[PATCH] Avoid taking global tasklist_lock for single threadedprocess at getrusage()
[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/capability.h>
23 #include <linux/device.h>
24 #include <linux/key.h>
25 #include <linux/times.h>
26 #include <linux/posix-timers.h>
27 #include <linux/security.h>
28 #include <linux/dcookies.h>
29 #include <linux/suspend.h>
30 #include <linux/tty.h>
31 #include <linux/signal.h>
32 #include <linux/cn_proc.h>
33
34 #include <linux/compat.h>
35 #include <linux/syscalls.h>
36 #include <linux/kprobes.h>
37
38 #include <asm/uaccess.h>
39 #include <asm/io.h>
40 #include <asm/unistd.h>
41
42 #ifndef SET_UNALIGN_CTL
43 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
44 #endif
45 #ifndef GET_UNALIGN_CTL
46 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
47 #endif
48 #ifndef SET_FPEMU_CTL
49 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
50 #endif
51 #ifndef GET_FPEMU_CTL
52 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
53 #endif
54 #ifndef SET_FPEXC_CTL
55 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
56 #endif
57 #ifndef GET_FPEXC_CTL
58 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
59 #endif
60
61 /*
62  * this is where the system-wide overflow UID and GID are defined, for
63  * architectures that now have 32-bit UID/GID but didn't in the past
64  */
65
66 int overflowuid = DEFAULT_OVERFLOWUID;
67 int overflowgid = DEFAULT_OVERFLOWGID;
68
69 #ifdef CONFIG_UID16
70 EXPORT_SYMBOL(overflowuid);
71 EXPORT_SYMBOL(overflowgid);
72 #endif
73
74 /*
75  * the same as above, but for filesystems which can only store a 16-bit
76  * UID and GID. as such, this is needed on all architectures
77  */
78
79 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
80 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
81
82 EXPORT_SYMBOL(fs_overflowuid);
83 EXPORT_SYMBOL(fs_overflowgid);
84
85 /*
86  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
87  */
88
89 int C_A_D = 1;
90 int cad_pid = 1;
91
92 /*
93  *      Notifier list for kernel code which wants to be called
94  *      at shutdown. This is used to stop any idling DMA operations
95  *      and the like. 
96  */
97
98 static struct notifier_block *reboot_notifier_list;
99 static DEFINE_RWLOCK(notifier_lock);
100
101 /**
102  *      notifier_chain_register - Add notifier to a notifier chain
103  *      @list: Pointer to root list pointer
104  *      @n: New entry in notifier chain
105  *
106  *      Adds a notifier to a notifier chain.
107  *
108  *      Currently always returns zero.
109  */
110  
111 int notifier_chain_register(struct notifier_block **list, struct notifier_block *n)
112 {
113         write_lock(&notifier_lock);
114         while(*list)
115         {
116                 if(n->priority > (*list)->priority)
117                         break;
118                 list= &((*list)->next);
119         }
120         n->next = *list;
121         *list=n;
122         write_unlock(&notifier_lock);
123         return 0;
124 }
125
126 EXPORT_SYMBOL(notifier_chain_register);
127
128 /**
129  *      notifier_chain_unregister - Remove notifier from a notifier chain
130  *      @nl: Pointer to root list pointer
131  *      @n: New entry in notifier chain
132  *
133  *      Removes a notifier from a notifier chain.
134  *
135  *      Returns zero on success, or %-ENOENT on failure.
136  */
137  
138 int notifier_chain_unregister(struct notifier_block **nl, struct notifier_block *n)
139 {
140         write_lock(&notifier_lock);
141         while((*nl)!=NULL)
142         {
143                 if((*nl)==n)
144                 {
145                         *nl=n->next;
146                         write_unlock(&notifier_lock);
147                         return 0;
148                 }
149                 nl=&((*nl)->next);
150         }
151         write_unlock(&notifier_lock);
152         return -ENOENT;
153 }
154
155 EXPORT_SYMBOL(notifier_chain_unregister);
156
157 /**
158  *      notifier_call_chain - Call functions in a notifier chain
159  *      @n: Pointer to root pointer of notifier chain
160  *      @val: Value passed unmodified to notifier function
161  *      @v: Pointer passed unmodified to notifier function
162  *
163  *      Calls each function in a notifier chain in turn.
164  *
165  *      If the return value of the notifier can be and'd
166  *      with %NOTIFY_STOP_MASK, then notifier_call_chain
167  *      will return immediately, with the return value of
168  *      the notifier function which halted execution.
169  *      Otherwise, the return value is the return value
170  *      of the last notifier function called.
171  */
172  
173 int __kprobes notifier_call_chain(struct notifier_block **n, unsigned long val, void *v)
174 {
175         int ret=NOTIFY_DONE;
176         struct notifier_block *nb = *n;
177
178         while(nb)
179         {
180                 ret=nb->notifier_call(nb,val,v);
181                 if(ret&NOTIFY_STOP_MASK)
182                 {
183                         return ret;
184                 }
185                 nb=nb->next;
186         }
187         return ret;
188 }
189
190 EXPORT_SYMBOL(notifier_call_chain);
191
192 /**
193  *      register_reboot_notifier - Register function to be called at reboot time
194  *      @nb: Info about notifier function to be called
195  *
196  *      Registers a function with the list of functions
197  *      to be called at reboot time.
198  *
199  *      Currently always returns zero, as notifier_chain_register
200  *      always returns zero.
201  */
202  
203 int register_reboot_notifier(struct notifier_block * nb)
204 {
205         return notifier_chain_register(&reboot_notifier_list, nb);
206 }
207
208 EXPORT_SYMBOL(register_reboot_notifier);
209
210 /**
211  *      unregister_reboot_notifier - Unregister previously registered reboot notifier
212  *      @nb: Hook to be unregistered
213  *
214  *      Unregisters a previously registered reboot
215  *      notifier function.
216  *
217  *      Returns zero on success, or %-ENOENT on failure.
