[NETLINK]: Add properly module refcounting for kernel netlink sockets.
[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 void emergency_restart(void)
365 {
366         machine_emergency_restart();
367 }
368 EXPORT_SYMBOL_GPL(emergency_restart);
369
370 void kernel_restart(char *cmd)
371 {
372         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
373         system_state = SYSTEM_RESTART;
374         device_shutdown();
375         if (!cmd) {
376                 printk(KERN_EMERG "Restarting system.\n");
377         } else {
378                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
379         }
380         printk(".\n");
381         machine_restart(cmd);
382 }
383 EXPORT_SYMBOL_GPL(kernel_restart);
384
385 void kernel_kexec(void)
386 {
387 #ifdef CONFIG_KEXEC
388         struct kimage *image;
389         image = xchg(&kexec_image, 0);
390         if (!image) {
391                 return;
392         }
393         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, NULL);
394         system_state = SYSTEM_RESTART;
395         device_shutdown();
396         printk(KERN_EMERG "Starting new kernel\n");
397         machine_shutdown();
398         machine_kexec(image);
399 #endif
400 }
401 EXPORT_SYMBOL_GPL(kernel_kexec);
402
403 void kernel_halt(void)
404 {
405         notifier_call_chain(&reboot_notifier_list, SYS_HALT, NULL);
406         system_state = SYSTEM_HALT;
407         device_shutdown();
408         printk(KERN_EMERG "System halted.\n");
409         machine_halt();
410 }
411 EXPORT_SYMBOL_GPL(kernel_halt);
412
413 void kernel_power_off(void)
414 {
415         notifier_call_chain(&reboot_notifier_list, SYS_POWER_OFF, NULL);
416         system_state = SYSTEM_POWER_OFF;
417         device_shutdown();
418         printk(KERN_EMERG "Power down.\n");
419         machine_power_off();
420 }
421 EXPORT_SYMBOL_GPL(kernel_power_off);
422
423 /*
424  * Reboot system call: for obvious reasons only root may call it,
425  * and even root needs to set up some magic numbers in the registers
426  * so that some mistake won't make this reboot the whole machine.
427  * You can also set the meaning of the ctrl-alt-del-key here.
428  *
429  * reboot doesn't sync: do that yourself before calling this.
430  */
431 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
432 {
433         char buffer[256];
434
435         /* We only trust the superuser with rebooting the system. */
436         if (!capable(CAP_SYS_BOOT))
437                 return -EPERM;
438
439         /* For safety, we require "magic" arguments. */
440         if (magic1 != LINUX_REBOOT_MAGIC1 ||
441             (magic2 != LINUX_REBOOT_MAGIC2 &&
442                         magic2 != LINUX_REBOOT_MAGIC2A &&
443                         magic2 != LINUX_REBOOT_MAGIC2B &&
444                         magic2 != LINUX_REBOOT_MAGIC2C))
445                 return -EINVAL;
446
447         lock_kernel();
448         switch (cmd) {
449         case LINUX_REBOOT_CMD_RESTART:
450                 kernel_restart(NULL);
451                 break;
452
453         case LINUX_REBOOT_CMD_CAD_ON:
454                 C_A_D = 1;
455                 break;
456
457         case LINUX_REBOOT_CMD_CAD_OFF:
458                 C_A_D = 0;
459                 break;
460
461         case LINUX_REBOOT_CMD_HALT:
462                 kernel_halt();
463                 unlock_kernel();
464                 do_exit(0);
465                 break;
466
467         case LINUX_REBOOT_CMD_POWER_OFF:
468                 kernel_power_off();
469                 unlock_kernel();
470                 do_exit(0);
471                 break;
472
473         case LINUX_REBOOT_CMD_RESTART2:
474                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
475                         unlock_kernel();
476                         return -EFAULT;
477                 }
478                 buffer[sizeof(buffer) - 1] = '\0';
479
480                 kernel_restart(buffer);
481                 break;
482
483         case LINUX_REBOOT_CMD_KEXEC:
484                 kernel_kexec();
485                 unlock_kernel();
486                 return -EINVAL;
487
488 #ifdef CONFIG_SOFTWARE_SUSPEND
489         case LINUX_REBOOT_CMD_SW_SUSPEND:
490                 {
491                         int ret = software_suspend();
492                         unlock_kernel();
493                         return ret;
494                 }
495 #endif
496
497         default:
498                 unlock_kernel();
499                 return -EINVAL;
500         }
501         unlock_kernel();
502         return 0;
503 }
504
505 static void deferred_cad(void *dummy)
506 {
507         kernel_restart(NULL);
508 }
509
510 /*
511  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
512  * As it's called within an interrupt, it may NOT sync: the only choice
513  * is whether to reboot at once, or just ignore the ctrl-alt-del.
514  */
515 void ctrl_alt_del(void)
516 {
517         static DECLARE_WORK(cad_work, deferred_cad, NULL);
518
519         if (C_A_D)
520                 schedule_work(&cad_work);
521         else
522                 kill_proc(cad_pid, SIGINT, 1);
523 }
524         
525
526 /*
527  * Unprivileged users may change the real gid to the effective gid
528  * or vice versa.  (BSD-style)
529  *
530  * If you set the real gid at all, or set the effective gid to a value not
531  * equal to the real gid, then the saved gid is set to the new effective gid.
532  *
533  * This makes it possible for a setgid program to completely drop its
534  * privileges, which is often a useful assertion to make when you are doing
535  * a security audit over a program.
