capabilities: implement per-process securebits
[linux-2.6.git] / kernel / sys.c
1 /*
2  *  linux/kernel/sys.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 #include <linux/module.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/smp_lock.h>
12 #include <linux/notifier.h>
13 #include <linux/reboot.h>
14 #include <linux/prctl.h>
15 #include <linux/highuid.h>
16 #include <linux/fs.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/kexec.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36
37 #include <linux/compat.h>
38 #include <linux/syscalls.h>
39 #include <linux/kprobes.h>
40 #include <linux/user_namespace.h>
41
42 #include <asm/uaccess.h>
43 #include <asm/io.h>
44 #include <asm/unistd.h>
45
46 #ifndef SET_UNALIGN_CTL
47 # define SET_UNALIGN_CTL(a,b)   (-EINVAL)
48 #endif
49 #ifndef GET_UNALIGN_CTL
50 # define GET_UNALIGN_CTL(a,b)   (-EINVAL)
51 #endif
52 #ifndef SET_FPEMU_CTL
53 # define SET_FPEMU_CTL(a,b)     (-EINVAL)
54 #endif
55 #ifndef GET_FPEMU_CTL
56 # define GET_FPEMU_CTL(a,b)     (-EINVAL)
57 #endif
58 #ifndef SET_FPEXC_CTL
59 # define SET_FPEXC_CTL(a,b)     (-EINVAL)
60 #endif
61 #ifndef GET_FPEXC_CTL
62 # define GET_FPEXC_CTL(a,b)     (-EINVAL)
63 #endif
64 #ifndef GET_ENDIAN
65 # define GET_ENDIAN(a,b)        (-EINVAL)
66 #endif
67 #ifndef SET_ENDIAN
68 # define SET_ENDIAN(a,b)        (-EINVAL)
69 #endif
70 #ifndef GET_TSC_CTL
71 # define GET_TSC_CTL(a)         (-EINVAL)
72 #endif
73 #ifndef SET_TSC_CTL
74 # define SET_TSC_CTL(a)         (-EINVAL)
75 #endif
76
77 /*
78  * this is where the system-wide overflow UID and GID are defined, for
79  * architectures that now have 32-bit UID/GID but didn't in the past
80  */
81
82 int overflowuid = DEFAULT_OVERFLOWUID;
83 int overflowgid = DEFAULT_OVERFLOWGID;
84
85 #ifdef CONFIG_UID16
86 EXPORT_SYMBOL(overflowuid);
87 EXPORT_SYMBOL(overflowgid);
88 #endif
89
90 /*
91  * the same as above, but for filesystems which can only store a 16-bit
92  * UID and GID. as such, this is needed on all architectures
93  */
94
95 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
96 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
97
98 EXPORT_SYMBOL(fs_overflowuid);
99 EXPORT_SYMBOL(fs_overflowgid);
100
101 /*
102  * this indicates whether you can reboot with ctrl-alt-del: the default is yes
103  */
104
105 int C_A_D = 1;
106 struct pid *cad_pid;
107 EXPORT_SYMBOL(cad_pid);
108
109 /*
110  * If set, this is used for preparing the system to power off.
111  */
112
113 void (*pm_power_off_prepare)(void);
114
115 static int set_one_prio(struct task_struct *p, int niceval, int error)
116 {
117         int no_nice;
118
119         if (p->uid != current->euid &&
120                 p->euid != current->euid && !capable(CAP_SYS_NICE)) {
121                 error = -EPERM;
122                 goto out;
123         }
124         if (niceval < task_nice(p) && !can_nice(p, niceval)) {
125                 error = -EACCES;
126                 goto out;
127         }
128         no_nice = security_task_setnice(p, niceval);
129         if (no_nice) {
130                 error = no_nice;
131                 goto out;
132         }
133         if (error == -ESRCH)
134                 error = 0;
135         set_user_nice(p, niceval);
136 out:
137         return error;
138 }
139
140 asmlinkage long sys_setpriority(int which, int who, int niceval)
141 {
142         struct task_struct *g, *p;
143         struct user_struct *user;
144         int error = -EINVAL;
145         struct pid *pgrp;
146
147         if (which > PRIO_USER || which < PRIO_PROCESS)
148                 goto out;
149
150         /* normalize: avoid signed division (rounding problems) */
151         error = -ESRCH;
152         if (niceval < -20)
153                 niceval = -20;
154         if (niceval > 19)
155                 niceval = 19;
156
157         read_lock(&tasklist_lock);
158         switch (which) {
159                 case PRIO_PROCESS:
160                         if (who)
161                                 p = find_task_by_vpid(who);
162                         else
163                                 p = current;
164                         if (p)
165                                 error = set_one_prio(p, niceval, error);
166                         break;
167                 case PRIO_PGRP:
168                         if (who)
169                                 pgrp = find_vpid(who);
170                         else
171                                 pgrp = task_pgrp(current);
172                         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
173                                 error = set_one_prio(p, niceval, error);
174                         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
175                         break;
176                 case PRIO_USER:
177                         user = current->user;
178                         if (!who)
179                                 who = current->uid;
180                         else
181                                 if ((who != current->uid) && !(user = find_user(who)))
182                                         goto out_unlock;        /* No processes for this user */
183
184                         do_each_thread(g, p)
185                                 if (p->uid == who)
186                                         error = set_one_prio(p, niceval, error);
187                         while_each_thread(g, p);
188                         if (who != current->uid)
189                                 free_uid(user);         /* For find_user() */
190                         break;
191         }
192 out_unlock:
193         read_unlock(&tasklist_lock);
194 out:
195         return error;
196 }
197
198 /*
199  * Ugh. To avoid negative return values, "getpriority()" will
200  * not return the normal nice-value, but a negated value that
201  * has been offset by 20 (ie it returns 40..1 instead of -20..19)
202  * to stay compatible.
203  */
204 asmlinkage long sys_getpriority(int which, int who)
205 {
206         struct task_struct *g, *p;
207         struct user_struct *user;
208         long niceval, retval = -ESRCH;
209         struct pid *pgrp;
210
211         if (which > PRIO_USER || which < PRIO_PROCESS)
212                 return -EINVAL;
213
214         read_lock(&tasklist_lock);
215         switch (which) {
216                 case PRIO_PROCESS:
217                         if (who)
218                                 p = find_task_by_vpid(who);
219                         else
220                                 p = current;
221                         if (p) {
222                                 niceval = 20 - task_nice(p);
223                                 if (niceval > retval)
224                                         retval = niceval;
225                         }
226                         break;
227                 case PRIO_PGRP:
228                         if (who)
229                                 pgrp = find_vpid(who);
230                         else
231                                 pgrp = task_pgrp(current);
232                         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
233                                 niceval = 20 - task_nice(p);
234                                 if (niceval > retval)
235                                         retval = niceval;
236                         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
237                         break;
238                 case PRIO_USER:
239                         user = current->user;
240                         if (!who)
241                                 who = current->uid;
242                         else
243                                 if ((who != current->uid) && !(user = find_user(who)))
244                                         goto out_unlock;        /* No processes for this user */
245
246                         do_each_thread(g, p)
247                                 if (p->uid == who) {
248                                         niceval = 20 - task_nice(p);
249                                         if (niceval > retval)
250                                                 retval = niceval;
251                                 }
252                         while_each_thread(g, p);
253                         if (who != current->uid)
254                                 free_uid(user);         /* for find_user() */
255                         break;
256         }
257 out_unlock:
258         read_unlock(&tasklist_lock);
259
260         return retval;
261 }
262
263 /**
264  *      emergency_restart - reboot the system
265  *
266  *      Without shutting down any hardware or taking any locks
267  *      reboot the system.  This is called when we know we are in
268  *      trouble so this is our best effort to reboot.  This is
269  *      safe to call in interrupt context.
