mm: remove last trace of shmem_get_unmapped_area
[linux-2.6.git] / kernel / sys.c
1 /*
2  *  linux/kernel/sys.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/module.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/notifier.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
15 #include <linux/fs.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/kexec.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/gfp.h>
40 #include <linux/syscore_ops.h>
41
42 #include <linux/compat.h>
43 #include <linux/syscalls.h>
44 #include <linux/kprobes.h>
45 #include <linux/user_namespace.h>
46
47 #include <linux/kmsg_dump.h>
48
49 #include <asm/uaccess.h>
50 #include <asm/io.h>
51 #include <asm/unistd.h>
52
53 #ifndef SET_UNALIGN_CTL
54 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
55 #endif
56 #ifndef GET_UNALIGN_CTL
57 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
58 #endif
59 #ifndef SET_FPEMU_CTL
60 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
61 #endif
62 #ifndef GET_FPEMU_CTL
63 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
64 #endif
65 #ifndef SET_FPEXC_CTL
66 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
67 #endif
68 #ifndef GET_FPEXC_CTL
69 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
70 #endif
71 #ifndef GET_ENDIAN
72 # define GET_ENDIAN(a,b)        (-EINVAL)
73 #endif
74 #ifndef SET_ENDIAN
75 # define SET_ENDIAN(a,b)        (-EINVAL)
76 #endif
77 #ifndef GET_TSC_CTL
78 # define GET_TSC_CTL(a)         (-EINVAL)
79 #endif
80 #ifndef SET_TSC_CTL
81 # define SET_TSC_CTL(a)         (-EINVAL)
82 #endif
83
84 /*
85  * this is where the system-wide overflow UID and GID are defined, for
86  * architectures that now have 32-bit UID/GID but didn't in the past
87  */
88
89 int overflowuid = DEFAULT_OVERFLOWUID;
90 int overflowgid = DEFAULT_OVERFLOWGID;
91
92 #ifdef CONFIG_UID16
93 EXPORT_SYMBOL(overflowuid);
94 EXPORT_SYMBOL(overflowgid);
95 #endif
96
97 /*
98  * the same as above, but for filesystems which can only store a 16-bit
99  * UID and GID. as such, this is needed on all architectures
100  */
101
102 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
103 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
104
105 EXPORT_SYMBOL(fs_overflowuid);
106 EXPORT_SYMBOL(fs_overflowgid);
107
108 /*
109  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
110  */
111
112 int C_A_D = 1;
113 struct pid *cad_pid;
114 EXPORT_SYMBOL(cad_pid);
115
116 /*
117  * If set, this is used for preparing the system to power off.
118  */
119
120 void (*pm_power_off_prepare)(void);
121
122 /*
123  * Returns true if current's euid is same as p's uid or euid,
124  * or has CAP_SYS_NICE to p's user_ns.
125  *
126  * Called with rcu_read_lock, creds are safe
127  */
128 static bool set_one_prio_perm(struct task_struct *p)
129 {
130         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
131
132         if (pcred->user->user_ns == cred->user->user_ns &&
133             (pcred->uid  == cred->euid ||
134              pcred->euid == cred->euid))
135                 return true;
136         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
137                 return true;
138         return false;
139 }
140
141 /*
142  * set the priority of a task
143  * - the caller must hold the RCU read lock
144  */
145 static int set_one_prio(struct task_struct *p, int niceval, int error)
146 {
147         int no_nice;
148
149         if (!set_one_prio_perm(p)) {
150                 error = -EPERM;
151                 goto out;
152         }
153         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
154                 error = -EACCES;
155                 goto out;
156         }
157         no_nice = security_task_setnice(p, niceval);
158         if (no_nice) {
159                 error = no_nice;
160                 goto out;
161         }
162         if (error == -ESRCH)
163                 error = 0;
164         set_user_nice(p, niceval);
165 out:
166         return error;
167 }
168
169 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
170 {
171         struct task_struct *g, *p;
172         struct user_struct *user;
173         const struct cred *cred = current_cred();
174         int error = -EINVAL;
175         struct pid *pgrp;
176
177         if (which > PRIO_USER || which < PRIO_PROCESS)
178                 goto out;
179
180         /* normalize: avoid signed division (rounding problems) */
181         error = -ESRCH;
182         if (niceval < -20)
183                 niceval = -20;
184         if (niceval > 19)
185                 niceval = 19;
186
187         rcu_read_lock();
188         read_lock(&tasklist_lock);
189         switch (which) {
190                 case PRIO_PROCESS:
191                         if (who)
192                                 p = find_task_by_vpid(who);
193                         else
194                                 p = current;
195                         if (p)
196                                 error = set_one_prio(p, niceval, error);
197                         break;
198                 case PRIO_PGRP:
199                         if (who)
200                                 pgrp = find_vpid(who);
201                         else
202                                 pgrp = task_pgrp(current);
203                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
204                                 error = set_one_prio(p, niceval, error);
205                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
206                         break;
207                 case PRIO_USER:
208                         user = (struct user_struct *) cred->user;
209                         if (!who)
210                                 who = cred->uid;
211                         else if ((who != cred->uid) &&
212                                  !(user = find_user(who)))
213                                 goto out_unlock;        /* No processes for this user */
214
215                         do_each_thread(g, p) {
216                                 if (__task_cred(p)->uid == who)
217                                         error = set_one_prio(p, niceval, error);
218                         } while_each_thread(g, p);
219                         if (who != cred->uid)
220                                 free_uid(user);         /* For find_user() */
221                         break;
222         }
223 out_unlock:
224         read_unlock(&tasklist_lock);
225         rcu_read_unlock();
226 out:
227         return error;
228 }
229
230 /*
231  * Ugh. To avoid negative return values, "getpriority()" will
232  * not return the normal nice-value, but a negated value that
233  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
234  * to stay compatible.