536  *
537  * The general idea is that a program which uses just setregid() will be
538  * 100% compatible with BSD.  A program which uses just setgid() will be
539  * 100% compatible with POSIX with saved IDs. 
540  *
541  * SMP: There are not races, the GIDs are checked only by filesystem
542  *      operations (as far as semantic preservation is concerned).
543  */
544 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
545 {
546         int old_rgid = current->gid;
547         int old_egid = current->egid;
548         int new_rgid = old_rgid;
549         int new_egid = old_egid;
550         int retval;
551
552         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
553         if (retval)
554                 return retval;
555
556         if (rgid != (gid_t) -1) {
557                 if ((old_rgid == rgid) ||
558                     (current->egid==rgid) ||
559                     capable(CAP_SETGID))
560                         new_rgid = rgid;
561                 else
562                         return -EPERM;
563         }
564         if (egid != (gid_t) -1) {
565                 if ((old_rgid == egid) ||
566                     (current->egid == egid) ||
567                     (current->sgid == egid) ||
568                     capable(CAP_SETGID))
569                         new_egid = egid;
570                 else {
571                         return -EPERM;
572                 }
573         }
574         if (new_egid != old_egid)
575         {
576                 current->mm->dumpable = suid_dumpable;
577                 smp_wmb();
578         }
579         if (rgid != (gid_t) -1 ||
580             (egid != (gid_t) -1 && egid != old_rgid))
581                 current->sgid = new_egid;
582         current->fsgid = new_egid;
583         current->egid = new_egid;
584         current->gid = new_rgid;
585         key_fsgid_changed(current);
586         return 0;
587 }
588
589 /*
590  * setgid() is implemented like SysV w/ SAVED_IDS 
591  *
592  * SMP: Same implicit races as above.
593  */
594 asmlinkage long sys_setgid(gid_t gid)
595 {
596         int old_egid = current->egid;
597         int retval;
598
599         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
600         if (retval)
601                 return retval;
602
603         if (capable(CAP_SETGID))
604         {
605                 if(old_egid != gid)
606                 {
607                         current->mm->dumpable = suid_dumpable;
608                         smp_wmb();
609                 }
610                 current->gid = current->egid = current->sgid = current->fsgid = gid;
611         }
612         else if ((gid == current->gid) || (gid == current->sgid))
613         {
614                 if(old_egid != gid)
615                 {
616                         current->mm->dumpable = suid_dumpable;
617                         smp_wmb();
618                 }
619                 current->egid = current->fsgid = gid;
620         }
621         else
622                 return -EPERM;
623
624         key_fsgid_changed(current);
625         return 0;
626 }
627   
628 static int set_user(uid_t new_ruid, int dumpclear)
629 {
630         struct user_struct *new_user;
631
632         new_user = alloc_uid(new_ruid);
633         if (!new_user)
634                 return -EAGAIN;
635
636         if (atomic_read(&new_user->processes) >=
637                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
638                         new_user != &root_user) {
639                 free_uid(new_user);
640                 return -EAGAIN;
641         }
642
643         switch_uid(new_user);
644
645         if(dumpclear)
646         {
647                 current->mm->dumpable = suid_dumpable;
648                 smp_wmb();
649         }
650         current->uid = new_ruid;
651         return 0;
652 }
653
654 /*
655  * Unprivileged users may change the real uid to the effective uid
656  * or vice versa.  (BSD-style)
657  *
658  * If you set the real uid at all, or set the effective uid to a value not
659  * equal to the real uid, then the saved uid is set to the new effective uid.
660  *
661  * This makes it possible for a setuid program to completely drop its
662  * privileges, which is often a useful assertion to make when you are doing
663  * a security audit over a program.
664  *
665  * The general idea is that a program which uses just setreuid() will be
666  * 100% compatible with BSD.  A program which uses just setuid() will be
667  * 100% compatible with POSIX with saved IDs. 
668  */
669 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
670 {
671         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
672         int retval;
673
674         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
675         if (retval)
676                 return retval;
677
678         new_ruid = old_ruid = current->uid;
679         new_euid = old_euid = current->euid;
680         old_suid = current->suid;
681
682         if (ruid != (uid_t) -1) {
683                 new_ruid = ruid;
684                 if ((old_ruid != ruid) &&
685                     (current->euid != ruid) &&
686                     !capable(CAP_SETUID))
687                         return -EPERM;
688         }
689
690         if (euid != (uid_t) -1) {
691                 new_euid = euid;
692                 if ((old_ruid != euid) &&
693                     (current->euid != euid) &&
694                     (current->suid != euid) &&
695                     !capable(CAP_SETUID))
696                         return -EPERM;
697         }
698
699         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
700                 return -EAGAIN;
701
702         if (new_euid != old_euid)
703         {
704                 current->mm->dumpable = suid_dumpable;
705                 smp_wmb();
706         }
707         current->fsuid = current->euid = new_euid;
708         if (ruid != (uid_t) -1 ||
709             (euid != (uid_t) -1 && euid != old_ruid))
710                 current->suid = current->euid;
711         current->fsuid = current->euid;
712
713         key_fsuid_changed(current);
714
715         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
716 }
717
718
719                 
720 /*
721  * setuid() is implemented like SysV with SAVED_IDS 
722  * 
723  * Note that SAVED_ID's is deficient in that a setuid root program
724  * like sendmail, for example, cannot set its uid to be a normal 
725  * user and then switch back, because if you're root, setuid() sets
726  * the saved uid too.  If you don't like this, blame the bright people
727  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
728  * will allow a root program to temporarily drop privileges and be able to
729  * regain them by swapping the real and effective uid.  
