Get rid of indirect include of fs_struct.h
[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/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/ptrace.h>
37 #include <linux/fs_struct.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 (!task_can_switch_user(new_user, current)) {
576                 free_uid(new_user);
577                 return -EINVAL;
578         }
579
580         if (atomic_read(&new_user->processes) >=
581                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
582                         new_user != INIT_USER) {
583                 free_uid(new_user);
584                 return -EAGAIN;
585         }
586
587         free_uid(new->user);
588         new->user = new_user;
589         return 0;
590 }
591
592 /*
593  * Unprivileged users may change the real uid to the effective uid
594  * or vice versa.  (BSD-style)
595  *
596  * If you set the real uid at all, or set the effective uid to a value not
597  * equal to the real uid, then the saved uid is set to the new effective uid.
598  *
599  * This makes it possible for a setuid program to completely drop its
600  * privileges, which is often a useful assertion to make when you are doing
601  * a security audit over a program.
602  *
603  * The general idea is that a program which uses just setreuid() will be
604  * 100% compatible with BSD.  A program which uses just setuid() will be
605  * 100% compatible with POSIX with saved IDs. 
606  */
607 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
608 {
609         const struct cred *old;
610         struct cred *new;
611         int retval;
612
613         new = prepare_creds();
614         if (!new)
615                 return -ENOMEM;
616         old = current_cred();
617
618         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
619         if (retval)
620                 goto error;
621
622         retval = -EPERM;
623         if (ruid != (uid_t) -1) {
624                 new->uid = ruid;
625                 if (old->uid != ruid &&
626                     old->euid != ruid &&
627                     !capable(CAP_SETUID))
628                         goto error;
629         }
630
631         if (euid != (uid_t) -1) {
632                 new->euid = euid;
633                 if (old->uid != euid &&
634                     old->euid != euid &&
635                     old->suid != euid &&
636                     !capable(CAP_SETUID))
637                         goto error;
638         }
639
640         if (new->uid != old->uid) {
641                 retval = set_user(new);
642                 if (retval < 0)
643                         goto error;
644         }
645         if (ruid != (uid_t) -1 ||
646             (euid != (uid_t) -1 && euid != old->uid))
647                 new->suid = new->euid;
648         new->fsuid = new->euid;
649
650         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
651         if (retval < 0)
652                 goto error;
653
654         return commit_creds(new);
655
656 error:
657         abort_creds(new);
658         return retval;
659 }
660                 
661 /*
662  * setuid() is implemented like SysV with SAVED_IDS 
663  * 
664  * Note that SAVED_ID's is deficient in that a setuid root program
665  * like sendmail, for example, cannot set its uid to be a normal 
666  * user and then switch back, because if you're root, setuid() sets
667  * the saved uid too.  If you don't like this, blame the bright people
668  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
669  * will allow a root program to temporarily drop privileges and be able to
670  * regain them by swapping the real and effective uid.  
671  */
672 SYSCALL_DEFINE1(setuid, uid_t, uid)
673 {
674         const struct cred *old;
675         struct cred *new;
676         int retval;
677
678         new = prepare_creds();
679         if (!new)
680                 return -ENOMEM;
681         old = current_cred();
682
683         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
684         if (retval)
685                 goto error;
686
687         retval = -EPERM;
688         if (capable(CAP_SETUID)) {
689                 new->suid = new->uid = uid;
690                 if (uid != old->uid) {
691                         retval = set_user(new);
692                         if (retval < 0)
693                                 goto error;
694                 }
695         } else if (uid != old->uid && uid != new->suid) {
696                 goto error;
697         }
698
699         new->fsuid = new->euid = uid;
700
701         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
702         if (retval < 0)
703                 goto error;
704
705         return commit_creds(new);
706
707 error:
708         abort_creds(new);
709         return retval;
710 }
711
712
713 /*
714  * This function implements a generic ability to update ruid, euid,
715  * and suid.  This allows you to implement the 4.4 compatible seteuid().
716  */
717 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
718 {
719         const struct cred *old;
720         struct cred *new;
721         int retval;
722
723         new = prepare_creds();
724         if (!new)
725                 return -ENOMEM;
726
727         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
728         if (retval)
729                 goto error;
730         old = current_cred();
731
732         retval = -EPERM;
733         if (!capable(CAP_SETUID)) {
734                 if (ruid != (uid_t) -1 && ruid != old->uid &&
735                     ruid != old->euid  && ruid != old->suid)
736                         goto error;
737                 if (euid != (uid_t) -1 && euid != old->uid &&
738                     euid != old->euid  && euid != old->suid)
739                         goto error;
740                 if (suid != (uid_t) -1 && suid != old->uid &&
741                     suid != old->euid  && suid != old->suid)
742                         goto error;
743         }
744
745         if (ruid != (uid_t) -1) {
746                 new->uid = ruid;
747                 if (ruid != old->uid) {
748                         retval = set_user(new);
749                         if (retval < 0)
750                                 goto error;
751                 }
752         }
753         if (euid != (uid_t) -1)
754                 new->euid = euid;
755         if (suid != (uid_t) -1)
756                 new->suid = suid;
757         new->fsuid = new->euid;
758
759         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
760         if (retval < 0)
761                 goto error;
762
763         return commit_creds(new);
764
765 error:
766         abort_creds(new);
767         return retval;
768 }
769
770 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
771 {
772         const struct cred *cred = current_cred();
773         int retval;
774
775         if (!(retval   = put_user(cred->uid,  ruid)) &&
776             !(retval   = put_user(cred->euid, euid)))
777                 retval = put_user(cred->suid, suid);
778
779         return retval;
780 }
781
782 /*
783  * Same as above, but for rgid, egid, sgid.
