reboot: disable usermodehelper to prevent fs access
[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  * Returns true if current's euid is same as p's uid or euid,
124  * or has CAP_SYS_NICE to p's user_ns.
125  *
126  * Called with rcu_read_lock, creds are safe
127  */
128 static bool set_one_prio_perm(struct task_struct *p)
129 {
130         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
131
132         if (pcred->user->user_ns == cred->user->user_ns &&
133             (pcred->uid  == cred->euid ||
134              pcred->euid == cred->euid))
135                 return true;
136         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
137                 return true;
138         return false;
139 }
140
141 /*
142  * set the priority of a task
143  * - the caller must hold the RCU read lock
144  */
145 static int set_one_prio(struct task_struct *p, int niceval, int error)
146 {
147         int no_nice;
148
149         if (!set_one_prio_perm(p)) {
150                 error = -EPERM;
151                 goto out;
152         }
153         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
154                 error = -EACCES;
155                 goto out;
156         }
157         no_nice = security_task_setnice(p, niceval);
158         if (no_nice) {
159                 error = no_nice;
160                 goto out;
161         }
162         if (error == -ESRCH)
163                 error = 0;
164         set_user_nice(p, niceval);
165 out:
166         return error;
167 }
168
169 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
170 {
171         struct task_struct *g, *p;
172         struct user_struct *user;
173         const struct cred *cred = current_cred();
174         int error = -EINVAL;
175         struct pid *pgrp;
176
177         if (which > PRIO_USER || which < PRIO_PROCESS)
178                 goto out;
179
180         /* normalize: avoid signed division (rounding problems) */
181         error = -ESRCH;
182         if (niceval < -20)
183                 niceval = -20;
184         if (niceval > 19)
185                 niceval = 19;
186
187         rcu_read_lock();
188         read_lock(&tasklist_lock);
189         switch (which) {
190                 case PRIO_PROCESS:
191                         if (who)
192                                 p = find_task_by_vpid(who);
193                         else
194                                 p = current;
195                         if (p)
196                                 error = set_one_prio(p, niceval, error);
197                         break;
198                 case PRIO_PGRP:
199                         if (who)
200                                 pgrp = find_vpid(who);
201                         else
202                                 pgrp = task_pgrp(current);
203                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
204                                 error = set_one_prio(p, niceval, error);
205                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
206                         break;
207                 case PRIO_USER:
208                         user = (struct user_struct *) cred->user;
209                         if (!who)
210                                 who = cred->uid;
211                         else if ((who != cred->uid) &&
212                                  !(user = find_user(who)))
213                                 goto out_unlock;        /* No processes for this user */
214
215                         do_each_thread(g, p) {
216                                 if (__task_cred(p)->uid == who)
217                                         error = set_one_prio(p, niceval, error);
218                         } while_each_thread(g, p);
219                         if (who != cred->uid)
220                                 free_uid(user);         /* For find_user() */
221                         break;
222         }
223 out_unlock:
224         read_unlock(&tasklist_lock);
225         rcu_read_unlock();
226 out:
227         return error;
228 }
229
230 /*
231  * Ugh. To avoid negative return values, "getpriority()" will
232  * not return the normal nice-value, but a negated value that
233  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
234  * to stay compatible.
235  */
236 SYSCALL_DEFINE2(getpriority, int, which, int, who)
237 {
238         struct task_struct *g, *p;
239         struct user_struct *user;
240         const struct cred *cred = current_cred();
241         long niceval, retval = -ESRCH;
242         struct pid *pgrp;
243
244         if (which > PRIO_USER || which < PRIO_PROCESS)
245                 return -EINVAL;
246
247         rcu_read_lock();
248         read_lock(&tasklist_lock);
249         switch (which) {
250                 case PRIO_PROCESS:
251                         if (who)
252                                 p = find_task_by_vpid(who);
253                         else
254                                 p = current;
255                         if (p) {
256                                 niceval = 20 - task_nice(p);
257                                 if (niceval > retval)
258                                         retval = niceval;
259                         }
260                         break;
261                 case PRIO_PGRP:
262                         if (who)
263                                 pgrp = find_vpid(who);
264                         else
265                                 pgrp = task_pgrp(current);
266                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
267                                 niceval = 20 - task_nice(p);
268                                 if (niceval > retval)
269                                         retval = niceval;
270                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
271                         break;
272                 case PRIO_USER:
273                         user = (struct user_struct *) cred->user;
274                         if (!who)
275                                 who = cred->uid;
276                         else if ((who != cred->uid) &&
277                                  !(user = find_user(who)))
278                                 goto out_unlock;        /* No processes for this user */
279
280                         do_each_thread(g, p) {
281                                 if (__task_cred(p)->uid == who) {
282                                         niceval = 20 - task_nice(p);
283                                         if (niceval > retval)
284                                                 retval = niceval;
285                                 }
286                         } while_each_thread(g, p);
287                         if (who != cred->uid)
288                                 free_uid(user);         /* for find_user() */
289                         break;
290         }
291 out_unlock:
292         read_unlock(&tasklist_lock);
293         rcu_read_unlock();
294
295         return retval;
296 }
297
298 /**
299  *      emergency_restart - reboot the system
300  *
301  *      Without shutting down any hardware or taking any locks
302  *      reboot the system.  This is called when we know we are in
303  *      trouble so this is our best effort to reboot.  This is
304  *      safe to call in interrupt context.
305  */
306 void emergency_restart(void)
307 {
308         kmsg_dump(KMSG_DUMP_EMERG);
309         machine_emergency_restart();
310 }
311 EXPORT_SYMBOL_GPL(emergency_restart);
312
313 void kernel_restart_prepare(char *cmd)
314 {
315         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
316         system_state = SYSTEM_RESTART;
317         usermodehelper_disable();
318         device_shutdown();
319         sysdev_shutdown();
320         syscore_shutdown();
321 }
322
323 /**
324  *      kernel_restart - reboot the system
325  *      @cmd: pointer to buffer containing command to execute for restart
326  *              or %NULL
327  *
328  *      Shutdown everything and perform a clean reboot.
329  *      This is not safe to call in interrupt context.
330  */
331 void kernel_restart(char *cmd)
332 {
333         kernel_restart_prepare(cmd);
334         if (!cmd)
335                 printk(KERN_EMERG "Restarting system.\n");
336         else
337                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
338         kmsg_dump(KMSG_DUMP_RESTART);
339         machine_restart(cmd);
340 }
341 EXPORT_SYMBOL_GPL(kernel_restart);
342
343 static void kernel_shutdown_prepare(enum system_states state)
344 {
345         blocking_notifier_call_chain(&reboot_notifier_list,
346                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
347         system_state = state;
348         usermodehelper_disable();
349         device_shutdown();
350 }
351 /**
352  *      kernel_halt - halt the system
353  *
354  *      Shutdown everything and perform a clean system halt.
