[PATCH] kexec: add kexec syscalls
[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/kernel.h>
20 #include <linux/kexec.h>
21 #include <linux/workqueue.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
32 #include <linux/compat.h>
33 #include <linux/syscalls.h>
34
35 #include <asm/uaccess.h>
36 #include <asm/io.h>
37 #include <asm/unistd.h>
38
39 #ifndef SET_UNALIGN_CTL
40 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
41 #endif
42 #ifndef GET_UNALIGN_CTL
43 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
44 #endif
45 #ifndef SET_FPEMU_CTL
46 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
47 #endif
48 #ifndef GET_FPEMU_CTL
49 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
50 #endif
51 #ifndef SET_FPEXC_CTL
52 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
53 #endif
54 #ifndef GET_FPEXC_CTL
55 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
56 #endif
57
58 /*
59  * this is where the system-wide overflow UID and GID are defined, for
60  * architectures that now have 32-bit UID/GID but didn't in the past
61  */
62
63 int overflowuid = DEFAULT_OVERFLOWUID;
64 int overflowgid = DEFAULT_OVERFLOWGID;
65
66 #ifdef CONFIG_UID16
67 EXPORT_SYMBOL(overflowuid);
68 EXPORT_SYMBOL(overflowgid);
69 #endif
70
71 /*
72  * the same as above, but for filesystems which can only store a 16-bit
73  * UID and GID. as such, this is needed on all architectures
74  */
75
76 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
77 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
78
79 EXPORT_SYMBOL(fs_overflowuid);
80 EXPORT_SYMBOL(fs_overflowgid);
81
82 /*
83  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
84  */
85
86 int C_A_D = 1;
87 int cad_pid = 1;
88
89 /*
90  *      Notifier list for kernel code which wants to be called
91  *      at shutdown. This is used to stop any idling DMA operations
92  *      and the like. 
93  */
94
95 static struct notifier_block *reboot_notifier_list;
96 static DEFINE_RWLOCK(notifier_lock);
97
98 /**
99  *      notifier_chain_register - Add notifier to a notifier chain
100  *      @list: Pointer to root list pointer
101  *      @n: New entry in notifier chain
102  *
103  *      Adds a notifier to a notifier chain.
104  *
105  *      Currently always returns zero.
106  */
107  
108 int notifier_chain_register(struct notifier_block **list, struct notifier_block *n)
109 {
110         write_lock(&notifier_lock);
111         while(*list)
112         {
113                 if(n->priority > (*list)->priority)
114                         break;
115                 list= &((*list)->next);
116         }
117         n->next = *list;
118         *list=n;
119         write_unlock(&notifier_lock);
120         return 0;
121 }
122
123 EXPORT_SYMBOL(notifier_chain_register);
124
125 /**
126  *      notifier_chain_unregister - Remove notifier from a notifier chain
127  *      @nl: Pointer to root list pointer
128  *      @n: New entry in notifier chain
129  *
130  *      Removes a notifier from a notifier chain.
131  *
132  *      Returns zero on success, or %-ENOENT on failure.
133  */
134  
135 int notifier_chain_unregister(struct notifier_block **nl, struct notifier_block *n)
136 {
137         write_lock(&notifier_lock);
138         while((*nl)!=NULL)
139         {
140                 if((*nl)==n)
141                 {
142                         *nl=n->next;
143                         write_unlock(&notifier_lock);
144                         return 0;
145                 }
146                 nl=&((*nl)->next);
147         }
148         write_unlock(&notifier_lock);
149         return -ENOENT;
150 }
151
152 EXPORT_SYMBOL(notifier_chain_unregister);
153
154 /**
155  *      notifier_call_chain - Call functions in a notifier chain
156  *      @n: Pointer to root pointer of notifier chain
157  *      @val: Value passed unmodified to notifier function
158  *      @v: Pointer passed unmodified to notifier function
159  *
160  *      Calls each function in a notifier chain in turn.
161  *
162  *      If the return value of the notifier can be and'd
163  *      with %NOTIFY_STOP_MASK, then notifier_call_chain
164  *      will return immediately, with the return value of
165  *      the notifier function which halted execution.
166  *      Otherwise, the return value is the return value
167  *      of the last notifier function called.
168  */
169  
170 int notifier_call_chain(struct notifier_block **n, unsigned long val, void *v)
171 {
172         int ret=NOTIFY_DONE;
173         struct notifier_block *nb = *n;
174
175         while(nb)
176         {
177                 ret=nb->notifier_call(nb,val,v);
178                 if(ret&NOTIFY_STOP_MASK)
179                 {
180                         return ret;
181                 }
182                 nb=nb->next;
183         }
184         return ret;
185 }
186
187 EXPORT_SYMBOL(notifier_call_chain);
188
189 /**
190  *      register_reboot_notifier - Register function to be called at reboot time
191  *      @nb: Info about notifier function to be called
192  *
193  *      Registers a function with the list of functions
194  *      to be called at reboot time.
195  *
196  *      Currently always returns zero, as notifier_chain_register
197  *      always returns zero.
198  */
199  
200 int register_reboot_notifier(struct notifier_block * nb)
201 {
202         return notifier_chain_register(&reboot_notifier_list, nb);
203 }
204
205 EXPORT_SYMBOL(register_reboot_notifier);
206
207 /**
208  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
209  *      @nb: Hook to be unregistered
210  *
211  *      Unregisters a previously registered reboot
212  *      notifier function.
213  *
214  *      Returns zero on success, or %-ENOENT on failure.
