dac10161ca2384c59340d4fab96eae4237f2f8af
[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/config.h>
8 #include <linux/module.h>
9 #include <linux/mm.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/smp_lock.h>
13 #include <linux/notifier.h>
14 #include <linux/reboot.h>
15 #include <linux/prctl.h>
16 #include <linux/init.h>
17 #include <linux/highuid.h>
18 #include <linux/fs.h>
19 #include <linux/workqueue.h>
20 #include <linux/device.h>
21 #include <linux/key.h>
22 #include <linux/times.h>
23 #include <linux/posix-timers.h>
24 #include <linux/security.h>
25 #include <linux/dcookies.h>
26 #include <linux/suspend.h>
27 #include <linux/tty.h>
28 #include <linux/signal.h>
29
30 #include <linux/compat.h>
31 #include <linux/syscalls.h>
32
33 #include <asm/uaccess.h>
34 #include <asm/io.h>
35 #include <asm/unistd.h>
36
37 #ifndef SET_UNALIGN_CTL
38 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
39 #endif
40 #ifndef GET_UNALIGN_CTL
41 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
42 #endif
43 #ifndef SET_FPEMU_CTL
44 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
45 #endif
46 #ifndef GET_FPEMU_CTL
47 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
48 #endif
49 #ifndef SET_FPEXC_CTL
50 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
51 #endif
52 #ifndef GET_FPEXC_CTL
53 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
54 #endif
55
56 /*
57  * this is where the system-wide overflow UID and GID are defined, for
58  * architectures that now have 32-bit UID/GID but didn't in the past
59  */
60
61 int overflowuid = DEFAULT_OVERFLOWUID;
62 int overflowgid = DEFAULT_OVERFLOWGID;
63
64 #ifdef CONFIG_UID16
65 EXPORT_SYMBOL(overflowuid);
66 EXPORT_SYMBOL(overflowgid);
67 #endif
68
69 /*
70  * the same as above, but for filesystems which can only store a 16-bit
71  * UID and GID. as such, this is needed on all architectures
72  */
73
74 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
75 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
76
77 EXPORT_SYMBOL(fs_overflowuid);
78 EXPORT_SYMBOL(fs_overflowgid);
79
80 /*
81  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
82  */
83
84 int C_A_D = 1;
85 int cad_pid = 1;
86
87 /*
88  *      Notifier list for kernel code which wants to be called
89  *      at shutdown. This is used to stop any idling DMA operations
90  *      and the like. 
91  */
92
93 static struct notifier_block *reboot_notifier_list;
94 static DEFINE_RWLOCK(notifier_lock);
95
96 /**
97  *      notifier_chain_register - Add notifier to a notifier chain
98  *      @list: Pointer to root list pointer
99  *      @n: New entry in notifier chain
100  *
101  *      Adds a notifier to a notifier chain.
102  *
103  *      Currently always returns zero.
104  */
105  
106 int notifier_chain_register(struct notifier_block **list, struct notifier_block *n)
107 {
108         write_lock(&notifier_lock);
109         while(*list)
110         {
111                 if(n->priority > (*list)->priority)
112                         break;
113                 list= &((*list)->next);
114         }
115         n->next = *list;
116         *list=n;
117         write_unlock(&notifier_lock);
118         return 0;
119 }
120
121 EXPORT_SYMBOL(notifier_chain_register);
122
123 /**
124  *      notifier_chain_unregister - Remove notifier from a notifier chain
125  *      @nl: Pointer to root list pointer
126  *      @n: New entry in notifier chain
127  *
128  *      Removes a notifier from a notifier chain.
129  *
130  *      Returns zero on success, or %-ENOENT on failure.
131  */
132  
133 int notifier_chain_unregister(struct notifier_block **nl, struct notifier_block *n)
134 {
135         write_lock(&notifier_lock);
136         while((*nl)!=NULL)
137         {
138                 if((*nl)==n)
139                 {
140                         *nl=n->next;
141                         write_unlock(&notifier_lock);
142                         return 0;
143                 }
144                 nl=&((*nl)->next);
145         }
146         write_unlock(&notifier_lock);
147         return -ENOENT;
148 }
149
150 EXPORT_SYMBOL(notifier_chain_unregister);
151
152 /**
153  *      notifier_call_chain - Call functions in a notifier chain
154  *      @n: Pointer to root pointer of notifier chain
155  *      @val: Value passed unmodified to notifier function
156  *      @v: Pointer passed unmodified to notifier function
157  *
158  *      Calls each function in a notifier chain in turn.
159  *
160  *      If the return value of the notifier can be and'd
161  *      with %NOTIFY_STOP_MASK, then notifier_call_chain
162  *      will return immediately, with the return value of
163  *      the notifier function which halted execution.
164  *      Otherwise, the return value is the return value
165  *      of the last notifier function called.
166  */
167  
168 int notifier_call_chain(struct notifier_block **n, unsigned long val, void *v)
169 {
170         int ret=NOTIFY_DONE;
171         struct notifier_block *nb = *n;
172
173         while(nb)
174         {
175                 ret=nb->notifier_call(nb,val,v);
176                 if(ret&NOTIFY_STOP_MASK)
177                 {
178                         return ret;
179                 }
180                 nb=nb->next;
181         }
182         return ret;
183 }
184
185 EXPORT_SYMBOL(notifier_call_chain);
186
187 /**
188  *      register_reboot_notifier - Register function to be called at reboot time
189  *      @nb: Info about notifier function to be called
190  *
191  *      Registers a function with the list of functions
192  *      to be called at reboot time.
193  *
194  *      Currently always returns zero, as notifier_chain_register
195  *      always returns zero.
196  */
197  
198 int register_reboot_notifier(struct notifier_block * nb)
199 {
200         return notifier_chain_register(&reboot_notifier_list, nb);
201 }
202
203 EXPORT_SYMBOL(register_reboot_notifier);
204
205 /**
206  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
207  *      @nb: Hook to be unregistered
208  *
209  *      Unregisters a previously registered reboot
210  *      notifier function.
211  *
212  *      Returns zero on success, or %-ENOENT on failure.