355  */
356 void kernel_halt(void)
357 {
358         kernel_shutdown_prepare(SYSTEM_HALT);
359         sysdev_shutdown();
360         syscore_shutdown();
361         printk(KERN_EMERG "System halted.\n");
362         kmsg_dump(KMSG_DUMP_HALT);
363         machine_halt();
364 }
365
366 EXPORT_SYMBOL_GPL(kernel_halt);
367
368 /**
369  *      kernel_power_off - power_off the system
370  *
371  *      Shutdown everything and perform a clean system power_off.
372  */
373 void kernel_power_off(void)
374 {
375         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
376         if (pm_power_off_prepare)
377                 pm_power_off_prepare();
378         disable_nonboot_cpus();
379         sysdev_shutdown();
380         syscore_shutdown();
381         printk(KERN_EMERG "Power down.\n");
382         kmsg_dump(KMSG_DUMP_POWEROFF);
383         machine_power_off();
384 }
385 EXPORT_SYMBOL_GPL(kernel_power_off);
386
387 static DEFINE_MUTEX(reboot_mutex);
388
389 /*
390  * Reboot system call: for obvious reasons only root may call it,
391  * and even root needs to set up some magic numbers in the registers
392  * so that some mistake won't make this reboot the whole machine.
393  * You can also set the meaning of the ctrl-alt-del-key here.
394  *
395  * reboot doesn't sync: do that yourself before calling this.
396  */
397 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
398                 void __user *, arg)
399 {
400         char buffer[256];
401         int ret = 0;
402
403         /* We only trust the superuser with rebooting the system. */
404         if (!capable(CAP_SYS_BOOT))
405                 return -EPERM;
406
407         /* For safety, we require "magic" arguments. */
408         if (magic1 != LINUX_REBOOT_MAGIC1 ||
409             (magic2 != LINUX_REBOOT_MAGIC2 &&
410                         magic2 != LINUX_REBOOT_MAGIC2A &&
411                         magic2 != LINUX_REBOOT_MAGIC2B &&
412                         magic2 != LINUX_REBOOT_MAGIC2C))
413                 return -EINVAL;
414
415         /* Instead of trying to make the power_off code look like
416          * halt when pm_power_off is not set do it the easy way.
417          */
418         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
419                 cmd = LINUX_REBOOT_CMD_HALT;
420
421         mutex_lock(&reboot_mutex);
422         switch (cmd) {
423         case LINUX_REBOOT_CMD_RESTART:
424                 kernel_restart(NULL);
425                 break;
426
427         case LINUX_REBOOT_CMD_CAD_ON:
428                 C_A_D = 1;
429                 break;
430
431         case LINUX_REBOOT_CMD_CAD_OFF:
432                 C_A_D = 0;
433                 break;
434
435         case LINUX_REBOOT_CMD_HALT:
436                 kernel_halt();
437                 do_exit(0);
438                 panic("cannot halt");
439
440         case LINUX_REBOOT_CMD_POWER_OFF:
441                 kernel_power_off();
442                 do_exit(0);
443                 break;
444
445         case LINUX_REBOOT_CMD_RESTART2:
446                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
447                         ret = -EFAULT;
448                         break;
449                 }
450                 buffer[sizeof(buffer) - 1] = '\0';
451
452                 kernel_restart(buffer);
453                 break;
454
455 #ifdef CONFIG_KEXEC
456         case LINUX_REBOOT_CMD_KEXEC:
457                 ret = kernel_kexec();
458                 break;
459 #endif
460
461 #ifdef CONFIG_HIBERNATION
462         case LINUX_REBOOT_CMD_SW_SUSPEND:
463                 ret = hibernate();
464                 break;
465 #endif
466
467         default:
468                 ret = -EINVAL;
469                 break;
470         }
471         mutex_unlock(&reboot_mutex);
472         return ret;
473 }
474
475 static void deferred_cad(struct work_struct *dummy)
476 {
477         kernel_restart(NULL);
478 }
479
480 /*
481  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
482  * As it's called within an interrupt, it may NOT sync: the only choice
483  * is whether to reboot at once, or just ignore the ctrl-alt-del.
484  */
485 void ctrl_alt_del(void)
486 {
487         static DECLARE_WORK(cad_work, deferred_cad);
488
489         if (C_A_D)
490                 schedule_work(&cad_work);
491         else
492                 kill_cad_pid(SIGINT, 1);
493 }
494         
495 /*
496  * Unprivileged users may change the real gid to the effective gid
497  * or vice versa.  (BSD-style)
498  *
499  * If you set the real gid at all, or set the effective gid to a value not
500  * equal to the real gid, then the saved gid is set to the new effective gid.
501  *
502  * This makes it possible for a setgid program to completely drop its
503  * privileges, which is often a useful assertion to make when you are doing
504  * a security audit over a program.
505  *
506  * The general idea is that a program which uses just setregid() will be
507  * 100% compatible with BSD.  A program which uses just setgid() will be
508  * 100% compatible with POSIX with saved IDs. 
509  *
510  * SMP: There are not races, the GIDs are checked only by filesystem
511  *      operations (as far as semantic preservation is concerned).
512  */
513 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
514 {
515         const struct cred *old;
516         struct cred *new;
517         int retval;
518
519         new = prepare_creds();
520         if (!new)
521                 return -ENOMEM;
522         old = current_cred();
523
524         retval = -EPERM;
525         if (rgid != (gid_t) -1) {
526                 if (old->gid == rgid ||
527                     old->egid == rgid ||
528                     nsown_capable(CAP_SETGID))
529                         new->gid = rgid;
530                 else
531                         goto error;
532         }
533         if (egid != (gid_t) -1) {
534                 if (old->gid == egid ||
535                     old->egid == egid ||
536                     old->sgid == egid ||
537                     nsown_capable(CAP_SETGID))
538                         new->egid = egid;
539                 else
540                         goto error;
541         }
542
543         if (rgid != (gid_t) -1 ||
544             (egid != (gid_t) -1 && egid != old->gid))
545                 new->sgid = new->egid;
546         new->fsgid = new->egid;
547
548         return commit_creds(new);
549
550 error:
551         abort_creds(new);
552         return retval;
553 }
554
555 /*
556  * setgid() is implemented like SysV w/ SAVED_IDS 
557  *
558  * SMP: Same implicit races as above.