215  */
216  
217 int unregister_reboot_notifier(struct notifier_block * nb)
218 {
219         return notifier_chain_unregister(&reboot_notifier_list, nb);
220 }
221
222 EXPORT_SYMBOL(unregister_reboot_notifier);
223
224 static int set_one_prio(struct task_struct *p, int niceval, int error)
225 {
226         int no_nice;
227
228         if (p->uid != current->euid &&
229                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
230                 error = -EPERM;
231                 goto out;
232         }
233         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
234                 error = -EACCES;
235                 goto out;
236         }
237         no_nice = security_task_setnice(p, niceval);
238         if (no_nice) {
239                 error = no_nice;
240                 goto out;
241         }
242         if (error == -ESRCH)
243                 error = 0;
244         set_user_nice(p, niceval);
245 out:
246         return error;
247 }
248
249 asmlinkage long sys_setpriority(int which, int who, int niceval)
250 {
251         struct task_struct *g, *p;
252         struct user_struct *user;
253         int error = -EINVAL;
254
255         if (which > 2 || which < 0)
256                 goto out;
257
258         /* normalize: avoid signed division (rounding problems) */
259         error = -ESRCH;
260         if (niceval < -20)
261                 niceval = -20;
262         if (niceval > 19)
263                 niceval = 19;
264
265         read_lock(&tasklist_lock);
266         switch (which) {
267                 case PRIO_PROCESS:
268                         if (!who)
269                                 who = current->pid;
270                         p = find_task_by_pid(who);
271                         if (p)
272                                 error = set_one_prio(p, niceval, error);
273                         break;
274                 case PRIO_PGRP:
275                         if (!who)
276                                 who = process_group(current);
277                         do_each_task_pid(who, PIDTYPE_PGID, p) {
278                                 error = set_one_prio(p, niceval, error);
279                         } while_each_task_pid(who, PIDTYPE_PGID, p);
280                         break;
281                 case PRIO_USER:
282                         user = current->user;
283                         if (!who)
284                                 who = current->uid;
285                         else
286                                 if ((who != current->uid) && !(user = find_user(who)))
287                                         goto out_unlock;        /* No processes for this user */
288
289                         do_each_thread(g, p)
290                                 if (p->uid == who)
291                                         error = set_one_prio(p, niceval, error);
292                         while_each_thread(g, p);
293                         if (who != current->uid)
294                                 free_uid(user);         /* For find_user() */
295                         break;
296         }
297 out_unlock:
298         read_unlock(&tasklist_lock);
299 out:
300         return error;
301 }
302
303 /*
304  * Ugh. To avoid negative return values, "getpriority()" will
305  * not return the normal nice-value, but a negated value that
306  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
307  * to stay compatible.
308  */
309 asmlinkage long sys_getpriority(int which, int who)
310 {
311         struct task_struct *g, *p;
312         struct user_struct *user;
313         long niceval, retval = -ESRCH;
314
315         if (which > 2 || which < 0)
316                 return -EINVAL;
317
318         read_lock(&tasklist_lock);
319         switch (which) {
320                 case PRIO_PROCESS:
321                         if (!who)
322                                 who = current->pid;
323                         p = find_task_by_pid(who);
324                         if (p) {
325                                 niceval = 20 - task_nice(p);
326                                 if (niceval > retval)
327                                         retval = niceval;
328                         }
329                         break;
330                 case PRIO_PGRP:
331                         if (!who)
332                                 who = process_group(current);
333                         do_each_task_pid(who, PIDTYPE_PGID, p) {
334                                 niceval = 20 - task_nice(p);
335                                 if (niceval > retval)
336                                         retval = niceval;
337                         } while_each_task_pid(who, PIDTYPE_PGID, p);
338                         break;
339                 case PRIO_USER:
340                         user = current->user;
341                         if (!who)
342                                 who = current->uid;
343                         else
344                                 if ((who != current->uid) && !(user = find_user(who)))
345                                         goto out_unlock;        /* No processes for this user */
346
347                         do_each_thread(g, p)
348                                 if (p->uid == who) {
349                                         niceval = 20 - task_nice(p);
350                                         if (niceval > retval)
351                                                 retval = niceval;
352                                 }
353                         while_each_thread(g, p);
354                         if (who != current->uid)
355                                 free_uid(user);         /* for find_user() */
356                         break;
357         }
358 out_unlock:
359         read_unlock(&tasklist_lock);
360
361         return retval;
362 }
363
364
365 /*
366  * Reboot system call: for obvious reasons only root may call it,
367  * and even root needs to set up some magic numbers in the registers
368  * so that some mistake won't make this reboot the whole machine.
369  * You can also set the meaning of the ctrl-alt-del-key here.
370  *
371  * reboot doesn't sync: do that yourself before calling this.
372  */
373 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
374 {
375         char buffer[256];
376
377         /* We only trust the superuser with rebooting the system. */
378         if (!capable(CAP_SYS_BOOT))
379                 return -EPERM;
380
381         /* For safety, we require "magic" arguments. */
382         if (magic1 != LINUX_REBOOT_MAGIC1 ||
383             (magic2 != LINUX_REBOOT_MAGIC2 &&
384                         magic2 != LINUX_REBOOT_MAGIC2A &&
385                         magic2 != LINUX_REBOOT_MAGIC2B &&
386                         magic2 != LINUX_REBOOT_MAGIC2C))
387                 return -EINVAL;
388
389         lock_kernel();
390         switch (cmd) {
391         case LINUX_REBOOT_CMD_RESTART:
392                 notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
393                 system_state = SYSTEM_RESTART;
394                 device_shutdown();
395                 printk(KERN_EMERG "Restarting system.\n");
396                 machine_restart(NULL);
397                 break;
398
399         case LINUX_REBOOT_CMD_CAD_ON:
400                 C_A_D = 1;
401                 break;
402
403         case LINUX_REBOOT_CMD_CAD_OFF:
404                 C_A_D = 0;
405                 break;
406
407         case LINUX_REBOOT_CMD_HALT:
408                 notifier_call_chain(&reboot_notifier_list, SYS_HALT, NULL);
409                 system_state = SYSTEM_HALT;
410                 device_suspend(PMSG_SUSPEND);
411                 device_shutdown();
412                 printk(KERN_EMERG "System halted.\n");
413                 machine_halt();
414                 unlock_kernel();
415                 do_exit(0);
416                 break;
417
418         case LINUX_REBOOT_CMD_POWER_OFF:
419                 notifier_call_chain(&reboot_notifier_list, SYS_POWER_OFF, NULL);
420                 system_state = SYSTEM_POWER_OFF;
421                 device_suspend(PMSG_SUSPEND);
422                 device_shutdown();
423                 printk(KERN_EMERG "Power down.\n");
424                 machine_power_off();
425                 unlock_kernel();
426                 do_exit(0);
427                 break;
428
429         case LINUX_REBOOT_CMD_RESTART2:
430                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
431                         unlock_kernel();
432                         return -EFAULT;
433                 }
434                 buffer[sizeof(buffer) - 1] = '\0';
435
436                 notifier_call_chain(&reboot_notifier_list, SYS_RESTART, buffer);
437                 system_state = SYSTEM_RESTART;
438                 device_suspend(PMSG_FREEZE);
439                 device_shutdown();
440                 printk(KERN_EMERG "Restarting system with command '%s'.\n", buffer);
441                 machine_restart(buffer);
442                 break;
443
444 #ifdef CONFIG_KEXEC
445         case LINUX_REBOOT_CMD_KEXEC:
446         {
447                 struct kimage *image;
448                 image = xchg(&kexec_image, 0);
449                 if (!image) {
450                         unlock_kernel();
451                         return -EINVAL;
452                 }
453                 notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
454                 system_state = SYSTEM_RESTART;
455                 device_shutdown();
456                 printk(KERN_EMERG "Starting new kernel\n");
457                 machine_shutdown();
458                 machine_kexec(image);
459                 break;
460         }
461 #endif
462 #ifdef CONFIG_SOFTWARE_SUSPEND
463         case LINUX_REBOOT_CMD_SW_SUSPEND:
464                 {
465                         int ret = software_suspend();
466                         unlock_kernel();
467                         return ret;
468                 }
469 #endif
470
471         default:
472                 unlock_kernel();
473                 return -EINVAL;
474         }
475         unlock_kernel();
476         return 0;
477 }
478
479 static void deferred_cad(void *dummy)
480 {
481         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
482         machine_restart(NULL);
483 }
484
485 /*
486  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
487  * As it's called within an interrupt, it may NOT sync: the only choice
488  * is whether to reboot at once, or just ignore the ctrl-alt-del.