218  */
219  
220 int unregister_reboot_notifier(struct notifier_block * nb)
221 {
222         return notifier_chain_unregister(&reboot_notifier_list, nb);
223 }
224
225 EXPORT_SYMBOL(unregister_reboot_notifier);
226
227 #ifndef CONFIG_SECURITY
228 int capable(int cap)
229 {
230         if (cap_raised(current->cap_effective, cap)) {
231                current->flags |= PF_SUPERPRIV;
232                return 1;
233         }
234         return 0;
235 }
236 EXPORT_SYMBOL(capable);
237 #endif
238
239 static int set_one_prio(struct task_struct *p, int niceval, int error)
240 {
241         int no_nice;
242
243         if (p->uid != current->euid &&
244                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
245                 error = -EPERM;
246                 goto out;
247         }
248         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
249                 error = -EACCES;
250                 goto out;
251         }
252         no_nice = security_task_setnice(p, niceval);
253         if (no_nice) {
254                 error = no_nice;
255                 goto out;
256         }
257         if (error == -ESRCH)
258                 error = 0;
259         set_user_nice(p, niceval);
260 out:
261         return error;
262 }
263
264 asmlinkage long sys_setpriority(int which, int who, int niceval)
265 {
266         struct task_struct *g, *p;
267         struct user_struct *user;
268         int error = -EINVAL;
269
270         if (which > 2 || which < 0)
271                 goto out;
272
273         /* normalize: avoid signed division (rounding problems) */
274         error = -ESRCH;
275         if (niceval < -20)
276                 niceval = -20;
277         if (niceval > 19)
278                 niceval = 19;
279
280         read_lock(&tasklist_lock);
281         switch (which) {
282                 case PRIO_PROCESS:
283                         if (!who)
284                                 who = current->pid;
285                         p = find_task_by_pid(who);
286                         if (p)
287                                 error = set_one_prio(p, niceval, error);
288                         break;
289                 case PRIO_PGRP:
290                         if (!who)
291                                 who = process_group(current);
292                         do_each_task_pid(who, PIDTYPE_PGID, p) {
293                                 error = set_one_prio(p, niceval, error);
294                         } while_each_task_pid(who, PIDTYPE_PGID, p);
295                         break;
296                 case PRIO_USER:
297                         user = current->user;
298                         if (!who)
299                                 who = current->uid;
300                         else
301                                 if ((who != current->uid) && !(user = find_user(who)))
302                                         goto out_unlock;        /* No processes for this user */
303
304                         do_each_thread(g, p)
305                                 if (p->uid == who)
306                                         error = set_one_prio(p, niceval, error);
307                         while_each_thread(g, p);
308                         if (who != current->uid)
309                                 free_uid(user);         /* For find_user() */
310                         break;
311         }
312 out_unlock:
313         read_unlock(&tasklist_lock);
314 out:
315         return error;
316 }
317
318 /*
319  * Ugh. To avoid negative return values, "getpriority()" will
320  * not return the normal nice-value, but a negated value that
321  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
322  * to stay compatible.
323  */
324 asmlinkage long sys_getpriority(int which, int who)
325 {
326         struct task_struct *g, *p;
327         struct user_struct *user;
328         long niceval, retval = -ESRCH;
329
330         if (which > 2 || which < 0)
331                 return -EINVAL;
332
333         read_lock(&tasklist_lock);
334         switch (which) {
335                 case PRIO_PROCESS:
336                         if (!who)
337                                 who = current->pid;
338                         p = find_task_by_pid(who);
339                         if (p) {
340                                 niceval = 20 - task_nice(p);
341                                 if (niceval > retval)
342                                         retval = niceval;
343                         }
344                         break;
345                 case PRIO_PGRP:
346                         if (!who)
347                                 who = process_group(current);
348                         do_each_task_pid(who, PIDTYPE_PGID, p) {
349                                 niceval = 20 - task_nice(p);
350                                 if (niceval > retval)
351                                         retval = niceval;
352                         } while_each_task_pid(who, PIDTYPE_PGID, p);
353                         break;
354                 case PRIO_USER:
355                         user = current->user;
356                         if (!who)
357                                 who = current->uid;
358                         else
359                                 if ((who != current->uid) && !(user = find_user(who)))
360                                         goto out_unlock;        /* No processes for this user */
361
362                         do_each_thread(g, p)
363                                 if (p->uid == who) {
364                                         niceval = 20 - task_nice(p);
365                                         if (niceval > retval)
366                                                 retval = niceval;
367                                 }
368                         while_each_thread(g, p);
369                         if (who != current->uid)
370                                 free_uid(user);         /* for find_user() */
371                         break;
372         }
373 out_unlock:
374         read_unlock(&tasklist_lock);
375
376         return retval;
377 }
378
379 /**
380  *      emergency_restart - reboot the system
381  *
382  *      Without shutting down any hardware or taking any locks
383  *      reboot the system.  This is called when we know we are in
384  *      trouble so this is our best effort to reboot.  This is
385  *      safe to call in interrupt context.
386  */
387 void emergency_restart(void)
388 {
389         machine_emergency_restart();
390 }
391 EXPORT_SYMBOL_GPL(emergency_restart);
392
393 void kernel_restart_prepare(char *cmd)
394 {
395         notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
396         system_state = SYSTEM_RESTART;
397         device_shutdown();
398 }
399
400 /**
401  *      kernel_restart - reboot the system
402  *      @cmd: pointer to buffer containing command to execute for restart
403  *              or %NULL
404  *
405  *      Shutdown everything and perform a clean reboot.
406  *      This is not safe to call in interrupt context.
407  */
408 void kernel_restart(char *cmd)
409 {
410         kernel_restart_prepare(cmd);
411         if (!cmd) {
412                 printk(KERN_EMERG "Restarting system.\n");
413         } else {
414                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
415         }
416         printk(".\n");
417         machine_restart(cmd);
418 }
419 EXPORT_SYMBOL_GPL(kernel_restart);
420
421 /**
422  *      kernel_kexec - reboot the system
423  *
424  *      Move into place and start executing a preloaded standalone
425  *      executable.  If nothing was preloaded return an error.
426  */
427 void kernel_kexec(void)
428 {
429 #ifdef CONFIG_KEXEC
430         struct kimage *image;
431         image = xchg(&kexec_image, NULL);
432         if (!image) {
433                 return;
434         }
435         kernel_restart_prepare(NULL);
436         printk(KERN_EMERG "Starting new kernel\n");
437         machine_shutdown();
438         machine_kexec(image);
439 #endif
440 }
441 EXPORT_SYMBOL_GPL(kernel_kexec);
442
443 void kernel_shutdown_prepare(enum system_states state)
444 {
445         notifier_call_chain(&reboot_notifier_list,
446                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
447         system_state = state;
448         device_shutdown();
449 }
450 /**
451  *      kernel_halt - halt the system
452  *
453  *      Shutdown everything and perform a clean system halt.
454  */
455 void kernel_halt(void)
456 {
457         kernel_shutdown_prepare(SYSTEM_HALT);
458         printk(KERN_EMERG "System halted.\n");
459         machine_halt();
460 }
461
462 EXPORT_SYMBOL_GPL(kernel_halt);
463
464 /**
465  *      kernel_power_off - power_off the system
466  *
467  *      Shutdown everything and perform a clean system power_off.
468  */
469 void kernel_power_off(void)
470 {
471         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
472         printk(KERN_EMERG "Power down.\n");
473         machine_power_off();
474 }
475 EXPORT_SYMBOL_GPL(kernel_power_off);
476 /*
477  * Reboot system call: for obvious reasons only root may call it,
478  * and even root needs to set up some magic numbers in the registers
479  * so that some mistake won't make this reboot the whole machine.
480  * You can also set the meaning of the ctrl-alt-del-key here.
481  *
482  * reboot doesn't sync: do that yourself before calling this.
483  */
484 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
485 {
486         char buffer[256];
487
488         /* We only trust the superuser with rebooting the system. */
489         if (!capable(CAP_SYS_BOOT))
490                 return -EPERM;
491
492         /* For safety, we require "magic" arguments. */
493         if (magic1 != LINUX_REBOOT_MAGIC1 ||
494             (magic2 != LINUX_REBOOT_MAGIC2 &&
495                         magic2 != LINUX_REBOOT_MAGIC2A &&
496                         magic2 != LINUX_REBOOT_MAGIC2B &&
497                         magic2 != LINUX_REBOOT_MAGIC2C))
498                 return -EINVAL;
499
500         /* Instead of trying to make the power_off code look like
501          * halt when pm_power_off is not set do it the easy way.