559  */
560 SYSCALL_DEFINE1(setgid, gid_t, gid)
561 {
562         const struct cred *old;
563         struct cred *new;
564         int retval;
565
566         new = prepare_creds();
567         if (!new)
568                 return -ENOMEM;
569         old = current_cred();
570
571         retval = -EPERM;
572         if (nsown_capable(CAP_SETGID))
573                 new->gid = new->egid = new->sgid = new->fsgid = gid;
574         else if (gid == old->gid || gid == old->sgid)
575                 new->egid = new->fsgid = gid;
576         else
577                 goto error;
578
579         return commit_creds(new);
580
581 error:
582         abort_creds(new);
583         return retval;
584 }
585
586 /*
587  * change the user struct in a credentials set to match the new UID
588  */
589 static int set_user(struct cred *new)
590 {
591         struct user_struct *new_user;
592
593         new_user = alloc_uid(current_user_ns(), new->uid);
594         if (!new_user)
595                 return -EAGAIN;
596
597         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
598                         new_user != INIT_USER) {
599                 free_uid(new_user);
600                 return -EAGAIN;
601         }
602
603         free_uid(new->user);
604         new->user = new_user;
605         return 0;
606 }
607
608 /*
609  * Unprivileged users may change the real uid to the effective uid
610  * or vice versa.  (BSD-style)
611  *
612  * If you set the real uid at all, or set the effective uid to a value not
613  * equal to the real uid, then the saved uid is set to the new effective uid.
614  *
615  * This makes it possible for a setuid program to completely drop its
616  * privileges, which is often a useful assertion to make when you are doing
617  * a security audit over a program.
618  *
619  * The general idea is that a program which uses just setreuid() will be
620  * 100% compatible with BSD.  A program which uses just setuid() will be
621  * 100% compatible with POSIX with saved IDs. 
622  */
623 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
624 {
625         const struct cred *old;
626         struct cred *new;
627         int retval;
628
629         new = prepare_creds();
630         if (!new)
631                 return -ENOMEM;
632         old = current_cred();
633
634         retval = -EPERM;
635         if (ruid != (uid_t) -1) {
636                 new->uid = ruid;
637                 if (old->uid != ruid &&
638                     old->euid != ruid &&
639                     !nsown_capable(CAP_SETUID))
640                         goto error;
641         }
642
643         if (euid != (uid_t) -1) {
644                 new->euid = euid;
645                 if (old->uid != euid &&
646                     old->euid != euid &&
647                     old->suid != euid &&
648                     !nsown_capable(CAP_SETUID))
649                         goto error;
650         }
651
652         if (new->uid != old->uid) {
653                 retval = set_user(new);
654                 if (retval < 0)
655                         goto error;
656         }
657         if (ruid != (uid_t) -1 ||
658             (euid != (uid_t) -1 && euid != old->uid))
659                 new->suid = new->euid;
660         new->fsuid = new->euid;
661
662         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
663         if (retval < 0)
664                 goto error;
665
666         return commit_creds(new);
667
668 error:
669         abort_creds(new);
670         return retval;
671 }
672                 
673 /*
674  * setuid() is implemented like SysV with SAVED_IDS 
675  * 
676  * Note that SAVED_ID's is deficient in that a setuid root program
677  * like sendmail, for example, cannot set its uid to be a normal 
678  * user and then switch back, because if you're root, setuid() sets
679  * the saved uid too.  If you don't like this, blame the bright people
680  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
681  * will allow a root program to temporarily drop privileges and be able to
682  * regain them by swapping the real and effective uid.  
683  */
684 SYSCALL_DEFINE1(setuid, uid_t, uid)
685 {
686         const struct cred *old;
687         struct cred *new;
688         int retval;
689
690         new = prepare_creds();
691         if (!new)
692                 return -ENOMEM;
693         old = current_cred();
694
695         retval = -EPERM;
696         if (nsown_capable(CAP_SETUID)) {
697                 new->suid = new->uid = uid;
698                 if (uid != old->uid) {
699                         retval = set_user(new);
700                         if (retval < 0)
701                                 goto error;
702                 }
703         } else if (uid != old->uid && uid != new->suid) {
704                 goto error;
705         }
706
707         new->fsuid = new->euid = uid;
708
709         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
710         if (retval < 0)
711                 goto error;
712
713         return commit_creds(new);
714
715 error:
716         abort_creds(new);
717         return retval;
718 }
719
720
721 /*
722  * This function implements a generic ability to update ruid, euid,
723  * and suid.  This allows you to implement the 4.4 compatible seteuid().
724  */
725 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
726 {
727         const struct cred *old;
728         struct cred *new;
729         int retval;
730
731         new = prepare_creds();
732         if (!new)
733                 return -ENOMEM;
734
735         old = current_cred();
736
737         retval = -EPERM;
738         if (!nsown_capable(CAP_SETUID)) {
739                 if (ruid != (uid_t) -1 && ruid != old->uid &&
740                     ruid != old->euid  && ruid != old->suid)
741                         goto error;
742                 if (euid != (uid_t) -1 && euid != old->uid &&
743                     euid != old->euid  && euid != old->suid)
744                         goto error;
745                 if (suid != (uid_t) -1 && suid != old->uid &&
746                     suid != old->euid  && suid != old->suid)
747                         goto error;
748         }
749
750         if (ruid != (uid_t) -1) {
751                 new->uid = ruid;
752                 if (ruid != old->uid) {
753                         retval = set_user(new);
754                         if (retval < 0)
755                                 goto error;
756                 }
757         }
758         if (euid != (uid_t) -1)
759                 new->euid = euid;
760         if (suid != (uid_t) -1)
761                 new->suid = suid;
762         new->fsuid = new->euid;
763
764         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
765         if (retval < 0)
766                 goto error;
767
768         return commit_creds(new);
769
770 error:
771         abort_creds(new);
772         return retval;
773 }
774
775 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
776 {
777         const struct cred *cred = current_cred();
778         int retval;
779
780         if (!(retval   = put_user(cred->uid,  ruid)) &&
781             !(retval   = put_user(cred->euid, euid)))
782                 retval = put_user(cred->suid, suid);
783
784         return retval;
785 }
786
787 /*
788  * Same as above, but for rgid, egid, sgid.