784  */
785 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
786 {
787         const struct cred *old;
788         struct cred *new;
789         int retval;
790
791         new = prepare_creds();
792         if (!new)
793                 return -ENOMEM;
794         old = current_cred();
795
796         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
797         if (retval)
798                 goto error;
799
800         retval = -EPERM;
801         if (!capable(CAP_SETGID)) {
802                 if (rgid != (gid_t) -1 && rgid != old->gid &&
803                     rgid != old->egid  && rgid != old->sgid)
804                         goto error;
805                 if (egid != (gid_t) -1 && egid != old->gid &&
806                     egid != old->egid  && egid != old->sgid)
807                         goto error;
808                 if (sgid != (gid_t) -1 && sgid != old->gid &&
809                     sgid != old->egid  && sgid != old->sgid)
810                         goto error;
811         }
812
813         if (rgid != (gid_t) -1)
814                 new->gid = rgid;
815         if (egid != (gid_t) -1)
816                 new->egid = egid;
817         if (sgid != (gid_t) -1)
818                 new->sgid = sgid;
819         new->fsgid = new->egid;
820
821         return commit_creds(new);
822
823 error:
824         abort_creds(new);
825         return retval;
826 }
827
828 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
829 {
830         const struct cred *cred = current_cred();
831         int retval;
832
833         if (!(retval   = put_user(cred->gid,  rgid)) &&
834             !(retval   = put_user(cred->egid, egid)))
835                 retval = put_user(cred->sgid, sgid);
836
837         return retval;
838 }
839
840
841 /*
842  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
843  * is used for "access()" and for the NFS daemon (letting nfsd stay at
844  * whatever uid it wants to). It normally shadows "euid", except when
845  * explicitly set by setfsuid() or for access..
846  */
847 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
848 {
849         const struct cred *old;
850         struct cred *new;
851         uid_t old_fsuid;
852
853         new = prepare_creds();
854         if (!new)
855                 return current_fsuid();
856         old = current_cred();
857         old_fsuid = old->fsuid;
858
859         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
860                 goto error;
861
862         if (uid == old->uid  || uid == old->euid  ||
863             uid == old->suid || uid == old->fsuid ||
864             capable(CAP_SETUID)) {
865                 if (uid != old_fsuid) {
866                         new->fsuid = uid;
867                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
868                                 goto change_okay;
869                 }
870         }
871
872 error:
873         abort_creds(new);
874         return old_fsuid;
875
876 change_okay:
877         commit_creds(new);
878         return old_fsuid;
879 }
880
881 /*
882  * Samma på svenska..
883  */
884 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
885 {
886         const struct cred *old;
887         struct cred *new;
888         gid_t old_fsgid;
889
890         new = prepare_creds();
891         if (!new)
892                 return current_fsgid();
893         old = current_cred();
894         old_fsgid = old->fsgid;
895
896         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
897                 goto error;
898
899         if (gid == old->gid  || gid == old->egid  ||
900             gid == old->sgid || gid == old->fsgid ||
901             capable(CAP_SETGID)) {
902                 if (gid != old_fsgid) {
903                         new->fsgid = gid;
904                         goto change_okay;
905                 }
906         }
907
908 error:
909         abort_creds(new);
910         return old_fsgid;
911
912 change_okay:
913         commit_creds(new);
914         return old_fsgid;
915 }
916
917 void do_sys_times(struct tms *tms)
918 {
919         struct task_cputime cputime;
920         cputime_t cutime, cstime;
921
922         thread_group_cputime(current, &cputime);
923         spin_lock_irq(&current->sighand->siglock);
924         cutime = current->signal->cutime;
925         cstime = current->signal->cstime;
926         spin_unlock_irq(&current->sighand->siglock);
927         tms->tms_utime = cputime_to_clock_t(cputime.utime);
928         tms->tms_stime = cputime_to_clock_t(cputime.stime);
929         tms->tms_cutime = cputime_to_clock_t(cutime);
930         tms->tms_cstime = cputime_to_clock_t(cstime);
931 }
932
933 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
934 {
935         if (tbuf) {
936                 struct tms tmp;
937
938                 do_sys_times(&tmp);
939                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
940                         return -EFAULT;
941         }
942         force_successful_syscall_return();
943         return (long) jiffies_64_to_clock_t(get_jiffies_64());
944 }
945
946 /*
947  * This needs some heavy checking ...