235  */
236 SYSCALL_DEFINE2(getpriority, int, which, int, who)
237 {
238         struct task_struct *g, *p;
239         struct user_struct *user;
240         const struct cred *cred = current_cred();
241         long niceval, retval = -ESRCH;
242         struct pid *pgrp;
243
244         if (which > PRIO_USER || which < PRIO_PROCESS)
245                 return -EINVAL;
246
247         rcu_read_lock();
248         read_lock(&tasklist_lock);
249         switch (which) {
250                 case PRIO_PROCESS:
251                         if (who)
252                                 p = find_task_by_vpid(who);
253                         else
254                                 p = current;
255                         if (p) {
256                                 niceval = 20 - task_nice(p);
257                                 if (niceval > retval)
258                                         retval = niceval;
259                         }
260                         break;
261                 case PRIO_PGRP:
262                         if (who)
263                                 pgrp = find_vpid(who);
264                         else
265                                 pgrp = task_pgrp(current);
266                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
267                                 niceval = 20 - task_nice(p);
268                                 if (niceval > retval)
269                                         retval = niceval;
270                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
271                         break;
272                 case PRIO_USER:
273                         user = (struct user_struct *) cred->user;
274                         if (!who)
275                                 who = cred->uid;
276                         else if ((who != cred->uid) &&
277                                  !(user = find_user(who)))
278                                 goto out_unlock;        /* No processes for this user */
279
280                         do_each_thread(g, p) {
281                                 if (__task_cred(p)->uid == who) {
282                                         niceval = 20 - task_nice(p);
283                                         if (niceval > retval)
284                                                 retval = niceval;
285                                 }
286                         } while_each_thread(g, p);
287                         if (who != cred->uid)
288                                 free_uid(user);         /* for find_user() */
289                         break;
290         }
291 out_unlock:
292         read_unlock(&tasklist_lock);
293         rcu_read_unlock();
294
295         return retval;
296 }
297
298 /**
299  *      emergency_restart - reboot the system
300  *
301  *      Without shutting down any hardware or taking any locks
302  *      reboot the system.  This is called when we know we are in
303  *      trouble so this is our best effort to reboot.  This is
304  *      safe to call in interrupt context.
305  */
306 void emergency_restart(void)
307 {
308         kmsg_dump(KMSG_DUMP_EMERG);
309         machine_emergency_restart();
310 }
311 EXPORT_SYMBOL_GPL(emergency_restart);
312
313 void kernel_restart_prepare(char *cmd)
314 {
315         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
316         system_state = SYSTEM_RESTART;
317         usermodehelper_disable();
318         device_shutdown();
319         syscore_shutdown();
320 }
321
322 /**
323  *      kernel_restart - reboot the system
324  *      @cmd: pointer to buffer containing command to execute for restart
325  *              or %NULL
326  *
327  *      Shutdown everything and perform a clean reboot.
328  *      This is not safe to call in interrupt context.
329  */
330 void kernel_restart(char *cmd)
331 {
332         kernel_restart_prepare(cmd);
333         if (!cmd)
334                 printk(KERN_EMERG "Restarting system.\n");
335         else
336                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
337         kmsg_dump(KMSG_DUMP_RESTART);
338         machine_restart(cmd);
339 }
340 EXPORT_SYMBOL_GPL(kernel_restart);
341
342 static void kernel_shutdown_prepare(enum system_states state)
343 {
344         blocking_notifier_call_chain(&reboot_notifier_list,
345                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
346         system_state = state;
347         usermodehelper_disable();
348         device_shutdown();
349 }
350 /**
351  *      kernel_halt - halt the system
352  *
353  *      Shutdown everything and perform a clean system halt.
354  */
355 void kernel_halt(void)
356 {
357         kernel_shutdown_prepare(SYSTEM_HALT);
358         syscore_shutdown();
359         printk(KERN_EMERG "System halted.\n");
360         kmsg_dump(KMSG_DUMP_HALT);
361         machine_halt();
362 }
363
364 EXPORT_SYMBOL_GPL(kernel_halt);
365
366 /**
367  *      kernel_power_off - power_off the system
368  *
369  *      Shutdown everything and perform a clean system power_off.
370  */
371 void kernel_power_off(void)
372 {
373         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
374         if (pm_power_off_prepare)
375                 pm_power_off_prepare();
376         disable_nonboot_cpus();
377         syscore_shutdown();
378         printk(KERN_EMERG "Power down.\n");
379         kmsg_dump(KMSG_DUMP_POWEROFF);
380         machine_power_off();
381 }
382 EXPORT_SYMBOL_GPL(kernel_power_off);
383
384 static DEFINE_MUTEX(reboot_mutex);
385
386 /*
387  * Reboot system call: for obvious reasons only root may call it,
388  * and even root needs to set up some magic numbers in the registers
389  * so that some mistake won't make this reboot the whole machine.
390  * You can also set the meaning of the ctrl-alt-del-key here.
391  *
392  * reboot doesn't sync: do that yourself before calling this.
393  */
394 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
395                 void __user *, arg)
396 {
397         char buffer[256];
398         int ret = 0;
399
400         /* We only trust the superuser with rebooting the system. */
401         if (!capable(CAP_SYS_BOOT))
402                 return -EPERM;
403
404         /* For safety, we require "magic" arguments. */
405         if (magic1 != LINUX_REBOOT_MAGIC1 ||
406             (magic2 != LINUX_REBOOT_MAGIC2 &&
407                         magic2 != LINUX_REBOOT_MAGIC2A &&
408                         magic2 != LINUX_REBOOT_MAGIC2B &&
409                         magic2 != LINUX_REBOOT_MAGIC2C))
410                 return -EINVAL;
411
412         /* Instead of trying to make the power_off code look like
413          * halt when pm_power_off is not set do it the easy way.
414          */
415         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
416                 cmd = LINUX_REBOOT_CMD_HALT;
417
418         mutex_lock(&reboot_mutex);
419         switch (cmd) {
420         case LINUX_REBOOT_CMD_RESTART:
421                 kernel_restart(NULL);
422                 break;
423
424         case LINUX_REBOOT_CMD_CAD_ON:
425                 C_A_D = 1;
426                 break;
427
428         case LINUX_REBOOT_CMD_CAD_OFF:
429                 C_A_D = 0;
430                 break;
431
432         case LINUX_REBOOT_CMD_HALT:
433                 kernel_halt();
434                 do_exit(0);
435                 panic("cannot halt");
436
437         case LINUX_REBOOT_CMD_POWER_OFF:
438                 kernel_power_off();
439                 do_exit(0);
440                 break;
441
442         case LINUX_REBOOT_CMD_RESTART2:
443                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
444                         ret = -EFAULT;
445                         break;
446                 }
447                 buffer[sizeof(buffer) - 1] = '\0';
448
449                 kernel_restart(buffer);
450                 break;
451
452 #ifdef CONFIG_KEXEC
453         case LINUX_REBOOT_CMD_KEXEC:
454                 ret = kernel_kexec();
455                 break;
456 #endif
457
458 #ifdef CONFIG_HIBERNATION
459         case LINUX_REBOOT_CMD_SW_SUSPEND:
460                 ret = hibernate();
461                 break;
462 #endif
463
464         default:
465                 ret = -EINVAL;
466                 break;
467         }
468         mutex_unlock(&reboot_mutex);
469         return ret;
470 }
471
472 static void deferred_cad(struct work_struct *dummy)
473 {
474         kernel_restart(NULL);
475 }
476
477 /*
478  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
479  * As it's called within an interrupt, it may NOT sync: the only choice
480  * is whether to reboot at once, or just ignore the ctrl-alt-del.