489  */
490 void ctrl_alt_del(void)
491 {
492         static DECLARE_WORK(cad_work, deferred_cad, NULL);
493
494         if (C_A_D)
495                 schedule_work(&cad_work);
496         else
497                 kill_proc(cad_pid, SIGINT, 1);
498 }
499         
500
501 /*
502  * Unprivileged users may change the real gid to the effective gid
503  * or vice versa.  (BSD-style)
504  *
505  * If you set the real gid at all, or set the effective gid to a value not
506  * equal to the real gid, then the saved gid is set to the new effective gid.
507  *
508  * This makes it possible for a setgid program to completely drop its
509  * privileges, which is often a useful assertion to make when you are doing
510  * a security audit over a program.
511  *
512  * The general idea is that a program which uses just setregid() will be
513  * 100% compatible with BSD.  A program which uses just setgid() will be
514  * 100% compatible with POSIX with saved IDs. 
515  *
516  * SMP: There are not races, the GIDs are checked only by filesystem
517  *      operations (as far as semantic preservation is concerned).
518  */
519 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
520 {
521         int old_rgid = current->gid;
522         int old_egid = current->egid;
523         int new_rgid = old_rgid;
524         int new_egid = old_egid;
525         int retval;
526
527         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
528         if (retval)
529                 return retval;
530
531         if (rgid != (gid_t) -1) {
532                 if ((old_rgid == rgid) ||
533                     (current->egid==rgid) ||
534                     capable(CAP_SETGID))
535                         new_rgid = rgid;
536                 else
537                         return -EPERM;
538         }
539         if (egid != (gid_t) -1) {
540                 if ((old_rgid == egid) ||
541                     (current->egid == egid) ||
542                     (current->sgid == egid) ||
543                     capable(CAP_SETGID))
544                         new_egid = egid;
545                 else {
546                         return -EPERM;
547                 }
548         }
549         if (new_egid != old_egid)
550         {
551                 current->mm->dumpable = suid_dumpable;
552                 smp_wmb();
553         }
554         if (rgid != (gid_t) -1 ||
555             (egid != (gid_t) -1 && egid != old_rgid))
556                 current->sgid = new_egid;
557         current->fsgid = new_egid;
558         current->egid = new_egid;
559         current->gid = new_rgid;
560         key_fsgid_changed(current);
561         return 0;
562 }
563
564 /*
565  * setgid() is implemented like SysV w/ SAVED_IDS 
566  *
567  * SMP: Same implicit races as above.
568  */
569 asmlinkage long sys_setgid(gid_t gid)
570 {
571         int old_egid = current->egid;
572         int retval;
573
574         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
575         if (retval)
576                 return retval;
577
578         if (capable(CAP_SETGID))
579         {
580                 if(old_egid != gid)
581                 {
582                         current->mm->dumpable = suid_dumpable;
583                         smp_wmb();
584                 }
585                 current->gid = current->egid = current->sgid = current->fsgid = gid;
586         }
587         else if ((gid == current->gid) || (gid == current->sgid))
588         {
589                 if(old_egid != gid)
590                 {
591                         current->mm->dumpable = suid_dumpable;
592                         smp_wmb();
593                 }
594                 current->egid = current->fsgid = gid;
595         }
596         else
597                 return -EPERM;
598
599         key_fsgid_changed(current);
600         return 0;
601 }
602   
603 static int set_user(uid_t new_ruid, int dumpclear)
604 {
605         struct user_struct *new_user;
606
607         new_user = alloc_uid(new_ruid);
608         if (!new_user)
609                 return -EAGAIN;
610
611         if (atomic_read(&new_user->processes) >=
612                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
613                         new_user != &root_user) {
614                 free_uid(new_user);
615                 return -EAGAIN;
616         }
617
618         switch_uid(new_user);
619
620         if(dumpclear)
621         {
622                 current->mm->dumpable = suid_dumpable;
623                 smp_wmb();
624         }
625         current->uid = new_ruid;
626         return 0;
627 }
628
629 /*
630  * Unprivileged users may change the real uid to the effective uid
631  * or vice versa.  (BSD-style)
632  *
633  * If you set the real uid at all, or set the effective uid to a value not
634  * equal to the real uid, then the saved uid is set to the new effective uid.
635  *
636  * This makes it possible for a setuid program to completely drop its
637  * privileges, which is often a useful assertion to make when you are doing
638  * a security audit over a program.
639  *
640  * The general idea is that a program which uses just setreuid() will be
641  * 100% compatible with BSD.  A program which uses just setuid() will be
642  * 100% compatible with POSIX with saved IDs. 
643  */
644 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
645 {
646         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
647         int retval;
648
649         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
650         if (retval)
651                 return retval;
652
653         new_ruid = old_ruid = current->uid;
654         new_euid = old_euid = current->euid;
655         old_suid = current->suid;
656
657         if (ruid != (uid_t) -1) {
658                 new_ruid = ruid;
659                 if ((old_ruid != ruid) &&
660                     (current->euid != ruid) &&
661                     !capable(CAP_SETUID))
662                         return -EPERM;
663         }
664
665         if (euid != (uid_t) -1) {
666                 new_euid = euid;
667                 if ((old_ruid != euid) &&
668                     (current->euid != euid) &&
669                     (current->suid != euid) &&
670                     !capable(CAP_SETUID))
671                         return -EPERM;
672         }
673
674         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
675                 return -EAGAIN;
676
677         if (new_euid != old_euid)
678         {
679                 current->mm->dumpable = suid_dumpable;
680                 smp_wmb();
681         }
682         current->fsuid = current->euid = new_euid;
683         if (ruid != (uid_t) -1 ||
684             (euid != (uid_t) -1 && euid != old_ruid))
685                 current->suid = current->euid;
686         current->fsuid = current->euid;
687
688         key_fsuid_changed(current);
689
690         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
691 }
692
693
694                 
695 /*
696  * setuid() is implemented like SysV with SAVED_IDS 
697  * 
698  * Note that SAVED_ID's is deficient in that a setuid root program
699  * like sendmail, for example, cannot set its uid to be a normal 
700  * user and then switch back, because if you're root, setuid() sets
701  * the saved uid too.  If you don't like this, blame the bright people
702  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
703  * will allow a root program to temporarily drop privileges and be able to
704  * regain them by swapping the real and effective uid.  