270  */
271 void emergency_restart(void)
272 {
273         machine_emergency_restart();
274 }
275 EXPORT_SYMBOL_GPL(emergency_restart);
276
277 static void kernel_restart_prepare(char *cmd)
278 {
279         blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
280         system_state = SYSTEM_RESTART;
281         device_shutdown();
282         sysdev_shutdown();
283 }
284
285 /**
286  *      kernel_restart - reboot the system
287  *      @cmd: pointer to buffer containing command to execute for restart
288  *              or %NULL
289  *
290  *      Shutdown everything and perform a clean reboot.
291  *      This is not safe to call in interrupt context.
292  */
293 void kernel_restart(char *cmd)
294 {
295         kernel_restart_prepare(cmd);
296         if (!cmd)
297                 printk(KERN_EMERG "Restarting system.\n");
298         else
299                 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
300         machine_restart(cmd);
301 }
302 EXPORT_SYMBOL_GPL(kernel_restart);
303
304 /**
305  *      kernel_kexec - reboot the system
306  *
307  *      Move into place and start executing a preloaded standalone
308  *      executable.  If nothing was preloaded return an error.
309  */
310 static void kernel_kexec(void)
311 {
312 #ifdef CONFIG_KEXEC
313         struct kimage *image;
314         image = xchg(&kexec_image, NULL);
315         if (!image)
316                 return;
317         kernel_restart_prepare(NULL);
318         printk(KERN_EMERG "Starting new kernel\n");
319         machine_shutdown();
320         machine_kexec(image);
321 #endif
322 }
323
324 static void kernel_shutdown_prepare(enum system_states state)
325 {
326         blocking_notifier_call_chain(&reboot_notifier_list,
327                 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
328         system_state = state;
329         device_shutdown();
330 }
331 /**
332  *      kernel_halt - halt the system
333  *
334  *      Shutdown everything and perform a clean system halt.
335  */
336 void kernel_halt(void)
337 {
338         kernel_shutdown_prepare(SYSTEM_HALT);
339         sysdev_shutdown();
340         printk(KERN_EMERG "System halted.\n");
341         machine_halt();
342 }
343
344 EXPORT_SYMBOL_GPL(kernel_halt);
345
346 /**
347  *      kernel_power_off - power_off the system
348  *
349  *      Shutdown everything and perform a clean system power_off.
350  */
351 void kernel_power_off(void)
352 {
353         kernel_shutdown_prepare(SYSTEM_POWER_OFF);
354         if (pm_power_off_prepare)
355                 pm_power_off_prepare();
356         disable_nonboot_cpus();
357         sysdev_shutdown();
358         printk(KERN_EMERG "Power down.\n");
359         machine_power_off();
360 }
361 EXPORT_SYMBOL_GPL(kernel_power_off);
362 /*
363  * Reboot system call: for obvious reasons only root may call it,
364  * and even root needs to set up some magic numbers in the registers
365  * so that some mistake won't make this reboot the whole machine.
366  * You can also set the meaning of the ctrl-alt-del-key here.
367  *
368  * reboot doesn't sync: do that yourself before calling this.
369  */
370 asmlinkage long sys_reboot(int magic1, int magic2, unsigned int cmd, void __user * arg)
371 {
372         char buffer[256];
373
374         /* We only trust the superuser with rebooting the system. */
375         if (!capable(CAP_SYS_BOOT))
376                 return -EPERM;
377
378         /* For safety, we require "magic" arguments. */
379         if (magic1 != LINUX_REBOOT_MAGIC1 ||
380             (magic2 != LINUX_REBOOT_MAGIC2 &&
381                         magic2 != LINUX_REBOOT_MAGIC2A &&
382                         magic2 != LINUX_REBOOT_MAGIC2B &&
383                         magic2 != LINUX_REBOOT_MAGIC2C))
384                 return -EINVAL;
385
386         /* Instead of trying to make the power_off code look like
387          * halt when pm_power_off is not set do it the easy way.
388          */
389         if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
390                 cmd = LINUX_REBOOT_CMD_HALT;
391
392         lock_kernel();
393         switch (cmd) {
394         case LINUX_REBOOT_CMD_RESTART:
395                 kernel_restart(NULL);
396                 break;
397
398         case LINUX_REBOOT_CMD_CAD_ON:
399                 C_A_D = 1;
400                 break;
401
402         case LINUX_REBOOT_CMD_CAD_OFF:
403                 C_A_D = 0;
404                 break;
405
406         case LINUX_REBOOT_CMD_HALT:
407                 kernel_halt();
408                 unlock_kernel();
409                 do_exit(0);
410                 break;
411
412         case LINUX_REBOOT_CMD_POWER_OFF:
413                 kernel_power_off();
414                 unlock_kernel();
415                 do_exit(0);
416                 break;
417
418         case LINUX_REBOOT_CMD_RESTART2:
419                 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
420                         unlock_kernel();
421                         return -EFAULT;
422                 }
423                 buffer[sizeof(buffer) - 1] = '\0';
424
425                 kernel_restart(buffer);
426                 break;
427
428         case LINUX_REBOOT_CMD_KEXEC:
429                 kernel_kexec();
430                 unlock_kernel();
431                 return -EINVAL;
432
433 #ifdef CONFIG_HIBERNATION
434         case LINUX_REBOOT_CMD_SW_SUSPEND:
435                 {
436                         int ret = hibernate();
437                         unlock_kernel();
438                         return ret;
439                 }
440 #endif
441
442         default:
443                 unlock_kernel();
444                 return -EINVAL;
445         }
446         unlock_kernel();
447         return 0;
448 }
449
450 static void deferred_cad(struct work_struct *dummy)
451 {
452         kernel_restart(NULL);
453 }
454
455 /*
456  * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
457  * As it's called within an interrupt, it may NOT sync: the only choice
458  * is whether to reboot at once, or just ignore the ctrl-alt-del.
459  */
460 void ctrl_alt_del(void)
461 {
462         static DECLARE_WORK(cad_work, deferred_cad);
463
464         if (C_A_D)
465                 schedule_work(&cad_work);
466         else
467                 kill_cad_pid(SIGINT, 1);
468 }
469         
470 /*
471  * Unprivileged users may change the real gid to the effective gid
472  * or vice versa.  (BSD-style)
473  *
474  * If you set the real gid at all, or set the effective gid to a value not
475  * equal to the real gid, then the saved gid is set to the new effective gid.
476  *
477  * This makes it possible for a setgid program to completely drop its
478  * privileges, which is often a useful assertion to make when you are doing
479  * a security audit over a program.
480  *
481  * The general idea is that a program which uses just setregid() will be
482  * 100% compatible with BSD.  A program which uses just setgid() will be
483  * 100% compatible with POSIX with saved IDs. 