481  */
482 void ctrl_alt_del(void)
483 {
484         static DECLARE_WORK(cad_work, deferred_cad);
485
486         if (C_A_D)
487                 schedule_work(&cad_work);
488         else
489                 kill_cad_pid(SIGINT, 1);
490 }
491         
492 /*
493  * Unprivileged users may change the real gid to the effective gid
494  * or vice versa.  (BSD-style)
495  *
496  * If you set the real gid at all, or set the effective gid to a value not
497  * equal to the real gid, then the saved gid is set to the new effective gid.
498  *
499  * This makes it possible for a setgid program to completely drop its
500  * privileges, which is often a useful assertion to make when you are doing
501  * a security audit over a program.
502  *
503  * The general idea is that a program which uses just setregid() will be
504  * 100% compatible with BSD.  A program which uses just setgid() will be
505  * 100% compatible with POSIX with saved IDs. 
506  *
507  * SMP: There are not races, the GIDs are checked only by filesystem
508  *      operations (as far as semantic preservation is concerned).
509  */
510 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
511 {
512         const struct cred *old;
513         struct cred *new;
514         int retval;
515
516         new = prepare_creds();
517         if (!new)
518                 return -ENOMEM;
519         old = current_cred();
520
521         retval = -EPERM;
522         if (rgid != (gid_t) -1) {
523                 if (old->gid == rgid ||
524                     old->egid == rgid ||
525                     nsown_capable(CAP_SETGID))
526                         new->gid = rgid;
527                 else
528                         goto error;
529         }
530         if (egid != (gid_t) -1) {
531                 if (old->gid == egid ||
532                     old->egid == egid ||
533                     old->sgid == egid ||
534                     nsown_capable(CAP_SETGID))
535                         new->egid = egid;
536                 else
537                         goto error;
538         }
539
540         if (rgid != (gid_t) -1 ||
541             (egid != (gid_t) -1 && egid != old->gid))
542                 new->sgid = new->egid;
543         new->fsgid = new->egid;
544
545         return commit_creds(new);
546
547 error:
548         abort_creds(new);
549         return retval;
550 }
551
552 /*
553  * setgid() is implemented like SysV w/ SAVED_IDS 
554  *
555  * SMP: Same implicit races as above.
556  */
557 SYSCALL_DEFINE1(setgid, gid_t, gid)
558 {
559         const struct cred *old;
560         struct cred *new;
561         int retval;
562
563         new = prepare_creds();
564         if (!new)
565                 return -ENOMEM;
566         old = current_cred();
567
568         retval = -EPERM;
569         if (nsown_capable(CAP_SETGID))
570                 new->gid = new->egid = new->sgid = new->fsgid = gid;
571         else if (gid == old->gid || gid == old->sgid)
572                 new->egid = new->fsgid = gid;
573         else
574                 goto error;
575
576         return commit_creds(new);
577
578 error:
579         abort_creds(new);
580         return retval;
581 }
582
583 /*
584  * change the user struct in a credentials set to match the new UID
585  */
586 static int set_user(struct cred *new)
587 {
588         struct user_struct *new_user;
589
590         new_user = alloc_uid(current_user_ns(), new->uid);
591         if (!new_user)
592                 return -EAGAIN;
593
594         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
595                         new_user != INIT_USER) {
596                 free_uid(new_user);
597                 return -EAGAIN;
598         }
599
600         free_uid(new->user);
601         new->user = new_user;
602         return 0;
603 }
604
605 /*
606  * Unprivileged users may change the real uid to the effective uid
607  * or vice versa.  (BSD-style)
608  *
609  * If you set the real uid at all, or set the effective uid to a value not
610  * equal to the real uid, then the saved uid is set to the new effective uid.
611  *
612  * This makes it possible for a setuid program to completely drop its
613  * privileges, which is often a useful assertion to make when you are doing
614  * a security audit over a program.
615  *
616  * The general idea is that a program which uses just setreuid() will be
617  * 100% compatible with BSD.  A program which uses just setuid() will be
618  * 100% compatible with POSIX with saved IDs. 
619  */
620 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
621 {
622         const struct cred *old;
623         struct cred *new;
624         int retval;
625
626         new = prepare_creds();
627         if (!new)
628                 return -ENOMEM;
629         old = current_cred();
630
631         retval = -EPERM;
632         if (ruid != (uid_t) -1) {
633                 new->uid = ruid;
634                 if (old->uid != ruid &&
635                     old->euid != ruid &&
636                     !nsown_capable(CAP_SETUID))
637                         goto error;
638         }
639
640         if (euid != (uid_t) -1) {
641                 new->euid = euid;
642                 if (old->uid != euid &&
643                     old->euid != euid &&
644                     old->suid != euid &&
645                     !nsown_capable(CAP_SETUID))
646                         goto error;
647         }
648
649         if (new->uid != old->uid) {
650                 retval = set_user(new);
651                 if (retval < 0)
652                         goto error;
653         }
654         if (ruid != (uid_t) -1 ||
655             (euid != (uid_t) -1 && euid != old->uid))
656                 new->suid = new->euid;
657         new->fsuid = new->euid;
658
659         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
660         if (retval < 0)
661                 goto error;
662
663         return commit_creds(new);
664
665 error:
666         abort_creds(new);
667         return retval;
668 }
669                 
670 /*
671  * setuid() is implemented like SysV with SAVED_IDS 
672  * 
673  * Note that SAVED_ID's is deficient in that a setuid root program
674  * like sendmail, for example, cannot set its uid to be a normal 
675  * user and then switch back, because if you're root, setuid() sets
676  * the saved uid too.  If you don't like this, blame the bright people
677  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
678  * will allow a root program to temporarily drop privileges and be able to
679  * regain them by swapping the real and effective uid.  
680  */
681 SYSCALL_DEFINE1(setuid, uid_t, uid)
682 {
683         const struct cred *old;
684         struct cred *new;
685         int retval;
686
687         new = prepare_creds();
688         if (!new)
689                 return -ENOMEM;
690         old = current_cred();
691
692         retval = -EPERM;
693         if (nsown_capable(CAP_SETUID)) {
694                 new->suid = new->uid = uid;
695                 if (uid != old->uid) {
696                         retval = set_user(new);
697                         if (retval < 0)
698                                 goto error;
699                 }
700         } else if (uid != old->uid && uid != new->suid) {
701                 goto error;
702         }
703
704         new->fsuid = new->euid = uid;
705
706         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
707         if (retval < 0)
708                 goto error;
709
710         return commit_creds(new);
711
712 error:
713         abort_creds(new);
714         return retval;
715 }
716
717
718 /*
719  * This function implements a generic ability to update ruid, euid,
720  * and suid.  This allows you to implement the 4.4 compatible seteuid().
