PM / Core: Introduce struct syscore_ops for core subsystems PM
[linux-2.6.git] / kernel / sys.c
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/notifier.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
15 #include <linux/fs.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/gfp.h>
40 #include <linux/syscore_ops.h>
41
42 #include <linux/compat.h>
43 #include <linux/syscalls.h>
44 #include <linux/kprobes.h>
45 #include <linux/user_namespace.h>
46
47 #include <linux/kmsg_dump.h>
48
49 #include <asm/uaccess.h>
50 #include <asm/io.h>
51 #include <asm/unistd.h>
52
53 #ifndef SET_UNALIGN_CTL
54 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
55 #endif
56 #ifndef GET_UNALIGN_CTL
57 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
58 #endif
59 #ifndef SET_FPEMU_CTL
60 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
61 #endif
62 #ifndef GET_FPEMU_CTL
63 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
64 #endif
65 #ifndef SET_FPEXC_CTL
66 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
67 #endif
68 #ifndef GET_FPEXC_CTL
69 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
70 #endif
71 #ifndef GET_ENDIAN
72 # define GET_ENDIAN(a,b)        (-EINVAL)
73 #endif
74 #ifndef SET_ENDIAN
75 # define SET_ENDIAN(a,b)        (-EINVAL)
76 #endif
77 #ifndef GET_TSC_CTL
78 # define GET_TSC_CTL(a)         (-EINVAL)
79 #endif
80 #ifndef SET_TSC_CTL
81 # define SET_TSC_CTL(a)         (-EINVAL)
82 #endif
83
84 /*
85  * this is where the system-wide overflow UID and GID are defined, for
86  * architectures that now have 32-bit UID/GID but didn't in the past
87  */
88
89 int overflowuid = DEFAULT_OVERFLOWUID;
90 int overflowgid = DEFAULT_OVERFLOWGID;
91
92 #ifdef CONFIG_UID16
93 EXPORT_SYMBOL(overflowuid);
94 EXPORT_SYMBOL(overflowgid);
95 #endif
96
97 /*
98  * the same as above, but for filesystems which can only store a 16-bit
99  * UID and GID. as such, this is needed on all architectures
100  */
101
102 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
103 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
104
105 EXPORT_SYMBOL(fs_overflowuid);
106 EXPORT_SYMBOL(fs_overflowgid);
107
108 /*
109  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
110  */
111
112 int C_A_D = 1;
113 struct pid *cad_pid;
114 EXPORT_SYMBOL(cad_pid);
115
116 /*
117  * If set, this is used for preparing the system to power off.
118  */
119
120 void (*pm_power_off_prepare)(void);
121
122 /*
123  * set the priority of a task
124  * - the caller must hold the RCU read lock
125  */
126 static int set_one_prio(struct task_struct *p, int niceval, int error)
127 {
128         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
129         int no_nice;
130
131         if (pcred->uid  != cred->euid &&
132             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
133                 error = -EPERM;
134                 goto out;
135         }
136         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
137                 error = -EACCES;
138                 goto out;
139         }
140         no_nice = security_task_setnice(p, niceval);
141         if (no_nice) {
142                 error = no_nice;
143                 goto out;
144         }
145         if (error == -ESRCH)
146                 error = 0;
147         set_user_nice(p, niceval);
148 out:
149         return error;
150 }
151
152 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
153 {
154         struct task_struct *g, *p;
155         struct user_struct *user;
156         const struct cred *cred = current_cred();
157         int error = -EINVAL;
158         struct pid *pgrp;
159
160         if (which > PRIO_USER || which < PRIO_PROCESS)
161                 goto out;
162
163         /* normalize: avoid signed division (rounding problems) */
164         error = -ESRCH;
165         if (niceval < -20)
166                 niceval = -20;
167         if (niceval > 19)
168                 niceval = 19;
169
170         rcu_read_lock();
171         read_lock(&tasklist_lock);
172         switch (which) {
173                 case PRIO_PROCESS:
174                         if (who)
175                                 p = find_task_by_vpid(who);
176                         else
177                                 p = current;
178                         if (p)
179                                 error = set_one_prio(p, niceval, error);
180                         break;
181                 case PRIO_PGRP:
182                         if (who)
183                                 pgrp = find_vpid(who);
184                         else
185                                 pgrp = task_pgrp(current);
186                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
187                                 error = set_one_prio(p, niceval, error);
188                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
189                         break;
190                 case PRIO_USER:
191                         user = (struct user_struct *) cred->user;
192                         if (!who)
193                                 who = cred->uid;
194                         else if ((who != cred->uid) &&
195                                  !(user = find_user(who)))
196                                 goto out_unlock;        /* No processes for this user */
197
198                         do_each_thread(g, p) {
199                                 if (__task_cred(p)->uid == who)
200                                         error = set_one_prio(p, niceval, error);
201                         } while_each_thread(g, p);
202                         if (who != cred->uid)
203                                 free_uid(user);         /* For find_user() */
204                         break;
205         }
206 out_unlock:
207         read_unlock(&tasklist_lock);
208         rcu_read_unlock();
209 out:
210         return error;
211 }
212
213 /*
214  * Ugh. To avoid negative return values, "getpriority()" will
215  * not return the normal nice-value, but a negated value that
216  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
217  * to stay compatible.
218  */
219 SYSCALL_DEFINE2(getpriority, int, which, int, who)
220 {
221         struct task_struct *g, *p;
222         struct user_struct *user;
223         const struct cred *cred = current_cred();
224         long niceval, retval = -ESRCH;
225         struct pid *pgrp;
226
227         if (which > PRIO_USER || which < PRIO_PROCESS)
228                 return -EINVAL;
229
230         rcu_read_lock();
231         read_lock(&tasklist_lock);
232         switch (which) {
233                 case PRIO_PROCESS:
234                         if (who)
235                                 p = find_task_by_vpid(who);
236                         else
237                                 p = current;
238                         if (p) {
239                                 niceval = 20 - task_nice(p);
240                                 if (niceval > retval)
241                                         retval = niceval;
242                         }
243                         break;
244                 case PRIO_PGRP:
245                         if (who)
246                                 pgrp = find_vpid(who);
247                         else
248                                 pgrp = task_pgrp(current);
249                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
250                                 niceval = 20 - task_nice(p);
251                                 if (niceval > retval)
252                                         retval = niceval;
253                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
254                         break;
255                 case PRIO_USER:
256                         user = (struct user_struct *) cred->user;
257                         if (!who)
258                                 who = cred->uid;
259                         else if ((who != cred->uid) &&
260                                  !(user = find_user(who)))
261                                 goto out_unlock;        /* No processes for this user */
262
263                         do_each_thread(g, p) {
264                                 if (__task_cred(p)->uid == who) {
265                                         niceval = 20 - task_nice(p);
266                                         if (niceval > retval)
267                                                 retval = niceval;
268                                 }
269                         } while_each_thread(g, p);
270                         if (who != cred->uid)
271                                 free_uid(user);         /* for find_user() */
272                         break;
273         }
274 out_unlock:
275         read_unlock(&tasklist_lock);
276         rcu_read_unlock();
277
278         return retval;
279 }
280
281 /**
282  *      emergency_restart - reboot the system
283  *
284  *      Without shutting down any hardware or taking any locks
285  *      reboot the system.  This is called when we know we are in
286  *      trouble so this is our best effort to reboot.  This is
287  *      safe to call in interrupt context.