730  */
731 asmlinkage long sys_setuid(uid_t uid)
732 {
733         int old_euid = current->euid;
734         int old_ruid, old_suid, new_ruid, new_suid;
735         int retval;
736
737         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
738         if (retval)
739                 return retval;
740
741         old_ruid = new_ruid = current->uid;
742         old_suid = current->suid;
743         new_suid = old_suid;
744         
745         if (capable(CAP_SETUID)) {
746                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
747                         return -EAGAIN;
748                 new_suid = uid;
749         } else if ((uid != current->uid) && (uid != new_suid))
750                 return -EPERM;
751
752         if (old_euid != uid)
753         {
754                 current->mm->dumpable = suid_dumpable;
755                 smp_wmb();
756         }
757         current->fsuid = current->euid = uid;
758         current->suid = new_suid;
759
760         key_fsuid_changed(current);
761
762         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
763 }
764
765
766 /*
767  * This function implements a generic ability to update ruid, euid,
768  * and suid.  This allows you to implement the 4.4 compatible seteuid().
769  */
770 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
771 {
772         int old_ruid = current->uid;
773         int old_euid = current->euid;
774         int old_suid = current->suid;
775         int retval;
776
777         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
778         if (retval)
779                 return retval;
780
781         if (!capable(CAP_SETUID)) {
782                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
783                     (ruid != current->euid) && (ruid != current->suid))
784                         return -EPERM;
785                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
786                     (euid != current->euid) && (euid != current->suid))
787                         return -EPERM;
788                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
789                     (suid != current->euid) && (suid != current->suid))
790                         return -EPERM;
791         }
792         if (ruid != (uid_t) -1) {
793                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
794                         return -EAGAIN;
795         }
796         if (euid != (uid_t) -1) {
797                 if (euid != current->euid)
798                 {
799                         current->mm->dumpable = suid_dumpable;
800                         smp_wmb();
801                 }
802                 current->euid = euid;
803         }
804         current->fsuid = current->euid;
805         if (suid != (uid_t) -1)
806                 current->suid = suid;
807
808         key_fsuid_changed(current);
809
810         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
811 }
812
813 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
814 {
815         int retval;
816
817         if (!(retval = put_user(current->uid, ruid)) &&
818             !(retval = put_user(current->euid, euid)))
819                 retval = put_user(current->suid, suid);
820
821         return retval;
822 }
823
824 /*
825  * Same as above, but for rgid, egid, sgid.
826  */
827 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
828 {
829         int retval;
830
831         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
832         if (retval)
833                 return retval;
834
835         if (!capable(CAP_SETGID)) {
836                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
837                     (rgid != current->egid) && (rgid != current->sgid))
838                         return -EPERM;
839                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
840                     (egid != current->egid) && (egid != current->sgid))
841                         return -EPERM;
842                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
843                     (sgid != current->egid) && (sgid != current->sgid))
844                         return -EPERM;
845         }
846         if (egid != (gid_t) -1) {
847                 if (egid != current->egid)
848                 {
849                         current->mm->dumpable = suid_dumpable;
850                         smp_wmb();
851                 }
852                 current->egid = egid;
853         }
854         current->fsgid = current->egid;
855         if (rgid != (gid_t) -1)
856                 current->gid = rgid;
857         if (sgid != (gid_t) -1)
858                 current->sgid = sgid;
859
860         key_fsgid_changed(current);
861         return 0;
862 }
863
864 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
865 {
866         int retval;
867
868         if (!(retval = put_user(current->gid, rgid)) &&
869             !(retval = put_user(current->egid, egid)))
870                 retval = put_user(current->sgid, sgid);
871
872         return retval;
873 }
874
875
876 /*
877  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
878  * is used for "access()" and for the NFS daemon (letting nfsd stay at
879  * whatever uid it wants to). It normally shadows "euid", except when
880  * explicitly set by setfsuid() or for access..
881  */
882 asmlinkage long sys_setfsuid(uid_t uid)
883 {
884         int old_fsuid;
885
886         old_fsuid = current->fsuid;
887         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
888                 return old_fsuid;
889
890         if (uid == current->uid || uid == current->euid ||
891             uid == current->suid || uid == current->fsuid || 
892             capable(CAP_SETUID))
893         {
894                 if (uid != old_fsuid)
895                 {
896                         current->mm->dumpable = suid_dumpable;
897                         smp_wmb();
898                 }
899                 current->fsuid = uid;
900         }
901
902         key_fsuid_changed(current);
903
904         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
905
906         return old_fsuid;
907 }
908
909 /*
910  * Samma på svenska..
911  */
912 asmlinkage long sys_setfsgid(gid_t gid)
913 {
914         int old_fsgid;
915
916         old_fsgid = current->fsgid;
917         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
918                 return old_fsgid;
919
920         if (gid == current->gid || gid == current->egid ||
921             gid == current->sgid || gid == current->fsgid || 
922             capable(CAP_SETGID))
923         {
924                 if (gid != old_fsgid)
925                 {
926                         current->mm->dumpable = suid_dumpable;
927                         smp_wmb();
928                 }
929                 current->fsgid = gid;
930                 key_fsgid_changed(current);
931         }
932         return old_fsgid;
933 }
934
935 asmlinkage long sys_times(struct tms __user * tbuf)
936 {
937         /*
938          *      In the SMP world we might just be unlucky and have one of
939          *      the times increment as we use it. Since the value is an
940          *      atomically safe type this is just fine. Conceptually its
941          *      as if the syscall took an instant longer to occur.
