reboot: rigrate shutdown/reboot to boot cpu
[opensuse:kernel.git] / kernel / sys.c
1 /*
2  *  linux/kernel/sys.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.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 #include <linux/version.h>
42 #include <linux/ctype.h>
43
44 #include <linux/compat.h>
45 #include <linux/syscalls.h>
46 #include <linux/kprobes.h>
47 #include <linux/user_namespace.h>
48
49 #include <linux/kmsg_dump.h>
50 /* Move somewhere else to avoid recompiling? */
51 #include <generated/utsrelease.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/io.h>
55 #include <asm/unistd.h>
56
57 #ifndef SET_UNALIGN_CTL
58 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
59 #endif
60 #ifndef GET_UNALIGN_CTL
61 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
62 #endif
63 #ifndef SET_FPEMU_CTL
64 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
65 #endif
66 #ifndef GET_FPEMU_CTL
67 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
68 #endif
69 #ifndef SET_FPEXC_CTL
70 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
71 #endif
72 #ifndef GET_FPEXC_CTL
73 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
74 #endif
75 #ifndef GET_ENDIAN
76 # define GET_ENDIAN(a,b)        (-EINVAL)
77 #endif
78 #ifndef SET_ENDIAN
79 # define SET_ENDIAN(a,b)        (-EINVAL)
80 #endif
81 #ifndef GET_TSC_CTL
82 # define GET_TSC_CTL(a)         (-EINVAL)
83 #endif
84 #ifndef SET_TSC_CTL
85 # define SET_TSC_CTL(a)         (-EINVAL)
86 #endif
87
88 /*
89  * this is where the system-wide overflow UID and GID are defined, for
90  * architectures that now have 32-bit UID/GID but didn't in the past
91  */
92
93 int overflowuid = DEFAULT_OVERFLOWUID;
94 int overflowgid = DEFAULT_OVERFLOWGID;
95
96 #ifdef CONFIG_UID16
97 EXPORT_SYMBOL(overflowuid);
98 EXPORT_SYMBOL(overflowgid);
99 #endif
100
101 /*
102  * the same as above, but for filesystems which can only store a 16-bit
103  * UID and GID. as such, this is needed on all architectures
104  */
105
106 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
107 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
108
109 EXPORT_SYMBOL(fs_overflowuid);
110 EXPORT_SYMBOL(fs_overflowgid);
111
112 /*
113  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
114  */
115
116 int C_A_D = 1;
117 struct pid *cad_pid;
118 EXPORT_SYMBOL(cad_pid);
119
120 /*
121  * If set, this is used for preparing the system to power off.
122  */
123
124 void (*pm_power_off_prepare)(void);
125
126 /*
127  * Returns true if current's euid is same as p's uid or euid,
128  * or has CAP_SYS_NICE to p's user_ns.
129  *
130  * Called with rcu_read_lock, creds are safe
131  */
132 static bool set_one_prio_perm(struct task_struct *p)
133 {
134         const struct cred *cred = current_cred(), *pcred = __task_cred(p);
135
136         if (pcred->user->user_ns == cred->user->user_ns &&
137             (pcred->uid  == cred->euid ||
138              pcred->euid == cred->euid))
139                 return true;
140         if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
141                 return true;
142         return false;
143 }
144
145 /*
146  * set the priority of a task
147  * - the caller must hold the RCU read lock
148  */
149 static int set_one_prio(struct task_struct *p, int niceval, int error)
150 {
151         int no_nice;
152
153         if (!set_one_prio_perm(p)) {
154                 error = -EPERM;
155                 goto out;
156         }
157         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
158                 error = -EACCES;
159                 goto out;
160         }
161         no_nice = security_task_setnice(p, niceval);
162         if (no_nice) {
163                 error = no_nice;
164                 goto out;
165         }
166         if (error == -ESRCH)
167                 error = 0;
168         set_user_nice(p, niceval);
169 out:
170         return error;
171 }
172
173 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
174 {
175         struct task_struct *g, *p;
176         struct user_struct *user;
177         const struct cred *cred = current_cred();
178         int error = -EINVAL;
179         struct pid *pgrp;
180
181         if (which > PRIO_USER || which < PRIO_PROCESS)
182                 goto out;
183
184         /* normalize: avoid signed division (rounding problems) */
185         error = -ESRCH;
186         if (niceval < -20)
187                 niceval = -20;
188         if (niceval > 19)
189                 niceval = 19;
190
191         rcu_read_lock();
192         read_lock(&tasklist_lock);
193         switch (which) {
194                 case PRIO_PROCESS:
195                         if (who)
196                                 p = find_task_by_vpid(who);
197                         else
198                                 p = current;
199                         if (p)
200                                 error = set_one_prio(p, niceval, error);
201                         break;
202                 case PRIO_PGRP:
203                         if (who)
204                                 pgrp = find_vpid(who);
205                         else
206                                 pgrp = task_pgrp(current);
207                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
208                                 error = set_one_prio(p, niceval, error);
209                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
210                         break;
211                 case PRIO_USER:
212                         user = (struct user_struct *) cred->user;
213                         if (!who)
214                                 who = cred->uid;
215                         else if ((who != cred->uid) &&
216                                  !(user = find_user(who)))
217                                 goto out_unlock;        /* No processes for this user */
218
219                         do_each_thread(g, p) {
220                                 if (__task_cred(p)->uid == who)
221                                         error = set_one_prio(p, niceval, error);
222                         } while_each_thread(g, p);
223                         if (who != cred->uid)
224                                 free_uid(user);         /* For find_user() */
225                         break;
226         }
227 out_unlock:
228         read_unlock(&tasklist_lock);
229         rcu_read_unlock();
230 out:
231         return error;
232 }
233
234 /*
235  * Ugh. To avoid negative return values, "getpriority()" will
236  * not return the normal nice-value, but a negated value that
237  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
238  * to stay compatible.
239  */
240 SYSCALL_DEFINE2(getpriority, int, which, int, who)
241 {
242         struct task_struct *g, *p;
243         struct user_struct *user;
244         const struct cred *cred = current_cred();
245         long niceval, retval = -ESRCH;
246         struct pid *pgrp;
247
248         if (which > PRIO_USER || which < PRIO_PROCESS)
249                 return -EINVAL;
250
251         rcu_read_lock();
252         read_lock(&tasklist_lock);
253         switch (which) {
254                 case PRIO_PROCESS:
255                         if (who)
256                                 p = find_task_by_vpid(who);
257                         else
258                                 p = current;
259                         if (p) {
260                                 niceval = 20 - task_nice(p);
261                                 if (niceval > retval)
262                                         retval = niceval;
263                         }
264                         break;
265                 case PRIO_PGRP:
266                         if (who)
267                                 pgrp = find_vpid(who);
268                         else
269                                 pgrp = task_pgrp(current);
270                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
271                                 niceval = 20 - task_nice(p);
272                                 if (niceval > retval)
273                                         retval = niceval;
274                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
275                         break;
276                 case PRIO_USER:
277                         user = (struct user_struct *) cred->user;
278                         if (!who)
279                                 who = cred->uid;
280                         else if ((who != cred->uid) &&
281                                  !(user = find_user(who)))
282                                 goto out_unlock;        /* No processes for this user */
283
284                         do_each_thread(g, p) {
285                                 if (__task_cred(p)->uid == who) {
286                                         niceval = 20 - task_nice(p);
287                                         if (niceval > retval)
288                                                 retval = niceval;
289                                 }
290                         } while_each_thread(g, p);
291                         if (who != cred->uid)
292                                 free_uid(user);         /* for find_user() */
293                         break;
294         }
295 out_unlock:
296         read_unlock(&tasklist_lock);
297         rcu_read_unlock();
298
299         return retval;
300 }
301
302 /**
303  *      emergency_restart - reboot the system
304  *
305  *      Without shutting down any hardware or taking any locks
306  *      reboot the system.  This is called when we know we are in
307  *      trouble so this is our best effort to reboot.  This is
308  *      safe to call in interrupt context.
