);
1080out:
1081 write_unlock_irq(&tasklist_lock);
1082 return err;
1083}
1084
1085/*
1086 * Supplementary group IDs
1087 */
1088
1089/* init to 2 - one for init_task, one to ensure it is never freed */
1090struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1091
1092struct group_info *groups_alloc(int gidsetsize)
1093{
1094 struct group_info *group_info;
1095 int nblocks;
1096 int i;
1097
1098 nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1099 /* Make sure we always allocate at least one indirect block pointer */
1100 nblocks = nblocks ? : 1;
1101 group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1102 if (!group_info)
1103 return NULL;
1104 group_info->ngroups = gidsetsize;
1105 group_info->nblocks = nblocks;
1106 atomic_set(&group_info->usage, 1);
1107
1108 if (gidsetsize <= NGROUPS_SMALL)
1109 group_info->blocks[0] = group_info->small_block;
1110 else {
1111 for (i = 0; i < nblocks; i++) {
1112 gid_t *b;
1113 b = (void *)__get_free_page(GFP_USER);
1114 if (!b)
1115 goto out_undo_partial_alloc;
1116 group_info->blocks[i] = b;
1117 }
1118 }
1119 return group_info;
1120
1121out_undo_partial_alloc:
1122 while (--i >= 0) {
1123 free_page((unsigned long)group_info->blocks[i]);
1124 }
1125 kfree(group_info);
1126 return NULL;
1127}
1128
1129EXPORT_SYMBOL(groups_alloc);
1130
1131void groups_free(struct group_info *group_info)
1132{
1133 if (group_info->blocks[0] != group_info->small_block) {
1134 int i;
1135 for (i = 0; i < group_info->nblocks; i++)
1136 free_page((unsigned long)group_info->blocks[i]);
1137 }
1138 kfree(group_info);
1139}
1140
1141EXPORT_SYMBOL(groups_free);
1142
1143/* export the group_info to a user-space array */
1144static int groups_to_user(gid_t __user *grouplist,
1145 struct group_info *group_info)
1146{
1147 int i;
1148 int count = group_info->ngroups;
1149
1150 for (i = 0; i < group_info->nblocks; i++) {
1151 int cp_count = min(NGROUPS_PER_BLOCK, count);
1152 int off = i * NGROUPS_PER_BLOCK;
1153 int len = cp_count * sizeof(*grouplist);
1154
1155 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1156 return -EFAULT;
1157
1158 count -= cp_count;
1159 }
1160 return 0;
1161}
1162
1163/* fill a group_info from a user-space array - it must be allocated already */
1164static int groups_from_user(struct group_info *group_info,
1165 gid_t __user *grouplist)
1166{
1167 int i;
1168 int count = group_info->ngroups;
1169
1170 for (i = 0; i < group_info->nblocks; i++) {
1171 int cp_count = min(NGROUPS_PER_BLOCK, count);
1172 int off = i * NGROUPS_PER_BLOCK;
1173 int len = cp_count * sizeof(*grouplist);
1174
1175 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1176 return -EFAULT;
1177
1178 count -= cp_count;
1179 }
1180 return 0;
1181}
1182
1183/* a simple Shell sort */
1184static void groups_sort(struct group_info *group_info)
1185{
1186 int base, max, stride;
1187 int gidsetsize = group_info->ngroups;
1188
1189 for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1190 ; /* nothing */
1191 stride /= 3;
1192
1193 while (stride) {
1194 max = gidsetsize - stride;
1195 for (base = 0; base < max; base++) {
1196 int left = base;
1197 int right = left + stride;
1198 gid_t tmp = GROUP_AT(group_info, right);
1199
1200 while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1201 GROUP_AT(group_info, right) =
1202 GROUP_AT(group_info, left);
1203 right = left;
1204 left -= stride;
1205 }
1206 GROUP_AT(group_info, right) = tmp;
1207 }
1208 stride /= 3;
1209 }
1210}
1211
1212/* a simple bsearch */
1213int groups_search(struct group_info *group_info, gid_t grp)
1214{
1215 unsigned int left, right;
1216
1217 if (!group_info)
1218 return 0;
1219
1220 left = 0;
1221 right = group_info->ngroups;
1222 while (left < right) {
1223 unsigned int mid = (left+right)/2;
1224 int cmp = grp - GROUP_AT(group_info, mid);
1225 if (cmp > 0)
1226 left = mid + 1;
1227 else if (cmp < 0)
1228 right = mid;
1229 else
1230 return 1;
1231 }
1232 return 0;
1233}
1234
1235/* validate and set current->group_info */
1236int set_current_groups(struct group_info *group_info)
1237{
1238 int retval;
1239 struct group_info *old_info;
1240
1241 retval = security_task_setgroups(group_info);
1242 if (retval)
1243 return retval;
1244
1245 groups_sort(group_info);
1246 get_group_info(group_info);
1247
1248 task_lock(current);
1249 old_info = current->group_info;
1250 current->group_info = group_info;
1251 task_unlock(current);
1252
1253 put_group_info(old_info);
1254
1255 return 0;
1256}
1257
1258EXPORT_SYMBOL(set_current_groups);
1259
1260asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1261{
1262 int i = 0;
1263
1264 /*
1265 * SMP: Nobody else can change our grouplist. Thus we are
1266 * safe.