948  * I just haven't the stomach for it. I also don't fully
949  * understand sessions/pgrp etc. Let somebody who does explain it.
950  *
951  * OK, I think I have the protection semantics right.... this is really
952  * only important on a multi-user system anyway, to make sure one user
953  * can't send a signal to a process owned by another.  -TYT, 12/12/91
954  *
955  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
956  * LBT 04.03.94
957  */
958 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
959 {
960         struct task_struct *p;
961         struct task_struct *group_leader = current->group_leader;
962         struct pid *pgrp;
963         int err;
964
965         if (!pid)
966                 pid = task_pid_vnr(group_leader);
967         if (!pgid)
968                 pgid = pid;
969         if (pgid < 0)
970                 return -EINVAL;
971
972         /* From this point forward we keep holding onto the tasklist lock
973          * so that our parent does not change from under us. -DaveM
974          */
975         write_lock_irq(&tasklist_lock);
976
977         err = -ESRCH;
978         p = find_task_by_vpid(pid);
979         if (!p)
980                 goto out;
981
982         err = -EINVAL;
983         if (!thread_group_leader(p))
984                 goto out;
985
986         if (same_thread_group(p->real_parent, group_leader)) {
987                 err = -EPERM;
988                 if (task_session(p) != task_session(group_leader))
989                         goto out;
990                 err = -EACCES;
991                 if (p->did_exec)
992                         goto out;
993         } else {
994                 err = -ESRCH;
995                 if (p != group_leader)
996                         goto out;
997         }
998
999         err = -EPERM;
1000         if (p->signal->leader)
1001                 goto out;
1002
1003         pgrp = task_pid(p);
1004         if (pgid != pid) {
1005                 struct task_struct *g;
1006
1007                 pgrp = find_vpid(pgid);
1008                 g = pid_task(pgrp, PIDTYPE_PGID);
1009                 if (!g || task_session(g) != task_session(group_leader))
1010                         goto out;
1011         }
1012
1013         err = security_task_setpgid(p, pgid);
1014         if (err)
1015                 goto out;
1016
1017         if (task_pgrp(p) != pgrp) {
1018                 change_pid(p, PIDTYPE_PGID, pgrp);
1019                 set_task_pgrp(p, pid_nr(pgrp));
1020         }
1021
1022         err = 0;
1023 out:
1024         /* All paths lead to here, thus we are safe. -DaveM */
1025         write_unlock_irq(&tasklist_lock);
1026         return err;
1027 }
1028
1029 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1030 {
1031         struct task_struct *p;
1032         struct pid *grp;
1033         int retval;
1034
1035         rcu_read_lock();
1036         if (!pid)
1037                 grp = task_pgrp(current);
1038         else {
1039                 retval = -ESRCH;
1040                 p = find_task_by_vpid(pid);
1041                 if (!p)
1042                         goto out;
1043                 grp = task_pgrp(p);
1044                 if (!grp)
1045                         goto out;
1046
1047                 retval = security_task_getpgid(p);
1048                 if (retval)
1049                         goto out;
1050         }
1051         retval = pid_vnr(grp);
1052 out:
1053         rcu_read_unlock();
1054         return retval;
1055 }
1056
1057 #ifdef __ARCH_WANT_SYS_GETPGRP
1058
1059 SYSCALL_DEFINE0(getpgrp)
1060 {
1061         return sys_getpgid(0);
1062 }
1063
1064 #endif
1065
1066 SYSCALL_DEFINE1(getsid, pid_t, pid)
1067 {
1068         struct task_struct *p;
1069         struct pid *sid;
1070         int retval;
1071
1072         rcu_read_lock();
1073         if (!pid)
1074                 sid = task_session(current);
1075         else {
1076                 retval = -ESRCH;
1077                 p = find_task_by_vpid(pid);
1078                 if (!p)
1079                         goto out;
1080                 sid = task_session(p);
1081                 if (!sid)
1082                         goto out;
1083
1084                 retval = security_task_getsid(p);
1085                 if (retval)
1086                         goto out;
1087         }
1088         retval = pid_vnr(sid);
1089 out:
1090         rcu_read_unlock();
1091         return retval;
1092 }
1093
1094 SYSCALL_DEFINE0(setsid)
1095 {
1096         struct task_struct *group_leader = current->group_leader;
1097         struct pid *sid = task_pid(group_leader);
1098         pid_t session = pid_vnr(sid);
1099         int err = -EPERM;
1100
1101         write_lock_irq(&tasklist_lock);
1102         /* Fail if I am already a session leader */
1103         if (group_leader->signal->leader)
1104                 goto out;
1105
1106         /* Fail if a process group id already exists that equals the
1107          * proposed session id.