705  */
706 asmlinkage long sys_setuid(uid_t uid)
707 {
708         int old_euid = current->euid;
709         int old_ruid, old_suid, new_ruid, new_suid;
710         int retval;
711
712         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
713         if (retval)
714                 return retval;
715
716         old_ruid = new_ruid = current->uid;
717         old_suid = current->suid;
718         new_suid = old_suid;
719         
720         if (capable(CAP_SETUID)) {
721                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
722                         return -EAGAIN;
723                 new_suid = uid;
724         } else if ((uid != current->uid) && (uid != new_suid))
725                 return -EPERM;
726
727         if (old_euid != uid)
728         {
729                 current->mm->dumpable = suid_dumpable;
730                 smp_wmb();
731         }
732         current->fsuid = current->euid = uid;
733         current->suid = new_suid;
734
735         key_fsuid_changed(current);
736
737         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
738 }
739
740
741 /*
742  * This function implements a generic ability to update ruid, euid,
743  * and suid.  This allows you to implement the 4.4 compatible seteuid().
744  */
745 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
746 {
747         int old_ruid = current->uid;
748         int old_euid = current->euid;
749         int old_suid = current->suid;
750         int retval;
751
752         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
753         if (retval)
754                 return retval;
755
756         if (!capable(CAP_SETUID)) {
757                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
758                     (ruid != current->euid) && (ruid != current->suid))
759                         return -EPERM;
760                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
761                     (euid != current->euid) && (euid != current->suid))
762                         return -EPERM;
763                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
764                     (suid != current->euid) && (suid != current->suid))
765                         return -EPERM;
766         }
767         if (ruid != (uid_t) -1) {
768                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
769                         return -EAGAIN;
770         }
771         if (euid != (uid_t) -1) {
772                 if (euid != current->euid)
773                 {
774                         current->mm->dumpable = suid_dumpable;
775                         smp_wmb();
776                 }
777                 current->euid = euid;
778         }
779         current->fsuid = current->euid;
780         if (suid != (uid_t) -1)
781                 current->suid = suid;
782
783         key_fsuid_changed(current);
784
785         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
786 }
787
788 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
789 {
790         int retval;
791
792         if (!(retval = put_user(current->uid, ruid)) &&
793             !(retval = put_user(current->euid, euid)))
794                 retval = put_user(current->suid, suid);
795
796         return retval;
797 }
798
799 /*
800  * Same as above, but for rgid, egid, sgid.
801  */
802 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
803 {
804         int retval;
805
806         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
807         if (retval)
808                 return retval;
809
810         if (!capable(CAP_SETGID)) {
811                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
812                     (rgid != current->egid) && (rgid != current->sgid))
813                         return -EPERM;
814                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
815                     (egid != current->egid) && (egid != current->sgid))
816                         return -EPERM;
817                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
818                     (sgid != current->egid) && (sgid != current->sgid))
819                         return -EPERM;
820         }
821         if (egid != (gid_t) -1) {
822                 if (egid != current->egid)
823                 {
824                         current->mm->dumpable = suid_dumpable;
825                         smp_wmb();
826                 }
827                 current->egid = egid;
828         }
829         current->fsgid = current->egid;
830         if (rgid != (gid_t) -1)
831                 current->gid = rgid;
832         if (sgid != (gid_t) -1)
833                 current->sgid = sgid;
834
835         key_fsgid_changed(current);
836         return 0;
837 }
838
839 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
840 {
841         int retval;
842
843         if (!(retval = put_user(current->gid, rgid)) &&
844             !(retval = put_user(current->egid, egid)))
845                 retval = put_user(current->sgid, sgid);
846
847         return retval;
848 }
849
850
851 /*
852  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
853  * is used for "access()" and for the NFS daemon (letting nfsd stay at
854  * whatever uid it wants to). It normally shadows "euid", except when
855  * explicitly set by setfsuid() or for access..
856  */
857 asmlinkage long sys_setfsuid(uid_t uid)
858 {
859         int old_fsuid;
860
861         old_fsuid = current->fsuid;
862         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
863                 return old_fsuid;
864
865         if (uid == current->uid || uid == current->euid ||
866             uid == current->suid || uid == current->fsuid || 
867             capable(CAP_SETUID))
868         {
869                 if (uid != old_fsuid)
870                 {
871                         current->mm->dumpable = suid_dumpable;
872                         smp_wmb();
873                 }
874                 current->fsuid = uid;
875         }
876
877         key_fsuid_changed(current);
878
879         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
880
881         return old_fsuid;
882 }
883
884 /*
885  * Samma på svenska..
886  */
887 asmlinkage long sys_setfsgid(gid_t gid)
888 {
889         int old_fsgid;
890
891         old_fsgid = current->fsgid;
892         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
893                 return old_fsgid;
894
895         if (gid == current->gid || gid == current->egid ||
896             gid == current->sgid || gid == current->fsgid || 
897             capable(CAP_SETGID))
898         {
899                 if (gid != old_fsgid)
900                 {
901                         current->mm->dumpable = suid_dumpable;
902                         smp_wmb();
903                 }
904                 current->fsgid = gid;
905                 key_fsgid_changed(current);
906         }
907         return old_fsgid;
908 }
909
910 asmlinkage long sys_times(struct tms __user * tbuf)
911 {
912         /*
913          *      In the SMP world we might just be unlucky and have one of
914          *      the times increment as we use it. Since the value is an
915          *      atomically safe type this is just fine. Conceptually its
916          *      as if the syscall took an instant longer to occur.
917          */
918         if (tbuf) {
919                 struct tms tmp;
920                 cputime_t utime, stime, cutime, cstime;
921
922 #ifdef CONFIG_SMP
923                 if (thread_group_empty(current)) {
924                         /*
925                          * Single thread case without the use of any locks.
926                          *
927                          * We may race with release_task if two threads are
928                          * executing. However, release task first adds up the
929                          * counters (__exit_signal) before  removing the task
930                          * from the process tasklist (__unhash_process).