789  */
790 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
791 {
792         const struct cred *old;
793         struct cred *new;
794         int retval;
795
796         new = prepare_creds();
797         if (!new)
798                 return -ENOMEM;
799         old = current_cred();
800
801         retval = -EPERM;
802         if (!nsown_capable(CAP_SETGID)) {
803                 if (rgid != (gid_t) -1 && rgid != old->gid &&
804                     rgid != old->egid  && rgid != old->sgid)
805                         goto error;
806                 if (egid != (gid_t) -1 && egid != old->gid &&
807                     egid != old->egid  && egid != old->sgid)
808                         goto error;
809                 if (sgid != (gid_t) -1 && sgid != old->gid &&
810                     sgid != old->egid  && sgid != old->sgid)
811                         goto error;
812         }
813
814         if (rgid != (gid_t) -1)
815                 new->gid = rgid;
816         if (egid != (gid_t) -1)
817                 new->egid = egid;
818         if (sgid != (gid_t) -1)
819                 new->sgid = sgid;
820         new->fsgid = new->egid;
821
822         return commit_creds(new);
823
824 error:
825         abort_creds(new);
826         return retval;
827 }
828
829 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
830 {
831         const struct cred *cred = current_cred();
832         int retval;
833
834         if (!(retval   = put_user(cred->gid,  rgid)) &&
835             !(retval   = put_user(cred->egid, egid)))
836                 retval = put_user(cred->sgid, sgid);
837
838         return retval;
839 }
840
841
842 /*
843  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
844  * is used for "access()" and for the NFS daemon (letting nfsd stay at
845  * whatever uid it wants to). It normally shadows "euid", except when
846  * explicitly set by setfsuid() or for access..
847  */
848 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
849 {
850         const struct cred *old;
851         struct cred *new;
852         uid_t old_fsuid;
853
854         new = prepare_creds();
855         if (!new)
856                 return current_fsuid();
857         old = current_cred();
858         old_fsuid = old->fsuid;
859
860         if (uid == old->uid  || uid == old->euid  ||
861             uid == old->suid || uid == old->fsuid ||
862             nsown_capable(CAP_SETUID)) {
863                 if (uid != old_fsuid) {
864                         new->fsuid = uid;
865                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
866                                 goto change_okay;
867                 }
868         }
869
870         abort_creds(new);
871         return old_fsuid;
872
873 change_okay:
874         commit_creds(new);
875         return old_fsuid;
876 }
877
878 /*
879  * Samma på svenska..
880  */
881 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
882 {
883         const struct cred *old;
884         struct cred *new;
885         gid_t old_fsgid;
886
887         new = prepare_creds();
888         if (!new)
889                 return current_fsgid();
890         old = current_cred();
891         old_fsgid = old->fsgid;
892
893         if (gid == old->gid  || gid == old->egid  ||
894             gid == old->sgid || gid == old->fsgid ||
895             nsown_capable(CAP_SETGID)) {
896                 if (gid != old_fsgid) {
897                         new->fsgid = gid;
898                         goto change_okay;
899                 }
900         }
901
902         abort_creds(new);
903         return old_fsgid;
904
905 change_okay:
906         commit_creds(new);
907         return old_fsgid;
908 }
909
910 void do_sys_times(struct tms *tms)
911 {
912         cputime_t tgutime, tgstime, cutime, cstime;
913
914         spin_lock_irq(&current->sighand->siglock);
915         thread_group_times(current, &tgutime, &tgstime);
916         cutime = current->signal->cutime;
917         cstime = current->signal->cstime;
918         spin_unlock_irq(&current->sighand->siglock);
919         tms->tms_utime = cputime_to_clock_t(tgutime);
920         tms->tms_stime = cputime_to_clock_t(tgstime);
921         tms->tms_cutime = cputime_to_clock_t(cutime);
922         tms->tms_cstime = cputime_to_clock_t(cstime);
923 }
924
925 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
926 {
927         if (tbuf) {
928                 struct tms tmp;
929
930                 do_sys_times(&tmp);
931                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
932                         return -EFAULT;
933         }
934         force_successful_syscall_return();
935         return (long) jiffies_64_to_clock_t(get_jiffies_64());
936 }
937
938 /*
939  * This needs some heavy checking ...
940  * I just haven't the stomach for it. I also don't fully
941  * understand sessions/pgrp etc. Let somebody who does explain it.
942  *
943  * OK, I think I have the protection semantics right.... this is really
944  * only important on a multi-user system anyway, to make sure one user
945  * can't send a signal to a process owned by another.  -TYT, 12/12/91
946  *
947  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
948  * LBT 04.03.94
949  */
950 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
951 {
952         struct task_struct *p;
953         struct task_struct *group_leader = current->group_leader;
954         struct pid *pgrp;
955         int err;
956
957         if (!pid)
958                 pid = task_pid_vnr(group_leader);
959         if (!pgid)
960                 pgid = pid;
961         if (pgid < 0)
962                 return -EINVAL;
963         rcu_read_lock();
964
965         /* From this point forward we keep holding onto the tasklist lock
966          * so that our parent does not change from under us. -DaveM
967          */
968         write_lock_irq(&tasklist_lock);
969
970         err = -ESRCH;
971         p = find_task_by_vpid(pid);
972         if (!p)
973                 goto out;
974
975         err = -EINVAL;
976         if (!thread_group_leader(p))
977                 goto out;
978
979         if (same_thread_group(p->real_parent, group_leader)) {
980                 err = -EPERM;
981                 if (task_session(p) != task_session(group_leader))
982                         goto out;
983                 err = -EACCES;
984                 if (p->did_exec)
985                         goto out;
986         } else {
987                 err = -ESRCH;
988                 if (p != group_leader)
989                         goto out;
990         }
991
992         err = -EPERM;
993         if (p->signal->leader)
994                 goto out;
995
996         pgrp = task_pid(p);
997         if (pgid != pid) {
998                 struct task_struct *g;
999
1000                 pgrp = find_vpid(pgid);
1001                 g = pid_task(pgrp, PIDTYPE_PGID);
1002                 if (!g || task_session(g) != task_session(group_leader))
1003                         goto out;
1004         }
1005
1006         err = security_task_setpgid(p, pgid);
1007         if (err)
1008                 goto out;
1009
1010         if (task_pgrp(p) != pgrp)
1011                 change_pid(p, PIDTYPE_PGID, pgrp);
1012
1013         err = 0;
1014 out:
1015         /* All paths lead to here, thus we are safe. -DaveM */
1016         write_unlock_irq(&tasklist_lock);
1017         rcu_read_unlock();
1018         return err;
1019 }
1020
1021 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1022 {
1023         struct task_struct *p;
1024         struct pid *grp;
1025         int retval;
1026
1027         rcu_read_lock();
1028         if (!pid)
1029                 grp = task_pgrp(current);
1030         else {
1031                 retval = -ESRCH;
1032                 p = find_task_by_vpid(pid);
1033                 if (!p)