288  */
289 void emergency_restart(void)
290 {
291         kmsg_dump(KMSG_DUMP_EMERG);
292         machine_emergency_restart();
293 }
294 EXPORT_SYMBOL_GPL(emergency_restart);
295
296 void kernel_restart_prepare(char *cmd)
297 {
298         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
299         system_state = SYSTEM_RESTART;
300         device_shutdown();
301         sysdev_shutdown();
302         syscore_shutdown();
303 }
304
305 /**
306  *      kernel_restart - reboot the system
307  *      @cmd: pointer to buffer containing command to execute for restart
308  *              or %NULL
309  *
310  *      Shutdown everything and perform a clean reboot.
311  *      This is not safe to call in interrupt context.
312  */
313 void kernel_restart(char *cmd)
314 {
315         kernel_restart_prepare(cmd);
316         if (!cmd)
317                 printk(KERN_EMERG "Restarting system.\n");
318         else
319                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
320         kmsg_dump(KMSG_DUMP_RESTART);
321         machine_restart(cmd);
322 }
323 EXPORT_SYMBOL_GPL(kernel_restart);
324
325 static void kernel_shutdown_prepare(enum system_states state)
326 {
327         blocking_notifier_call_chain(&reboot_notifier_list,
328                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
329         system_state = state;
330         device_shutdown();
331 }
332 /**
333  *      kernel_halt - halt the system
334  *
335  *      Shutdown everything and perform a clean system halt.
336  */
337 void kernel_halt(void)
338 {
339         kernel_shutdown_prepare(SYSTEM_HALT);
340         sysdev_shutdown();
341         syscore_shutdown();
342         printk(KERN_EMERG "System halted.\n");
343         kmsg_dump(KMSG_DUMP_HALT);
344         machine_halt();
345 }
346
347 EXPORT_SYMBOL_GPL(kernel_halt);
348
349 /**
350  *      kernel_power_off - power_off the system
351  *
352  *      Shutdown everything and perform a clean system power_off.
353  */
354 void kernel_power_off(void)
355 {
356         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
357         if (pm_power_off_prepare)
358                 pm_power_off_prepare();
359         disable_nonboot_cpus();
360         sysdev_shutdown();
361         syscore_shutdown();
362         printk(KERN_EMERG "Power down.\n");
363         kmsg_dump(KMSG_DUMP_POWEROFF);
364         machine_power_off();
365 }
366 EXPORT_SYMBOL_GPL(kernel_power_off);
367
368 static DEFINE_MUTEX(reboot_mutex);
369
370 /*
371  * Reboot system call: for obvious reasons only root may call it,
372  * and even root needs to set up some magic numbers in the registers
373  * so that some mistake won't make this reboot the whole machine.
374  * You can also set the meaning of the ctrl-alt-del-key here.
375  *
376  * reboot doesn't sync: do that yourself before calling this.
377  */
378 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
379                 void __user *, arg)
380 {
381         char buffer[256];
382         int ret = 0;
383
384         /* We only trust the superuser with rebooting the system. */
385         if (!capable(CAP_SYS_BOOT))
386                 return -EPERM;
387
388         /* For safety, we require "magic" arguments. */
389         if (magic1 != LINUX_REBOOT_MAGIC1 ||
390             (magic2 != LINUX_REBOOT_MAGIC2 &&
391                         magic2 != LINUX_REBOOT_MAGIC2A &&
392                         magic2 != LINUX_REBOOT_MAGIC2B &&
393                         magic2 != LINUX_REBOOT_MAGIC2C))
394                 return -EINVAL;
395
396         /* Instead of trying to make the power_off code look like
397          * halt when pm_power_off is not set do it the easy way.
398          */
399         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
400                 cmd = LINUX_REBOOT_CMD_HALT;
401
402         mutex_lock(&reboot_mutex);
403         switch (cmd) {
404         case LINUX_REBOOT_CMD_RESTART:
405                 kernel_restart(NULL);
406                 break;
407
408         case LINUX_REBOOT_CMD_CAD_ON:
409                 C_A_D = 1;
410                 break;
411
412         case LINUX_REBOOT_CMD_CAD_OFF:
413                 C_A_D = 0;
414                 break;
415
416         case LINUX_REBOOT_CMD_HALT:
417                 kernel_halt();
418                 do_exit(0);
419                 panic("cannot halt");
420
421         case LINUX_REBOOT_CMD_POWER_OFF:
422                 kernel_power_off();
423                 do_exit(0);
424                 break;
425
426         case LINUX_REBOOT_CMD_RESTART2:
427                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
428                         ret = -EFAULT;
429                         break;
430                 }
431                 buffer[sizeof(buffer) - 1] = '\0';
432
433                 kernel_restart(buffer);
434                 break;
435
436 #ifdef CONFIG_KEXEC
437         case LINUX_REBOOT_CMD_KEXEC:
438                 ret = kernel_kexec();
439                 break;
440 #endif
441
442 #ifdef CONFIG_HIBERNATION
443         case LINUX_REBOOT_CMD_SW_SUSPEND:
444                 ret = hibernate();
445                 break;
446 #endif
447
448         default:
449                 ret = -EINVAL;
450                 break;
451         }
452         mutex_unlock(&reboot_mutex);
453         return ret;
454 }
455
456 static void deferred_cad(struct work_struct *dummy)
457 {
458         kernel_restart(NULL);
459 }
460
461 /*
462  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
463  * As it's called within an interrupt, it may NOT sync: the only choice
464  * is whether to reboot at once, or just ignore the ctrl-alt-del.
465  */
466 void ctrl_alt_del(void)
467 {
468         static DECLARE_WORK(cad_work, deferred_cad);
469
470         if (C_A_D)
471                 schedule_work(&cad_work);
472         else
473                 kill_cad_pid(SIGINT, 1);
474 }
475         
476 /*
477  * Unprivileged users may change the real gid to the effective gid
478  * or vice versa.  (BSD-style)
479  *
480  * If you set the real gid at all, or set the effective gid to a value not
481  * equal to the real gid, then the saved gid is set to the new effective gid.