213  */
214  
215 int unregister_reboot_notifier(struct notifier_block * nb)
216 {
217         return notifier_chain_unregister(&reboot_notifier_list, nb);
218 }
219
220 EXPORT_SYMBOL(unregister_reboot_notifier);
221
222 static int set_one_prio(struct task_struct *p, int niceval, int error)
223 {
224         int no_nice;
225
226         if (p->uid != current->euid &&
227                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
228                 error = -EPERM;
229                 goto out;
230         }
231         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
232                 error = -EACCES;
233                 goto out;
234         }
235         no_nice = security_task_setnice(p, niceval);
236         if (no_nice) {
237                 error = no_nice;
238                 goto out;
239         }
240         if (error == -ESRCH)
241                 error = 0;
242         set_user_nice(p, niceval);
243 out:
244         return error;
245 }
246
247 asmlinkage long sys_setpriority(int which, int who, int niceval)
248 {
249         struct task_struct *g, *p;
250         struct user_struct *user;
251         int error = -EINVAL;
252
253         if (which > 2 || which < 0)
254                 goto out;
255
256         /* normalize: avoid signed division (rounding problems) */
257         error = -ESRCH;
258         if (niceval < -20)
259                 niceval = -20;
260         if (niceval > 19)
261                 niceval = 19;
262
263         read_lock(&tasklist_lock);
264         switch (which) {
265                 case PRIO_PROCESS:
266                         if (!who)
267                                 who = current->pid;
268                         p = find_task_by_pid(who);
269                         if (p)
270                                 error = set_one_prio(p, niceval, error);
271                         break;
272                 case PRIO_PGRP:
273                         if (!who)
274                                 who = process_group(current);
275                         do_each_task_pid(who, PIDTYPE_PGID, p) {
276                                 error = set_one_prio(p, niceval, error);
277                         } while_each_task_pid(who, PIDTYPE_PGID, p);
278                         break;
279                 case PRIO_USER:
280                         user = current->user;
281                         if (!who)
282                                 who = current->uid;
283                         else
284                                 if ((who != current->uid) && !(user = find_user(who)))
285                                         goto out_unlock;        /* No processes for this user */
286
287                         do_each_thread(g, p)
288                                 if (p->uid == who)
289                                         error = set_one_prio(p, niceval, error);
290                         while_each_thread(g, p);
291                         if (who != current->uid)
292                                 free_uid(user);         /* For find_user() */
293                         break;
294         }
295 out_unlock:
296         read_unlock(&tasklist_lock);
297 out:
298         return error;
299 }
300
301 /*
302  * Ugh. To avoid negative return values, "getpriority()" will
303  * not return the normal nice-value, but a negated value that
304  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
305  * to stay compatible.
306  */
307 asmlinkage long sys_getpriority(int which, int who)
308 {
309         struct task_struct *g, *p;
310         struct user_struct *user;
311         long niceval, retval = -ESRCH;
312
313         if (which > 2 || which < 0)
314                 return -EINVAL;
315
316         read_lock(&tasklist_lock);
317         switch (which) {
318                 case PRIO_PROCESS:
319                         if (!who)
320                                 who = current->pid;
321                         p = find_task_by_pid(who);
322                         if (p) {
323                                 niceval = 20 - task_nice(p);
324                                 if (niceval > retval)
325                                         retval = niceval;
326                         }
327                         break;
328                 case PRIO_PGRP:
329                         if (!who)
330                                 who = process_group(current);
331                         do_each_task_pid(who, PIDTYPE_PGID, p) {
332                                 niceval = 20 - task_nice(p);
333                                 if (niceval > retval)
334                                         retval = niceval;
335                         } while_each_task_pid(who, PIDTYPE_PGID, p);
336                         break;
337                 case PRIO_USER:
338                         user = current->user;
339                         if (!who)
340                                 who = current->uid;
341                         else
342                                 if ((who != current->uid) && !(user = find_user(who)))
343                                         goto out_unlock;        /* No processes for this user */
344
345                         do_each_thread(g, p)
346                                 if (p->uid == who) {
347                                         niceval = 20 - task_nice(p);
348                                         if (niceval > retval)
349                                                 retval = niceval;
350                                 }
351                         while_each_thread(g, p);
352                         if (who != current->uid)
353                                 free_uid(user);         /* for find_user() */
354                         break;
355         }
356 out_unlock:
357         read_unlock(&tasklist_lock);
358
359         return retval;
360 }
361
362
363 /*
364  * Reboot system call: for obvious reasons only root may call it,
365  * and even root needs to set up some magic numbers in the registers
366  * so that some mistake won't make this reboot the whole machine.
367  * You can also set the meaning of the ctrl-alt-del-key here.
368  *
369  * reboot doesn't sync: do that yourself before calling this.
370  */
371 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
372 {
373         char buffer[256];
374
375         /* We only trust the superuser with rebooting the system. */
376         if (!capable(CAP_SYS_BOOT))
377                 return -EPERM;
378
379         /* For safety, we require "magic" arguments. */
380         if (magic1 != LINUX_REBOOT_MAGIC1 ||
381             (magic2 != LINUX_REBOOT_MAGIC2 &&
382                         magic2 != LINUX_REBOOT_MAGIC2A &&
383                         magic2 != LINUX_REBOOT_MAGIC2B &&
384                         magic2 != LINUX_REBOOT_MAGIC2C))
385                 return -EINVAL;
386
387         lock_kernel();
388         switch (cmd) {
389         case LINUX_REBOOT_CMD_RESTART:
390                 notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
391                 system_state = SYSTEM_RESTART;
392                 device_shutdown();
393                 printk(KERN_EMERG "Restarting system.\n");
394                 machine_restart(NULL);
395                 break;
396
397         case LINUX_REBOOT_CMD_CAD_ON:
398                 C_A_D = 1;
399                 break;
400
401         case LINUX_REBOOT_CMD_CAD_OFF:
402                 C_A_D = 0;
403                 break;
404
405         case LINUX_REBOOT_CMD_HALT:
406                 notifier_call_chain(&reboot_notifier_list, SYS_HALT, NULL);
407                 system_state = SYSTEM_HALT;
408                 device_suspend(PMSG_SUSPEND);
409                 device_shutdown();
410                 printk(KERN_EMERG "System halted.\n");
411                 machine_halt();
412                 unlock_kernel();
413                 do_exit(0);
414                 break;
415
416         case LINUX_REBOOT_CMD_POWER_OFF:
417                 notifier_call_chain(&reboot_notifier_list, SYS_POWER_OFF, NULL);
418                 system_state = SYSTEM_POWER_OFF;
419                 device_suspend(PMSG_SUSPEND);
420                 device_shutdown();
421                 printk(KERN_EMERG "Power down.\n");
422                 machine_power_off();
423                 unlock_kernel();
424                 do_exit(0);
425                 break;
426
427         case LINUX_REBOOT_CMD_RESTART2:
428                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
429                         unlock_kernel();
430                         return -EFAULT;
431                 }
432                 buffer[sizeof(buffer) - 1] = '\0';
433
434                 notifier_call_chain(&reboot_notifier_list, SYS_RESTART, buffer);
435                 system_state = SYSTEM_RESTART;
436                 device_suspend(PMSG_FREEZE);
437                 device_shutdown();
438                 printk(KERN_EMERG "Restarting system with command '%s'.\n", buffer);
439                 machine_restart(buffer);
440                 break;
441
442 #ifdef CONFIG_SOFTWARE_SUSPEND
443         case LINUX_REBOOT_CMD_SW_SUSPEND:
444                 {
445                         int ret = software_suspend();
446                         unlock_kernel();
447                         return ret;
448                 }
449 #endif
450
451         default:
452                 unlock_kernel();
453                 return -EINVAL;
454         }
455         unlock_kernel();
456         return 0;
457 }
458
459 static void deferred_cad(void *dummy)
460 {
461         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
462         machine_restart(NULL);
463 }
464
465 /*
466  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
467  * As it's called within an interrupt, it may NOT sync: the only choice
468  * is whether to reboot at once, or just ignore the ctrl-alt-del.
