Merge branch 'linus' into perfcounters/core-v2
[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 #include <linux/fs_struct.h>
39
40 #include <linux/compat.h>
41 #include <linux/syscalls.h>
42 #include <linux/kprobes.h>
43 #include <linux/user_namespace.h>
44
45 #include <asm/uaccess.h>
46 #include <asm/io.h>
47 #include <asm/unistd.h>
48
49 #ifndef SET_UNALIGN_CTL
50 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
51 #endif
52 #ifndef GET_UNALIGN_CTL
53 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
54 #endif
55 #ifndef SET_FPEMU_CTL
56 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
57 #endif
58 #ifndef GET_FPEMU_CTL
59 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
60 #endif
61 #ifndef SET_FPEXC_CTL
62 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
63 #endif
64 #ifndef GET_FPEXC_CTL
65 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
66 #endif
67 #ifndef GET_ENDIAN
68 # define GET_ENDIAN(a,b)        (-EINVAL)
69 #endif
70 #ifndef SET_ENDIAN
71 # define SET_ENDIAN(a,b)        (-EINVAL)
72 #endif
73 #ifndef GET_TSC_CTL
74 # define GET_TSC_CTL(a)         (-EINVAL)
75 #endif
76 #ifndef SET_TSC_CTL
77 # define SET_TSC_CTL(a)         (-EINVAL)
78 #endif
79
80 /*
81  * this is where the system-wide overflow UID and GID are defined, for
82  * architectures that now have 32-bit UID/GID but didn't in the past
83  */
84
85 int overflowuid = DEFAULT_OVERFLOWUID;
86 int overflowgid = DEFAULT_OVERFLOWGID;
87
88 #ifdef CONFIG_UID16
89 EXPORT_SYMBOL(overflowuid);
90 EXPORT_SYMBOL(overflowgid);
91 #endif
92
93 /*
94  * the same as above, but for filesystems which can only store a 16-bit
95  * UID and GID. as such, this is needed on all architectures
96  */
97
98 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
99 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
100
101 EXPORT_SYMBOL(fs_overflowuid);
102 EXPORT_SYMBOL(fs_overflowgid);
103
104 /*
105  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
106  */
107
108 int C_A_D = 1;
109 struct pid *cad_pid;
110 EXPORT_SYMBOL(cad_pid);
111
112 /*
113  * If set, this is used for preparing the system to power off.
114  */
115
116 void (*pm_power_off_prepare)(void);
117
118 /*
119  * set the priority of a task
120  * - the caller must hold the RCU read lock
121  */
122 static int set_one_prio(struct task_struct *p, int niceval, int error)
123 {
124         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
125         int no_nice;
126
127         if (pcred->uid  != cred->euid &&
128             pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
129                 error = -EPERM;
130                 goto out;
131         }
132         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
133                 error = -EACCES;
134                 goto out;
135         }
136         no_nice = security_task_setnice(p, niceval);
137         if (no_nice) {
138                 error = no_nice;
139                 goto out;
140         }
141         if (error == -ESRCH)
142                 error = 0;
143         set_user_nice(p, niceval);
144 out:
145         return error;
146 }
147
148 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
149 {
150         struct task_struct *g, *p;
151         struct user_struct *user;
152         const struct cred *cred = current_cred();
153         int error = -EINVAL;
154         struct pid *pgrp;
155
156         if (which > PRIO_USER || which < PRIO_PROCESS)
157                 goto out;
158
159         /* normalize: avoid signed division (rounding problems) */
160         error = -ESRCH;
161         if (niceval < -20)
162                 niceval = -20;
163         if (niceval > 19)
164                 niceval = 19;
165
166         read_lock(&tasklist_lock);
167         switch (which) {
168                 case PRIO_PROCESS:
169                         if (who)
170                                 p = find_task_by_vpid(who);
171                         else
172                                 p = current;
173                         if (p)
174                                 error = set_one_prio(p, niceval, error);
175                         break;
176                 case PRIO_PGRP:
177                         if (who)
178                                 pgrp = find_vpid(who);
179                         else
180                                 pgrp = task_pgrp(current);
181                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
182                                 error = set_one_prio(p, niceval, error);
183                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
184                         break;
185                 case PRIO_USER:
186                         user = (struct user_struct *) cred->user;
187                         if (!who)
188                                 who = cred->uid;
189                         else if ((who != cred->uid) &&
190                                  !(user = find_user(who)))
191                                 goto out_unlock;        /* No processes for this user */
192
193                         do_each_thread(g, p)
194                                 if (__task_cred(p)->uid == who)
195                                         error = set_one_prio(p, niceval, error);
196                         while_each_thread(g, p);
197                         if (who != cred->uid)
198                                 free_uid(user);         /* For find_user() */
199                         break;
200         }
201 out_unlock:
202         read_unlock(&tasklist_lock);
203 out:
204         return error;
205 }
206
207 /*
208  * Ugh. To avoid negative return values, "getpriority()" will
209  * not return the normal nice-value, but a negated value that
210  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
211  * to stay compatible.
212  */
213 SYSCALL_DEFINE2(getpriority, int, which, int, who)
214 {
215         struct task_struct *g, *p;
216         struct user_struct *user;
217         const struct cred *cred = current_cred();
218         long niceval, retval = -ESRCH;
219         struct pid *pgrp;
220
221         if (which > PRIO_USER || which < PRIO_PROCESS)
222                 return -EINVAL;
223
224         read_lock(&tasklist_lock);
225         switch (which) {
226                 case PRIO_PROCESS:
227                         if (who)
228                                 p = find_task_by_vpid(who);
229                         else
230                                 p = current;
231                         if (p) {
232                                 niceval = 20 - task_nice(p);
233                                 if (niceval > retval)
234                                         retval = niceval;
235                         }
236                         break;
237                 case PRIO_PGRP:
238                         if (who)
239                                 pgrp = find_vpid(who);
240                         else
241                                 pgrp = task_pgrp(current);
242                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
243                                 niceval = 20 - task_nice(p);
244                                 if (niceval > retval)
245                                         retval = niceval;
246                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
247                         break;
248                 case PRIO_USER:
249                         user = (struct user_struct *) cred->user;
250                         if (!who)
251                                 who = cred->uid;
252                         else if ((who != cred->uid) &&
253                                  !(user = find_user(who)))
254                                 goto out_unlock;        /* No processes for this user */
255
256                         do_each_thread(g, p)
257                                 if (__task_cred(p)->uid == who) {
258                                         niceval = 20 - task_nice(p);
259                                         if (niceval > retval)
260                                                 retval = niceval;
261                                 }
262                         while_each_thread(g, p);
263                         if (who != cred->uid)
264                                 free_uid(user);         /* for find_user() */
265                         break;
266         }
267 out_unlock:
268         read_unlock(&tasklist_lock);
269
270         return retval;
271 }
272
273 /**
274  *      emergency_restart - reboot the system
275  *
276  *      Without shutting down any hardware or taking any locks
277  *      reboot the system.  This is called when we know we are in
278  *      trouble so this is our best effort to reboot.  This is
279  *      safe to call in interrupt context.
