Merge commit 'v2.6.29-rc4' into perfcounters/core
[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/smp_lock.h>
12 #include <linux/notifier.h>
13 #include <linux/reboot.h>
14 #include <linux/prctl.h>
15 #include <linux/highuid.h>
16 #include <linux/fs.h>
17 #include <linux/perf_counter.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/kexec.h>
21 #include <linux/workqueue.h>
22 #include <linux/capability.h>
23 #include <linux/device.h>
24 #include <linux/key.h>
25 #include <linux/times.h>
26 #include <linux/posix-timers.h>
27 #include <linux/security.h>
28 #include <linux/dcookies.h>
29 #include <linux/suspend.h>
30 #include <linux/tty.h>
31 #include <linux/signal.h>
32 #include <linux/cn_proc.h>
33 #include <linux/getcpu.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/seccomp.h>
36 #include <linux/cpu.h>
37 #include <linux/ptrace.h>
38
39 #include <linux/compat.h>
40 #include <linux/syscalls.h>
41 #include <linux/kprobes.h>
42 #include <linux/user_namespace.h>
43
44 #include <asm/uaccess.h>
45 #include <asm/io.h>
46 #include <asm/unistd.h>
47
48 #ifndef SET_UNALIGN_CTL
49 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
50 #endif
51 #ifndef GET_UNALIGN_CTL
52 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
53 #endif
54 #ifndef SET_FPEMU_CTL
55 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
56 #endif
57 #ifndef GET_FPEMU_CTL
58 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
59 #endif
60 #ifndef SET_FPEXC_CTL
61 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
62 #endif
63 #ifndef GET_FPEXC_CTL
64 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
65 #endif
66 #ifndef GET_ENDIAN
67 # define GET_ENDIAN(a,b)        (-EINVAL)
68 #endif
69 #ifndef SET_ENDIAN
70 # define SET_ENDIAN(a,b)        (-EINVAL)
71 #endif
72 #ifndef GET_TSC_CTL
73 # define GET_TSC_CTL(a)         (-EINVAL)
74 #endif
75 #ifndef SET_TSC_CTL
76 # define SET_TSC_CTL(a)         (-EINVAL)
77 #endif
78
79 /*
80  * this is where the system-wide overflow UID and GID are defined, for
81  * architectures that now have 32-bit UID/GID but didn't in the past
82  */
83
84 int overflowuid = DEFAULT_OVERFLOWUID;
85 int overflowgid = DEFAULT_OVERFLOWGID;
86
87 #ifdef CONFIG_UID16
88 EXPORT_SYMBOL(overflowuid);
89 EXPORT_SYMBOL(overflowgid);
90 #endif
91
92 /*
93  * the same as above, but for filesystems which can only store a 16-bit
94  * UID and GID. as such, this is needed on all architectures
95  */
96
97 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
98 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
99
100 EXPORT_SYMBOL(fs_overflowuid);
101 EXPORT_SYMBOL(fs_overflowgid);
102
103 /*
104  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
105  */
106
107 int C_A_D = 1;
108 struct pid *cad_pid;
109 EXPORT_SYMBOL(cad_pid);
110
111 /*
112  * If set, this is used for preparing the system to power off.
113  */
114
115 void (*pm_power_off_prepare)(void);
116
117 /*
118  * set the priority of a task
119  * - the caller must hold the RCU read lock
120  */
121 static int set_one_prio(struct task_struct *p, int niceval, int error)
122 {
123         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
124         int no_nice;
125
126         if (pcred->uid  != cred->euid &&
127             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
128                 error = -EPERM;
129                 goto out;
130         }
131         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
132                 error = -EACCES;
133                 goto out;
134         }
135         no_nice = security_task_setnice(p, niceval);
136         if (no_nice) {
137                 error = no_nice;
138                 goto out;
139         }
140         if (error == -ESRCH)
141                 error = 0;
142         set_user_nice(p, niceval);
143 out:
144         return error;
145 }
146
147 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
148 {
149         struct task_struct *g, *p;
150         struct user_struct *user;
151         const struct cred *cred = current_cred();
152         int error = -EINVAL;
153         struct pid *pgrp;
154
155         if (which > PRIO_USER || which < PRIO_PROCESS)
156                 goto out;
157
158         /* normalize: avoid signed division (rounding problems) */
159         error = -ESRCH;
160         if (niceval < -20)
161                 niceval = -20;
162         if (niceval > 19)
163                 niceval = 19;
164
165         read_lock(&tasklist_lock);
166         switch (which) {
167                 case PRIO_PROCESS:
168                         if (who)
169                                 p = find_task_by_vpid(who);
170                         else
171                                 p = current;
172                         if (p)
173                                 error = set_one_prio(p, niceval, error);
174                         break;
175                 case PRIO_PGRP:
176                         if (who)
177                                 pgrp = find_vpid(who);
178                         else
179                                 pgrp = task_pgrp(current);
180                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
181                                 error = set_one_prio(p, niceval, error);
182                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
183                         break;
184                 case PRIO_USER:
185                         user = (struct user_struct *) cred->user;
186                         if (!who)
187                                 who = cred->uid;
188                         else if ((who != cred->uid) &&
189                                  !(user = find_user(who)))
190                                 goto out_unlock;        /* No processes for this user */
191
192                         do_each_thread(g, p)
193                                 if (__task_cred(p)->uid == who)
194                                         error = set_one_prio(p, niceval, error);
195                         while_each_thread(g, p);
196                         if (who != cred->uid)
197                                 free_uid(user);         /* For find_user() */
198                         break;
199         }
200 out_unlock:
201         read_unlock(&tasklist_lock);
202 out:
203         return error;
204 }
205
206 /*
207  * Ugh. To avoid negative return values, "getpriority()" will
208  * not return the normal nice-value, but a negated value that
209  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
210  * to stay compatible.
211  */
212 SYSCALL_DEFINE2(getpriority, int, which, int, who)
213 {
214         struct task_struct *g, *p;
215         struct user_struct *user;
216         const struct cred *cred = current_cred();
217         long niceval, retval = -ESRCH;
218         struct pid *pgrp;
219
220         if (which > PRIO_USER || which < PRIO_PROCESS)
221                 return -EINVAL;
222
223         read_lock(&tasklist_lock);
224         switch (which) {
225                 case PRIO_PROCESS:
226                         if (who)
227                                 p = find_task_by_vpid(who);
228                         else
229                                 p = current;
230                         if (p) {
231                                 niceval = 20 - task_nice(p);
232                                 if (niceval > retval)
233                                         retval = niceval;
234                         }
235                         break;
236                 case PRIO_PGRP:
237                         if (who)
238                                 pgrp = find_vpid(who);
239                         else
240                                 pgrp = task_pgrp(current);
241                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
242                                 niceval = 20 - task_nice(p);
243                                 if (niceval > retval)
244                                         retval = niceval;
245                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
246                         break;
247                 case PRIO_USER:
248                         user = (struct user_struct *) cred->user;
249                         if (!who)
250                                 who = cred->uid;
251                         else if ((who != cred->uid) &&
252                                  !(user = find_user(who)))
253                                 goto out_unlock;        /* No processes for this user */
254
255                         do_each_thread(g, p)
256                                 if (__task_cred(p)->uid == who) {
257                                         niceval = 20 - task_nice(p);
258                                         if (niceval > retval)
259                                                 retval = niceval;
260                                 }
261                         while_each_thread(g, p);
262                         if (who != cred->uid)
263                                 free_uid(user);         /* for find_user() */
264                         break;
265         }
266 out_unlock:
267         read_unlock(&tasklist_lock);
268
269         return retval;
270 }
271
272 /**
273  *      emergency_restart - reboot the system
274  *
275  *      Without shutting down any hardware or taking any locks
276  *      reboot the system.  This is called when we know we are in
277  *      trouble so this is our best effort to reboot.  This is
278  *      safe to call in interrupt context.
