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