1108          */
1109         if (pid_task(sid, PIDTYPE_PGID))
1110                 goto out;
1111
1112         group_leader->signal->leader = 1;
1113         __set_special_pids(sid);
1114
1115         proc_clear_tty(group_leader);
1116
1117         err = session;
1118 out:
1119         write_unlock_irq(&tasklist_lock);
1120         return err;
1121 }
1122
1123 /*
1124  * Supplementary group IDs
1125  */
1126
1127 /* init to 2 - one for init_task, one to ensure it is never freed */
1128 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1129
1130 struct group_info *groups_alloc(int gidsetsize)
1131 {
1132         struct group_info *group_info;
1133         int nblocks;
1134         int i;
1135
1136         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1137         /* Make sure we always allocate at least one indirect block pointer */
1138         nblocks = nblocks ? : 1;
1139         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1140         if (!group_info)
1141                 return NULL;
1142         group_info->ngroups = gidsetsize;
1143         group_info->nblocks = nblocks;
1144         atomic_set(&group_info->usage, 1);
1145
1146         if (gidsetsize <= NGROUPS_SMALL)
1147                 group_info->blocks[0] = group_info->small_block;
1148         else {
1149                 for (i = 0; i < nblocks; i++) {
1150                         gid_t *b;
1151                         b = (void *)__get_free_page(GFP_USER);
1152                         if (!b)
1153                                 goto out_undo_partial_alloc;
1154                         group_info->blocks[i] = b;
1155                 }
1156         }
1157         return group_info;
1158
1159 out_undo_partial_alloc:
1160         while (--i >= 0) {
1161                 free_page((unsigned long)group_info->blocks[i]);
1162         }
1163         kfree(group_info);
1164         return NULL;
1165 }
1166
1167 EXPORT_SYMBOL(groups_alloc);
1168
1169 void groups_free(struct group_info *group_info)
1170 {
1171         if (group_info->blocks[0] != group_info->small_block) {
1172                 int i;
1173                 for (i = 0; i < group_info->nblocks; i++)
1174                         free_page((unsigned long)group_info->blocks[i]);
1175         }
1176         kfree(group_info);
1177 }
1178
1179 EXPORT_SYMBOL(groups_free);
1180
1181 /* export the group_info to a user-space array */
1182 static int groups_to_user(gid_t __user *grouplist,
1183                           const struct group_info *group_info)
1184 {
1185         int i;
1186         unsigned int count = group_info->ngroups;
1187
1188         for (i = 0; i < group_info->nblocks; i++) {
1189                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1190                 unsigned int len = cp_count * sizeof(*grouplist);
1191
1192                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1193                         return -EFAULT;
1194
1195                 grouplist += NGROUPS_PER_BLOCK;
1196                 count -= cp_count;
1197         }
1198         return 0;
1199 }
1200
1201 /* fill a group_info from a user-space array - it must be allocated already */
1202 static int groups_from_user(struct group_info *group_info,
1203     gid_t __user *grouplist)
1204 {
1205         int i;
1206         unsigned int count = group_info->ngroups;
1207
1208         for (i = 0; i < group_info->nblocks; i++) {
1209                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1210                 unsigned int len = cp_count * sizeof(*grouplist);
1211
1212                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1213                         return -EFAULT;
1214
1215                 grouplist += NGROUPS_PER_BLOCK;
1216                 count -= cp_count;
1217         }
1218         return 0;
1219 }
1220
1221 /* a simple Shell sort */
1222 static void groups_sort(struct group_info *group_info)
1223 {
1224         int base, max, stride;
1225         int gidsetsize = group_info->ngroups;
1226
1227         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1228                 ; /* nothing */
1229         stride /= 3;
1230
1231         while (stride) {
1232                 max = gidsetsize - stride;
1233                 for (base = 0; base < max; base++) {
1234                         int left = base;
1235                         int right = left + stride;
1236                         gid_t tmp = GROUP_AT(group_info, right);
1237
1238                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1239                                 GROUP_AT(group_info, right) =
1240                                     GROUP_AT(group_info, left);
1241                                 right = left;
1242                                 left -= stride;
1243                         }
1244                         GROUP_AT(group_info, right) = tmp;
1245                 }
1246                 stride /= 3;
1247         }
1248 }
1249
1250 /* a simple bsearch */
1251 int groups_search(const struct group_info *group_info, gid_t grp)
1252 {
1253         unsigned int left, right;
1254
1255         if (!group_info)
1256                 return 0;
1257
1258         left = 0;
1259         right = group_info->ngroups;
1260         while (left < right) {
1261                 unsigned int mid = (left+right)/2;
1262                 int cmp = grp - GROUP_AT(group_info, mid);
1263                 if (cmp > 0)
1264                         left = mid + 1;
1265                 else if (cmp < 0)
1266                         right = mid;
1267                 else
1268                         return 1;
1269         }
1270         return 0;
1271 }
1272
1273 /**
1274  * set_groups - Change a group subscription in a set of credentials
1275  * @new: The newly prepared set of credentials to alter
1276  * @group_info: The group list to install
1277  *
1278  * Validate a group subscription and, if valid, insert it into a set
1279  * of credentials.