279  */
280 void emergency_restart(void)
281 {
282         machine_emergency_restart();
283 }
284 EXPORT_SYMBOL_GPL(emergency_restart);
285
286 void kernel_restart_prepare(char *cmd)
287 {
288         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
289         system_state = SYSTEM_RESTART;
290         device_shutdown();
291         sysdev_shutdown();
292 }
293
294 /**
295  *      kernel_restart - reboot the system
296  *      @cmd: pointer to buffer containing command to execute for restart
297  *              or %NULL
298  *
299  *      Shutdown everything and perform a clean reboot.
300  *      This is not safe to call in interrupt context.
301  */
302 void kernel_restart(char *cmd)
303 {
304         kernel_restart_prepare(cmd);
305         if (!cmd)
306                 printk(KERN_EMERG "Restarting system.\n");
307         else
308                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
309         machine_restart(cmd);
310 }
311 EXPORT_SYMBOL_GPL(kernel_restart);
312
313 static void kernel_shutdown_prepare(enum system_states state)
314 {
315         blocking_notifier_call_chain(&reboot_notifier_list,
316                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
317         system_state = state;
318         device_shutdown();
319 }
320 /**
321  *      kernel_halt - halt the system
322  *
323  *      Shutdown everything and perform a clean system halt.
324  */
325 void kernel_halt(void)
326 {
327         kernel_shutdown_prepare(SYSTEM_HALT);
328         sysdev_shutdown();
329         printk(KERN_EMERG "System halted.\n");
330         machine_halt();
331 }
332
333 EXPORT_SYMBOL_GPL(kernel_halt);
334
335 /**
336  *      kernel_power_off - power_off the system
337  *
338  *      Shutdown everything and perform a clean system power_off.
339  */
340 void kernel_power_off(void)
341 {
342         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
343         if (pm_power_off_prepare)
344                 pm_power_off_prepare();
345         disable_nonboot_cpus();
346         sysdev_shutdown();
347         printk(KERN_EMERG "Power down.\n");
348         machine_power_off();
349 }
350 EXPORT_SYMBOL_GPL(kernel_power_off);
351 /*
352  * Reboot system call: for obvious reasons only root may call it,
353  * and even root needs to set up some magic numbers in the registers
354  * so that some mistake won't make this reboot the whole machine.
355  * You can also set the meaning of the ctrl-alt-del-key here.
356  *
357  * reboot doesn't sync: do that yourself before calling this.
358  */
359 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
360                 void __user *, arg)
361 {
362         char buffer[256];
363
364         /* We only trust the superuser with rebooting the system. */
365         if (!capable(CAP_SYS_BOOT))
366                 return -EPERM;
367
368         /* For safety, we require "magic" arguments. */
369         if (magic1 != LINUX_REBOOT_MAGIC1 ||
370             (magic2 != LINUX_REBOOT_MAGIC2 &&
371                         magic2 != LINUX_REBOOT_MAGIC2A &&
372                         magic2 != LINUX_REBOOT_MAGIC2B &&
373                         magic2 != LINUX_REBOOT_MAGIC2C))
374                 return -EINVAL;
375
376         /* Instead of trying to make the power_off code look like
377          * halt when pm_power_off is not set do it the easy way.
378          */
379         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
380                 cmd = LINUX_REBOOT_CMD_HALT;
381
382         lock_kernel();
383         switch (cmd) {
384         case LINUX_REBOOT_CMD_RESTART:
385                 kernel_restart(NULL);
386                 break;
387
388         case LINUX_REBOOT_CMD_CAD_ON:
389                 C_A_D = 1;
390                 break;
391
392         case LINUX_REBOOT_CMD_CAD_OFF:
393                 C_A_D = 0;
394                 break;
395
396         case LINUX_REBOOT_CMD_HALT:
397                 kernel_halt();
398                 unlock_kernel();
399                 do_exit(0);
400                 break;
401
402         case LINUX_REBOOT_CMD_POWER_OFF:
403                 kernel_power_off();
404                 unlock_kernel();
405                 do_exit(0);
406                 break;
407
408         case LINUX_REBOOT_CMD_RESTART2:
409                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
410                         unlock_kernel();
411                         return -EFAULT;
412                 }
413                 buffer[sizeof(buffer) - 1] = '\0';
414
415                 kernel_restart(buffer);
416                 break;
417
418 #ifdef CONFIG_KEXEC
419         case LINUX_REBOOT_CMD_KEXEC:
420                 {
421                         int ret;
422                         ret = kernel_kexec();
423                         unlock_kernel();
424                         return ret;
425                 }
426 #endif
427
428 #ifdef CONFIG_HIBERNATION
429         case LINUX_REBOOT_CMD_SW_SUSPEND:
430                 {
431                         int ret = hibernate();
432                         unlock_kernel();
433                         return ret;
434                 }
435 #endif
436
437         default:
438                 unlock_kernel();
439                 return -EINVAL;
440         }
441         unlock_kernel();
442         return 0;
443 }
444
445 static void deferred_cad(struct work_struct *dummy)
446 {
447         kernel_restart(NULL);
448 }
449
450 /*
451  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
452  * As it's called within an interrupt, it may NOT sync: the only choice
453  * is whether to reboot at once, or just ignore the ctrl-alt-del.
454  */
455 void ctrl_alt_del(void)
456 {
457         static DECLARE_WORK(cad_work, deferred_cad);
458
459         if (C_A_D)
460                 schedule_work(&cad_work);
461         else
462                 kill_cad_pid(SIGINT, 1);
463 }
464         
465 /*
466  * Unprivileged users may change the real gid to the effective gid
467  * or vice versa.  (BSD-style)
468  *
469  * If you set the real gid at all, or set the effective gid to a value not
470  * equal to the real gid, then the saved gid is set to the new effective gid.
471  *
472  * This makes it possible for a setgid program to completely drop its
473  * privileges, which is often a useful assertion to make when you are doing
474  * a security audit over a program.