502          */
503         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
504                 cmd = LINUX_REBOOT_CMD_HALT;
505
506         lock_kernel();
507         switch (cmd) {
508         case LINUX_REBOOT_CMD_RESTART:
509                 kernel_restart(NULL);
510                 break;
511
512         case LINUX_REBOOT_CMD_CAD_ON:
513                 C_A_D = 1;
514                 break;
515
516         case LINUX_REBOOT_CMD_CAD_OFF:
517                 C_A_D = 0;
518                 break;
519
520         case LINUX_REBOOT_CMD_HALT:
521                 kernel_halt();
522                 unlock_kernel();
523                 do_exit(0);
524                 break;
525
526         case LINUX_REBOOT_CMD_POWER_OFF:
527                 kernel_power_off();
528                 unlock_kernel();
529                 do_exit(0);
530                 break;
531
532         case LINUX_REBOOT_CMD_RESTART2:
533                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
534                         unlock_kernel();
535                         return -EFAULT;
536                 }
537                 buffer[sizeof(buffer) - 1] = '\0';
538
539                 kernel_restart(buffer);
540                 break;
541
542         case LINUX_REBOOT_CMD_KEXEC:
543                 kernel_kexec();
544                 unlock_kernel();
545                 return -EINVAL;
546
547 #ifdef CONFIG_SOFTWARE_SUSPEND
548         case LINUX_REBOOT_CMD_SW_SUSPEND:
549                 {
550                         int ret = software_suspend();
551                         unlock_kernel();
552                         return ret;
553                 }
554 #endif
555
556         default:
557                 unlock_kernel();
558                 return -EINVAL;
559         }
560         unlock_kernel();
561         return 0;
562 }
563
564 static void deferred_cad(void *dummy)
565 {
566         kernel_restart(NULL);
567 }
568
569 /*
570  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
571  * As it's called within an interrupt, it may NOT sync: the only choice
572  * is whether to reboot at once, or just ignore the ctrl-alt-del.
573  */
574 void ctrl_alt_del(void)
575 {
576         static DECLARE_WORK(cad_work, deferred_cad, NULL);
577
578         if (C_A_D)
579                 schedule_work(&cad_work);
580         else
581                 kill_proc(cad_pid, SIGINT, 1);
582 }
583         
584
585 /*
586  * Unprivileged users may change the real gid to the effective gid
587  * or vice versa.  (BSD-style)
588  *
589  * If you set the real gid at all, or set the effective gid to a value not
590  * equal to the real gid, then the saved gid is set to the new effective gid.
591  *
592  * This makes it possible for a setgid program to completely drop its
593  * privileges, which is often a useful assertion to make when you are doing
594  * a security audit over a program.
595  *
596  * The general idea is that a program which uses just setregid() will be
597  * 100% compatible with BSD.  A program which uses just setgid() will be
598  * 100% compatible with POSIX with saved IDs. 
599  *
600  * SMP: There are not races, the GIDs are checked only by filesystem
601  *      operations (as far as semantic preservation is concerned).
602  */
603 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
604 {
605         int old_rgid = current->gid;
606         int old_egid = current->egid;
607         int new_rgid = old_rgid;
608         int new_egid = old_egid;
609         int retval;
610
611         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
612         if (retval)
613                 return retval;
614
615         if (rgid != (gid_t) -1) {
616                 if ((old_rgid == rgid) ||
617                     (current->egid==rgid) ||
618                     capable(CAP_SETGID))
619                         new_rgid = rgid;
620                 else
621                         return -EPERM;
622         }
623         if (egid != (gid_t) -1) {
624                 if ((old_rgid == egid) ||
625                     (current->egid == egid) ||
626                     (current->sgid == egid) ||
627                     capable(CAP_SETGID))
628                         new_egid = egid;
629                 else {
630                         return -EPERM;
631                 }
632         }
633         if (new_egid != old_egid)
634         {
635                 current->mm->dumpable = suid_dumpable;
636                 smp_wmb();
637         }
638         if (rgid != (gid_t) -1 ||
639             (egid != (gid_t) -1 && egid != old_rgid))
640                 current->sgid = new_egid;
641         current->fsgid = new_egid;
642         current->egid = new_egid;
643         current->gid = new_rgid;
644         key_fsgid_changed(current);
645         proc_id_connector(current, PROC_EVENT_GID);
646         return 0;
647 }
648
649 /*
650  * setgid() is implemented like SysV w/ SAVED_IDS 
651  *
652  * SMP: Same implicit races as above.
653  */
654 asmlinkage long sys_setgid(gid_t gid)
655 {
656         int old_egid = current->egid;
657         int retval;
658
659         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
660         if (retval)
661                 return retval;
662
663         if (capable(CAP_SETGID))
664         {
665                 if(old_egid != gid)
666                 {
667                         current->mm->dumpable = suid_dumpable;
668                         smp_wmb();
669                 }
670                 current->gid = current->egid = current->sgid = current->fsgid = gid;
671         }
672         else if ((gid == current->gid) || (gid == current->sgid))
673         {
674                 if(old_egid != gid)
675                 {
676                         current->mm->dumpable = suid_dumpable;
677                         smp_wmb();
678                 }
679                 current->egid = current->fsgid = gid;
680         }
681         else
682                 return -EPERM;
683
684         key_fsgid_changed(current);
685         proc_id_connector(current, PROC_EVENT_GID);
686         return 0;
687 }
688   
689 static int set_user(uid_t new_ruid, int dumpclear)
690 {
691         struct user_struct *new_user;
692
693         new_user = alloc_uid(new_ruid);
694         if (!new_user)
695                 return -EAGAIN;
696
697         if (atomic_read(&new_user->processes) >=
698                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
699                         new_user != &root_user) {
700                 free_uid(new_user);
701                 return -EAGAIN;
702         }
703
704         switch_uid(new_user);
705
706         if(dumpclear)
707         {
708                 current->mm->dumpable = suid_dumpable;
709                 smp_wmb();
710         }
711         current->uid = new_ruid;
712         return 0;
713 }
714
715 /*
716  * Unprivileged users may change the real uid to the effective uid
717  * or vice versa.  (BSD-style)
718  *
719  * If you set the real uid at all, or set the effective uid to a value not
720  * equal to the real uid, then the saved uid is set to the new effective uid.
721  *
722  * This makes it possible for a setuid program to completely drop its
723  * privileges, which is often a useful assertion to make when you are doing
724  * a security audit over a program.
725  *
726  * The general idea is that a program which uses just setreuid() will be
727  * 100% compatible with BSD.  A program which uses just setuid() will be
728  * 100% compatible with POSIX with saved IDs. 