469  */
470 void ctrl_alt_del(void)
471 {
472         static DECLARE_WORK(cad_work, deferred_cad, NULL);
473
474         if (C_A_D)
475                 schedule_work(&cad_work);
476         else
477                 kill_proc(cad_pid, SIGINT, 1);
478 }
479         
480
481 /*
482  * Unprivileged users may change the real gid to the effective gid
483  * or vice versa.  (BSD-style)
484  *
485  * If you set the real gid at all, or set the effective gid to a value not
486  * equal to the real gid, then the saved gid is set to the new effective gid.
487  *
488  * This makes it possible for a setgid program to completely drop its
489  * privileges, which is often a useful assertion to make when you are doing
490  * a security audit over a program.
491  *
492  * The general idea is that a program which uses just setregid() will be
493  * 100% compatible with BSD.  A program which uses just setgid() will be
494  * 100% compatible with POSIX with saved IDs. 
495  *
496  * SMP: There are not races, the GIDs are checked only by filesystem
497  *      operations (as far as semantic preservation is concerned).
498  */
499 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
500 {
501         int old_rgid = current->gid;
502         int old_egid = current->egid;
503         int new_rgid = old_rgid;
504         int new_egid = old_egid;
505         int retval;
506
507         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
508         if (retval)
509                 return retval;
510
511         if (rgid != (gid_t) -1) {
512                 if ((old_rgid == rgid) ||
513                     (current->egid==rgid) ||
514                     capable(CAP_SETGID))
515                         new_rgid = rgid;
516                 else
517                         return -EPERM;
518         }
519         if (egid != (gid_t) -1) {
520                 if ((old_rgid == egid) ||
521                     (current->egid == egid) ||
522                     (current->sgid == egid) ||
523                     capable(CAP_SETGID))
524                         new_egid = egid;
525                 else {
526                         return -EPERM;
527                 }
528         }
529         if (new_egid != old_egid)
530         {
531                 current->mm->dumpable = suid_dumpable;
532                 smp_wmb();
533         }
534         if (rgid != (gid_t) -1 ||
535             (egid != (gid_t) -1 && egid != old_rgid))
536                 current->sgid = new_egid;
537         current->fsgid = new_egid;
538         current->egid = new_egid;
539         current->gid = new_rgid;
540         key_fsgid_changed(current);
541         return 0;
542 }
543
544 /*
545  * setgid() is implemented like SysV w/ SAVED_IDS 
546  *
547  * SMP: Same implicit races as above.
548  */
549 asmlinkage long sys_setgid(gid_t gid)
550 {
551         int old_egid = current->egid;
552         int retval;
553
554         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
555         if (retval)
556                 return retval;
557
558         if (capable(CAP_SETGID))
559         {
560                 if(old_egid != gid)
561                 {
562                         current->mm->dumpable = suid_dumpable;
563                         smp_wmb();
564                 }
565                 current->gid = current->egid = current->sgid = current->fsgid = gid;
566         }
567         else if ((gid == current->gid) || (gid == current->sgid))
568         {
569                 if(old_egid != gid)
570                 {
571                         current->mm->dumpable = suid_dumpable;
572                         smp_wmb();
573                 }
574                 current->egid = current->fsgid = gid;
575         }
576         else
577                 return -EPERM;
578
579         key_fsgid_changed(current);
580         return 0;
581 }
582   
583 static int set_user(uid_t new_ruid, int dumpclear)
584 {
585         struct user_struct *new_user;
586
587         new_user = alloc_uid(new_ruid);
588         if (!new_user)
589                 return -EAGAIN;
590
591         if (atomic_read(&new_user->processes) >=
592                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
593                         new_user != &root_user) {
594                 free_uid(new_user);
595                 return -EAGAIN;
596         }
597
598         switch_uid(new_user);
599
600         if(dumpclear)
601         {
602                 current->mm->dumpable = suid_dumpable;
603                 smp_wmb();
604         }
605         current->uid = new_ruid;
606         return 0;
607 }
608
609 /*
610  * Unprivileged users may change the real uid to the effective uid
611  * or vice versa.  (BSD-style)
612  *
613  * If you set the real uid at all, or set the effective uid to a value not
614  * equal to the real uid, then the saved uid is set to the new effective uid.
615  *
616  * This makes it possible for a setuid program to completely drop its
617  * privileges, which is often a useful assertion to make when you are doing
618  * a security audit over a program.
619  *
620  * The general idea is that a program which uses just setreuid() will be
621  * 100% compatible with BSD.  A program which uses just setuid() will be
622  * 100% compatible with POSIX with saved IDs. 
623  */
624 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
625 {
626         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
627         int retval;
628
629         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
630         if (retval)
631                 return retval;
632
633         new_ruid = old_ruid = current->uid;
634         new_euid = old_euid = current->euid;
635         old_suid = current->suid;
636
637         if (ruid != (uid_t) -1) {
638                 new_ruid = ruid;
639                 if ((old_ruid != ruid) &&
640                     (current->euid != ruid) &&
641                     !capable(CAP_SETUID))
642                         return -EPERM;
643         }
644
645         if (euid != (uid_t) -1) {
646                 new_euid = euid;
647                 if ((old_ruid != euid) &&
648                     (current->euid != euid) &&
649                     (current->suid != euid) &&
650                     !capable(CAP_SETUID))
651                         return -EPERM;
652         }
653
654         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
655                 return -EAGAIN;
656
657         if (new_euid != old_euid)
658         {
659                 current->mm->dumpable = suid_dumpable;
660                 smp_wmb();
661         }
662         current->fsuid = current->euid = new_euid;
663         if (ruid != (uid_t) -1 ||
664             (euid != (uid_t) -1 && euid != old_ruid))
665                 current->suid = current->euid;
666         current->fsuid = current->euid;
667
668         key_fsuid_changed(current);
669
670         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
671 }
672
673
674                 
675 /*
676  * setuid() is implemented like SysV with SAVED_IDS 
677  * 
678  * Note that SAVED_ID's is deficient in that a setuid root program
679  * like sendmail, for example, cannot set its uid to be a normal 
680  * user and then switch back, because if you're root, setuid() sets
681  * the saved uid too.  If you don't like this, blame the bright people
682  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
683  * will allow a root program to temporarily drop privileges and be able to
684  * regain them by swapping the real and effective uid.  
