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