729  */
730 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
731 {
732         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
733         int retval;
734
735         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
736         if (retval)
737                 return retval;
738
739         new_ruid = old_ruid = current->uid;
740         new_euid = old_euid = current->euid;
741         old_suid = current->suid;
742
743         if (ruid != (uid_t) -1) {
744                 new_ruid = ruid;
745                 if ((old_ruid != ruid) &&
746                     (current->euid != ruid) &&
747                     !capable(CAP_SETUID))
748                         return -EPERM;
749         }
750
751         if (euid != (uid_t) -1) {
752                 new_euid = euid;
753                 if ((old_ruid != euid) &&
754                     (current->euid != euid) &&
755                     (current->suid != euid) &&
756                     !capable(CAP_SETUID))
757                         return -EPERM;
758         }
759
760         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
761                 return -EAGAIN;
762
763         if (new_euid != old_euid)
764         {
765                 current->mm->dumpable = suid_dumpable;
766                 smp_wmb();
767         }
768         current->fsuid = current->euid = new_euid;
769         if (ruid != (uid_t) -1 ||
770             (euid != (uid_t) -1 && euid != old_ruid))
771                 current->suid = current->euid;
772         current->fsuid = current->euid;
773
774         key_fsuid_changed(current);
775         proc_id_connector(current, PROC_EVENT_UID);
776
777         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
778 }
779
780
781                 
782 /*
783  * setuid() is implemented like SysV with SAVED_IDS 
784  * 
785  * Note that SAVED_ID's is deficient in that a setuid root program
786  * like sendmail, for example, cannot set its uid to be a normal 
787  * user and then switch back, because if you're root, setuid() sets
788  * the saved uid too.  If you don't like this, blame the bright people
789  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
790  * will allow a root program to temporarily drop privileges and be able to
791  * regain them by swapping the real and effective uid.  
792  */
793 asmlinkage long sys_setuid(uid_t uid)
794 {
795         int old_euid = current->euid;
796         int old_ruid, old_suid, new_ruid, new_suid;
797         int retval;
798
799         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
800         if (retval)
801                 return retval;
802
803         old_ruid = new_ruid = current->uid;
804         old_suid = current->suid;
805         new_suid = old_suid;
806         
807         if (capable(CAP_SETUID)) {
808                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
809                         return -EAGAIN;
810                 new_suid = uid;
811         } else if ((uid != current->uid) && (uid != new_suid))
812                 return -EPERM;
813
814         if (old_euid != uid)
815         {
816                 current->mm->dumpable = suid_dumpable;
817                 smp_wmb();
818         }
819         current->fsuid = current->euid = uid;
820         current->suid = new_suid;
821
822         key_fsuid_changed(current);
823         proc_id_connector(current, PROC_EVENT_UID);
824
825         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
826 }
827
828
829 /*
830  * This function implements a generic ability to update ruid, euid,
831  * and suid.  This allows you to implement the 4.4 compatible seteuid().
832  */
833 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
834 {
835         int old_ruid = current->uid;
836         int old_euid = current->euid;
837         int old_suid = current->suid;
838         int retval;
839
840         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
841         if (retval)
842                 return retval;
843
844         if (!capable(CAP_SETUID)) {
845                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
846                     (ruid != current->euid) && (ruid != current->suid))
847                         return -EPERM;
848                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
849                     (euid != current->euid) && (euid != current->suid))
850                         return -EPERM;
851                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
852                     (suid != current->euid) && (suid != current->suid))
853                         return -EPERM;
854         }
855         if (ruid != (uid_t) -1) {
856                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
857                         return -EAGAIN;
858         }
859         if (euid != (uid_t) -1) {
860                 if (euid != current->euid)
861                 {
862                         current->mm->dumpable = suid_dumpable;
863                         smp_wmb();
864                 }
865                 current->euid = euid;
866         }
867         current->fsuid = current->euid;
868         if (suid != (uid_t) -1)
869                 current->suid = suid;
870
871         key_fsuid_changed(current);
872         proc_id_connector(current, PROC_EVENT_UID);
873
874         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
875 }
876
877 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
878 {
879         int retval;
880
881         if (!(retval = put_user(current->uid, ruid)) &&
882             !(retval = put_user(current->euid, euid)))
883                 retval = put_user(current->suid, suid);
884
885         return retval;
886 }
887
888 /*
889  * Same as above, but for rgid, egid, sgid.
890  */
891 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
892 {
893         int retval;
894
895         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
896         if (retval)
897                 return retval;
898
899         if (!capable(CAP_SETGID)) {
900                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
901                     (rgid != current->egid) && (rgid != current->sgid))
902                         return -EPERM;
903                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
904                     (egid != current->egid) && (egid != current->sgid))
905                         return -EPERM;
906                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
907                     (sgid != current->egid) && (sgid != current->sgid))
908                         return -EPERM;
909         }
910         if (egid != (gid_t) -1) {
911                 if (egid != current->egid)
912                 {
913                         current->mm->dumpable = suid_dumpable;
914                         smp_wmb();
915                 }
916                 current->egid = egid;
917         }
918         current->fsgid = current->egid;
919         if (rgid != (gid_t) -1)
920                 current->gid = rgid;
921         if (sgid != (gid_t) -1)
922                 current->sgid = sgid;
923
924         key_fsgid_changed(current);
925         proc_id_connector(current, PROC_EVENT_GID);
926         return 0;
927 }
928
929 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
930 {
931         int retval;
932
933         if (!(retval = put_user(current->gid, rgid)) &&
934             !(retval = put_user(current->egid, egid)))
935                 retval = put_user(current->sgid, sgid);
936
937         return retval;
938 }
939
940
941 /*
942  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
943  * is used for "access()" and for the NFS daemon (letting nfsd stay at
944  * whatever uid it wants to). It normally shadows "euid", except when
945  * explicitly set by setfsuid() or for access..
946  */
947 asmlinkage long sys_setfsuid(uid_t uid)
948 {
949         int old_fsuid;
950
951         old_fsuid = current->fsuid;
952         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
953                 return old_fsuid;
954
955         if (uid == current->uid || uid == current->euid ||
956             uid == current->suid || uid == current->fsuid || 
957             capable(CAP_SETUID))
958         {
959                 if (uid != old_fsuid)
960                 {
961                         current->mm->dumpable = suid_dumpable;
962                         smp_wmb();
963                 }
964                 current->fsuid = uid;
965         }
966
967         key_fsuid_changed(current);
968         proc_id_connector(current, PROC_EVENT_UID);
969
970         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
971
972         return old_fsuid;
973 }
974
975 /*
976  * Samma på svenska..
977  */
978 asmlinkage long sys_setfsgid(gid_t gid)
979 {
980         int old_fsgid;
981
982         old_fsgid = current->fsgid;
983         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
984                 return old_fsgid;
985
986         if (gid == current->gid || gid == current->egid ||
987             gid == current->sgid || gid == current->fsgid || 
988             capable(CAP_SETGID))
989         {
990                 if (gid != old_fsgid)
991                 {
992                         current->mm->dumpable = suid_dumpable;
993                         smp_wmb();
994                 }
995                 current->fsgid = gid;
996                 key_fsgid_changed(current);
997                 proc_id_connector(current, PROC_EVENT_GID);
998         }
999         return old_fsgid;
1000 }
1001
1002 asmlinkage long sys_times(struct tms __user * tbuf)
1003 {
1004         /*
1005          *      In the SMP world we might just be unlucky and have one of
1006          *      the times increment as we use it. Since the value is an
1007          *      atomically safe type this is just fine. Conceptually its
1008          *      as if the syscall took an instant longer to occur.