685  */
686 asmlinkage long sys_setuid(uid_t uid)
687 {
688         int old_euid = current->euid;
689         int old_ruid, old_suid, new_ruid, new_suid;
690         int retval;
691
692         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
693         if (retval)
694                 return retval;
695
696         old_ruid = new_ruid = current->uid;
697         old_suid = current->suid;
698         new_suid = old_suid;
699         
700         if (capable(CAP_SETUID)) {
701                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
702                         return -EAGAIN;
703                 new_suid = uid;
704         } else if ((uid != current->uid) && (uid != new_suid))
705                 return -EPERM;
706
707         if (old_euid != uid)
708         {
709                 current->mm->dumpable = suid_dumpable;
710                 smp_wmb();
711         }
712         current->fsuid = current->euid = uid;
713         current->suid = new_suid;
714
715         key_fsuid_changed(current);
716
717         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
718 }
719
720
721 /*
722  * This function implements a generic ability to update ruid, euid,
723  * and suid.  This allows you to implement the 4.4 compatible seteuid().
724  */
725 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
726 {
727         int old_ruid = current->uid;
728         int old_euid = current->euid;
729         int old_suid = current->suid;
730         int retval;
731
732         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
733         if (retval)
734                 return retval;
735
736         if (!capable(CAP_SETUID)) {
737                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
738                     (ruid != current->euid) && (ruid != current->suid))
739                         return -EPERM;
740                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
741                     (euid != current->euid) && (euid != current->suid))
742                         return -EPERM;
743                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
744                     (suid != current->euid) && (suid != current->suid))
745                         return -EPERM;
746         }
747         if (ruid != (uid_t) -1) {
748                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
749                         return -EAGAIN;
750         }
751         if (euid != (uid_t) -1) {
752                 if (euid != current->euid)
753                 {
754                         current->mm->dumpable = suid_dumpable;
755                         smp_wmb();
756                 }
757                 current->euid = euid;
758         }
759         current->fsuid = current->euid;
760         if (suid != (uid_t) -1)
761                 current->suid = suid;
762
763         key_fsuid_changed(current);
764
765         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
766 }
767
768 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
769 {
770         int retval;
771
772         if (!(retval = put_user(current->uid, ruid)) &&
773             !(retval = put_user(current->euid, euid)))
774                 retval = put_user(current->suid, suid);
775
776         return retval;
777 }
778
779 /*
780  * Same as above, but for rgid, egid, sgid.
781  */
782 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
783 {
784         int retval;
785
786         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
787         if (retval)
788                 return retval;
789
790         if (!capable(CAP_SETGID)) {
791                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
792                     (rgid != current->egid) && (rgid != current->sgid))
793                         return -EPERM;
794                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
795                     (egid != current->egid) && (egid != current->sgid))
796                         return -EPERM;
797                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
798                     (sgid != current->egid) && (sgid != current->sgid))
799                         return -EPERM;
800         }
801         if (egid != (gid_t) -1) {
802                 if (egid != current->egid)
803                 {
804                         current->mm->dumpable = suid_dumpable;
805                         smp_wmb();
806                 }
807                 current->egid = egid;
808         }
809         current->fsgid = current->egid;
810         if (rgid != (gid_t) -1)
811                 current->gid = rgid;
812         if (sgid != (gid_t) -1)
813                 current->sgid = sgid;
814
815         key_fsgid_changed(current);
816         return 0;
817 }
818
819 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
820 {
821         int retval;
822
823         if (!(retval = put_user(current->gid, rgid)) &&
824             !(retval = put_user(current->egid, egid)))
825                 retval = put_user(current->sgid, sgid);
826
827         return retval;
828 }
829
830
831 /*
832  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
833  * is used for "access()" and for the NFS daemon (letting nfsd stay at
834  * whatever uid it wants to). It normally shadows "euid", except when
835  * explicitly set by setfsuid() or for access..
836  */
837 asmlinkage long sys_setfsuid(uid_t uid)
838 {
839         int old_fsuid;
840
841         old_fsuid = current->fsuid;
842         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
843                 return old_fsuid;
844
845         if (uid == current->uid || uid == current->euid ||
846             uid == current->suid || uid == current->fsuid || 
847             capable(CAP_SETUID))
848         {
849                 if (uid != old_fsuid)
850                 {
851                         current->mm->dumpable = suid_dumpable;
852                         smp_wmb();
853                 }
854                 current->fsuid = uid;
855         }
856
857         key_fsuid_changed(current);
858
859         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
860
861         return old_fsuid;
862 }
863
864 /*
865  * Samma på svenska..
866  */
867 asmlinkage long sys_setfsgid(gid_t gid)
868 {
869         int old_fsgid;
870
871         old_fsgid = current->fsgid;
872         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
873                 return old_fsgid;
874
875         if (gid == current->gid || gid == current->egid ||
876             gid == current->sgid || gid == current->fsgid || 
877             capable(CAP_SETGID))
878         {
879                 if (gid != old_fsgid)
880                 {
881                         current->mm->dumpable = suid_dumpable;
882                         smp_wmb();
883                 }
884                 current->fsgid = gid;
885                 key_fsgid_changed(current);
886         }
887         return old_fsgid;
888 }
889
890 asmlinkage long sys_times(struct tms __user * tbuf)
891 {
892         /*
893          *      In the SMP world we might just be unlucky and have one of
894          *      the times increment as we use it. Since the value is an
895          *      atomically safe type this is just fine. Conceptually its
896          *      as if the syscall took an instant longer to occur.
897          */
898         if (tbuf) {
899                 struct tms tmp;
900                 cputime_t utime, stime, cutime, cstime;
901
902 #ifdef CONFIG_SMP
903                 if (thread_group_empty(current)) {
904                         /*
905                          * Single thread case without the use of any locks.
906                          *
907                          * We may race with release_task if two threads are
908                          * executing. However, release task first adds up the
909                          * counters (__exit_signal) before  removing the task
910                          * from the process tasklist (__unhash_process).
911                          * __exit_signal also acquires and releases the
912                          * siglock which results in the proper memory ordering
913                          * so that the list modifications are always visible
914                          * after the counters have been updated.
