?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[SCSI] pm80xx: remove unneeded NULL check
[linux-2.6.git] / kernel / sys.c
blobb95d3c72ba211a44c955ce3ae58b07fa72980cb1
1 /*
2 * linux/kernel/sys.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/export.h>
8 #include <linux/mm.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
14 #include <linux/fs.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.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 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
45
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/binfmts.h>
51
52 #include <linux/sched.h>
53 #include <linux/rcupdate.h>
54 #include <linux/uidgid.h>
55 #include <linux/cred.h>
56
57 #include <linux/kmsg_dump.h>
58 /* Move somewhere else to avoid recompiling? */
59 #include <generated/utsrelease.h>
60
61 #include <asm/uaccess.h>
62 #include <asm/io.h>
63 #include <asm/unistd.h>
64
65 #ifndef SET_UNALIGN_CTL
66 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
67 #endif
68 #ifndef GET_UNALIGN_CTL
69 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
70 #endif
71 #ifndef SET_FPEMU_CTL
72 # define SET_FPEMU_CTL(a,b) (-EINVAL)
73 #endif
74 #ifndef GET_FPEMU_CTL
75 # define GET_FPEMU_CTL(a,b) (-EINVAL)
76 #endif
77 #ifndef SET_FPEXC_CTL
78 # define SET_FPEXC_CTL(a,b) (-EINVAL)
79 #endif
80 #ifndef GET_FPEXC_CTL
81 # define GET_FPEXC_CTL(a,b) (-EINVAL)
82 #endif
83 #ifndef GET_ENDIAN
84 # define GET_ENDIAN(a,b) (-EINVAL)
85 #endif
86 #ifndef SET_ENDIAN
87 # define SET_ENDIAN(a,b) (-EINVAL)
88 #endif
89 #ifndef GET_TSC_CTL
90 # define GET_TSC_CTL(a) (-EINVAL)
91 #endif
92 #ifndef SET_TSC_CTL
93 # define SET_TSC_CTL(a) (-EINVAL)
94 #endif
95
96 /*
97 * this is where the system-wide overflow UID and GID are defined, for
98 * architectures that now have 32-bit UID/GID but didn't in the past
99 */
100
101 int overflowuid = DEFAULT_OVERFLOWUID;
102 int overflowgid = DEFAULT_OVERFLOWGID;
103
104 EXPORT_SYMBOL(overflowuid);
105 EXPORT_SYMBOL(overflowgid);
106
107 /*
108 * the same as above, but for filesystems which can only store a 16-bit
109 * UID and GID. as such, this is needed on all architectures
110 */
111
112 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
113 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
114
115 EXPORT_SYMBOL(fs_overflowuid);
116 EXPORT_SYMBOL(fs_overflowgid);
117
118 /*
119 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
120 */
121
122 int C_A_D = 1;
123 struct pid *cad_pid;
124 EXPORT_SYMBOL(cad_pid);
125
126 /*
127 * If set, this is used for preparing the system to power off.
128 */
129
130 void (*pm_power_off_prepare)(void);
131
132 /*
133 * Returns true if current's euid is same as p's uid or euid,
134 * or has CAP_SYS_NICE to p's user_ns.
135 *
136 * Called with rcu_read_lock, creds are safe
137 */
138 static bool set_one_prio_perm(struct task_struct *p)
139 {
140 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
141
142 if (uid_eq(pcred->uid, cred->euid) ||
143 uid_eq(pcred->euid, cred->euid))
144 return true;
145 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
146 return true;
147 return false;
148 }
149
150 /*
151 * set the priority of a task
152 * - the caller must hold the RCU read lock
153 */
154 static int set_one_prio(struct task_struct *p, int niceval, int error)
155 {
156 int no_nice;
157
158 if (!set_one_prio_perm(p)) {
159 error = -EPERM;
160 goto out;
161 }
162 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
163 error = -EACCES;
164 goto out;
165 }
166 no_nice = security_task_setnice(p, niceval);
167 if (no_nice) {
168 error = no_nice;
169 goto out;
170 }
171 if (error == -ESRCH)
172 error = 0;
173 set_user_nice(p, niceval);
174 out:
175 return error;
176 }
177
178 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
179 {
180 struct task_struct *g, *p;
181 struct user_struct *user;
182 const struct cred *cred = current_cred();
183 int error = -EINVAL;
184 struct pid *pgrp;
185 kuid_t uid;
186
187 if (which > PRIO_USER || which < PRIO_PROCESS)
188 goto out;
189
190 /* normalize: avoid signed division (rounding problems) */
191 error = -ESRCH;
192 if (niceval < -20)
193 niceval = -20;
194 if (niceval > 19)
195 niceval = 19;
196
197 rcu_read_lock();
198 read_lock(&tasklist_lock);
199 switch (which) {
200 case PRIO_PROCESS:
201 if (who)
202 p = find_task_by_vpid(who);
203 else
204 p = current;
205 if (p)
206 error = set_one_prio(p, niceval, error);
207 break;
208 case PRIO_PGRP:
209 if (who)
210 pgrp = find_vpid(who);
211 else
212 pgrp = task_pgrp(current);
213 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
214 error = set_one_prio(p, niceval, error);
215 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
216 break;
217 case PRIO_USER:
218 uid = make_kuid(cred->user_ns, who);
219 user = cred->user;
220 if (!who)
221 uid = cred->uid;
222 else if (!uid_eq(uid, cred->uid) &&
223 !(user = find_user(uid)))
224 goto out_unlock; /* No processes for this user */
225
226 do_each_thread(g, p) {
227 if (uid_eq(task_uid(p), uid))
228 error = set_one_prio(p, niceval, error);
229 } while_each_thread(g, p);
230 if (!uid_eq(uid, cred->uid))
231 free_uid(user); /* For find_user() */
232 break;
233 }
234 out_unlock:
235 read_unlock(&tasklist_lock);
236 rcu_read_unlock();
237 out:
238 return error;
239 }
240
241 /*
242 * Ugh. To avoid negative return values, "getpriority()" will
243 * not return the normal nice-value, but a negated value that
244 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
245 * to stay compatible.
246 */
247 SYSCALL_DEFINE2(getpriority, int, which, int, who)
248 {
249 struct task_struct *g, *p;
250 struct user_struct *user;
251 const struct cred *cred = current_cred();
252 long niceval, retval = -ESRCH;
253 struct pid *pgrp;
254 kuid_t uid;
255
256 if (which > PRIO_USER || which < PRIO_PROCESS)
257 return -EINVAL;
258
259 rcu_read_lock();
260 read_lock(&tasklist_lock);
261 switch (which) {
262 case PRIO_PROCESS:
263 if (who)
264 p = find_task_by_vpid(who);
265 else
266 p = current;
267 if (p) {
268 niceval = 20 - task_nice(p);
269 if (niceval > retval)
270 retval = niceval;
271 }
272 break;
273 case PRIO_PGRP:
274 if (who)
275 pgrp = find_vpid(who);
276 else
277 pgrp = task_pgrp(current);
278 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
279 niceval = 20 - task_nice(p);
280 if (niceval > retval)
281 retval = niceval;
282 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
283 break;
284 case PRIO_USER:
285 uid = make_kuid(cred->user_ns, who);
286 user = cred->user;
287 if (!who)
288 uid = cred->uid;
289 else if (!uid_eq(uid, cred->uid) &&
290 !(user = find_user(uid)))
291 goto out_unlock; /* No processes for this user */
292
293 do_each_thread(g, p) {
294 if (uid_eq(task_uid(p), uid)) {
295 niceval = 20 - task_nice(p);
296 if (niceval > retval)
297 retval = niceval;
298 }
299 } while_each_thread(g, p);
300 if (!uid_eq(uid, cred->uid))
301 free_uid(user); /* for find_user() */
302 break;
303 }
304 out_unlock:
305 read_unlock(&tasklist_lock);
306 rcu_read_unlock();
307
308 return retval;
309 }
310
311 /**
312 * emergency_restart - reboot the system
313 *
314 * Without shutting down any hardware or taking any locks
315 * reboot the system. This is called when we know we are in
316 * trouble so this is our best effort to reboot. This is
317 * safe to call in interrupt context.
318 */
319 void emergency_restart(void)
320 {
321 kmsg_dump(KMSG_DUMP_EMERG);
322 machine_emergency_restart();
323 }
324 EXPORT_SYMBOL_GPL(emergency_restart);
325
326 void kernel_restart_prepare(char *cmd)
327 {
328 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
329 system_state = SYSTEM_RESTART;
330 usermodehelper_disable();
331 device_shutdown();
332 }
333
334 /**
335 * register_reboot_notifier - Register function to be called at reboot time
336 * @nb: Info about notifier function to be called
337 *
338 * Registers a function with the list of functions
339 * to be called at reboot time.
340 *
341 * Currently always returns zero, as blocking_notifier_chain_register()
342 * always returns zero.