1034                         goto out;
1035                 grp = task_pgrp(p);
1036                 if (!grp)
1037                         goto out;
1038
1039                 retval = security_task_getpgid(p);
1040                 if (retval)
1041                         goto out;
1042         }
1043         retval = pid_vnr(grp);
1044 out:
1045         rcu_read_unlock();
1046         return retval;
1047 }
1048
1049 #ifdef __ARCH_WANT_SYS_GETPGRP
1050
1051 SYSCALL_DEFINE0(getpgrp)
1052 {
1053         return sys_getpgid(0);
1054 }
1055
1056 #endif
1057
1058 SYSCALL_DEFINE1(getsid, pid_t, pid)
1059 {
1060         struct task_struct *p;
1061         struct pid *sid;
1062         int retval;
1063
1064         rcu_read_lock();
1065         if (!pid)
1066                 sid = task_session(current);
1067         else {
1068                 retval = -ESRCH;
1069                 p = find_task_by_vpid(pid);
1070                 if (!p)
1071                         goto out;
1072                 sid = task_session(p);
1073                 if (!sid)
1074                         goto out;
1075
1076                 retval = security_task_getsid(p);
1077                 if (retval)
1078                         goto out;
1079         }
1080         retval = pid_vnr(sid);
1081 out:
1082         rcu_read_unlock();
1083         return retval;
1084 }
1085
1086 SYSCALL_DEFINE0(setsid)
1087 {
1088         struct task_struct *group_leader = current->group_leader;
1089         struct pid *sid = task_pid(group_leader);
1090         pid_t session = pid_vnr(sid);
1091         int err = -EPERM;
1092
1093         write_lock_irq(&tasklist_lock);
1094         /* Fail if I am already a session leader */
1095         if (group_leader->signal->leader)
1096                 goto out;
1097
1098         /* Fail if a process group id already exists that equals the
1099          * proposed session id.
1100          */
1101         if (pid_task(sid, PIDTYPE_PGID))
1102                 goto out;
1103
1104         group_leader->signal->leader = 1;
1105         __set_special_pids(sid);
1106
1107         proc_clear_tty(group_leader);
1108
1109         err = session;
1110 out:
1111         write_unlock_irq(&tasklist_lock);
1112         if (err > 0) {
1113                 proc_sid_connector(group_leader);
1114                 sched_autogroup_create_attach(group_leader);
1115         }
1116         return err;
1117 }
1118
1119 DECLARE_RWSEM(uts_sem);
1120
1121 #ifdef COMPAT_UTS_MACHINE
1122 #define override_architecture(name) \
1123         (personality(current->personality) == PER_LINUX32 && \
1124          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1125                       sizeof(COMPAT_UTS_MACHINE)))
1126 #else
1127 #define override_architecture(name)     0
1128 #endif
1129
1130 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1131 {
1132         int errno = 0;
1133
1134         down_read(&uts_sem);
1135         if (copy_to_user(name, utsname(), sizeof *name))
1136                 errno = -EFAULT;
1137         up_read(&uts_sem);
1138
1139         if (!errno && override_architecture(name))
1140                 errno = -EFAULT;
1141         return errno;
1142 }
1143
1144 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1145 /*
1146  * Old cruft
1147  */
1148 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1149 {
1150         int error = 0;
1151
1152         if (!name)
1153                 return -EFAULT;
1154
1155         down_read(&uts_sem);
1156         if (copy_to_user(name, utsname(), sizeof(*name)))
1157                 error = -EFAULT;
1158         up_read(&uts_sem);
1159
1160         if (!error && override_architecture(name))
1161                 error = -EFAULT;
1162         return error;
1163 }
1164
1165 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1166 {
1167         int error;
1168
1169         if (!name)
1170                 return -EFAULT;
1171         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1172                 return -EFAULT;
1173
1174         down_read(&uts_sem);
1175         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1176                                __OLD_UTS_LEN);
1177         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1178         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1179                                 __OLD_UTS_LEN);
1180         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1181         error |= __copy_to_user(&name->release, &utsname()->release,
1182                                 __OLD_UTS_LEN);
1183         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1184         error |= __copy_to_user(&name->version, &utsname()->version,
1185                                 __OLD_UTS_LEN);
1186         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1187         error |= __copy_to_user(&name->machine, &utsname()->machine,
1188                                 __OLD_UTS_LEN);
1189         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1190         up_read(&uts_sem);
1191
1192         if (!error && override_architecture(name))
1193                 error = -EFAULT;
1194         return error ? -EFAULT : 0;
1195 }
1196 #endif
1197
1198 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1199 {
1200         int errno;
1201         char tmp[__NEW_UTS_LEN];
1202
1203         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1204                 return -EPERM;
1205
1206         if (len < 0 || len > __NEW_UTS_LEN)
1207                 return -EINVAL;
1208         down_write(&uts_sem);
1209         errno = -EFAULT;
1210         if (!copy_from_user(tmp, name, len)) {
1211                 struct new_utsname *u = utsname();
1212
1213                 memcpy(u->nodename, tmp, len);
1214                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1215                 errno = 0;
1216         }
1217         up_write(&uts_sem);
1218         return errno;
1219 }
1220
1221 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1222
1223 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1224 {
1225         int i, errno;
1226         struct new_utsname *u;
1227
1228         if (len < 0)
1229                 return -EINVAL;
1230         down_read(&uts_sem);
1231         u = utsname();
1232         i = 1 + strlen(u->nodename);
1233         if (i > len)
1234                 i = len;
1235         errno = 0;
1236         if (copy_to_user(name, u->nodename, i))
1237                 errno = -EFAULT;
1238         up_read(&uts_sem);
1239         return errno;
1240 }
1241
1242 #endif
1243
1244 /*
1245  * Only setdomainname; getdomainname can be implemented by calling
1246  * uname()
1247  */
1248 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1249 {
1250         int errno;
1251         char tmp[__NEW_UTS_LEN];
1252
1253         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1254                 return -EPERM;
1255         if (len < 0 || len > __NEW_UTS_LEN)
1256                 return -EINVAL;
1257
1258         down_write(&uts_sem);
1259         errno = -EFAULT;
1260         if (!copy_from_user(tmp, name, len)) {
1261                 struct new_utsname *u = utsname();
1262
1263                 memcpy(u->domainname, tmp, len);
1264                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1265                 errno = 0;
1266         }
1267         up_write(&uts_sem);
1268         return errno;
1269 }
1270
1271 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1272 {
1273         struct rlimit value;
1274         int ret;
1275
1276         ret = do_prlimit(current, resource, NULL, &value);
1277         if (!ret)
1278                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1279
1280         return ret;
1281 }
1282
1283 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1284
1285 /*
1286  *      Back compatibility for getrlimit. Needed for some apps.