1009          */
1010         if (tbuf) {
1011                 struct tms tmp;
1012                 cputime_t utime, stime, cutime, cstime;
1013
1014 #ifdef CONFIG_SMP
1015                 if (thread_group_empty(current)) {
1016                         /*
1017                          * Single thread case without the use of any locks.
1018                          *
1019                          * We may race with release_task if two threads are
1020                          * executing. However, release task first adds up the
1021                          * counters (__exit_signal) before  removing the task
1022                          * from the process tasklist (__unhash_process).
1023                          * __exit_signal also acquires and releases the
1024                          * siglock which results in the proper memory ordering
1025                          * so that the list modifications are always visible
1026                          * after the counters have been updated.
1027                          *
1028                          * If the counters have been updated by the second thread
1029                          * but the thread has not yet been removed from the list
1030                          * then the other branch will be executing which will
1031                          * block on tasklist_lock until the exit handling of the
1032                          * other task is finished.
1033                          *
1034                          * This also implies that the sighand->siglock cannot
1035                          * be held by another processor. So we can also
1036                          * skip acquiring that lock.
1037                          */
1038                         utime = cputime_add(current->signal->utime, current->utime);
1039                         stime = cputime_add(current->signal->utime, current->stime);
1040                         cutime = current->signal->cutime;
1041                         cstime = current->signal->cstime;
1042                 } else
1043 #endif
1044                 {
1045
1046                         /* Process with multiple threads */
1047                         struct task_struct *tsk = current;
1048                         struct task_struct *t;
1049
1050                         read_lock(&tasklist_lock);
1051                         utime = tsk->signal->utime;
1052                         stime = tsk->signal->stime;
1053                         t = tsk;
1054                         do {
1055                                 utime = cputime_add(utime, t->utime);
1056                                 stime = cputime_add(stime, t->stime);
1057                                 t = next_thread(t);
1058                         } while (t != tsk);
1059
1060                         /*
1061                          * While we have tasklist_lock read-locked, no dying thread
1062                          * can be updating current->signal->[us]time.  Instead,
1063                          * we got their counts included in the live thread loop.
1064                          * However, another thread can come in right now and
1065                          * do a wait call that updates current->signal->c[us]time.
1066                          * To make sure we always see that pair updated atomically,
1067                          * we take the siglock around fetching them.
1068                          */
1069                         spin_lock_irq(&tsk->sighand->siglock);
1070                         cutime = tsk->signal->cutime;
1071                         cstime = tsk->signal->cstime;
1072                         spin_unlock_irq(&tsk->sighand->siglock);
1073                         read_unlock(&tasklist_lock);
1074                 }
1075                 tmp.tms_utime = cputime_to_clock_t(utime);
1076                 tmp.tms_stime = cputime_to_clock_t(stime);
1077                 tmp.tms_cutime = cputime_to_clock_t(cutime);
1078                 tmp.tms_cstime = cputime_to_clock_t(cstime);
1079                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1080                         return -EFAULT;
1081         }
1082         return (long) jiffies_64_to_clock_t(get_jiffies_64());
1083 }
1084
1085 /*
1086  * This needs some heavy checking ...
1087  * I just haven't the stomach for it. I also don't fully
1088  * understand sessions/pgrp etc. Let somebody who does explain it.
1089  *
1090  * OK, I think I have the protection semantics right.... this is really
1091  * only important on a multi-user system anyway, to make sure one user
1092  * can't send a signal to a process owned by another.  -TYT, 12/12/91
1093  *
1094  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1095  * LBT 04.03.94
1096  */
1097
1098 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
1099 {
1100         struct task_struct *p;
1101         struct task_struct *group_leader = current->group_leader;
1102         int err = -EINVAL;
1103
1104         if (!pid)
1105                 pid = group_leader->pid;
1106         if (!pgid)
1107                 pgid = pid;
1108         if (pgid < 0)
1109                 return -EINVAL;
1110
1111         /* From this point forward we keep holding onto the tasklist lock
1112          * so that our parent does not change from under us. -DaveM
1113          */
1114         write_lock_irq(&tasklist_lock);
1115
1116         err = -ESRCH;
1117         p = find_task_by_pid(pid);
1118         if (!p)
1119                 goto out;
1120
1121         err = -EINVAL;
1122         if (!thread_group_leader(p))
1123                 goto out;
1124
1125         if (p->real_parent == group_leader) {
1126                 err = -EPERM;
1127                 if (p->signal->session != group_leader->signal->session)
1128                         goto out;
1129                 err = -EACCES;
1130                 if (p->did_exec)
1131                         goto out;
1132         } else {
1133                 err = -ESRCH;
1134                 if (p != group_leader)
1135                         goto out;
1136         }
1137
1138         err = -EPERM;
1139         if (p->signal->leader)
1140                 goto out;
1141
1142         if (pgid != pid) {
1143                 struct task_struct *p;
1144
1145                 do_each_task_pid(pgid, PIDTYPE_PGID, p) {
1146                         if (p->signal->session == group_leader->signal->session)
1147                                 goto ok_pgid;
1148                 } while_each_task_pid(pgid, PIDTYPE_PGID, p);
1149                 goto out;
1150         }
1151
1152 ok_pgid:
1153         err = security_task_setpgid(p, pgid);
1154         if (err)
1155                 goto out;
1156
1157         if (process_group(p) != pgid) {
1158                 detach_pid(p, PIDTYPE_PGID);
1159                 p->signal->pgrp = pgid;
1160                 attach_pid(p, PIDTYPE_PGID, pgid);
1161         }
1162
1163         err = 0;
1164 out:
1165         /* All paths lead to here, thus we are safe. -DaveM */
1166         write_unlock_irq(&tasklist_lock);
1167         return err;
1168 }
1169
1170 asmlinkage long sys_getpgid(pid_t pid)
1171 {
1172         if (!pid) {
1173                 return process_group(current);
1174         } else {
1175                 int retval;
1176                 struct task_struct *p;
1177
1178                 read_lock(&tasklist_lock);
1179                 p = find_task_by_pid(pid);
1180
1181                 retval = -ESRCH;
1182                 if (p) {
1183                         retval = security_task_getpgid(p);
1184                         if (!retval)
1185                                 retval = process_group(p);
1186                 }
1187                 read_unlock(&tasklist_lock);
1188                 return retval;
1189         }
1190 }
1191
1192 #ifdef __ARCH_WANT_SYS_GETPGRP
1193
1194 asmlinkage long sys_getpgrp(void)
1195 {
1196         /* SMP - assuming writes are word atomic this is fine */
1197         return process_group(current);
1198 }
1199
1200 #endif
1201
1202 asmlinkage long sys_getsid(pid_t pid)
1203 {
1204         if (!pid) {
1205                 return current->signal->session;
1206         } else {
1207                 int retval;