309  */
310 void emergency_restart(void)
311 {
312         kmsg_dump(KMSG_DUMP_EMERG);
313         machine_emergency_restart();
314 }
315 EXPORT_SYMBOL_GPL(emergency_restart);
316
317 void kernel_restart_prepare(char *cmd)
318 {
319         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
320         system_state = SYSTEM_RESTART;
321         usermodehelper_disable();
322         device_shutdown();
323 }
324
325 /**
326  *      register_reboot_notifier - Register function to be called at reboot time
327  *      @nb: Info about notifier function to be called
328  *
329  *      Registers a function with the list of functions
330  *      to be called at reboot time.
331  *
332  *      Currently always returns zero, as blocking_notifier_chain_register()
333  *      always returns zero.
334  */
335 int register_reboot_notifier(struct notifier_block *nb)
336 {
337         return blocking_notifier_chain_register(&reboot_notifier_list, nb);
338 }
339 EXPORT_SYMBOL(register_reboot_notifier);
340
341 /**
342  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
343  *      @nb: Hook to be unregistered
344  *
345  *      Unregisters a previously registered reboot
346  *      notifier function.
347  *
348  *      Returns zero on success, or %-ENOENT on failure.
349  */
350 int unregister_reboot_notifier(struct notifier_block *nb)
351 {
352         return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
353 }
354 EXPORT_SYMBOL(unregister_reboot_notifier);
355
356 /* Add backwards compatibility for stable trees. */
357 #ifndef PF_NO_SETAFFINITY
358 #define PF_NO_SETAFFINITY               PF_THREAD_BOUND
359 #endif
360
361 static void migrate_to_reboot_cpu(void)
362 {
363         /* The boot cpu is always logical cpu 0 */
364         int cpu = 0;
365
366         cpu_hotplug_disable();
367
368         /* Make certain the cpu I'm about to reboot on is online */
369         if (!cpu_online(cpu))
370                 cpu = cpumask_first(cpu_online_mask);
371
372         /* Prevent races with other tasks migrating this task */
373         current->flags |= PF_NO_SETAFFINITY;
374
375         /* Make certain I only run on the appropriate processor */
376         set_cpus_allowed_ptr(current, cpumask_of(cpu));
377 }
378
379 /**
380  *      kernel_restart - reboot the system
381  *      @cmd: pointer to buffer containing command to execute for restart
382  *              or %NULL
383  *
384  *      Shutdown everything and perform a clean reboot.
385  *      This is not safe to call in interrupt context.
386  */
387 void kernel_restart(char *cmd)
388 {
389         kernel_restart_prepare(cmd);
390         migrate_to_reboot_cpu();
391         syscore_shutdown();
392         if (!cmd)
393                 printk(KERN_EMERG "Restarting system.\n");
394         else
395                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
396         kmsg_dump(KMSG_DUMP_RESTART);
397         machine_restart(cmd);
398 }
399 EXPORT_SYMBOL_GPL(kernel_restart);
400
401 static void kernel_shutdown_prepare(enum system_states state)
402 {
403         blocking_notifier_call_chain(&reboot_notifier_list,
404                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
405         system_state = state;
406         usermodehelper_disable();
407         device_shutdown();
408 }
409 /**
410  *      kernel_halt - halt the system
411  *
412  *      Shutdown everything and perform a clean system halt.
413  */
414 void kernel_halt(void)
415 {
416         kernel_shutdown_prepare(SYSTEM_HALT);
417         migrate_to_reboot_cpu();
418         syscore_shutdown();
419         printk(KERN_EMERG "System halted.\n");
420         kmsg_dump(KMSG_DUMP_HALT);
421         machine_halt();
422 }
423
424 EXPORT_SYMBOL_GPL(kernel_halt);
425
426 /**
427  *      kernel_power_off - power_off the system
428  *
429  *      Shutdown everything and perform a clean system power_off.
430  */
431 void kernel_power_off(void)
432 {
433         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
434         if (pm_power_off_prepare)
435                 pm_power_off_prepare();
436         migrate_to_reboot_cpu();
437         syscore_shutdown();
438         printk(KERN_EMERG "Power down.\n");
439         kmsg_dump(KMSG_DUMP_POWEROFF);
440         machine_power_off();
441 }
442 EXPORT_SYMBOL_GPL(kernel_power_off);
443
444 static DEFINE_MUTEX(reboot_mutex);
445
446 /*
447  * Reboot system call: for obvious reasons only root may call it,
448  * and even root needs to set up some magic numbers in the registers
449  * so that some mistake won't make this reboot the whole machine.
450  * You can also set the meaning of the ctrl-alt-del-key here.
451  *
452  * reboot doesn't sync: do that yourself before calling this.
453  */
454 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
455                 void __user *, arg)
456 {
457         char buffer[256];
458         int ret = 0;
459
460         /* We only trust the superuser with rebooting the system. */
461         if (!capable(CAP_SYS_BOOT))
462                 return -EPERM;
463
464         /* For safety, we require "magic" arguments. */
465         if (magic1 != LINUX_REBOOT_MAGIC1 ||
466             (magic2 != LINUX_REBOOT_MAGIC2 &&
467                         magic2 != LINUX_REBOOT_MAGIC2A &&
468                         magic2 != LINUX_REBOOT_MAGIC2B &&
469                         magic2 != LINUX_REBOOT_MAGIC2C))
470                 return -EINVAL;
471
472         /*
473          * If pid namespaces are enabled and the current task is in a child
474          * pid_namespace, the command is handled by reboot_pid_ns() which will
475          * call do_exit().
476          */
477         ret = reboot_pid_ns(task_active_pid_ns(current), cmd);
478         if (ret)
479                 return ret;
480
481         /* Instead of trying to make the power_off code look like
482          * halt when pm_power_off is not set do it the easy way.
483          */
484         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
485                 cmd = LINUX_REBOOT_CMD_HALT;
486
487         mutex_lock(&reboot_mutex);
488         switch (cmd) {
489         case LINUX_REBOOT_CMD_RESTART:
490                 kernel_restart(NULL);
491                 break;
492
493         case LINUX_REBOOT_CMD_CAD_ON:
494                 C_A_D = 1;
495                 break;
496
497         case LINUX_REBOOT_CMD_CAD_OFF:
498                 C_A_D = 0;
499                 break;
500
501         case LINUX_REBOOT_CMD_HALT:
502                 kernel_halt();
503                 do_exit(0);
504                 panic("cannot halt");
505
506         case LINUX_REBOOT_CMD_POWER_OFF:
507                 kernel_power_off();
508                 do_exit(0);
509                 break;
510
511         case LINUX_REBOOT_CMD_RESTART2:
512                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
513                         ret = -EFAULT;
514                         break;
515                 }
516                 buffer[sizeof(buffer) - 1] = '\0';
517
518                 kernel_restart(buffer);
519                 break;
520
521 #ifdef CONFIG_KEXEC
522         case LINUX_REBOOT_CMD_KEXEC:
523                 ret = kernel_kexec();
524                 break;
525 #endif
526
527 #ifdef CONFIG_HIBERNATION
528         case LINUX_REBOOT_CMD_SW_SUSPEND:
529                 ret = hibernate();
530                 break;
531 #endif
532
533         default:
534                 ret = -EINVAL;
535                 break;
536         }
537         mutex_unlock(&reboot_mutex);
538         return ret;
539 }
540
541 static void deferred_cad(struct work_struct *dummy)
542 {
543         kernel_restart(NULL);
544 }
545
546 /*
547  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
548  * As it's called within an interrupt, it may NOT sync: the only choice
549  * is whether to reboot at once, or just ignore the ctrl-alt-del.