1280  */
1281 int set_groups(struct cred *new, struct group_info *group_info)
1282 {
1283         int retval;
1284
1285         retval = security_task_setgroups(group_info);
1286         if (retval)
1287                 return retval;
1288
1289         put_group_info(new->group_info);
1290         groups_sort(group_info);
1291         get_group_info(group_info);
1292         new->group_info = group_info;
1293         return 0;
1294 }
1295
1296 EXPORT_SYMBOL(set_groups);
1297
1298 /**
1299  * set_current_groups - Change current's group subscription
1300  * @group_info: The group list to impose
1301  *
1302  * Validate a group subscription and, if valid, impose it upon current's task
1303  * security record.
1304  */
1305 int set_current_groups(struct group_info *group_info)
1306 {
1307         struct cred *new;
1308         int ret;
1309
1310         new = prepare_creds();
1311         if (!new)
1312                 return -ENOMEM;
1313
1314         ret = set_groups(new, group_info);
1315         if (ret < 0) {
1316                 abort_creds(new);
1317                 return ret;
1318         }
1319
1320         return commit_creds(new);
1321 }
1322
1323 EXPORT_SYMBOL(set_current_groups);
1324
1325 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1326 {
1327         const struct cred *cred = current_cred();
1328         int i;
1329
1330         if (gidsetsize < 0)
1331                 return -EINVAL;
1332
1333         /* no need to grab task_lock here; it cannot change */
1334         i = cred->group_info->ngroups;
1335         if (gidsetsize) {
1336                 if (i > gidsetsize) {
1337                         i = -EINVAL;
1338                         goto out;
1339                 }
1340                 if (groups_to_user(grouplist, cred->group_info)) {
1341                         i = -EFAULT;
1342                         goto out;
1343                 }
1344         }
1345 out:
1346         return i;
1347 }
1348
1349 /*
1350  *      SMP: Our groups are copy-on-write. We can set them safely
1351  *      without another task interfering.
1352  */
1353  
1354 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1355 {
1356         struct group_info *group_info;
1357         int retval;
1358
1359         if (!capable(CAP_SETGID))
1360                 return -EPERM;
1361         if ((unsigned)gidsetsize > NGROUPS_MAX)
1362                 return -EINVAL;
1363
1364         group_info = groups_alloc(gidsetsize);
1365         if (!group_info)
1366                 return -ENOMEM;
1367         retval = groups_from_user(group_info, grouplist);
1368         if (retval) {
1369                 put_group_info(group_info);
1370                 return retval;
1371         }
1372
1373         retval = set_current_groups(group_info);
1374         put_group_info(group_info);
1375
1376         return retval;
1377 }
1378
1379 /*
1380  * Check whether we're fsgid/egid or in the supplemental group..
1381  */
1382 int in_group_p(gid_t grp)
1383 {
1384         const struct cred *cred = current_cred();
1385         int retval = 1;
1386
1387         if (grp != cred->fsgid)
1388                 retval = groups_search(cred->group_info, grp);
1389         return retval;
1390 }
1391
1392 EXPORT_SYMBOL(in_group_p);
1393
1394 int in_egroup_p(gid_t grp)
1395 {
1396         const struct cred *cred = current_cred();
1397         int retval = 1;
1398
1399         if (grp != cred->egid)
1400                 retval = groups_search(cred->group_info, grp);
1401         return retval;
1402 }
1403
1404 EXPORT_SYMBOL(in_egroup_p);
1405
1406 DECLARE_RWSEM(uts_sem);
1407
1408 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1409 {
1410         int errno = 0;
1411
1412         down_read(&uts_sem);
1413         if (copy_to_user(name, utsname(), sizeof *name))
1414                 errno = -EFAULT;
1415         up_read(&uts_sem);
1416         return errno;
1417 }
1418
1419 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1420 {
1421         int errno;
1422         char tmp[__NEW_UTS_LEN];
1423
1424         if (!capable(CAP_SYS_ADMIN))
1425                 return -EPERM;
1426         if (len < 0 || len > __NEW_UTS_LEN)
1427                 return -EINVAL;
1428         down_write(&uts_sem);
1429         errno = -EFAULT;
1430         if (!copy_from_user(tmp, name, len)) {
1431                 struct new_utsname *u = utsname();
1432
1433                 memcpy(u->nodename, tmp, len);
1434                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1435                 errno = 0;
1436         }
1437         up_write(&uts_sem);
1438         return errno;
1439 }
1440
1441 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1442
1443 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1444 {
1445         int i, errno;
1446         struct new_utsname *u;
1447
1448         if (len < 0)
1449                 return -EINVAL;
1450         down_read(&uts_sem);
1451         u = utsname();
1452         i = 1 + strlen(u->nodename);
1453         if (i > len)
1454                 i = len;
1455         errno = 0;
1456         if (copy_to_user(name, u->nodename, i))
1457                 errno = -EFAULT;
1458         up_read(&uts_sem);
1459         return errno;
1460 }
1461
1462 #endif
1463
1464 /*