721  */
722 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
723 {
724         const struct cred *old;
725         struct cred *new;
726         int retval;
727
728         new = prepare_creds();
729         if (!new)
730                 return -ENOMEM;
731
732         old = current_cred();
733
734         retval = -EPERM;
735         if (!nsown_capable(CAP_SETUID)) {
736                 if (ruid != (uid_t) -1 && ruid != old->uid &&
737                     ruid != old->euid  && ruid != old->suid)
738                         goto error;
739                 if (euid != (uid_t) -1 && euid != old->uid &&
740                     euid != old->euid  && euid != old->suid)
741                         goto error;
742                 if (suid != (uid_t) -1 && suid != old->uid &&
743                     suid != old->euid  && suid != old->suid)
744                         goto error;
745         }
746
747         if (ruid != (uid_t) -1) {
748                 new->uid = ruid;
749                 if (ruid != old->uid) {
750                         retval = set_user(new);
751                         if (retval < 0)
752                                 goto error;
753                 }
754         }
755         if (euid != (uid_t) -1)
756                 new->euid = euid;
757         if (suid != (uid_t) -1)
758                 new->suid = suid;
759         new->fsuid = new->euid;
760
761         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
762         if (retval < 0)
763                 goto error;
764
765         return commit_creds(new);
766
767 error:
768         abort_creds(new);
769         return retval;
770 }
771
772 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
773 {
774         const struct cred *cred = current_cred();
775         int retval;
776
777         if (!(retval   = put_user(cred->uid,  ruid)) &&
778             !(retval   = put_user(cred->euid, euid)))
779                 retval = put_user(cred->suid, suid);
780
781         return retval;
782 }
783
784 /*
785  * Same as above, but for rgid, egid, sgid.
786  */
787 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
788 {
789         const struct cred *old;
790         struct cred *new;
791         int retval;
792
793         new = prepare_creds();
794         if (!new)
795                 return -ENOMEM;
796         old = current_cred();
797
798         retval = -EPERM;
799         if (!nsown_capable(CAP_SETGID)) {
800                 if (rgid != (gid_t) -1 && rgid != old->gid &&
801                     rgid != old->egid  && rgid != old->sgid)
802                         goto error;
803                 if (egid != (gid_t) -1 && egid != old->gid &&
804                     egid != old->egid  && egid != old->sgid)
805                         goto error;
806                 if (sgid != (gid_t) -1 && sgid != old->gid &&
807                     sgid != old->egid  && sgid != old->sgid)
808                         goto error;
809         }
810
811         if (rgid != (gid_t) -1)
812                 new->gid = rgid;
813         if (egid != (gid_t) -1)
814                 new->egid = egid;
815         if (sgid != (gid_t) -1)
816                 new->sgid = sgid;
817         new->fsgid = new->egid;
818
819         return commit_creds(new);
820
821 error:
822         abort_creds(new);
823         return retval;
824 }
825
826 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
827 {
828         const struct cred *cred = current_cred();
829         int retval;
830
831         if (!(retval   = put_user(cred->gid,  rgid)) &&
832             !(retval   = put_user(cred->egid, egid)))
833                 retval = put_user(cred->sgid, sgid);
834
835         return retval;
836 }
837
838
839 /*
840  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
841  * is used for "access()" and for the NFS daemon (letting nfsd stay at
842  * whatever uid it wants to). It normally shadows "euid", except when
843  * explicitly set by setfsuid() or for access..
844  */
845 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
846 {
847         const struct cred *old;
848         struct cred *new;
849         uid_t old_fsuid;
850
851         new = prepare_creds();
852         if (!new)
853                 return current_fsuid();
854         old = current_cred();
855         old_fsuid = old->fsuid;
856
857         if (uid == old->uid  || uid == old->euid  ||
858             uid == old->suid || uid == old->fsuid ||
859             nsown_capable(CAP_SETUID)) {
860                 if (uid != old_fsuid) {
861                         new->fsuid = uid;
862                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
863                                 goto change_okay;
864                 }
865         }
866
867         abort_creds(new);
868         return old_fsuid;
869
870 change_okay:
871         commit_creds(new);
872         return old_fsuid;
873 }
874
875 /*
876  * Samma på svenska..
877  */
878 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
879 {
880         const struct cred *old;
881         struct cred *new;
882         gid_t old_fsgid;
883
884         new = prepare_creds();
885         if (!new)
886                 return current_fsgid();
887         old = current_cred();
888         old_fsgid = old->fsgid;
889
890         if (gid == old->gid  || gid == old->egid  ||
891             gid == old->sgid || gid == old->fsgid ||
892             nsown_capable(CAP_SETGID)) {
893                 if (gid != old_fsgid) {
894                         new->fsgid = gid;
895                         goto change_okay;
896                 }
897         }
898
899         abort_creds(new);
900         return old_fsgid;
901
902 change_okay:
903         commit_creds(new);
904         return old_fsgid;
905 }
906
907 void do_sys_times(struct tms *tms)
908 {
909         cputime_t tgutime, tgstime, cutime, cstime;
910
911         spin_lock_irq(&current->sighand->siglock);
912         thread_group_times(current, &tgutime, &tgstime);
913         cutime = current->signal->cutime;
914         cstime = current->signal->cstime;
915         spin_unlock_irq(&current->sighand->siglock);
916         tms->tms_utime = cputime_to_clock_t(tgutime);
917         tms->tms_stime = cputime_to_clock_t(tgstime);
918         tms->tms_cutime = cputime_to_clock_t(cutime);
919         tms->tms_cstime = cputime_to_clock_t(cstime);
920 }
921
922 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
923 {
924         if (tbuf) {
925                 struct tms tmp;
926
927                 do_sys_times(&tmp);
928                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
929                         return -EFAULT;
930         }
931         force_successful_syscall_return();
932         return (long) jiffies_64_to_clock_t(get_jiffies_64());
933 }
934
935 /*
936  * This needs some heavy checking ...
937  * I just haven't the stomach for it. I also don't fully
938  * understand sessions/pgrp etc. Let somebody who does explain it.