550  */
551 void ctrl_alt_del(void)
552 {
553         static DECLARE_WORK(cad_work, deferred_cad);
554
555         if (C_A_D)
556                 schedule_work(&cad_work);
557         else
558                 kill_cad_pid(SIGINT, 1);
559 }
560         
561 /*
562  * Unprivileged users may change the real gid to the effective gid
563  * or vice versa.  (BSD-style)
564  *
565  * If you set the real gid at all, or set the effective gid to a value not
566  * equal to the real gid, then the saved gid is set to the new effective gid.
567  *
568  * This makes it possible for a setgid program to completely drop its
569  * privileges, which is often a useful assertion to make when you are doing
570  * a security audit over a program.
571  *
572  * The general idea is that a program which uses just setregid() will be
573  * 100% compatible with BSD.  A program which uses just setgid() will be
574  * 100% compatible with POSIX with saved IDs. 
575  *
576  * SMP: There are not races, the GIDs are checked only by filesystem
577  *      operations (as far as semantic preservation is concerned).
578  */
579 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
580 {
581         const struct cred *old;
582         struct cred *new;
583         int retval;
584
585         new = prepare_creds();
586         if (!new)
587                 return -ENOMEM;
588         old = current_cred();
589
590         retval = -EPERM;
591         if (rgid != (gid_t) -1) {
592                 if (old->gid == rgid ||
593                     old->egid == rgid ||
594                     nsown_capable(CAP_SETGID))
595                         new->gid = rgid;
596                 else
597                         goto error;
598         }
599         if (egid != (gid_t) -1) {
600                 if (old->gid == egid ||
601                     old->egid == egid ||
602                     old->sgid == egid ||
603                     nsown_capable(CAP_SETGID))
604                         new->egid = egid;
605                 else
606                         goto error;
607         }
608
609         if (rgid != (gid_t) -1 ||
610             (egid != (gid_t) -1 && egid != old->gid))
611                 new->sgid = new->egid;
612         new->fsgid = new->egid;
613
614         return commit_creds(new);
615
616 error:
617         abort_creds(new);
618         return retval;
619 }
620
621 /*
622  * setgid() is implemented like SysV w/ SAVED_IDS 
623  *
624  * SMP: Same implicit races as above.
625  */
626 SYSCALL_DEFINE1(setgid, gid_t, gid)
627 {
628         const struct cred *old;
629         struct cred *new;
630         int retval;
631
632         new = prepare_creds();
633         if (!new)
634                 return -ENOMEM;
635         old = current_cred();
636
637         retval = -EPERM;
638         if (nsown_capable(CAP_SETGID))
639                 new->gid = new->egid = new->sgid = new->fsgid = gid;
640         else if (gid == old->gid || gid == old->sgid)
641                 new->egid = new->fsgid = gid;
642         else
643                 goto error;
644
645         return commit_creds(new);
646
647 error:
648         abort_creds(new);
649         return retval;
650 }
651
652 /*
653  * change the user struct in a credentials set to match the new UID
654  */
655 static int set_user(struct cred *new)
656 {
657         struct user_struct *new_user;
658
659         new_user = alloc_uid(current_user_ns(), new->uid);
660         if (!new_user)
661                 return -EAGAIN;
662
663         /*
664          * We don't fail in case of NPROC limit excess here because too many
665          * poorly written programs don't check set*uid() return code, assuming
666          * it never fails if called by root.  We may still enforce NPROC limit
667          * for programs doing set*uid()+execve() by harmlessly deferring the
668          * failure to the execve() stage.
669          */
670         if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
671                         new_user != INIT_USER)
672                 current->flags |= PF_NPROC_EXCEEDED;
673         else
674                 current->flags &= ~PF_NPROC_EXCEEDED;
675
676         free_uid(new->user);
677         new->user = new_user;
678         return 0;
679 }
680
681 /*
682  * Unprivileged users may change the real uid to the effective uid
683  * or vice versa.  (BSD-style)
684  *
685  * If you set the real uid at all, or set the effective uid to a value not
686  * equal to the real uid, then the saved uid is set to the new effective uid.
687  *
688  * This makes it possible for a setuid program to completely drop its
689  * privileges, which is often a useful assertion to make when you are doing
690  * a security audit over a program.
691  *
692  * The general idea is that a program which uses just setreuid() will be
693  * 100% compatible with BSD.  A program which uses just setuid() will be
694  * 100% compatible with POSIX with saved IDs. 
695  */
696 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
697 {
698         const struct cred *old;
699         struct cred *new;
700         int retval;
701
702         new = prepare_creds();
703         if (!new)
704                 return -ENOMEM;
705         old = current_cred();
706
707         retval = -EPERM;
708         if (ruid != (uid_t) -1) {
709                 new->uid = ruid;
710                 if (old->uid != ruid &&
711                     old->euid != ruid &&
712                     !nsown_capable(CAP_SETUID))
713                         goto error;
714         }
715
716         if (euid != (uid_t) -1) {
717                 new->euid = euid;
718                 if (old->uid != euid &&
719                     old->euid != euid &&
720                     old->suid != euid &&
721                     !nsown_capable(CAP_SETUID))
722                         goto error;
723         }
724
725         if (new->uid != old->uid) {
726                 retval = set_user(new);
727                 if (retval < 0)
728                         goto error;
729         }
730         if (ruid != (uid_t) -1 ||
731             (euid != (uid_t) -1 && euid != old->uid))
732                 new->suid = new->euid;
733         new->fsuid = new->euid;
734
735         retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
736         if (retval < 0)
737                 goto error;
738
739         return commit_creds(new);
740
741 error:
742         abort_creds(new);
743         return retval;
744 }
745                 
746 /*
747  * setuid() is implemented like SysV with SAVED_IDS 
748  * 
749  * Note that SAVED_ID's is deficient in that a setuid root program
750  * like sendmail, for example, cannot set its uid to be a normal 
751  * user and then switch back, because if you're root, setuid() sets
752  * the saved uid too.  If you don't like this, blame the bright people
753  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
754  * will allow a root program to temporarily drop privileges and be able to
755  * regain them by swapping the real and effective uid.  
756  */
757 SYSCALL_DEFINE1(setuid, uid_t, uid)
758 {
759         const struct cred *old;
760         struct cred *new;
761         int retval;
762
763         new = prepare_creds();
764         if (!new)
765                 return -ENOMEM;
766         old = current_cred();
767
768         retval = -EPERM;
769         if (nsown_capable(CAP_SETUID)) {
770                 new->suid = new->uid = uid;
771                 if (uid != old->uid) {
772                         retval = set_user(new);
773                         if (retval < 0)
774                                 goto error;
775                 }
776         } else if (uid != old->uid && uid != new->suid) {
777                 goto error;
778         }
779
780         new->fsuid = new->euid = uid;
781
782         retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
783         if (retval < 0)
784                 goto error;
785
786         return commit_creds(new);
787
788 error:
789         abort_creds(new);
790         return retval;
791 }
792
793
794 /*
795  * This function implements a generic ability to update ruid, euid,
796  * and suid.  This allows you to implement the 4.4 compatible seteuid().
797  */
798 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
799 {
800         const struct cred *old;
801         struct cred *new;
802         int retval;
803
804         new = prepare_creds();
805         if (!new)
806                 return -ENOMEM;
807
808         old = current_cred();
809
810         retval = -EPERM;
811         if (!nsown_capable(CAP_SETUID)) {
812                 if (ruid != (uid_t) -1 && ruid != old->uid &&
813                     ruid != old->euid  && ruid != old->suid)
814                         goto error;
815                 if (euid != (uid_t) -1 && euid != old->uid &&
816                     euid != old->euid  && euid != old->suid)
817                         goto error;
818                 if (suid != (uid_t) -1 && suid != old->uid &&
819                     suid != old->euid  && suid != old->suid)
820                         goto error;
821         }
822
823         if (ruid != (uid_t) -1) {
824                 new->uid = ruid;
825                 if (ruid != old->uid) {
826                         retval = set_user(new);
827                         if (retval < 0)
828                                 goto error;
829                 }
830         }
831         if (euid != (uid_t) -1)
832                 new->euid = euid;
833         if (suid != (uid_t) -1)
834                 new->suid = suid;
835         new->fsuid = new->euid;
836
837         retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
838         if (retval < 0)
839                 goto error;
840
841         return commit_creds(new);
842
843 error:
844         abort_creds(new);
845         return retval;
846 }
847
848 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
849 {
850         const struct cred *cred = current_cred();
851         int retval;
852
853         if (!(retval   = put_user(cred->uid,  ruid)) &&
854             !(retval   = put_user(cred->euid, euid)))
855                 retval = put_user(cred->suid, suid);
856
857         return retval;
858 }
859
860 /*
861  * Same as above, but for rgid, egid, sgid.