343 */
344 int register_reboot_notifier(struct notifier_block *nb)
345 {
346 return blocking_notifier_chain_register(&reboot_notifier_list, nb);
347 }
348 EXPORT_SYMBOL(register_reboot_notifier);
349
350 /**
351 * unregister_reboot_notifier - Unregister previously registered reboot notifier
352 * @nb: Hook to be unregistered
353 *
354 * Unregisters a previously registered reboot
355 * notifier function.
356 *
357 * Returns zero on success, or %-ENOENT on failure.
358 */
359 int unregister_reboot_notifier(struct notifier_block *nb)
360 {
361 return blocking_notifier_chain_unregister(&reboot_notifier_list, nb);
362 }
363 EXPORT_SYMBOL(unregister_reboot_notifier);
364
365 /**
366 * kernel_restart - reboot the system
367 * @cmd: pointer to buffer containing command to execute for restart
368 * or %NULL
369 *
370 * Shutdown everything and perform a clean reboot.
371 * This is not safe to call in interrupt context.
372 */
373 void kernel_restart(char *cmd)
374 {
375 kernel_restart_prepare(cmd);
376 disable_nonboot_cpus();
377 syscore_shutdown();
378 if (!cmd)
379 printk(KERN_EMERG "Restarting system.\n");
380 else
381 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
382 kmsg_dump(KMSG_DUMP_RESTART);
383 machine_restart(cmd);
384 }
385 EXPORT_SYMBOL_GPL(kernel_restart);
386
387 static void kernel_shutdown_prepare(enum system_states state)
388 {
389 blocking_notifier_call_chain(&reboot_notifier_list,
390 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
391 system_state = state;
392 usermodehelper_disable();
393 device_shutdown();
394 }
395 /**
396 * kernel_halt - halt the system
397 *
398 * Shutdown everything and perform a clean system halt.
399 */
400 void kernel_halt(void)
401 {
402 kernel_shutdown_prepare(SYSTEM_HALT);
403 disable_nonboot_cpus();
404 syscore_shutdown();
405 printk(KERN_EMERG "System halted.\n");
406 kmsg_dump(KMSG_DUMP_HALT);
407 machine_halt();
408 }
409
410 EXPORT_SYMBOL_GPL(kernel_halt);
411
412 /**
413 * kernel_power_off - power_off the system
414 *
415 * Shutdown everything and perform a clean system power_off.
416 */
417 void kernel_power_off(void)
418 {
419 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
420 if (pm_power_off_prepare)
421 pm_power_off_prepare();
422 disable_nonboot_cpus();
423 syscore_shutdown();
424 printk(KERN_EMERG "Power down.\n");
425 kmsg_dump(KMSG_DUMP_POWEROFF);
426 machine_power_off();
427 }
428 EXPORT_SYMBOL_GPL(kernel_power_off);
429
430 static DEFINE_MUTEX(reboot_mutex);
431
432 /*
433 * Reboot system call: for obvious reasons only root may call it,
434 * and even root needs to set up some magic numbers in the registers
435 * so that some mistake won't make this reboot the whole machine.
436 * You can also set the meaning of the ctrl-alt-del-key here.
437 *
438 * reboot doesn't sync: do that yourself before calling this.
439 */
440 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
441 void __user *, arg)
442 {
443 struct pid_namespace *pid_ns = task_active_pid_ns(current);
444 char buffer[256];
445 int ret = 0;
446
447 /* We only trust the superuser with rebooting the system. */
448 if (!ns_capable(pid_ns->user_ns, CAP_SYS_BOOT))
449 return -EPERM;
450
451 /* For safety, we require "magic" arguments. */
452 if (magic1 != LINUX_REBOOT_MAGIC1 ||
453 (magic2 != LINUX_REBOOT_MAGIC2 &&
454 magic2 != LINUX_REBOOT_MAGIC2A &&
455 magic2 != LINUX_REBOOT_MAGIC2B &&
456 magic2 != LINUX_REBOOT_MAGIC2C))
457 return -EINVAL;
458
459 /*
460 * If pid namespaces are enabled and the current task is in a child
461 * pid_namespace, the command is handled by reboot_pid_ns() which will
462 * call do_exit().
463 */
464 ret = reboot_pid_ns(pid_ns, cmd);
465 if (ret)
466 return ret;
467
468 /* Instead of trying to make the power_off code look like
469 * halt when pm_power_off is not set do it the easy way.
470 */
471 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
472 cmd = LINUX_REBOOT_CMD_HALT;
473
474 mutex_lock(&reboot_mutex);
475 switch (cmd) {
476 case LINUX_REBOOT_CMD_RESTART:
477 kernel_restart(NULL);
478 break;
479
480 case LINUX_REBOOT_CMD_CAD_ON:
481 C_A_D = 1;
482 break;
483
484 case LINUX_REBOOT_CMD_CAD_OFF:
485 C_A_D = 0;
486 break;
487
488 case LINUX_REBOOT_CMD_HALT:
489 kernel_halt();
490 do_exit(0);
491 panic("cannot halt");
492
493 case LINUX_REBOOT_CMD_POWER_OFF:
494 kernel_power_off();
495 do_exit(0);
496 break;
497
498 case LINUX_REBOOT_CMD_RESTART2:
499 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
500 ret = -EFAULT;
501 break;
502 }
503 buffer[sizeof(buffer) - 1] = '\0';
504
505 kernel_restart(buffer);
506 break;
507
508 #ifdef CONFIG_KEXEC
509 case LINUX_REBOOT_CMD_KEXEC:
510 ret = kernel_kexec();
511 break;
512 #endif
513
514 #ifdef CONFIG_HIBERNATION
515 case LINUX_REBOOT_CMD_SW_SUSPEND:
516 ret = hibernate();
517 break;
518 #endif
519
520 default:
521 ret = -EINVAL;
522 break;
523 }
524 mutex_unlock(&reboot_mutex);
525 return ret;
526 }
527
528 static void deferred_cad(struct work_struct *dummy)
529 {
530 kernel_restart(NULL);
531 }
532
533 /*
534 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
535 * As it's called within an interrupt, it may NOT sync: the only choice
536 * is whether to reboot at once, or just ignore the ctrl-alt-del.
537 */
538 void ctrl_alt_del(void)
539 {
540 static DECLARE_WORK(cad_work, deferred_cad);
541
542 if (C_A_D)
543 schedule_work(&cad_work);
544 else
545 kill_cad_pid(SIGINT, 1);
546 }
547
548 /*
549 * Unprivileged users may change the real gid to the effective gid
550 * or vice versa. (BSD-style)
551 *
552 * If you set the real gid at all, or set the effective gid to a value not
553 * equal to the real gid, then the saved gid is set to the new effective gid.
554 *
555 * This makes it possible for a setgid program to completely drop its
556 * privileges, which is often a useful assertion to make when you are doing
557 * a security audit over a program.
558 *
559 * The general idea is that a program which uses just setregid() will be
560 * 100% compatible with BSD. A program which uses just setgid() will be
561 * 100% compatible with POSIX with saved IDs.
562 *
563 * SMP: There are not races, the GIDs are checked only by filesystem
564 * operations (as far as semantic preservation is concerned).
565 */
566 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
567 {
568 struct user_namespace *ns = current_user_ns();
569 const struct cred *old;
570 struct cred *new;
571 int retval;
572 kgid_t krgid, kegid;
573
574 krgid = make_kgid(ns, rgid);
575 kegid = make_kgid(ns, egid);
576
577 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
578 return -EINVAL;
579 if ((egid != (gid_t) -1) && !gid_valid(kegid))
580 return -EINVAL;
581
582 new = prepare_creds();
583 if (!new)
584 return -ENOMEM;
585 old = current_cred();
586
587 retval = -EPERM;
588 if (rgid != (gid_t) -1) {
589 if (gid_eq(old->gid, krgid) ||
590 gid_eq(old->egid, krgid) ||
591 nsown_capable(CAP_SETGID))
592 new->gid = krgid;
593 else
594 goto error;
595 }
596 if (egid != (gid_t) -1) {
597 if (gid_eq(old->gid, kegid) ||
598 gid_eq(old->egid, kegid) ||
599 gid_eq(old->sgid, kegid) ||
600 nsown_capable(CAP_SETGID))
601 new->egid = kegid;
602 else
603 goto error;
604 }
605
606 if (rgid != (gid_t) -1 ||
607 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
608 new->sgid = new->egid;
609 new->fsgid = new->egid;
610
611 return commit_creds(new);
612
613 error:
614 abort_creds(new);
615 return retval;
616 }
617
618 /*
619 * setgid() is implemented like SysV w/ SAVED_IDS
620 *
621 * SMP: Same implicit races as above.