942          */
943         if (tbuf) {
944                 struct tms tmp;
945                 cputime_t utime, stime, cutime, cstime;
946
947 #ifdef CONFIG_SMP
948                 if (thread_group_empty(current)) {
949                         /*
950                          * Single thread case without the use of any locks.
951                          *
952                          * We may race with release_task if two threads are
953                          * executing. However, release task first adds up the
954                          * counters (__exit_signal) before  removing the task
955                          * from the process tasklist (__unhash_process).
956                          * __exit_signal also acquires and releases the
957                          * siglock which results in the proper memory ordering
958                          * so that the list modifications are always visible
959                          * after the counters have been updated.
960                          *
961                          * If the counters have been updated by the second thread
962                          * but the thread has not yet been removed from the list
963                          * then the other branch will be executing which will
964                          * block on tasklist_lock until the exit handling of the
965                          * other task is finished.
966                          *
967                          * This also implies that the sighand->siglock cannot
968                          * be held by another processor. So we can also
969                          * skip acquiring that lock.
970                          */
971                         utime = cputime_add(current->signal->utime, current->utime);
972                         stime = cputime_add(current->signal->utime, current->stime);
973                         cutime = current->signal->cutime;
974                         cstime = current->signal->cstime;
975                 } else
976 #endif
977                 {
978
979                         /* Process with multiple threads */
980                         struct task_struct *tsk = current;
981                         struct task_struct *t;
982
983                         read_lock(&tasklist_lock);
984                         utime = tsk->signal->utime;
985                         stime = tsk->signal->stime;
986                         t = tsk;
987                         do {
988                                 utime = cputime_add(utime, t->utime);
989                                 stime = cputime_add(stime, t->stime);
990                                 t = next_thread(t);
991                         } while (t != tsk);
992
993                         /*
994                          * While we have tasklist_lock read-locked, no dying thread
995                          * can be updating current->signal->[us]time.  Instead,
996                          * we got their counts included in the live thread loop.
997                          * However, another thread can come in right now and
998                          * do a wait call that updates current->signal->c[us]time.
999                          * To make sure we always see that pair updated atomically,
1000                          * we take the siglock around fetching them.
1001                          */
1002                         spin_lock_irq(&tsk->sighand->siglock);
1003                         cutime = tsk->signal->cutime;
1004                         cstime = tsk->signal->cstime;
1005                         spin_unlock_irq(&tsk->sighand->siglock);
1006                         read_unlock(&tasklist_lock);
1007                 }
1008                 tmp.tms_utime = cputime_to_clock_t(utime);
1009                 tmp.tms_stime = cputime_to_clock_t(stime);
1010                 tmp.tms_cutime = cputime_to_clock_t(cutime);
1011                 tmp.tms_cstime = cputime_to_clock_t(cstime);
1012                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1013                         return -EFAULT;
1014         }
1015         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1016 }
1017
1018 /*
1019  * This needs some heavy checking ...
1020  * I just haven't the stomach for it. I also don't fully
1021  * understand sessions/pgrp etc. Let somebody who does explain it.
1022  *
1023  * OK, I think I have the protection semantics right.... this is really
1024  * only important on a multi-user system anyway, to make sure one user
1025  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1026  *
1027  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1028  * LBT 04.03.94
1029  */
1030
1031 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1032 {
1033         struct task_struct *p;
1034         int err = -EINVAL;
1035
1036         if (!pid)
1037                 pid = current->pid;
1038         if (!pgid)
1039                 pgid = pid;
1040         if (pgid < 0)
1041                 return -EINVAL;
1042
1043         /* From this point forward we keep holding onto the tasklist lock
1044          * so that our parent does not change from under us. -DaveM
1045          */
1046         write_lock_irq(&tasklist_lock);
1047
1048         err = -ESRCH;
1049         p = find_task_by_pid(pid);
1050         if (!p)
1051                 goto out;
1052
1053         err = -EINVAL;
1054         if (!thread_group_leader(p))
1055                 goto out;
1056
1057         if (p->parent == current || p->real_parent == current) {
1058                 err = -EPERM;
1059                 if (p->signal->session != current->signal->session)
1060                         goto out;
1061                 err = -EACCES;
1062                 if (p->did_exec)
1063                         goto out;
1064         } else {
1065                 err = -ESRCH;
1066                 if (p != current)
1067                         goto out;
1068         }
1069
1070         err = -EPERM;
1071         if (p->signal->leader)
1072                 goto out;
1073