862  */
863 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
864 {
865         const struct cred *old;
866         struct cred *new;
867         int retval;
868
869         new = prepare_creds();
870         if (!new)
871                 return -ENOMEM;
872         old = current_cred();
873
874         retval = -EPERM;
875         if (!nsown_capable(CAP_SETGID)) {
876                 if (rgid != (gid_t) -1 && rgid != old->gid &&
877                     rgid != old->egid  && rgid != old->sgid)
878                         goto error;
879                 if (egid != (gid_t) -1 && egid != old->gid &&
880                     egid != old->egid  && egid != old->sgid)
881                         goto error;
882                 if (sgid != (gid_t) -1 && sgid != old->gid &&
883                     sgid != old->egid  && sgid != old->sgid)
884                         goto error;
885         }
886
887         if (rgid != (gid_t) -1)
888                 new->gid = rgid;
889         if (egid != (gid_t) -1)
890                 new->egid = egid;
891         if (sgid != (gid_t) -1)
892                 new->sgid = sgid;
893         new->fsgid = new->egid;
894
895         return commit_creds(new);
896
897 error:
898         abort_creds(new);
899         return retval;
900 }
901
902 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
903 {
904         const struct cred *cred = current_cred();
905         int retval;
906
907         if (!(retval   = put_user(cred->gid,  rgid)) &&
908             !(retval   = put_user(cred->egid, egid)))
909                 retval = put_user(cred->sgid, sgid);
910
911         return retval;
912 }
913
914
915 /*
916  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
917  * is used for "access()" and for the NFS daemon (letting nfsd stay at
918  * whatever uid it wants to). It normally shadows "euid", except when
919  * explicitly set by setfsuid() or for access..
920  */
921 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
922 {
923         const struct cred *old;
924         struct cred *new;
925         uid_t old_fsuid;
926
927         new = prepare_creds();
928         if (!new)
929                 return current_fsuid();
930         old = current_cred();
931         old_fsuid = old->fsuid;
932
933         if (uid == old->uid  || uid == old->euid  ||
934             uid == old->suid || uid == old->fsuid ||
935             nsown_capable(CAP_SETUID)) {
936                 if (uid != old_fsuid) {
937                         new->fsuid = uid;
938                         if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
939                                 goto change_okay;
940                 }
941         }
942
943         abort_creds(new);
944         return old_fsuid;
945
946 change_okay:
947         commit_creds(new);
948         return old_fsuid;
949 }
950
951 /*
952  * Samma på svenska..
953  */
954 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
955 {
956         const struct cred *old;
957         struct cred *new;
958         gid_t old_fsgid;
959
960         new = prepare_creds();
961         if (!new)
962                 return current_fsgid();
963         old = current_cred();
964         old_fsgid = old->fsgid;
965
966         if (gid == old->gid  || gid == old->egid  ||
967             gid == old->sgid || gid == old->fsgid ||
968             nsown_capable(CAP_SETGID)) {
969                 if (gid != old_fsgid) {
970                         new->fsgid = gid;
971                         goto change_okay;
972                 }
973         }
974
975         abort_creds(new);
976         return old_fsgid;
977
978 change_okay:
979         commit_creds(new);
980         return old_fsgid;
981 }
982
983 void do_sys_times(struct tms *tms)
984 {
985         cputime_t tgutime, tgstime, cutime, cstime;
986
987         spin_lock_irq(&current->sighand->siglock);
988         thread_group_times(current, &tgutime, &tgstime);
989         cutime = current->signal->cutime;
990         cstime = current->signal->cstime;
991         spin_unlock_irq(&current->sighand->siglock);
992         tms->tms_utime = cputime_to_clock_t(tgutime);
993         tms->tms_stime = cputime_to_clock_t(tgstime);
994         tms->tms_cutime = cputime_to_clock_t(cutime);
995         tms->tms_cstime = cputime_to_clock_t(cstime);
996 }
997
998 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
999 {
1000         if (tbuf) {
1001                 struct tms tmp;
1002
1003                 do_sys_times(&tmp);
1004                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1005                         return -EFAULT;
1006         }
1007         force_successful_syscall_return();
1008         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1009 }
1010
1011 /*
1012  * This needs some heavy checking ...
1013  * I just haven't the stomach for it. I also don't fully
1014  * understand sessions/pgrp etc. Let somebody who does explain it.
1015  *
1016  * OK, I think I have the protection semantics right.... this is really
1017  * only important on a multi-user system anyway, to make sure one user
1018  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1019  *
1020  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1021  * LBT 04.03.94
1022  */
1023 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1024 {
1025         struct task_struct *p;
1026         struct task_struct *group_leader = current->group_leader;
1027         struct pid *pgrp;
1028         int err;
1029
1030         if (!pid)
1031                 pid = task_pid_vnr(group_leader);
1032         if (!pgid)
1033                 pgid = pid;
1034         if (pgid < 0)
1035                 return -EINVAL;
1036         rcu_read_lock();
1037
1038         /* From this point forward we keep holding onto the tasklist lock
1039          * so that our parent does not change from under us. -DaveM
1040          */
1041         write_lock_irq(&tasklist_lock);
1042
1043         err = -ESRCH;
1044         p = find_task_by_vpid(pid);
1045         if (!p)
1046                 goto out;
1047
1048         err = -EINVAL;
1049         if (!thread_group_leader(p))
1050                 goto out;
1051
1052         if (same_thread_group(p->real_parent, group_leader)) {
1053                 err = -EPERM;
1054                 if (task_session(p) != task_session(group_leader))
1055                         goto out;
1056                 err = -EACCES;
1057                 if (p->did_exec)
1058                         goto out;
1059         } else {
1060                 err = -ESRCH;
1061                 if (p != group_leader)
1062                         goto out;
1063         }
1064
1065         err = -EPERM;
1066         if (p->signal->leader)
1067                 goto out;
1068
1069         pgrp = task_pid(p);
1070         if (pgid != pid) {
1071                 struct task_struct *g;
1072
1073                 pgrp = find_vpid(pgid);
1074                 g = pid_task(pgrp, PIDTYPE_PGID);
1075                 if (!g || task_session(g) != task_session(group_leader))
1076                         goto out;
1077         }
1078
1079         err = security_task_setpgid(p, pgid);
1080         if (err)
1081                 goto out;
1082
1083         if (task_pgrp(p) != pgrp)
1084                 change_pid(p, PIDTYPE_PGID, pgrp);
1085
1086         err = 0;
1087 out:
1088         /* All paths lead to here, thus we are safe. -DaveM */
1089         write_unlock_irq(&tasklist_lock);
1090         rcu_read_unlock();
1091         return err;
1092 }
1093
1094 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1095 {
1096         struct task_struct *p;
1097         struct pid *grp;
1098         int retval;
1099
1100         rcu_read_lock();
1101         if (!pid)
1102                 grp = task_pgrp(current);
1103         else {
1104                 retval = -ESRCH;
1105                 p = find_task_by_vpid(pid);
1106                 if (!p)
1107                         goto out;
1108                 grp = task_pgrp(p);
1109                 if (!grp)
1110                         goto out;
1111
1112                 retval = security_task_getpgid(p);
1113                 if (retval)
1114                         goto out;
1115         }
1116         retval = pid_vnr(grp);
1117 out:
1118         rcu_read_unlock();
1119         return retval;
1120 }
1121
1122 #ifdef __ARCH_WANT_SYS_GETPGRP
1123
1124 SYSCALL_DEFINE0(getpgrp)
1125 {
1126         return sys_getpgid(0);
1127 }
1128
1129 #endif
1130
1131 SYSCALL_DEFINE1(getsid, pid_t, pid)
1132 {
1133         struct task_struct *p;
1134         struct pid *sid;
1135         int retval;
1136
1137         rcu_read_lock();
1138         if (!pid)
1139                 sid = task_session(current);
1140         else {
1141                 retval = -ESRCH;
1142                 p = find_task_by_vpid(pid);
1143                 if (!p)
1144                         goto out;
1145                 sid = task_session(p);
1146                 if (!sid)
1147                         goto out;
1148
1149                 retval = security_task_getsid(p);
1150                 if (retval)
1151                         goto out;
1152         }
1153         retval = pid_vnr(sid);
1154 out:
1155         rcu_read_unlock();
1156         return retval;
1157 }
1158
1159 SYSCALL_DEFINE0(setsid)
1160 {
1161         struct task_struct *group_leader = current->group_leader;
1162         struct pid *sid = task_pid(group_leader);
1163         pid_t session = pid_vnr(sid);
1164         int err = -EPERM;
1165
1166         write_lock_irq(&tasklist_lock);
1167         /* Fail if I am already a session leader */
1168         if (group_leader->signal->leader)
1169                 goto out;
1170
1171         /* Fail if a process group id already exists that equals the
1172          * proposed session id.