482  *
483  * This makes it possible for a setgid program to completely drop its
484  * privileges, which is often a useful assertion to make when you are doing
485  * a security audit over a program.
486  *
487  * The general idea is that a program which uses just setregid() will be
488  * 100% compatible with BSD.  A program which uses just setgid() will be
489  * 100% compatible with POSIX with saved IDs. 
490  *
491  * SMP: There are not races, the GIDs are checked only by filesystem
492  *      operations (as far as semantic preservation is concerned).
493  */
494 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
495 {
496         const struct cred *old;
497         struct cred *new;
498         int retval;
499
500         new = prepare_creds();
501         if (!new)
502                 return -ENOMEM;
503         old = current_cred();
504
505         retval = -EPERM;
506         if (rgid != (gid_t) -1) {
507                 if (old->gid == rgid ||
508                     old->egid == rgid ||
509                     capable(CAP_SETGID))
510                         new->gid = rgid;
511                 else
512                         goto error;
513         }
514         if (egid != (gid_t) -1) {
515                 if (old->gid == egid ||
516                     old->egid == egid ||
517                     old->sgid == egid ||
518                     capable(CAP_SETGID))
519                         new->egid = egid;
520                 else
521                         goto error;
522         }
523
524         if (rgid != (gid_t) -1 ||
525             (egid != (gid_t) -1 && egid != old->gid))
526                 new->sgid = new->egid;
527         new->fsgid = new->egid;
528
529         return commit_creds(new);
530
531 error:
532         abort_creds(new);
533         return retval;
534 }
535
536 /*
537  * setgid() is implemented like SysV w/ SAVED_IDS 
538  *
539  * SMP: Same implicit races as above.
540  */
541 SYSCALL_DEFINE1(setgid, gid_t, gid)
542 {
543         const struct cred *old;
544         struct cred *new;
545         int retval;
546
547         new = prepare_creds();
548         if (!new)
549                 return -ENOMEM;
550         old = current_cred();
551
552         retval = -EPERM;
553         if (capable(CAP_SETGID))
554                 new->gid = new->egid = new->sgid = new->fsgid = gid;
555         else if (gid == old->gid || gid == old->sgid)
556                 new->egid = new->fsgid = gid;
557         else
558                 goto error;
559
560         return commit_creds(new);
561
562 error:
563         abort_creds(new);
564         return retval;
565 }
566
567 /*
568  * change the user struct in a credentials set to match the new UID
569  */
570 static int set_user(struct cred *new)
571 {
572         struct user_struct *new_user;
573
574         new_user = alloc_uid(current_user_ns(), new->uid);
575         if (!new_user)
576                 return -EAGAIN;
577
578         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
579                         new_user != INIT_USER) {
580                 free_uid(new_user);
581                 return -EAGAIN;
582         }
583
584         free_uid(new->user);
585         new->user = new_user;
586         return 0;
587 }
588
589 /*
590  * Unprivileged users may change the real uid to the effective uid
591  * or vice versa.  (BSD-style)
592  *
593  * If you set the real uid at all, or set the effective uid to a value not
594  * equal to the real uid, then the saved uid is set to the new effective uid.
595  *
596  * This makes it possible for a setuid program to completely drop its
597  * privileges, which is often a useful assertion to make when you are doing
598  * a security audit over a program.
599  *
600  * The general idea is that a program which uses just setreuid() will be
601  * 100% compatible with BSD.  A program which uses just setuid() will be
602  * 100% compatible with POSIX with saved IDs. 
603  */
604 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
605 {
606         const struct cred *old;
607         struct cred *new;
608         int retval;
609
610         new = prepare_creds();
611         if (!new)
612                 return -ENOMEM;
613         old = current_cred();
614
615         retval = -EPERM;
616         if (ruid != (uid_t) -1) {
617                 new->uid = ruid;
618                 if (old->uid != ruid &&
619                     old->euid != ruid &&
620                     !capable(CAP_SETUID))
621                         goto error;
622         }
623
624         if (euid != (uid_t) -1) {
625                 new->euid = euid;
626                 if (old->uid != euid &&
627                     old->euid != euid &&
628                     old->suid != euid &&
629                     !capable(CAP_SETUID))
630                         goto error;
631         }
632
633         if (new->uid != old->uid) {
634                 retval = set_user(new);
635                 if (retval < 0)
636                         goto error;
637         }
638         if (ruid != (uid_t) -1 ||
639             (euid != (uid_t) -1 && euid != old->uid))
640                 new->suid = new->euid;
641         new->fsuid = new->euid;
642
643         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
644         if (retval < 0)
645                 goto error;
646
647         return commit_creds(new);
648
649 error:
650         abort_creds(new);
651         return retval;
652 }
653                 
654 /*
655  * setuid() is implemented like SysV with SAVED_IDS 
656  * 
657  * Note that SAVED_ID's is deficient in that a setuid root program
658  * like sendmail, for example, cannot set its uid to be a normal 
659  * user and then switch back, because if you're root, setuid() sets
660  * the saved uid too.  If you don't like this, blame the bright people
661  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
662  * will allow a root program to temporarily drop privileges and be able to
663  * regain them by swapping the real and effective uid.  
664  */
665 SYSCALL_DEFINE1(setuid, uid_t, uid)
666 {
667         const struct cred *old;
668         struct cred *new;
669         int retval;
670
671         new = prepare_creds();
672         if (!new)
673                 return -ENOMEM;
674         old = current_cred();
675
676         retval = -EPERM;
677         if (capable(CAP_SETUID)) {
678                 new->suid = new->uid = uid;
679                 if (uid != old->uid) {
680                         retval = set_user(new);
681                         if (retval < 0)
682                                 goto error;
683                 }
684         } else if (uid != old->uid && uid != new->suid) {
685                 goto error;
686         }
687
688         new->fsuid = new->euid = uid;
689
690         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
691         if (retval < 0)
692                 goto error;
693
694         return commit_creds(new);
695
696 error:
697         abort_creds(new);
698         return retval;
699 }
700
701
702 /*
703  * This function implements a generic ability to update ruid, euid,
704  * and suid.  This allows you to implement the 4.4 compatible seteuid().