1465  * Only setdomainname; getdomainname can be implemented by calling
1466  * uname()
1467  */
1468 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1469 {
1470         int errno;
1471         char tmp[__NEW_UTS_LEN];
1472
1473         if (!capable(CAP_SYS_ADMIN))
1474                 return -EPERM;
1475         if (len < 0 || len > __NEW_UTS_LEN)
1476                 return -EINVAL;
1477
1478         down_write(&uts_sem);
1479         errno = -EFAULT;
1480         if (!copy_from_user(tmp, name, len)) {
1481                 struct new_utsname *u = utsname();
1482
1483                 memcpy(u->domainname, tmp, len);
1484                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1485                 errno = 0;
1486         }
1487         up_write(&uts_sem);
1488         return errno;
1489 }
1490
1491 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1492 {
1493         if (resource >= RLIM_NLIMITS)
1494                 return -EINVAL;
1495         else {
1496                 struct rlimit value;
1497                 task_lock(current->group_leader);
1498                 value = current->signal->rlim[resource];
1499                 task_unlock(current->group_leader);
1500                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1501         }
1502 }
1503
1504 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1505
1506 /*
1507  *      Back compatibility for getrlimit. Needed for some apps.
1508  */
1509  
1510 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1511                 struct rlimit __user *, rlim)
1512 {
1513         struct rlimit x;
1514         if (resource >= RLIM_NLIMITS)
1515                 return -EINVAL;
1516
1517         task_lock(current->group_leader);
1518         x = current->signal->rlim[resource];
1519         task_unlock(current->group_leader);
1520         if (x.rlim_cur > 0x7FFFFFFF)
1521                 x.rlim_cur = 0x7FFFFFFF;
1522         if (x.rlim_max > 0x7FFFFFFF)
1523                 x.rlim_max = 0x7FFFFFFF;
1524         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1525 }
1526
1527 #endif
1528
1529 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1530 {
1531         struct rlimit new_rlim, *old_rlim;
1532         int retval;
1533
1534         if (resource >= RLIM_NLIMITS)
1535                 return -EINVAL;
1536         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1537                 return -EFAULT;
1538         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1539                 return -EINVAL;
1540         old_rlim = current->signal->rlim + resource;
1541         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1542             !capable(CAP_SYS_RESOURCE))
1543                 return -EPERM;
1544         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1545                 return -EPERM;
1546
1547         retval = security_task_setrlimit(resource, &new_rlim);
1548         if (retval)
1549                 return retval;
1550
1551         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1552                 /*
1553                  * The caller is asking for an immediate RLIMIT_CPU
1554                  * expiry.  But we use the zero value to mean "it was
1555                  * never set".  So let's cheat and make it one second
1556                  * instead
1557                  */
1558                 new_rlim.rlim_cur = 1;
1559         }
1560
1561         task_lock(current->group_leader);
1562         *old_rlim = new_rlim;
1563         task_unlock(current->group_leader);
1564
1565         if (resource != RLIMIT_CPU)
1566                 goto out;
1567
1568         /*
1569          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1570          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1571          * very long-standing error, and fixing it now risks breakage of
1572          * applications, so we live with it
1573          */
1574         if (new_rlim.rlim_cur == RLIM_INFINITY)
1575                 goto out;
1576
1577         update_rlimit_cpu(new_rlim.rlim_cur);
1578 out:
1579         return 0;
1580 }
1581
1582 /*
1583  * It would make sense to put struct rusage in the task_struct,
1584  * except that would make the task_struct be *really big*.  After
1585  * task_struct gets moved into malloc'ed memory, it would
1586  * make sense to do this.  It will make moving the rest of the information
1587  * a lot simpler!  (Which we're not doing right now because we're not
1588  * measuring them yet).
1589  *
1590  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1591  * races with threads incrementing their own counters.  But since word
1592  * reads are atomic, we either get new values or old values and we don't
1593  * care which for the sums.  We always take the siglock to protect reading
1594  * the c* fields from p->signal from races with exit.c updating those
1595  * fields when reaping, so a sample either gets all the additions of a
1596  * given child after it's reaped, or none so this sample is before reaping.
1597  *
1598  * Locking:
1599  * We need to take the siglock for CHILDEREN, SELF and BOTH
1600  * for  the cases current multithreaded, non-current single threaded
1601  * non-current multithreaded.  Thread traversal is now safe with
1602  * the siglock held.