280  */
281 void emergency_restart(void)
282 {
283         machine_emergency_restart();
284 }
285 EXPORT_SYMBOL_GPL(emergency_restart);
286
287 void kernel_restart_prepare(char *cmd)
288 {
289         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
290         system_state = SYSTEM_RESTART;
291         device_shutdown();
292         sysdev_shutdown();
293 }
294
295 /**
296  *      kernel_restart - reboot the system
297  *      @cmd: pointer to buffer containing command to execute for restart
298  *              or %NULL
299  *
300  *      Shutdown everything and perform a clean reboot.
301  *      This is not safe to call in interrupt context.
302  */
303 void kernel_restart(char *cmd)
304 {
305         kernel_restart_prepare(cmd);
306         if (!cmd)
307                 printk(KERN_EMERG "Restarting system.\n");
308         else
309                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
310         machine_restart(cmd);
311 }
312 EXPORT_SYMBOL_GPL(kernel_restart);
313
314 static void kernel_shutdown_prepare(enum system_states state)
315 {
316         blocking_notifier_call_chain(&reboot_notifier_list,
317                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
318         system_state = state;
319         device_shutdown();
320 }
321 /**
322  *      kernel_halt - halt the system
323  *
324  *      Shutdown everything and perform a clean system halt.
325  */
326 void kernel_halt(void)
327 {
328         kernel_shutdown_prepare(SYSTEM_HALT);
329         sysdev_shutdown();
330         printk(KERN_EMERG "System halted.\n");
331         machine_halt();
332 }
333
334 EXPORT_SYMBOL_GPL(kernel_halt);
335
336 /**
337  *      kernel_power_off - power_off the system
338  *
339  *      Shutdown everything and perform a clean system power_off.
340  */
341 void kernel_power_off(void)
342 {
343         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
344         if (pm_power_off_prepare)
345                 pm_power_off_prepare();
346         disable_nonboot_cpus();
347         sysdev_shutdown();
348         printk(KERN_EMERG "Power down.\n");
349         machine_power_off();
350 }
351 EXPORT_SYMBOL_GPL(kernel_power_off);
352 /*
353  * Reboot system call: for obvious reasons only root may call it,
354  * and even root needs to set up some magic numbers in the registers
355  * so that some mistake won't make this reboot the whole machine.
356  * You can also set the meaning of the ctrl-alt-del-key here.
357  *
358  * reboot doesn't sync: do that yourself before calling this.
359  */
360 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
361                 void __user *, arg)
362 {
363         char buffer[256];
364
365         /* We only trust the superuser with rebooting the system. */
366         if (!capable(CAP_SYS_BOOT))
367                 return -EPERM;
368
369         /* For safety, we require "magic" arguments. */
370         if (magic1 != LINUX_REBOOT_MAGIC1 ||
371             (magic2 != LINUX_REBOOT_MAGIC2 &&
372                         magic2 != LINUX_REBOOT_MAGIC2A &&
373                         magic2 != LINUX_REBOOT_MAGIC2B &&
374                         magic2 != LINUX_REBOOT_MAGIC2C))
375                 return -EINVAL;
376
377         /* Instead of trying to make the power_off code look like
378          * halt when pm_power_off is not set do it the easy way.
379          */
380         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
381                 cmd = LINUX_REBOOT_CMD_HALT;
382
383         lock_kernel();
384         switch (cmd) {
385         case LINUX_REBOOT_CMD_RESTART:
386                 kernel_restart(NULL);
387                 break;
388
389         case LINUX_REBOOT_CMD_CAD_ON:
390                 C_A_D = 1;
391                 break;
392
393         case LINUX_REBOOT_CMD_CAD_OFF:
394                 C_A_D = 0;
395                 break;
396
397         case LINUX_REBOOT_CMD_HALT:
398                 kernel_halt();
399                 unlock_kernel();
400                 do_exit(0);
401                 break;
402
403         case LINUX_REBOOT_CMD_POWER_OFF:
404                 kernel_power_off();
405                 unlock_kernel();
406                 do_exit(0);
407                 break;
408
409         case LINUX_REBOOT_CMD_RESTART2:
410                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
411                         unlock_kernel();
412                         return -EFAULT;
413                 }
414                 buffer[sizeof(buffer) - 1] = '\0';
415
416                 kernel_restart(buffer);
417                 break;
418
419 #ifdef CONFIG_KEXEC
420         case LINUX_REBOOT_CMD_KEXEC:
421                 {
422                         int ret;
423                         ret = kernel_kexec();
424                         unlock_kernel();
425                         return ret;
426                 }
427 #endif
428
429 #ifdef CONFIG_HIBERNATION
430         case LINUX_REBOOT_CMD_SW_SUSPEND:
431                 {
432                         int ret = hibernate();
433                         unlock_kernel();
434                         return ret;
435                 }
436 #endif
437
438         default:
439                 unlock_kernel();
440                 return -EINVAL;
441         }
442         unlock_kernel();
443         return 0;
444 }
445
446 static void deferred_cad(struct work_struct *dummy)
447 {
448         kernel_restart(NULL);
449 }
450
451 /*
452  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
453  * As it's called within an interrupt, it may NOT sync: the only choice
454  * is whether to reboot at once, or just ignore the ctrl-alt-del.
455  */
456 void ctrl_alt_del(void)
457 {
458         static DECLARE_WORK(cad_work, deferred_cad);
459
460         if (C_A_D)
461                 schedule_work(&cad_work);
462         else
463                 kill_cad_pid(SIGINT, 1);
464 }
465         
466 /*
467  * Unprivileged users may change the real gid to the effective gid
468  * or vice versa.  (BSD-style)
469  *
470  * If you set the real gid at all, or set the effective gid to a value not
471  * equal to the real gid, then the saved gid is set to the new effective gid.
472  *
473  * This makes it possible for a setgid program to completely drop its
474  * privileges, which is often a useful assertion to make when you are doing
475  * a security audit over a program.
476  *
477  * The general idea is that a program which uses just setregid() will be
478  * 100% compatible with BSD.  A program which uses just setgid() will be
479  * 100% compatible with POSIX with saved IDs. 
480  *
481  * SMP: There are not races, the GIDs are checked only by filesystem
482  *      operations (as far as semantic preservation is concerned).