484  *
485  * SMP: There are not races, the GIDs are checked only by filesystem
486  *      operations (as far as semantic preservation is concerned).
487  */
488 asmlinkage long sys_setregid(gid_t rgid, gid_t egid)
489 {
490         int old_rgid = current->gid;
491         int old_egid = current->egid;
492         int new_rgid = old_rgid;
493         int new_egid = old_egid;
494         int retval;
495
496         retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
497         if (retval)
498                 return retval;
499
500         if (rgid != (gid_t) -1) {
501                 if ((old_rgid == rgid) ||
502                     (current->egid==rgid) ||
503                     capable(CAP_SETGID))
504                         new_rgid = rgid;
505                 else
506                         return -EPERM;
507         }
508         if (egid != (gid_t) -1) {
509                 if ((old_rgid == egid) ||
510                     (current->egid == egid) ||
511                     (current->sgid == egid) ||
512                     capable(CAP_SETGID))
513                         new_egid = egid;
514                 else
515                         return -EPERM;
516         }
517         if (new_egid != old_egid) {
518                 set_dumpable(current->mm, suid_dumpable);
519                 smp_wmb();
520         }
521         if (rgid != (gid_t) -1 ||
522             (egid != (gid_t) -1 && egid != old_rgid))
523                 current->sgid = new_egid;
524         current->fsgid = new_egid;
525         current->egid = new_egid;
526         current->gid = new_rgid;
527         key_fsgid_changed(current);
528         proc_id_connector(current, PROC_EVENT_GID);
529         return 0;
530 }
531
532 /*
533  * setgid() is implemented like SysV w/ SAVED_IDS 
534  *
535  * SMP: Same implicit races as above.
536  */
537 asmlinkage long sys_setgid(gid_t gid)
538 {
539         int old_egid = current->egid;
540         int retval;
541
542         retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
543         if (retval)
544                 return retval;
545
546         if (capable(CAP_SETGID)) {
547                 if (old_egid != gid) {
548                         set_dumpable(current->mm, suid_dumpable);
549                         smp_wmb();
550                 }
551                 current->gid = current->egid = current->sgid = current->fsgid = gid;
552         } else if ((gid == current->gid) || (gid == current->sgid)) {
553                 if (old_egid != gid) {
554                         set_dumpable(current->mm, suid_dumpable);
555                         smp_wmb();
556                 }
557                 current->egid = current->fsgid = gid;
558         }
559         else
560                 return -EPERM;
561
562         key_fsgid_changed(current);
563         proc_id_connector(current, PROC_EVENT_GID);
564         return 0;
565 }
566   
567 static int set_user(uid_t new_ruid, int dumpclear)
568 {
569         struct user_struct *new_user;
570
571         new_user = alloc_uid(current->nsproxy->user_ns, new_ruid);
572         if (!new_user)
573                 return -EAGAIN;
574
575         if (atomic_read(&new_user->processes) >=
576                                 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
577                         new_user != current->nsproxy->user_ns->root_user) {
578                 free_uid(new_user);
579                 return -EAGAIN;
580         }
581
582         switch_uid(new_user);
583
584         if (dumpclear) {
585                 set_dumpable(current->mm, suid_dumpable);
586                 smp_wmb();
587         }
588         current->uid = new_ruid;
589         return 0;
590 }
591
592 /*
593  * Unprivileged users may change the real uid to the effective uid
594  * or vice versa.  (BSD-style)
595  *
596  * If you set the real uid at all, or set the effective uid to a value not
597  * equal to the real uid, then the saved uid is set to the new effective uid.
598  *
599  * This makes it possible for a setuid program to completely drop its
600  * privileges, which is often a useful assertion to make when you are doing
601  * a security audit over a program.
602  *
603  * The general idea is that a program which uses just setreuid() will be
604  * 100% compatible with BSD.  A program which uses just setuid() will be
605  * 100% compatible with POSIX with saved IDs. 
606  */
607 asmlinkage long sys_setreuid(uid_t ruid, uid_t euid)
608 {
609         int old_ruid, old_euid, old_suid, new_ruid, new_euid;
610         int retval;
611
612         retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
613         if (retval)
614                 return retval;
615
616         new_ruid = old_ruid = current->uid;
617         new_euid = old_euid = current->euid;
618         old_suid = current->suid;
619
620         if (ruid != (uid_t) -1) {
621                 new_ruid = ruid;
622                 if ((old_ruid != ruid) &&
623                     (current->euid != ruid) &&
624                     !capable(CAP_SETUID))
625                         return -EPERM;
626         }
627
628         if (euid != (uid_t) -1) {
629                 new_euid = euid;
630                 if ((old_ruid != euid) &&
631                     (current->euid != euid) &&
632                     (current->suid != euid) &&
633                     !capable(CAP_SETUID))
634                         return -EPERM;
635         }
636
637         if (new_ruid != old_ruid && set_user(new_ruid, new_euid != old_euid) < 0)
638                 return -EAGAIN;
639
640         if (new_euid != old_euid) {
641                 set_dumpable(current->mm, suid_dumpable);
642                 smp_wmb();
643         }
644         current->fsuid = current->euid = new_euid;
645         if (ruid != (uid_t) -1 ||
646             (euid != (uid_t) -1 && euid != old_ruid))
647                 current->suid = current->euid;
648         current->fsuid = current->euid;
649
650         key_fsuid_changed(current);
651         proc_id_connector(current, PROC_EVENT_UID);
652
653         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RE);
654 }
655
656
657                 
658 /*
659  * setuid() is implemented like SysV with SAVED_IDS 
660  * 
661  * Note that SAVED_ID's is deficient in that a setuid root program
662  * like sendmail, for example, cannot set its uid to be a normal 
663  * user and then switch back, because if you're root, setuid() sets
664  * the saved uid too.  If you don't like this, blame the bright people
665  * in the POSIX committee and/or USG.  Note that the BSD-style setreuid()
666  * will allow a root program to temporarily drop privileges and be able to
667  * regain them by swapping the real and effective uid.  
668  */
669 asmlinkage long sys_setuid(uid_t uid)
670 {
671         int old_euid = current->euid;
672         int old_ruid, old_suid, new_suid;
673         int retval;
674
675         retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
676         if (retval)
677                 return retval;
678
679         old_ruid = current->uid;
680         old_suid = current->suid;
681         new_suid = old_suid;
682         
683         if (capable(CAP_SETUID)) {
684                 if (uid != old_ruid && set_user(uid, old_euid != uid) < 0)
685                         return -EAGAIN;
686                 new_suid = uid;
687         } else if ((uid != current->uid) && (uid != new_suid))
688                 return -EPERM;
689
690         if (old_euid != uid) {
691                 set_dumpable(current->mm, suid_dumpable);
692                 smp_wmb();
693         }
694         current->fsuid = current->euid = uid;
695         current->suid = new_suid;
696
697         key_fsuid_changed(current);
698         proc_id_connector(current, PROC_EVENT_UID);
699
700         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_ID);
701 }
702
703
704 /*
705  * This function implements a generic ability to update ruid, euid,
706  * and suid.  This allows you to implement the 4.4 compatible seteuid().