475  *
476  * The general idea is that a program which uses just setregid() will be
477  * 100% compatible with BSD.  A program which uses just setgid() will be
478  * 100% compatible with POSIX with saved IDs. 
479  *
480  * SMP: There are not races, the GIDs are checked only by filesystem
481  *      operations (as far as semantic preservation is concerned).
482  */
483 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
484 {
485         const struct cred *old;
486         struct cred *new;
487         int retval;
488
489         new = prepare_creds();
490         if (!new)
491                 return -ENOMEM;
492         old = current_cred();
493
494         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
495         if (retval)
496                 goto error;
497
498         retval = -EPERM;
499         if (rgid != (gid_t) -1) {
500                 if (old->gid == rgid ||
501                     old->egid == rgid ||
502                     capable(CAP_SETGID))
503                         new->gid = rgid;
504                 else
505                         goto error;
506         }
507         if (egid != (gid_t) -1) {
508                 if (old->gid == egid ||
509                     old->egid == egid ||
510                     old->sgid == egid ||
511                     capable(CAP_SETGID))
512                         new->egid = egid;
513                 else
514                         goto error;
515         }
516
517         if (rgid != (gid_t) -1 ||
518             (egid != (gid_t) -1 && egid != old->gid))
519                 new->sgid = new->egid;
520         new->fsgid = new->egid;
521
522         return commit_creds(new);
523
524 error:
525         abort_creds(new);
526         return retval;
527 }
528
529 /*
530  * setgid() is implemented like SysV w/ SAVED_IDS 
531  *
532  * SMP: Same implicit races as above.
533  */
534 SYSCALL_DEFINE1(setgid, gid_t, gid)
535 {
536         const struct cred *old;
537         struct cred *new;
538         int retval;
539
540         new = prepare_creds();
541         if (!new)
542                 return -ENOMEM;
543         old = current_cred();
544
545         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
546         if (retval)
547                 goto error;
548
549         retval = -EPERM;
550         if (capable(CAP_SETGID))
551                 new->gid = new->egid = new->sgid = new->fsgid = gid;
552         else if (gid == old->gid || gid == old->sgid)
553                 new->egid = new->fsgid = gid;
554         else
555                 goto error;
556
557         return commit_creds(new);
558
559 error:
560         abort_creds(new);
561         return retval;
562 }
563   
564 /*
565  * change the user struct in a credentials set to match the new UID
566  */
567 static int set_user(struct cred *new)
568 {
569         struct user_struct *new_user;
570
571         new_user = alloc_uid(current_user_ns(), new->uid);
572         if (!new_user)
573                 return -EAGAIN;
574
575         if (atomic_read(&new_user->processes) >=
576                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
577                         new_user != INIT_USER) {
578                 free_uid(new_user);
579                 return -EAGAIN;
580         }
581
582         free_uid(new->user);
583         new->user = new_user;
584         return 0;
585 }
586
587 /*
588  * Unprivileged users may change the real uid to the effective uid
589  * or vice versa.  (BSD-style)
590  *
591  * If you set the real uid at all, or set the effective uid to a value not
592  * equal to the real uid, then the saved uid is set to the new effective uid.
593  *
594  * This makes it possible for a setuid program to completely drop its
595  * privileges, which is often a useful assertion to make when you are doing
596  * a security audit over a program.
597  *
598  * The general idea is that a program which uses just setreuid() will be
599  * 100% compatible with BSD.  A program which uses just setuid() will be
600  * 100% compatible with POSIX with saved IDs. 
601  */
602 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
603 {
604         const struct cred *old;
605         struct cred *new;
606         int retval;
607
608         new = prepare_creds();
609         if (!new)
610                 return -ENOMEM;
611         old = current_cred();
612
613         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
614         if (retval)
615                 goto error;
616
617         retval = -EPERM;
618         if (ruid != (uid_t) -1) {
619                 new->uid = ruid;
620                 if (old->uid != ruid &&
621                     old->euid != ruid &&
622                     !capable(CAP_SETUID))
623                         goto error;
624         }
625
626         if (euid != (uid_t) -1) {
627                 new->euid = euid;
628                 if (old->uid != euid &&
629                     old->euid != euid &&
630                     old->suid != euid &&
631                     !capable(CAP_SETUID))
632                         goto error;
633         }
634
635         retval = -EAGAIN;
636         if (new->uid != old->uid && set_user(new) < 0)
637                 goto error;
638
639         if (ruid != (uid_t) -1 ||
640             (euid != (uid_t) -1 && euid != old->uid))
641                 new->suid = new->euid;
642         new->fsuid = new->euid;
643
644         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
645         if (retval < 0)
646                 goto error;
647
648         return commit_creds(new);
649
650 error:
651         abort_creds(new);
652         return retval;
653 }
654                 
655 /*
656  * setuid() is implemented like SysV with SAVED_IDS 
657  * 
658  * Note that SAVED_ID's is deficient in that a setuid root program
659  * like sendmail, for example, cannot set its uid to be a normal 
660  * user and then switch back, because if you're root, setuid() sets
661  * the saved uid too.  If you don't like this, blame the bright people
662  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
663  * will allow a root program to temporarily drop privileges and be able to
664  * regain them by swapping the real and effective uid.  
665  */
666 SYSCALL_DEFINE1(setuid, uid_t, uid)
667 {
668         const struct cred *old;
669         struct cred *new;
670         int retval;
671
672         new = prepare_creds();
673         if (!new)
674                 return -ENOMEM;
675         old = current_cred();
676
677         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
678         if (retval)
679                 goto error;
680
681         retval = -EPERM;
682         if (capable(CAP_SETUID)) {
683                 new->suid = new->uid = uid;
684                 if (uid != old->uid && set_user(new) < 0) {
685                         retval = -EAGAIN;
686                         goto error;
687                 }
688         } else if (uid != old->uid && uid != new->suid) {
689                 goto error;
690         }
691
692         new->fsuid = new->euid = uid;
693
694         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
695         if (retval < 0)
696                 goto error;
697
698         return commit_creds(new);
699
700 error:
701         abort_creds(new);
702         return retval;
703 }
704
705
706 /*
707  * This function implements a generic ability to update ruid, euid,
708  * and suid.  This allows you to implement the 4.4 compatible seteuid().