483  */
484 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
485 {
486         const struct cred *old;
487         struct cred *new;
488         int retval;
489
490         new = prepare_creds();
491         if (!new)
492                 return -ENOMEM;
493         old = current_cred();
494
495         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
496         if (retval)
497                 goto error;
498
499         retval = -EPERM;
500         if (rgid != (gid_t) -1) {
501                 if (old->gid == rgid ||
502                     old->egid == rgid ||
503                     capable(CAP_SETGID))
504                         new->gid = rgid;
505                 else
506                         goto error;
507         }
508         if (egid != (gid_t) -1) {
509                 if (old->gid == egid ||
510                     old->egid == egid ||
511                     old->sgid == egid ||
512                     capable(CAP_SETGID))
513                         new->egid = egid;
514                 else
515                         goto error;
516         }
517
518         if (rgid != (gid_t) -1 ||
519             (egid != (gid_t) -1 && egid != old->gid))
520                 new->sgid = new->egid;
521         new->fsgid = new->egid;
522
523         return commit_creds(new);
524
525 error:
526         abort_creds(new);
527         return retval;
528 }
529
530 /*
531  * setgid() is implemented like SysV w/ SAVED_IDS 
532  *
533  * SMP: Same implicit races as above.
534  */
535 SYSCALL_DEFINE1(setgid, gid_t, gid)
536 {
537         const struct cred *old;
538         struct cred *new;
539         int retval;
540
541         new = prepare_creds();
542         if (!new)
543                 return -ENOMEM;
544         old = current_cred();
545
546         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
547         if (retval)
548                 goto error;
549
550         retval = -EPERM;
551         if (capable(CAP_SETGID))
552                 new->gid = new->egid = new->sgid = new->fsgid = gid;
553         else if (gid == old->gid || gid == old->sgid)
554                 new->egid = new->fsgid = gid;
555         else
556                 goto error;
557
558         return commit_creds(new);
559
560 error:
561         abort_creds(new);
562         return retval;
563 }
564
565 /*
566  * change the user struct in a credentials set to match the new UID
567  */
568 static int set_user(struct cred *new)
569 {
570         struct user_struct *new_user;
571
572         new_user = alloc_uid(current_user_ns(), new->uid);
573         if (!new_user)
574                 return -EAGAIN;
575
576         if (!task_can_switch_user(new_user, current)) {
577                 free_uid(new_user);
578                 return -EINVAL;
579         }
580
581         if (atomic_read(&new_user->processes) >=
582                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
583                         new_user != INIT_USER) {
584                 free_uid(new_user);
585                 return -EAGAIN;
586         }
587
588         free_uid(new->user);
589         new->user = new_user;
590         return 0;
591 }
592
593 /*
594  * Unprivileged users may change the real uid to the effective uid
595  * or vice versa.  (BSD-style)
596  *
597  * If you set the real uid at all, or set the effective uid to a value not
598  * equal to the real uid, then the saved uid is set to the new effective uid.
599  *
600  * This makes it possible for a setuid program to completely drop its
601  * privileges, which is often a useful assertion to make when you are doing
602  * a security audit over a program.
603  *
604  * The general idea is that a program which uses just setreuid() will be
605  * 100% compatible with BSD.  A program which uses just setuid() will be
606  * 100% compatible with POSIX with saved IDs. 
607  */
608 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
609 {
610         const struct cred *old;
611         struct cred *new;
612         int retval;
613
614         new = prepare_creds();
615         if (!new)
616                 return -ENOMEM;
617         old = current_cred();
618
619         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
620         if (retval)
621                 goto error;
622
623         retval = -EPERM;
624         if (ruid != (uid_t) -1) {
625                 new->uid = ruid;
626                 if (old->uid != ruid &&
627                     old->euid != ruid &&
628                     !capable(CAP_SETUID))
629                         goto error;
630         }
631
632         if (euid != (uid_t) -1) {
633                 new->euid = euid;
634                 if (old->uid != euid &&
635                     old->euid != euid &&
636                     old->suid != euid &&
637                     !capable(CAP_SETUID))
638                         goto error;
639         }
640
641         if (new->uid != old->uid) {
642                 retval = set_user(new);
643                 if (retval < 0)
644                         goto error;
645         }
646         if (ruid != (uid_t) -1 ||
647             (euid != (uid_t) -1 && euid != old->uid))
648                 new->suid = new->euid;
649         new->fsuid = new->euid;
650
651         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
652         if (retval < 0)
653                 goto error;
654
655         return commit_creds(new);
656
657 error:
658         abort_creds(new);
659         return retval;
660 }
661                 
662 /*
663  * setuid() is implemented like SysV with SAVED_IDS 
664  * 
665  * Note that SAVED_ID's is deficient in that a setuid root program
666  * like sendmail, for example, cannot set its uid to be a normal 
667  * user and then switch back, because if you're root, setuid() sets
668  * the saved uid too.  If you don't like this, blame the bright people
669  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
670  * will allow a root program to temporarily drop privileges and be able to
671  * regain them by swapping the real and effective uid.  
672  */
673 SYSCALL_DEFINE1(setuid, uid_t, uid)
674 {
675         const struct cred *old;
676         struct cred *new;
677         int retval;
678
679         new = prepare_creds();
680         if (!new)
681                 return -ENOMEM;
682         old = current_cred();
683
684         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
685         if (retval)
686                 goto error;
687
688         retval = -EPERM;
689         if (capable(CAP_SETUID)) {
690                 new->suid = new->uid = uid;
691                 if (uid != old->uid) {
692                         retval = set_user(new);
693                         if (retval < 0)
694                                 goto error;
695                 }
696         } else if (uid != old->uid && uid != new->suid) {
697                 goto error;
698         }
699
700         new->fsuid = new->euid = uid;
701
702         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
703         if (retval < 0)
704                 goto error;
705
706         return commit_creds(new);
707
708 error:
709         abort_creds(new);
710         return retval;
711 }
712
713
714 /*
715  * This function implements a generic ability to update ruid, euid,
716  * and suid.  This allows you to implement the 4.4 compatible seteuid().
717  */
718 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
719 {
720         const struct cred *old;
721         struct cred *new;
722         int retval;
723
724         new = prepare_creds();
725         if (!new)
726                 return -ENOMEM;
727
728         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
729         if (retval)
730                 goto error;
731         old = current_cred();
732
733         retval = -EPERM;
734         if (!capable(CAP_SETUID)) {
735                 if (ruid != (uid_t) -1 && ruid != old->uid &&
736                     ruid != old->euid  && ruid != old->suid)
737                         goto error;
738                 if (euid != (uid_t) -1 && euid != old->uid &&
739                     euid != old->euid  && euid != old->suid)
740                         goto error;
741                 if (suid != (uid_t) -1 && suid != old->uid &&
742                     suid != old->euid  && suid != old->suid)
743                         goto error;
744         }
745
746         if (ruid != (uid_t) -1) {
747                 new->uid = ruid;
748                 if (ruid != old->uid) {
749                         retval = set_user(new);
750                         if (retval < 0)
751                                 goto error;
752                 }
753         }
754         if (euid != (uid_t) -1)
755                 new->euid = euid;
756         if (suid != (uid_t) -1)
757                 new->suid = suid;
758         new->fsuid = new->euid;
759
760         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
761         if (retval < 0)
762                 goto error;
763
764         return commit_creds(new);
765
766 error:
767         abort_creds(new);
768         return retval;
769 }
770
771 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
772 {
773         const struct cred *cred = current_cred();
774         int retval;
775
776         if (!(retval   = put_user(cred->uid,  ruid)) &&
777             !(retval   = put_user(cred->euid, euid)))
778                 retval = put_user(cred->suid, suid);
779
780         return retval;
781 }
782
783 /*
784  * Same as above, but for rgid, egid, sgid.
