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