1074         if (pgid != pid) {
1075                 struct task_struct *p;
1076
1077                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1078                         if (p->signal->session == current->signal->session)
1079                                 goto ok_pgid;
1080                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1081                 goto out;
1082         }
1083
1084 ok_pgid:
1085         err = security_task_setpgid(p, pgid);
1086         if (err)
1087                 goto out;
1088
1089         if (process_group(p) != pgid) {
1090                 detach_pid(p, PIDTYPE_PGID);
1091                 p->signal->pgrp = pgid;
1092                 attach_pid(p, PIDTYPE_PGID, pgid);
1093         }
1094
1095         err = 0;
1096 out:
1097         /* All paths lead to here, thus we are safe. -DaveM */
1098         write_unlock_irq(&tasklist_lock);
1099         return err;
1100 }
1101
1102 asmlinkage long sys_getpgid(pid_t pid)
1103 {
1104         if (!pid) {
1105                 return process_group(current);
1106         } else {
1107                 int retval;
1108                 struct task_struct *p;
1109
1110                 read_lock(&tasklist_lock);
1111                 p = find_task_by_pid(pid);
1112
1113                 retval = -ESRCH;
1114                 if (p) {
1115                         retval = security_task_getpgid(p);
1116                         if (!retval)
1117                                 retval = process_group(p);
1118                 }
1119                 read_unlock(&tasklist_lock);
1120                 return retval;
1121         }
1122 }
1123
1124 #ifdef __ARCH_WANT_SYS_GETPGRP
1125
1126 asmlinkage long sys_getpgrp(void)
1127 {
1128         /* SMP - assuming writes are word atomic this is fine */
1129         return process_group(current);
1130 }
1131
1132 #endif
1133
1134 asmlinkage long sys_getsid(pid_t pid)
1135 {
1136         if (!pid) {
1137                 return current->signal->session;
1138         } else {
1139                 int retval;
1140                 struct task_struct *p;
1141
1142                 read_lock(&tasklist_lock);
1143                 p = find_task_by_pid(pid);
1144
1145                 retval = -ESRCH;
1146                 if(p) {
1147                         retval = security_task_getsid(p);
1148                         if (!retval)
1149                                 retval = p->signal->session;
1150                 }
1151                 read_unlock(&tasklist_lock);
1152                 return retval;
1153         }
1154 }
1155
1156 asmlinkage long sys_setsid(void)
1157 {
1158         struct pid *pid;
1159         int err = -EPERM;
1160
1161         if (!thread_group_leader(current))
1162                 return -EINVAL;
1163
1164         down(&tty_sem);
1165         write_lock_irq(&tasklist_lock);
1166
1167         pid = find_pid(PIDTYPE_PGID, current->pid);
1168         if (pid)
1169                 goto out;
1170
1171         current->signal->leader = 1;
1172         __set_special_pids(current->pid, current->pid);
1173         current->signal->tty = NULL;
1174         current->signal->tty_old_pgrp = 0;
1175         err = process_group(current);
1176 out:
1177         write_unlock_irq(&tasklist_lock);
1178         up(&tty_sem);
1179         return err;
1180 }
1181
1182 /*
1183  * Supplementary group IDs
1184  */
1185
1186 /* init to 2 - one for init_task, one to ensure it is never freed */
1187 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1188
1189 struct group_info *groups_alloc(int gidsetsize)
1190 {
1191         struct group_info *group_info;
1192         int nblocks;
1193         int i;
1194
1195         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1196         /* Make sure we always allocate at least one indirect block pointer */
1197         nblocks = nblocks ? : 1;
1198         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1199         if (!group_info)
1200                 return NULL;
1201         group_info->ngroups = gidsetsize;
1202         group_info->nblocks = nblocks;
1203         atomic_set(&group_info->usage, 1);
1204
1205         if (gidsetsize <= NGROUPS_SMALL) {
1206                 group_info->blocks[0] = group_info->small_block;
1207         } else {
1208                 for (i = 0; i < nblocks; i++) {
1209                         gid_t *b;
1210                         b = (void *)__get_free_page(GFP_USER);
1211                         if (!b)
1212                                 goto out_undo_partial_alloc;
1213                         group_info->blocks[i] = b;
1214                 }
1215         }
1216         return group_info;
1217
1218 out_undo_partial_alloc:
1219         while (--i >= 0) {
1220                 free_page((unsigned long)group_info->blocks[i]);
1221         }
1222         kfree(group_info);
1223         return NULL;
1224 }
1225
1226 EXPORT_SYMBOL(groups_alloc);
1227
1228 void groups_free(struct group_info *group_info)
1229 {
1230         if (group_info->blocks[0] != group_info->small_block) {
1231                 int i;
1232                 for (i = 0; i < group_info->nblocks; i++)
1233                         free_page((unsigned long)group_info->blocks[i]);
1234         }
1235         kfree(group_info);
1236 }
1237
1238 EXPORT_SYMBOL(groups_free);
1239
1240 /* export the group_info to a user-space array */
1241 static int groups_to_user(gid_t __user *grouplist,
1242     struct group_info *group_info)
1243 {
1244         int i;
1245         int count = group_info->ngroups;
1246
1247         for (i = 0; i < group_info->nblocks; i++) {
1248                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1249                 int off = i * NGROUPS_PER_BLOCK;
1250                 int len = cp_count * sizeof(*grouplist);
1251
1252                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1253                         return -EFAULT;
1254
1255                 count -= cp_count;
1256         }
1257         return 0;
1258 }
1259
1260 /* fill a group_info from a user-space array - it must be allocated already */