915                          *
916                          * If the counters have been updated by the second thread
917                          * but the thread has not yet been removed from the list
918                          * then the other branch will be executing which will
919                          * block on tasklist_lock until the exit handling of the
920                          * other task is finished.
921                          *
922                          * This also implies that the sighand->siglock cannot
923                          * be held by another processor. So we can also
924                          * skip acquiring that lock.
925                          */
926                         utime = cputime_add(current->signal->utime, current->utime);
927                         stime = cputime_add(current->signal->utime, current->stime);
928                         cutime = current->signal->cutime;
929                         cstime = current->signal->cstime;
930                 } else
931 #endif
932                 {
933
934                         /* Process with multiple threads */
935                         struct task_struct *tsk = current;
936                         struct task_struct *t;
937
938                         read_lock(&tasklist_lock);
939                         utime = tsk->signal->utime;
940                         stime = tsk->signal->stime;
941                         t = tsk;
942                         do {
943                                 utime = cputime_add(utime, t->utime);
944                                 stime = cputime_add(stime, t->stime);
945                                 t = next_thread(t);
946                         } while (t != tsk);
947
948                         /*
949                          * While we have tasklist_lock read-locked, no dying thread
950                          * can be updating current->signal->[us]time.  Instead,
951                          * we got their counts included in the live thread loop.
952                          * However, another thread can come in right now and
953                          * do a wait call that updates current->signal->c[us]time.
954                          * To make sure we always see that pair updated atomically,
955                          * we take the siglock around fetching them.
956                          */
957                         spin_lock_irq(&tsk->sighand->siglock);
958                         cutime = tsk->signal->cutime;
959                         cstime = tsk->signal->cstime;
960                         spin_unlock_irq(&tsk->sighand->siglock);
961                         read_unlock(&tasklist_lock);
962                 }
963                 tmp.tms_utime = cputime_to_clock_t(utime);
964                 tmp.tms_stime = cputime_to_clock_t(stime);
965                 tmp.tms_cutime = cputime_to_clock_t(cutime);
966                 tmp.tms_cstime = cputime_to_clock_t(cstime);
967                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
968                         return -EFAULT;
969         }
970         return (long) jiffies_64_to_clock_t(get_jiffies_64());
971 }
972
973 /*
974  * This needs some heavy checking ...
975  * I just haven't the stomach for it. I also don't fully
976  * understand sessions/pgrp etc. Let somebody who does explain it.
977  *
978  * OK, I think I have the protection semantics right.... this is really
979  * only important on a multi-user system anyway, to make sure one user
980  * can't send a signal to a process owned by another.  -TYT, 12/12/91
981  *
982  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
983  * LBT 04.03.94
984  */
985
986 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
987 {
988         struct task_struct *p;
989         int err = -EINVAL;
990
991         if (!pid)
992                 pid = current->pid;
993         if (!pgid)
994                 pgid = pid;
995         if (pgid < 0)
996                 return -EINVAL;
997
998         /* From this point forward we keep holding onto the tasklist lock
999          * so that our parent does not change from under us. -DaveM
1000          */
1001         write_lock_irq(&tasklist_lock);
1002
1003         err = -ESRCH;
1004         p = find_task_by_pid(pid);
1005         if (!p)
1006                 goto out;
1007
1008         err = -EINVAL;
1009         if (!thread_group_leader(p))
1010                 goto out;
1011
1012         if (p->parent == current || p->real_parent == current) {
1013                 err = -EPERM;
1014                 if (p->signal->session != current->signal->session)
1015                         goto out;
1016                 err = -EACCES;
1017                 if (p->did_exec)
1018                         goto out;
1019         } else {
1020                 err = -ESRCH;
1021                 if (p != current)
1022                         goto out;
1023         }
1024
1025         err = -EPERM;
1026         if (p->signal->leader)
1027                 goto out;
1028
1029         if (pgid != pid) {
1030                 struct task_struct *p;
1031
1032                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1033                         if (p->signal->session == current->signal->session)
1034                                 goto ok_pgid;
1035                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1036                 goto out;
1037         }
1038
1039 ok_pgid:
1040         err = security_task_setpgid(p, pgid);
1041         if (err)
1042                 goto out;
1043
1044         if (process_group(p) != pgid) {
1045                 detach_pid(p, PIDTYPE_PGID);
1046                 p->signal->pgrp = pgid;
1047                 attach_pid(p, PIDTYPE_PGID, pgid);
1048         }
1049
1050         err = 0;
1051 out:
1052         /* All paths lead to here, thus we are safe. -DaveM */
1053         write_unlock_irq(&tasklist_lock);
1054         return err;
1055 }
1056
1057 asmlinkage long sys_getpgid(pid_t pid)
1058 {
1059         if (!pid) {
1060                 return process_group(current);
1061         } else {
1062                 int retval;
1063                 struct task_struct *p;
1064
1065                 read_lock(&tasklist_lock);
1066                 p = find_task_by_pid(pid);
1067
1068                 retval = -ESRCH;
1069                 if (p) {
1070                         retval = security_task_getpgid(p);
1071                         if (!retval)
1072                                 retval = process_group(p);
1073                 }
1074                 read_unlock(&tasklist_lock);
1075                 return retval;
1076         }
1077 }
1078
1079 #ifdef __ARCH_WANT_SYS_GETPGRP
1080
1081 asmlinkage long sys_getpgrp(void)
1082 {
1083         /* SMP - assuming writes are word atomic this is fine */
1084         return process_group(current);
1085 }
1086
1087 #endif
1088
1089 asmlinkage long sys_getsid(pid_t pid)
1090 {
1091         if (!pid) {
1092                 return current->signal->session;
1093         } else {
1094                 int retval;
1095                 struct task_struct *p;
1096
1097                 read_lock(&tasklist_lock);
1098                 p = find_task_by_pid(pid);
1099
1100                 retval = -ESRCH;
1101                 if(p) {
1102                         retval = security_task_getsid(p);
1103                         if (!retval)
1104                                 retval = p->signal->session;
1105                 }
1106                 read_unlock(&tasklist_lock);
1107                 return retval;
1108         }
1109 }
1110
1111 asmlinkage long sys_setsid(void)
1112 {
1113         struct pid *pid;
1114         int err = -EPERM;
1115
1116         if (!thread_group_leader(current))
1117                 return -EINVAL;
1118
1119         down(&tty_sem);
1120         write_lock_irq(&tasklist_lock);
1121
1122         pid = find_pid(PIDTYPE_PGID, current->pid);
1123         if (pid)
1124                 goto out;
1125
1126         current->signal->leader = 1;
1127         __set_special_pids(current->pid, current->pid);