1208                 struct task_struct *p;
1209
1210                 read_lock(&tasklist_lock);
1211                 p = find_task_by_pid(pid);
1212
1213                 retval = -ESRCH;
1214                 if(p) {
1215                         retval = security_task_getsid(p);
1216                         if (!retval)
1217                                 retval = p->signal->session;
1218                 }
1219                 read_unlock(&tasklist_lock);
1220                 return retval;
1221         }
1222 }
1223
1224 asmlinkage long sys_setsid(void)
1225 {
1226         struct task_struct *group_leader = current->group_leader;
1227         struct pid *pid;
1228         int err = -EPERM;
1229
1230         down(&tty_sem);
1231         write_lock_irq(&tasklist_lock);
1232
1233         pid = find_pid(PIDTYPE_PGID, group_leader->pid);
1234         if (pid)
1235                 goto out;
1236
1237         group_leader->signal->leader = 1;
1238         __set_special_pids(group_leader->pid, group_leader->pid);
1239         group_leader->signal->tty = NULL;
1240         group_leader->signal->tty_old_pgrp = 0;
1241         err = process_group(group_leader);
1242 out:
1243         write_unlock_irq(&tasklist_lock);
1244         up(&tty_sem);
1245         return err;
1246 }
1247
1248 /*
1249  * Supplementary group IDs
1250  */
1251
1252 /* init to 2 - one for init_task, one to ensure it is never freed */
1253 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1254
1255 struct group_info *groups_alloc(int gidsetsize)
1256 {
1257         struct group_info *group_info;
1258         int nblocks;
1259         int i;
1260
1261         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1262         /* Make sure we always allocate at least one indirect block pointer */
1263         nblocks = nblocks ? : 1;
1264         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1265         if (!group_info)
1266                 return NULL;
1267         group_info->ngroups = gidsetsize;
1268         group_info->nblocks = nblocks;
1269         atomic_set(&group_info->usage, 1);
1270
1271         if (gidsetsize <= NGROUPS_SMALL) {
1272                 group_info->blocks[0] = group_info->small_block;
1273         } else {
1274                 for (i = 0; i < nblocks; i++) {
1275                         gid_t *b;
1276                         b = (void *)__get_free_page(GFP_USER);
1277                         if (!b)
1278                                 goto out_undo_partial_alloc;
1279                         group_info->blocks[i] = b;
1280                 }
1281         }
1282         return group_info;
1283
1284 out_undo_partial_alloc:
1285         while (--i >= 0) {
1286                 free_page((unsigned long)group_info->blocks[i]);
1287         }
1288         kfree(group_info);
1289         return NULL;
1290 }
1291
1292 EXPORT_SYMBOL(groups_alloc);
1293
1294 void groups_free(struct group_info *group_info)
1295 {
1296         if (group_info->blocks[0] != group_info->small_block) {
1297                 int i;
1298                 for (i = 0; i < group_info->nblocks; i++)
1299                         free_page((unsigned long)group_info->blocks[i]);
1300         }
1301         kfree(group_info);
1302 }
1303
1304 EXPORT_SYMBOL(groups_free);
1305
1306 /* export the group_info to a user-space array */
1307 static int groups_to_user(gid_t __user *grouplist,
1308     struct group_info *group_info)
1309 {
1310         int i;
1311         int count = group_info->ngroups;
1312
1313         for (i = 0; i < group_info->nblocks; i++) {
1314                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1315                 int off = i * NGROUPS_PER_BLOCK;
1316                 int len = cp_count * sizeof(*grouplist);
1317
1318                 if (copy_to_user(grouplist+off, group_info->blocks[i], len))
1319                         return -EFAULT;
1320
1321                 count -= cp_count;
1322         }
1323         return 0;
1324 }
1325
1326 /* fill a group_info from a user-space array - it must be allocated already */
1327 static int groups_from_user(struct group_info *group_info,
1328     gid_t __user *grouplist)
1329  {
1330         int i;
1331         int count = group_info->ngroups;
1332
1333         for (i = 0; i < group_info->nblocks; i++) {
1334                 int cp_count = min(NGROUPS_PER_BLOCK, count);
1335                 int off = i * NGROUPS_PER_BLOCK;
1336                 int len = cp_count * sizeof(*grouplist);
1337
1338                 if (copy_from_user(group_info->blocks[i], grouplist+off, len))
1339                         return -EFAULT;
1340
1341                 count -= cp_count;
1342         }
1343         return 0;
1344 }
1345
1346 /* a simple Shell sort */
1347 static void groups_sort(struct group_info *group_info)
1348 {
1349         int base, max, stride;
1350         int gidsetsize = group_info->ngroups;
1351
1352         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1353                 ; /* nothing */
1354         stride /= 3;
1355
1356         while (stride) {
1357                 max = gidsetsize - stride;
1358                 for (base = 0; base < max; base++) {
1359                         int left = base;
1360                         int right = left + stride;
1361                         gid_t tmp = GROUP_AT(group_info, right);
1362
1363                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1364                                 GROUP_AT(group_info, right) =
1365                                     GROUP_AT(group_info, left);
1366                                 right = left;
1367                                 left -= stride;
1368                         }
1369                         GROUP_AT(group_info, right) = tmp;
1370                 }
1371                 stride /= 3;
1372         }
1373 }
1374
1375 /* a simple bsearch */
1376 int groups_search(struct group_info *group_info, gid_t grp)
1377 {
1378         int left, right;
1379
1380         if (!group_info)
1381                 return 0;
1382
1383         left = 0;
1384         right = group_info->ngroups;
1385         while (left < right) {
1386                 int mid = (left+right)/2;
1387                 int cmp = grp - GROUP_AT(group_info, mid);
1388                 if (cmp > 0)
1389                         left = mid + 1;
1390                 else if (cmp < 0)
1391                         right = mid;
1392                 else
1393                         return 1;
1394         }
1395         return 0;
1396 }
1397
1398 /* validate and set current->group_info */
1399 int set_current_groups(struct group_info *group_info)
1400 {
1401         int retval;
1402         struct group_info *old_info;
1403
1404         retval = security_task_setgroups(group_info);
1405         if (retval)
1406                 return retval;
1407
1408         groups_sort(group_info);
1409         get_group_info(group_info);
1410
1411         task_lock(current);
1412         old_info = current->group_info;
1413         current->group_info = group_info;
1414         task_unlock(current);
1415
1416         put_group_info(old_info);
1417
1418         return 0;
1419 }
1420
1421 EXPORT_SYMBOL(set_current_groups);
1422
1423 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1424 {
1425         int i = 0;
1426
1427         /*
1428          *      SMP: Nobody else can change our grouplist. Thus we are
1429          *      safe.
1430          */
1431
1432         if (gidsetsize < 0)
1433                 return -EINVAL;
1434
1435         /* no need to grab task_lock here; it cannot change */
1436         get_group_info(current->group_info);
1437         i = current->group_info->ngroups;
1438         if (gidsetsize) {
1439                 if (i > gidsetsize) {
1440                         i = -EINVAL;
1441                         goto out;
1442                 }
1443                 if (groups_to_user(grouplist, current->group_info)) {
1444                         i = -EFAULT;
1445                         goto out;
1446                 }
1447         }
1448 out:
1449         put_group_info(current->group_info);
1450         return i;
1451 }
1452
1453 /*
1454  *      SMP: Our groups are copy-on-write. We can set them safely
1455  *      without another task interfering.