1287  */
1288  
1289 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1290                 struct rlimit __user *, rlim)
1291 {
1292         struct rlimit x;
1293         if (resource >= RLIM_NLIMITS)
1294                 return -EINVAL;
1295
1296         task_lock(current->group_leader);
1297         x = current->signal->rlim[resource];
1298         task_unlock(current->group_leader);
1299         if (x.rlim_cur > 0x7FFFFFFF)
1300                 x.rlim_cur = 0x7FFFFFFF;
1301         if (x.rlim_max > 0x7FFFFFFF)
1302                 x.rlim_max = 0x7FFFFFFF;
1303         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1304 }
1305
1306 #endif
1307
1308 static inline bool rlim64_is_infinity(__u64 rlim64)
1309 {
1310 #if BITS_PER_LONG < 64
1311         return rlim64 >= ULONG_MAX;
1312 #else
1313         return rlim64 == RLIM64_INFINITY;
1314 #endif
1315 }
1316
1317 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1318 {
1319         if (rlim->rlim_cur == RLIM_INFINITY)
1320                 rlim64->rlim_cur = RLIM64_INFINITY;
1321         else
1322                 rlim64->rlim_cur = rlim->rlim_cur;
1323         if (rlim->rlim_max == RLIM_INFINITY)
1324                 rlim64->rlim_max = RLIM64_INFINITY;
1325         else
1326                 rlim64->rlim_max = rlim->rlim_max;
1327 }
1328
1329 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1330 {
1331         if (rlim64_is_infinity(rlim64->rlim_cur))
1332                 rlim->rlim_cur = RLIM_INFINITY;
1333         else
1334                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1335         if (rlim64_is_infinity(rlim64->rlim_max))
1336                 rlim->rlim_max = RLIM_INFINITY;
1337         else
1338                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1339 }
1340
1341 /* make sure you are allowed to change @tsk limits before calling this */
1342 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1343                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1344 {
1345         struct rlimit *rlim;
1346         int retval = 0;
1347
1348         if (resource >= RLIM_NLIMITS)
1349                 return -EINVAL;
1350         if (new_rlim) {
1351                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1352                         return -EINVAL;
1353                 if (resource == RLIMIT_NOFILE &&
1354                                 new_rlim->rlim_max > sysctl_nr_open)
1355                         return -EPERM;
1356         }
1357
1358         /* protect tsk->signal and tsk->sighand from disappearing */
1359         read_lock(&tasklist_lock);
1360         if (!tsk->sighand) {
1361                 retval = -ESRCH;
1362                 goto out;
1363         }
1364
1365         rlim = tsk->signal->rlim + resource;
1366         task_lock(tsk->group_leader);
1367         if (new_rlim) {
1368                 /* Keep the capable check against init_user_ns until
1369                    cgroups can contain all limits */
1370                 if (new_rlim->rlim_max > rlim->rlim_max &&
1371                                 !capable(CAP_SYS_RESOURCE))
1372                         retval = -EPERM;
1373                 if (!retval)
1374                         retval = security_task_setrlimit(tsk->group_leader,
1375                                         resource, new_rlim);
1376                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1377                         /*
1378                          * The caller is asking for an immediate RLIMIT_CPU
1379                          * expiry.  But we use the zero value to mean "it was
1380                          * never set".  So let's cheat and make it one second
1381                          * instead
1382                          */
1383                         new_rlim->rlim_cur = 1;
1384                 }
1385         }
1386         if (!retval) {
1387                 if (old_rlim)
1388                         *old_rlim = *rlim;
1389                 if (new_rlim)
1390                         *rlim = *new_rlim;
1391         }
1392         task_unlock(tsk->group_leader);
1393
1394         /*
1395          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1396          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1397          * very long-standing error, and fixing it now risks breakage of
1398          * applications, so we live with it
1399          */
1400          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1401                          new_rlim->rlim_cur != RLIM_INFINITY)
1402                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1403 out:
1404         read_unlock(&tasklist_lock);
1405         return retval;
1406 }
1407
1408 /* rcu lock must be held */
1409 static int check_prlimit_permission(struct task_struct *task)
1410 {
1411         const struct cred *cred = current_cred(), *tcred;
1412
1413         if (current == task)
1414                 return 0;
1415
1416         tcred = __task_cred(task);
1417         if (cred->user->user_ns == tcred->user->user_ns &&
1418             (cred->uid == tcred->euid &&
1419              cred->uid == tcred->suid &&
1420              cred->uid == tcred->uid  &&
1421              cred->gid == tcred->egid &&
1422              cred->gid == tcred->sgid &&
1423              cred->gid == tcred->gid))
1424                 return 0;
1425         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1426                 return 0;
1427
1428         return -EPERM;
1429 }
1430
1431 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1432                 const struct rlimit64 __user *, new_rlim,
1433                 struct rlimit64 __user *, old_rlim)
1434 {
1435         struct rlimit64 old64, new64;
1436         struct rlimit old, new;
1437         struct task_struct *tsk;
1438         int ret;
1439
1440         if (new_rlim) {
1441                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1442                         return -EFAULT;
1443                 rlim64_to_rlim(&new64, &new);
1444         }
1445
1446         rcu_read_lock();
1447         tsk = pid ? find_task_by_vpid(pid) : current;
1448         if (!tsk) {
1449                 rcu_read_unlock();
1450                 return -ESRCH;
1451         }
1452         ret = check_prlimit_permission(tsk);
1453         if (ret) {
1454                 rcu_read_unlock();
1455                 return ret;
1456         }
1457         get_task_struct(tsk);
1458         rcu_read_unlock();
1459
1460         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1461                         old_rlim ? &old : NULL);
1462
1463         if (!ret && old_rlim) {
1464                 rlim_to_rlim64(&old, &old64);
1465                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1466                         ret = -EFAULT;
1467         }
1468
1469         put_task_struct(tsk);
1470         return ret;
1471 }
1472
1473 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1474 {
1475         struct rlimit new_rlim;
1476
1477         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1478                 return -EFAULT;
1479         return do_prlimit(current, resource, &new_rlim, NULL);
1480 }
1481
1482 /*
1483  * It would make sense to put struct rusage in the task_struct,
1484  * except that would make the task_struct be *really big*.  After
1485  * task_struct gets moved into malloc'ed memory, it would
1486  * make sense to do this.  It will make moving the rest of the information
1487  * a lot simpler!  (Which we're not doing right now because we're not
1488  * measuring them yet).