1603  * Strictly speaking, we donot need to take the siglock if we are current and
1604  * single threaded,  as no one else can take our signal_struct away, no one
1605  * else can  reap the  children to update signal->c* counters, and no one else
1606  * can race with the signal-> fields. If we do not take any lock, the
1607  * signal-> fields could be read out of order while another thread was just
1608  * exiting. So we should  place a read memory barrier when we avoid the lock.
1609  * On the writer side,  write memory barrier is implied in  __exit_signal
1610  * as __exit_signal releases  the siglock spinlock after updating the signal->
1611  * fields. But we don't do this yet to keep things simple.
1612  *
1613  */
1614
1615 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1616 {
1617         r->ru_nvcsw += t->nvcsw;
1618         r->ru_nivcsw += t->nivcsw;
1619         r->ru_minflt += t->min_flt;
1620         r->ru_majflt += t->maj_flt;
1621         r->ru_inblock += task_io_get_inblock(t);
1622         r->ru_oublock += task_io_get_oublock(t);
1623 }
1624
1625 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1626 {
1627         struct task_struct *t;
1628         unsigned long flags;
1629         cputime_t utime, stime;
1630         struct task_cputime cputime;
1631
1632         memset((char *) r, 0, sizeof *r);
1633         utime = stime = cputime_zero;
1634
1635         if (who == RUSAGE_THREAD) {
1636                 utime = task_utime(current);
1637                 stime = task_stime(current);
1638                 accumulate_thread_rusage(p, r);
1639                 goto out;
1640         }
1641
1642         if (!lock_task_sighand(p, &flags))
1643                 return;
1644
1645         switch (who) {
1646                 case RUSAGE_BOTH:
1647                 case RUSAGE_CHILDREN:
1648                         utime = p->signal->cutime;
1649                         stime = p->signal->cstime;
1650                         r->ru_nvcsw = p->signal->cnvcsw;
1651                         r->ru_nivcsw = p->signal->cnivcsw;
1652                         r->ru_minflt = p->signal->cmin_flt;
1653                         r->ru_majflt = p->signal->cmaj_flt;
1654                         r->ru_inblock = p->signal->cinblock;
1655                         r->ru_oublock = p->signal->coublock;
1656
1657                         if (who == RUSAGE_CHILDREN)
1658                                 break;
1659
1660                 case RUSAGE_SELF:
1661                         thread_group_cputime(p, &cputime);
1662                         utime = cputime_add(utime, cputime.utime);
1663                         stime = cputime_add(stime, cputime.stime);
1664                         r->ru_nvcsw += p->signal->nvcsw;
1665                         r->ru_nivcsw += p->signal->nivcsw;
1666                         r->ru_minflt += p->signal->min_flt;
1667                         r->ru_majflt += p->signal->maj_flt;
1668                         r->ru_inblock += p->signal->inblock;
1669                         r->ru_oublock += p->signal->oublock;
1670                         t = p;
1671                         do {
1672                                 accumulate_thread_rusage(t, r);
1673                                 t = next_thread(t);
1674                         } while (t != p);
1675                         break;
1676
1677                 default:
1678                         BUG();
1679         }
1680         unlock_task_sighand(p, &flags);
1681
1682 out:
1683         cputime_to_timeval(utime, &r->ru_utime);
1684         cputime_to_timeval(stime, &r->ru_stime);
1685 }
1686
1687 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1688 {
1689         struct rusage r;
1690         k_getrusage(p, who, &r);
1691         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1692 }
1693
1694 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1695 {
1696         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1697             who != RUSAGE_THREAD)
1698                 return -EINVAL;
1699         return getrusage(current, who, ru);
1700 }
1701
1702 SYSCALL_DEFINE1(umask, int, mask)
1703 {
1704         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1705         return mask;
1706 }
1707
1708 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1709                 unsigned long, arg4, unsigned long, arg5)
1710 {
1711         struct task_struct *me = current;
1712         unsigned char comm[sizeof(me->comm)];
1713         long error;
1714
1715         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1716         if (error != -ENOSYS)
1717                 return error;
1718
1719         error = 0;
1720         switch (option) {
1721                 case PR_SET_PDEATHSIG:
1722                         if (!valid_signal(arg2)) {
1723                                 error = -EINVAL;
1724                                 break;
1725                         }
1726                         me->pdeath_signal = arg2;
1727                         error = 0;
1728                         break;
1729                 case PR_GET_PDEATHSIG:
1730                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1731                         break;
1732                 case PR_GET_DUMPABLE:
1733                         error = get_dumpable(me->mm);
1734                         break;
1735                 case PR_SET_DUMPABLE:
1736                         if (arg2 < 0 || arg2 > 1) {
1737                                 error = -EINVAL;
1738                                 break;
1739                         }
1740                         set_dumpable(me->mm, arg2);
1741                         error = 0;
1742                         break;
1743
1744                 case PR_SET_UNALIGN:
1745                         error = SET_UNALIGN_CTL(me, arg2);