931                          * __exit_signal also acquires and releases the
932                          * siglock which results in the proper memory ordering
933                          * so that the list modifications are always visible
934                          * after the counters have been updated.
935                          *
936                          * If the counters have been updated by the second thread
937                          * but the thread has not yet been removed from the list
938                          * then the other branch will be executing which will
939                          * block on tasklist_lock until the exit handling of the
940                          * other task is finished.
941                          *
942                          * This also implies that the sighand->siglock cannot
943                          * be held by another processor. So we can also
944                          * skip acquiring that lock.
945                          */
946                         utime = cputime_add(current->signal->utime, current->utime);
947                         stime = cputime_add(current->signal->utime, current->stime);
948                         cutime = current->signal->cutime;
949                         cstime = current->signal->cstime;
950                 } else
951 #endif
952                 {
953
954                         /* Process with multiple threads */
955                         struct task_struct *tsk = current;
956                         struct task_struct *t;
957
958                         read_lock(&tasklist_lock);
959                         utime = tsk->signal->utime;
960                         stime = tsk->signal->stime;
961                         t = tsk;
962                         do {
963                                 utime = cputime_add(utime, t->utime);
964                                 stime = cputime_add(stime, t->stime);
965                                 t = next_thread(t);
966                         } while (t != tsk);
967
968                         /*
969                          * While we have tasklist_lock read-locked, no dying thread
970                          * can be updating current->signal->[us]time.  Instead,
971                          * we got their counts included in the live thread loop.
972                          * However, another thread can come in right now and
973                          * do a wait call that updates current->signal->c[us]time.
974                          * To make sure we always see that pair updated atomically,
975                          * we take the siglock around fetching them.
976                          */
977                         spin_lock_irq(&tsk->sighand->siglock);
978                         cutime = tsk->signal->cutime;
979                         cstime = tsk->signal->cstime;
980                         spin_unlock_irq(&tsk->sighand->siglock);
981                         read_unlock(&tasklist_lock);
982                 }
983                 tmp.tms_utime = cputime_to_clock_t(utime);
984                 tmp.tms_stime = cputime_to_clock_t(stime);
985                 tmp.tms_cutime = cputime_to_clock_t(cutime);
986                 tmp.tms_cstime = cputime_to_clock_t(cstime);
987                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
988                         return -EFAULT;
989         }
990         return (long) jiffies_64_to_clock_t(get_jiffies_64());
991 }
992
993 /*
994  * This needs some heavy checking ...
995  * I just haven't the stomach for it. I also don't fully
996  * understand sessions/pgrp etc. Let somebody who does explain it.
997  *
998  * OK, I think I have the protection semantics right.... this is really
999  * only important on a multi-user system anyway, to make sure one user
1000  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1001  *
1002  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1003  * LBT 04.03.94
1004  */
1005
1006 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1007 {
1008         struct task_struct *p;
1009         int err = -EINVAL;
1010
1011         if (!pid)
1012                 pid = current->pid;
1013         if (!pgid)
1014                 pgid = pid;
1015         if (pgid < 0)
1016                 return -EINVAL;
1017
1018         /* From this point forward we keep holding onto the tasklist lock
1019          * so that our parent does not change from under us. -DaveM
1020          */
1021         write_lock_irq(&tasklist_lock);
1022
1023         err = -ESRCH;
1024         p = find_task_by_pid(pid);
1025         if (!p)
1026                 goto out;
1027
1028         err = -EINVAL;
1029         if (!thread_group_leader(p))
1030                 goto out;
1031
1032         if (p->parent == current || p->real_parent == current) {
1033                 err = -EPERM;
1034                 if (p->signal->session != current->signal->session)
1035                         goto out;
1036                 err = -EACCES;
1037                 if (p->did_exec)
1038                         goto out;
1039         } else {
1040                 err = -ESRCH;
1041                 if (p != current)
1042                         goto out;
1043         }
1044
1045         err = -EPERM;
1046         if (p->signal->leader)
1047                 goto out;
1048
1049         if (pgid != pid) {
1050                 struct task_struct *p;
1051
1052                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1053                         if (p->signal->session == current->signal->session)
1054                                 goto ok_pgid;
1055                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1056                 goto out;
1057         }
1058
1059 ok_pgid:
1060         err = security_task_setpgid(p, pgid);
1061         if (err)
1062                 goto out;
1063
1064         if (process_group(p) != pgid) {
1065                 detach_pid(p, PIDTYPE_PGID);
1066                 p->signal->pgrp = pgid;
1067                 attach_pid(p, PIDTYPE_PGID, pgid);
1068         }
1069
1070         err = 0;
1071 out:
1072         /* All paths lead to here, thus we are safe. -DaveM */
1073         write_unlock_irq(&tasklist_lock);
1074         return err;
1075 }
1076
1077 asmlinkage long sys_getpgid(pid_t pid)
1078 {
1079         if (!pid) {
1080                 return process_group(current);
1081         } else {
1082                 int retval;
1083                 struct task_struct *p;
1084
1085                 read_lock(&tasklist_lock);
1086                 p = find_task_by_pid(pid);
1087
1088                 retval = -ESRCH;
1089                 if (p) {
1090                         retval = security_task_getpgid(p);
1091                         if (!retval)
1092                                 retval = process_group(p);
1093                 }
1094                 read_unlock(&tasklist_lock);
1095                 return retval;
1096         }
1097 }
1098
1099 #ifdef __ARCH_WANT_SYS_GETPGRP
1100
1101 asmlinkage long sys_getpgrp(void)
1102 {
1103         /* SMP - assuming writes are word atomic this is fine */
1104         return process_group(current);
1105 }
1106
1107 #endif
1108
1109 asmlinkage long sys_getsid(pid_t pid)
1110 {
1111         if (!pid) {
1112                 return current->signal->session;
1113         } else {
1114                 int retval;
1115                 struct task_struct *p;
1116
1117                 read_lock(&tasklist_lock);
1118                 p = find_task_by_pid(pid);
1119
1120                 retval = -ESRCH;
1121                 if(p) {
1122                         retval = security_task_getsid(p);
1123                         if (!retval)
1124                                 retval = p->signal->session;
1125                 }
1126                 read_unlock(&tasklist_lock);
1127                 return retval;
1128         }
1129 }
1130
1131 asmlinkage long sys_setsid(void)
1132 {
1133         struct pid *pid;
1134         int err = -EPERM;
1135
1136         if (!thread_group_leader(current))
1137                 return -EINVAL;