1261 static int groups_from_user(struct group_info *group_info,
1262     gid_t __user *grouplist)
1263  {
1264         int i;
1265         int count = group_info->ngroups;
1266
1267         for (i = 0; i < group_info->nblocks; i++) {
1268                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1269                 int off = i * NGROUPS_PER_BLOCK;
1270                 int len = cp_count * sizeof(*grouplist);
1271
1272                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1273                         return -EFAULT;
1274
1275                 count -= cp_count;
1276         }
1277         return 0;
1278 }
1279
1280 /* a simple Shell sort */
1281 static void groups_sort(struct group_info *group_info)
1282 {
1283         int base, max, stride;
1284         int gidsetsize = group_info->ngroups;
1285
1286         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1287                 ; /* nothing */
1288         stride /= 3;
1289
1290         while (stride) {
1291                 max = gidsetsize - stride;
1292                 for (base = 0; base < max; base++) {
1293                         int left = base;
1294                         int right = left + stride;
1295                         gid_t tmp = GROUP_AT(group_info, right);
1296
1297                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1298                                 GROUP_AT(group_info, right) =
1299                                     GROUP_AT(group_info, left);
1300                                 right = left;
1301                                 left -= stride;
1302                         }
1303                         GROUP_AT(group_info, right) = tmp;
1304                 }
1305                 stride /= 3;
1306         }
1307 }
1308
1309 /* a simple bsearch */
1310 int groups_search(struct group_info *group_info, gid_t grp)
1311 {
1312         int left, right;
1313
1314         if (!group_info)
1315                 return 0;
1316
1317         left = 0;
1318         right = group_info->ngroups;
1319         while (left < right) {
1320                 int mid = (left+right)/2;
1321                 int cmp = grp - GROUP_AT(group_info, mid);
1322                 if (cmp > 0)
1323                         left = mid + 1;
1324                 else if (cmp < 0)
1325                         right = mid;
1326                 else
1327                         return 1;
1328         }
1329         return 0;
1330 }
1331
1332 /* validate and set current->group_info */
1333 int set_current_groups(struct group_info *group_info)
1334 {
1335         int retval;
1336         struct group_info *old_info;
1337
1338         retval = security_task_setgroups(group_info);
1339         if (retval)
1340                 return retval;
1341
1342         groups_sort(group_info);
1343         get_group_info(group_info);
1344
1345         task_lock(current);
1346         old_info = current->group_info;
1347         current->group_info = group_info;
1348         task_unlock(current);
1349
1350         put_group_info(old_info);
1351
1352         return 0;
1353 }
1354
1355 EXPORT_SYMBOL(set_current_groups);
1356
1357 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1358 {
1359         int i = 0;
1360
1361         /*
1362          *      SMP: Nobody else can change our grouplist. Thus we are
1363          *      safe.
1364          */
1365
1366         if (gidsetsize < 0)
1367                 return -EINVAL;
1368
1369         /* no need to grab task_lock here; it cannot change */
1370         get_group_info(current->group_info);
1371         i = current->group_info->ngroups;
1372         if (gidsetsize) {
1373                 if (i > gidsetsize) {
1374                         i = -EINVAL;
1375                         goto out;
1376                 }
1377                 if (groups_to_user(grouplist, current->group_info)) {
1378                         i = -EFAULT;
1379                         goto out;
1380                 }
1381         }
1382 out:
1383         put_group_info(current->group_info);
1384         return i;
1385 }
1386
1387 /*
1388  *      SMP: Our groups are copy-on-write. We can set them safely
1389  *      without another task interfering.
1390  */
1391  
1392 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1393 {
1394         struct group_info *group_info;
1395         int retval;
1396
1397         if (!capable(CAP_SETGID))
1398                 return -EPERM;
1399         if ((unsigned)gidsetsize > NGROUPS_MAX)
1400                 return -EINVAL;
1401
1402         group_info = groups_alloc(gidsetsize);
1403         if (!group_info)
1404                 return -ENOMEM;
1405         retval = groups_from_user(group_info, grouplist);
1406         if (retval) {
1407                 put_group_info(group_info);
1408                 return retval;
1409         }
1410
1411         retval = set_current_groups(group_info);
1412         put_group_info(group_info);
1413
1414         return retval;
1415 }
1416
1417 /*
1418  * Check whether we're fsgid/egid or in the supplemental group..
1419  */
1420 int in_group_p(gid_t grp)
1421 {
1422         int retval = 1;
1423         if (grp != current->fsgid) {
1424                 get_group_info(current->group_info);
1425                 retval = groups_search(current->group_info, grp);
1426                 put_group_info(current->group_info);
1427         }
1428         return retval;
1429 }
1430
1431 EXPORT_SYMBOL(in_group_p);
1432
1433 int in_egroup_p(gid_t grp)
1434 {
1435         int retval = 1;
1436         if (grp != current->egid) {
1437                 get_group_info(current->group_info);
1438                 retval = groups_search(current->group_info, grp);
1439                 put_group_info(current->group_info);
1440         }
1441         return retval;
1442 }
1443
1444 EXPORT_SYMBOL(in_egroup_p);
1445
1446 DECLARE_RWSEM(uts_sem);
1447
1448 EXPORT_SYMBOL(uts_sem);
1449
1450 asmlinkage long sys_newuname(struct new_utsname __user * name)
1451 {
1452         int errno = 0;
1453
1454         down_read(&uts_sem);