1746                         break;
1747                 case PR_GET_UNALIGN:
1748                         error = GET_UNALIGN_CTL(me, arg2);
1749                         break;
1750                 case PR_SET_FPEMU:
1751                         error = SET_FPEMU_CTL(me, arg2);
1752                         break;
1753                 case PR_GET_FPEMU:
1754                         error = GET_FPEMU_CTL(me, arg2);
1755                         break;
1756                 case PR_SET_FPEXC:
1757                         error = SET_FPEXC_CTL(me, arg2);
1758                         break;
1759                 case PR_GET_FPEXC:
1760                         error = GET_FPEXC_CTL(me, arg2);
1761                         break;
1762                 case PR_GET_TIMING:
1763                         error = PR_TIMING_STATISTICAL;
1764                         break;
1765                 case PR_SET_TIMING:
1766                         if (arg2 != PR_TIMING_STATISTICAL)
1767                                 error = -EINVAL;
1768                         else
1769                                 error = 0;
1770                         break;
1771
1772                 case PR_SET_NAME:
1773                         comm[sizeof(me->comm)-1] = 0;
1774                         if (strncpy_from_user(comm, (char __user *)arg2,
1775                                               sizeof(me->comm) - 1) < 0)
1776                                 return -EFAULT;
1777                         set_task_comm(me, comm);
1778                         return 0;
1779                 case PR_GET_NAME:
1780                         get_task_comm(comm, me);
1781                         if (copy_to_user((char __user *)arg2, comm,
1782                                          sizeof(comm)))
1783                                 return -EFAULT;
1784                         return 0;
1785                 case PR_GET_ENDIAN:
1786                         error = GET_ENDIAN(me, arg2);
1787                         break;
1788                 case PR_SET_ENDIAN:
1789                         error = SET_ENDIAN(me, arg2);
1790                         break;
1791
1792                 case PR_GET_SECCOMP:
1793                         error = prctl_get_seccomp();
1794                         break;
1795                 case PR_SET_SECCOMP:
1796                         error = prctl_set_seccomp(arg2);
1797                         break;
1798                 case PR_GET_TSC:
1799                         error = GET_TSC_CTL(arg2);
1800                         break;
1801                 case PR_SET_TSC:
1802                         error = SET_TSC_CTL(arg2);
1803                         break;
1804                 case PR_GET_TIMERSLACK:
1805                         error = current->timer_slack_ns;
1806                         break;
1807                 case PR_SET_TIMERSLACK:
1808                         if (arg2 <= 0)
1809                                 current->timer_slack_ns =
1810                                         current->default_timer_slack_ns;
1811                         else
1812                                 current->timer_slack_ns = arg2;
1813                         error = 0;
1814                         break;
1815                 default:
1816                         error = -EINVAL;
1817                         break;
1818         }
1819         return error;
1820 }
1821
1822 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1823                 struct getcpu_cache __user *, unused)
1824 {
1825         int err = 0;
1826         int cpu = raw_smp_processor_id();
1827         if (cpup)
1828                 err |= put_user(cpu, cpup);
1829         if (nodep)
1830                 err |= put_user(cpu_to_node(cpu), nodep);
1831         return err ? -EFAULT : 0;
1832 }
1833
1834 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1835
1836 static void argv_cleanup(char **argv, char **envp)
1837 {
1838         argv_free(argv);
1839 }
1840
1841 /**
1842  * orderly_poweroff - Trigger an orderly system poweroff
1843  * @force: force poweroff if command execution fails
1844  *
1845  * This may be called from any context to trigger a system shutdown.
1846  * If the orderly shutdown fails, it will force an immediate shutdown.
1847  */
1848 int orderly_poweroff(bool force)
1849 {
1850         int argc;
1851         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1852         static char *envp[] = {
1853                 "HOME=/",
1854                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1855                 NULL
1856         };
1857         int ret = -ENOMEM;
1858         struct subprocess_info *info;
1859
1860         if (argv == NULL) {
1861                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1862                        __func__, poweroff_cmd);
1863                 goto out;
1864         }
1865
1866         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1867         if (info == NULL) {
1868                 argv_free(argv);
1869                 goto out;
1870         }
1871
1872         call_usermodehelper_setcleanup(info, argv_cleanup);
1873
1874         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1875
1876   out:
1877         if (ret && force) {
1878                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1879                        "forcing the issue\n");
1880
1881                 /* I guess this should try to kick off some daemon to
1882                    sync and poweroff asap.  Or not even bother syncing
1883                    if we're doing an emergency shutdown? */
1884                 emergency_sync();
1885                 kernel_power_off();
1886         }
1887
1888         return ret;
1889 }
1890 EXPORT_SYMBOL_GPL(orderly_poweroff);