707  */
708 asmlinkage long sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
709 {
710         int old_ruid = current->uid;
711         int old_euid = current->euid;
712         int old_suid = current->suid;
713         int retval;
714
715         retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
716         if (retval)
717                 return retval;
718
719         if (!capable(CAP_SETUID)) {
720                 if ((ruid != (uid_t) -1) && (ruid != current->uid) &&
721                     (ruid != current->euid) && (ruid != current->suid))
722                         return -EPERM;
723                 if ((euid != (uid_t) -1) && (euid != current->uid) &&
724                     (euid != current->euid) && (euid != current->suid))
725                         return -EPERM;
726                 if ((suid != (uid_t) -1) && (suid != current->uid) &&
727                     (suid != current->euid) && (suid != current->suid))
728                         return -EPERM;
729         }
730         if (ruid != (uid_t) -1) {
731                 if (ruid != current->uid && set_user(ruid, euid != current->euid) < 0)
732                         return -EAGAIN;
733         }
734         if (euid != (uid_t) -1) {
735                 if (euid != current->euid) {
736                         set_dumpable(current->mm, suid_dumpable);
737                         smp_wmb();
738                 }
739                 current->euid = euid;
740         }
741         current->fsuid = current->euid;
742         if (suid != (uid_t) -1)
743                 current->suid = suid;
744
745         key_fsuid_changed(current);
746         proc_id_connector(current, PROC_EVENT_UID);
747
748         return security_task_post_setuid(old_ruid, old_euid, old_suid, LSM_SETID_RES);
749 }
750
751 asmlinkage long sys_getresuid(uid_t __user *ruid, uid_t __user *euid, uid_t __user *suid)
752 {
753         int retval;
754
755         if (!(retval = put_user(current->uid, ruid)) &&
756             !(retval = put_user(current->euid, euid)))
757                 retval = put_user(current->suid, suid);
758
759         return retval;
760 }
761
762 /*
763  * Same as above, but for rgid, egid, sgid.
764  */
765 asmlinkage long sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
766 {
767         int retval;
768
769         retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
770         if (retval)
771                 return retval;
772
773         if (!capable(CAP_SETGID)) {
774                 if ((rgid != (gid_t) -1) && (rgid != current->gid) &&
775                     (rgid != current->egid) && (rgid != current->sgid))
776                         return -EPERM;
777                 if ((egid != (gid_t) -1) && (egid != current->gid) &&
778                     (egid != current->egid) && (egid != current->sgid))
779                         return -EPERM;
780                 if ((sgid != (gid_t) -1) && (sgid != current->gid) &&
781                     (sgid != current->egid) && (sgid != current->sgid))
782                         return -EPERM;
783         }
784         if (egid != (gid_t) -1) {
785                 if (egid != current->egid) {
786                         set_dumpable(current->mm, suid_dumpable);
787                         smp_wmb();
788                 }
789                 current->egid = egid;
790         }
791         current->fsgid = current->egid;
792         if (rgid != (gid_t) -1)
793                 current->gid = rgid;
794         if (sgid != (gid_t) -1)
795                 current->sgid = sgid;
796
797         key_fsgid_changed(current);
798         proc_id_connector(current, PROC_EVENT_GID);
799         return 0;
800 }
801
802 asmlinkage long sys_getresgid(gid_t __user *rgid, gid_t __user *egid, gid_t __user *sgid)
803 {
804         int retval;
805
806         if (!(retval = put_user(current->gid, rgid)) &&
807             !(retval = put_user(current->egid, egid)))
808                 retval = put_user(current->sgid, sgid);
809
810         return retval;
811 }
812
813
814 /*
815  * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
816  * is used for "access()" and for the NFS daemon (letting nfsd stay at
817  * whatever uid it wants to). It normally shadows "euid", except when
818  * explicitly set by setfsuid() or for access..
819  */
820 asmlinkage long sys_setfsuid(uid_t uid)
821 {
822         int old_fsuid;
823
824         old_fsuid = current->fsuid;
825         if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS))
826                 return old_fsuid;
827
828         if (uid == current->uid || uid == current->euid ||
829             uid == current->suid || uid == current->fsuid || 
830             capable(CAP_SETUID)) {
831                 if (uid != old_fsuid) {
832                         set_dumpable(current->mm, suid_dumpable);
833                         smp_wmb();
834                 }
835                 current->fsuid = uid;
836         }
837
838         key_fsuid_changed(current);
839         proc_id_connector(current, PROC_EVENT_UID);
840
841         security_task_post_setuid(old_fsuid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS);
842
843         return old_fsuid;
844 }
845
846 /*
847  * Samma på svenska..
848  */
849 asmlinkage long sys_setfsgid(gid_t gid)
850 {
851         int old_fsgid;
852
853         old_fsgid = current->fsgid;
854         if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
855                 return old_fsgid;
856
857         if (gid == current->gid || gid == current->egid ||
858             gid == current->sgid || gid == current->fsgid || 
859             capable(CAP_SETGID)) {
860                 if (gid != old_fsgid) {
861                         set_dumpable(current->mm, suid_dumpable);
862                         smp_wmb();
863                 }
864                 current->fsgid = gid;
865                 key_fsgid_changed(current);
866                 proc_id_connector(current, PROC_EVENT_GID);
867         }
868         return old_fsgid;
869 }
870
871 asmlinkage long sys_times(struct tms __user * tbuf)
872 {
873         /*
874          *      In the SMP world we might just be unlucky and have one of
875          *      the times increment as we use it. Since the value is an
876          *      atomically safe type this is just fine. Conceptually its
877          *      as if the syscall took an instant longer to occur.
878          */
879         if (tbuf) {
880                 struct tms tmp;
881                 struct task_struct *tsk = current;
882                 struct task_struct *t;
883                 cputime_t utime, stime, cutime, cstime;
884
885                 spin_lock_irq(&tsk->sighand->siglock);
886                 utime = tsk->signal->utime;
887                 stime = tsk->signal->stime;
888                 t = tsk;
889                 do {
890                         utime = cputime_add(utime, t->utime);
891                         stime = cputime_add(stime, t->stime);
892                         t = next_thread(t);
893                 } while (t != tsk);
894
895                 cutime = tsk->signal->cutime;
896                 cstime = tsk->signal->cstime;
897                 spin_unlock_irq(&tsk->sighand->siglock);
898
899                 tmp.tms_utime = cputime_to_clock_t(utime);
900                 tmp.tms_stime = cputime_to_clock_t(stime);
901                 tmp.tms_cutime = cputime_to_clock_t(cutime);
902                 tmp.tms_cstime = cputime_to_clock_t(cstime);
903                 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
904                         return -EFAULT;
905         }
906         return (long) jiffies_64_to_clock_t(get_jiffies_64());
907 }
908
909 /*
910  * This needs some heavy checking ...
911  * I just haven't the stomach for it. I also don't fully
912  * understand sessions/pgrp etc. Let somebody who does explain it.