1455         if (copy_to_user(name,&system_utsname,sizeof *name))
1456                 errno = -EFAULT;
1457         up_read(&uts_sem);
1458         return errno;
1459 }
1460
1461 asmlinkage long sys_sethostname(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         down_write(&uts_sem);
1471         errno = -EFAULT;
1472         if (!copy_from_user(tmp, name, len)) {
1473                 memcpy(system_utsname.nodename, tmp, len);
1474                 system_utsname.nodename[len] = 0;
1475                 errno = 0;
1476         }
1477         up_write(&uts_sem);
1478         return errno;
1479 }
1480
1481 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1482
1483 asmlinkage long sys_gethostname(char __user *name, int len)
1484 {
1485         int i, errno;
1486
1487         if (len < 0)
1488                 return -EINVAL;
1489         down_read(&uts_sem);
1490         i = 1 + strlen(system_utsname.nodename);
1491         if (i > len)
1492                 i = len;
1493         errno = 0;
1494         if (copy_to_user(name, system_utsname.nodename, i))
1495                 errno = -EFAULT;
1496         up_read(&uts_sem);
1497         return errno;
1498 }
1499
1500 #endif
1501
1502 /*
1503  * Only setdomainname; getdomainname can be implemented by calling
1504  * uname()
1505  */
1506 asmlinkage long sys_setdomainname(char __user *name, int len)
1507 {
1508         int errno;
1509         char tmp[__NEW_UTS_LEN];
1510
1511         if (!capable(CAP_SYS_ADMIN))
1512                 return -EPERM;
1513         if (len < 0 || len > __NEW_UTS_LEN)
1514                 return -EINVAL;
1515
1516         down_write(&uts_sem);
1517         errno = -EFAULT;
1518         if (!copy_from_user(tmp, name, len)) {
1519                 memcpy(system_utsname.domainname, tmp, len);
1520                 system_utsname.domainname[len] = 0;
1521                 errno = 0;
1522         }
1523         up_write(&uts_sem);
1524         return errno;
1525 }
1526
1527 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1528 {
1529         if (resource >= RLIM_NLIMITS)
1530                 return -EINVAL;
1531         else {
1532                 struct rlimit value;
1533                 task_lock(current->group_leader);
1534                 value = current->signal->rlim[resource];
1535                 task_unlock(current->group_leader);
1536                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1537         }
1538 }
1539
1540 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1541
1542 /*
1543  *      Back compatibility for getrlimit. Needed for some apps.
1544  */
1545  
1546 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1547 {
1548         struct rlimit x;
1549         if (resource >= RLIM_NLIMITS)
1550                 return -EINVAL;
1551
1552         task_lock(current->group_leader);
1553         x = current->signal->rlim[resource];
1554         task_unlock(current->group_leader);
1555         if(x.rlim_cur > 0x7FFFFFFF)
1556                 x.rlim_cur = 0x7FFFFFFF;
1557         if(x.rlim_max > 0x7FFFFFFF)
1558                 x.rlim_max = 0x7FFFFFFF;
1559         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1560 }
1561
1562 #endif
1563
1564 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1565 {
1566         struct rlimit new_rlim, *old_rlim;
1567         int retval;
1568
1569         if (resource >= RLIM_NLIMITS)
1570                 return -EINVAL;
1571         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1572                 return -EFAULT;
1573        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1574                return -EINVAL;
1575         old_rlim = current->signal->rlim + resource;
1576         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1577             !capable(CAP_SYS_RESOURCE))
1578                 return -EPERM;
1579         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1580                         return -EPERM;
1581
1582         retval = security_task_setrlimit(resource, &new_rlim);
1583         if (retval)
1584                 return retval;
1585
1586         task_lock(current->group_leader);
1587         *old_rlim = new_rlim;
1588         task_unlock(current->group_leader);
1589
1590         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1591             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1592              new_rlim.rlim_cur <= cputime_to_secs(
1593                      current->signal->it_prof_expires))) {
1594                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1595                 read_lock(&tasklist_lock);
1596                 spin_lock_irq(&current->sighand->siglock);
1597                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1598                                       &cputime, NULL);
1599                 spin_unlock_irq(&current->sighand->siglock);
1600                 read_unlock(&tasklist_lock);
1601         }
1602
1603         return 0;
1604 }
1605
1606 /*
1607  * It would make sense to put struct rusage in the task_struct,
1608  * except that would make the task_struct be *really big*.  After
1609  * task_struct gets moved into malloc'ed memory, it would
1610  * make sense to do this.  It will make moving the rest of the information
1611  * a lot simpler!  (Which we're not doing right now because we're not
1612  * measuring them yet).
1613  *
1614  * This expects to be called with tasklist_lock read-locked or better,
1615  * and the siglock not locked.  It may momentarily take the siglock.
1616  *
1617  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1618  * races with threads incrementing their own counters.  But since word
1619  * reads are atomic, we either get new values or old values and we don't
1620  * care which for the sums.  We always take the siglock to protect reading
1621  * the c* fields from p->signal from races with exit.c updating those
1622  * fields when reaping, so a sample either gets all the additions of a
1623  * given child after it's reaped, or none so this sample is before reaping.