705  */
706 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
707 {
708         const struct cred *old;
709         struct cred *new;
710         int retval;
711
712         new = prepare_creds();
713         if (!new)
714                 return -ENOMEM;
715
716         old = current_cred();
717
718         retval = -EPERM;
719         if (!capable(CAP_SETUID)) {
720                 if (ruid != (uid_t) -1 && ruid != old->uid &&
721                     ruid != old->euid  && ruid != old->suid)
722                         goto error;
723                 if (euid != (uid_t) -1 && euid != old->uid &&
724                     euid != old->euid  && euid != old->suid)
725                         goto error;
726                 if (suid != (uid_t) -1 && suid != old->uid &&
727                     suid != old->euid  && suid != old->suid)
728                         goto error;
729         }
730
731         if (ruid != (uid_t) -1) {
732                 new->uid = ruid;
733                 if (ruid != old->uid) {
734                         retval = set_user(new);
735                         if (retval < 0)
736                                 goto error;
737                 }
738         }
739         if (euid != (uid_t) -1)
740                 new->euid = euid;
741         if (suid != (uid_t) -1)
742                 new->suid = suid;
743         new->fsuid = new->euid;
744
745         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
746         if (retval < 0)
747                 goto error;
748
749         return commit_creds(new);
750
751 error:
752         abort_creds(new);
753         return retval;
754 }
755
756 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
757 {
758         const struct cred *cred = current_cred();
759         int retval;
760
761         if (!(retval   = put_user(cred->uid,  ruid)) &&
762             !(retval   = put_user(cred->euid, euid)))
763                 retval = put_user(cred->suid, suid);
764
765         return retval;
766 }
767
768 /*
769  * Same as above, but for rgid, egid, sgid.
770  */
771 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
772 {
773         const struct cred *old;
774         struct cred *new;
775         int retval;
776
777         new = prepare_creds();
778         if (!new)
779                 return -ENOMEM;
780         old = current_cred();
781
782         retval = -EPERM;
783         if (!capable(CAP_SETGID)) {
784                 if (rgid != (gid_t) -1 && rgid != old->gid &&
785                     rgid != old->egid  && rgid != old->sgid)
786                         goto error;
787                 if (egid != (gid_t) -1 && egid != old->gid &&
788                     egid != old->egid  && egid != old->sgid)
789                         goto error;
790                 if (sgid != (gid_t) -1 && sgid != old->gid &&
791                     sgid != old->egid  && sgid != old->sgid)
792                         goto error;
793         }
794
795         if (rgid != (gid_t) -1)
796                 new->gid = rgid;
797         if (egid != (gid_t) -1)
798                 new->egid = egid;
799         if (sgid != (gid_t) -1)
800                 new->sgid = sgid;
801         new->fsgid = new->egid;
802
803         return commit_creds(new);
804
805 error:
806         abort_creds(new);
807         return retval;
808 }
809
810 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
811 {
812         const struct cred *cred = current_cred();
813         int retval;
814
815         if (!(retval   = put_user(cred->gid,  rgid)) &&
816             !(retval   = put_user(cred->egid, egid)))
817                 retval = put_user(cred->sgid, sgid);
818
819         return retval;
820 }
821
822
823 /*
824  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
825  * is used for "access()" and for the NFS daemon (letting nfsd stay at
826  * whatever uid it wants to). It normally shadows "euid", except when
827  * explicitly set by setfsuid() or for access..
828  */
829 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
830 {
831         const struct cred *old;
832         struct cred *new;
833         uid_t old_fsuid;
834
835         new = prepare_creds();
836         if (!new)
837                 return current_fsuid();
838         old = current_cred();
839         old_fsuid = old->fsuid;
840
841         if (uid == old->uid  || uid == old->euid  ||
842             uid == old->suid || uid == old->fsuid ||
843             capable(CAP_SETUID)) {
844                 if (uid != old_fsuid) {
845                         new->fsuid = uid;
846                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
847                                 goto change_okay;
848                 }
849         }
850
851         abort_creds(new);
852         return old_fsuid;
853
854 change_okay:
855         commit_creds(new);
856         return old_fsuid;
857 }
858
859 /*
860  * Samma på svenska..
861  */
862 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
863 {
864         const struct cred *old;
865         struct cred *new;
866         gid_t old_fsgid;
867
868         new = prepare_creds();
869         if (!new)
870                 return current_fsgid();
871         old = current_cred();
872         old_fsgid = old->fsgid;
873
874         if (gid == old->gid  || gid == old->egid  ||
875             gid == old->sgid || gid == old->fsgid ||
876             capable(CAP_SETGID)) {
877                 if (gid != old_fsgid) {
878                         new->fsgid = gid;
879                         goto change_okay;
880                 }
881         }
882
883         abort_creds(new);
884         return old_fsgid;
885
886 change_okay:
887         commit_creds(new);
888         return old_fsgid;
889 }
890
891 void do_sys_times(struct tms *tms)
892 {
893         cputime_t tgutime, tgstime, cutime, cstime;
894
895         spin_lock_irq(&current->sighand->siglock);
896         thread_group_times(current, &tgutime, &tgstime);
897         cutime = current->signal->cutime;
898         cstime = current->signal->cstime;
899         spin_unlock_irq(&current->sighand->siglock);
900         tms->tms_utime = cputime_to_clock_t(tgutime);
901         tms->tms_stime = cputime_to_clock_t(tgstime);
902         tms->tms_cutime = cputime_to_clock_t(cutime);
903         tms->tms_cstime = cputime_to_clock_t(cstime);
904 }
905
906 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
907 {
908         if (tbuf) {
909                 struct tms tmp;
910
911                 do_sys_times(&tmp);
912                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
913                         return -EFAULT;
914         }
915         force_successful_syscall_return();
916         return (long) jiffies_64_to_clock_t(get_jiffies_64());
917 }
918
919 /*
920  * This needs some heavy checking ...
921  * I just haven't the stomach for it. I also don't fully
922  * understand sessions/pgrp etc. Let somebody who does explain it.