1173          */
1174         if (pid_task(sid, PIDTYPE_PGID))
1175                 goto out;
1176
1177         group_leader->signal->leader = 1;
1178         __set_special_pids(sid);
1179
1180         proc_clear_tty(group_leader);
1181
1182         err = session;
1183 out:
1184         write_unlock_irq(&tasklist_lock);
1185         if (err > 0) {
1186                 proc_sid_connector(group_leader);
1187                 sched_autogroup_create_attach(group_leader);
1188         }
1189         return err;
1190 }
1191
1192 DECLARE_RWSEM(uts_sem);
1193
1194 #ifdef COMPAT_UTS_MACHINE
1195 #define override_architecture(name) \
1196         (personality(current->personality) == PER_LINUX32 && \
1197          copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1198                       sizeof(COMPAT_UTS_MACHINE)))
1199 #else
1200 #define override_architecture(name)     0
1201 #endif
1202
1203 /*
1204  * Work around broken programs that cannot handle "Linux 3.0".
1205  * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1206  */
1207 static int override_release(char __user *release, size_t len)
1208 {
1209         int ret = 0;
1210
1211         if (current->personality & UNAME26) {
1212                 const char *rest = UTS_RELEASE;
1213                 char buf[65] = { 0 };
1214                 int ndots = 0;
1215                 unsigned v;
1216                 size_t copy;
1217
1218                 while (*rest) {
1219                         if (*rest == '.' && ++ndots >= 3)
1220                                 break;
1221                         if (!isdigit(*rest) && *rest != '.')
1222                                 break;
1223                         rest++;
1224                 }
1225                 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1226                 copy = clamp_t(size_t, len, 1, sizeof(buf));
1227                 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1228                 ret = copy_to_user(release, buf, copy + 1);
1229         }
1230         return ret;
1231 }
1232
1233 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1234 {
1235         int errno = 0;
1236
1237         down_read(&uts_sem);
1238         if (copy_to_user(name, utsname(), sizeof *name))
1239                 errno = -EFAULT;
1240         up_read(&uts_sem);
1241
1242         if (!errno && override_release(name->release, sizeof(name->release)))
1243                 errno = -EFAULT;
1244         if (!errno && override_architecture(name))
1245                 errno = -EFAULT;
1246         return errno;
1247 }
1248
1249 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1250 /*
1251  * Old cruft
1252  */
1253 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1254 {
1255         int error = 0;
1256
1257         if (!name)
1258                 return -EFAULT;
1259
1260         down_read(&uts_sem);
1261         if (copy_to_user(name, utsname(), sizeof(*name)))
1262                 error = -EFAULT;
1263         up_read(&uts_sem);
1264
1265         if (!error && override_release(name->release, sizeof(name->release)))
1266                 error = -EFAULT;
1267         if (!error && override_architecture(name))
1268                 error = -EFAULT;
1269         return error;
1270 }
1271
1272 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1273 {
1274         int error;
1275
1276         if (!name)
1277                 return -EFAULT;
1278         if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1279                 return -EFAULT;
1280
1281         down_read(&uts_sem);
1282         error = __copy_to_user(&name->sysname, &utsname()->sysname,
1283                                __OLD_UTS_LEN);
1284         error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1285         error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1286                                 __OLD_UTS_LEN);
1287         error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1288         error |= __copy_to_user(&name->release, &utsname()->release,
1289                                 __OLD_UTS_LEN);
1290         error |= __put_user(0, name->release + __OLD_UTS_LEN);
1291         error |= __copy_to_user(&name->version, &utsname()->version,
1292                                 __OLD_UTS_LEN);
1293         error |= __put_user(0, name->version + __OLD_UTS_LEN);
1294         error |= __copy_to_user(&name->machine, &utsname()->machine,
1295                                 __OLD_UTS_LEN);
1296         error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1297         up_read(&uts_sem);
1298
1299         if (!error && override_architecture(name))
1300                 error = -EFAULT;
1301         if (!error && override_release(name->release, sizeof(name->release)))
1302                 error = -EFAULT;
1303         return error ? -EFAULT : 0;
1304 }
1305 #endif
1306
1307 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1308 {
1309         int errno;
1310         char tmp[__NEW_UTS_LEN];
1311
1312         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1313                 return -EPERM;
1314
1315         if (len < 0 || len > __NEW_UTS_LEN)
1316                 return -EINVAL;
1317         down_write(&uts_sem);
1318         errno = -EFAULT;
1319         if (!copy_from_user(tmp, name, len)) {
1320                 struct new_utsname *u = utsname();
1321
1322                 memcpy(u->nodename, tmp, len);
1323                 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1324                 errno = 0;
1325         }
1326         uts_proc_notify(UTS_PROC_HOSTNAME);
1327         up_write(&uts_sem);
1328         return errno;
1329 }
1330
1331 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1332
1333 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1334 {
1335         int i, errno;
1336         struct new_utsname *u;
1337
1338         if (len < 0)
1339                 return -EINVAL;
1340         down_read(&uts_sem);
1341         u = utsname();
1342         i = 1 + strlen(u->nodename);
1343         if (i > len)
1344                 i = len;
1345         errno = 0;
1346         if (copy_to_user(name, u->nodename, i))
1347                 errno = -EFAULT;
1348         up_read(&uts_sem);
1349         return errno;
1350 }
1351
1352 #endif
1353
1354 /*
1355  * Only setdomainname; getdomainname can be implemented by calling
1356  * uname()
1357  */
1358 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1359 {
1360         int errno;
1361         char tmp[__NEW_UTS_LEN];
1362
1363         if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1364                 return -EPERM;
1365         if (len < 0 || len > __NEW_UTS_LEN)
1366                 return -EINVAL;
1367
1368         down_write(&uts_sem);
1369         errno = -EFAULT;
1370         if (!copy_from_user(tmp, name, len)) {
1371                 struct new_utsname *u = utsname();
1372
1373                 memcpy(u->domainname, tmp, len);
1374                 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1375                 errno = 0;
1376         }
1377         uts_proc_notify(UTS_PROC_DOMAINNAME);
1378         up_write(&uts_sem);
1379         return errno;
1380 }
1381
1382 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1383 {
1384         struct rlimit value;
1385         int ret;
1386
1387         ret = do_prlimit(current, resource, NULL, &value);
1388         if (!ret)
1389                 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1390
1391         return ret;
1392 }
1393
1394 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1395
1396 /*
1397  *      Back compatibility for getrlimit. Needed for some apps.