785  */
786 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
787 {
788         const struct cred *old;
789         struct cred *new;
790         int retval;
791
792         new = prepare_creds();
793         if (!new)
794                 return -ENOMEM;
795         old = current_cred();
796
797         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
798         if (retval)
799                 goto error;
800
801         retval = -EPERM;
802         if (!capable(CAP_SETGID)) {
803                 if (rgid != (gid_t) -1 && rgid != old->gid &&
804                     rgid != old->egid  && rgid != old->sgid)
805                         goto error;
806                 if (egid != (gid_t) -1 && egid != old->gid &&
807                     egid != old->egid  && egid != old->sgid)
808                         goto error;
809                 if (sgid != (gid_t) -1 && sgid != old->gid &&
810                     sgid != old->egid  && sgid != old->sgid)
811                         goto error;
812         }
813
814         if (rgid != (gid_t) -1)
815                 new->gid = rgid;
816         if (egid != (gid_t) -1)
817                 new->egid = egid;
818         if (sgid != (gid_t) -1)
819                 new->sgid = sgid;
820         new->fsgid = new->egid;
821
822         return commit_creds(new);
823
824 error:
825         abort_creds(new);
826         return retval;
827 }
828
829 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
830 {
831         const struct cred *cred = current_cred();
832         int retval;
833
834         if (!(retval   = put_user(cred->gid,  rgid)) &&
835             !(retval   = put_user(cred->egid, egid)))
836                 retval = put_user(cred->sgid, sgid);
837
838         return retval;
839 }
840
841
842 /*
843  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
844  * is used for "access()" and for the NFS daemon (letting nfsd stay at
845  * whatever uid it wants to). It normally shadows "euid", except when
846  * explicitly set by setfsuid() or for access..
847  */
848 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
849 {
850         const struct cred *old;
851         struct cred *new;
852         uid_t old_fsuid;
853
854         new = prepare_creds();
855         if (!new)
856                 return current_fsuid();
857         old = current_cred();
858         old_fsuid = old->fsuid;
859
860         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
861                 goto error;
862
863         if (uid == old->uid  || uid == old->euid  ||
864             uid == old->suid || uid == old->fsuid ||
865             capable(CAP_SETUID)) {
866                 if (uid != old_fsuid) {
867                         new->fsuid = uid;
868                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
869                                 goto change_okay;
870                 }
871         }
872
873 error:
874         abort_creds(new);
875         return old_fsuid;
876
877 change_okay:
878         commit_creds(new);
879         return old_fsuid;
880 }
881
882 /*
883  * Samma på svenska..
884  */
885 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
886 {
887         const struct cred *old;
888         struct cred *new;
889         gid_t old_fsgid;
890
891         new = prepare_creds();
892         if (!new)
893                 return current_fsgid();
894         old = current_cred();
895         old_fsgid = old->fsgid;
896
897         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
898                 goto error;
899
900         if (gid == old->gid  || gid == old->egid  ||
901             gid == old->sgid || gid == old->fsgid ||
902             capable(CAP_SETGID)) {
903                 if (gid != old_fsgid) {
904                         new->fsgid = gid;
905                         goto change_okay;
906                 }
907         }
908
909 error:
910         abort_creds(new);
911         return old_fsgid;
912
913 change_okay:
914         commit_creds(new);
915         return old_fsgid;
916 }
917
918 void do_sys_times(struct tms *tms)
919 {
920         struct task_cputime cputime;
921         cputime_t cutime, cstime;
922
923         thread_group_cputime(current, &cputime);
924         spin_lock_irq(&current->sighand->siglock);
925         cutime = current->signal->cutime;
926         cstime = current->signal->cstime;
927         spin_unlock_irq(&current->sighand->siglock);
928         tms->tms_utime = cputime_to_clock_t(cputime.utime);
929         tms->tms_stime = cputime_to_clock_t(cputime.stime);
930         tms->tms_cutime = cputime_to_clock_t(cutime);
931         tms->tms_cstime = cputime_to_clock_t(cstime);
932 }
933
934 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
935 {
936         if (tbuf) {
937                 struct tms tmp;
938
939                 do_sys_times(&tmp);
940                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
941                         return -EFAULT;
942         }
943         force_successful_syscall_return();
944         return (long) jiffies_64_to_clock_t(get_jiffies_64());
945 }
946
947 /*
948  * This needs some heavy checking ...
949  * I just haven't the stomach for it. I also don't fully
950  * understand sessions/pgrp etc. Let somebody who does explain it.