709  */
710 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
711 {
712         const struct cred *old;
713         struct cred *new;
714         int retval;
715
716         new = prepare_creds();
717         if (!new)
718                 return -ENOMEM;
719
720         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
721         if (retval)
722                 goto error;
723         old = current_cred();
724
725         retval = -EPERM;
726         if (!capable(CAP_SETUID)) {
727                 if (ruid != (uid_t) -1 && ruid != old->uid &&
728                     ruid != old->euid  && ruid != old->suid)
729                         goto error;
730                 if (euid != (uid_t) -1 && euid != old->uid &&
731                     euid != old->euid  && euid != old->suid)
732                         goto error;
733                 if (suid != (uid_t) -1 && suid != old->uid &&
734                     suid != old->euid  && suid != old->suid)
735                         goto error;
736         }
737
738         retval = -EAGAIN;
739         if (ruid != (uid_t) -1) {
740                 new->uid = ruid;
741                 if (ruid != old->uid && set_user(new) < 0)
742                         goto error;
743         }
744         if (euid != (uid_t) -1)
745                 new->euid = euid;
746         if (suid != (uid_t) -1)
747                 new->suid = suid;
748         new->fsuid = new->euid;
749
750         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
751         if (retval < 0)
752                 goto error;
753
754         return commit_creds(new);
755
756 error:
757         abort_creds(new);
758         return retval;
759 }
760
761 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
762 {
763         const struct cred *cred = current_cred();
764         int retval;
765
766         if (!(retval   = put_user(cred->uid,  ruid)) &&
767             !(retval   = put_user(cred->euid, euid)))
768                 retval = put_user(cred->suid, suid);
769
770         return retval;
771 }
772
773 /*
774  * Same as above, but for rgid, egid, sgid.
775  */
776 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
777 {
778         const struct cred *old;
779         struct cred *new;
780         int retval;
781
782         new = prepare_creds();
783         if (!new)
784                 return -ENOMEM;
785         old = current_cred();
786
787         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
788         if (retval)
789                 goto error;
790
791         retval = -EPERM;
792         if (!capable(CAP_SETGID)) {
793                 if (rgid != (gid_t) -1 && rgid != old->gid &&
794                     rgid != old->egid  && rgid != old->sgid)
795                         goto error;
796                 if (egid != (gid_t) -1 && egid != old->gid &&
797                     egid != old->egid  && egid != old->sgid)
798                         goto error;
799                 if (sgid != (gid_t) -1 && sgid != old->gid &&
800                     sgid != old->egid  && sgid != old->sgid)
801                         goto error;
802         }
803
804         if (rgid != (gid_t) -1)
805                 new->gid = rgid;
806         if (egid != (gid_t) -1)
807                 new->egid = egid;
808         if (sgid != (gid_t) -1)
809                 new->sgid = sgid;
810         new->fsgid = new->egid;
811
812         return commit_creds(new);
813
814 error:
815         abort_creds(new);
816         return retval;
817 }
818
819 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
820 {
821         const struct cred *cred = current_cred();
822         int retval;
823
824         if (!(retval   = put_user(cred->gid,  rgid)) &&
825             !(retval   = put_user(cred->egid, egid)))
826                 retval = put_user(cred->sgid, sgid);
827
828         return retval;
829 }
830
831
832 /*
833  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
834  * is used for "access()" and for the NFS daemon (letting nfsd stay at
835  * whatever uid it wants to). It normally shadows "euid", except when
836  * explicitly set by setfsuid() or for access..
837  */
838 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
839 {
840         const struct cred *old;
841         struct cred *new;
842         uid_t old_fsuid;
843
844         new = prepare_creds();
845         if (!new)
846                 return current_fsuid();
847         old = current_cred();
848         old_fsuid = old->fsuid;
849
850         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
851                 goto error;
852
853         if (uid == old->uid  || uid == old->euid  ||
854             uid == old->suid || uid == old->fsuid ||
855             capable(CAP_SETUID)) {
856                 if (uid != old_fsuid) {
857                         new->fsuid = uid;
858                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
859                                 goto change_okay;
860                 }
861         }
862
863 error:
864         abort_creds(new);
865         return old_fsuid;
866
867 change_okay:
868         commit_creds(new);
869         return old_fsuid;
870 }
871
872 /*
873  * Samma på svenska..
874  */
875 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
876 {
877         const struct cred *old;
878         struct cred *new;
879         gid_t old_fsgid;
880
881         new = prepare_creds();
882         if (!new)
883                 return current_fsgid();
884         old = current_cred();
885         old_fsgid = old->fsgid;
886
887         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
888                 goto error;
889
890         if (gid == old->gid  || gid == old->egid  ||
891             gid == old->sgid || gid == old->fsgid ||
892             capable(CAP_SETGID)) {
893                 if (gid != old_fsgid) {
894                         new->fsgid = gid;
895                         goto change_okay;
896                 }
897         }
898
899 error:
900         abort_creds(new);
901         return old_fsgid;
902
903 change_okay:
904         commit_creds(new);
905         return old_fsgid;
906 }
907
908 void do_sys_times(struct tms *tms)
909 {
910         struct task_cputime cputime;
911         cputime_t cutime, cstime;
912
913         thread_group_cputime(current, &cputime);
914         spin_lock_irq(&current->sighand->siglock);
915         cutime = current->signal->cutime;
916         cstime = current->signal->cstime;
917         spin_unlock_irq(&current->sighand->siglock);
918         tms->tms_utime = cputime_to_clock_t(cputime.utime);
919         tms->tms_stime = cputime_to_clock_t(cputime.stime);
920         tms->tms_cutime = cputime_to_clock_t(cutime);
921         tms->tms_cstime = cputime_to_clock_t(cstime);
922 }
923
924 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
925 {
926         if (tbuf) {
927                 struct tms tmp;
928
929                 do_sys_times(&tmp);
930                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
931                         return -EFAULT;
932         }
933         force_successful_syscall_return();
934         return (long) jiffies_64_to_clock_t(get_jiffies_64());
935 }
936
937 /*
938  * This needs some heavy checking ...
939  * I just haven't the stomach for it. I also don't fully
940  * understand sessions/pgrp etc. Let somebody who does explain it.