913  *
914  * OK, I think I have the protection semantics right.... this is really
915  * only important on a multi-user system anyway, to make sure one user
916  * can't send a signal to a process owned by another.  -TYT, 12/12/91
917  *
918  * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
919  * LBT 04.03.94
920  */
921 asmlinkage long sys_setpgid(pid_t pid, pid_t pgid)
922 {
923         struct task_struct *p;
924         struct task_struct *group_leader = current->group_leader;
925         struct pid *pgrp;
926         int err;
927
928         if (!pid)
929                 pid = task_pid_vnr(group_leader);
930         if (!pgid)
931                 pgid = pid;
932         if (pgid < 0)
933                 return -EINVAL;
934
935         /* From this point forward we keep holding onto the tasklist lock
936          * so that our parent does not change from under us. -DaveM
937          */
938         write_lock_irq(&tasklist_lock);
939
940         err = -ESRCH;
941         p = find_task_by_vpid(pid);
942         if (!p)
943                 goto out;
944
945         err = -EINVAL;
946         if (!thread_group_leader(p))
947                 goto out;
948
949         if (same_thread_group(p->real_parent, group_leader)) {
950                 err = -EPERM;
951                 if (task_session(p) != task_session(group_leader))
952                         goto out;
953                 err = -EACCES;
954                 if (p->did_exec)
955                         goto out;
956         } else {
957                 err = -ESRCH;
958                 if (p != group_leader)
959                         goto out;
960         }
961
962         err = -EPERM;
963         if (p->signal->leader)
964                 goto out;
965
966         pgrp = task_pid(p);
967         if (pgid != pid) {
968                 struct task_struct *g;
969
970                 pgrp = find_vpid(pgid);
971                 g = pid_task(pgrp, PIDTYPE_PGID);
972                 if (!g || task_session(g) != task_session(group_leader))
973                         goto out;
974         }
975
976         err = security_task_setpgid(p, pgid);
977         if (err)
978                 goto out;
979
980         if (task_pgrp(p) != pgrp) {
981                 detach_pid(p, PIDTYPE_PGID);
982                 attach_pid(p, PIDTYPE_PGID, pgrp);
983                 set_task_pgrp(p, pid_nr(pgrp));
984         }
985
986         err = 0;
987 out:
988         /* All paths lead to here, thus we are safe. -DaveM */
989         write_unlock_irq(&tasklist_lock);
990         return err;
991 }
992
993 asmlinkage long sys_getpgid(pid_t pid)
994 {
995         if (!pid)
996                 return task_pgrp_vnr(current);
997         else {
998                 int retval;
999                 struct task_struct *p;
1000
1001                 read_lock(&tasklist_lock);
1002                 p = find_task_by_vpid(pid);
1003                 retval = -ESRCH;
1004                 if (p) {
1005                         retval = security_task_getpgid(p);
1006                         if (!retval)
1007                                 retval = task_pgrp_vnr(p);
1008                 }
1009                 read_unlock(&tasklist_lock);
1010                 return retval;
1011         }
1012 }
1013
1014 #ifdef __ARCH_WANT_SYS_GETPGRP
1015
1016 asmlinkage long sys_getpgrp(void)
1017 {
1018         /* SMP - assuming writes are word atomic this is fine */
1019         return task_pgrp_vnr(current);
1020 }
1021
1022 #endif
1023
1024 asmlinkage long sys_getsid(pid_t pid)
1025 {
1026         if (!pid)
1027                 return task_session_vnr(current);
1028         else {
1029                 int retval;
1030                 struct task_struct *p;
1031
1032                 rcu_read_lock();
1033                 p = find_task_by_vpid(pid);
1034                 retval = -ESRCH;
1035                 if (p) {
1036                         retval = security_task_getsid(p);
1037                         if (!retval)
1038                                 retval = task_session_vnr(p);
1039                 }
1040                 rcu_read_unlock();
1041                 return retval;
1042         }
1043 }
1044
1045 asmlinkage long sys_setsid(void)
1046 {
1047         struct task_struct *group_leader = current->group_leader;
1048         struct pid *sid = task_pid(group_leader);
1049         pid_t session = pid_vnr(sid);
1050         int err = -EPERM;
1051
1052         write_lock_irq(&tasklist_lock);
1053         /* Fail if I am already a session leader */
1054         if (group_leader->signal->leader)
1055                 goto out;
1056
1057         /* Fail if a process group id already exists that equals the
1058          * proposed session id.
1059          */
1060         if (pid_task(sid, PIDTYPE_PGID))
1061                 goto out;
1062
1063         group_leader->signal->leader = 1;
1064         __set_special_pids(sid);
1065
1066         spin_lock(&group_leader->sighand->siglock);
1067         group_leader->signal->tty = NULL;
1068         spin_unlock(&group_leader->sighand->siglock);
1069
1070         err = session;
1071 out:
1072         write_unlock_irq(&tasklist_lock);
1073         return err;
1074 }
1075
1076 /*
1077  * Supplementary group IDs
1078  */
1079
1080 /* init to 2 - one for init_task, one to ensure it is never freed */
1081 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1082
1083 struct group_info *groups_alloc(int gidsetsize)
1084 {
1085         struct group_info *group_info;
1086         int nblocks;
1087         int i;
1088
1089         nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1090         /* Make sure we always allocate at least one indirect block pointer */
1091         nblocks = nblocks ? : 1;
1092         group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1093         if (!group_info)
1094                 return NULL;
1095         group_info->ngroups = gidsetsize;
1096         group_info->nblocks = nblocks;
1097         atomic_set(&group_info->usage, 1);
1098
1099         if (gidsetsize <= NGROUPS_SMALL)
1100                 group_info->blocks[0] = group_info->small_block;
1101         else {
1102                 for (i = 0; i < nblocks; i++) {
1103                         gid_t *b;
1104                         b = (void *)__get_free_page(GFP_USER);
1105                         if (!b)
1106                                 goto out_undo_partial_alloc;
1107                         group_info->blocks[i] = b;
1108                 }
1109         }
1110         return group_info;
1111
1112 out_undo_partial_alloc:
1113         while (--i >= 0) {
1114                 free_page((unsigned long)group_info->blocks[i]);
1115         }
1116         kfree(group_info);
1117         return NULL;
1118 }
1119
1120 EXPORT_SYMBOL(groups_alloc);
1121
1122 void groups_free(struct group_info *group_info)
1123 {
1124         if (group_info->blocks[0] != group_info->small_block) {
1125                 int i;
1126                 for (i = 0; i < group_info->nblocks; i++)
1127                         free_page((unsigned long)group_info->blocks[i]);
1128         }
1129         kfree(group_info);
1130 }
1131
1132 EXPORT_SYMBOL(groups_free);
1133
1134 /* export the group_info to a user-space array */
1135 static int groups_to_user(gid_t __user *grouplist,
1136     struct group_info *group_info)
1137 {
1138         int i;
1139         unsigned int count = group_info->ngroups;
1140
1141         for (i = 0; i < group_info->nblocks; i++) {
1142                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1143                 unsigned int len = cp_count * sizeof(*grouplist);
1144
1145                 if (copy_to_user(grouplist, group_info->blocks[i], len))
1146                         return -EFAULT;
1147
1148                 grouplist += NGROUPS_PER_BLOCK;
1149                 count -= cp_count;
1150         }
1151         return 0;
1152 }
1153
1154 /* fill a group_info from a user-space array - it must be allocated already */
1155 static int groups_from_user(struct group_info *group_info,
1156     gid_t __user *grouplist)
1157 {
1158         int i;
1159         unsigned int count = group_info->ngroups;
1160
1161         for (i = 0; i < group_info->nblocks; i++) {
