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