1456  */
1457  
1458 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1459 {
1460         struct group_info *group_info;
1461         int retval;
1462
1463         if (!capable(CAP_SETGID))
1464                 return -EPERM;
1465         if ((unsigned)gidsetsize > NGROUPS_MAX)
1466                 return -EINVAL;
1467
1468         group_info = groups_alloc(gidsetsize);
1469         if (!group_info)
1470                 return -ENOMEM;
1471         retval = groups_from_user(group_info, grouplist);
1472         if (retval) {
1473                 put_group_info(group_info);
1474                 return retval;
1475         }
1476
1477         retval = set_current_groups(group_info);
1478         put_group_info(group_info);
1479
1480         return retval;
1481 }
1482
1483 /*
1484  * Check whether we're fsgid/egid or in the supplemental group..
1485  */
1486 int in_group_p(gid_t grp)
1487 {
1488         int retval = 1;
1489         if (grp != current->fsgid) {
1490                 get_group_info(current->group_info);
1491                 retval = groups_search(current->group_info, grp);
1492                 put_group_info(current->group_info);
1493         }
1494         return retval;
1495 }
1496
1497 EXPORT_SYMBOL(in_group_p);
1498
1499 int in_egroup_p(gid_t grp)
1500 {
1501         int retval = 1;
1502         if (grp != current->egid) {
1503                 get_group_info(current->group_info);
1504                 retval = groups_search(current->group_info, grp);
1505                 put_group_info(current->group_info);
1506         }
1507         return retval;
1508 }
1509
1510 EXPORT_SYMBOL(in_egroup_p);
1511
1512 DECLARE_RWSEM(uts_sem);
1513
1514 EXPORT_SYMBOL(uts_sem);
1515
1516 asmlinkage long sys_newuname(struct new_utsname __user * name)
1517 {
1518         int errno = 0;
1519
1520         down_read(&uts_sem);
1521         if (copy_to_user(name,&system_utsname,sizeof *name))
1522                 errno = -EFAULT;
1523         up_read(&uts_sem);
1524         return errno;
1525 }
1526
1527 asmlinkage long sys_sethostname(char __user *name, int len)
1528 {
1529         int errno;
1530         char tmp[__NEW_UTS_LEN];
1531
1532         if (!capable(CAP_SYS_ADMIN))
1533                 return -EPERM;
1534         if (len < 0 || len > __NEW_UTS_LEN)
1535                 return -EINVAL;
1536         down_write(&uts_sem);
1537         errno = -EFAULT;
1538         if (!copy_from_user(tmp, name, len)) {
1539                 memcpy(system_utsname.nodename, tmp, len);
1540                 system_utsname.nodename[len] = 0;
1541                 errno = 0;
1542         }
1543         up_write(&uts_sem);
1544         return errno;
1545 }
1546
1547 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1548
1549 asmlinkage long sys_gethostname(char __user *name, int len)
1550 {
1551         int i, errno;
1552
1553         if (len < 0)
1554                 return -EINVAL;
1555         down_read(&uts_sem);
1556         i = 1 + strlen(system_utsname.nodename);
1557         if (i > len)
1558                 i = len;
1559         errno = 0;
1560         if (copy_to_user(name, system_utsname.nodename, i))
1561                 errno = -EFAULT;
1562         up_read(&uts_sem);
1563         return errno;
1564 }
1565
1566 #endif
1567
1568 /*
1569  * Only setdomainname; getdomainname can be implemented by calling
1570  * uname()
1571  */
1572 asmlinkage long sys_setdomainname(char __user *name, int len)
1573 {
1574         int errno;
1575         char tmp[__NEW_UTS_LEN];
1576
1577         if (!capable(CAP_SYS_ADMIN))
1578                 return -EPERM;
1579         if (len < 0 || len > __NEW_UTS_LEN)
1580                 return -EINVAL;
1581
1582         down_write(&uts_sem);
1583         errno = -EFAULT;
1584         if (!copy_from_user(tmp, name, len)) {
1585                 memcpy(system_utsname.domainname, tmp, len);
1586                 system_utsname.domainname[len] = 0;
1587                 errno = 0;
1588         }
1589         up_write(&uts_sem);
1590         return errno;
1591 }
1592
1593 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1594 {
1595         if (resource >= RLIM_NLIMITS)
1596                 return -EINVAL;
1597         else {
1598                 struct rlimit value;
1599                 task_lock(current->group_leader);
1600                 value = current->signal->rlim[resource];
1601                 task_unlock(current->group_leader);
1602                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1603         }
1604 }
1605
1606 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1607
1608 /*
1609  *      Back compatibility for getrlimit. Needed for some apps.
1610  */
1611  
1612 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1613 {
1614         struct rlimit x;
1615         if (resource >= RLIM_NLIMITS)
1616                 return -EINVAL;
1617
1618         task_lock(current->group_leader);
1619         x = current->signal->rlim[resource];
1620         task_unlock(current->group_leader);
1621         if(x.rlim_cur > 0x7FFFFFFF)
1622                 x.rlim_cur = 0x7FFFFFFF;
1623         if(x.rlim_max > 0x7FFFFFFF)
1624                 x.rlim_max = 0x7FFFFFFF;
1625         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1626 }
1627
1628 #endif
1629
1630 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1631 {
1632         struct rlimit new_rlim, *old_rlim;
1633         int retval;
1634
1635         if (resource >= RLIM_NLIMITS)
1636                 return -EINVAL;
1637         if(copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1638                 return -EFAULT;
1639        if (new_rlim.rlim_cur > new_rlim.rlim_max)
1640                return -EINVAL;
1641         old_rlim = current->signal->rlim + resource;
1642         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1643             !capable(CAP_SYS_RESOURCE))
1644                 return -EPERM;
1645         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > NR_OPEN)
1646                         return -EPERM;
1647
1648         retval = security_task_setrlimit(resource, &new_rlim);
1649         if (retval)
1650                 return retval;
1651
1652         task_lock(current->group_leader);
1653         *old_rlim = new_rlim;
1654         task_unlock(current->group_leader);
1655
1656         if (resource == RLIMIT_CPU && new_rlim.rlim_cur != RLIM_INFINITY &&
1657             (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
1658              new_rlim.rlim_cur <= cputime_to_secs(
1659                      current->signal->it_prof_expires))) {
1660                 cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
1661                 read_lock(&tasklist_lock);
1662                 spin_lock_irq(&current->sighand->siglock);
1663                 set_process_cpu_timer(current, CPUCLOCK_PROF,
1664                                       &cputime, NULL);
1665                 spin_unlock_irq(&current->sighand->siglock);
1666                 read_unlock(&tasklist_lock);
1667         }
1668
1669         return 0;
1670 }
1671
1672 /*
1673  * It would make sense to put struct rusage in the task_struct,
1674  * except that would make the task_struct be *really big*.  After
1675  * task_struct gets moved into malloc'ed memory, it would
1676  * make sense to do this.  It will make moving the rest of the information
1677  * a lot simpler!  (Which we're not doing right now because we're not
1678  * measuring them yet).
1679  *
1680  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1681  * races with threads incrementing their own counters.  But since word
1682  * reads are atomic, we either get new values or old values and we don't
1683  * care which for the sums.  We always take the siglock to protect reading
1684  * the c* fields from p->signal from races with exit.c updating those
1685  * fields when reaping, so a sample either gets all the additions of a
1686  * given child after it's reaped, or none so this sample is before reaping.