1398  */
1399  
1400 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1401                 struct rlimit __user *, rlim)
1402 {
1403         struct rlimit x;
1404         if (resource >= RLIM_NLIMITS)
1405                 return -EINVAL;
1406
1407         task_lock(current->group_leader);
1408         x = current->signal->rlim[resource];
1409         task_unlock(current->group_leader);
1410         if (x.rlim_cur > 0x7FFFFFFF)
1411                 x.rlim_cur = 0x7FFFFFFF;
1412         if (x.rlim_max > 0x7FFFFFFF)
1413                 x.rlim_max = 0x7FFFFFFF;
1414         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1415 }
1416
1417 #endif
1418
1419 static inline bool rlim64_is_infinity(__u64 rlim64)
1420 {
1421 #if BITS_PER_LONG < 64
1422         return rlim64 >= ULONG_MAX;
1423 #else
1424         return rlim64 == RLIM64_INFINITY;
1425 #endif
1426 }
1427
1428 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1429 {
1430         if (rlim->rlim_cur == RLIM_INFINITY)
1431                 rlim64->rlim_cur = RLIM64_INFINITY;
1432         else
1433                 rlim64->rlim_cur = rlim->rlim_cur;
1434         if (rlim->rlim_max == RLIM_INFINITY)
1435                 rlim64->rlim_max = RLIM64_INFINITY;
1436         else
1437                 rlim64->rlim_max = rlim->rlim_max;
1438 }
1439
1440 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1441 {
1442         if (rlim64_is_infinity(rlim64->rlim_cur))
1443                 rlim->rlim_cur = RLIM_INFINITY;
1444         else
1445                 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1446         if (rlim64_is_infinity(rlim64->rlim_max))
1447                 rlim->rlim_max = RLIM_INFINITY;
1448         else
1449                 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1450 }
1451
1452 /* make sure you are allowed to change @tsk limits before calling this */
1453 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1454                 struct rlimit *new_rlim, struct rlimit *old_rlim)
1455 {
1456         struct rlimit *rlim;
1457         int retval = 0;
1458
1459         if (resource >= RLIM_NLIMITS)
1460                 return -EINVAL;
1461         if (new_rlim) {
1462                 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1463                         return -EINVAL;
1464                 if (resource == RLIMIT_NOFILE &&
1465                                 new_rlim->rlim_max > sysctl_nr_open)
1466                         return -EPERM;
1467         }
1468
1469         /* protect tsk->signal and tsk->sighand from disappearing */
1470         read_lock(&tasklist_lock);
1471         if (!tsk->sighand) {
1472                 retval = -ESRCH;
1473                 goto out;
1474         }
1475
1476         rlim = tsk->signal->rlim + resource;
1477         task_lock(tsk->group_leader);
1478         if (new_rlim) {
1479                 /* Keep the capable check against init_user_ns until
1480                    cgroups can contain all limits */
1481                 if (new_rlim->rlim_max > rlim->rlim_max &&
1482                                 !capable(CAP_SYS_RESOURCE))
1483                         retval = -EPERM;
1484                 if (!retval)
1485                         retval = security_task_setrlimit(tsk->group_leader,
1486                                         resource, new_rlim);
1487                 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1488                         /*
1489                          * The caller is asking for an immediate RLIMIT_CPU
1490                          * expiry.  But we use the zero value to mean "it was
1491                          * never set".  So let's cheat and make it one second
1492                          * instead
1493                          */
1494                         new_rlim->rlim_cur = 1;
1495                 }
1496         }
1497         if (!retval) {
1498                 if (old_rlim)
1499                         *old_rlim = *rlim;
1500                 if (new_rlim)
1501                         *rlim = *new_rlim;
1502         }
1503         task_unlock(tsk->group_leader);
1504
1505         /*
1506          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1507          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1508          * very long-standing error, and fixing it now risks breakage of
1509          * applications, so we live with it
1510          */
1511          if (!retval && new_rlim && resource == RLIMIT_CPU &&
1512                          new_rlim->rlim_cur != RLIM_INFINITY)
1513                 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1514 out:
1515         read_unlock(&tasklist_lock);
1516         return retval;
1517 }
1518
1519 /* rcu lock must be held */
1520 static int check_prlimit_permission(struct task_struct *task)
1521 {
1522         const struct cred *cred = current_cred(), *tcred;
1523
1524         if (current == task)
1525                 return 0;
1526
1527         tcred = __task_cred(task);
1528         if (cred->user->user_ns == tcred->user->user_ns &&
1529             (cred->uid == tcred->euid &&
1530              cred->uid == tcred->suid &&
1531              cred->uid == tcred->uid  &&
1532              cred->gid == tcred->egid &&
1533              cred->gid == tcred->sgid &&
1534              cred->gid == tcred->gid))
1535                 return 0;
1536         if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1537                 return 0;
1538
1539         return -EPERM;
1540 }
1541
1542 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1543                 const struct rlimit64 __user *, new_rlim,
1544                 struct rlimit64 __user *, old_rlim)
1545 {
1546         struct rlimit64 old64, new64;
1547         struct rlimit old, new;
1548         struct task_struct *tsk;
1549         int ret;
1550
1551         if (new_rlim) {
1552                 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1553                         return -EFAULT;
1554                 rlim64_to_rlim(&new64, &new);
1555         }
1556
1557         rcu_read_lock();
1558         tsk = pid ? find_task_by_vpid(pid) : current;
1559         if (!tsk) {
1560                 rcu_read_unlock();
1561                 return -ESRCH;
1562         }
1563         ret = check_prlimit_permission(tsk);
1564         if (ret) {
1565                 rcu_read_unlock();
1566                 return ret;
1567         }
1568         get_task_struct(tsk);
1569         rcu_read_unlock();
1570
1571         ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1572                         old_rlim ? &old : NULL);
1573
1574         if (!ret && old_rlim) {
1575                 rlim_to_rlim64(&old, &old64);
1576                 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1577                         ret = -EFAULT;
1578         }
1579
1580         put_task_struct(tsk);
1581         return ret;
1582 }
1583
1584 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1585 {
1586         struct rlimit new_rlim;
1587
1588         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1589                 return -EFAULT;
1590         return do_prlimit(current, resource, &new_rlim, NULL);
1591 }
1592
1593 /*
1594  * It would make sense to put struct rusage in the task_struct,
1595  * except that would make the task_struct be *really big*.  After
1596  * task_struct gets moved into malloc'ed memory, it would
1597  * make sense to do this.  It will make moving the rest of the information
1598  * a lot simpler!  (Which we're not doing right now because we're not
1599  * measuring them yet).
1600  *
1601  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1602  * races with threads incrementing their own counters.  But since word
1603  * reads are atomic, we either get new values or old values and we don't
1604  * care which for the sums.  We always take the siglock to protect reading
1605  * the c* fields from p->signal from races with exit.c updating those
1606  * fields when reaping, so a sample either gets all the additions of a
1607  * given child after it's reaped, or none so this sample is before reaping.
1608  *
1609  * Locking:
1610  * We need to take the siglock for CHILDEREN, SELF and BOTH
1611  * for  the cases current multithreaded, non-current single threaded
1612  * non-current multithreaded.  Thread traversal is now safe with
1613  * the siglock held.
1614  * Strictly speaking, we donot need to take the siglock if we are current and
1615  * single threaded,  as no one else can take our signal_struct away, no one
1616  * else can  reap the  children to update signal->c* counters, and no one else
1617  * can race with the signal-> fields. If we do not take any lock, the
1618  * signal-> fields could be read out of order while another thread was just
1619  * exiting. So we should  place a read memory barrier when we avoid the lock.
1620  * On the writer side,  write memory barrier is implied in  __exit_signal
1621  * as __exit_signal releases  the siglock spinlock after updating the signal->
1622  * fields. But we don't do this yet to keep things simple.