1128         current->signal->tty = NULL;
1129         current->signal->tty_old_pgrp = 0;
1130         err = process_group(current);
1131 out:
1132         write_unlock_irq(&tasklist_lock);
1133         up(&tty_sem);
1134         return err;
1135 }
1136
1137 /*
1138  * Supplementary group IDs
1139  */
1140
1141 /* init to 2 - one for init_task, one to ensure it is never freed */
1142 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1143
1144 struct group_info *groups_alloc(int gidsetsize)
1145 {
1146         struct group_info *group_info;
1147         int nblocks;
1148         int i;
1149
1150         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1151         /* Make sure we always allocate at least one indirect block pointer */
1152         nblocks = nblocks ? : 1;
1153         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1154         if (!group_info)
1155                 return NULL;
1156         group_info->ngroups = gidsetsize;
1157         group_info->nblocks = nblocks;
1158         atomic_set(&group_info->usage, 1);
1159
1160         if (gidsetsize <= NGROUPS_SMALL) {
1161                 group_info->blocks[0] = group_info->small_block;
1162         } else {
1163                 for (i = 0; i < nblocks; i++) {
1164                         gid_t *b;
1165                         b = (void *)__get_free_page(GFP_USER);
1166                         if (!b)
1167                                 goto out_undo_partial_alloc;
1168                         group_info->blocks[i] = b;
1169                 }
1170         }
1171         return group_info;
1172
1173 out_undo_partial_alloc:
1174         while (--i >= 0) {
1175                 free_page((unsigned long)group_info->blocks[i]);
1176         }
1177         kfree(group_info);
1178         return NULL;
1179 }
1180
1181 EXPORT_SYMBOL(groups_alloc);
1182
1183 void groups_free(struct group_info *group_info)
1184 {
1185         if (group_info->blocks[0] != group_info->small_block) {
1186                 int i;
1187                 for (i = 0; i < group_info->nblocks; i++)
1188                         free_page((unsigned long)group_info->blocks[i]);
1189         }
1190         kfree(group_info);
1191 }
1192
1193 EXPORT_SYMBOL(groups_free);
1194
1195 /* export the group_info to a user-space array */
1196 static int groups_to_user(gid_t __user *grouplist,
1197     struct group_info *group_info)
1198 {
1199         int i;
1200         int count = group_info->ngroups;
1201
1202         for (i = 0; i < group_info->nblocks; i++) {
1203                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1204                 int off = i * NGROUPS_PER_BLOCK;
1205                 int len = cp_count * sizeof(*grouplist);
1206
1207                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1208                         return -EFAULT;
1209
1210                 count -= cp_count;
1211         }
1212         return 0;
1213 }
1214
1215 /* fill a group_info from a user-space array - it must be allocated already */
1216 static int groups_from_user(struct group_info *group_info,
1217     gid_t __user *grouplist)
1218  {
1219         int i;
1220         int count = group_info->ngroups;
1221
1222         for (i = 0; i < group_info->nblocks; i++) {
1223                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1224                 int off = i * NGROUPS_PER_BLOCK;
1225                 int len = cp_count * sizeof(*grouplist);
1226
1227                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1228                         return -EFAULT;
1229
1230                 count -= cp_count;
1231         }
1232         return 0;
1233 }
1234
1235 /* a simple Shell sort */
1236 static void groups_sort(struct group_info *group_info)
1237 {
1238         int base, max, stride;
1239         int gidsetsize = group_info->ngroups;
1240
1241         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1242                 ; /* nothing */
1243         stride /= 3;
1244
1245         while (stride) {
1246                 max = gidsetsize - stride;
1247                 for (base = 0; base < max; base++) {
1248                         int left = base;
1249                         int right = left + stride;
1250                         gid_t tmp = GROUP_AT(group_info, right);
1251
1252                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1253                                 GROUP_AT(group_info, right) =
1254                                     GROUP_AT(group_info, left);
1255                                 right = left;
1256                                 left -= stride;
1257                         }
1258                         GROUP_AT(group_info, right) = tmp;
1259                 }
1260                 stride /= 3;
1261         }
1262 }
1263
1264 /* a simple bsearch */
1265 int groups_search(struct group_info *group_info, gid_t grp)
1266 {
1267         int left, right;
1268
1269         if (!group_info)
1270                 return 0;
1271
1272         left = 0;
1273         right = group_info->ngroups;
1274         while (left < right) {
1275                 int mid = (left+right)/2;
1276                 int cmp = grp - GROUP_AT(group_info, mid);
1277                 if (cmp > 0)
1278                         left = mid + 1;
1279                 else if (cmp < 0)
1280                         right = mid;
1281                 else
1282                         return 1;
1283         }
1284         return 0;
1285 }
1286
1287 /* validate and set current->group_info */
1288 int set_current_groups(struct group_info *group_info)
1289 {
1290         int retval;
1291         struct group_info *old_info;
1292
1293         retval = security_task_setgroups(group_info);
1294         if (retval)
1295                 return retval;
1296
1297         groups_sort(group_info);
1298         get_group_info(group_info);
1299
1300         task_lock(current);
1301         old_info = current->group_info;
1302         current->group_info = group_info;
1303         task_unlock(current);
1304
1305         put_group_info(old_info);
1306
1307         return 0;
1308 }
1309
1310 EXPORT_SYMBOL(set_current_groups);
1311
1312 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1313 {
1314         int i = 0;
1315
1316         /*
1317          *      SMP: Nobody else can change our grouplist. Thus we are
1318          *      safe.
1319          */
1320
1321         if (gidsetsize < 0)
1322                 return -EINVAL;
1323
1324         /* no need to grab task_lock here; it cannot change */
1325         get_group_info(current->group_info);
1326         i = current->group_info->ngroups;
1327         if (gidsetsize) {
1328                 if (i > gidsetsize) {
1329                         i = -EINVAL;
1330                         goto out;
1331                 }
1332                 if (groups_to_user(grouplist, current->group_info)) {
1333                         i = -EFAULT;
1334                         goto out;
1335                 }
1336         }
1337 out:
1338         put_group_info(current->group_info);
1339         return i;
1340 }
1341
1342 /*
1343  *      SMP: Our groups are copy-on-write. We can set them safely
1344  *      without another task interfering.
1345  */
1346  
1347 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1348 {
1349         struct group_info *group_info;
1350         int retval;
1351
1352         if (!capable(CAP_SETGID))
1353                 return -EPERM;
1354         if ((unsigned)gidsetsize > NGROUPS_MAX)
1355                 return -EINVAL;
1356
1357         group_info = groups_alloc(gidsetsize);
1358         if (!group_info)
1359                 return -ENOMEM;
1360         retval = groups_from_user(group_info, grouplist);
1361         if (retval) {
1362                 put_group_info(group_info);
1363                 return retval;
1364         }
1365
1366         retval = set_current_groups(group_info);
1367         put_group_info(group_info);
1368
1369         return retval;
1370 }
1371
1372 /*
1373  * Check whether we're fsgid/egid or in the supplemental group..