941  *
942  * OK, I think I have the protection semantics right.... this is really
943  * only important on a multi-user system anyway, to make sure one user
944  * can't send a signal to a process owned by another.  -TYT, 12/12/91
945  *
946  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
947  * LBT 04.03.94
948  */
949 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
950 {
951         struct task_struct *p;
952         struct task_struct *group_leader = current->group_leader;
953         struct pid *pgrp;
954         int err;
955
956         if (!pid)
957                 pid = task_pid_vnr(group_leader);
958         if (!pgid)
959                 pgid = pid;
960         if (pgid < 0)
961                 return -EINVAL;
962
963         /* From this point forward we keep holding onto the tasklist lock
964          * so that our parent does not change from under us. -DaveM
965          */
966         write_lock_irq(&tasklist_lock);
967
968         err = -ESRCH;
969         p = find_task_by_vpid(pid);
970         if (!p)
971                 goto out;
972
973         err = -EINVAL;
974         if (!thread_group_leader(p))
975                 goto out;
976
977         if (same_thread_group(p->real_parent, group_leader)) {
978                 err = -EPERM;
979                 if (task_session(p) != task_session(group_leader))
980                         goto out;
981                 err = -EACCES;
982                 if (p->did_exec)
983                         goto out;
984         } else {
985                 err = -ESRCH;
986                 if (p != group_leader)
987                         goto out;
988         }
989
990         err = -EPERM;
991         if (p->signal->leader)
992                 goto out;
993
994         pgrp = task_pid(p);
995         if (pgid != pid) {
996                 struct task_struct *g;
997
998                 pgrp = find_vpid(pgid);
999                 g = pid_task(pgrp, PIDTYPE_PGID);
1000                 if (!g || task_session(g) != task_session(group_leader))
1001                         goto out;
1002         }
1003
1004         err = security_task_setpgid(p, pgid);
1005         if (err)
1006                 goto out;
1007
1008         if (task_pgrp(p) != pgrp) {
1009                 change_pid(p, PIDTYPE_PGID, pgrp);
1010                 set_task_pgrp(p, pid_nr(pgrp));
1011         }
1012
1013         err = 0;
1014 out:
1015         /* All paths lead to here, thus we are safe. -DaveM */
1016         write_unlock_irq(&tasklist_lock);
1017         return err;
1018 }
1019
1020 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1021 {
1022         struct task_struct *p;
1023         struct pid *grp;
1024         int retval;
1025
1026         rcu_read_lock();
1027         if (!pid)
1028                 grp = task_pgrp(current);
1029         else {
1030                 retval = -ESRCH;
1031                 p = find_task_by_vpid(pid);
1032                 if (!p)
1033                         goto out;
1034                 grp = task_pgrp(p);
1035                 if (!grp)
1036                         goto out;
1037
1038                 retval = security_task_getpgid(p);
1039                 if (retval)
1040                         goto out;
1041         }
1042         retval = pid_vnr(grp);
1043 out:
1044         rcu_read_unlock();
1045         return retval;
1046 }
1047
1048 #ifdef __ARCH_WANT_SYS_GETPGRP
1049
1050 SYSCALL_DEFINE0(getpgrp)
1051 {
1052         return sys_getpgid(0);
1053 }
1054
1055 #endif
1056
1057 SYSCALL_DEFINE1(getsid, pid_t, pid)
1058 {
1059         struct task_struct *p;
1060         struct pid *sid;
1061         int retval;
1062
1063         rcu_read_lock();
1064         if (!pid)
1065                 sid = task_session(current);
1066         else {
1067                 retval = -ESRCH;
1068                 p = find_task_by_vpid(pid);
1069                 if (!p)
1070                         goto out;
1071                 sid = task_session(p);
1072                 if (!sid)
1073                         goto out;
1074
1075                 retval = security_task_getsid(p);
1076                 if (retval)
1077                         goto out;
1078         }
1079         retval = pid_vnr(sid);
1080 out:
1081         rcu_read_unlock();
1082         return retval;
1083 }
1084
1085 SYSCALL_DEFINE0(setsid)
1086 {
1087         struct task_struct *group_leader = current->group_leader;
1088         struct pid *sid = task_pid(group_leader);
1089         pid_t session = pid_vnr(sid);
1090         int err = -EPERM;
1091
1092         write_lock_irq(&tasklist_lock);
1093         /* Fail if I am already a session leader */
1094         if (group_leader->signal->leader)
1095                 goto out;
1096
1097         /* Fail if a process group id already exists that equals the
1098          * proposed session id.
1099          */
1100         if (pid_task(sid, PIDTYPE_PGID))
1101                 goto out;
1102
1103         group_leader->signal->leader = 1;
1104         __set_special_pids(sid);
1105
1106         proc_clear_tty(group_leader);
1107
1108         err = session;
1109 out:
1110         write_unlock_irq(&tasklist_lock);
1111         return err;
1112 }
1113
1114 /*
1115  * Supplementary group IDs
1116  */
1117
1118 /* init to 2 - one for init_task, one to ensure it is never freed */
1119 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1120
1121 struct group_info *groups_alloc(int gidsetsize)
1122 {
1123         struct group_info *group_info;
1124         int nblocks;
1125         int i;
1126
1127         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1128         /* Make sure we always allocate at least one indirect block pointer */
1129         nblocks = nblocks ? : 1;
1130         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1131         if (!group_info)
1132                 return NULL;
1133         group_info->ngroups = gidsetsize;
1134         group_info->nblocks = nblocks;
1135         atomic_set(&group_info->usage, 1);
1136
1137         if (gidsetsize <= NGROUPS_SMALL)
1138                 group_info->blocks[0] = group_info->small_block;
1139         else {
1140                 for (i = 0; i < nblocks; i++) {
1141                         gid_t *b;
1142                         b = (void *)__get_free_page(GFP_USER);
1143                         if (!b)
1144                                 goto out_undo_partial_alloc;
1145                         group_info->blocks[i] = b;
1146                 }
1147         }
1148         return group_info;
1149
1150 out_undo_partial_alloc:
1151         while (--i >= 0) {
1152                 free_page((unsigned long)group_info->blocks[i]);
1153         }
1154         kfree(group_info);
1155         return NULL;
1156 }
1157
1158 EXPORT_SYMBOL(groups_alloc);
1159
1160 void groups_free(struct group_info *group_info)
1161 {
1162         if (group_info->blocks[0] != group_info->small_block) {
1163                 int i;
1164                 for (i = 0; i < group_info->nblocks; i++)
1165                         free_page((unsigned long)group_info->blocks[i]);
1166         }
1167         kfree(group_info);
1168 }
1169
1170 EXPORT_SYMBOL(groups_free);
1171
1172 /* export the group_info to a user-space array */
1173 static int groups_to_user(gid_t __user *grouplist,
1174                           const struct group_info *group_info)
1175 {
1176         int i;
1177         unsigned int count = group_info->ngroups;
1178
1179         for (i = 0; i < group_info->nblocks; i++) {
1180                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1181                 unsigned int len = cp_count * sizeof(*grouplist);
1182
1183                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1184                         return -EFAULT;
1185
1186                 grouplist += NGROUPS_PER_BLOCK;
1187                 count -= cp_count;
1188         }
1189         return 0;
1190 }
1191
1192 /* fill a group_info from a user-space array - it must be allocated already */
1193 static int groups_from_user(struct group_info *group_info,
1194     gid_t __user *grouplist)
1195 {
1196         int i;
1197         unsigned int count = group_info->ngroups;
1198
1199         for (i = 0; i < group_info->nblocks; i++) {
1200                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1201                 unsigned int len = cp_count * sizeof(*grouplist);
1202
1203                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1204                         return -EFAULT;
1205
1206                 grouplist += NGROUPS_PER_BLOCK;
1207                 count -= cp_count;
1208         }
1209         return 0;
1210 }
1211
1212 /* a simple Shell sort */
1213 static void groups_sort(struct group_info *group_info)
1214 {
1215         int base, max, stride;
1216         int gidsetsize = group_info->ngroups;
1217
1218         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1219                 ; /* nothing */
1220         stride /= 3;
1221
1222         while (stride) {
1223                 max = gidsetsize - stride;
1224                 for (base = 0; base < max; base++) {
1225                         int left = base;
1226                         int right = left + stride;
1227                         gid_t tmp = GROUP_AT(group_info, right);
1228
1229                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1230                                 GROUP_AT(group_info, right) =
1231                                     GROUP_AT(group_info, left);
1232                                 right = left;
1233                                 left -= stride;
1234                         }
1235                         GROUP_AT(group_info, right) = tmp;
1236                 }
1237                 stride /= 3;
1238         }
1239 }
1240
1241 /* a simple bsearch */
1242 int groups_search(const struct group_info *group_info, gid_t grp)
1243 {
1244         unsigned int left, right;
1245
1246         if (!group_info)
1247                 return 0;
1248
1249         left = 0;
1250         right = group_info->ngroups;
1251         while (left < right) {
1252                 unsigned int mid = (left+right)/2;
1253                 int cmp = grp - GROUP_AT(group_info, mid);
1254                 if (cmp > 0)
1255                         left = mid + 1;
1256                 else if (cmp < 0)
1257                         right = mid;
1258                 else
1259                         return 1;
1260         }
1261         return 0;
1262 }
1263
1264 /**
1265  * set_groups - Change a group subscription in a set of credentials
1266  * @new: The newly prepared set of credentials to alter
1267  * @group_info: The group list to install
1268  *
1269  * Validate a group subscription and, if valid, insert it into a set
1270  * of credentials.