622 */
623 SYSCALL_DEFINE1(setgid, gid_t, gid)
624 {
625 struct user_namespace *ns = current_user_ns();
626 const struct cred *old;
627 struct cred *new;
628 int retval;
629 kgid_t kgid;
630
631 kgid = make_kgid(ns, gid);
632 if (!gid_valid(kgid))
633 return -EINVAL;
634
635 new = prepare_creds();
636 if (!new)
637 return -ENOMEM;
638 old = current_cred();
639
640 retval = -EPERM;
641 if (nsown_capable(CAP_SETGID))
642 new->gid = new->egid = new->sgid = new->fsgid = kgid;
643 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
644 new->egid = new->fsgid = kgid;
645 else
646 goto error;
647
648 return commit_creds(new);
649
650 error:
651 abort_creds(new);
652 return retval;
653 }
654
655 /*
656 * change the user struct in a credentials set to match the new UID
657 */
658 static int set_user(struct cred *new)
659 {
660 struct user_struct *new_user;
661
662 new_user = alloc_uid(new->uid);
663 if (!new_user)
664 return -EAGAIN;
665
666 /*
667 * We don't fail in case of NPROC limit excess here because too many
668 * poorly written programs don't check set*uid() return code, assuming
669 * it never fails if called by root. We may still enforce NPROC limit
670 * for programs doing set*uid()+execve() by harmlessly deferring the
671 * failure to the execve() stage.
672 */
673 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
674 new_user != INIT_USER)
675 current->flags |= PF_NPROC_EXCEEDED;
676 else
677 current->flags &= ~PF_NPROC_EXCEEDED;
678
679 free_uid(new->user);
680 new->user = new_user;
681 return 0;
682 }
683
684 /*
685 * Unprivileged users may change the real uid to the effective uid
686 * or vice versa. (BSD-style)
687 *
688 * If you set the real uid at all, or set the effective uid to a value not
689 * equal to the real uid, then the saved uid is set to the new effective uid.
690 *
691 * This makes it possible for a setuid program to completely drop its
692 * privileges, which is often a useful assertion to make when you are doing
693 * a security audit over a program.
694 *
695 * The general idea is that a program which uses just setreuid() will be
696 * 100% compatible with BSD. A program which uses just setuid() will be
697 * 100% compatible with POSIX with saved IDs.
698 */
699 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
700 {
701 struct user_namespace *ns = current_user_ns();
702 const struct cred *old;
703 struct cred *new;
704 int retval;
705 kuid_t kruid, keuid;
706
707 kruid = make_kuid(ns, ruid);
708 keuid = make_kuid(ns, euid);
709
710 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
711 return -EINVAL;
712 if ((euid != (uid_t) -1) && !uid_valid(keuid))
713 return -EINVAL;
714
715 new = prepare_creds();
716 if (!new)
717 return -ENOMEM;
718 old = current_cred();
719
720 retval = -EPERM;
721 if (ruid != (uid_t) -1) {
722 new->uid = kruid;
723 if (!uid_eq(old->uid, kruid) &&
724 !uid_eq(old->euid, kruid) &&
725 !nsown_capable(CAP_SETUID))
726 goto error;
727 }
728
729 if (euid != (uid_t) -1) {
730 new->euid = keuid;
731 if (!uid_eq(old->uid, keuid) &&
732 !uid_eq(old->euid, keuid) &&
733 !uid_eq(old->suid, keuid) &&
734 !nsown_capable(CAP_SETUID))
735 goto error;
736 }
737
738 if (!uid_eq(new->uid, old->uid)) {
739 retval = set_user(new);
740 if (retval < 0)
741 goto error;
742 }
743 if (ruid != (uid_t) -1 ||
744 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
745 new->suid = new->euid;
746 new->fsuid = new->euid;
747
748 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
749 if (retval < 0)
750 goto error;
751
752 return commit_creds(new);
753
754 error:
755 abort_creds(new);
756 return retval;
757 }
758
759 /*
760 * setuid() is implemented like SysV with SAVED_IDS
761 *
762 * Note that SAVED_ID's is deficient in that a setuid root program
763 * like sendmail, for example, cannot set its uid to be a normal
764 * user and then switch back, because if you're root, setuid() sets
765 * the saved uid too. If you don't like this, blame the bright people
766 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
767 * will allow a root program to temporarily drop privileges and be able to
768 * regain them by swapping the real and effective uid.
769 */
770 SYSCALL_DEFINE1(setuid, uid_t, uid)
771 {
772 struct user_namespace *ns = current_user_ns();
773 const struct cred *old;
774 struct cred *new;
775 int retval;
776 kuid_t kuid;
777
778 kuid = make_kuid(ns, uid);
779 if (!uid_valid(kuid))
780 return -EINVAL;
781
782 new = prepare_creds();
783 if (!new)
784 return -ENOMEM;
785 old = current_cred();
786
787 retval = -EPERM;
788 if (nsown_capable(CAP_SETUID)) {
789 new->suid = new->uid = kuid;
790 if (!uid_eq(kuid, old->uid)) {
791 retval = set_user(new);
792 if (retval < 0)
793 goto error;
794 }
795 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
796 goto error;
797 }
798
799 new->fsuid = new->euid = kuid;
800
801 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
802 if (retval < 0)
803 goto error;
804
805 return commit_creds(new);
806
807 error:
808 abort_creds(new);
809 return retval;
810 }
811
812
813 /*
814 * This function implements a generic ability to update ruid, euid,
815 * and suid. This allows you to implement the 4.4 compatible seteuid().
816 */
817 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
818 {
819 struct user_namespace *ns = current_user_ns();
820 const struct cred *old;
821 struct cred *new;
822 int retval;
823 kuid_t kruid, keuid, ksuid;
824
825 kruid = make_kuid(ns, ruid);
826 keuid = make_kuid(ns, euid);
827 ksuid = make_kuid(ns, suid);
828
829 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
830 return -EINVAL;
831
832 if ((euid != (uid_t) -1) && !uid_valid(keuid))
833 return -EINVAL;
834
835 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
836 return -EINVAL;
837
838 new = prepare_creds();
839 if (!new)
840 return -ENOMEM;
841
842 old = current_cred();
843
844 retval = -EPERM;
845 if (!nsown_capable(CAP_SETUID)) {
846 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
847 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
848 goto error;
849 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
850 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
851 goto error;
852 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
853 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
854 goto error;
855 }
856
857 if (ruid != (uid_t) -1) {
858 new->uid = kruid;
859 if (!uid_eq(kruid, old->uid)) {
860 retval = set_user(new);
861 if (retval < 0)
862 goto error;
863 }
864 }
865 if (euid != (uid_t) -1)
866 new->euid = keuid;
867 if (suid != (uid_t) -1)
868 new->suid = ksuid;
869 new->fsuid = new->euid;
870
871 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
872 if (retval < 0)
873 goto error;
874
875 return commit_creds(new);
876
877 error:
878 abort_creds(new);
879 return retval;
880 }
881
882 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
883 {
884 const struct cred *cred = current_cred();
885 int retval;
886 uid_t ruid, euid, suid;
887
888 ruid = from_kuid_munged(cred->user_ns, cred->uid);
889 euid = from_kuid_munged(cred->user_ns, cred->euid);
890 suid = from_kuid_munged(cred->user_ns, cred->suid);
891
892 if (!(retval = put_user(ruid, ruidp)) &&
893 !(retval = put_user(euid, euidp)))
894 retval = put_user(suid, suidp);
895
896 return retval;
897 }
898
899 /*
900 * Same as above, but for rgid, egid, sgid.
901 */
902 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
903 {
904 struct user_namespace *ns = current_user_ns();
905 const struct cred *old;
906 struct cred *new;
907 int retval;
908 kgid_t krgid, kegid, ksgid;
909
910 krgid = make_kgid(ns, rgid);
911 kegid = make_kgid(ns, egid);
912 ksgid = make_kgid(ns, sgid);
913
914 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
915 return -EINVAL;
916 if ((egid != (gid_t) -1) && !gid_valid(kegid))
917 return -EINVAL;
918 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
919 return -EINVAL;
920
921 new = prepare_creds();
922 if (!new)
923 return -ENOMEM;
924 old = current_cred();
925
926 retval = -EPERM;
927 if (!nsown_capable(CAP_SETGID)) {
928 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
929 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
930 goto error;
931 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
932 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
933 goto error;
934 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
935 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
936 goto error;
937 }
938
939 if (rgid != (gid_t) -1)
940 new->gid = krgid;
941 if (egid != (gid_t) -1)
942 new->egid = kegid;
943 if (sgid != (gid_t) -1)
944 new->sgid = ksgid;
945 new->fsgid = new->egid;
946
947 return commit_creds(new);
948
949 error:
950 abort_creds(new);
951 return retval;
952 }
953
954 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
955 {
956 const struct cred *cred = current_cred();
957 int retval;
958 gid_t rgid, egid, sgid;
959
960 rgid = from_kgid_munged(cred->user_ns, cred->gid);
961 egid = from_kgid_munged(cred->user_ns, cred->egid);
962 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
963
964 if (!(retval = put_user(rgid, rgidp)) &&
965 !(retval = put_user(egid, egidp)))
966 retval = put_user(sgid, sgidp);
967
968 return retval;
969 }
970
971
972 /*
973 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
974 * is used for "access()" and for the NFS daemon (letting nfsd stay at
975 * whatever uid it wants to). It normally shadows "euid", except when
976 * explicitly set by setfsuid() or for access..