1267 */
1268
1269 if (gidsetsize < 0)
1270 return -EINVAL;
1271
1272 /* no need to grab task_lock here; it cannot change */
1273 i = current->group_info->ngroups;
1274 if (gidsetsize) {
1275 if (i > gidsetsize) {
1276 i = -EINVAL;
1277 goto out;
1278 }
1279 if (groups_to_user(grouplist, current->group_info)) {
1280 i = -EFAULT;
1281 goto out;
1282 }
1283 }
1284out:
1285 return i;
1286}
1287
1288/*
1289 * SMP: Our groups are copy-on-write. We can set them safely
1290 * without another task interfering.
1291 */
1292
1293asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1294{
1295 struct group_info *group_info;
1296 int retval;
1297
1298 if (!capable(CAP_SETGID))
1299 return -EPERM;
1300 if ((unsigned)gidsetsize > NGROUPS_MAX)
1301 return -EINVAL;
1302
1303 group_info = groups_alloc(gidsetsize);
1304 if (!group_info)
1305 return -ENOMEM;
1306 retval = groups_from_user(group_info, grouplist);
1307 if (retval) {
1308 put_group_info(group_info);
1309 return retval;
1310 }
1311
1312 retval = set_current_groups(group_info);
1313 put_group_info(group_info);
1314
1315 return retval;
1316}
1317
1318/*
1319 * Check whether we're fsgid/egid or in the supplemental group..
1320 */
1321int in_group_p(gid_t grp)
1322{
1323 int retval = 1;
1324 if (grp != current->fsgid)
1325 retval = groups_search(current->group_info, grp);
1326 return retval;
1327}
1328
1329EXPORT_SYMBOL(in_group_p);
1330
1331int in_egroup_p(gid_t grp)
1332{
1333 int retval = 1;
1334 if (grp != current->egid)
1335 retval = groups_search(current->group_info, grp);
1336 return retval;
1337}
1338
1339EXPORT_SYMBOL(in_egroup_p);
1340
1341DECLARE_RWSEM(uts_sem);
1342
1343EXPORT_SYMBOL(uts_sem);
1344
1345asmlinkage long sys_newuname(struct new_utsname __user * name)
1346{
1347 int errno = 0;
1348
1349 down_read(&uts_sem);
1350 if (copy_to_user(name, utsname(), sizeof *name))
1351 errno = -EFAULT;
1352 up_read(&uts_sem);
1353 return errno;
1354}
1355
1356asmlinkage long sys_sethostname(char __user *name, int len)
1357{
1358 int errno;
1359 char tmp[__NEW_UTS_LEN];
1360
1361 if (!capable(CAP_SYS_ADMIN))
1362 return -EPERM;
1363 if (len < 0 || len > __NEW_UTS_LEN)
1364 return -EINVAL;
1365 down_write(&uts_sem);
1366 errno = -EFAULT;
1367 if (!copy_from_user(tmp, name, len)) {
1368 memcpy(utsname()->nodename, tmp, len);
1369 utsname()->nodename[len] = 0;
1370 errno = 0;
1371 }
1372 up_write(&uts_sem);
1373 return errno;
1374}
1375
1376#ifdef __ARCH_WANT_SYS_GETHOSTNAME
1377
1378asmlinkage long sys_gethostname(char __user *name, int len)
1379{
1380 int i, errno;
1381
1382 if (len < 0)
1383 return -EINVAL;
1384 down_read(&uts_sem);
1385 i = 1 + strlen(utsname()->nodename);
1386 if (i > len)
1387 i = len;
1388 errno = 0;
1389 if (copy_to_user(name, utsname()->nodename, i))
1390 errno = -EFAULT;
1391 up_read(&uts_sem);
1392 return errno;
1393}
1394
1395#endif
1396
1397/*
1398 * Only setdomainname; getdomainname can be implemented by calling
1399 * uname()
1400 */
1401asmlinkage long sys_setdomainname(char __user *name, int len)
1402{
1403 int errno;
1404 char tmp[__NEW_UTS_LEN];
1405
1406 if (!capable(CAP_SYS_ADMIN))
1407 return -EPERM;
1408 if (len < 0 || len > __NEW_UTS_LEN)
1409 return -EINVAL;
1410
1411 down_write(&uts_sem);
1412 errno = -EFAULT;
1413 if (!copy_from_user(tmp, name, len)) {
1414 memcpy(utsname()->domainname, tmp, len);
1415 utsname()->domainname[len] = 0;
1416 errno = 0;
1417 }
1418 up_write(&uts_sem);
1419 return errno;
1420}
1421
1422asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1423{
1424 if (resource >= RLIM_NLIMITS)
1425 return -EINVAL;
1426 else {
1427 struct rlimit value;
1428 task_lock(current->group_leader);
1429 value = current->signal->rlim[resource];
1430 task_unlock(current->group_leader);
1431 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1432 }
1433}
1434
1435#ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1436
1437/*
1438 * Back compatibility for getrlimit. Needed for some apps.