1271  */
1272 int set_groups(struct cred *new, struct group_info *group_info)
1273 {
1274         int retval;
1275
1276         retval = security_task_setgroups(group_info);
1277         if (retval)
1278                 return retval;
1279
1280         put_group_info(new->group_info);
1281         groups_sort(group_info);
1282         get_group_info(group_info);
1283         new->group_info = group_info;
1284         return 0;
1285 }
1286
1287 EXPORT_SYMBOL(set_groups);
1288
1289 /**
1290  * set_current_groups - Change current's group subscription
1291  * @group_info: The group list to impose
1292  *
1293  * Validate a group subscription and, if valid, impose it upon current's task
1294  * security record.
1295  */
1296 int set_current_groups(struct group_info *group_info)
1297 {
1298         struct cred *new;
1299         int ret;
1300
1301         new = prepare_creds();
1302         if (!new)
1303                 return -ENOMEM;
1304
1305         ret = set_groups(new, group_info);
1306         if (ret < 0) {
1307                 abort_creds(new);
1308                 return ret;
1309         }
1310
1311         return commit_creds(new);
1312 }
1313
1314 EXPORT_SYMBOL(set_current_groups);
1315
1316 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1317 {
1318         const struct cred *cred = current_cred();
1319         int i;
1320
1321         if (gidsetsize < 0)
1322                 return -EINVAL;
1323
1324         /* no need to grab task_lock here; it cannot change */
1325         i = cred->group_info->ngroups;
1326         if (gidsetsize) {
1327                 if (i > gidsetsize) {
1328                         i = -EINVAL;
1329                         goto out;
1330                 }
1331                 if (groups_to_user(grouplist, cred->group_info)) {
1332                         i = -EFAULT;
1333                         goto out;
1334                 }
1335         }
1336 out:
1337         return i;
1338 }
1339
1340 /*
1341  *      SMP: Our groups are copy-on-write. We can set them safely
1342  *      without another task interfering.
1343  */
1344  
1345 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1346 {
1347         struct group_info *group_info;
1348         int retval;
1349
1350         if (!capable(CAP_SETGID))
1351                 return -EPERM;
1352         if ((unsigned)gidsetsize > NGROUPS_MAX)
1353                 return -EINVAL;
1354
1355         group_info = groups_alloc(gidsetsize);
1356         if (!group_info)
1357                 return -ENOMEM;
1358         retval = groups_from_user(group_info, grouplist);
1359         if (retval) {
1360                 put_group_info(group_info);
1361                 return retval;
1362         }
1363
1364         retval = set_current_groups(group_info);
1365         put_group_info(group_info);
1366
1367         return retval;
1368 }
1369
1370 /*
1371  * Check whether we're fsgid/egid or in the supplemental group..
1372  */
1373 int in_group_p(gid_t grp)
1374 {
1375         const struct cred *cred = current_cred();
1376         int retval = 1;
1377
1378         if (grp != cred->fsgid)
1379                 retval = groups_search(cred->group_info, grp);
1380         return retval;
1381 }
1382
1383 EXPORT_SYMBOL(in_group_p);
1384
1385 int in_egroup_p(gid_t grp)
1386 {
1387         const struct cred *cred = current_cred();
1388         int retval = 1;
1389
1390         if (grp != cred->egid)
1391                 retval = groups_search(cred->group_info, grp);
1392         return retval;
1393 }
1394
1395 EXPORT_SYMBOL(in_egroup_p);
1396
1397 DECLARE_RWSEM(uts_sem);
1398
1399 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1400 {
1401         int errno = 0;
1402
1403         down_read(&uts_sem);
1404         if (copy_to_user(name, utsname(), sizeof *name))
1405                 errno = -EFAULT;
1406         up_read(&uts_sem);
1407         return errno;
1408 }
1409
1410 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1411 {
1412         int errno;
1413         char tmp[__NEW_UTS_LEN];
1414
1415         if (!capable(CAP_SYS_ADMIN))
1416                 return -EPERM;
1417         if (len < 0 || len > __NEW_UTS_LEN)
1418                 return -EINVAL;
1419         down_write(&uts_sem);
1420         errno = -EFAULT;
1421         if (!copy_from_user(tmp, name, len)) {
1422                 struct new_utsname *u = utsname();
1423
1424                 memcpy(u->nodename, tmp, len);
1425                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1426                 errno = 0;
1427         }
1428         up_write(&uts_sem);
1429         return errno;
1430 }
1431
1432 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1433
1434 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1435 {
1436         int i, errno;
1437         struct new_utsname *u;
1438
1439         if (len < 0)
1440                 return -EINVAL;
1441         down_read(&uts_sem);
1442         u = utsname();
1443         i = 1 + strlen(u->nodename);
1444         if (i > len)
1445                 i = len;
1446         errno = 0;
1447         if (copy_to_user(name, u->nodename, i))
1448                 errno = -EFAULT;
1449         up_read(&uts_sem);
1450         return errno;
1451 }
1452
1453 #endif
1454
1455 /*
1456  * Only setdomainname; getdomainname can be implemented by calling
1457  * uname()
1458  */
1459 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1460 {
1461         int errno;
1462         char tmp[__NEW_UTS_LEN];
1463
1464         if (!capable(CAP_SYS_ADMIN))
1465                 return -EPERM;
1466         if (len < 0 || len > __NEW_UTS_LEN)
1467                 return -EINVAL;
1468
1469         down_write(&uts_sem);
1470         errno = -EFAULT;
1471         if (!copy_from_user(tmp, name, len)) {
1472                 struct new_utsname *u = utsname();
1473
1474                 memcpy(u->domainname, tmp, len);
1475                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1476                 errno = 0;
1477         }
1478         up_write(&uts_sem);
1479         return errno;
1480 }
1481
1482 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1483 {
1484         if (resource >= RLIM_NLIMITS)
1485                 return -EINVAL;
1486         else {
1487                 struct rlimit value;
1488                 task_lock(current->group_leader);
1489                 value = current->signal->rlim[resource];
1490                 task_unlock(current->group_leader);
1491                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1492         }
1493 }
1494
1495 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1496
1497 /*
1498  *      Back compatibility for getrlimit. Needed for some apps.