977 */
978 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
979 {
980 const struct cred *old;
981 struct cred *new;
982 uid_t old_fsuid;
983 kuid_t kuid;
984
985 old = current_cred();
986 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
987
988 kuid = make_kuid(old->user_ns, uid);
989 if (!uid_valid(kuid))
990 return old_fsuid;
991
992 new = prepare_creds();
993 if (!new)
994 return old_fsuid;
995
996 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
997 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
998 nsown_capable(CAP_SETUID)) {
999 if (!uid_eq(kuid, old->fsuid)) {
1000 new->fsuid = kuid;
1001 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
1002 goto change_okay;
1003 }
1004 }
1005
1006 abort_creds(new);
1007 return old_fsuid;
1008
1009 change_okay:
1010 commit_creds(new);
1011 return old_fsuid;
1012 }
1013
1014 /*
1015 * Samma på svenska..
1016 */
1017 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
1018 {
1019 const struct cred *old;
1020 struct cred *new;
1021 gid_t old_fsgid;
1022 kgid_t kgid;
1023
1024 old = current_cred();
1025 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
1026
1027 kgid = make_kgid(old->user_ns, gid);
1028 if (!gid_valid(kgid))
1029 return old_fsgid;
1030
1031 new = prepare_creds();
1032 if (!new)
1033 return old_fsgid;
1034
1035 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
1036 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
1037 nsown_capable(CAP_SETGID)) {
1038 if (!gid_eq(kgid, old->fsgid)) {
1039 new->fsgid = kgid;
1040 goto change_okay;
1041 }
1042 }
1043
1044 abort_creds(new);
1045 return old_fsgid;
1046
1047 change_okay:
1048 commit_creds(new);
1049 return old_fsgid;
1050 }
1051
1052 /**
1053 * sys_getpid - return the thread group id of the current process
1054 *
1055 * Note, despite the name, this returns the tgid not the pid. The tgid and
1056 * the pid are identical unless CLONE_THREAD was specified on clone() in
1057 * which case the tgid is the same in all threads of the same group.
1058 *
1059 * This is SMP safe as current->tgid does not change.
1060 */
1061 SYSCALL_DEFINE0(getpid)
1062 {
1063 return task_tgid_vnr(current);
1064 }
1065
1066 /* Thread ID - the internal kernel "pid" */
1067 SYSCALL_DEFINE0(gettid)
1068 {
1069 return task_pid_vnr(current);
1070 }
1071
1072 /*
1073 * Accessing ->real_parent is not SMP-safe, it could
1074 * change from under us. However, we can use a stale
1075 * value of ->real_parent under rcu_read_lock(), see
1076 * release_task()->call_rcu(delayed_put_task_struct).
1077 */
1078 SYSCALL_DEFINE0(getppid)
1079 {
1080 int pid;
1081
1082 rcu_read_lock();
1083 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
1084 rcu_read_unlock();
1085
1086 return pid;
1087 }
1088
1089 SYSCALL_DEFINE0(getuid)
1090 {
1091 /* Only we change this so SMP safe */
1092 return from_kuid_munged(current_user_ns(), current_uid());
1093 }
1094
1095 SYSCALL_DEFINE0(geteuid)
1096 {
1097 /* Only we change this so SMP safe */
1098 return from_kuid_munged(current_user_ns(), current_euid());
1099 }
1100
1101 SYSCALL_DEFINE0(getgid)
1102 {
1103 /* Only we change this so SMP safe */
1104 return from_kgid_munged(current_user_ns(), current_gid());
1105 }
1106
1107 SYSCALL_DEFINE0(getegid)
1108 {
1109 /* Only we change this so SMP safe */
1110 return from_kgid_munged(current_user_ns(), current_egid());
1111 }
1112
1113 void do_sys_times(struct tms *tms)
1114 {
1115 cputime_t tgutime, tgstime, cutime, cstime;
1116
1117 spin_lock_irq(&current->sighand->siglock);
1118 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1119 cutime = current->signal->cutime;
1120 cstime = current->signal->cstime;
1121 spin_unlock_irq(&current->sighand->siglock);
1122 tms->tms_utime = cputime_to_clock_t(tgutime);
1123 tms->tms_stime = cputime_to_clock_t(tgstime);
1124 tms->tms_cutime = cputime_to_clock_t(cutime);
1125 tms->tms_cstime = cputime_to_clock_t(cstime);
1126 }
1127
1128 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1129 {
1130 if (tbuf) {
1131 struct tms tmp;
1132
1133 do_sys_times(&tmp);
1134 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1135 return -EFAULT;
1136 }
1137 force_successful_syscall_return();
1138 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1139 }
1140
1141 /*
1142 * This needs some heavy checking ...
1143 * I just haven't the stomach for it. I also don't fully
1144 * understand sessions/pgrp etc. Let somebody who does explain it.
1145 *
1146 * OK, I think I have the protection semantics right.... this is really
1147 * only important on a multi-user system anyway, to make sure one user
1148 * can't send a signal to a process owned by another. -TYT, 12/12/91
1149 *
1150 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
1151 * LBT 04.03.94
1152 */
1153 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1154 {
1155 struct task_struct *p;
1156 struct task_struct *group_leader = current->group_leader;
1157 struct pid *pgrp;
1158 int err;
1159
1160 if (!pid)
1161 pid = task_pid_vnr(group_leader);
1162 if (!pgid)
1163 pgid = pid;
1164 if (pgid < 0)
1165 return -EINVAL;
1166 rcu_read_lock();
1167
1168 /* From this point forward we keep holding onto the tasklist lock
1169 * so that our parent does not change from under us. -DaveM
1170 */
1171 write_lock_irq(&tasklist_lock);
1172
1173 err = -ESRCH;
1174 p = find_task_by_vpid(pid);
1175 if (!p)
1176 goto out;
1177
1178 err = -EINVAL;
1179 if (!thread_group_leader(p))
1180 goto out;
1181
1182 if (same_thread_group(p->real_parent, group_leader)) {
1183 err = -EPERM;
1184 if (task_session(p) != task_session(group_leader))
1185 goto out;
1186 err = -EACCES;
1187 if (p->did_exec)
1188 goto out;
1189 } else {
1190 err = -ESRCH;
1191 if (p != group_leader)
1192 goto out;
1193 }
1194
1195 err = -EPERM;
1196 if (p->signal->leader)
1197 goto out;
1198
1199 pgrp = task_pid(p);
1200 if (pgid != pid) {
1201 struct task_struct *g;
1202
1203 pgrp = find_vpid(pgid);
1204 g = pid_task(pgrp, PIDTYPE_PGID);
1205 if (!g || task_session(g) != task_session(group_leader))
1206 goto out;
1207 }
1208
1209 err = security_task_setpgid(p, pgid);
1210 if (err)
1211 goto out;
1212
1213 if (task_pgrp(p) != pgrp)
1214 change_pid(p, PIDTYPE_PGID, pgrp);
1215
1216 err = 0;
1217 out:
1218 /* All paths lead to here, thus we are safe. -DaveM */
1219 write_unlock_irq(&tasklist_lock);
1220 rcu_read_unlock();
1221 return err;
1222 }
1223
1224 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1225 {
1226 struct task_struct *p;
1227 struct pid *grp;
1228 int retval;
1229
1230 rcu_read_lock();
1231 if (!pid)
1232 grp = task_pgrp(current);
1233 else {
1234 retval = -ESRCH;
1235 p = find_task_by_vpid(pid);
1236 if (!p)
1237 goto out;
1238 grp = task_pgrp(p);
1239 if (!grp)
1240 goto out;
1241
1242 retval = security_task_getpgid(p);
1243 if (retval)
1244 goto out;
1245 }
1246 retval = pid_vnr(grp);
1247 out:
1248 rcu_read_unlock();
1249 return retval;
1250 }
1251
1252 #ifdef __ARCH_WANT_SYS_GETPGRP
1253
1254 SYSCALL_DEFINE0(getpgrp)
1255 {
1256 return sys_getpgid(0);
1257 }
1258
1259 #endif
1260
1261 SYSCALL_DEFINE1(getsid, pid_t, pid)
1262 {
1263 struct task_struct *p;
1264 struct pid *sid;
1265 int retval;
1266
1267 rcu_read_lock();
1268 if (!pid)
1269 sid = task_session(current);
1270 else {
1271 retval = -ESRCH;
1272 p = find_task_by_vpid(pid);
1273 if (!p)
1274 goto out;
1275 sid = task_session(p);
1276 if (!sid)
1277 goto out;
1278
1279 retval = security_task_getsid(p);
1280 if (retval)
1281 goto out;
1282 }
1283 retval = pid_vnr(sid);
1284 out:
1285 rcu_read_unlock();
1286 return retval;
1287 }
1288
1289 SYSCALL_DEFINE0(setsid)
1290 {
1291 struct task_struct *group_leader = current->group_leader;
1292 struct pid *sid = task_pid(group_leader);
1293 pid_t session = pid_vnr(sid);
1294 int err = -EPERM;
1295
1296 write_lock_irq(&tasklist_lock);
1297 /* Fail if I am already a session leader */
1298 if (group_leader->signal->leader)
1299 goto out;
1300
1301 /* Fail if a process group id already exists that equals the
1302 * proposed session id.