939  *
940  * OK, I think I have the protection semantics right.... this is really
941  * only important on a multi-user system anyway, to make sure one user
942  * can't send a signal to a process owned by another.  -TYT, 12/12/91
943  *
944  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
945  * LBT 04.03.94
946  */
947 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
948 {
949         struct task_struct *p;
950         struct task_struct *group_leader = current->group_leader;
951         struct pid *pgrp;
952         int err;
953
954         if (!pid)
955                 pid = task_pid_vnr(group_leader);
956         if (!pgid)
957                 pgid = pid;
958         if (pgid < 0)
959                 return -EINVAL;
960         rcu_read_lock();
961
962         /* From this point forward we keep holding onto the tasklist lock
963          * so that our parent does not change from under us. -DaveM
964          */
965         write_lock_irq(&tasklist_lock);
966
967         err = -ESRCH;
968         p = find_task_by_vpid(pid);
969         if (!p)
970                 goto out;
971
972         err = -EINVAL;
973         if (!thread_group_leader(p))
974                 goto out;
975
976         if (same_thread_group(p->real_parent, group_leader)) {
977                 err = -EPERM;
978                 if (task_session(p) != task_session(group_leader))
979                         goto out;
980                 err = -EACCES;
981                 if (p->did_exec)
982                         goto out;
983         } else {
984                 err = -ESRCH;
985                 if (p != group_leader)
986                         goto out;
987         }
988
989         err = -EPERM;
990         if (p->signal->leader)
991                 goto out;
992
993         pgrp = task_pid(p);
994         if (pgid != pid) {
995                 struct task_struct *g;
996
997                 pgrp = find_vpid(pgid);
998                 g = pid_task(pgrp, PIDTYPE_PGID);
999                 if (!g || task_session(g) != task_session(group_leader))
1000                         goto out;
1001         }
1002
1003         err = security_task_setpgid(p, pgid);
1004         if (err)
1005                 goto out;
1006
1007         if (task_pgrp(p) != pgrp)
1008                 change_pid(p, PIDTYPE_PGID, pgrp);
1009
1010         err = 0;
1011 out:
1012         /* All paths lead to here, thus we are safe. -DaveM */
1013         write_unlock_irq(&tasklist_lock);
1014         rcu_read_unlock();
1015         return err;
1016 }
1017
1018 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1019 {
1020         struct task_struct *p;
1021         struct pid *grp;
1022         int retval;
1023
1024         rcu_read_lock();
1025         if (!pid)
1026                 grp = task_pgrp(current);
1027         else {
1028                 retval = -ESRCH;
1029                 p = find_task_by_vpid(pid);
1030                 if (!p)
1031                         goto out;
1032                 grp = task_pgrp(p);
1033                 if (!grp)
1034                         goto out;
1035
1036                 retval = security_task_getpgid(p);
1037                 if (retval)
1038                         goto out;
1039         }
1040         retval = pid_vnr(grp);
1041 out:
1042         rcu_read_unlock();
1043         return retval;
1044 }
1045
1046 #ifdef __ARCH_WANT_SYS_GETPGRP
1047
1048 SYSCALL_DEFINE0(getpgrp)
1049 {
1050         return sys_getpgid(0);
1051 }
1052
1053 #endif
1054
1055 SYSCALL_DEFINE1(getsid, pid_t, pid)
1056 {
1057         struct task_struct *p;
1058         struct pid *sid;
1059         int retval;
1060
1061         rcu_read_lock();
1062         if (!pid)
1063                 sid = task_session(current);
1064         else {
1065                 retval = -ESRCH;
1066                 p = find_task_by_vpid(pid);
1067                 if (!p)
1068                         goto out;
1069                 sid = task_session(p);
1070                 if (!sid)
1071                         goto out;
1072
1073                 retval = security_task_getsid(p);
1074                 if (retval)
1075                         goto out;
1076         }
1077         retval = pid_vnr(sid);
1078 out:
1079         rcu_read_unlock();
1080         return retval;
1081 }
1082
1083 SYSCALL_DEFINE0(setsid)
1084 {
1085         struct task_struct *group_leader = current->group_leader;
1086         struct pid *sid = task_pid(group_leader);
1087         pid_t session = pid_vnr(sid);
1088         int err = -EPERM;
1089
1090         write_lock_irq(&tasklist_lock);
1091         /* Fail if I am already a session leader */
1092         if (group_leader->signal->leader)
1093                 goto out;
1094
1095         /* Fail if a process group id already exists that equals the
1096          * proposed session id.
1097          */
1098         if (pid_task(sid, PIDTYPE_PGID))
1099                 goto out;
1100
1101         group_leader->signal->leader = 1;
1102         __set_special_pids(sid);
1103
1104         proc_clear_tty(group_leader);
1105
1106         err = session;
1107 out:
1108         write_unlock_irq(&tasklist_lock);
1109         if (err > 0) {
1110                 proc_sid_connector(group_leader);
1111                 sched_autogroup_create_attach(group_leader);
1112         }
1113         return err;
1114 }
1115
1116 DECLARE_RWSEM(uts_sem);
1117
1118 #ifdef COMPAT_UTS_MACHINE
1119 #define override_architecture(name) \
1120         (personality(current->personality) == PER_LINUX32 && \
1121          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1122                       sizeof(COMPAT_UTS_MACHINE)))
1123 #else
1124 #define override_architecture(name)     0
1125 #endif
1126
1127 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1128 {
1129         int errno = 0;
1130
1131         down_read(&uts_sem);
1132         if (copy_to_user(name, utsname(), sizeof *name))
1133                 errno = -EFAULT;
1134         up_read(&uts_sem);
1135
1136         if (!errno && override_architecture(name))
1137                 errno = -EFAULT;
1138         return errno;
1139 }
1140
1141 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1142 /*
1143  * Old cruft
1144  */
1145 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1146 {
1147         int error = 0;
1148
1149         if (!name)
1150                 return -EFAULT;
1151
1152         down_read(&uts_sem);
1153         if (copy_to_user(name, utsname(), sizeof(*name)))
1154                 error = -EFAULT;
1155         up_read(&uts_sem);
1156
1157         if (!error && override_architecture(name))
1158                 error = -EFAULT;
1159         return error;
1160 }
1161
1162 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1163 {
1164         int error;
1165
1166         if (!name)
1167                 return -EFAULT;
1168         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1169                 return -EFAULT;
1170
1171         down_read(&uts_sem);
1172         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1173                                __OLD_UTS_LEN);
1174         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1175         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1176                                 __OLD_UTS_LEN);
1177         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1178         error |= __copy_to_user(&name->release, &utsname()->release,
1179                                 __OLD_UTS_LEN);
1180         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1181         error |= __copy_to_user(&name->version, &utsname()->version,
1182                                 __OLD_UTS_LEN);
1183         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1184         error |= __copy_to_user(&name->machine, &utsname()->machine,
1185                                 __OLD_UTS_LEN);
1186         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1187         up_read(&uts_sem);
1188
1189         if (!error && override_architecture(name))
1190                 error = -EFAULT;
1191         return error ? -EFAULT : 0;
1192 }
1193 #endif
1194
1195 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1196 {
1197         int errno;
1198         char tmp[__NEW_UTS_LEN];
1199
1200         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1201                 return -EPERM;
1202
1203         if (len < 0 || len > __NEW_UTS_LEN)
1204                 return -EINVAL;
1205         down_write(&uts_sem);
1206         errno = -EFAULT;
1207         if (!copy_from_user(tmp, name, len)) {
1208                 struct new_utsname *u = utsname();
1209
1210                 memcpy(u->nodename, tmp, len);
1211                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1212                 errno = 0;
1213         }
1214         up_write(&uts_sem);
1215         return errno;
1216 }
1217
1218 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1219
1220 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1221 {
1222         int i, errno;
1223         struct new_utsname *u;
1224
1225         if (len < 0)
1226                 return -EINVAL;
1227         down_read(&uts_sem);
1228         u = utsname();
1229         i = 1 + strlen(u->nodename);
1230         if (i > len)
1231                 i = len;
1232         errno = 0;
1233         if (copy_to_user(name, u->nodename, i))
1234                 errno = -EFAULT;
1235         up_read(&uts_sem);
1236         return errno;
1237 }
1238
1239 #endif
1240
1241 /*
1242  * Only setdomainname; getdomainname can be implemented by calling
1243  * uname()
1244  */
1245 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1246 {
1247         int errno;
1248         char tmp[__NEW_UTS_LEN];
1249
1250         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1251                 return -EPERM;
1252         if (len < 0 || len > __NEW_UTS_LEN)
1253                 return -EINVAL;
1254
1255         down_write(&uts_sem);
1256         errno = -EFAULT;
1257         if (!copy_from_user(tmp, name, len)) {
1258                 struct new_utsname *u = utsname();
1259
1260                 memcpy(u->domainname, tmp, len);
1261                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1262                 errno = 0;
1263         }
1264         up_write(&uts_sem);
1265         return errno;
1266 }
1267
1268 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1269 {
1270         struct rlimit value;
1271         int ret;
1272
1273         ret = do_prlimit(current, resource, NULL, &value);
1274         if (!ret)
1275                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1276
1277         return ret;
1278 }
1279
1280 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1281
1282 /*
1283  *      Back compatibility for getrlimit. Needed for some apps.