1162                 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1163                 unsigned int len = cp_count * sizeof(*grouplist);
1164
1165                 if (copy_from_user(group_info->blocks[i], grouplist, len))
1166                         return -EFAULT;
1167
1168                 grouplist += NGROUPS_PER_BLOCK;
1169                 count -= cp_count;
1170         }
1171         return 0;
1172 }
1173
1174 /* a simple Shell sort */
1175 static void groups_sort(struct group_info *group_info)
1176 {
1177         int base, max, stride;
1178         int gidsetsize = group_info->ngroups;
1179
1180         for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1181                 ; /* nothing */
1182         stride /= 3;
1183
1184         while (stride) {
1185                 max = gidsetsize - stride;
1186                 for (base = 0; base < max; base++) {
1187                         int left = base;
1188                         int right = left + stride;
1189                         gid_t tmp = GROUP_AT(group_info, right);
1190
1191                         while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1192                                 GROUP_AT(group_info, right) =
1193                                     GROUP_AT(group_info, left);
1194                                 right = left;
1195                                 left -= stride;
1196                         }
1197                         GROUP_AT(group_info, right) = tmp;
1198                 }
1199                 stride /= 3;
1200         }
1201 }
1202
1203 /* a simple bsearch */
1204 int groups_search(struct group_info *group_info, gid_t grp)
1205 {
1206         unsigned int left, right;
1207
1208         if (!group_info)
1209                 return 0;
1210
1211         left = 0;
1212         right = group_info->ngroups;
1213         while (left < right) {
1214                 unsigned int mid = (left+right)/2;
1215                 int cmp = grp - GROUP_AT(group_info, mid);
1216                 if (cmp > 0)
1217                         left = mid + 1;
1218                 else if (cmp < 0)
1219                         right = mid;
1220                 else
1221                         return 1;
1222         }
1223         return 0;
1224 }
1225
1226 /* validate and set current->group_info */
1227 int set_current_groups(struct group_info *group_info)
1228 {
1229         int retval;
1230         struct group_info *old_info;
1231
1232         retval = security_task_setgroups(group_info);
1233         if (retval)
1234                 return retval;
1235
1236         groups_sort(group_info);
1237         get_group_info(group_info);
1238
1239         task_lock(current);
1240         old_info = current->group_info;
1241         current->group_info = group_info;
1242         task_unlock(current);
1243
1244         put_group_info(old_info);
1245
1246         return 0;
1247 }
1248
1249 EXPORT_SYMBOL(set_current_groups);
1250
1251 asmlinkage long sys_getgroups(int gidsetsize, gid_t __user *grouplist)
1252 {
1253         int i = 0;
1254
1255         /*
1256          *      SMP: Nobody else can change our grouplist. Thus we are
1257          *      safe.
1258          */
1259
1260         if (gidsetsize < 0)
1261                 return -EINVAL;
1262
1263         /* no need to grab task_lock here; it cannot change */
1264         i = current->group_info->ngroups;
1265         if (gidsetsize) {
1266                 if (i > gidsetsize) {
1267                         i = -EINVAL;
1268                         goto out;
1269                 }
1270                 if (groups_to_user(grouplist, current->group_info)) {
1271                         i = -EFAULT;
1272                         goto out;
1273                 }
1274         }
1275 out:
1276         return i;
1277 }
1278
1279 /*
1280  *      SMP: Our groups are copy-on-write. We can set them safely
1281  *      without another task interfering.
1282  */
1283  
1284 asmlinkage long sys_setgroups(int gidsetsize, gid_t __user *grouplist)
1285 {
1286         struct group_info *group_info;
1287         int retval;
1288
1289         if (!capable(CAP_SETGID))
1290                 return -EPERM;
1291         if ((unsigned)gidsetsize > NGROUPS_MAX)
1292                 return -EINVAL;
1293
1294         group_info = groups_alloc(gidsetsize);
1295         if (!group_info)
1296                 return -ENOMEM;
1297         retval = groups_from_user(group_info, grouplist);
1298         if (retval) {
1299                 put_group_info(group_info);
1300                 return retval;
1301         }
1302
1303         retval = set_current_groups(group_info);
1304         put_group_info(group_info);
1305
1306         return retval;
1307 }
1308
1309 /*
1310  * Check whether we're fsgid/egid or in the supplemental group..
1311  */
1312 int in_group_p(gid_t grp)
1313 {
1314         int retval = 1;
1315         if (grp != current->fsgid)
1316                 retval = groups_search(current->group_info, grp);
1317         return retval;
1318 }
1319
1320 EXPORT_SYMBOL(in_group_p);
1321
1322 int in_egroup_p(gid_t grp)
1323 {
1324         int retval = 1;
1325         if (grp != current->egid)
1326                 retval = groups_search(current->group_info, grp);
1327         return retval;
1328 }
1329
1330 EXPORT_SYMBOL(in_egroup_p);
1331
1332 DECLARE_RWSEM(uts_sem);
1333
1334 EXPORT_SYMBOL(uts_sem);
1335
1336 asmlinkage long sys_newuname(struct new_utsname __user * name)
1337 {
1338         int errno = 0;
1339
1340         down_read(&uts_sem);
1341         if (copy_to_user(name, utsname(), sizeof *name))
1342                 errno = -EFAULT;
1343         up_read(&uts_sem);
1344         return errno;
1345 }
1346
1347 asmlinkage long sys_sethostname(char __user *name, int len)
1348 {
1349         int errno;
1350         char tmp[__NEW_UTS_LEN];
1351
1352         if (!capable(CAP_SYS_ADMIN))
1353                 return -EPERM;
1354         if (len < 0 || len > __NEW_UTS_LEN)
1355                 return -EINVAL;
1356         down_write(&uts_sem);
1357         errno = -EFAULT;
1358         if (!copy_from_user(tmp, name, len)) {
1359                 memcpy(utsname()->nodename, tmp, len);
1360                 utsname()->nodename[len] = 0;
1361                 errno = 0;
1362         }
1363         up_write(&uts_sem);
1364         return errno;
1365 }
1366
1367 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1368
1369 asmlinkage long sys_gethostname(char __user *name, int len)
1370 {
1371         int i, errno;
1372
1373         if (len < 0)
1374                 return -EINVAL;
1375         down_read(&uts_sem);
1376         i = 1 + strlen(utsname()->nodename);
1377         if (i > len)
1378                 i = len;
1379         errno = 0;
1380         if (copy_to_user(name, utsname()->nodename, i))
1381                 errno = -EFAULT;
1382         up_read(&uts_sem);
1383         return errno;
1384 }
1385
1386 #endif
1387
1388 /*
1389  * Only setdomainname; getdomainname can be implemented by calling
1390  * uname()
1391  */
1392 asmlinkage long sys_setdomainname(char __user *name, int len)
1393 {
1394         int errno;
1395         char tmp[__NEW_UTS_LEN];
1396
1397         if (!capable(CAP_SYS_ADMIN))
1398                 return -EPERM;
1399         if (len < 0 || len > __NEW_UTS_LEN)
1400                 return -EINVAL;
1401
1402         down_write(&uts_sem);
1403         errno = -EFAULT;
1404         if (!copy_from_user(tmp, name, len)) {
1405                 memcpy(utsname()->domainname, tmp, len);
1406                 utsname()->domainname[len] = 0;
1407                 errno = 0;
1408         }
1409         up_write(&uts_sem);
1410         return errno;
1411 }
1412
1413 asmlinkage long sys_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1414 {
1415         if (resource >= RLIM_NLIMITS)
1416                 return -EINVAL;
1417         else {
1418                 struct rlimit value;
1419                 task_lock(current->group_leader);
1420                 value = current->signal->rlim[resource];
1421                 task_unlock(current->group_leader);
1422                 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1423         }
1424 }
1425
1426 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1427
1428 /*
1429  *      Back compatibility for getrlimit. Needed for some apps.