923  *
924  * OK, I think I have the protection semantics right.... this is really
925  * only important on a multi-user system anyway, to make sure one user
926  * can't send a signal to a process owned by another.  -TYT, 12/12/91
927  *
928  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
929  * LBT 04.03.94
930  */
931 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
932 {
933         struct task_struct *p;
934         struct task_struct *group_leader = current->group_leader;
935         struct pid *pgrp;
936         int err;
937
938         if (!pid)
939                 pid = task_pid_vnr(group_leader);
940         if (!pgid)
941                 pgid = pid;
942         if (pgid < 0)
943                 return -EINVAL;
944         rcu_read_lock();
945
946         /* From this point forward we keep holding onto the tasklist lock
947          * so that our parent does not change from under us. -DaveM
948          */
949         write_lock_irq(&tasklist_lock);
950
951         err = -ESRCH;
952         p = find_task_by_vpid(pid);
953         if (!p)
954                 goto out;
955
956         err = -EINVAL;
957         if (!thread_group_leader(p))
958                 goto out;
959
960         if (same_thread_group(p->real_parent, group_leader)) {
961                 err = -EPERM;
962                 if (task_session(p) != task_session(group_leader))
963                         goto out;
964                 err = -EACCES;
965                 if (p->did_exec)
966                         goto out;
967         } else {
968                 err = -ESRCH;
969                 if (p != group_leader)
970                         goto out;
971         }
972
973         err = -EPERM;
974         if (p->signal->leader)
975                 goto out;
976
977         pgrp = task_pid(p);
978         if (pgid != pid) {
979                 struct task_struct *g;
980
981                 pgrp = find_vpid(pgid);
982                 g = pid_task(pgrp, PIDTYPE_PGID);
983                 if (!g || task_session(g) != task_session(group_leader))
984                         goto out;
985         }
986
987         err = security_task_setpgid(p, pgid);
988         if (err)
989                 goto out;
990
991         if (task_pgrp(p) != pgrp)
992                 change_pid(p, PIDTYPE_PGID, pgrp);
993
994         err = 0;
995 out:
996         /* All paths lead to here, thus we are safe. -DaveM */
997         write_unlock_irq(&tasklist_lock);
998         rcu_read_unlock();
999         return err;
1000 }
1001
1002 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1003 {
1004         struct task_struct *p;
1005         struct pid *grp;
1006         int retval;
1007
1008         rcu_read_lock();
1009         if (!pid)
1010                 grp = task_pgrp(current);
1011         else {
1012                 retval = -ESRCH;
1013                 p = find_task_by_vpid(pid);
1014                 if (!p)
1015                         goto out;
1016                 grp = task_pgrp(p);
1017                 if (!grp)
1018                         goto out;
1019
1020                 retval = security_task_getpgid(p);
1021                 if (retval)
1022                         goto out;
1023         }
1024         retval = pid_vnr(grp);
1025 out:
1026         rcu_read_unlock();
1027         return retval;
1028 }
1029
1030 #ifdef __ARCH_WANT_SYS_GETPGRP
1031
1032 SYSCALL_DEFINE0(getpgrp)
1033 {
1034         return sys_getpgid(0);
1035 }
1036
1037 #endif
1038
1039 SYSCALL_DEFINE1(getsid, pid_t, pid)
1040 {
1041         struct task_struct *p;
1042         struct pid *sid;
1043         int retval;
1044
1045         rcu_read_lock();
1046         if (!pid)
1047                 sid = task_session(current);
1048         else {
1049                 retval = -ESRCH;
1050                 p = find_task_by_vpid(pid);
1051                 if (!p)
1052                         goto out;
1053                 sid = task_session(p);
1054                 if (!sid)
1055                         goto out;
1056
1057                 retval = security_task_getsid(p);
1058                 if (retval)
1059                         goto out;
1060         }
1061         retval = pid_vnr(sid);
1062 out:
1063         rcu_read_unlock();
1064         return retval;
1065 }
1066
1067 SYSCALL_DEFINE0(setsid)
1068 {
1069         struct task_struct *group_leader = current->group_leader;
1070         struct pid *sid = task_pid(group_leader);
1071         pid_t session = pid_vnr(sid);
1072         int err = -EPERM;
1073
1074         write_lock_irq(&tasklist_lock);
1075         /* Fail if I am already a session leader */
1076         if (group_leader->signal->leader)
1077                 goto out;
1078
1079         /* Fail if a process group id already exists that equals the
1080          * proposed session id.
1081          */
1082         if (pid_task(sid, PIDTYPE_PGID))
1083                 goto out;
1084
1085         group_leader->signal->leader = 1;
1086         __set_special_pids(sid);
1087
1088         proc_clear_tty(group_leader);
1089
1090         err = session;
1091 out:
1092         write_unlock_irq(&tasklist_lock);
1093         if (err > 0) {
1094                 proc_sid_connector(group_leader);
1095                 sched_autogroup_create_attach(group_leader);
1096         }
1097         return err;
1098 }
1099
1100 DECLARE_RWSEM(uts_sem);
1101
1102 #ifdef COMPAT_UTS_MACHINE
1103 #define override_architecture(name) \
1104         (personality(current->personality) == PER_LINUX32 && \
1105          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1106                       sizeof(COMPAT_UTS_MACHINE)))
1107 #else
1108 #define override_architecture(name)     0
1109 #endif
1110
1111 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1112 {
1113         int errno = 0;
1114
1115         down_read(&uts_sem);
1116         if (copy_to_user(name, utsname(), sizeof *name))
1117                 errno = -EFAULT;
1118         up_read(&uts_sem);
1119
1120         if (!errno && override_architecture(name))
1121                 errno = -EFAULT;
1122         return errno;
1123 }
1124
1125 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1126 /*
1127  * Old cruft
1128  */
1129 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1130 {
1131         int error = 0;
1132
1133         if (!name)
1134                 return -EFAULT;
1135
1136         down_read(&uts_sem);
1137         if (copy_to_user(name, utsname(), sizeof(*name)))
1138                 error = -EFAULT;
1139         up_read(&uts_sem);
1140
1141         if (!error && override_architecture(name))
1142                 error = -EFAULT;
1143         return error;
1144 }
1145
1146 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1147 {
1148         int error;
1149
1150         if (!name)
1151                 return -EFAULT;
1152         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1153                 return -EFAULT;
1154
1155         down_read(&uts_sem);
1156         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1157                                __OLD_UTS_LEN);
1158         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1159         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1160                                 __OLD_UTS_LEN);
1161         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1162         error |= __copy_to_user(&name->release, &utsname()->release,
1163                                 __OLD_UTS_LEN);
1164         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1165         error |= __copy_to_user(&name->version, &utsname()->version,
1166                                 __OLD_UTS_LEN);
1167         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1168         error |= __copy_to_user(&name->machine, &utsname()->machine,
1169                                 __OLD_UTS_LEN);
1170         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1171         up_read(&uts_sem);
1172
1173         if (!error && override_architecture(name))
1174                 error = -EFAULT;
1175         return error ? -EFAULT : 0;
1176 }
1177 #endif
1178
1179 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1180 {
1181         int errno;
1182         char tmp[__NEW_UTS_LEN];
1183
1184         if (!capable(CAP_SYS_ADMIN))
1185                 return -EPERM;
1186         if (len < 0 || len > __NEW_UTS_LEN)
1187                 return -EINVAL;
1188         down_write(&uts_sem);
1189         errno = -EFAULT;
1190         if (!copy_from_user(tmp, name, len)) {
1191                 struct new_utsname *u = utsname();
1192
1193                 memcpy(u->nodename, tmp, len);
1194                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1195                 errno = 0;
1196         }
1197         up_write(&uts_sem);
1198         return errno;
1199 }
1200
1201 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1202
1203 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1204 {
1205         int i, errno;
1206         struct new_utsname *u;
1207
1208         if (len < 0)
1209                 return -EINVAL;
1210         down_read(&uts_sem);
1211         u = utsname();
1212         i = 1 + strlen(u->nodename);
1213         if (i > len)
1214                 i = len;
1215         errno = 0;
1216         if (copy_to_user(name, u->nodename, i))
1217                 errno = -EFAULT;
1218         up_read(&uts_sem);
1219         return errno;
1220 }
1221
1222 #endif
1223
1224 /*
1225  * Only setdomainname; getdomainname can be implemented by calling
1226  * uname()
1227  */
1228 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1229 {
1230         int errno;
1231         char tmp[__NEW_UTS_LEN];
1232
1233         if (!capable(CAP_SYS_ADMIN))
1234                 return -EPERM;
1235         if (len < 0 || len > __NEW_UTS_LEN)
1236                 return -EINVAL;
1237
1238         down_write(&uts_sem);
1239         errno = -EFAULT;
1240         if (!copy_from_user(tmp, name, len)) {
1241                 struct new_utsname *u = utsname();
1242
1243                 memcpy(u->domainname, tmp, len);
1244                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1245                 errno = 0;
1246         }
1247         up_write(&uts_sem);
1248         return errno;
1249 }
1250
1251 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1252 {
1253         struct rlimit value;
1254         int ret;
1255
1256         ret = do_prlimit(current, resource, NULL, &value);
1257         if (!ret)
1258                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1259
1260         return ret;
1261 }
1262
1263 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1264
1265 /*
1266  *      Back compatibility for getrlimit. Needed for some apps.