1623  *
1624  */
1625
1626 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1627 {
1628         r->ru_nvcsw += t->nvcsw;
1629         r->ru_nivcsw += t->nivcsw;
1630         r->ru_minflt += t->min_flt;
1631         r->ru_majflt += t->maj_flt;
1632         r->ru_inblock += task_io_get_inblock(t);
1633         r->ru_oublock += task_io_get_oublock(t);
1634 }
1635
1636 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1637 {
1638         struct task_struct *t;
1639         unsigned long flags;
1640         cputime_t tgutime, tgstime, utime, stime;
1641         unsigned long maxrss = 0;
1642
1643         memset((char *) r, 0, sizeof *r);
1644         utime = stime = 0;
1645
1646         if (who == RUSAGE_THREAD) {
1647                 task_times(current, &utime, &stime);
1648                 accumulate_thread_rusage(p, r);
1649                 maxrss = p->signal->maxrss;
1650                 goto out;
1651         }
1652
1653         if (!lock_task_sighand(p, &flags))
1654                 return;
1655
1656         switch (who) {
1657                 case RUSAGE_BOTH:
1658                 case RUSAGE_CHILDREN:
1659                         utime = p->signal->cutime;
1660                         stime = p->signal->cstime;
1661                         r->ru_nvcsw = p->signal->cnvcsw;
1662                         r->ru_nivcsw = p->signal->cnivcsw;
1663                         r->ru_minflt = p->signal->cmin_flt;
1664                         r->ru_majflt = p->signal->cmaj_flt;
1665                         r->ru_inblock = p->signal->cinblock;
1666                         r->ru_oublock = p->signal->coublock;
1667                         maxrss = p->signal->cmaxrss;
1668
1669                         if (who == RUSAGE_CHILDREN)
1670                                 break;
1671
1672                 case RUSAGE_SELF:
1673                         thread_group_times(p, &tgutime, &tgstime);
1674                         utime += tgutime;
1675                         stime += tgstime;
1676                         r->ru_nvcsw += p->signal->nvcsw;
1677                         r->ru_nivcsw += p->signal->nivcsw;
1678                         r->ru_minflt += p->signal->min_flt;
1679                         r->ru_majflt += p->signal->maj_flt;
1680                         r->ru_inblock += p->signal->inblock;
1681                         r->ru_oublock += p->signal->oublock;
1682                         if (maxrss < p->signal->maxrss)
1683                                 maxrss = p->signal->maxrss;
1684                         t = p;
1685                         do {
1686                                 accumulate_thread_rusage(t, r);
1687                                 t = next_thread(t);
1688                         } while (t != p);
1689                         break;
1690
1691                 default:
1692                         BUG();
1693         }
1694         unlock_task_sighand(p, &flags);
1695
1696 out:
1697         cputime_to_timeval(utime, &r->ru_utime);
1698         cputime_to_timeval(stime, &r->ru_stime);
1699
1700         if (who != RUSAGE_CHILDREN) {
1701                 struct mm_struct *mm = get_task_mm(p);
1702                 if (mm) {
1703                         setmax_mm_hiwater_rss(&maxrss, mm);
1704                         mmput(mm);
1705                 }
1706         }
1707         r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1708 }
1709
1710 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1711 {
1712         struct rusage r;
1713         k_getrusage(p, who, &r);
1714         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1715 }
1716
1717 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1718 {
1719         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1720             who != RUSAGE_THREAD)
1721                 return -EINVAL;
1722         return getrusage(current, who, ru);
1723 }
1724
1725 SYSCALL_DEFINE1(umask, int, mask)
1726 {
1727         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1728         return mask;
1729 }
1730
1731 #ifdef CONFIG_CHECKPOINT_RESTORE
1732 static int prctl_set_mm(int opt, unsigned long addr,
1733                         unsigned long arg4, unsigned long arg5)
1734 {
1735         unsigned long rlim = rlimit(RLIMIT_DATA);
1736         unsigned long vm_req_flags;
1737         unsigned long vm_bad_flags;
1738         struct vm_area_struct *vma;
1739         int error = 0;
1740         struct mm_struct *mm = current->mm;
1741
1742         if (arg4 | arg5)
1743                 return -EINVAL;
1744
1745         if (!capable(CAP_SYS_RESOURCE))
1746                 return -EPERM;
1747
1748         if (addr >= TASK_SIZE)
1749                 return -EINVAL;
1750
1751         down_read(&mm->mmap_sem);
1752         vma = find_vma(mm, addr);
1753
1754         if (opt != PR_SET_MM_START_BRK && opt != PR_SET_MM_BRK) {
1755                 /* It must be existing VMA */
1756                 if (!vma || vma->vm_start > addr)
1757                         goto out;
1758         }
1759
1760         error = -EINVAL;
1761         switch (opt) {
1762         case PR_SET_MM_START_CODE:
1763         case PR_SET_MM_END_CODE:
1764                 vm_req_flags = VM_READ | VM_EXEC;
1765                 vm_bad_flags = VM_WRITE | VM_MAYSHARE;
1766
1767                 if ((vma->vm_flags & vm_req_flags) != vm_req_flags ||
1768                     (vma->vm_flags & vm_bad_flags))
1769                         goto out;
1770
1771                 if (opt == PR_SET_MM_START_CODE)
1772                         mm->start_code = addr;
1773                 else
1774                         mm->end_code = addr;
1775                 break;
1776
1777         case PR_SET_MM_START_DATA:
1778         case PR_SET_MM_END_DATA:
1779                 vm_req_flags = VM_READ | VM_WRITE;
1780                 vm_bad_flags = VM_EXEC | VM_MAYSHARE;
1781
1782                 if ((vma->vm_flags & vm_req_flags) != vm_req_flags ||
1783                     (vma->vm_flags & vm_bad_flags))
1784                         goto out;
1785
1786                 if (opt == PR_SET_MM_START_DATA)
1787                         mm->start_data = addr;
1788                 else
1789                         mm->end_data = addr;
1790                 break;
1791
1792         case PR_SET_MM_START_STACK:
1793
1794 #ifdef CONFIG_STACK_GROWSUP
1795                 vm_req_flags = VM_READ | VM_WRITE | VM_GROWSUP;
1796 #else
1797                 vm_req_flags = VM_READ | VM_WRITE | VM_GROWSDOWN;
1798 #endif
1799                 if ((vma->vm_flags & vm_req_flags) != vm_req_flags)
1800                         goto out;
1801
1802                 mm->start_stack = addr;
1803                 break;
1804
1805         case PR_SET_MM_START_BRK:
1806                 if (addr <= mm->end_data)
1807                         goto out;
1808
1809                 if (rlim < RLIM_INFINITY &&
1810                     (mm->brk - addr) +
1811                     (mm->end_data - mm->start_data) > rlim)
1812                         goto out;
1813
1814                 mm->start_brk = addr;
1815                 break;
1816
1817         case PR_SET_MM_BRK:
1818                 if (addr <= mm->end_data)
1819                         goto out;
1820
1821                 if (rlim < RLIM_INFINITY &&
1822                     (addr - mm->start_brk) +
1823                     (mm->end_data - mm->start_data) > rlim)
1824                         goto out;
1825
1826                 mm->brk = addr;
1827                 break;
1828
1829         default:
1830                 error = -EINVAL;
1831                 goto out;
1832         }
1833
1834         error = 0;
1835
1836 out:
1837         up_read(&mm->mmap_sem);
1838
1839         return error;
1840 }
1841 #else /* CONFIG_CHECKPOINT_RESTORE */
1842 static int prctl_set_mm(int opt, unsigned long addr,
1843                         unsigned long arg4, unsigned long arg5)
1844 {
1845         return -EINVAL;
1846 }
1847 #endif
1848
1849 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1850                 unsigned long, arg4, unsigned long, arg5)
1851 {
1852         struct task_struct *me = current;
1853         unsigned char comm[sizeof(me->comm)];
1854         long error;
1855
1856         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1857         if (error != -ENOSYS)
1858                 return error;
1859
1860         error = 0;
1861         switch (option) {
1862                 case PR_SET_PDEATHSIG:
1863                         if (!valid_signal(arg2)) {
1864                                 error = -EINVAL;
1865                                 break;
1866                         }
1867                         me->pdeath_signal = arg2;
1868                         error = 0;
1869                         break;
1870                 case PR_GET_PDEATHSIG:
1871                         error = put_user(me->pdeath_signal, (int __user *)arg2);
1872                         break;
1873                 case PR_GET_DUMPABLE:
1874                         error = get_dumpable(me->mm);
1875                         break;
1876                 case PR_SET_DUMPABLE:
1877                         if (arg2 < 0 || arg2 > 1) {
1878                                 error = -EINVAL;
1879                                 break;
1880                         }
1881                         set_dumpable(me->mm, arg2);
1882                         error = 0;
1883                         break;
1884
1885                 case PR_SET_UNALIGN:
1886                         error = SET_UNALIGN_CTL(me, arg2);
1887                         break;
1888                 case PR_GET_UNALIGN:
1889                         error = GET_UNALIGN_CTL(me, arg2);
1890                         break;
1891                 case PR_SET_FPEMU:
1892                         error = SET_FPEMU_CTL(me, arg2);
1893                         break;
1894                 case PR_GET_FPEMU:
1895                         error = GET_FPEMU_CTL(me, arg2);
1896                         break;
1897                 case PR_SET_FPEXC:
1898                         error = SET_FPEXC_CTL(me, arg2);
1899                         break;
1900                 case PR_GET_FPEXC:
1901                         error = GET_FPEXC_CTL(me, arg2);
1902                         break;
1903                 case PR_GET_TIMING:
1904                         error = PR_TIMING_STATISTICAL;
1905                         break;
1906                 case PR_SET_TIMING:
1907                         if (arg2 != PR_TIMING_STATISTICAL)
1908                                 error = -EINVAL;
1909                         else
1910                                 error = 0;
1911                         break;
1912
1913                 case PR_SET_NAME:
1914                         comm[sizeof(me->comm)-1] = 0;
1915                         if (strncpy_from_user(comm, (char __user *)arg2,
1916                                               sizeof(me->comm) - 1) < 0)
1917                                 return -EFAULT;
1918                         set_task_comm(me, comm);
1919                         proc_comm_connector(me);
1920                         return 0;
1921                 case PR_GET_NAME:
1922                         get_task_comm(comm, me);
1923                         if (copy_to_user((char __user *)arg2, comm,
1924                                          sizeof(comm)))
1925                                 return -EFAULT;
1926                         return 0;
1927                 case PR_GET_ENDIAN:
1928                         error = GET_ENDIAN(me, arg2);
1929                         break;
1930                 case PR_SET_ENDIAN:
1931                         error = SET_ENDIAN(me, arg2);
1932                         break;
1933
1934                 case PR_GET_SECCOMP:
1935                         error = prctl_get_seccomp();
1936                         break;
1937                 case PR_SET_SECCOMP:
1938                         error = prctl_set_seccomp(arg2);
1939                         break;
1940                 case PR_GET_TSC:
1941                         error = GET_TSC_CTL(arg2);
1942                         break;
1943                 case PR_SET_TSC:
1944                         error = SET_TSC_CTL(arg2);
1945                         break;
1946                 case PR_TASK_PERF_EVENTS_DISABLE:
1947                         error = perf_event_task_disable();
1948                         break;
1949                 case PR_TASK_PERF_EVENTS_ENABLE:
1950                         error = perf_event_task_enable();
1951                         break;
1952                 case PR_GET_TIMERSLACK:
1953                         error = current->timer_slack_ns;
1954                         break;
1955                 case PR_SET_TIMERSLACK:
1956                         if (arg2 <= 0)
1957                                 current->timer_slack_ns =
1958                                         current->default_timer_slack_ns;
1959                         else
1960                                 current->timer_slack_ns = arg2;
1961                         error = 0;
1962                         break;
1963                 case PR_MCE_KILL:
1964                         if (arg4 | arg5)
1965                                 return -EINVAL;
1966                         switch (arg2) {
1967                         case PR_MCE_KILL_CLEAR:
1968                                 if (arg3 != 0)
1969                                         return -EINVAL;
1970                                 current->flags &= ~PF_MCE_PROCESS;
1971                                 break;
1972                         case PR_MCE_KILL_SET:
1973                                 current->flags |= PF_MCE_PROCESS;
1974                                 if (arg3 == PR_MCE_KILL_EARLY)
1975                                         current->flags |= PF_MCE_EARLY;
1976                                 else if (arg3 == PR_MCE_KILL_LATE)
1977                                         current->flags &= ~PF_MCE_EARLY;
1978                                 else if (arg3 == PR_MCE_KILL_DEFAULT)
1979                                         current->flags &=
1980                                                 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1981                                 else
1982                                         return -EINVAL;
1983                                 break;
1984                         default:
1985                                 return -EINVAL;
1986                         }
1987                         error = 0;
1988                         break;
1989                 case PR_MCE_KILL_GET:
1990                         if (arg2 | arg3 | arg4 | arg5)
1991                                 return -EINVAL;
1992                         if (current->flags & PF_MCE_PROCESS)
1993                                 error = (current->flags & PF_MCE_EARLY) ?
1994                                         PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1995                         else
1996                                 error = PR_MCE_KILL_DEFAULT;
1997                         break;
1998                 case PR_SET_MM:
1999                         error = prctl_set_mm(arg2, arg3, arg4, arg5);
2000                         break;
2001                 case PR_SET_CHILD_SUBREAPER:
2002                         me->signal->is_child_subreaper = !!arg2;
2003                         error = 0;
2004                         break;
2005                 case PR_GET_CHILD_SUBREAPER:
2006                         error = put_user(me->signal->is_child_subreaper,
2007                                          (int __user *) arg2);
2008                         break;
2009                 default:
2010                         error = -EINVAL;
2011                         break;
2012         }
2013         return error;
2014 }
2015
2016 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2017                 struct getcpu_cache __user *, unused)
2018 {
2019         int err = 0;
2020         int cpu = raw_smp_processor_id();
2021         if (cpup)
2022                 err |= put_user(cpu, cpup);
2023         if (nodep)
2024                 err |= put_user(cpu_to_node(cpu), nodep);
2025         return err ? -EFAULT : 0;
2026 }
2027
2028 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2029
2030 static void argv_cleanup(struct subprocess_info *info)
2031 {
2032         argv_free(info->argv);
2033 }
2034
2035 /**
2036  * orderly_poweroff - Trigger an orderly system poweroff
2037  * @force: force poweroff if command execution fails
2038  *
2039  * This may be called from any context to trigger a system shutdown.
2040  * If the orderly shutdown fails, it will force an immediate shutdown.
2041  */
2042 int orderly_poweroff(bool force)
2043 {
2044         int argc;
2045         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
2046         static char *envp[] = {
2047                 "HOME=/",
2048                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2049                 NULL
2050         };
2051         int ret = -ENOMEM;
2052         struct subprocess_info *info;
2053
2054         if (argv == NULL) {
2055                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2056                        __func__, poweroff_cmd);
2057                 goto out;
2058         }
2059
2060         info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
2061         if (info == NULL) {
2062                 argv_free(argv);
2063                 goto out;
2064         }
2065
2066         call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
2067
2068         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
2069
2070   out:
2071         if (ret && force) {
2072                 printk(KERN_WARNING "Failed to start orderly shutdown: "
2073                        "forcing the issue\n");
2074
2075                 /* I guess this should try to kick off some daemon to
2076                    sync and poweroff asap.  Or not even bother syncing
2077                    if we're doing an emergency shutdown? */
2078                 emergency_sync();
2079                 kernel_power_off();
2080         }
2081
2082         return ret;
2083 }
2084 EXPORT_SYMBOL_GPL(orderly_poweroff);