1430  */
1431  
1432 asmlinkage long sys_old_getrlimit(unsigned int resource, struct rlimit __user *rlim)
1433 {
1434         struct rlimit x;
1435         if (resource >= RLIM_NLIMITS)
1436                 return -EINVAL;
1437
1438         task_lock(current->group_leader);
1439         x = current->signal->rlim[resource];
1440         task_unlock(current->group_leader);
1441         if (x.rlim_cur > 0x7FFFFFFF)
1442                 x.rlim_cur = 0x7FFFFFFF;
1443         if (x.rlim_max > 0x7FFFFFFF)
1444                 x.rlim_max = 0x7FFFFFFF;
1445         return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1446 }
1447
1448 #endif
1449
1450 asmlinkage long sys_setrlimit(unsigned int resource, struct rlimit __user *rlim)
1451 {
1452         struct rlimit new_rlim, *old_rlim;
1453         unsigned long it_prof_secs;
1454         int retval;
1455
1456         if (resource >= RLIM_NLIMITS)
1457                 return -EINVAL;
1458         if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1459                 return -EFAULT;
1460         if (new_rlim.rlim_cur > new_rlim.rlim_max)
1461                 return -EINVAL;
1462         old_rlim = current->signal->rlim + resource;
1463         if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1464             !capable(CAP_SYS_RESOURCE))
1465                 return -EPERM;
1466         if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1467                 return -EPERM;
1468
1469         retval = security_task_setrlimit(resource, &new_rlim);
1470         if (retval)
1471                 return retval;
1472
1473         if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1474                 /*
1475                  * The caller is asking for an immediate RLIMIT_CPU
1476                  * expiry.  But we use the zero value to mean "it was
1477                  * never set".  So let's cheat and make it one second
1478                  * instead
1479                  */
1480                 new_rlim.rlim_cur = 1;
1481         }
1482
1483         task_lock(current->group_leader);
1484         *old_rlim = new_rlim;
1485         task_unlock(current->group_leader);
1486
1487         if (resource != RLIMIT_CPU)
1488                 goto out;
1489
1490         /*
1491          * RLIMIT_CPU handling.   Note that the kernel fails to return an error
1492          * code if it rejected the user's attempt to set RLIMIT_CPU.  This is a
1493          * very long-standing error, and fixing it now risks breakage of
1494          * applications, so we live with it
1495          */
1496         if (new_rlim.rlim_cur == RLIM_INFINITY)
1497                 goto out;
1498
1499         it_prof_secs = cputime_to_secs(current->signal->it_prof_expires);
1500         if (it_prof_secs == 0 || new_rlim.rlim_cur <= it_prof_secs) {
1501                 unsigned long rlim_cur = new_rlim.rlim_cur;
1502                 cputime_t cputime;
1503
1504                 cputime = secs_to_cputime(rlim_cur);
1505                 read_lock(&tasklist_lock);
1506                 spin_lock_irq(&current->sighand->siglock);
1507                 set_process_cpu_timer(current, CPUCLOCK_PROF, &cputime, NULL);
1508                 spin_unlock_irq(&current->sighand->siglock);
1509                 read_unlock(&tasklist_lock);
1510         }
1511 out:
1512         return 0;
1513 }
1514
1515 /*
1516  * It would make sense to put struct rusage in the task_struct,
1517  * except that would make the task_struct be *really big*.  After
1518  * task_struct gets moved into malloc'ed memory, it would
1519  * make sense to do this.  It will make moving the rest of the information
1520  * a lot simpler!  (Which we're not doing right now because we're not
1521  * measuring them yet).
1522  *
1523  * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1524  * races with threads incrementing their own counters.  But since word
1525  * reads are atomic, we either get new values or old values and we don't
1526  * care which for the sums.  We always take the siglock to protect reading
1527  * the c* fields from p->signal from races with exit.c updating those
1528  * fields when reaping, so a sample either gets all the additions of a
1529  * given child after it's reaped, or none so this sample is before reaping.
1530  *
1531  * Locking:
1532  * We need to take the siglock for CHILDEREN, SELF and BOTH
1533  * for  the cases current multithreaded, non-current single threaded
1534  * non-current multithreaded.  Thread traversal is now safe with
1535  * the siglock held.
1536  * Strictly speaking, we donot need to take the siglock if we are current and
1537  * single threaded,  as no one else can take our signal_struct away, no one
1538  * else can  reap the  children to update signal->c* counters, and no one else
1539  * can race with the signal-> fields. If we do not take any lock, the
1540  * signal-> fields could be read out of order while another thread was just
1541  * exiting. So we should  place a read memory barrier when we avoid the lock.
1542  * On the writer side,  write memory barrier is implied in  __exit_signal
1543  * as __exit_signal releases  the siglock spinlock after updating the signal->
1544  * fields. But we don't do this yet to keep things simple.