1267  */
1268  
1269 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1270                 struct rlimit __user *, rlim)
1271 {
1272         struct rlimit x;
1273         if (resource >= RLIM_NLIMITS)
1274                 return -EINVAL;
1275
1276         task_lock(current->group_leader);
1277         x = current->signal->rlim[resource];
1278         task_unlock(current->group_leader);
1279         if (x.rlim_cur > 0x7FFFFFFF)
1280                 x.rlim_cur = 0x7FFFFFFF;
1281         if (x.rlim_max > 0x7FFFFFFF)
1282                 x.rlim_max = 0x7FFFFFFF;
1283         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1284 }
1285
1286 #endif
1287
1288 static inline bool rlim64_is_infinity(__u64 rlim64)
1289 {
1290 #if BITS_PER_LONG < 64
1291         return rlim64 >= ULONG_MAX;
1292 #else
1293         return rlim64 == RLIM64_INFINITY;
1294 #endif
1295 }
1296
1297 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1298 {
1299         if (rlim->rlim_cur == RLIM_INFINITY)
1300                 rlim64->rlim_cur = RLIM64_INFINITY;
1301         else
1302                 rlim64->rlim_cur = rlim->rlim_cur;
1303         if (rlim->rlim_max == RLIM_INFINITY)
1304                 rlim64->rlim_max = RLIM64_INFINITY;
1305         else
1306                 rlim64->rlim_max = rlim->rlim_max;
1307 }
1308
1309 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1310 {
1311         if (rlim64_is_infinity(rlim64->rlim_cur))
1312                 rlim->rlim_cur = RLIM_INFINITY;
1313         else
1314                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1315         if (rlim64_is_infinity(rlim64->rlim_max))
1316                 rlim->rlim_max = RLIM_INFINITY;
1317         else
1318                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1319 }
1320
1321 /* make sure you are allowed to change @tsk limits before calling this */
1322 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1323                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1324 {
1325         struct rlimit *rlim;
1326         int retval = 0;
1327
1328         if (resource >= RLIM_NLIMITS)
1329                 return -EINVAL;
1330         if (new_rlim) {
1331                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1332                         return -EINVAL;
1333                 if (resource == RLIMIT_NOFILE &&
1334                                 new_rlim->rlim_max > sysctl_nr_open)
1335                         return -EPERM;
1336         }
1337
1338         /* protect tsk->signal and tsk->sighand from disappearing */
1339         read_lock(&tasklist_lock);
1340         if (!tsk->sighand) {
1341                 retval = -ESRCH;
1342                 goto out;
1343         }
1344
1345         rlim = tsk->signal->rlim + resource;
1346         task_lock(tsk->group_leader);
1347         if (new_rlim) {
1348                 if (new_rlim->rlim_max > rlim->rlim_max &&
1349                                 !capable(CAP_SYS_RESOURCE))
1350                         retval = -EPERM;
1351                 if (!retval)
1352                         retval = security_task_setrlimit(tsk->group_leader,
1353                                         resource, new_rlim);
1354                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1355                         /*
1356                          * The caller is asking for an immediate RLIMIT_CPU
1357                          * expiry.  But we use the zero value to mean "it was
1358                          * never set".  So let's cheat and make it one second
1359                          * instead
1360                          */
1361                         new_rlim->rlim_cur = 1;
1362                 }
1363         }
1364         if (!retval) {
1365                 if (old_rlim)
1366                         *old_rlim = *rlim;
1367                 if (new_rlim)
1368                         *rlim = *new_rlim;
1369         }
1370         task_unlock(tsk->group_leader);
1371
1372         /*
1373          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1374          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1375          * very long-standing error, and fixing it now risks breakage of
1376          * applications, so we live with it
1377          */
1378          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1379                          new_rlim->rlim_cur != RLIM_INFINITY)
1380                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1381 out:
1382         read_unlock(&tasklist_lock);
1383         return retval;
1384 }
1385
1386 /* rcu lock must be held */
1387 static int check_prlimit_permission(struct task_struct *task)
1388 {
1389         const struct cred *cred = current_cred(), *tcred;
1390
1391         tcred = __task_cred(task);
1392         if (current != task &&
1393             (cred->uid != tcred->euid ||
1394              cred->uid != tcred->suid ||
1395              cred->uid != tcred->uid  ||
1396              cred->gid != tcred->egid ||
1397              cred->gid != tcred->sgid ||
1398              cred->gid != tcred->gid) &&
1399              !capable(CAP_SYS_RESOURCE)) {
1400                 return -EPERM;
1401         }
1402
1403         return 0;
1404 }
1405
1406 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1407                 const struct rlimit64 __user *, new_rlim,
1408                 struct rlimit64 __user *, old_rlim)
1409 {
1410         struct rlimit64 old64, new64;
1411         struct rlimit old, new;
1412         struct task_struct *tsk;
1413         int ret;
1414
1415         if (new_rlim) {
1416                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1417                         return -EFAULT;
1418                 rlim64_to_rlim(&new64, &new);
1419         }
1420
1421         rcu_read_lock();
1422         tsk = pid ? find_task_by_vpid(pid) : current;
1423         if (!tsk) {
1424                 rcu_read_unlock();
1425                 return -ESRCH;
1426         }
1427         ret = check_prlimit_permission(tsk);
1428         if (ret) {
1429                 rcu_read_unlock();
1430                 return ret;
1431         }
1432         get_task_struct(tsk);
1433         rcu_read_unlock();
1434
1435         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1436                         old_rlim ? &old : NULL);
1437
1438         if (!ret && old_rlim) {
1439                 rlim_to_rlim64(&old, &old64);
1440                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1441                         ret = -EFAULT;
1442         }
1443
1444         put_task_struct(tsk);
1445         return ret;
1446 }
1447
1448 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1449 {
1450         struct rlimit new_rlim;
1451
1452         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1453                 return -EFAULT;
1454         return do_prlimit(current, resource, &new_rlim, NULL);
1455 }
1456
1457 /*
1458  * It would make sense to put struct rusage in the task_struct,
1459  * except that would make the task_struct be *really big*.  After
1460  * task_struct gets moved into malloc'ed memory, it would
1461  * make sense to do this.  It will make moving the rest of the information
1462  * a lot simpler!  (Which we're not doing right now because we're not
1463  * measuring them yet).