1303 */
1304 if (pid_task(sid, PIDTYPE_PGID))
1305 goto out;
1306
1307 group_leader->signal->leader = 1;
1308 __set_special_pids(sid);
1309
1310 proc_clear_tty(group_leader);
1311
1312 err = session;
1313 out:
1314 write_unlock_irq(&tasklist_lock);
1315 if (err > 0) {
1316 proc_sid_connector(group_leader);
1317 sched_autogroup_create_attach(group_leader);
1318 }
1319 return err;
1320 }
1321
1322 DECLARE_RWSEM(uts_sem);
1323
1324 #ifdef COMPAT_UTS_MACHINE
1325 #define override_architecture(name) \
1326 (personality(current->personality) == PER_LINUX32 && \
1327 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1328 sizeof(COMPAT_UTS_MACHINE)))
1329 #else
1330 #define override_architecture(name) 0
1331 #endif
1332
1333 /*
1334 * Work around broken programs that cannot handle "Linux 3.0".
1335 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1336 */
1337 static int override_release(char __user *release, size_t len)
1338 {
1339 int ret = 0;
1340
1341 if (current->personality & UNAME26) {
1342 const char *rest = UTS_RELEASE;
1343 char buf[65] = { 0 };
1344 int ndots = 0;
1345 unsigned v;
1346 size_t copy;
1347
1348 while (*rest) {
1349 if (*rest == '.' && ++ndots >= 3)
1350 break;
1351 if (!isdigit(*rest) && *rest != '.')
1352 break;
1353 rest++;
1354 }
1355 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1356 copy = clamp_t(size_t, len, 1, sizeof(buf));
1357 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1358 ret = copy_to_user(release, buf, copy + 1);
1359 }
1360 return ret;
1361 }
1362
1363 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1364 {
1365 int errno = 0;
1366
1367 down_read(&uts_sem);
1368 if (copy_to_user(name, utsname(), sizeof *name))
1369 errno = -EFAULT;
1370 up_read(&uts_sem);
1371
1372 if (!errno && override_release(name->release, sizeof(name->release)))
1373 errno = -EFAULT;
1374 if (!errno && override_architecture(name))
1375 errno = -EFAULT;
1376 return errno;
1377 }
1378
1379 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1380 /*
1381 * Old cruft
1382 */
1383 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1384 {
1385 int error = 0;
1386
1387 if (!name)
1388 return -EFAULT;
1389
1390 down_read(&uts_sem);
1391 if (copy_to_user(name, utsname(), sizeof(*name)))
1392 error = -EFAULT;
1393 up_read(&uts_sem);
1394
1395 if (!error && override_release(name->release, sizeof(name->release)))
1396 error = -EFAULT;
1397 if (!error && override_architecture(name))
1398 error = -EFAULT;
1399 return error;
1400 }
1401
1402 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1403 {
1404 int error;
1405
1406 if (!name)
1407 return -EFAULT;
1408 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1409 return -EFAULT;
1410
1411 down_read(&uts_sem);
1412 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1413 __OLD_UTS_LEN);
1414 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1415 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1416 __OLD_UTS_LEN);
1417 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1418 error |= __copy_to_user(&name->release, &utsname()->release,
1419 __OLD_UTS_LEN);
1420 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1421 error |= __copy_to_user(&name->version, &utsname()->version,
1422 __OLD_UTS_LEN);
1423 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1424 error |= __copy_to_user(&name->machine, &utsname()->machine,
1425 __OLD_UTS_LEN);
1426 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1427 up_read(&uts_sem);
1428
1429 if (!error && override_architecture(name))
1430 error = -EFAULT;
1431 if (!error && override_release(name->release, sizeof(name->release)))
1432 error = -EFAULT;
1433 return error ? -EFAULT : 0;
1434 }
1435 #endif
1436
1437 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1438 {
1439 int errno;
1440 char tmp[__NEW_UTS_LEN];
1441
1442 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1443 return -EPERM;
1444
1445 if (len < 0 || len > __NEW_UTS_LEN)
1446 return -EINVAL;
1447 down_write(&uts_sem);
1448 errno = -EFAULT;
1449 if (!copy_from_user(tmp, name, len)) {
1450 struct new_utsname *u = utsname();
1451
1452 memcpy(u->nodename, tmp, len);
1453 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1454 errno = 0;
1455 uts_proc_notify(UTS_PROC_HOSTNAME);
1456 }
1457 up_write(&uts_sem);
1458 return errno;
1459 }
1460
1461 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1462
1463 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1464 {
1465 int i, errno;
1466 struct new_utsname *u;
1467
1468 if (len < 0)
1469 return -EINVAL;
1470 down_read(&uts_sem);
1471 u = utsname();
1472 i = 1 + strlen(u->nodename);
1473 if (i > len)
1474 i = len;
1475 errno = 0;
1476 if (copy_to_user(name, u->nodename, i))
1477 errno = -EFAULT;
1478 up_read(&uts_sem);
1479 return errno;
1480 }
1481
1482 #endif
1483
1484 /*
1485 * Only setdomainname; getdomainname can be implemented by calling
1486 * uname()
1487 */
1488 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1489 {
1490 int errno;
1491 char tmp[__NEW_UTS_LEN];
1492
1493 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1494 return -EPERM;
1495 if (len < 0 || len > __NEW_UTS_LEN)
1496 return -EINVAL;
1497
1498 down_write(&uts_sem);
1499 errno = -EFAULT;
1500 if (!copy_from_user(tmp, name, len)) {
1501 struct new_utsname *u = utsname();
1502
1503 memcpy(u->domainname, tmp, len);
1504 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1505 errno = 0;
1506 uts_proc_notify(UTS_PROC_DOMAINNAME);
1507 }
1508 up_write(&uts_sem);
1509 return errno;
1510 }
1511
1512 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1513 {
1514 struct rlimit value;
1515 int ret;
1516
1517 ret = do_prlimit(current, resource, NULL, &value);
1518 if (!ret)
1519 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1520
1521 return ret;
1522 }
1523
1524 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1525
1526 /*
1527 * Back compatibility for getrlimit. Needed for some apps.
1528 */
1529
1530 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1531 struct rlimit __user *, rlim)
1532 {
1533 struct rlimit x;
1534 if (resource >= RLIM_NLIMITS)
1535 return -EINVAL;
1536
1537 task_lock(current->group_leader);
1538 x = current->signal->rlim[resource];
1539 task_unlock(current->group_leader);
1540 if (x.rlim_cur > 0x7FFFFFFF)
1541 x.rlim_cur = 0x7FFFFFFF;
1542 if (x.rlim_max > 0x7FFFFFFF)
1543 x.rlim_max = 0x7FFFFFFF;
1544 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1545 }
1546
1547 #endif
1548
1549 static inline bool rlim64_is_infinity(__u64 rlim64)
1550 {
1551 #if BITS_PER_LONG < 64
1552 return rlim64 >= ULONG_MAX;
1553 #else
1554 return rlim64 == RLIM64_INFINITY;
1555 #endif
1556 }
1557
1558 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1559 {
1560 if (rlim->rlim_cur == RLIM_INFINITY)
1561 rlim64->rlim_cur = RLIM64_INFINITY;
1562 else
1563 rlim64->rlim_cur = rlim->rlim_cur;
1564 if (rlim->rlim_max == RLIM_INFINITY)
1565 rlim64->rlim_max = RLIM64_INFINITY;
1566 else
1567 rlim64->rlim_max = rlim->rlim_max;
1568 }
1569
1570 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1571 {
1572 if (rlim64_is_infinity(rlim64->rlim_cur))
1573 rlim->rlim_cur = RLIM_INFINITY;
1574 else
1575 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1576 if (rlim64_is_infinity(rlim64->rlim_max))
1577 rlim->rlim_max = RLIM_INFINITY;
1578 else
1579 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1580 }
1581
1582 /* make sure you are allowed to change @tsk limits before calling this */
1583 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1584 struct rlimit *new_rlim, struct rlimit *old_rlim)
1585 {
1586 struct rlimit *rlim;
1587 int retval = 0;
1588
1589 if (resource >= RLIM_NLIMITS)
1590 return -EINVAL;
1591 if (new_rlim) {
1592 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1593 return -EINVAL;
1594 if (resource == RLIMIT_NOFILE &&
1595 new_rlim->rlim_max > sysctl_nr_open)
1596 return -EPERM;
1597 }
1598
1599 /* protect tsk->signal and tsk->sighand from disappearing */
1600 read_lock(&tasklist_lock);
1601 if (!tsk->sighand) {
1602 retval = -ESRCH;
1603 goto out;
1604 }
1605
1606 rlim = tsk->signal->rlim + resource;
1607 task_lock(tsk->group_leader);
1608 if (new_rlim) {
1609 /* Keep the capable check against init_user_ns until
1610 cgroups can contain all limits */
1611 if (new_rlim->rlim_max > rlim->rlim_max &&
1612 !capable(CAP_SYS_RESOURCE))
1613 retval = -EPERM;
1614 if (!retval)
1615 retval = security_task_setrlimit(tsk->group_leader,
1616 resource, new_rlim);
1617 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1618 /*