1374  */
1375 int in_group_p(gid_t grp)
1376 {
1377         int retval = 1;
1378         if (grp != current->fsgid) {
1379                 get_group_info(current->group_info);
1380                 retval = groups_search(current->group_info, grp);
1381                 put_group_info(current->group_info);
1382         }
1383         return retval;
1384 }
1385
1386 EXPORT_SYMBOL(in_group_p);
1387
1388 int in_egroup_p(gid_t grp)
1389 {
1390         int retval = 1;
1391         if (grp != current->egid) {
1392                 get_group_info(current->group_info);
1393                 retval = groups_search(current->group_info, grp);
1394                 put_group_info(current->group_info);
1395         }
1396         return retval;
1397 }
1398
1399 EXPORT_SYMBOL(in_egroup_p);
1400
1401 DECLARE_RWSEM(uts_sem);
1402
1403 EXPORT_SYMBOL(uts_sem);
1404
1405 asmlinkage long sys_newuname(struct new_utsname __user * name)
1406 {
1407         int errno = 0;
1408
1409         down_read(&uts_sem);
1410         if (copy_to_user(name,&system_utsname,sizeof *name))
1411                 errno = -EFAULT;
1412         up_read(&uts_sem);
1413         return errno;
1414 }
1415
1416 asmlinkage long sys_sethostname(char __user *name, int len)
1417 {
1418         int errno;
1419         char tmp[__NEW_UTS_LEN];
1420
1421         if (!capable(CAP_SYS_ADMIN))
1422                 return -EPERM;
1423         if (len < 0 || len > __NEW_UTS_LEN)
1424                 return -EINVAL;
1425         down_write(&uts_sem);
1426         errno = -EFAULT;
1427         if (!copy_from_user(tmp, name, len)) {
1428                 memcpy(system_utsname.nodename, tmp, len);
1429                 system_utsname.nodename[len] = 0;
1430                 errno = 0;
1431         }
1432         up_write(&uts_sem);
1433         return errno;
1434 }
1435
1436 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1437
1438 asmlinkage long sys_gethostname(char __user *name, int len)
1439 {
1440         int i, errno;
1441
1442         if (len < 0)
1443                 return -EINVAL;
1444         down_read(&uts_sem);
1445         i = 1 + strlen(system_utsname.nodename);
1446         if (i > len)
1447                 i = len;
1448         errno = 0;
1449         if (copy_to_user(name, system_utsname.nodename, i))
1450                 errno = -EFAULT;
1451         up_read(&uts_sem);
1452         return errno;
1453 }
1454
1455 #endif
1456
1457 /*
1458  * Only setdomainname; getdomainname can be implemented by calling
1459  * uname()
1460  */
1461 asmlinkage long sys_setdomainname(char __user *name, int len)
1462 {
1463         int errno;
1464         char tmp[__NEW_UTS_LEN];
1465
1466         if (!capable(CAP_SYS_ADMIN))
1467                 return -EPERM;
1468         if (len < 0 || len > __NEW_UTS_LEN)
1469                 return -EINVAL;
1470
1471         down_write(&uts_sem);
1472         errno = -EFAULT;
1473         if (!copy_from_user(tmp, name, len)) {
1474                 memcpy(system_utsname.domainname, tmp, len);
1475                 system_utsname.domainname[len] = 0;
1476                 errno = 0;
1477         }
1478         up_write(&uts_sem);
1479         return errno;
1480 }
1481
1482 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1483 {
1484         if (resource >= RLIM_NLIMITS)
1485                 return -EINVAL;
1486         else {
1487                 struct rlimit value;
1488                 task_lock(current->group_leader);
1489                 value = current->signal->rlim[resource];
1490                 task_unlock(current->group_leader);
1491                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1492         }
1493 }
1494
1495 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1496
1497 /*
1498  *      Back compatibility for getrlimit. Needed for some apps.
1499  */
1500  
1501 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1502 {
1503         struct rlimit x;
1504         if (resource >= RLIM_NLIMITS)
1505                 return -EINVAL;
1506
1507         task_lock(current->group_leader);
1508         x = current->signal->rlim[resource];
1509         task_unlock(current->group_leader);
1510         if(x.rlim_cur > 0x7FFFFFFF)
1511                 x.rlim_cur = 0x7FFFFFFF;
1512         if(x.rlim_max > 0x7FFFFFFF)
1513                 x.rlim_max = 0x7FFFFFFF;
1514         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1515 }
1516
1517 #endif
1518
1519 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1520 {
1521         struct rlimit new_rlim, *old_rlim;
1522         int retval;
1523
1524         if (resource >= RLIM_NLIMITS)
1525                 return -EINVAL;
1526         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1527                 return -EFAULT;
1528        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1529                return -EINVAL;
1530         old_rlim = current->signal->rlim + resource;
1531         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1532             !capable(CAP_SYS_RESOURCE))
1533                 return -EPERM;
1534         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1535                         return -EPERM;
1536
1537         retval = security_task_setrlimit(resource, &new_rlim);
1538         if (retval)
1539                 return retval;
1540
1541         task_lock(current->group_leader);
1542         *old_rlim = new_rlim;
1543         task_unlock(current->group_leader);
1544
1545         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1546             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1547              new_rlim.rlim_cur <= cputime_to_secs(
1548                      current->signal->it_prof_expires))) {
1549                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1550                 read_lock(&tasklist_lock);
1551                 spin_lock_irq(&current->sighand->siglock);
1552                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1553                                       &cputime, NULL);
1554                 spin_unlock_irq(&current->sighand->siglock);
1555                 read_unlock(&tasklist_lock);
1556         }
1557
1558         return 0;
1559 }
1560
1561 /*
1562  * It would make sense to put struct rusage in the task_struct,
1563  * except that would make the task_struct be *really big*.  After
1564  * task_struct gets moved into malloc'ed memory, it would
1565  * make sense to do this.  It will make moving the rest of the information
1566  * a lot simpler!  (Which we're not doing right now because we're not
1567  * measuring them yet).
1568  *
1569  * This expects to be called with tasklist_lock read-locked or better,
1570  * and the siglock not locked.  It may momentarily take the siglock.