951  *
952  * OK, I think I have the protection semantics right.... this is really
953  * only important on a multi-user system anyway, to make sure one user
954  * can't send a signal to a process owned by another.  -TYT, 12/12/91
955  *
956  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
957  * LBT 04.03.94
958  */
959 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
960 {
961         struct task_struct *p;
962         struct task_struct *group_leader = current->group_leader;
963         struct pid *pgrp;
964         int err;
965
966         if (!pid)
967                 pid = task_pid_vnr(group_leader);
968         if (!pgid)
969                 pgid = pid;
970         if (pgid < 0)
971                 return -EINVAL;
972
973         /* From this point forward we keep holding onto the tasklist lock
974          * so that our parent does not change from under us. -DaveM
975          */
976         write_lock_irq(&tasklist_lock);
977
978         err = -ESRCH;
979         p = find_task_by_vpid(pid);
980         if (!p)
981                 goto out;
982
983         err = -EINVAL;
984         if (!thread_group_leader(p))
985                 goto out;
986
987         if (same_thread_group(p->real_parent, group_leader)) {
988                 err = -EPERM;
989                 if (task_session(p) != task_session(group_leader))
990                         goto out;
991                 err = -EACCES;
992                 if (p->did_exec)
993                         goto out;
994         } else {
995                 err = -ESRCH;
996                 if (p != group_leader)
997                         goto out;
998         }
999
1000         err = -EPERM;
1001         if (p->signal->leader)
1002                 goto out;
1003
1004         pgrp = task_pid(p);
1005         if (pgid != pid) {
1006                 struct task_struct *g;
1007
1008                 pgrp = find_vpid(pgid);
1009                 g = pid_task(pgrp, PIDTYPE_PGID);
1010                 if (!g || task_session(g) != task_session(group_leader))
1011                         goto out;
1012         }
1013
1014         err = security_task_setpgid(p, pgid);
1015         if (err)
1016                 goto out;
1017
1018         if (task_pgrp(p) != pgrp)
1019                 change_pid(p, PIDTYPE_PGID, pgrp);
1020
1021         err = 0;
1022 out:
1023         /* All paths lead to here, thus we are safe. -DaveM */
1024         write_unlock_irq(&tasklist_lock);
1025         return err;
1026 }
1027
1028 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1029 {
1030         struct task_struct *p;
1031         struct pid *grp;
1032         int retval;
1033
1034         rcu_read_lock();
1035         if (!pid)
1036                 grp = task_pgrp(current);
1037         else {
1038                 retval = -ESRCH;
1039                 p = find_task_by_vpid(pid);
1040                 if (!p)
1041                         goto out;
1042                 grp = task_pgrp(p);
1043                 if (!grp)
1044                         goto out;
1045
1046                 retval = security_task_getpgid(p);
1047                 if (retval)
1048                         goto out;
1049         }
1050         retval = pid_vnr(grp);
1051 out:
1052         rcu_read_unlock();
1053         return retval;
1054 }
1055
1056 #ifdef __ARCH_WANT_SYS_GETPGRP
1057
1058 SYSCALL_DEFINE0(getpgrp)
1059 {
1060         return sys_getpgid(0);
1061 }
1062
1063 #endif
1064
1065 SYSCALL_DEFINE1(getsid, pid_t, pid)
1066 {
1067         struct task_struct *p;
1068         struct pid *sid;
1069         int retval;
1070
1071         rcu_read_lock();
1072         if (!pid)
1073                 sid = task_session(current);
1074         else {
1075                 retval = -ESRCH;
1076                 p = find_task_by_vpid(pid);
1077                 if (!p)
1078                         goto out;
1079                 sid = task_session(p);
1080                 if (!sid)
1081                         goto out;
1082
1083                 retval = security_task_getsid(p);
1084                 if (retval)
1085                         goto out;
1086         }
1087         retval = pid_vnr(sid);
1088 out:
1089         rcu_read_unlock();
1090         return retval;
1091 }
1092
1093 SYSCALL_DEFINE0(setsid)
1094 {
1095         struct task_struct *group_leader = current->group_leader;
1096         struct pid *sid = task_pid(group_leader);
1097         pid_t session = pid_vnr(sid);
1098         int err = -EPERM;
1099
1100         write_lock_irq(&tasklist_lock);
1101         /* Fail if I am already a session leader */
1102         if (group_leader->signal->leader)
1103                 goto out;
1104
1105         /* Fail if a process group id already exists that equals the
1106          * proposed session id.
1107          */
1108         if (pid_task(sid, PIDTYPE_PGID))
1109                 goto out;
1110
1111         group_leader->signal->leader = 1;
1112         __set_special_pids(sid);
1113
1114         proc_clear_tty(group_leader);
1115
1116         err = session;
1117 out:
1118         write_unlock_irq(&tasklist_lock);
1119         return err;
1120 }
1121
1122 /*
1123  * Supplementary group IDs
1124  */
1125
1126 /* init to 2 - one for init_task, one to ensure it is never freed */
1127 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1128
1129 struct group_info *groups_alloc(int gidsetsize)
1130 {
1131         struct group_info *group_info;
1132         int nblocks;
1133         int i;
1134
1135         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1136         /* Make sure we always allocate at least one indirect block pointer */
1137         nblocks = nblocks ? : 1;
1138         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1139         if (!group_info)
1140                 return NULL;
1141         group_info->ngroups = gidsetsize;
1142         group_info->nblocks = nblocks;
1143         atomic_set(&group_info->usage, 1);
1144
1145         if (gidsetsize <= NGROUPS_SMALL)
1146                 group_info->blocks[0] = group_info->small_block;
1147         else {
1148                 for (i = 0; i < nblocks; i++) {
1149                         gid_t *b;
1150                         b = (void *)__get_free_page(GFP_USER);
1151                         if (!b)
1152                                 goto out_undo_partial_alloc;
1153                         group_info->blocks[i] = b;
1154                 }
1155         }
1156         return group_info;
1157
1158 out_undo_partial_alloc:
1159         while (--i >= 0) {
1160                 free_page((unsigned long)group_info->blocks[i]);
1161         }
1162         kfree(group_info);
1163         return NULL;
1164 }
1165
1166 EXPORT_SYMBOL(groups_alloc);
1167
1168 void groups_free(struct group_info *group_info)
1169 {
1170         if (group_info->blocks[0] != group_info->small_block) {
1171                 int i;
1172                 for (i = 0; i < group_info->nblocks; i++)
1173                         free_page((unsigned long)group_info->blocks[i]);
1174         }
1175         kfree(group_info);
1176 }
1177
1178 EXPORT_SYMBOL(groups_free);
1179
1180 /* export the group_info to a user-space array */
1181 static int groups_to_user(gid_t __user *grouplist,
1182                           const struct group_info *group_info)
1183 {
1184         int i;
1185         unsigned int count = group_info->ngroups;
1186
1187         for (i = 0; i < group_info->nblocks; i++) {
1188                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1189                 unsigned int len = cp_count * sizeof(*grouplist);
1190
1191                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1192                         return -EFAULT;
1193
1194                 grouplist += NGROUPS_PER_BLOCK;
1195                 count -= cp_count;
1196         }
1197         return 0;
1198 }
1199
1200 /* fill a group_info from a user-space array - it must be allocated already */
1201 static int groups_from_user(struct group_info *group_info,
1202     gid_t __user *grouplist)
1203 {
1204         int i;
1205         unsigned int count = group_info->ngroups;
1206
1207         for (i = 0; i < group_info->nblocks; i++) {
1208                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1209                 unsigned int len = cp_count * sizeof(*grouplist);
1210
1211                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1212                         return -EFAULT;
1213
1214                 grouplist += NGROUPS_PER_BLOCK;
1215                 count -= cp_count;
1216         }
1217         return 0;
1218 }
1219
1220 /* a simple Shell sort */
1221 static void groups_sort(struct group_info *group_info)
1222 {
1223         int base, max, stride;
1224         int gidsetsize = group_info->ngroups;
1225
1226         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1227                 ; /* nothing */
1228         stride /= 3;
1229
1230         while (stride) {
1231                 max = gidsetsize - stride;
1232                 for (base = 0; base < max; base++) {
1233                         int left = base;
1234                         int right = left + stride;
1235                         gid_t tmp = GROUP_AT(group_info, right);
1236
1237                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1238                                 GROUP_AT(group_info, right) =
1239                                     GROUP_AT(group_info, left);
1240                                 right = left;
1241                                 left -= stride;
1242                         }
1243                         GROUP_AT(group_info, right) = tmp;
1244                 }
1245                 stride /= 3;
1246         }
1247 }
1248
1249 /* a simple bsearch */
1250 int groups_search(const struct group_info *group_info, gid_t grp)
1251 {
1252         unsigned int left, right;
1253
1254         if (!group_info)
1255                 return 0;
1256
1257         left = 0;
1258         right = group_info->ngroups;
1259         while (left < right) {
1260                 unsigned int mid = (left+right)/2;
1261                 int cmp = grp - GROUP_AT(group_info, mid);
1262                 if (cmp > 0)
1263                         left = mid + 1;
1264                 else if (cmp < 0)
1265                         right = mid;
1266                 else
1267                         return 1;
1268         }
1269         return 0;
1270 }
1271
1272 /**
1273  * set_groups - Change a group subscription in a set of credentials
1274  * @new: The newly prepared set of credentials to alter
1275  * @group_info: The group list to install
1276  *
1277  * Validate a group subscription and, if valid, insert it into a set
1278  * of credentials.