1284  */
1285  
1286 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1287                 struct rlimit __user *, rlim)
1288 {
1289         struct rlimit x;
1290         if (resource >= RLIM_NLIMITS)
1291                 return -EINVAL;
1292
1293         task_lock(current->group_leader);
1294         x = current->signal->rlim[resource];
1295         task_unlock(current->group_leader);
1296         if (x.rlim_cur > 0x7FFFFFFF)
1297                 x.rlim_cur = 0x7FFFFFFF;
1298         if (x.rlim_max > 0x7FFFFFFF)
1299                 x.rlim_max = 0x7FFFFFFF;
1300         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1301 }
1302
1303 #endif
1304
1305 static inline bool rlim64_is_infinity(__u64 rlim64)
1306 {
1307 #if BITS_PER_LONG < 64
1308         return rlim64 >= ULONG_MAX;
1309 #else
1310         return rlim64 == RLIM64_INFINITY;
1311 #endif
1312 }
1313
1314 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1315 {
1316         if (rlim->rlim_cur == RLIM_INFINITY)
1317                 rlim64->rlim_cur = RLIM64_INFINITY;
1318         else
1319                 rlim64->rlim_cur = rlim->rlim_cur;
1320         if (rlim->rlim_max == RLIM_INFINITY)
1321                 rlim64->rlim_max = RLIM64_INFINITY;
1322         else
1323                 rlim64->rlim_max = rlim->rlim_max;
1324 }
1325
1326 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1327 {
1328         if (rlim64_is_infinity(rlim64->rlim_cur))
1329                 rlim->rlim_cur = RLIM_INFINITY;
1330         else
1331                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1332         if (rlim64_is_infinity(rlim64->rlim_max))
1333                 rlim->rlim_max = RLIM_INFINITY;
1334         else
1335                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1336 }
1337
1338 /* make sure you are allowed to change @tsk limits before calling this */
1339 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1340                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1341 {
1342         struct rlimit *rlim;
1343         int retval = 0;
1344
1345         if (resource >= RLIM_NLIMITS)
1346                 return -EINVAL;
1347         if (new_rlim) {
1348                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1349                         return -EINVAL;
1350                 if (resource == RLIMIT_NOFILE &&
1351                                 new_rlim->rlim_max > sysctl_nr_open)
1352                         return -EPERM;
1353         }
1354
1355         /* protect tsk->signal and tsk->sighand from disappearing */
1356         read_lock(&tasklist_lock);
1357         if (!tsk->sighand) {
1358                 retval = -ESRCH;
1359                 goto out;
1360         }
1361
1362         rlim = tsk->signal->rlim + resource;
1363         task_lock(tsk->group_leader);
1364         if (new_rlim) {
1365                 /* Keep the capable check against init_user_ns until
1366                    cgroups can contain all limits */
1367                 if (new_rlim->rlim_max > rlim->rlim_max &&
1368                                 !capable(CAP_SYS_RESOURCE))
1369                         retval = -EPERM;
1370                 if (!retval)
1371                         retval = security_task_setrlimit(tsk->group_leader,
1372                                         resource, new_rlim);
1373                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1374                         /*
1375                          * The caller is asking for an immediate RLIMIT_CPU
1376                          * expiry.  But we use the zero value to mean "it was
1377                          * never set".  So let's cheat and make it one second
1378                          * instead
1379                          */
1380                         new_rlim->rlim_cur = 1;
1381                 }
1382         }
1383         if (!retval) {
1384                 if (old_rlim)
1385                         *old_rlim = *rlim;
1386                 if (new_rlim)
1387                         *rlim = *new_rlim;
1388         }
1389         task_unlock(tsk->group_leader);
1390
1391         /*
1392          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1393          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1394          * very long-standing error, and fixing it now risks breakage of
1395          * applications, so we live with it
1396          */
1397          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1398                          new_rlim->rlim_cur != RLIM_INFINITY)
1399                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1400 out:
1401         read_unlock(&tasklist_lock);
1402         return retval;
1403 }
1404
1405 /* rcu lock must be held */
1406 static int check_prlimit_permission(struct task_struct *task)
1407 {
1408         const struct cred *cred = current_cred(), *tcred;
1409
1410         if (current == task)
1411                 return 0;
1412
1413         tcred = __task_cred(task);
1414         if (cred->user->user_ns == tcred->user->user_ns &&
1415             (cred->uid == tcred->euid &&
1416              cred->uid == tcred->suid &&
1417              cred->uid == tcred->uid  &&
1418              cred->gid == tcred->egid &&
1419              cred->gid == tcred->sgid &&
1420              cred->gid == tcred->gid))
1421                 return 0;
1422         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1423                 return 0;
1424
1425         return -EPERM;
1426 }
1427
1428 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1429                 const struct rlimit64 __user *, new_rlim,
1430                 struct rlimit64 __user *, old_rlim)
1431 {
1432         struct rlimit64 old64, new64;
1433         struct rlimit old, new;
1434         struct task_struct *tsk;
1435         int ret;
1436
1437         if (new_rlim) {
1438                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1439                         return -EFAULT;
1440                 rlim64_to_rlim(&new64, &new);
1441         }
1442
1443         rcu_read_lock();
1444         tsk = pid ? find_task_by_vpid(pid) : current;
1445         if (!tsk) {
1446                 rcu_read_unlock();
1447                 return -ESRCH;
1448         }
1449         ret = check_prlimit_permission(tsk);
1450         if (ret) {
1451                 rcu_read_unlock();
1452                 return ret;
1453         }
1454         get_task_struct(tsk);
1455         rcu_read_unlock();
1456
1457         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1458                         old_rlim ? &old : NULL);
1459
1460         if (!ret && old_rlim) {
1461                 rlim_to_rlim64(&old, &old64);
1462                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1463                         ret = -EFAULT;
1464         }
1465
1466         put_task_struct(tsk);
1467         return ret;
1468 }
1469
1470 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1471 {
1472         struct rlimit new_rlim;
1473
1474         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1475                 return -EFAULT;
1476         return do_prlimit(current, resource, &new_rlim, NULL);
1477 }
1478
1479 /*
1480  * It would make sense to put struct rusage in the task_struct,
1481  * except that would make the task_struct be *really big*.  After
1482  * task_struct gets moved into malloc'ed memory, it would
1483  * make sense to do this.  It will make moving the rest of the information
1484  * a lot simpler!  (Which we're not doing right now because we're not
1485  * measuring them yet).