1499  */
1500  
1501 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1502                 struct rlimit __user *, rlim)
1503 {
1504         struct rlimit x;
1505         if (resource >= RLIM_NLIMITS)
1506                 return -EINVAL;
1507
1508         task_lock(current->group_leader);
1509         x = current->signal->rlim[resource];
1510         task_unlock(current->group_leader);
1511         if (x.rlim_cur > 0x7FFFFFFF)
1512                 x.rlim_cur = 0x7FFFFFFF;
1513         if (x.rlim_max > 0x7FFFFFFF)
1514                 x.rlim_max = 0x7FFFFFFF;
1515         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1516 }
1517
1518 #endif
1519
1520 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1521 {
1522         struct rlimit new_rlim, *old_rlim;
1523         int retval;
1524
1525         if (resource >= RLIM_NLIMITS)
1526                 return -EINVAL;
1527         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1528                 return -EFAULT;
1529         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1530                 return -EINVAL;
1531         old_rlim = current->signal->rlim + resource;
1532         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1533             !capable(CAP_SYS_RESOURCE))
1534                 return -EPERM;
1535         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1536                 return -EPERM;
1537
1538         retval = security_task_setrlimit(resource, &new_rlim);
1539         if (retval)
1540                 return retval;
1541
1542         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1543                 /*
1544                  * The caller is asking for an immediate RLIMIT_CPU
1545                  * expiry.  But we use the zero value to mean "it was
1546                  * never set".  So let's cheat and make it one second
1547                  * instead
1548                  */
1549                 new_rlim.rlim_cur = 1;
1550         }
1551
1552         task_lock(current->group_leader);
1553         *old_rlim = new_rlim;
1554         task_unlock(current->group_leader);
1555
1556         if (resource != RLIMIT_CPU)
1557                 goto out;
1558
1559         /*
1560          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1561          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1562          * very long-standing error, and fixing it now risks breakage of
1563          * applications, so we live with it
1564          */
1565         if (new_rlim.rlim_cur == RLIM_INFINITY)
1566                 goto out;
1567
1568         update_rlimit_cpu(new_rlim.rlim_cur);
1569 out:
1570         return 0;
1571 }
1572
1573 /*
1574  * It would make sense to put struct rusage in the task_struct,
1575  * except that would make the task_struct be *really big*.  After
1576  * task_struct gets moved into malloc'ed memory, it would
1577  * make sense to do this.  It will make moving the rest of the information
1578  * a lot simpler!  (Which we're not doing right now because we're not
1579  * measuring them yet).
1580  *
1581  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1582  * races with threads incrementing their own counters.  But since word
1583  * reads are atomic, we either get new values or old values and we don't
1584  * care which for the sums.  We always take the siglock to protect reading
1585  * the c* fields from p->signal from races with exit.c updating those
1586  * fields when reaping, so a sample either gets all the additions of a
1587  * given child after it's reaped, or none so this sample is before reaping.
1588  *
1589  * Locking:
1590  * We need to take the siglock for CHILDEREN, SELF and BOTH
1591  * for  the cases current multithreaded, non-current single threaded
1592  * non-current multithreaded.  Thread traversal is now safe with
1593  * the siglock held.
1594  * Strictly speaking, we donot need to take the siglock if we are current and
1595  * single threaded,  as no one else can take our signal_struct away, no one
1596  * else can  reap the  children to update signal->c* counters, and no one else
1597  * can race with the signal-> fields. If we do not take any lock, the
1598  * signal-> fields could be read out of order while another thread was just
1599  * exiting. So we should  place a read memory barrier when we avoid the lock.
1600  * On the writer side,  write memory barrier is implied in  __exit_signal
1601  * as __exit_signal releases  the siglock spinlock after updating the signal->
1602  * fields. But we don't do this yet to keep things simple.
1603  *
1604  */
1605
1606 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1607 {
1608         r->ru_nvcsw += t->nvcsw;
1609         r->ru_nivcsw += t->nivcsw;
1610         r->ru_minflt += t->min_flt;
1611         r->ru_majflt += t->maj_flt;
1612         r->ru_inblock += task_io_get_inblock(t);
1613         r->ru_oublock += task_io_get_oublock(t);
1614 }
1615
1616 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1617 {
1618         struct task_struct *t;
1619         unsigned long flags;
1620         cputime_t utime, stime;
1621         struct task_cputime cputime;
1622
1623         memset((char *) r, 0, sizeof *r);
1624         utime = stime = cputime_zero;
1625
1626         if (who == RUSAGE_THREAD) {
1627                 utime = task_utime(current);
1628                 stime = task_stime(current);
1629                 accumulate_thread_rusage(p, r);
1630                 goto out;
1631         }
1632
1633         if (!lock_task_sighand(p, &flags))
1634                 return;
1635
1636         switch (who) {
1637                 case RUSAGE_BOTH:
1638                 case RUSAGE_CHILDREN:
1639                         utime = p->signal->cutime;
1640                         stime = p->signal->cstime;
1641                         r->ru_nvcsw = p->signal->cnvcsw;
1642                         r->ru_nivcsw = p->signal->cnivcsw;
1643                         r->ru_minflt = p->signal->cmin_flt;
1644                         r->ru_majflt = p->signal->cmaj_flt;
1645                         r->ru_inblock = p->signal->cinblock;
1646                         r->ru_oublock = p->signal->coublock;
1647
1648                         if (who == RUSAGE_CHILDREN)
1649                                 break;
1650
1651                 case RUSAGE_SELF:
1652                         thread_group_cputime(p, &cputime);
1653                         utime = cputime_add(utime, cputime.utime);
1654                         stime = cputime_add(stime, cputime.stime);
1655                         r->ru_nvcsw += p->signal->nvcsw;
1656                         r->ru_nivcsw += p->signal->nivcsw;
1657                         r->ru_minflt += p->signal->min_flt;
1658                         r->ru_majflt += p->signal->maj_flt;
1659                         r->ru_inblock += p->signal->inblock;
1660                         r->ru_oublock += p->signal->oublock;
1661                         t = p;
1662                         do {
1663                                 accumulate_thread_rusage(t, r);
1664                                 t = next_thread(t);
1665                         } while (t != p);
1666                         break;
1667
1668                 default:
1669                         BUG();
1670         }
1671         unlock_task_sighand(p, &flags);
1672
1673 out:
1674         cputime_to_timeval(utime, &r->ru_utime);
1675         cputime_to_timeval(stime, &r->ru_stime);
1676 }
1677
1678 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1679 {
1680         struct rusage r;
1681         k_getrusage(p, who, &r);
1682         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1683 }
1684
1685 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1686 {
1687         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1688             who != RUSAGE_THREAD)
1689                 return -EINVAL;