1464  *
1465  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1466  * races with threads incrementing their own counters.  But since word
1467  * reads are atomic, we either get new values or old values and we don't
1468  * care which for the sums.  We always take the siglock to protect reading
1469  * the c* fields from p->signal from races with exit.c updating those
1470  * fields when reaping, so a sample either gets all the additions of a
1471  * given child after it's reaped, or none so this sample is before reaping.
1472  *
1473  * Locking:
1474  * We need to take the siglock for CHILDEREN, SELF and BOTH
1475  * for  the cases current multithreaded, non-current single threaded
1476  * non-current multithreaded.  Thread traversal is now safe with
1477  * the siglock held.
1478  * Strictly speaking, we donot need to take the siglock if we are current and
1479  * single threaded,  as no one else can take our signal_struct away, no one
1480  * else can  reap the  children to update signal->c* counters, and no one else
1481  * can race with the signal-> fields. If we do not take any lock, the
1482  * signal-> fields could be read out of order while another thread was just
1483  * exiting. So we should  place a read memory barrier when we avoid the lock.
1484  * On the writer side,  write memory barrier is implied in  __exit_signal
1485  * as __exit_signal releases  the siglock spinlock after updating the signal->
1486  * fields. But we don't do this yet to keep things simple.
1487  *
1488  */
1489
1490 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1491 {
1492         r->ru_nvcsw += t->nvcsw;
1493         r->ru_nivcsw += t->nivcsw;
1494         r->ru_minflt += t->min_flt;
1495         r->ru_majflt += t->maj_flt;
1496         r->ru_inblock += task_io_get_inblock(t);
1497         r->ru_oublock += task_io_get_oublock(t);
1498 }
1499
1500 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1501 {
1502         struct task_struct *t;
1503         unsigned long flags;
1504         cputime_t tgutime, tgstime, utime, stime;
1505         unsigned long maxrss = 0;
1506
1507         memset((char *) r, 0, sizeof *r);
1508         utime = stime = cputime_zero;
1509
1510         if (who == RUSAGE_THREAD) {
1511                 task_times(current, &utime, &stime);
1512                 accumulate_thread_rusage(p, r);
1513                 maxrss = p->signal->maxrss;
1514                 goto out;
1515         }
1516
1517         if (!lock_task_sighand(p, &flags))
1518                 return;
1519
1520         switch (who) {
1521                 case RUSAGE_BOTH:
1522                 case RUSAGE_CHILDREN:
1523                         utime = p->signal->cutime;
1524                         stime = p->signal->cstime;
1525                         r->ru_nvcsw = p->signal->cnvcsw;
1526                         r->ru_nivcsw = p->signal->cnivcsw;
1527                         r->ru_minflt = p->signal->cmin_flt;
1528                         r->ru_majflt = p->signal->cmaj_flt;
1529                         r->ru_inblock = p->signal->cinblock;
1530                         r->ru_oublock = p->signal->coublock;
1531                         maxrss = p->signal->cmaxrss;
1532
1533                         if (who == RUSAGE_CHILDREN)
1534                                 break;
1535
1536                 case RUSAGE_SELF:
1537                         thread_group_times(p, &tgutime, &tgstime);
1538                         utime = cputime_add(utime, tgutime);
1539                         stime = cputime_add(stime, tgstime);
1540                         r->ru_nvcsw += p->signal->nvcsw;
1541                         r->ru_nivcsw += p->signal->nivcsw;
1542                         r->ru_minflt += p->signal->min_flt;
1543                         r->ru_majflt += p->signal->maj_flt;
1544                         r->ru_inblock += p->signal->inblock;
1545                         r->ru_oublock += p->signal->oublock;
1546                         if (maxrss < p->signal->maxrss)
1547                                 maxrss = p->signal->maxrss;
1548                         t = p;
1549                         do {
1550                                 accumulate_thread_rusage(t, r);
1551                                 t = next_thread(t);
1552                         } while (t != p);
1553                         break;
1554
1555                 default:
1556                         BUG();
1557         }
1558         unlock_task_sighand(p, &flags);
1559
1560 out:
1561         cputime_to_timeval(utime, &r->ru_utime);
1562         cputime_to_timeval(stime, &r->ru_stime);
1563
1564         if (who != RUSAGE_CHILDREN) {
1565                 struct mm_struct *mm = get_task_mm(p);
1566                 if (mm) {
1567                         setmax_mm_hiwater_rss(&maxrss, mm);
1568                         mmput(mm);
1569                 }
1570         }
1571         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1572 }
1573
1574 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1575 {
1576         struct rusage r;
1577         k_getrusage(p, who, &r);
1578         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1579 }
1580
1581 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1582 {
1583         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1584             who != RUSAGE_THREAD)
1585                 return -EINVAL;
1586         return getrusage(current, who, ru);
1587 }
1588
1589 SYSCALL_DEFINE1(umask, int, mask)
1590 {
1591         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1592         return mask;
1593 }
1594
1595 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1596                 unsigned long, arg4, unsigned long, arg5)
1597 {
1598         struct task_struct *me = current;
1599         unsigned char comm[sizeof(me->comm)];
1600         long error;
1601
1602         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1603         if (error != -ENOSYS)
1604                 return error;
1605
1606         error = 0;
1607         switch (option) {
1608                 case PR_SET_PDEATHSIG:
1609                         if (!valid_signal(arg2)) {
1610                                 error = -EINVAL;
1611                                 break;
1612                         }
1613                         me->pdeath_signal = arg2;
1614                         error = 0;
1615                         break;
1616                 case PR_GET_PDEATHSIG:
1617                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1618                         break;
1619                 case PR_GET_DUMPABLE:
1620                         error = get_dumpable(me->mm);
1621                         break;
1622                 case PR_SET_DUMPABLE:
1623                         if (arg2 < 0 || arg2 > 1) {
1624                                 error = -EINVAL;
1625                                 break;
1626                         }
1627                         set_dumpable(me->mm, arg2);
1628                         error = 0;
1629                         break;
1630
1631                 case PR_SET_UNALIGN:
1632                         error = SET_UNALIGN_CTL(me, arg2);
1633                         break;
1634                 case PR_GET_UNALIGN:
1635                         error = GET_UNALIGN_CTL(me, arg2);
1636                         break;
1637                 case PR_SET_FPEMU:
1638                         error = SET_FPEMU_CTL(me, arg2);
1639                         break;
1640                 case PR_GET_FPEMU:
1641                         error = GET_FPEMU_CTL(me, arg2);