1486  *
1487  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1488  * races with threads incrementing their own counters.  But since word
1489  * reads are atomic, we either get new values or old values and we don't
1490  * care which for the sums.  We always take the siglock to protect reading
1491  * the c* fields from p->signal from races with exit.c updating those
1492  * fields when reaping, so a sample either gets all the additions of a
1493  * given child after it's reaped, or none so this sample is before reaping.
1494  *
1495  * Locking:
1496  * We need to take the siglock for CHILDEREN, SELF and BOTH
1497  * for  the cases current multithreaded, non-current single threaded
1498  * non-current multithreaded.  Thread traversal is now safe with
1499  * the siglock held.
1500  * Strictly speaking, we donot need to take the siglock if we are current and
1501  * single threaded,  as no one else can take our signal_struct away, no one
1502  * else can  reap the  children to update signal->c* counters, and no one else
1503  * can race with the signal-> fields. If we do not take any lock, the
1504  * signal-> fields could be read out of order while another thread was just
1505  * exiting. So we should  place a read memory barrier when we avoid the lock.
1506  * On the writer side,  write memory barrier is implied in  __exit_signal
1507  * as __exit_signal releases  the siglock spinlock after updating the signal->
1508  * fields. But we don't do this yet to keep things simple.
1509  *
1510  */
1511
1512 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1513 {
1514         r->ru_nvcsw += t->nvcsw;
1515         r->ru_nivcsw += t->nivcsw;
1516         r->ru_minflt += t->min_flt;
1517         r->ru_majflt += t->maj_flt;
1518         r->ru_inblock += task_io_get_inblock(t);
1519         r->ru_oublock += task_io_get_oublock(t);
1520 }
1521
1522 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1523 {
1524         struct task_struct *t;
1525         unsigned long flags;
1526         cputime_t tgutime, tgstime, utime, stime;
1527         unsigned long maxrss = 0;
1528
1529         memset((char *) r, 0, sizeof *r);
1530         utime = stime = cputime_zero;
1531
1532         if (who == RUSAGE_THREAD) {
1533                 task_times(current, &utime, &stime);
1534                 accumulate_thread_rusage(p, r);
1535                 maxrss = p->signal->maxrss;
1536                 goto out;
1537         }
1538
1539         if (!lock_task_sighand(p, &flags))
1540                 return;
1541
1542         switch (who) {
1543                 case RUSAGE_BOTH:
1544                 case RUSAGE_CHILDREN:
1545                         utime = p->signal->cutime;
1546                         stime = p->signal->cstime;
1547                         r->ru_nvcsw = p->signal->cnvcsw;
1548                         r->ru_nivcsw = p->signal->cnivcsw;
1549                         r->ru_minflt = p->signal->cmin_flt;
1550                         r->ru_majflt = p->signal->cmaj_flt;
1551                         r->ru_inblock = p->signal->cinblock;
1552                         r->ru_oublock = p->signal->coublock;
1553                         maxrss = p->signal->cmaxrss;
1554
1555                         if (who == RUSAGE_CHILDREN)
1556                                 break;
1557
1558                 case RUSAGE_SELF:
1559                         thread_group_times(p, &tgutime, &tgstime);
1560                         utime = cputime_add(utime, tgutime);
1561                         stime = cputime_add(stime, tgstime);
1562                         r->ru_nvcsw += p->signal->nvcsw;
1563                         r->ru_nivcsw += p->signal->nivcsw;
1564                         r->ru_minflt += p->signal->min_flt;
1565                         r->ru_majflt += p->signal->maj_flt;
1566                         r->ru_inblock += p->signal->inblock;
1567                         r->ru_oublock += p->signal->oublock;
1568                         if (maxrss < p->signal->maxrss)
1569                                 maxrss = p->signal->maxrss;
1570                         t = p;
1571                         do {
1572                                 accumulate_thread_rusage(t, r);
1573                                 t = next_thread(t);
1574                         } while (t != p);
1575                         break;
1576
1577                 default:
1578                         BUG();
1579         }
1580         unlock_task_sighand(p, &flags);
1581
1582 out:
1583         cputime_to_timeval(utime, &r->ru_utime);
1584         cputime_to_timeval(stime, &r->ru_stime);
1585
1586         if (who != RUSAGE_CHILDREN) {
1587                 struct mm_struct *mm = get_task_mm(p);
1588                 if (mm) {
1589                         setmax_mm_hiwater_rss(&maxrss, mm);
1590                         mmput(mm);
1591                 }
1592         }
1593         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1594 }
1595
1596 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1597 {
1598         struct rusage r;
1599         k_getrusage(p, who, &r);
1600         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1601 }
1602
1603 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1604 {
1605         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1606             who != RUSAGE_THREAD)
1607                 return -EINVAL;
1608         return getrusage(current, who, ru);
1609 }
1610
1611 SYSCALL_DEFINE1(umask, int, mask)
1612 {
1613         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1614         return mask;
1615 }
1616
1617 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1618                 unsigned long, arg4, unsigned long, arg5)
1619 {
1620         struct task_struct *me = current;
1621         unsigned char comm[sizeof(me->comm)];
1622         long error;
1623
1624         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1625         if (error != -ENOSYS)
1626                 return error;
1627
1628         error = 0;
1629         switch (option) {
1630                 case PR_SET_PDEATHSIG:
1631                         if (!valid_signal(arg2)) {
1632                                 error = -EINVAL;
1633                                 break;
1634                         }
1635                         me->pdeath_signal = arg2;
1636                         error = 0;
1637                         break;
1638                 case PR_GET_PDEATHSIG:
1639                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1640                         break;
1641                 case PR_GET_DUMPABLE:
1642                         error = get_dumpable(me->mm);
1643                         break;
1644                 case PR_SET_DUMPABLE:
1645                         if (arg2 < 0 || arg2 > 1) {
1646                                 error = -EINVAL;
1647                                 break;
1648                         }
1649                         set_dumpable(me->mm, arg2);
1650                         error = 0;
1651                         break;
1652
1653                 case PR_SET_UNALIGN:
1654                         error = SET_UNALIGN_CTL(me, arg2);
1655                         break;
1656                 case PR_GET_UNALIGN:
1657                         error = GET_UNALIGN_CTL(me, arg2);
1658                         break;
1659                 case PR_SET_FPEMU:
1660                         error = SET_FPEMU_CTL(me, arg2);
1661                         break;
1662                 case PR_GET_FPEMU:
1663                         error = GET_FPEMU_CTL(me, arg2);