1690         return getrusage(current, who, ru);
1691 }
1692
1693 SYSCALL_DEFINE1(umask, int, mask)
1694 {
1695         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1696         return mask;
1697 }
1698
1699 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1700                 unsigned long, arg4, unsigned long, arg5)
1701 {
1702         struct task_struct *me = current;
1703         unsigned char comm[sizeof(me->comm)];
1704         long error;
1705
1706         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1707         if (error != -ENOSYS)
1708                 return error;
1709
1710         error = 0;
1711         switch (option) {
1712                 case PR_SET_PDEATHSIG:
1713                         if (!valid_signal(arg2)) {
1714                                 error = -EINVAL;
1715                                 break;
1716                         }
1717                         me->pdeath_signal = arg2;
1718                         error = 0;
1719                         break;
1720                 case PR_GET_PDEATHSIG:
1721                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1722                         break;
1723                 case PR_GET_DUMPABLE:
1724                         error = get_dumpable(me->mm);
1725                         break;
1726                 case PR_SET_DUMPABLE:
1727                         if (arg2 < 0 || arg2 > 1) {
1728                                 error = -EINVAL;
1729                                 break;
1730                         }
1731                         set_dumpable(me->mm, arg2);
1732                         error = 0;
1733                         break;
1734
1735                 case PR_SET_UNALIGN:
1736                         error = SET_UNALIGN_CTL(me, arg2);
1737                         break;
1738                 case PR_GET_UNALIGN:
1739                         error = GET_UNALIGN_CTL(me, arg2);
1740                         break;
1741                 case PR_SET_FPEMU:
1742                         error = SET_FPEMU_CTL(me, arg2);
1743                         break;
1744                 case PR_GET_FPEMU:
1745                         error = GET_FPEMU_CTL(me, arg2);
1746                         break;
1747                 case PR_SET_FPEXC:
1748                         error = SET_FPEXC_CTL(me, arg2);
1749                         break;
1750                 case PR_GET_FPEXC:
1751                         error = GET_FPEXC_CTL(me, arg2);
1752                         break;
1753                 case PR_GET_TIMING:
1754                         error = PR_TIMING_STATISTICAL;
1755                         break;
1756                 case PR_SET_TIMING:
1757                         if (arg2 != PR_TIMING_STATISTICAL)
1758                                 error = -EINVAL;
1759                         else
1760                                 error = 0;
1761                         break;
1762
1763                 case PR_SET_NAME:
1764                         comm[sizeof(me->comm)-1] = 0;
1765                         if (strncpy_from_user(comm, (char __user *)arg2,
1766                                               sizeof(me->comm) - 1) < 0)
1767                                 return -EFAULT;
1768                         set_task_comm(me, comm);
1769                         return 0;
1770                 case PR_GET_NAME:
1771                         get_task_comm(comm, me);
1772                         if (copy_to_user((char __user *)arg2, comm,
1773                                          sizeof(comm)))
1774                                 return -EFAULT;
1775                         return 0;
1776                 case PR_GET_ENDIAN:
1777                         error = GET_ENDIAN(me, arg2);
1778                         break;
1779                 case PR_SET_ENDIAN:
1780                         error = SET_ENDIAN(me, arg2);
1781                         break;
1782
1783                 case PR_GET_SECCOMP:
1784                         error = prctl_get_seccomp();
1785                         break;
1786                 case PR_SET_SECCOMP:
1787                         error = prctl_set_seccomp(arg2);
1788                         break;
1789                 case PR_GET_TSC:
1790                         error = GET_TSC_CTL(arg2);
1791                         break;
1792                 case PR_SET_TSC:
1793                         error = SET_TSC_CTL(arg2);
1794                         break;
1795                 case PR_TASK_PERF_COUNTERS_DISABLE:
1796                         error = perf_counter_task_disable();
1797                         break;
1798                 case PR_TASK_PERF_COUNTERS_ENABLE:
1799                         error = perf_counter_task_enable();
1800                         break;
1801                 case PR_GET_TIMERSLACK:
1802                         error = current->timer_slack_ns;
1803                         break;
1804                 case PR_SET_TIMERSLACK:
1805                         if (arg2 <= 0)
1806                                 current->timer_slack_ns =
1807                                         current->default_timer_slack_ns;
1808                         else
1809                                 current->timer_slack_ns = arg2;
1810                         error = 0;
1811                         break;
1812                 default:
1813                         error = -EINVAL;
1814                         break;
1815         }
1816         return error;
1817 }
1818
1819 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1820                 struct getcpu_cache __user *, unused)
1821 {
1822         int err = 0;
1823         int cpu = raw_smp_processor_id();
1824         if (cpup)
1825                 err |= put_user(cpu, cpup);
1826         if (nodep)
1827                 err |= put_user(cpu_to_node(cpu), nodep);
1828         return err ? -EFAULT : 0;
1829 }
1830
1831 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1832
1833 static void argv_cleanup(char **argv, char **envp)
1834 {
1835         argv_free(argv);
1836 }
1837
1838 /**
1839  * orderly_poweroff - Trigger an orderly system poweroff
1840  * @force: force poweroff if command execution fails
1841  *
1842  * This may be called from any context to trigger a system shutdown.
1843  * If the orderly shutdown fails, it will force an immediate shutdown.
1844  */
1845 int orderly_poweroff(bool force)
1846 {
1847         int argc;
1848         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1849         static char *envp[] = {
1850                 "HOME=/",
1851                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1852                 NULL
1853         };
1854         int ret = -ENOMEM;
1855         struct subprocess_info *info;
1856
1857         if (argv == NULL) {
1858                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1859                        __func__, poweroff_cmd);
1860                 goto out;
1861         }
1862
1863         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1864         if (info == NULL) {
1865                 argv_free(argv);
1866                 goto out;
1867         }
1868
1869         call_usermodehelper_setcleanup(info, argv_cleanup);
1870
1871         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1872
1873   out:
1874         if (ret && force) {
1875                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1876                        "forcing the issue\n");
1877
1878                 /* I guess this should try to kick off some daemon to
1879                    sync and poweroff asap.  Or not even bother syncing
1880                    if we're doing an emergency shutdown? */
1881                 emergency_sync();
1882                 kernel_power_off();
1883         }
1884
1885         return ret;
1886 }
1887 EXPORT_SYMBOL_GPL(orderly_poweroff);