1138
1139         down(&tty_sem);
1140         write_lock_irq(&tasklist_lock);
1141
1142         pid = find_pid(PIDTYPE_PGID, current->pid);
1143         if (pid)
1144                 goto out;
1145
1146         current->signal->leader = 1;
1147         __set_special_pids(current->pid, current->pid);
1148         current->signal->tty = NULL;
1149         current->signal->tty_old_pgrp = 0;
1150         err = process_group(current);
1151 out:
1152         write_unlock_irq(&tasklist_lock);
1153         up(&tty_sem);
1154         return err;
1155 }
1156
1157 /*
1158  * Supplementary group IDs
1159  */
1160
1161 /* init to 2 - one for init_task, one to ensure it is never freed */
1162 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1163
1164 struct group_info *groups_alloc(int gidsetsize)
1165 {
1166         struct group_info *group_info;
1167         int nblocks;
1168         int i;
1169
1170         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1171         /* Make sure we always allocate at least one indirect block pointer */
1172         nblocks = nblocks ? : 1;
1173         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1174         if (!group_info)
1175                 return NULL;
1176         group_info->ngroups = gidsetsize;
1177         group_info->nblocks = nblocks;
1178         atomic_set(&group_info->usage, 1);
1179
1180         if (gidsetsize <= NGROUPS_SMALL) {
1181                 group_info->blocks[0] = group_info->small_block;
1182         } else {
1183                 for (i = 0; i < nblocks; i++) {
1184                         gid_t *b;
1185                         b = (void *)__get_free_page(GFP_USER);
1186                         if (!b)
1187                                 goto out_undo_partial_alloc;
1188                         group_info->blocks[i] = b;
1189                 }
1190         }
1191         return group_info;
1192
1193 out_undo_partial_alloc:
1194         while (--i >= 0) {
1195                 free_page((unsigned long)group_info->blocks[i]);
1196         }
1197         kfree(group_info);
1198         return NULL;
1199 }
1200
1201 EXPORT_SYMBOL(groups_alloc);
1202
1203 void groups_free(struct group_info *group_info)
1204 {
1205         if (group_info->blocks[0] != group_info->small_block) {
1206                 int i;
1207                 for (i = 0; i < group_info->nblocks; i++)
1208                         free_page((unsigned long)group_info->blocks[i]);
1209         }
1210         kfree(group_info);
1211 }
1212
1213 EXPORT_SYMBOL(groups_free);
1214
1215 /* export the group_info to a user-space array */
1216 static int groups_to_user(gid_t __user *grouplist,
1217     struct group_info *group_info)
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_to_user(grouplist+off, group_info->blocks[i], len))
1228                         return -EFAULT;
1229
1230                 count -= cp_count;
1231         }
1232         return 0;
1233 }
1234
1235 /* fill a group_info from a user-space array - it must be allocated already */
1236 static int groups_from_user(struct group_info *group_info,
1237     gid_t __user *grouplist)
1238  {
1239         int i;
1240         int count = group_info->ngroups;
1241
1242         for (i = 0; i < group_info->nblocks; i++) {
1243                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1244                 int off = i * NGROUPS_PER_BLOCK;
1245                 int len = cp_count * sizeof(*grouplist);
1246
1247                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1248                         return -EFAULT;
1249
1250                 count -= cp_count;
1251         }
1252         return 0;
1253 }
1254
1255 /* a simple Shell sort */
1256 static void groups_sort(struct group_info *group_info)
1257 {
1258         int base, max, stride;
1259         int gidsetsize = group_info->ngroups;
1260
1261         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1262                 ; /* nothing */
1263         stride /= 3;
1264
1265         while (stride) {
1266                 max = gidsetsize - stride;
1267                 for (base = 0; base < max; base++) {
1268                         int left = base;
1269                         int right = left + stride;
1270                         gid_t tmp = GROUP_AT(group_info, right);
1271
1272                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1273                                 GROUP_AT(group_info, right) =
1274                                     GROUP_AT(group_info, left);
1275                                 right = left;
1276                                 left -= stride;
1277                         }
1278                         GROUP_AT(group_info, right) = tmp;
1279                 }
1280                 stride /= 3;
1281         }
1282 }
1283
1284 /* a simple bsearch */
1285 int groups_search(struct group_info *group_info, gid_t grp)
1286 {
1287         int left, right;
1288
1289         if (!group_info)
1290                 return 0;
1291
1292         left = 0;
1293         right = group_info->ngroups;
1294         while (left < right) {
1295                 int mid = (left+right)/2;
1296                 int cmp = grp - GROUP_AT(group_info, mid);
1297                 if (cmp > 0)
1298                         left = mid + 1;
1299                 else if (cmp < 0)
1300                         right = mid;
1301                 else
1302                         return 1;
1303         }
1304         return 0;
1305 }
1306
1307 /* validate and set current->group_info */
1308 int set_current_groups(struct group_info *group_info)
1309 {
1310         int retval;
1311         struct group_info *old_info;
1312
1313         retval = security_task_setgroups(group_info);
1314         if (retval)
1315                 return retval;
1316
1317         groups_sort(group_info);
1318         get_group_info(group_info);
1319
1320         task_lock(current);
1321         old_info = current->group_info;
1322         current->group_info = group_info;
1323         task_unlock(current);
1324
1325         put_group_info(old_info);
1326
1327         return 0;
1328 }
1329
1330 EXPORT_SYMBOL(set_current_groups);
1331
1332 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1333 {
1334         int i = 0;
1335
1336         /*
1337          *      SMP: Nobody else can change our grouplist. Thus we are
1338          *      safe.
1339          */
1340
1341         if (gidsetsize < 0)
1342                 return -EINVAL;
1343
1344         /* no need to grab task_lock here; it cannot change */
1345         get_group_info(current->group_info);
1346         i = current->group_info->ngroups;
1347         if (gidsetsize) {
1348                 if (i > gidsetsize) {
1349                         i = -EINVAL;
1350                         goto out;
1351                 }
1352                 if (groups_to_user(grouplist, current->group_info)) {
1353                         i = -EFAULT;
1354                         goto out;
1355                 }
1356         }
1357 out:
1358         put_group_info(current->group_info);
1359         return i;
1360 }
1361
1362 /*
1363  *      SMP: Our groups are copy-on-write. We can set them safely
1364  *      without another task interfering.
1365  */
1366  
1367 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1368 {
1369         struct group_info *group_info;
1370         int retval;
1371
1372         if (!capable(CAP_SETGID))
1373                 return -EPERM;
1374         if ((unsigned)gidsetsize > NGROUPS_MAX)
1375                 return -EINVAL;
1376
1377         group_info = groups_alloc(gidsetsize);
1378         if (!group_info)
1379                 return -ENOMEM;
1380         retval = groups_from_user(group_info, grouplist);
1381         if (retval) {
1382                 put_group_info(group_info);
1383                 return retval;
1384         }
1385
1386         retval = set_current_groups(group_info);
1387         put_group_info(group_info);
1388
1389         return retval;
1390 }
1391
1392 /*
1393  * Check whether we're fsgid/egid or in the supplemental group..