1619 * The caller is asking for an immediate RLIMIT_CPU
1620 * expiry. But we use the zero value to mean "it was
1621 * never set". So let's cheat and make it one second
1622 * instead
1623 */
1624 new_rlim->rlim_cur = 1;
1625 }
1626 }
1627 if (!retval) {
1628 if (old_rlim)
1629 *old_rlim = *rlim;
1630 if (new_rlim)
1631 *rlim = *new_rlim;
1632 }
1633 task_unlock(tsk->group_leader);
1634
1635 /*
1636 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1637 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1638 * very long-standing error, and fixing it now risks breakage of
1639 * applications, so we live with it
1640 */
1641 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1642 new_rlim->rlim_cur != RLIM_INFINITY)
1643 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1644 out:
1645 read_unlock(&tasklist_lock);
1646 return retval;
1647 }
1648
1649 /* rcu lock must be held */
1650 static int check_prlimit_permission(struct task_struct *task)
1651 {
1652 const struct cred *cred = current_cred(), *tcred;
1653
1654 if (current == task)
1655 return 0;
1656
1657 tcred = __task_cred(task);
1658 if (uid_eq(cred->uid, tcred->euid) &&
1659 uid_eq(cred->uid, tcred->suid) &&
1660 uid_eq(cred->uid, tcred->uid) &&
1661 gid_eq(cred->gid, tcred->egid) &&
1662 gid_eq(cred->gid, tcred->sgid) &&
1663 gid_eq(cred->gid, tcred->gid))
1664 return 0;
1665 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1666 return 0;
1667
1668 return -EPERM;
1669 }
1670
1671 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1672 const struct rlimit64 __user *, new_rlim,
1673 struct rlimit64 __user *, old_rlim)
1674 {
1675 struct rlimit64 old64, new64;
1676 struct rlimit old, new;
1677 struct task_struct *tsk;
1678 int ret;
1679
1680 if (new_rlim) {
1681 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1682 return -EFAULT;
1683 rlim64_to_rlim(&new64, &new);
1684 }
1685
1686 rcu_read_lock();
1687 tsk = pid ? find_task_by_vpid(pid) : current;
1688 if (!tsk) {
1689 rcu_read_unlock();
1690 return -ESRCH;
1691 }
1692 ret = check_prlimit_permission(tsk);
1693 if (ret) {
1694 rcu_read_unlock();
1695 return ret;
1696 }
1697 get_task_struct(tsk);
1698 rcu_read_unlock();
1699
1700 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1701 old_rlim ? &old : NULL);
1702
1703 if (!ret && old_rlim) {
1704 rlim_to_rlim64(&old, &old64);
1705 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1706 ret = -EFAULT;
1707 }
1708
1709 put_task_struct(tsk);
1710 return ret;
1711 }
1712
1713 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1714 {
1715 struct rlimit new_rlim;
1716
1717 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1718 return -EFAULT;
1719 return do_prlimit(current, resource, &new_rlim, NULL);
1720 }
1721
1722 /*
1723 * It would make sense to put struct rusage in the task_struct,
1724 * except that would make the task_struct be *really big*. After
1725 * task_struct gets moved into malloc'ed memory, it would
1726 * make sense to do this. It will make moving the rest of the information
1727 * a lot simpler! (Which we're not doing right now because we're not
1728 * measuring them yet).
1729 *
1730 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1731 * races with threads incrementing their own counters. But since word
1732 * reads are atomic, we either get new values or old values and we don't
1733 * care which for the sums. We always take the siglock to protect reading
1734 * the c* fields from p->signal from races with exit.c updating those
1735 * fields when reaping, so a sample either gets all the additions of a
1736 * given child after it's reaped, or none so this sample is before reaping.
1737 *
1738 * Locking:
1739 * We need to take the siglock for CHILDEREN, SELF and BOTH
1740 * for the cases current multithreaded, non-current single threaded
1741 * non-current multithreaded. Thread traversal is now safe with
1742 * the siglock held.
1743 * Strictly speaking, we donot need to take the siglock if we are current and
1744 * single threaded, as no one else can take our signal_struct away, no one
1745 * else can reap the children to update signal->c* counters, and no one else
1746 * can race with the signal-> fields. If we do not take any lock, the
1747 * signal-> fields could be read out of order while another thread was just
1748 * exiting. So we should place a read memory barrier when we avoid the lock.
1749 * On the writer side, write memory barrier is implied in __exit_signal
1750 * as __exit_signal releases the siglock spinlock after updating the signal->
1751 * fields. But we don't do this yet to keep things simple.
1752 *
1753 */
1754
1755 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1756 {
1757 r->ru_nvcsw += t->nvcsw;
1758 r->ru_nivcsw += t->nivcsw;
1759 r->ru_minflt += t->min_flt;
1760 r->ru_majflt += t->maj_flt;
1761 r->ru_inblock += task_io_get_inblock(t);
1762 r->ru_oublock += task_io_get_oublock(t);
1763 }
1764
1765 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1766 {
1767 struct task_struct *t;
1768 unsigned long flags;
1769 cputime_t tgutime, tgstime, utime, stime;
1770 unsigned long maxrss = 0;
1771
1772 memset((char *) r, 0, sizeof *r);
1773 utime = stime = 0;
1774
1775 if (who == RUSAGE_THREAD) {
1776 task_cputime_adjusted(current, &utime, &stime);
1777 accumulate_thread_rusage(p, r);
1778 maxrss = p->signal->maxrss;
1779 goto out;
1780 }
1781
1782 if (!lock_task_sighand(p, &flags))
1783 return;
1784
1785 switch (who) {
1786 case RUSAGE_BOTH:
1787 case RUSAGE_CHILDREN:
1788 utime = p->signal->cutime;
1789 stime = p->signal->cstime;
1790 r->ru_nvcsw = p->signal->cnvcsw;
1791 r->ru_nivcsw = p->signal->cnivcsw;
1792 r->ru_minflt = p->signal->cmin_flt;
1793 r->ru_majflt = p->signal->cmaj_flt;
1794 r->ru_inblock = p->signal->cinblock;
1795 r->ru_oublock = p->signal->coublock;
1796 maxrss = p->signal->cmaxrss;
1797
1798 if (who == RUSAGE_CHILDREN)
1799 break;
1800
1801 case RUSAGE_SELF:
1802 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1803 utime += tgutime;
1804 stime += tgstime;
1805 r->ru_nvcsw += p->signal->nvcsw;
1806 r->ru_nivcsw += p->signal->nivcsw;
1807 r->ru_minflt += p->signal->min_flt;
1808 r->ru_majflt += p->signal->maj_flt;
1809 r->ru_inblock += p->signal->inblock;
1810 r->ru_oublock += p->signal->oublock;
1811 if (maxrss < p->signal->maxrss)
1812 maxrss = p->signal->maxrss;
1813 t = p;
1814 do {
1815 accumulate_thread_rusage(t, r);
1816 t = next_thread(t);
1817 } while (t != p);
1818 break;
1819
1820 default:
1821 BUG();
1822 }
1823 unlock_task_sighand(p, &flags);
1824
1825 out:
1826 cputime_to_timeval(utime, &r->ru_utime);
1827 cputime_to_timeval(stime, &r->ru_stime);
1828
1829 if (who != RUSAGE_CHILDREN) {
1830 struct mm_struct *mm = get_task_mm(p);
1831 if (mm) {
1832 setmax_mm_hiwater_rss(&maxrss, mm);
1833 mmput(mm);
1834 }
1835 }
1836 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1837 }
1838
1839 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1840 {
1841 struct rusage r;
1842 k_getrusage(p, who, &r);
1843 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1844 }
1845
1846 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1847 {
1848 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1849 who != RUSAGE_THREAD)
1850 return -EINVAL;
1851 return getrusage(current, who, ru);
1852 }
1853
1854 #ifdef CONFIG_COMPAT
1855 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1856 {
1857 struct rusage r;
1858
1859 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1860 who != RUSAGE_THREAD)
1861 return -EINVAL;
1862
1863 k_getrusage(current, who, &r);
1864 return put_compat_rusage(&r, ru);
1865 }
1866 #endif
1867
1868 SYSCALL_DEFINE1(umask, int, mask)
1869 {
1870 mask = xchg(&current->fs->umask, mask & S_IRWXUGO);
1871 return mask;
1872 }
1873
1874 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1875 {
1876 struct fd exe;
1877 struct inode *inode;
1878 int err;
1879
1880 exe = fdget(fd);
1881 if (!exe.file)
1882 return -EBADF;
1883
1884 inode = file_inode(exe.file);
1885
1886 /*
1887 * Because the original mm->exe_file points to executable file, make
1888 * sure that this one is executable as well, to avoid breaking an
1889 * overall picture.
1890 */
1891 err = -EACCES;
1892 if (!S_ISREG(inode->i_mode) ||
1893 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1894 goto exit;
1895
1896 err = inode_permission(inode, MAY_EXEC);
1897 if (err)
1898 goto exit;
1899
1900 down_write(&mm->mmap_sem);
1901
1902 /*
1903 * Forbid mm->exe_file change if old file still mapped.