1642                         break;
1643                 case PR_SET_FPEXC:
1644                         error = SET_FPEXC_CTL(me, arg2);
1645                         break;
1646                 case PR_GET_FPEXC:
1647                         error = GET_FPEXC_CTL(me, arg2);
1648                         break;
1649                 case PR_GET_TIMING:
1650                         error = PR_TIMING_STATISTICAL;
1651                         break;
1652                 case PR_SET_TIMING:
1653                         if (arg2 != PR_TIMING_STATISTICAL)
1654                                 error = -EINVAL;
1655                         else
1656                                 error = 0;
1657                         break;
1658
1659                 case PR_SET_NAME:
1660                         comm[sizeof(me->comm)-1] = 0;
1661                         if (strncpy_from_user(comm, (char __user *)arg2,
1662                                               sizeof(me->comm) - 1) < 0)
1663                                 return -EFAULT;
1664                         set_task_comm(me, comm);
1665                         return 0;
1666                 case PR_GET_NAME:
1667                         get_task_comm(comm, me);
1668                         if (copy_to_user((char __user *)arg2, comm,
1669                                          sizeof(comm)))
1670                                 return -EFAULT;
1671                         return 0;
1672                 case PR_GET_ENDIAN:
1673                         error = GET_ENDIAN(me, arg2);
1674                         break;
1675                 case PR_SET_ENDIAN:
1676                         error = SET_ENDIAN(me, arg2);
1677                         break;
1678
1679                 case PR_GET_SECCOMP:
1680                         error = prctl_get_seccomp();
1681                         break;
1682                 case PR_SET_SECCOMP:
1683                         error = prctl_set_seccomp(arg2);
1684                         break;
1685                 case PR_GET_TSC:
1686                         error = GET_TSC_CTL(arg2);
1687                         break;
1688                 case PR_SET_TSC:
1689                         error = SET_TSC_CTL(arg2);
1690                         break;
1691                 case PR_TASK_PERF_EVENTS_DISABLE:
1692                         error = perf_event_task_disable();
1693                         break;
1694                 case PR_TASK_PERF_EVENTS_ENABLE:
1695                         error = perf_event_task_enable();
1696                         break;
1697                 case PR_GET_TIMERSLACK:
1698                         error = current->timer_slack_ns;
1699                         break;
1700                 case PR_SET_TIMERSLACK:
1701                         if (arg2 <= 0)
1702                                 current->timer_slack_ns =
1703                                         current->default_timer_slack_ns;
1704                         else
1705                                 current->timer_slack_ns = arg2;
1706                         error = 0;
1707                         break;
1708                 case PR_MCE_KILL:
1709                         if (arg4 | arg5)
1710                                 return -EINVAL;
1711                         switch (arg2) {
1712                         case PR_MCE_KILL_CLEAR:
1713                                 if (arg3 != 0)
1714                                         return -EINVAL;
1715                                 current->flags &= ~PF_MCE_PROCESS;
1716                                 break;
1717                         case PR_MCE_KILL_SET:
1718                                 current->flags |= PF_MCE_PROCESS;
1719                                 if (arg3 == PR_MCE_KILL_EARLY)
1720                                         current->flags |= PF_MCE_EARLY;
1721                                 else if (arg3 == PR_MCE_KILL_LATE)
1722                                         current->flags &= ~PF_MCE_EARLY;
1723                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1724                                         current->flags &=
1725                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1726                                 else
1727                                         return -EINVAL;
1728                                 break;
1729                         default:
1730                                 return -EINVAL;
1731                         }
1732                         error = 0;
1733                         break;
1734                 case PR_MCE_KILL_GET:
1735                         if (arg2 | arg3 | arg4 | arg5)
1736                                 return -EINVAL;
1737                         if (current->flags & PF_MCE_PROCESS)
1738                                 error = (current->flags & PF_MCE_EARLY) ?
1739                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1740                         else
1741                                 error = PR_MCE_KILL_DEFAULT;
1742                         break;
1743                 default:
1744                         error = -EINVAL;
1745                         break;
1746         }
1747         return error;
1748 }
1749
1750 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1751                 struct getcpu_cache __user *, unused)
1752 {
1753         int err = 0;
1754         int cpu = raw_smp_processor_id();
1755         if (cpup)
1756                 err |= put_user(cpu, cpup);
1757         if (nodep)
1758                 err |= put_user(cpu_to_node(cpu), nodep);
1759         return err ? -EFAULT : 0;
1760 }
1761
1762 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1763
1764 static void argv_cleanup(struct subprocess_info *info)
1765 {
1766         argv_free(info->argv);
1767 }
1768
1769 /**
1770  * orderly_poweroff - Trigger an orderly system poweroff
1771  * @force: force poweroff if command execution fails
1772  *
1773  * This may be called from any context to trigger a system shutdown.
1774  * If the orderly shutdown fails, it will force an immediate shutdown.
1775  */
1776 int orderly_poweroff(bool force)
1777 {
1778         int argc;
1779         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1780         static char *envp[] = {
1781                 "HOME=/",
1782                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1783                 NULL
1784         };
1785         int ret = -ENOMEM;
1786         struct subprocess_info *info;
1787
1788         if (argv == NULL) {
1789                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1790                        __func__, poweroff_cmd);
1791                 goto out;
1792         }
1793
1794         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1795         if (info == NULL) {
1796                 argv_free(argv);
1797                 goto out;
1798         }
1799
1800         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1801
1802         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1803
1804   out:
1805         if (ret && force) {
1806                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1807                        "forcing the issue\n");
1808
1809                 /* I guess this should try to kick off some daemon to
1810                    sync and poweroff asap.  Or not even bother syncing
1811                    if we're doing an emergency shutdown? */
1812                 emergency_sync();
1813                 kernel_power_off();
1814         }
1815
1816         return ret;
1817 }
1818 EXPORT_SYMBOL_GPL(orderly_poweroff);