1394  */
1395 int in_group_p(gid_t grp)
1396 {
1397         int retval = 1;
1398         if (grp != current->fsgid) {
1399                 get_group_info(current->group_info);
1400                 retval = groups_search(current->group_info, grp);
1401                 put_group_info(current->group_info);
1402         }
1403         return retval;
1404 }
1405
1406 EXPORT_SYMBOL(in_group_p);
1407
1408 int in_egroup_p(gid_t grp)
1409 {
1410         int retval = 1;
1411         if (grp != current->egid) {
1412                 get_group_info(current->group_info);
1413                 retval = groups_search(current->group_info, grp);
1414                 put_group_info(current->group_info);
1415         }
1416         return retval;
1417 }
1418
1419 EXPORT_SYMBOL(in_egroup_p);
1420
1421 DECLARE_RWSEM(uts_sem);
1422
1423 EXPORT_SYMBOL(uts_sem);
1424
1425 asmlinkage long sys_newuname(struct new_utsname __user * name)
1426 {
1427         int errno = 0;
1428
1429         down_read(&uts_sem);
1430         if (copy_to_user(name,&system_utsname,sizeof *name))
1431                 errno = -EFAULT;
1432         up_read(&uts_sem);
1433         return errno;
1434 }
1435
1436 asmlinkage long sys_sethostname(char __user *name, int len)
1437 {
1438         int errno;
1439         char tmp[__NEW_UTS_LEN];
1440
1441         if (!capable(CAP_SYS_ADMIN))
1442                 return -EPERM;
1443         if (len < 0 || len > __NEW_UTS_LEN)
1444                 return -EINVAL;
1445         down_write(&uts_sem);
1446         errno = -EFAULT;
1447         if (!copy_from_user(tmp, name, len)) {
1448                 memcpy(system_utsname.nodename, tmp, len);
1449                 system_utsname.nodename[len] = 0;
1450                 errno = 0;
1451         }
1452         up_write(&uts_sem);
1453         return errno;
1454 }
1455
1456 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1457
1458 asmlinkage long sys_gethostname(char __user *name, int len)
1459 {
1460         int i, errno;
1461
1462         if (len < 0)
1463                 return -EINVAL;
1464         down_read(&uts_sem);
1465         i = 1 + strlen(system_utsname.nodename);
1466         if (i > len)
1467                 i = len;
1468         errno = 0;
1469         if (copy_to_user(name, system_utsname.nodename, i))
1470                 errno = -EFAULT;
1471         up_read(&uts_sem);
1472         return errno;
1473 }
1474
1475 #endif
1476
1477 /*
1478  * Only setdomainname; getdomainname can be implemented by calling
1479  * uname()
1480  */
1481 asmlinkage long sys_setdomainname(char __user *name, int len)
1482 {
1483         int errno;
1484         char tmp[__NEW_UTS_LEN];
1485
1486         if (!capable(CAP_SYS_ADMIN))
1487                 return -EPERM;
1488         if (len < 0 || len > __NEW_UTS_LEN)
1489                 return -EINVAL;
1490
1491         down_write(&uts_sem);
1492         errno = -EFAULT;
1493         if (!copy_from_user(tmp, name, len)) {
1494                 memcpy(system_utsname.domainname, tmp, len);
1495                 system_utsname.domainname[len] = 0;
1496                 errno = 0;
1497         }
1498         up_write(&uts_sem);
1499         return errno;
1500 }
1501
1502 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1503 {
1504         if (resource >= RLIM_NLIMITS)
1505                 return -EINVAL;
1506         else {
1507                 struct rlimit value;
1508                 task_lock(current->group_leader);
1509                 value = current->signal->rlim[resource];
1510                 task_unlock(current->group_leader);
1511                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1512         }
1513 }
1514
1515 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1516
1517 /*
1518  *      Back compatibility for getrlimit. Needed for some apps.
1519  */
1520  
1521 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1522 {
1523         struct rlimit x;
1524         if (resource >= RLIM_NLIMITS)
1525                 return -EINVAL;
1526
1527         task_lock(current->group_leader);
1528         x = current->signal->rlim[resource];
1529         task_unlock(current->group_leader);
1530         if(x.rlim_cur > 0x7FFFFFFF)
1531                 x.rlim_cur = 0x7FFFFFFF;
1532         if(x.rlim_max > 0x7FFFFFFF)
1533                 x.rlim_max = 0x7FFFFFFF;
1534         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1535 }
1536
1537 #endif
1538
1539 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1540 {
1541         struct rlimit new_rlim, *old_rlim;
1542         int retval;
1543
1544         if (resource >= RLIM_NLIMITS)
1545                 return -EINVAL;
1546         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1547                 return -EFAULT;
1548        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1549                return -EINVAL;
1550         old_rlim = current->signal->rlim + resource;
1551         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1552             !capable(CAP_SYS_RESOURCE))
1553                 return -EPERM;
1554         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1555                         return -EPERM;
1556
1557         retval = security_task_setrlimit(resource, &new_rlim);
1558         if (retval)
1559                 return retval;
1560
1561         task_lock(current->group_leader);
1562         *old_rlim = new_rlim;
1563         task_unlock(current->group_leader);
1564
1565         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1566             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1567              new_rlim.rlim_cur <= cputime_to_secs(
1568                      current->signal->it_prof_expires))) {
1569                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1570                 read_lock(&tasklist_lock);
1571                 spin_lock_irq(&current->sighand->siglock);
1572                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1573                                       &cputime, NULL);
1574                 spin_unlock_irq(&current->sighand->siglock);
1575                 read_unlock(&tasklist_lock);
1576         }
1577
1578         return 0;
1579 }
1580
1581 /*
1582  * It would make sense to put struct rusage in the task_struct,
1583  * except that would make the task_struct be *really big*.  After
1584  * task_struct gets moved into malloc'ed memory, it would
1585  * make sense to do this.  It will make moving the rest of the information
1586  * a lot simpler!  (Which we're not doing right now because we're not
1587  * measuring them yet).
1588  *
1589  * This expects to be called with tasklist_lock read-locked or better,
1590  * and the siglock not locked.  It may momentarily take the siglock.
1591  *
1592  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1593  * races with threads incrementing their own counters.  But since word
1594  * reads are atomic, we either get new values or old values and we don't
1595  * care which for the sums.  We always take the siglock to protect reading
1596  * the c* fields from p->signal from races with exit.c updating those
1597  * fields when reaping, so a sample either gets all the additions of a
1598  * given child after it's reaped, or none so this sample is before reaping.