1571  *
1572  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1573  * races with threads incrementing their own counters.  But since word
1574  * reads are atomic, we either get new values or old values and we don't
1575  * care which for the sums.  We always take the siglock to protect reading
1576  * the c* fields from p->signal from races with exit.c updating those
1577  * fields when reaping, so a sample either gets all the additions of a
1578  * given child after it's reaped, or none so this sample is before reaping.
1579  */
1580
1581 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1582 {
1583         struct task_struct *t;
1584         unsigned long flags;
1585         cputime_t utime, stime;
1586
1587         memset((char *) r, 0, sizeof *r);
1588
1589         if (unlikely(!p->signal))
1590                 return;
1591
1592         switch (who) {
1593                 case RUSAGE_CHILDREN:
1594                         spin_lock_irqsave(&p->sighand->siglock, flags);
1595                         utime = p->signal->cutime;
1596                         stime = p->signal->cstime;
1597                         r->ru_nvcsw = p->signal->cnvcsw;
1598                         r->ru_nivcsw = p->signal->cnivcsw;
1599                         r->ru_minflt = p->signal->cmin_flt;
1600                         r->ru_majflt = p->signal->cmaj_flt;
1601                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1602                         cputime_to_timeval(utime, &r->ru_utime);
1603                         cputime_to_timeval(stime, &r->ru_stime);
1604                         break;
1605                 case RUSAGE_SELF:
1606                         spin_lock_irqsave(&p->sighand->siglock, flags);
1607                         utime = stime = cputime_zero;
1608                         goto sum_group;
1609                 case RUSAGE_BOTH:
1610                         spin_lock_irqsave(&p->sighand->siglock, flags);
1611                         utime = p->signal->cutime;
1612                         stime = p->signal->cstime;
1613                         r->ru_nvcsw = p->signal->cnvcsw;
1614                         r->ru_nivcsw = p->signal->cnivcsw;
1615                         r->ru_minflt = p->signal->cmin_flt;
1616                         r->ru_majflt = p->signal->cmaj_flt;
1617                 sum_group:
1618                         utime = cputime_add(utime, p->signal->utime);
1619                         stime = cputime_add(stime, p->signal->stime);
1620                         r->ru_nvcsw += p->signal->nvcsw;
1621                         r->ru_nivcsw += p->signal->nivcsw;
1622                         r->ru_minflt += p->signal->min_flt;
1623                         r->ru_majflt += p->signal->maj_flt;
1624                         t = p;
1625                         do {
1626                                 utime = cputime_add(utime, t->utime);
1627                                 stime = cputime_add(stime, t->stime);
1628                                 r->ru_nvcsw += t->nvcsw;
1629                                 r->ru_nivcsw += t->nivcsw;
1630                                 r->ru_minflt += t->min_flt;
1631                                 r->ru_majflt += t->maj_flt;
1632                                 t = next_thread(t);
1633                         } while (t != p);
1634                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1635                         cputime_to_timeval(utime, &r->ru_utime);
1636                         cputime_to_timeval(stime, &r->ru_stime);
1637                         break;
1638                 default:
1639                         BUG();
1640         }
1641 }
1642
1643 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1644 {
1645         struct rusage r;
1646         read_lock(&tasklist_lock);
1647         k_getrusage(p, who, &r);
1648         read_unlock(&tasklist_lock);
1649         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1650 }
1651
1652 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1653 {
1654         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1655                 return -EINVAL;
1656         return getrusage(current, who, ru);
1657 }
1658
1659 asmlinkage long sys_umask(int mask)
1660 {
1661         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1662         return mask;
1663 }
1664     
1665 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1666                           unsigned long arg4, unsigned long arg5)
1667 {
1668         long error;
1669         int sig;
1670
1671         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1672         if (error)
1673                 return error;
1674
1675         switch (option) {
1676                 case PR_SET_PDEATHSIG:
1677                         sig = arg2;
1678                         if (!valid_signal(sig)) {
1679                                 error = -EINVAL;
1680                                 break;
1681                         }
1682                         current->pdeath_signal = sig;
1683                         break;
1684                 case PR_GET_PDEATHSIG:
1685                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1686                         break;
1687                 case PR_GET_DUMPABLE:
1688                         if (current->mm->dumpable)
1689                                 error = 1;
1690                         break;
1691                 case PR_SET_DUMPABLE:
1692                         if (arg2 < 0 || arg2 > 2) {
1693                                 error = -EINVAL;
1694                                 break;
1695                         }
1696                         current->mm->dumpable = arg2;
1697                         break;
1698
1699                 case PR_SET_UNALIGN:
1700                         error = SET_UNALIGN_CTL(current, arg2);
1701                         break;
1702                 case PR_GET_UNALIGN:
1703                         error = GET_UNALIGN_CTL(current, arg2);
1704                         break;
1705                 case PR_SET_FPEMU:
1706                         error = SET_FPEMU_CTL(current, arg2);
1707                         break;
1708                 case PR_GET_FPEMU:
1709                         error = GET_FPEMU_CTL(current, arg2);
1710                         break;
1711                 case PR_SET_FPEXC:
1712                         error = SET_FPEXC_CTL(current, arg2);
1713                         break;
1714                 case PR_GET_FPEXC:
1715                         error = GET_FPEXC_CTL(current, arg2);
1716                         break;
1717                 case PR_GET_TIMING:
1718                         error = PR_TIMING_STATISTICAL;
1719                         break;
1720                 case PR_SET_TIMING:
1721                         if (arg2 == PR_TIMING_STATISTICAL)
1722                                 error = 0;
1723                         else
1724                                 error = -EINVAL;
1725                         break;
1726
1727                 case PR_GET_KEEPCAPS:
1728                         if (current->keep_capabilities)
1729                                 error = 1;
1730                         break;
1731                 case PR_SET_KEEPCAPS:
1732                         if (arg2 != 0 && arg2 != 1) {
1733                                 error = -EINVAL;
1734                                 break;
1735                         }
1736                         current->keep_capabilities = arg2;
1737                         break;
1738                 case PR_SET_NAME: {
1739                         struct task_struct *me = current;
1740                         unsigned char ncomm[sizeof(me->comm)];
1741
1742                         ncomm[sizeof(me->comm)-1] = 0;
1743                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1744                                                 sizeof(me->comm)-1) < 0)
1745                                 return -EFAULT;
1746                         set_task_comm(me, ncomm);
1747                         return 0;
1748                 }
1749                 case PR_GET_NAME: {
1750                         struct task_struct *me = current;
1751                         unsigned char tcomm[sizeof(me->comm)];
1752
1753                         get_task_comm(tcomm, me);
1754                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1755                                 return -EFAULT;
1756                         return 0;
1757                 }
1758                 default:
1759                         error = -EINVAL;
1760                         break;
1761         }
1762         return error;
1763 }