1624  */
1625
1626 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1627 {
1628         struct task_struct *t;
1629         unsigned long flags;
1630         cputime_t utime, stime;
1631
1632         memset((char *) r, 0, sizeof *r);
1633
1634         if (unlikely(!p->signal))
1635                 return;
1636
1637         switch (who) {
1638                 case RUSAGE_CHILDREN:
1639                         spin_lock_irqsave(&p->sighand->siglock, flags);
1640                         utime = p->signal->cutime;
1641                         stime = p->signal->cstime;
1642                         r->ru_nvcsw = p->signal->cnvcsw;
1643                         r->ru_nivcsw = p->signal->cnivcsw;
1644                         r->ru_minflt = p->signal->cmin_flt;
1645                         r->ru_majflt = p->signal->cmaj_flt;
1646                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1647                         cputime_to_timeval(utime, &r->ru_utime);
1648                         cputime_to_timeval(stime, &r->ru_stime);
1649                         break;
1650                 case RUSAGE_SELF:
1651                         spin_lock_irqsave(&p->sighand->siglock, flags);
1652                         utime = stime = cputime_zero;
1653                         goto sum_group;
1654                 case RUSAGE_BOTH:
1655                         spin_lock_irqsave(&p->sighand->siglock, flags);
1656                         utime = p->signal->cutime;
1657                         stime = p->signal->cstime;
1658                         r->ru_nvcsw = p->signal->cnvcsw;
1659                         r->ru_nivcsw = p->signal->cnivcsw;
1660                         r->ru_minflt = p->signal->cmin_flt;
1661                         r->ru_majflt = p->signal->cmaj_flt;
1662                 sum_group:
1663                         utime = cputime_add(utime, p->signal->utime);
1664                         stime = cputime_add(stime, p->signal->stime);
1665                         r->ru_nvcsw += p->signal->nvcsw;
1666                         r->ru_nivcsw += p->signal->nivcsw;
1667                         r->ru_minflt += p->signal->min_flt;
1668                         r->ru_majflt += p->signal->maj_flt;
1669                         t = p;
1670                         do {
1671                                 utime = cputime_add(utime, t->utime);
1672                                 stime = cputime_add(stime, t->stime);
1673                                 r->ru_nvcsw += t->nvcsw;
1674                                 r->ru_nivcsw += t->nivcsw;
1675                                 r->ru_minflt += t->min_flt;
1676                                 r->ru_majflt += t->maj_flt;
1677                                 t = next_thread(t);
1678                         } while (t != p);
1679                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1680                         cputime_to_timeval(utime, &r->ru_utime);
1681                         cputime_to_timeval(stime, &r->ru_stime);
1682                         break;
1683                 default:
1684                         BUG();
1685         }
1686 }
1687
1688 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1689 {
1690         struct rusage r;
1691         read_lock(&tasklist_lock);
1692         k_getrusage(p, who, &r);
1693         read_unlock(&tasklist_lock);
1694         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1695 }
1696
1697 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1698 {
1699         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1700                 return -EINVAL;
1701         return getrusage(current, who, ru);
1702 }
1703
1704 asmlinkage long sys_umask(int mask)
1705 {
1706         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1707         return mask;
1708 }
1709     
1710 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1711                           unsigned long arg4, unsigned long arg5)
1712 {
1713         long error;
1714         int sig;
1715
1716         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1717         if (error)
1718                 return error;
1719
1720         switch (option) {
1721                 case PR_SET_PDEATHSIG:
1722                         sig = arg2;
1723                         if (!valid_signal(sig)) {
1724                                 error = -EINVAL;
1725                                 break;
1726                         }
1727                         current->pdeath_signal = sig;
1728                         break;
1729                 case PR_GET_PDEATHSIG:
1730                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1731                         break;
1732                 case PR_GET_DUMPABLE:
1733                         if (current->mm->dumpable)
1734                                 error = 1;
1735                         break;
1736                 case PR_SET_DUMPABLE:
1737                         if (arg2 < 0 || arg2 > 2) {
1738                                 error = -EINVAL;
1739                                 break;
1740                         }
1741                         current->mm->dumpable = arg2;
1742                         break;
1743
1744                 case PR_SET_UNALIGN:
1745                         error = SET_UNALIGN_CTL(current, arg2);
1746                         break;
1747                 case PR_GET_UNALIGN:
1748                         error = GET_UNALIGN_CTL(current, arg2);
1749                         break;
1750                 case PR_SET_FPEMU:
1751                         error = SET_FPEMU_CTL(current, arg2);
1752                         break;
1753                 case PR_GET_FPEMU:
1754                         error = GET_FPEMU_CTL(current, arg2);
1755                         break;
1756                 case PR_SET_FPEXC:
1757                         error = SET_FPEXC_CTL(current, arg2);
1758                         break;
1759                 case PR_GET_FPEXC:
1760                         error = GET_FPEXC_CTL(current, arg2);
1761                         break;
1762                 case PR_GET_TIMING:
1763                         error = PR_TIMING_STATISTICAL;
1764                         break;
1765                 case PR_SET_TIMING:
1766                         if (arg2 == PR_TIMING_STATISTICAL)
1767                                 error = 0;
1768                         else
1769                                 error = -EINVAL;
1770                         break;
1771
1772                 case PR_GET_KEEPCAPS:
1773                         if (current->keep_capabilities)
1774                                 error = 1;
1775                         break;
1776                 case PR_SET_KEEPCAPS:
1777                         if (arg2 != 0 && arg2 != 1) {
1778                                 error = -EINVAL;
1779                                 break;
1780                         }
1781                         current->keep_capabilities = arg2;
1782                         break;
1783                 case PR_SET_NAME: {
1784                         struct task_struct *me = current;
1785                         unsigned char ncomm[sizeof(me->comm)];
1786
1787                         ncomm[sizeof(me->comm)-1] = 0;
1788                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1789                                                 sizeof(me->comm)-1) < 0)
1790                                 return -EFAULT;
1791                         set_task_comm(me, ncomm);
1792                         return 0;
1793                 }
1794                 case PR_GET_NAME: {
1795                         struct task_struct *me = current;
1796                         unsigned char tcomm[sizeof(me->comm)];
1797
1798                         get_task_comm(tcomm, me);
1799                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1800                                 return -EFAULT;
1801                         return 0;
1802                 }
1803                 default:
1804                         error = -EINVAL;
1805                         break;
1806         }
1807         return error;
1808 }