1904 */
1905 err = -EBUSY;
1906 if (mm->exe_file) {
1907 struct vm_area_struct *vma;
1908
1909 for (vma = mm->mmap; vma; vma = vma->vm_next)
1910 if (vma->vm_file &&
1911 path_equal(&vma->vm_file->f_path,
1912 &mm->exe_file->f_path))
1913 goto exit_unlock;
1914 }
1915
1916 /*
1917 * The symlink can be changed only once, just to disallow arbitrary
1918 * transitions malicious software might bring in. This means one
1919 * could make a snapshot over all processes running and monitor
1920 * /proc/pid/exe changes to notice unusual activity if needed.
1921 */
1922 err = -EPERM;
1923 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1924 goto exit_unlock;
1925
1926 err = 0;
1927 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1928 exit_unlock:
1929 up_write(&mm->mmap_sem);
1930
1931 exit:
1932 fdput(exe);
1933 return err;
1934 }
1935
1936 static int prctl_set_mm(int opt, unsigned long addr,
1937 unsigned long arg4, unsigned long arg5)
1938 {
1939 unsigned long rlim = rlimit(RLIMIT_DATA);
1940 struct mm_struct *mm = current->mm;
1941 struct vm_area_struct *vma;
1942 int error;
1943
1944 if (arg5 || (arg4 && opt != PR_SET_MM_AUXV))
1945 return -EINVAL;
1946
1947 if (!capable(CAP_SYS_RESOURCE))
1948 return -EPERM;
1949
1950 if (opt == PR_SET_MM_EXE_FILE)
1951 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1952
1953 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1954 return -EINVAL;
1955
1956 error = -EINVAL;
1957
1958 down_read(&mm->mmap_sem);
1959 vma = find_vma(mm, addr);
1960
1961 switch (opt) {
1962 case PR_SET_MM_START_CODE:
1963 mm->start_code = addr;
1964 break;
1965 case PR_SET_MM_END_CODE:
1966 mm->end_code = addr;
1967 break;
1968 case PR_SET_MM_START_DATA:
1969 mm->start_data = addr;
1970 break;
1971 case PR_SET_MM_END_DATA:
1972 mm->end_data = addr;
1973 break;
1974
1975 case PR_SET_MM_START_BRK:
1976 if (addr <= mm->end_data)
1977 goto out;
1978
1979 if (rlim < RLIM_INFINITY &&
1980 (mm->brk - addr) +
1981 (mm->end_data - mm->start_data) > rlim)
1982 goto out;
1983
1984 mm->start_brk = addr;
1985 break;
1986
1987 case PR_SET_MM_BRK:
1988 if (addr <= mm->end_data)
1989 goto out;
1990
1991 if (rlim < RLIM_INFINITY &&
1992 (addr - mm->start_brk) +
1993 (mm->end_data - mm->start_data) > rlim)
1994 goto out;
1995
1996 mm->brk = addr;
1997 break;
1998
1999 /*
2000 * If command line arguments and environment
2001 * are placed somewhere else on stack, we can
2002 * set them up here, ARG_START/END to setup
2003 * command line argumets and ENV_START/END
2004 * for environment.
2005 */
2006 case PR_SET_MM_START_STACK:
2007 case PR_SET_MM_ARG_START:
2008 case PR_SET_MM_ARG_END:
2009 case PR_SET_MM_ENV_START:
2010 case PR_SET_MM_ENV_END:
2011 if (!vma) {
2012 error = -EFAULT;
2013 goto out;
2014 }
2015 if (opt == PR_SET_MM_START_STACK)
2016 mm->start_stack = addr;
2017 else if (opt == PR_SET_MM_ARG_START)
2018 mm->arg_start = addr;
2019 else if (opt == PR_SET_MM_ARG_END)
2020 mm->arg_end = addr;
2021 else if (opt == PR_SET_MM_ENV_START)
2022 mm->env_start = addr;
2023 else if (opt == PR_SET_MM_ENV_END)
2024 mm->env_end = addr;
2025 break;
2026
2027 /*
2028 * This doesn't move auxiliary vector itself
2029 * since it's pinned to mm_struct, but allow
2030 * to fill vector with new values. It's up
2031 * to a caller to provide sane values here
2032 * otherwise user space tools which use this
2033 * vector might be unhappy.
2034 */
2035 case PR_SET_MM_AUXV: {
2036 unsigned long user_auxv[AT_VECTOR_SIZE];
2037
2038 if (arg4 > sizeof(user_auxv))
2039 goto out;
2040 up_read(&mm->mmap_sem);
2041
2042 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
2043 return -EFAULT;
2044
2045 /* Make sure the last entry is always AT_NULL */
2046 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2047 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2048
2049 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2050
2051 task_lock(current);
2052 memcpy(mm->saved_auxv, user_auxv, arg4);
2053 task_unlock(current);
2054
2055 return 0;
2056 }
2057 default:
2058 goto out;
2059 }
2060
2061 error = 0;
2062 out:
2063 up_read(&mm->mmap_sem);
2064 return error;
2065 }
2066
2067 #ifdef CONFIG_CHECKPOINT_RESTORE
2068 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2069 {
2070 return put_user(me->clear_child_tid, tid_addr);
2071 }
2072 #else
2073 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2074 {
2075 return -EINVAL;
2076 }
2077 #endif
2078
2079 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2080 unsigned long, arg4, unsigned long, arg5)
2081 {
2082 struct task_struct *me = current;
2083 unsigned char comm[sizeof(me->comm)];
2084 long error;
2085
2086 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2087 if (error != -ENOSYS)
2088 return error;
2089
2090 error = 0;
2091 switch (option) {
2092 case PR_SET_PDEATHSIG:
2093 if (!valid_signal(arg2)) {
2094 error = -EINVAL;
2095 break;
2096 }
2097 me->pdeath_signal = arg2;
2098 break;
2099 case PR_GET_PDEATHSIG:
2100 error = put_user(me->pdeath_signal, (int __user *)arg2);
2101 break;
2102 case PR_GET_DUMPABLE:
2103 error = get_dumpable(me->mm);
2104 break;
2105 case PR_SET_DUMPABLE:
2106 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2107 error = -EINVAL;
2108 break;
2109 }
2110 set_dumpable(me->mm, arg2);
2111 break;
2112
2113 case PR_SET_UNALIGN:
2114 error = SET_UNALIGN_CTL(me, arg2);
2115 break;
2116 case PR_GET_UNALIGN:
2117 error = GET_UNALIGN_CTL(me, arg2);
2118 break;
2119 case PR_SET_FPEMU:
2120 error = SET_FPEMU_CTL(me, arg2);
2121 break;
2122 case PR_GET_FPEMU:
2123 error = GET_FPEMU_CTL(me, arg2);
2124 break;
2125 case PR_SET_FPEXC:
2126 error = SET_FPEXC_CTL(me, arg2);
2127 break;
2128 case PR_GET_FPEXC:
2129 error = GET_FPEXC_CTL(me, arg2);
2130 break;
2131 case PR_GET_TIMING:
2132 error = PR_TIMING_STATISTICAL;
2133 break;
2134 case PR_SET_TIMING:
2135 if (arg2 != PR_TIMING_STATISTICAL)
2136 error = -EINVAL;
2137 break;
2138 case PR_SET_NAME:
2139 comm[sizeof(me->comm) - 1] = 0;
2140 if (strncpy_from_user(comm, (char __user *)arg2,
2141 sizeof(me->comm) - 1) < 0)
2142 return -EFAULT;
2143 set_task_comm(me, comm);
2144 proc_comm_connector(me);
2145 break;
2146 case PR_GET_NAME:
2147 get_task_comm(comm, me);
2148 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2149 return -EFAULT;
2150 break;
2151 case PR_GET_ENDIAN:
2152 error = GET_ENDIAN(me, arg2);
2153 break;
2154 case PR_SET_ENDIAN:
2155 error = SET_ENDIAN(me, arg2);
2156 break;
2157 case PR_GET_SECCOMP:
2158 error = prctl_get_seccomp();
2159 break;
2160 case PR_SET_SECCOMP:
2161 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2162 break;
2163 case PR_GET_TSC:
2164 error = GET_TSC_CTL(arg2);
2165 break;
2166 case PR_SET_TSC:
2167 error = SET_TSC_CTL(arg2);
2168 break;
2169 case PR_TASK_PERF_EVENTS_DISABLE:
2170 error = perf_event_task_disable();
2171 break;
2172 case PR_TASK_PERF_EVENTS_ENABLE:
2173 error = perf_event_task_enable();
2174 break;
2175 case PR_GET_TIMERSLACK:
2176 error = current->timer_slack_ns;
2177 break;
2178 case PR_SET_TIMERSLACK:
2179 if (arg2 <= 0)
2180 current->timer_slack_ns =
2181 current->default_timer_slack_ns;
2182 else
2183 current->timer_slack_ns = arg2;
2184 break;
2185 case PR_MCE_KILL:
2186 if (arg4 | arg5)
2187 return -EINVAL;
2188 switch (arg2) {
2189 case PR_MCE_KILL_CLEAR:
2190 if (arg3 != 0)
2191 return -EINVAL;
2192 current->flags &= ~PF_MCE_PROCESS;
2193 break;
2194 case PR_MCE_KILL_SET:
2195 current->flags |= PF_MCE_PROCESS;
2196 if (arg3 == PR_MCE_KILL_EARLY)
2197 current->flags |= PF_MCE_EARLY;
2198 else if (arg3 == PR_MCE_KILL_LATE)
2199 current->flags &= ~PF_MCE_EARLY;
2200 else if (arg3 == PR_MCE_KILL_DEFAULT)
2201 current->flags &=
2202 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2203 else
2204 return -EINVAL;
2205 break;
2206 default:
2207 return -EINVAL;
2208 }
2209 break;
2210 case PR_MCE_KILL_GET:
2211 if (arg2 | arg3 | arg4 | arg5)
2212 return -EINVAL;
2213 if (current->flags & PF_MCE_PROCESS)
2214 error = (current->flags & PF_MCE_EARLY) ?