1439 */
1440
1441asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1442{
1443 struct rlimit x;
1444 if (resource >= RLIM_NLIMITS)
1445 return -EINVAL;
1446
1447 task_lock(current->group_leader);
1448 x = current->signal->rlim[resource];
1449 task_unlock(current->group_leader);
1450 if (x.rlim_cur > 0x7FFFFFFF)
1451 x.rlim_cur = 0x7FFFFFFF;
1452 if (x.rlim_max > 0x7FFFFFFF)
1453 x.rlim_max = 0x7FFFFFFF;
1454 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1455}
1456
1457#endif
1458
1459asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1460{
1461 struct rlimit new_rlim, *old_rlim;
1462 unsigned long it_prof_secs;
1463 int retval;
1464
1465 if (resource >= RLIM_NLIMITS)
1466 return -EINVAL;
1467 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1468 return -EFAULT;
1469 if (new_rlim.rlim_cur > new_rlim.rlim_max)
1470 return -EINVAL;
1471 old_rlim = current->signal->rlim + resource;
1472 if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1473 !capable(CAP_SYS_RESOURCE))
1474 return -EPERM;
1475 if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1476 return -EPERM;
1477
1478 retval = security_task_setrlimit(resource, &new_rlim);
1479 if (retval)
1480 return retval;
1481
1482 if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1483 /*
1484 * The caller is asking for an immediate RLIMIT_CPU
1485 * expiry. But we use the zero value to mean "it was
1486 * never set". So let's cheat and make it one second
1487 * instead
1488 */
1489 new_rlim.rlim_cur = 1;
1490 }
1491
1492 task_lock(current->group_leader);
1493 *old_rlim = new_rlim;
1494 task_unlock(current->group_leader);
1495
1496 if (resource != RLIMIT_CPU)
1497 goto out;
1498
1499 /*
1500 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1501 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1502 * very long-standing error, and fixing it now risks breakage of
1503 * applications, so we live with it
1504 */
1505 if (new_rlim.rlim_cur == RLIM_INFINITY)
1506 goto out;
1507
1508 it_prof_secs = cputime_to_secs(current->signal->it_prof_expires);
1509 if (it_prof_secs == 0 || new_rlim.rlim_cur <= it_prof_secs) {
1510 unsigned long rlim_cur = new_rlim.rlim_cur;
1511 cputime_t cputime;
1512
1513 cputime = secs_to_cputime(rlim_cur);
1514 read_lock(&tasklist_lock);
1515 spin_lock_irq(&current->sighand->siglock);
1516 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
1517 spin_unlock_irq(&current->sighand->siglock);
1518 read_unlock(&tasklist_lock);
1519 }
1520out:
1521 return 0;
1522}
1523
1524/*
1525 * It would make sense to put struct rusage in the task_struct,
1526 * except that would make the task_struct be *really big*. After
1527 * task_struct gets moved into malloc'ed memory, it would
1528 * make sense to do this. It will make moving the rest of the information
1529 * a lot simpler! (Which we're not doing right now because we're not
1530 * measuring them yet).
1531 *
1532 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1533 * races with threads incrementing their own counters. But since word
1534 * reads are atomic, we either get new values or old values and we don't
1535 * care which for the sums. We always take the siglock to protect reading
1536 * the c* fields from p->signal from races with exit.c updating those
1537 * fields when reaping, so a sample either gets all the additions of a
1538 * given child after it's reaped, or none so this sample is before reaping.