1489  *
1490  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1491  * races with threads incrementing their own counters.  But since word
1492  * reads are atomic, we either get new values or old values and we don't
1493  * care which for the sums.  We always take the siglock to protect reading
1494  * the c* fields from p->signal from races with exit.c updating those
1495  * fields when reaping, so a sample either gets all the additions of a
1496  * given child after it's reaped, or none so this sample is before reaping.
1497  *
1498  * Locking:
1499  * We need to take the siglock for CHILDEREN, SELF and BOTH
1500  * for  the cases current multithreaded, non-current single threaded
1501  * non-current multithreaded.  Thread traversal is now safe with
1502  * the siglock held.
1503  * Strictly speaking, we donot need to take the siglock if we are current and
1504  * single threaded,  as no one else can take our signal_struct away, no one
1505  * else can  reap the  children to update signal->c* counters, and no one else
1506  * can race with the signal-> fields. If we do not take any lock, the
1507  * signal-> fields could be read out of order while another thread was just
1508  * exiting. So we should  place a read memory barrier when we avoid the lock.
1509  * On the writer side,  write memory barrier is implied in  __exit_signal
1510  * as __exit_signal releases  the siglock spinlock after updating the signal->
1511  * fields. But we don't do this yet to keep things simple.
1512  *
1513  */
1514
1515 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1516 {
1517         r->ru_nvcsw += t->nvcsw;
1518         r->ru_nivcsw += t->nivcsw;
1519         r->ru_minflt += t->min_flt;
1520         r->ru_majflt += t->maj_flt;
1521         r->ru_inblock += task_io_get_inblock(t);
1522         r->ru_oublock += task_io_get_oublock(t);
1523 }
1524
1525 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1526 {
1527         struct task_struct *t;
1528         unsigned long flags;
1529         cputime_t tgutime, tgstime, utime, stime;
1530         unsigned long maxrss = 0;
1531
1532         memset((char *) r, 0, sizeof *r);
1533         utime = stime = cputime_zero;
1534
1535         if (who == RUSAGE_THREAD) {
1536                 task_times(current, &utime, &stime);
1537                 accumulate_thread_rusage(p, r);
1538                 maxrss = p->signal->maxrss;
1539                 goto out;
1540         }
1541
1542         if (!lock_task_sighand(p, &flags))
1543                 return;
1544
1545         switch (who) {
1546                 case RUSAGE_BOTH:
1547                 case RUSAGE_CHILDREN:
1548                         utime = p->signal->cutime;
1549                         stime = p->signal->cstime;
1550                         r->ru_nvcsw = p->signal->cnvcsw;
1551                         r->ru_nivcsw = p->signal->cnivcsw;
1552                         r->ru_minflt = p->signal->cmin_flt;
1553                         r->ru_majflt = p->signal->cmaj_flt;
1554                         r->ru_inblock = p->signal->cinblock;
1555                         r->ru_oublock = p->signal->coublock;
1556                         maxrss = p->signal->cmaxrss;
1557
1558                         if (who == RUSAGE_CHILDREN)
1559                                 break;
1560
1561                 case RUSAGE_SELF:
1562                         thread_group_times(p, &tgutime, &tgstime);
1563                         utime = cputime_add(utime, tgutime);
1564                         stime = cputime_add(stime, tgstime);
1565                         r->ru_nvcsw += p->signal->nvcsw;
1566                         r->ru_nivcsw += p->signal->nivcsw;
1567                         r->ru_minflt += p->signal->min_flt;
1568                         r->ru_majflt += p->signal->maj_flt;
1569                         r->ru_inblock += p->signal->inblock;
1570                         r->ru_oublock += p->signal->oublock;
1571                         if (maxrss < p->signal->maxrss)
1572                                 maxrss = p->signal->maxrss;
1573                         t = p;
1574                         do {
1575                                 accumulate_thread_rusage(t, r);
1576                                 t = next_thread(t);
1577                         } while (t != p);
1578                         break;
1579
1580                 default:
1581                         BUG();
1582         }
1583         unlock_task_sighand(p, &flags);
1584
1585 out:
1586         cputime_to_timeval(utime, &r->ru_utime);
1587         cputime_to_timeval(stime, &r->ru_stime);
1588
1589         if (who != RUSAGE_CHILDREN) {
1590                 struct mm_struct *mm = get_task_mm(p);
1591                 if (mm) {
1592                         setmax_mm_hiwater_rss(&maxrss, mm);
1593                         mmput(mm);
1594                 }
1595         }
1596         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1597 }
1598
1599 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1600 {
1601         struct rusage r;
1602         k_getrusage(p, who, &r);
1603         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1604 }
1605
1606 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1607 {
1608         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1609             who != RUSAGE_THREAD)
1610                 return -EINVAL;
1611         return getrusage(current, who, ru);
1612 }
1613
1614 SYSCALL_DEFINE1(umask, int, mask)
1615 {
1616         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1617         return mask;
1618 }
1619
1620 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1621                 unsigned long, arg4, unsigned long, arg5)
1622 {
1623         struct task_struct *me = current;
1624         unsigned char comm[sizeof(me->comm)];
1625         long error;
1626
1627         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1628         if (error != -ENOSYS)
1629                 return error;
1630
1631         error = 0;
1632         switch (option) {
1633                 case PR_SET_PDEATHSIG:
1634                         if (!valid_signal(arg2)) {
1635                                 error = -EINVAL;
1636                                 break;
1637                         }
1638                         me->pdeath_signal = arg2;
1639                         error = 0;
1640                         break;
1641                 case PR_GET_PDEATHSIG:
1642                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1643                         break;
1644                 case PR_GET_DUMPABLE:
1645                         error = get_dumpable(me->mm);
1646                         break;
1647                 case PR_SET_DUMPABLE:
1648                         if (arg2 < 0 || arg2 > 1) {
1649                                 error = -EINVAL;
1650                                 break;
1651                         }
1652                         set_dumpable(me->mm, arg2);
1653                         error = 0;
1654                         break;
1655
1656                 case PR_SET_UNALIGN:
1657                         error = SET_UNALIGN_CTL(me, arg2);
1658                         break;
1659                 case PR_GET_UNALIGN:
1660                         error = GET_UNALIGN_CTL(me, arg2);
1661                         break;
1662                 case PR_SET_FPEMU:
1663                         error = SET_FPEMU_CTL(me, arg2);
1664                         break;
1665                 case PR_GET_FPEMU:
1666                         error = GET_FPEMU_CTL(me, arg2);