1599  */
1600
1601 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1602 {
1603         struct task_struct *t;
1604         unsigned long flags;
1605         cputime_t utime, stime;
1606
1607         memset((char *) r, 0, sizeof *r);
1608
1609         if (unlikely(!p->signal))
1610                 return;
1611
1612         switch (who) {
1613                 case RUSAGE_CHILDREN:
1614                         spin_lock_irqsave(&p->sighand->siglock, flags);
1615                         utime = p->signal->cutime;
1616                         stime = p->signal->cstime;
1617                         r->ru_nvcsw = p->signal->cnvcsw;
1618                         r->ru_nivcsw = p->signal->cnivcsw;
1619                         r->ru_minflt = p->signal->cmin_flt;
1620                         r->ru_majflt = p->signal->cmaj_flt;
1621                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1622                         cputime_to_timeval(utime, &r->ru_utime);
1623                         cputime_to_timeval(stime, &r->ru_stime);
1624                         break;
1625                 case RUSAGE_SELF:
1626                         spin_lock_irqsave(&p->sighand->siglock, flags);
1627                         utime = stime = cputime_zero;
1628                         goto sum_group;
1629                 case RUSAGE_BOTH:
1630                         spin_lock_irqsave(&p->sighand->siglock, flags);
1631                         utime = p->signal->cutime;
1632                         stime = p->signal->cstime;
1633                         r->ru_nvcsw = p->signal->cnvcsw;
1634                         r->ru_nivcsw = p->signal->cnivcsw;
1635                         r->ru_minflt = p->signal->cmin_flt;
1636                         r->ru_majflt = p->signal->cmaj_flt;
1637                 sum_group:
1638                         utime = cputime_add(utime, p->signal->utime);
1639                         stime = cputime_add(stime, p->signal->stime);
1640                         r->ru_nvcsw += p->signal->nvcsw;
1641                         r->ru_nivcsw += p->signal->nivcsw;
1642                         r->ru_minflt += p->signal->min_flt;
1643                         r->ru_majflt += p->signal->maj_flt;
1644                         t = p;
1645                         do {
1646                                 utime = cputime_add(utime, t->utime);
1647                                 stime = cputime_add(stime, t->stime);
1648                                 r->ru_nvcsw += t->nvcsw;
1649                                 r->ru_nivcsw += t->nivcsw;
1650                                 r->ru_minflt += t->min_flt;
1651                                 r->ru_majflt += t->maj_flt;
1652                                 t = next_thread(t);
1653                         } while (t != p);
1654                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1655                         cputime_to_timeval(utime, &r->ru_utime);
1656                         cputime_to_timeval(stime, &r->ru_stime);
1657                         break;
1658                 default:
1659                         BUG();
1660         }
1661 }
1662
1663 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1664 {
1665         struct rusage r;
1666         read_lock(&tasklist_lock);
1667         k_getrusage(p, who, &r);
1668         read_unlock(&tasklist_lock);
1669         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1670 }
1671
1672 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1673 {
1674         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1675                 return -EINVAL;
1676         return getrusage(current, who, ru);
1677 }
1678
1679 asmlinkage long sys_umask(int mask)
1680 {
1681         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1682         return mask;
1683 }
1684     
1685 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1686                           unsigned long arg4, unsigned long arg5)
1687 {
1688         long error;
1689         int sig;
1690
1691         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1692         if (error)
1693                 return error;
1694
1695         switch (option) {
1696                 case PR_SET_PDEATHSIG:
1697                         sig = arg2;
1698                         if (!valid_signal(sig)) {
1699                                 error = -EINVAL;
1700                                 break;
1701                         }
1702                         current->pdeath_signal = sig;
1703                         break;
1704                 case PR_GET_PDEATHSIG:
1705                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1706                         break;
1707                 case PR_GET_DUMPABLE:
1708                         if (current->mm->dumpable)
1709                                 error = 1;
1710                         break;
1711                 case PR_SET_DUMPABLE:
1712                         if (arg2 < 0 || arg2 > 2) {
1713                                 error = -EINVAL;
1714                                 break;
1715                         }
1716                         current->mm->dumpable = arg2;
1717                         break;
1718
1719                 case PR_SET_UNALIGN:
1720                         error = SET_UNALIGN_CTL(current, arg2);
1721                         break;
1722                 case PR_GET_UNALIGN:
1723                         error = GET_UNALIGN_CTL(current, arg2);
1724                         break;
1725                 case PR_SET_FPEMU:
1726                         error = SET_FPEMU_CTL(current, arg2);
1727                         break;
1728                 case PR_GET_FPEMU:
1729                         error = GET_FPEMU_CTL(current, arg2);
1730                         break;
1731                 case PR_SET_FPEXC:
1732                         error = SET_FPEXC_CTL(current, arg2);
1733                         break;
1734                 case PR_GET_FPEXC:
1735                         error = GET_FPEXC_CTL(current, arg2);
1736                         break;
1737                 case PR_GET_TIMING:
1738                         error = PR_TIMING_STATISTICAL;
1739                         break;
1740                 case PR_SET_TIMING:
1741                         if (arg2 == PR_TIMING_STATISTICAL)
1742                                 error = 0;
1743                         else
1744                                 error = -EINVAL;
1745                         break;
1746
1747                 case PR_GET_KEEPCAPS:
1748                         if (current->keep_capabilities)
1749                                 error = 1;
1750                         break;
1751                 case PR_SET_KEEPCAPS:
1752                         if (arg2 != 0 && arg2 != 1) {
1753                                 error = -EINVAL;
1754                                 break;
1755                         }
1756                         current->keep_capabilities = arg2;
1757                         break;
1758                 case PR_SET_NAME: {
1759                         struct task_struct *me = current;
1760                         unsigned char ncomm[sizeof(me->comm)];
1761
1762                         ncomm[sizeof(me->comm)-1] = 0;
1763                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1764                                                 sizeof(me->comm)-1) < 0)
1765                                 return -EFAULT;
1766                         set_task_comm(me, ncomm);
1767                         return 0;
1768                 }
1769                 case PR_GET_NAME: {
1770                         struct task_struct *me = current;
1771                         unsigned char tcomm[sizeof(me->comm)];
1772
1773                         get_task_comm(tcomm, me);
1774                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1775                                 return -EFAULT;
1776                         return 0;
1777                 }
1778                 default:
1779                         error = -EINVAL;
1780                         break;
1781         }
1782         return error;
1783 }