1687  *
1688  * tasklist_lock locking optimisation:
1689  * If we are current and single threaded, we do not need to take the tasklist
1690  * lock or the siglock.  No one else can take our signal_struct away,
1691  * no one else can reap the children to update signal->c* counters, and
1692  * no one else can race with the signal-> fields.
1693  * If we do not take the tasklist_lock, the signal-> fields could be read
1694  * out of order while another thread was just exiting. So we place a
1695  * read memory barrier when we avoid the lock.  On the writer side,
1696  * write memory barrier is implied in  __exit_signal as __exit_signal releases
1697  * the siglock spinlock after updating the signal-> fields.
1698  *
1699  * We don't really need the siglock when we access the non c* fields
1700  * of the signal_struct (for RUSAGE_SELF) even in multithreaded
1701  * case, since we take the tasklist lock for read and the non c* signal->
1702  * fields are updated only in __exit_signal, which is called with
1703  * tasklist_lock taken for write, hence these two threads cannot execute
1704  * concurrently.
1705  *
1706  */
1707
1708 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1709 {
1710         struct task_struct *t;
1711         unsigned long flags;
1712         cputime_t utime, stime;
1713         int need_lock = 0;
1714
1715         memset((char *) r, 0, sizeof *r);
1716         utime = stime = cputime_zero;
1717
1718         if (p != current || !thread_group_empty(p))
1719                 need_lock = 1;
1720
1721         if (need_lock) {
1722                 read_lock(&tasklist_lock);
1723                 if (unlikely(!p->signal)) {
1724                         read_unlock(&tasklist_lock);
1725                         return;
1726                 }
1727         } else
1728                 /* See locking comments above */
1729                 smp_rmb();
1730
1731         switch (who) {
1732                 case RUSAGE_BOTH:
1733                 case RUSAGE_CHILDREN:
1734                         spin_lock_irqsave(&p->sighand->siglock, flags);
1735                         utime = p->signal->cutime;
1736                         stime = p->signal->cstime;
1737                         r->ru_nvcsw = p->signal->cnvcsw;
1738                         r->ru_nivcsw = p->signal->cnivcsw;
1739                         r->ru_minflt = p->signal->cmin_flt;
1740                         r->ru_majflt = p->signal->cmaj_flt;
1741                         spin_unlock_irqrestore(&p->sighand->siglock, flags);
1742
1743                         if (who == RUSAGE_CHILDREN)
1744                                 break;
1745
1746                 case RUSAGE_SELF:
1747                         utime = cputime_add(utime, p->signal->utime);
1748                         stime = cputime_add(stime, p->signal->stime);
1749                         r->ru_nvcsw += p->signal->nvcsw;
1750                         r->ru_nivcsw += p->signal->nivcsw;
1751                         r->ru_minflt += p->signal->min_flt;
1752                         r->ru_majflt += p->signal->maj_flt;
1753                         t = p;
1754                         do {
1755                                 utime = cputime_add(utime, t->utime);
1756                                 stime = cputime_add(stime, t->stime);
1757                                 r->ru_nvcsw += t->nvcsw;
1758                                 r->ru_nivcsw += t->nivcsw;
1759                                 r->ru_minflt += t->min_flt;
1760                                 r->ru_majflt += t->maj_flt;
1761                                 t = next_thread(t);
1762                         } while (t != p);
1763                         break;
1764
1765                 default:
1766                         BUG();
1767         }
1768
1769         if (need_lock)
1770                 read_unlock(&tasklist_lock);
1771         cputime_to_timeval(utime, &r->ru_utime);
1772         cputime_to_timeval(stime, &r->ru_stime);
1773 }
1774
1775 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1776 {
1777         struct rusage r;
1778         k_getrusage(p, who, &r);
1779         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1780 }
1781
1782 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1783 {
1784         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1785                 return -EINVAL;
1786         return getrusage(current, who, ru);
1787 }
1788
1789 asmlinkage long sys_umask(int mask)
1790 {
1791         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1792         return mask;
1793 }
1794     
1795 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1796                           unsigned long arg4, unsigned long arg5)
1797 {
1798         long error;
1799
1800         error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1801         if (error)
1802                 return error;
1803
1804         switch (option) {
1805                 case PR_SET_PDEATHSIG:
1806                         if (!valid_signal(arg2)) {
1807                                 error = -EINVAL;
1808                                 break;
1809                         }
1810                         current->pdeath_signal = arg2;
1811                         break;
1812                 case PR_GET_PDEATHSIG:
1813                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1814                         break;
1815                 case PR_GET_DUMPABLE:
1816                         error = current->mm->dumpable;
1817                         break;
1818                 case PR_SET_DUMPABLE:
1819                         if (arg2 < 0 || arg2 > 2) {
1820                                 error = -EINVAL;
1821                                 break;
1822                         }
1823                         current->mm->dumpable = arg2;
1824                         break;
1825
1826                 case PR_SET_UNALIGN:
1827                         error = SET_UNALIGN_CTL(current, arg2);
1828                         break;
1829                 case PR_GET_UNALIGN:
1830                         error = GET_UNALIGN_CTL(current, arg2);
1831                         break;
1832                 case PR_SET_FPEMU:
1833                         error = SET_FPEMU_CTL(current, arg2);
1834                         break;
1835                 case PR_GET_FPEMU:
1836                         error = GET_FPEMU_CTL(current, arg2);
1837                         break;
1838                 case PR_SET_FPEXC:
1839                         error = SET_FPEXC_CTL(current, arg2);
1840                         break;
1841                 case PR_GET_FPEXC:
1842                         error = GET_FPEXC_CTL(current, arg2);
1843                         break;
1844                 case PR_GET_TIMING:
1845                         error = PR_TIMING_STATISTICAL;
1846                         break;
1847                 case PR_SET_TIMING:
1848                         if (arg2 == PR_TIMING_STATISTICAL)
1849                                 error = 0;
1850                         else
1851                                 error = -EINVAL;
1852                         break;
1853
1854                 case PR_GET_KEEPCAPS:
1855                         if (current->keep_capabilities)
1856                                 error = 1;
1857                         break;
1858                 case PR_SET_KEEPCAPS:
1859                         if (arg2 != 0 && arg2 != 1) {
1860                                 error = -EINVAL;
1861                                 break;
1862                         }
1863                         current->keep_capabilities = arg2;
1864                         break;
1865                 case PR_SET_NAME: {
1866                         struct task_struct *me = current;
1867                         unsigned char ncomm[sizeof(me->comm)];
1868
1869                         ncomm[sizeof(me->comm)-1] = 0;
1870                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1871                                                 sizeof(me->comm)-1) < 0)
1872                                 return -EFAULT;
1873                         set_task_comm(me, ncomm);
1874                         return 0;
1875                 }
1876                 case PR_GET_NAME: {
1877                         struct task_struct *me = current;
1878                         unsigned char tcomm[sizeof(me->comm)];
1879
1880                         get_task_comm(tcomm, me);
1881                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1882                                 return -EFAULT;
1883                         return 0;
1884                 }
1885                 default:
1886                         error = -EINVAL;
1887                         break;
1888         }
1889         return error;
1890 }