1279  */
1280 int set_groups(struct cred *new, struct group_info *group_info)
1281 {
1282         int retval;
1283
1284         retval = security_task_setgroups(group_info);
1285         if (retval)
1286                 return retval;
1287
1288         put_group_info(new->group_info);
1289         groups_sort(group_info);
1290         get_group_info(group_info);
1291         new->group_info = group_info;
1292         return 0;
1293 }
1294
1295 EXPORT_SYMBOL(set_groups);
1296
1297 /**
1298  * set_current_groups - Change current's group subscription
1299  * @group_info: The group list to impose
1300  *
1301  * Validate a group subscription and, if valid, impose it upon current's task
1302  * security record.
1303  */
1304 int set_current_groups(struct group_info *group_info)
1305 {
1306         struct cred *new;
1307         int ret;
1308
1309         new = prepare_creds();
1310         if (!new)
1311                 return -ENOMEM;
1312
1313         ret = set_groups(new, group_info);
1314         if (ret < 0) {
1315                 abort_creds(new);
1316                 return ret;
1317         }
1318
1319         return commit_creds(new);
1320 }
1321
1322 EXPORT_SYMBOL(set_current_groups);
1323
1324 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1325 {
1326         const struct cred *cred = current_cred();
1327         int i;
1328
1329         if (gidsetsize < 0)
1330                 return -EINVAL;
1331
1332         /* no need to grab task_lock here; it cannot change */
1333         i = cred->group_info->ngroups;
1334         if (gidsetsize) {
1335                 if (i > gidsetsize) {
1336                         i = -EINVAL;
1337                         goto out;
1338                 }
1339                 if (groups_to_user(grouplist, cred->group_info)) {
1340                         i = -EFAULT;
1341                         goto out;
1342                 }
1343         }
1344 out:
1345         return i;
1346 }
1347
1348 /*
1349  *      SMP: Our groups are copy-on-write. We can set them safely
1350  *      without another task interfering.
1351  */
1352  
1353 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1354 {
1355         struct group_info *group_info;
1356         int retval;
1357
1358         if (!capable(CAP_SETGID))
1359                 return -EPERM;
1360         if ((unsigned)gidsetsize > NGROUPS_MAX)
1361                 return -EINVAL;
1362
1363         group_info = groups_alloc(gidsetsize);
1364         if (!group_info)
1365                 return -ENOMEM;
1366         retval = groups_from_user(group_info, grouplist);
1367         if (retval) {
1368                 put_group_info(group_info);
1369                 return retval;
1370         }
1371
1372         retval = set_current_groups(group_info);
1373         put_group_info(group_info);
1374
1375         return retval;
1376 }
1377
1378 /*
1379  * Check whether we're fsgid/egid or in the supplemental group..
1380  */
1381 int in_group_p(gid_t grp)
1382 {
1383         const struct cred *cred = current_cred();
1384         int retval = 1;
1385
1386         if (grp != cred->fsgid)
1387                 retval = groups_search(cred->group_info, grp);
1388         return retval;
1389 }
1390
1391 EXPORT_SYMBOL(in_group_p);
1392
1393 int in_egroup_p(gid_t grp)
1394 {
1395         const struct cred *cred = current_cred();
1396         int retval = 1;
1397
1398         if (grp != cred->egid)
1399                 retval = groups_search(cred->group_info, grp);
1400         return retval;
1401 }
1402
1403 EXPORT_SYMBOL(in_egroup_p);
1404
1405 DECLARE_RWSEM(uts_sem);
1406
1407 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1408 {
1409         int errno = 0;
1410
1411         down_read(&uts_sem);
1412         if (copy_to_user(name, utsname(), sizeof *name))
1413                 errno = -EFAULT;
1414         up_read(&uts_sem);
1415         return errno;
1416 }
1417
1418 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1419 {
1420         int errno;
1421         char tmp[__NEW_UTS_LEN];
1422
1423         if (!capable(CAP_SYS_ADMIN))
1424                 return -EPERM;
1425         if (len < 0 || len > __NEW_UTS_LEN)
1426                 return -EINVAL;
1427         down_write(&uts_sem);
1428         errno = -EFAULT;
1429         if (!copy_from_user(tmp, name, len)) {
1430                 struct new_utsname *u = utsname();
1431
1432                 memcpy(u->nodename, tmp, len);
1433                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1434                 errno = 0;
1435         }
1436         up_write(&uts_sem);
1437         return errno;
1438 }
1439
1440 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1441
1442 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1443 {
1444         int i, errno;
1445         struct new_utsname *u;
1446
1447         if (len < 0)
1448                 return -EINVAL;
1449         down_read(&uts_sem);
1450         u = utsname();
1451         i = 1 + strlen(u->nodename);
1452         if (i > len)
1453                 i = len;
1454         errno = 0;
1455         if (copy_to_user(name, u->nodename, i))
1456                 errno = -EFAULT;
1457         up_read(&uts_sem);
1458         return errno;
1459 }
1460
1461 #endif
1462
1463 /*
1464  * Only setdomainname; getdomainname can be implemented by calling
1465  * uname()
1466  */
1467 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1468 {
1469         int errno;
1470         char tmp[__NEW_UTS_LEN];
1471
1472         if (!capable(CAP_SYS_ADMIN))
1473                 return -EPERM;
1474         if (len < 0 || len > __NEW_UTS_LEN)
1475                 return -EINVAL;
1476
1477         down_write(&uts_sem);
1478         errno = -EFAULT;
1479         if (!copy_from_user(tmp, name, len)) {
1480                 struct new_utsname *u = utsname();
1481
1482                 memcpy(u->domainname, tmp, len);
1483                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1484                 errno = 0;
1485         }
1486         up_write(&uts_sem);
1487         return errno;
1488 }
1489
1490 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1491 {
1492         if (resource >= RLIM_NLIMITS)
1493                 return -EINVAL;
1494         else {
1495                 struct rlimit value;
1496                 task_lock(current->group_leader);
1497                 value = current->signal->rlim[resource];
1498                 task_unlock(current->group_leader);
1499                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1500         }
1501 }
1502
1503 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1504
1505 /*
1506  *      Back compatibility for getrlimit. Needed for some apps.