1539 *
1540 * Locking:
1541 * We need to take the siglock for CHILDEREN, SELF and BOTH
1542 * for the cases current multithreaded, non-current single threaded
1543 * non-current multithreaded. Thread traversal is now safe with
1544 * the siglock held.
1545 * Strictly speaking, we donot need to take the siglock if we are current and
1546 * single threaded, as no one else can take our signal_struct away, no one
1547 * else can reap the children to update signal->c* counters, and no one else
1548 * can race with the signal-> fields. If we do not take any lock, the
1549 * signal-> fields could be read out of order while another thread was just
1550 * exiting. So we should place a read memory barrier when we avoid the lock.
1551 * On the writer side, write memory barrier is implied in __exit_signal
1552 * as __exit_signal releases the siglock spinlock after updating the signal->
1553 * fields. But we don't do this yet to keep things simple.
1554 *
1555 */
1556
1557static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1558{
1559 struct task_struct *t;
1560 unsigned long flags;
1561 cputime_t utime, stime;
1562
1563 memset((char *) r, 0, sizeof *r);
1564 utime = stime = cputime_zero;
1565
1566 rcu_read_lock();
1567 if (!lock_task_sighand(p, &flags)) {
1568 rcu_read_unlock();
1569 return;
1570 }
1571
1572 switch (who) {
1573 case RUSAGE_BOTH:
1574 case RUSAGE_CHILDREN:
1575 utime = p->signal->cutime;
1576 stime = p->signal->cstime;
1577 r->ru_nvcsw = p->signal->cnvcsw;
1578 r->ru_nivcsw = p->signal->cnivcsw;
1579 r->ru_minflt = p->signal->cmin_flt;
1580 r->ru_majflt = p->signal->cmaj_flt;
1581 r->ru_inblock = p->signal->cinblock;
1582 r->ru_oublock = p->signal->coublock;
1583
1584 if (who == RUSAGE_CHILDREN)
1585 break;
1586
1587 case RUSAGE_SELF:
1588 utime = cputime_add(utime, p->signal->utime);
1589 stime = cputime_add(stime, p->signal->stime);
1590 r->ru_nvcsw += p->signal->nvcsw;
1591 r->ru_nivcsw += p->signal->nivcsw;
1592 r->ru_minflt += p->signal->min_flt;
1593 r->ru_majflt += p->signal->maj_flt;
1594 r->ru_inblock += p->signal->inblock;
1595 r->ru_oublock += p->signal->oublock;
1596 t = p;
1597 do {
1598 utime = cputime_add(utime, t->utime);
1599 stime = cputime_add(stime, t->stime);
1600 r->ru_nvcsw += t->nvcsw;
1601 r->ru_nivcsw += t->nivcsw;
1602 r->ru_minflt += t->min_flt;
1603 r->ru_majflt += t->maj_flt;
1604 r->ru_inblock += task_io_get_inblock(t);
1605 r->ru_oublock += task_io_get_oublock(t);
1606 t = next_thread(t);
1607 } while (t != p);
1608 break;
1609
1610 default:
1611 BUG();
1612 }
1613
1614 unlock_task_sighand(p, &flags);
1615 rcu_read_unlock();
1616
1617 cputime_to_timeval(utime, &r->ru_utime);
1618 cputime_to_timeval(stime, &r->ru_stime);
1619}
1620
1621int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1622{
1623 struct rusage r;
1624 k_getrusage(p, who, &r);
1625 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1626}
1627
1628asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1629{
1630 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1631 return -EINVAL;
1632 return getrusage(current, who, ru);
1633}
1634
1635asmlinkage long sys_umask(int mask)
1636{
1637 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1638 return mask;
1639}
1640
1641asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1642 unsigned long arg4, unsigned long arg5)
1643{
1644 long error;
1645
1646 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1647 if (error)
1648 return error;
1649
1650 switch (option) {
1651 case PR_SET_PDEATHSIG:
1652 if (!valid_signal(arg2)) {
1653 error = -EINVAL;
1654 break;
1655 }
1656 current->pdeath_signal = arg2;
1657 break;
1658 case PR_GET_PDEATHSIG:
1659 error = put_user(current->pdeath_signal, (int __user *)arg2);
1660 break;
1661 case PR_GET_DUMPABLE:
1662 error = get_dumpable(current->mm);
1663 break;
1664 case PR_SET_DUMPABLE:
1665 if (arg2 < 0 || arg2 > 1) {