1667                         break;
1668                 case PR_SET_FPEXC:
1669                         error = SET_FPEXC_CTL(me, arg2);
1670                         break;
1671                 case PR_GET_FPEXC:
1672                         error = GET_FPEXC_CTL(me, arg2);
1673                         break;
1674                 case PR_GET_TIMING:
1675                         error = PR_TIMING_STATISTICAL;
1676                         break;
1677                 case PR_SET_TIMING:
1678                         if (arg2 != PR_TIMING_STATISTICAL)
1679                                 error = -EINVAL;
1680                         else
1681                                 error = 0;
1682                         break;
1683
1684                 case PR_SET_NAME:
1685                         comm[sizeof(me->comm)-1] = 0;
1686                         if (strncpy_from_user(comm, (char __user *)arg2,
1687                                               sizeof(me->comm) - 1) < 0)
1688                                 return -EFAULT;
1689                         set_task_comm(me, comm);
1690                         return 0;
1691                 case PR_GET_NAME:
1692                         get_task_comm(comm, me);
1693                         if (copy_to_user((char __user *)arg2, comm,
1694                                          sizeof(comm)))
1695                                 return -EFAULT;
1696                         return 0;
1697                 case PR_GET_ENDIAN:
1698                         error = GET_ENDIAN(me, arg2);
1699                         break;
1700                 case PR_SET_ENDIAN:
1701                         error = SET_ENDIAN(me, arg2);
1702                         break;
1703
1704                 case PR_GET_SECCOMP:
1705                         error = prctl_get_seccomp();
1706                         break;
1707                 case PR_SET_SECCOMP:
1708                         error = prctl_set_seccomp(arg2);
1709                         break;
1710                 case PR_GET_TSC:
1711                         error = GET_TSC_CTL(arg2);
1712                         break;
1713                 case PR_SET_TSC:
1714                         error = SET_TSC_CTL(arg2);
1715                         break;
1716                 case PR_TASK_PERF_EVENTS_DISABLE:
1717                         error = perf_event_task_disable();
1718                         break;
1719                 case PR_TASK_PERF_EVENTS_ENABLE:
1720                         error = perf_event_task_enable();
1721                         break;
1722                 case PR_GET_TIMERSLACK:
1723                         error = current->timer_slack_ns;
1724                         break;
1725                 case PR_SET_TIMERSLACK:
1726                         if (arg2 <= 0)
1727                                 current->timer_slack_ns =
1728                                         current->default_timer_slack_ns;
1729                         else
1730                                 current->timer_slack_ns = arg2;
1731                         error = 0;
1732                         break;
1733                 case PR_MCE_KILL:
1734                         if (arg4 | arg5)
1735                                 return -EINVAL;
1736                         switch (arg2) {
1737                         case PR_MCE_KILL_CLEAR:
1738                                 if (arg3 != 0)
1739                                         return -EINVAL;
1740                                 current->flags &= ~PF_MCE_PROCESS;
1741                                 break;
1742                         case PR_MCE_KILL_SET:
1743                                 current->flags |= PF_MCE_PROCESS;
1744                                 if (arg3 == PR_MCE_KILL_EARLY)
1745                                         current->flags |= PF_MCE_EARLY;
1746                                 else if (arg3 == PR_MCE_KILL_LATE)
1747                                         current->flags &= ~PF_MCE_EARLY;
1748                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1749                                         current->flags &=
1750                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1751                                 else
1752                                         return -EINVAL;
1753                                 break;
1754                         default:
1755                                 return -EINVAL;
1756                         }
1757                         error = 0;
1758                         break;
1759                 case PR_MCE_KILL_GET:
1760                         if (arg2 | arg3 | arg4 | arg5)
1761                                 return -EINVAL;
1762                         if (current->flags & PF_MCE_PROCESS)
1763                                 error = (current->flags & PF_MCE_EARLY) ?
1764                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1765                         else
1766                                 error = PR_MCE_KILL_DEFAULT;
1767                         break;
1768                 default:
1769                         error = -EINVAL;
1770                         break;
1771         }
1772         return error;
1773 }
1774
1775 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1776                 struct getcpu_cache __user *, unused)
1777 {
1778         int err = 0;
1779         int cpu = raw_smp_processor_id();
1780         if (cpup)
1781                 err |= put_user(cpu, cpup);
1782         if (nodep)
1783                 err |= put_user(cpu_to_node(cpu), nodep);
1784         return err ? -EFAULT : 0;
1785 }
1786
1787 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1788
1789 static void argv_cleanup(struct subprocess_info *info)
1790 {
1791         argv_free(info->argv);
1792 }
1793
1794 /**
1795  * orderly_poweroff - Trigger an orderly system poweroff
1796  * @force: force poweroff if command execution fails
1797  *
1798  * This may be called from any context to trigger a system shutdown.
1799  * If the orderly shutdown fails, it will force an immediate shutdown.
1800  */
1801 int orderly_poweroff(bool force)
1802 {
1803         int argc;
1804         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1805         static char *envp[] = {
1806                 "HOME=/",
1807                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1808                 NULL
1809         };
1810         int ret = -ENOMEM;
1811         struct subprocess_info *info;
1812
1813         if (argv == NULL) {
1814                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1815                        __func__, poweroff_cmd);
1816                 goto out;
1817         }
1818
1819         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1820         if (info == NULL) {
1821                 argv_free(argv);
1822                 goto out;
1823         }
1824
1825         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1826
1827         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1828
1829   out:
1830         if (ret && force) {
1831                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1832                        "forcing the issue\n");
1833
1834                 /* I guess this should try to kick off some daemon to
1835                    sync and poweroff asap.  Or not even bother syncing
1836                    if we're doing an emergency shutdown? */
1837                 emergency_sync();
1838                 kernel_power_off();
1839         }
1840
1841         return ret;
1842 }
1843 EXPORT_SYMBOL_GPL(orderly_poweroff);