1507  */
1508  
1509 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1510                 struct rlimit __user *, rlim)
1511 {
1512         struct rlimit x;
1513         if (resource >= RLIM_NLIMITS)
1514                 return -EINVAL;
1515
1516         task_lock(current->group_leader);
1517         x = current->signal->rlim[resource];
1518         task_unlock(current->group_leader);
1519         if (x.rlim_cur > 0x7FFFFFFF)
1520                 x.rlim_cur = 0x7FFFFFFF;
1521         if (x.rlim_max > 0x7FFFFFFF)
1522                 x.rlim_max = 0x7FFFFFFF;
1523         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1524 }
1525
1526 #endif
1527
1528 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1529 {
1530         struct rlimit new_rlim, *old_rlim;
1531         int retval;
1532
1533         if (resource >= RLIM_NLIMITS)
1534                 return -EINVAL;
1535         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1536                 return -EFAULT;
1537         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1538                 return -EINVAL;
1539         old_rlim = current->signal->rlim + resource;
1540         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1541             !capable(CAP_SYS_RESOURCE))
1542                 return -EPERM;
1543         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1544                 return -EPERM;
1545
1546         retval = security_task_setrlimit(resource, &new_rlim);
1547         if (retval)
1548                 return retval;
1549
1550         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1551                 /*
1552                  * The caller is asking for an immediate RLIMIT_CPU
1553                  * expiry.  But we use the zero value to mean "it was
1554                  * never set".  So let's cheat and make it one second
1555                  * instead
1556                  */
1557                 new_rlim.rlim_cur = 1;
1558         }
1559
1560         task_lock(current->group_leader);
1561         *old_rlim = new_rlim;
1562         task_unlock(current->group_leader);
1563
1564         if (resource != RLIMIT_CPU)
1565                 goto out;
1566
1567         /*
1568          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1569          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1570          * very long-standing error, and fixing it now risks breakage of
1571          * applications, so we live with it
1572          */
1573         if (new_rlim.rlim_cur == RLIM_INFINITY)
1574                 goto out;
1575
1576         update_rlimit_cpu(new_rlim.rlim_cur);
1577 out:
1578         return 0;
1579 }
1580
1581 /*
1582  * It would make sense to put struct rusage in the task_struct,
1583  * except that would make the task_struct be *really big*.  After
1584  * task_struct gets moved into malloc'ed memory, it would
1585  * make sense to do this.  It will make moving the rest of the information
1586  * a lot simpler!  (Which we're not doing right now because we're not
1587  * measuring them yet).
1588  *
1589  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1590  * races with threads incrementing their own counters.  But since word
1591  * reads are atomic, we either get new values or old values and we don't
1592  * care which for the sums.  We always take the siglock to protect reading
1593  * the c* fields from p->signal from races with exit.c updating those
1594  * fields when reaping, so a sample either gets all the additions of a
1595  * given child after it's reaped, or none so this sample is before reaping.
1596  *
1597  * Locking:
1598  * We need to take the siglock for CHILDEREN, SELF and BOTH
1599  * for  the cases current multithreaded, non-current single threaded
1600  * non-current multithreaded.  Thread traversal is now safe with
1601  * the siglock held.
1602  * Strictly speaking, we donot need to take the siglock if we are current and
1603  * single threaded,  as no one else can take our signal_struct away, no one
1604  * else can  reap the  children to update signal->c* counters, and no one else
1605  * can race with the signal-> fields. If we do not take any lock, the
1606  * signal-> fields could be read out of order while another thread was just
1607  * exiting. So we should  place a read memory barrier when we avoid the lock.
1608  * On the writer side,  write memory barrier is implied in  __exit_signal
1609  * as __exit_signal releases  the siglock spinlock after updating the signal->
1610  * fields. But we don't do this yet to keep things simple.
1611  *
1612  */
1613
1614 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1615 {
1616         r->ru_nvcsw += t->nvcsw;
1617         r->ru_nivcsw += t->nivcsw;
1618         r->ru_minflt += t->min_flt;
1619         r->ru_majflt += t->maj_flt;
1620         r->ru_inblock += task_io_get_inblock(t);
1621         r->ru_oublock += task_io_get_oublock(t);
1622 }
1623
1624 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1625 {
1626         struct task_struct *t;
1627         unsigned long flags;
1628         cputime_t utime, stime;
1629         struct task_cputime cputime;
1630
1631         memset((char *) r, 0, sizeof *r);
1632         utime = stime = cputime_zero;
1633
1634         if (who == RUSAGE_THREAD) {
1635                 utime = task_utime(current);
1636                 stime = task_stime(current);
1637                 accumulate_thread_rusage(p, r);
1638                 goto out;
1639         }
1640
1641         if (!lock_task_sighand(p, &flags))
1642                 return;
1643
1644         switch (who) {
1645                 case RUSAGE_BOTH:
1646                 case RUSAGE_CHILDREN:
1647                         utime = p->signal->cutime;
1648                         stime = p->signal->cstime;
1649                         r->ru_nvcsw = p->signal->cnvcsw;
1650                         r->ru_nivcsw = p->signal->cnivcsw;
1651                         r->ru_minflt = p->signal->cmin_flt;
1652                         r->ru_majflt = p->signal->cmaj_flt;
1653                         r->ru_inblock = p->signal->cinblock;
1654                         r->ru_oublock = p->signal->coublock;
1655
1656                         if (who == RUSAGE_CHILDREN)
1657                                 break;
1658
1659                 case RUSAGE_SELF:
1660                         thread_group_cputime(p, &cputime);
1661                         utime = cputime_add(utime, cputime.utime);
1662                         stime = cputime_add(stime, cputime.stime);
1663                         r->ru_nvcsw += p->signal->nvcsw;
1664                         r->ru_nivcsw += p->signal->nivcsw;
1665                         r->ru_minflt += p->signal->min_flt;
1666                         r->ru_majflt += p->signal->maj_flt;
1667                         r->ru_inblock += p->signal->inblock;
1668                         r->ru_oublock += p->signal->oublock;
1669                         t = p;
1670                         do {
1671                                 accumulate_thread_rusage(t, r);
1672                                 t = next_thread(t);
1673                         } while (t != p);
1674                         break;
1675
1676                 default:
1677                         BUG();
1678         }
1679         unlock_task_sighand(p, &flags);
1680
1681 out:
1682         cputime_to_timeval(utime, &r->ru_utime);
1683         cputime_to_timeval(stime, &r->ru_stime);
1684 }
1685
1686 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1687 {
1688         struct rusage r;
1689         k_getrusage(p, who, &r);
1690         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1691 }
1692
1693 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1694 {
1695         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1696             who != RUSAGE_THREAD)
1697                 return -EINVAL;