1666 error = -EINVAL;
1667 break;
1668 }
1669 set_dumpable(current->mm, arg2);
1670 break;
1671
1672 case PR_SET_UNALIGN:
1673 error = SET_UNALIGN_CTL(current, arg2);
1674 break;
1675 case PR_GET_UNALIGN:
1676 error = GET_UNALIGN_CTL(current, arg2);
1677 break;
1678 case PR_SET_FPEMU:
1679 error = SET_FPEMU_CTL(current, arg2);
1680 break;
1681 case PR_GET_FPEMU:
1682 error = GET_FPEMU_CTL(current, arg2);
1683 break;
1684 case PR_SET_FPEXC:
1685 error = SET_FPEXC_CTL(current, arg2);
1686 break;
1687 case PR_GET_FPEXC:
1688 error = GET_FPEXC_CTL(current, arg2);
1689 break;
1690 case PR_GET_TIMING:
1691 error = PR_TIMING_STATISTICAL;
1692 break;
1693 case PR_SET_TIMING:
1694 if (arg2 == PR_TIMING_STATISTICAL)
1695 error = 0;
1696 else
1697 error = -EINVAL;
1698 break;
1699
1700 case PR_GET_KEEPCAPS:
1701 if (current->keep_capabilities)
1702 error = 1;
1703 break;
1704 case PR_SET_KEEPCAPS:
1705 if (arg2 != 0 && arg2 != 1) {
1706 error = -EINVAL;
1707 break;
1708 }
1709 current->keep_capabilities = arg2;
1710 break;
1711 case PR_SET_NAME: {
1712 struct task_struct *me = current;
1713 unsigned char ncomm[sizeof(me->comm)];
1714
1715 ncomm[sizeof(me->comm)-1] = 0;
1716 if (strncpy_from_user(ncomm, (char __user *)arg2,
1717 sizeof(me->comm)-1) < 0)
1718 return -EFAULT;
1719 set_task_comm(me, ncomm);
1720 return 0;
1721 }
1722 case PR_GET_NAME: {
1723 struct task_struct *me = current;
1724 unsigned char tcomm[sizeof(me->comm)];
1725
1726 get_task_comm(tcomm, me);
1727 if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1728 return -EFAULT;
1729 return 0;
1730 }
1731 case PR_GET_ENDIAN:
1732 error = GET_ENDIAN(current, arg2);
1733 break;
1734 case PR_SET_ENDIAN:
1735 error = SET_ENDIAN(current, arg2);
1736 break;
1737
1738 case PR_GET_SECCOMP:
1739 error = prctl_get_seccomp();
1740 break;
1741 case PR_SET_SECCOMP:
1742 error = prctl_set_seccomp(arg2);
1743 break;
1744
1745 default:
1746 error = -EINVAL;
1747 break;
1748 }
1749 return error;
1750}
1751
1752asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
1753 struct getcpu_cache __user *unused)
1754{
1755 int err = 0;
1756 int cpu = raw_smp_processor_id();
1757 if (cpup)
1758 err |= put_user(cpu, cpup);
1759 if (nodep)
1760 err |= put_user(cpu_to_node(cpu), nodep);
1761 return err ? -EFAULT : 0;
1762}
1763
1764char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1765
1766static void argv_cleanup(char **argv, char **envp)
1767{
1768 argv_free(argv);
1769}
1770
1771/**
1772 * orderly_poweroff - Trigger an orderly system poweroff
1773 * @force: force poweroff if command execution fails
1774 *
1775 * This may be called from any context to trigger a system shutdown.
1776 * If the orderly shutdown fails, it will force an immediate shutdown.
1777 */
1778int orderly_poweroff(bool force)
1779{
1780 int argc;
1781 char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1782 static char *envp[] = {
1783 "HOME=/",
1784 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1785 NULL
1786 };
1787 int ret = -ENOMEM;
1788 struct subprocess_info *info;
1789
1790 if (argv == NULL) {
1791 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1792 __func__, poweroff_cmd);
1793 goto out;
1794 }
1795
1796 info = call_usermodehelper_setup(argv[0], argv, envp);
1797 if (info == NULL) {
1798 argv_free(argv);
1799 goto out;
1800 }
1801
1802 call_usermodehelper_setcleanup(info, argv_cleanup);
1803
1804 ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1805
1806 out:
1807 if (ret && force) {
1808 printk(KERN_WARNING "Failed to start orderly shutdown: "
1809 "forcing the issue\n");
1810
1811 /* I guess this should try to kick off some daemon to
1812 sync and poweroff asap. Or not even bother syncing
1813 if we're doing an emergency shutdown? */
1814 emergency_sync();
1815 kernel_power_off();
1816 }
1817
1818 return ret;
1819}
1820EXPORT_SYMBOL_GPL(orderly_poweroff);