2215 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2216 else
2217 error = PR_MCE_KILL_DEFAULT;
2218 break;
2219 case PR_SET_MM:
2220 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2221 break;
2222 case PR_GET_TID_ADDRESS:
2223 error = prctl_get_tid_address(me, (int __user **)arg2);
2224 break;
2225 case PR_SET_CHILD_SUBREAPER:
2226 me->signal->is_child_subreaper = !!arg2;
2227 break;
2228 case PR_GET_CHILD_SUBREAPER:
2229 error = put_user(me->signal->is_child_subreaper,
2230 (int __user *)arg2);
2231 break;
2232 case PR_SET_NO_NEW_PRIVS:
2233 if (arg2 != 1 || arg3 || arg4 || arg5)
2234 return -EINVAL;
2235
2236 current->no_new_privs = 1;
2237 break;
2238 case PR_GET_NO_NEW_PRIVS:
2239 if (arg2 || arg3 || arg4 || arg5)
2240 return -EINVAL;
2241 return current->no_new_privs ? 1 : 0;
2242 default:
2243 error = -EINVAL;
2244 break;
2245 }
2246 return error;
2247 }
2248
2249 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2250 struct getcpu_cache __user *, unused)
2251 {
2252 int err = 0;
2253 int cpu = raw_smp_processor_id();
2254 if (cpup)
2255 err |= put_user(cpu, cpup);
2256 if (nodep)
2257 err |= put_user(cpu_to_node(cpu), nodep);
2258 return err ? -EFAULT : 0;
2259 }
2260
2261 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
2262
2263 static int __orderly_poweroff(bool force)
2264 {
2265 char **argv;
2266 static char *envp[] = {
2267 "HOME=/",
2268 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
2269 NULL
2270 };
2271 int ret;
2272
2273 argv = argv_split(GFP_KERNEL, poweroff_cmd, NULL);
2274 if (argv) {
2275 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
2276 argv_free(argv);
2277 } else {
2278 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
2279 __func__, poweroff_cmd);
2280 ret = -ENOMEM;
2281 }
2282
2283 if (ret && force) {
2284 printk(KERN_WARNING "Failed to start orderly shutdown: "
2285 "forcing the issue\n");
2286 /*
2287 * I guess this should try to kick off some daemon to sync and
2288 * poweroff asap. Or not even bother syncing if we're doing an
2289 * emergency shutdown?
2290 */
2291 emergency_sync();
2292 kernel_power_off();
2293 }
2294
2295 return ret;
2296 }
2297
2298 static bool poweroff_force;
2299
2300 static void poweroff_work_func(struct work_struct *work)
2301 {
2302 __orderly_poweroff(poweroff_force);
2303 }
2304
2305 static DECLARE_WORK(poweroff_work, poweroff_work_func);
2306
2307 /**
2308 * orderly_poweroff - Trigger an orderly system poweroff
2309 * @force: force poweroff if command execution fails
2310 *
2311 * This may be called from any context to trigger a system shutdown.
2312 * If the orderly shutdown fails, it will force an immediate shutdown.
2313 */
2314 int orderly_poweroff(bool force)
2315 {
2316 if (force) /* do not override the pending "true" */
2317 poweroff_force = true;
2318 schedule_work(&poweroff_work);
2319 return 0;
2320 }
2321 EXPORT_SYMBOL_GPL(orderly_poweroff);
2322
2323 /**
2324 * do_sysinfo - fill in sysinfo struct
2325 * @info: pointer to buffer to fill
2326 */
2327 static int do_sysinfo(struct sysinfo *info)
2328 {
2329 unsigned long mem_total, sav_total;
2330 unsigned int mem_unit, bitcount;
2331 struct timespec tp;
2332
2333 memset(info, 0, sizeof(struct sysinfo));
2334
2335 ktime_get_ts(&tp);
2336 monotonic_to_bootbased(&tp);
2337 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2338
2339 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2340
2341 info->procs = nr_threads;
2342
2343 si_meminfo(info);
2344 si_swapinfo(info);
2345
2346 /*
2347 * If the sum of all the available memory (i.e. ram + swap)
2348 * is less than can be stored in a 32 bit unsigned long then
2349 * we can be binary compatible with 2.2.x kernels. If not,
2350 * well, in that case 2.2.x was broken anyways...
2351 *
2352 * -Erik Andersen <andersee@debian.org>
2353 */
2354
2355 mem_total = info->totalram + info->totalswap;
2356 if (mem_total < info->totalram || mem_total < info->totalswap)
2357 goto out;
2358 bitcount = 0;
2359 mem_unit = info->mem_unit;
2360 while (mem_unit > 1) {
2361 bitcount++;
2362 mem_unit >>= 1;
2363 sav_total = mem_total;
2364 mem_total <<= 1;
2365 if (mem_total < sav_total)
2366 goto out;
2367 }
2368
2369 /*
2370 * If mem_total did not overflow, multiply all memory values by
2371 * info->mem_unit and set it to 1. This leaves things compatible
2372 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2373 * kernels...
2374 */
2375
2376 info->mem_unit = 1;
2377 info->totalram <<= bitcount;
2378 info->freeram <<= bitcount;
2379 info->sharedram <<= bitcount;
2380 info->bufferram <<= bitcount;
2381 info->totalswap <<= bitcount;
2382 info->freeswap <<= bitcount;
2383 info->totalhigh <<= bitcount;
2384 info->freehigh <<= bitcount;
2385
2386 out:
2387 return 0;
2388 }
2389
2390 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2391 {
2392 struct sysinfo val;
2393
2394 do_sysinfo(&val);
2395
2396 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2397 return -EFAULT;
2398
2399 return 0;
2400 }
2401
2402 #ifdef CONFIG_COMPAT
2403 struct compat_sysinfo {
2404 s32 uptime;
2405 u32 loads[3];
2406 u32 totalram;
2407 u32 freeram;
2408 u32 sharedram;
2409 u32 bufferram;
2410 u32 totalswap;
2411 u32 freeswap;
2412 u16 procs;
2413 u16 pad;
2414 u32 totalhigh;
2415 u32 freehigh;
2416 u32 mem_unit;
2417 char _f[20-2*sizeof(u32)-sizeof(int)];
2418 };
2419
2420 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2421 {
2422 struct sysinfo s;
2423
2424 do_sysinfo(&s);
2425
2426 /* Check to see if any memory value is too large for 32-bit and scale
2427 * down if needed
2428 */
2429 if ((s.totalram >> 32) || (s.totalswap >> 32)) {
2430 int bitcount = 0;
2431
2432 while (s.mem_unit < PAGE_SIZE) {
2433 s.mem_unit <<= 1;
2434 bitcount++;
2435 }
2436
2437 s.totalram >>= bitcount;
2438 s.freeram >>= bitcount;
2439 s.sharedram >>= bitcount;
2440 s.bufferram >>= bitcount;
2441 s.totalswap >>= bitcount;
2442 s.freeswap >>= bitcount;
2443 s.totalhigh >>= bitcount;
2444 s.freehigh >>= bitcount;
2445 }
2446
2447 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2448 __put_user(s.uptime, &info->uptime) ||
2449 __put_user(s.loads[0], &info->loads[0]) ||
2450 __put_user(s.loads[1], &info->loads[1]) ||
2451 __put_user(s.loads[2], &info->loads[2]) ||
2452 __put_user(s.totalram, &info->totalram) ||
2453 __put_user(s.freeram, &info->freeram) ||
2454 __put_user(s.sharedram, &info->sharedram) ||
2455 __put_user(s.bufferram, &info->bufferram) ||
2456 __put_user(s.totalswap, &info->totalswap) ||
2457 __put_user(s.freeswap, &info->freeswap) ||
2458 __put_user(s.procs, &info->procs) ||
2459 __put_user(s.totalhigh, &info->totalhigh) ||
2460 __put_user(s.freehigh, &info->freehigh) ||
2461 __put_user(s.mem_unit, &info->mem_unit))
2462 return -EFAULT;
2463
2464 return 0;
2465 }
2466 #endif /* CONFIG_COMPAT */
Linus' kernel tree