1545  *
1546  */
1547
1548 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1549 {
1550         struct task_struct *t;
1551         unsigned long flags;
1552         cputime_t utime, stime;
1553
1554         memset((char *) r, 0, sizeof *r);
1555         utime = stime = cputime_zero;
1556
1557         rcu_read_lock();
1558         if (!lock_task_sighand(p, &flags)) {
1559                 rcu_read_unlock();
1560                 return;
1561         }
1562
1563         switch (who) {
1564                 case RUSAGE_BOTH:
1565                 case RUSAGE_CHILDREN:
1566                         utime = p->signal->cutime;
1567                         stime = p->signal->cstime;
1568                         r->ru_nvcsw = p->signal->cnvcsw;
1569                         r->ru_nivcsw = p->signal->cnivcsw;
1570                         r->ru_minflt = p->signal->cmin_flt;
1571                         r->ru_majflt = p->signal->cmaj_flt;
1572                         r->ru_inblock = p->signal->cinblock;
1573                         r->ru_oublock = p->signal->coublock;
1574
1575                         if (who == RUSAGE_CHILDREN)
1576                                 break;
1577
1578                 case RUSAGE_SELF:
1579                         utime = cputime_add(utime, p->signal->utime);
1580                         stime = cputime_add(stime, p->signal->stime);
1581                         r->ru_nvcsw += p->signal->nvcsw;
1582                         r->ru_nivcsw += p->signal->nivcsw;
1583                         r->ru_minflt += p->signal->min_flt;
1584                         r->ru_majflt += p->signal->maj_flt;
1585                         r->ru_inblock += p->signal->inblock;
1586                         r->ru_oublock += p->signal->oublock;
1587                         t = p;
1588                         do {
1589                                 utime = cputime_add(utime, t->utime);
1590                                 stime = cputime_add(stime, t->stime);
1591                                 r->ru_nvcsw += t->nvcsw;
1592                                 r->ru_nivcsw += t->nivcsw;
1593                                 r->ru_minflt += t->min_flt;
1594                                 r->ru_majflt += t->maj_flt;
1595                                 r->ru_inblock += task_io_get_inblock(t);
1596                                 r->ru_oublock += task_io_get_oublock(t);
1597                                 t = next_thread(t);
1598                         } while (t != p);
1599                         break;
1600
1601                 default:
1602                         BUG();
1603         }
1604
1605         unlock_task_sighand(p, &flags);
1606         rcu_read_unlock();
1607
1608         cputime_to_timeval(utime, &r->ru_utime);
1609         cputime_to_timeval(stime, &r->ru_stime);
1610 }
1611
1612 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1613 {
1614         struct rusage r;
1615         k_getrusage(p, who, &r);
1616         return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1617 }
1618
1619 asmlinkage long sys_getrusage(int who, struct rusage __user *ru)
1620 {
1621         if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN)
1622                 return -EINVAL;
1623         return getrusage(current, who, ru);
1624 }
1625
1626 asmlinkage long sys_umask(int mask)
1627 {
1628         mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1629         return mask;
1630 }
1631
1632 asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1633                           unsigned long arg4, unsigned long arg5)
1634 {
1635         long uninitialized_var(error);
1636
1637         if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error))
1638                 return error;
1639
1640         switch (option) {
1641                 case PR_SET_PDEATHSIG:
1642                         if (!valid_signal(arg2)) {
1643                                 error = -EINVAL;
1644                                 break;
1645                         }
1646                         current->pdeath_signal = arg2;
1647                         break;
1648                 case PR_GET_PDEATHSIG:
1649                         error = put_user(current->pdeath_signal, (int __user *)arg2);
1650                         break;
1651                 case PR_GET_DUMPABLE:
1652                         error = get_dumpable(current->mm);
1653                         break;
1654                 case PR_SET_DUMPABLE:
1655                         if (arg2 < 0 || arg2 > 1) {
1656                                 error = -EINVAL;
1657                                 break;
1658                         }
1659                         set_dumpable(current->mm, arg2);
1660                         break;
1661
1662                 case PR_SET_UNALIGN:
1663                         error = SET_UNALIGN_CTL(current, arg2);
1664                         break;
1665                 case PR_GET_UNALIGN:
1666                         error = GET_UNALIGN_CTL(current, arg2);
1667                         break;
1668                 case PR_SET_FPEMU:
1669                         error = SET_FPEMU_CTL(current, arg2);
1670                         break;
1671                 case PR_GET_FPEMU:
1672                         error = GET_FPEMU_CTL(current, arg2);
1673                         break;
1674                 case PR_SET_FPEXC:
1675                         error = SET_FPEXC_CTL(current, arg2);
1676                         break;
1677                 case PR_GET_FPEXC:
1678                         error = GET_FPEXC_CTL(current, arg2);
1679                         break;
1680                 case PR_GET_TIMING:
1681                         error = PR_TIMING_STATISTICAL;
1682                         break;
1683                 case PR_SET_TIMING:
1684                         if (arg2 == PR_TIMING_STATISTICAL)
1685                                 error = 0;
1686                         else
1687                                 error = -EINVAL;
1688                         break;
1689
1690                 case PR_SET_NAME: {
1691                         struct task_struct *me = current;
1692                         unsigned char ncomm[sizeof(me->comm)];
1693
1694                         ncomm[sizeof(me->comm)-1] = 0;
1695                         if (strncpy_from_user(ncomm, (char __user *)arg2,
1696                                                 sizeof(me->comm)-1) < 0)
1697                                 return -EFAULT;
1698                         set_task_comm(me, ncomm);
1699                         return 0;
1700                 }
1701                 case PR_GET_NAME: {
1702                         struct task_struct *me = current;
1703                         unsigned char tcomm[sizeof(me->comm)];
1704
1705                         get_task_comm(tcomm, me);
1706                         if (copy_to_user((char __user *)arg2, tcomm, sizeof(tcomm)))
1707                                 return -EFAULT;
1708                         return 0;
1709                 }
1710                 case PR_GET_ENDIAN:
1711                         error = GET_ENDIAN(current, arg2);
1712                         break;
1713                 case PR_SET_ENDIAN:
1714                         error = SET_ENDIAN(current, arg2);
1715                         break;
1716
1717                 case PR_GET_SECCOMP:
1718                         error = prctl_get_seccomp();
1719                         break;
1720                 case PR_SET_SECCOMP:
1721                         error = prctl_set_seccomp(arg2);
1722                         break;
1723                 case PR_GET_TSC:
1724                         error = GET_TSC_CTL(arg2);
1725                         break;
1726                 case PR_SET_TSC:
1727                         error = SET_TSC_CTL(arg2);
1728                         break;
1729                 default:
1730                         error = -EINVAL;
1731                         break;
1732         }
1733         return error;
1734 }
1735
1736 asmlinkage long sys_getcpu(unsigned __user *cpup, unsigned __user *nodep,
1737                            struct getcpu_cache __user *unused)
1738 {
1739         int err = 0;
1740         int cpu = raw_smp_processor_id();
1741         if (cpup)
1742                 err |= put_user(cpu, cpup);
1743         if (nodep)
1744                 err |= put_user(cpu_to_node(cpu), nodep);
1745         return err ? -EFAULT : 0;
1746 }
1747
1748 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1749
1750 static void argv_cleanup(char **argv, char **envp)
1751 {
1752         argv_free(argv);
1753 }
1754
1755 /**
1756  * orderly_poweroff - Trigger an orderly system poweroff
1757  * @force: force poweroff if command execution fails
1758  *
1759  * This may be called from any context to trigger a system shutdown.
1760  * If the orderly shutdown fails, it will force an immediate shutdown.
1761  */
1762 int orderly_poweroff(bool force)
1763 {
1764         int argc;
1765         char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1766         static char *envp[] = {
1767                 "HOME=/",
1768                 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1769                 NULL
1770         };
1771         int ret = -ENOMEM;
1772         struct subprocess_info *info;
1773
1774         if (argv == NULL) {
1775                 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1776                        __func__, poweroff_cmd);
1777                 goto out;
1778         }
1779
1780         info = call_usermodehelper_setup(argv[0], argv, envp);
1781         if (info == NULL) {
1782                 argv_free(argv);
1783                 goto out;
1784         }
1785
1786         call_usermodehelper_setcleanup(info, argv_cleanup);
1787
1788         ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1789
1790   out:
1791         if (ret && force) {
1792                 printk(KERN_WARNING "Failed to start orderly shutdown: "
1793                        "forcing the issue\n");
1794
1795                 /* I guess this should try to kick off some daemon to
1796                    sync and poweroff asap.  Or not even bother syncing
1797                    if we're doing an emergency shutdown? */
1798                 emergency_sync();
1799                 kernel_power_off();
1800         }
1801
1802         return ret;
1803 }
1804 EXPORT_SYMBOL_GPL(orderly_poweroff);