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