1698         return getrusage(current, who, ru);
1699 }
1700
1701 SYSCALL_DEFINE1(umask, int, mask)
1702 {
1703         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1704         return mask;
1705 }
1706
1707 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1708                 unsigned long, arg4, unsigned long, arg5)
1709 {
1710         struct task_struct *me = current;
1711         unsigned char comm[sizeof(me->comm)];
1712         long error;
1713
1714         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1715         if (error != -ENOSYS)
1716                 return error;
1717
1718         error = 0;
1719         switch (option) {
1720                 case PR_SET_PDEATHSIG:
1721                         if (!valid_signal(arg2)) {
1722                                 error = -EINVAL;
1723                                 break;
1724                         }
1725                         me->pdeath_signal = arg2;
1726                         error = 0;
1727                         break;
1728                 case PR_GET_PDEATHSIG:
1729                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1730                         break;
1731                 case PR_GET_DUMPABLE:
1732                         error = get_dumpable(me->mm);
1733                         break;
1734                 case PR_SET_DUMPABLE:
1735                         if (arg2 < 0 || arg2 > 1) {
1736                                 error = -EINVAL;
1737                                 break;
1738                         }
1739                         set_dumpable(me->mm, arg2);
1740                         error = 0;
1741                         break;
1742
1743                 case PR_SET_UNALIGN:
1744                         error = SET_UNALIGN_CTL(me, arg2);
1745                         break;
1746                 case PR_GET_UNALIGN:
1747                         error = GET_UNALIGN_CTL(me, arg2);
1748                         break;
1749                 case PR_SET_FPEMU:
1750                         error = SET_FPEMU_CTL(me, arg2);
1751                         break;
1752                 case PR_GET_FPEMU:
1753                         error = GET_FPEMU_CTL(me, arg2);
1754                         break;
1755                 case PR_SET_FPEXC:
1756                         error = SET_FPEXC_CTL(me, arg2);
1757                         break;
1758                 case PR_GET_FPEXC:
1759                         error = GET_FPEXC_CTL(me, arg2);
1760                         break;
1761                 case PR_GET_TIMING:
1762                         error = PR_TIMING_STATISTICAL;
1763                         break;
1764                 case PR_SET_TIMING:
1765                         if (arg2 != PR_TIMING_STATISTICAL)
1766                                 error = -EINVAL;
1767                         else
1768                                 error = 0;
1769                         break;
1770
1771                 case PR_SET_NAME:
1772                         comm[sizeof(me->comm)-1] = 0;
1773                         if (strncpy_from_user(comm, (char __user *)arg2,
1774                                               sizeof(me->comm) - 1) < 0)
1775                                 return -EFAULT;
1776                         set_task_comm(me, comm);
1777                         return 0;
1778                 case PR_GET_NAME:
1779                         get_task_comm(comm, me);
1780                         if (copy_to_user((char __user *)arg2, comm,
1781                                          sizeof(comm)))
1782                                 return -EFAULT;
1783                         return 0;
1784                 case PR_GET_ENDIAN:
1785                         error = GET_ENDIAN(me, arg2);
1786                         break;
1787                 case PR_SET_ENDIAN:
1788                         error = SET_ENDIAN(me, arg2);
1789                         break;
1790
1791                 case PR_GET_SECCOMP:
1792                         error = prctl_get_seccomp();
1793                         break;
1794                 case PR_SET_SECCOMP:
1795                         error = prctl_set_seccomp(arg2);
1796                         break;
1797                 case PR_GET_TSC:
1798                         error = GET_TSC_CTL(arg2);
1799                         break;
1800                 case PR_SET_TSC:
1801                         error = SET_TSC_CTL(arg2);
1802                         break;
1803                 case PR_TASK_PERF_COUNTERS_DISABLE:
1804                         error = perf_counter_task_disable();
1805                         break;
1806                 case PR_TASK_PERF_COUNTERS_ENABLE:
1807                         error = perf_counter_task_enable();
1808                         break;
1809                 case PR_GET_TIMERSLACK:
1810                         error = current->timer_slack_ns;
1811                         break;
1812                 case PR_SET_TIMERSLACK:
1813                         if (arg2 <= 0)
1814                                 current->timer_slack_ns =
1815                                         current->default_timer_slack_ns;
1816                         else
1817                                 current->timer_slack_ns = arg2;
1818                         error = 0;
1819                         break;
1820                 default:
1821                         error = -EINVAL;
1822                         break;
1823         }
1824         return error;
1825 }
1826
1827 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1828                 struct getcpu_cache __user *, unused)
1829 {
1830         int err = 0;
1831         int cpu = raw_smp_processor_id();
1832         if (cpup)
1833                 err |= put_user(cpu, cpup);
1834         if (nodep)
1835                 err |= put_user(cpu_to_node(cpu), nodep);
1836         return err ? -EFAULT : 0;
1837 }
1838
1839 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1840
1841 static void argv_cleanup(char **argv, char **envp)
1842 {
1843         argv_free(argv);
1844 }
1845
1846 /**
1847  * orderly_poweroff - Trigger an orderly system poweroff
1848  * @force: force poweroff if command execution fails
1849  *
1850  * This may be called from any context to trigger a system shutdown.
1851  * If the orderly shutdown fails, it will force an immediate shutdown.
1852  */
1853 int orderly_poweroff(bool force)
1854 {
1855         int argc;
1856         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1857         static char *envp[] = {
1858                 "HOME=/",
1859                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1860                 NULL
1861         };
1862         int ret = -ENOMEM;
1863         struct subprocess_info *info;
1864
1865         if (argv == NULL) {
1866                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1867                        __func__, poweroff_cmd);
1868                 goto out;
1869         }
1870
1871         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1872         if (info == NULL) {
1873                 argv_free(argv);
1874                 goto out;
1875         }
1876
1877         call_usermodehelper_setcleanup(info, argv_cleanup);
1878
1879         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1880
1881   out:
1882         if (ret && force) {
1883                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1884                        "forcing the issue\n");
1885
1886                 /* I guess this should try to kick off some daemon to
1887                    sync and poweroff asap.  Or not even bother syncing
1888                    if we're doing an emergency shutdown? */
1889                 emergency_sync();
1890                 kernel_power_off();
1891         }
1892
1893         return ret;
1894 }
1895 EXPORT_SYMBOL_GPL(orderly_poweroff);