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