1664                         break;
1665                 case PR_SET_FPEXC:
1666                         error = SET_FPEXC_CTL(me, arg2);
1667                         break;
1668                 case PR_GET_FPEXC:
1669                         error = GET_FPEXC_CTL(me, arg2);
1670                         break;
1671                 case PR_GET_TIMING:
1672                         error = PR_TIMING_STATISTICAL;
1673                         break;
1674                 case PR_SET_TIMING:
1675                         if (arg2 != PR_TIMING_STATISTICAL)
1676                                 error = -EINVAL;
1677                         else
1678                                 error = 0;
1679                         break;
1680
1681                 case PR_SET_NAME:
1682                         comm[sizeof(me->comm)-1] = 0;
1683                         if (strncpy_from_user(comm, (char __user *)arg2,
1684                                               sizeof(me->comm) - 1) < 0)
1685                                 return -EFAULT;
1686                         set_task_comm(me, comm);
1687                         return 0;
1688                 case PR_GET_NAME:
1689                         get_task_comm(comm, me);
1690                         if (copy_to_user((char __user *)arg2, comm,
1691                                          sizeof(comm)))
1692                                 return -EFAULT;
1693                         return 0;
1694                 case PR_GET_ENDIAN:
1695                         error = GET_ENDIAN(me, arg2);
1696                         break;
1697                 case PR_SET_ENDIAN:
1698                         error = SET_ENDIAN(me, arg2);
1699                         break;
1700
1701                 case PR_GET_SECCOMP:
1702                         error = prctl_get_seccomp();
1703                         break;
1704                 case PR_SET_SECCOMP:
1705                         error = prctl_set_seccomp(arg2);
1706                         break;
1707                 case PR_GET_TSC:
1708                         error = GET_TSC_CTL(arg2);
1709                         break;
1710                 case PR_SET_TSC:
1711                         error = SET_TSC_CTL(arg2);
1712                         break;
1713                 case PR_TASK_PERF_EVENTS_DISABLE:
1714                         error = perf_event_task_disable();
1715                         break;
1716                 case PR_TASK_PERF_EVENTS_ENABLE:
1717                         error = perf_event_task_enable();
1718                         break;
1719                 case PR_GET_TIMERSLACK:
1720                         error = current->timer_slack_ns;
1721                         break;
1722                 case PR_SET_TIMERSLACK:
1723                         if (arg2 <= 0)
1724                                 current->timer_slack_ns =
1725                                         current->default_timer_slack_ns;
1726                         else
1727                                 current->timer_slack_ns = arg2;
1728                         error = 0;
1729                         break;
1730                 case PR_MCE_KILL:
1731                         if (arg4 | arg5)
1732                                 return -EINVAL;
1733                         switch (arg2) {
1734                         case PR_MCE_KILL_CLEAR:
1735                                 if (arg3 != 0)
1736                                         return -EINVAL;
1737                                 current->flags &= ~PF_MCE_PROCESS;
1738                                 break;
1739                         case PR_MCE_KILL_SET:
1740                                 current->flags |= PF_MCE_PROCESS;
1741                                 if (arg3 == PR_MCE_KILL_EARLY)
1742                                         current->flags |= PF_MCE_EARLY;
1743                                 else if (arg3 == PR_MCE_KILL_LATE)
1744                                         current->flags &= ~PF_MCE_EARLY;
1745                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1746                                         current->flags &=
1747                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1748                                 else
1749                                         return -EINVAL;
1750                                 break;
1751                         default:
1752                                 return -EINVAL;
1753                         }
1754                         error = 0;
1755                         break;
1756                 case PR_MCE_KILL_GET:
1757                         if (arg2 | arg3 | arg4 | arg5)
1758                                 return -EINVAL;
1759                         if (current->flags & PF_MCE_PROCESS)
1760                                 error = (current->flags & PF_MCE_EARLY) ?
1761                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1762                         else
1763                                 error = PR_MCE_KILL_DEFAULT;
1764                         break;
1765                 default:
1766                         error = -EINVAL;
1767                         break;
1768         }
1769         return error;
1770 }
1771
1772 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1773                 struct getcpu_cache __user *, unused)
1774 {
1775         int err = 0;
1776         int cpu = raw_smp_processor_id();
1777         if (cpup)
1778                 err |= put_user(cpu, cpup);
1779         if (nodep)
1780                 err |= put_user(cpu_to_node(cpu), nodep);
1781         return err ? -EFAULT : 0;
1782 }
1783
1784 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1785
1786 static void argv_cleanup(struct subprocess_info *info)
1787 {
1788         argv_free(info->argv);
1789 }
1790
1791 /**
1792  * orderly_poweroff - Trigger an orderly system poweroff
1793  * @force: force poweroff if command execution fails
1794  *
1795  * This may be called from any context to trigger a system shutdown.
1796  * If the orderly shutdown fails, it will force an immediate shutdown.
1797  */
1798 int orderly_poweroff(bool force)
1799 {
1800         int argc;
1801         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1802         static char *envp[] = {
1803                 "HOME=/",
1804                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1805                 NULL
1806         };
1807         int ret = -ENOMEM;
1808         struct subprocess_info *info;
1809
1810         if (argv == NULL) {
1811                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1812                        __func__, poweroff_cmd);
1813                 goto out;
1814         }
1815
1816         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1817         if (info == NULL) {
1818                 argv_free(argv);
1819                 goto out;
1820         }
1821
1822         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1823
1824         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1825
1826   out:
1827         if (ret && force) {
1828                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1829                        "forcing the issue\n");
1830
1831                 /* I guess this should try to kick off some daemon to
1832                    sync and poweroff asap.  Or not even bother syncing
1833                    if we're doing an emergency shutdown? */
1834                 emergency_sync();
1835                 kernel_power_off();
1836         }
1837
1838         return ret;
1839 }
1840 EXPORT_SYMBOL_GPL(orderly_poweroff);