4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/module.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>
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>
37 #include <linux/compat.h>
38 #include <linux/syscalls.h>
39 #include <linux/kprobes.h>
40 #include <linux/user_namespace.h>
42 #include <asm/uaccess.h>
44 #include <asm/unistd.h>
46 #ifndef SET_UNALIGN_CTL
47 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
49 #ifndef GET_UNALIGN_CTL
50 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
53 # define SET_FPEMU_CTL(a,b) (-EINVAL)
56 # define GET_FPEMU_CTL(a,b) (-EINVAL)
59 # define SET_FPEXC_CTL(a,b) (-EINVAL)
62 # define GET_FPEXC_CTL(a,b) (-EINVAL)
65 # define GET_ENDIAN(a,b) (-EINVAL)
68 # define SET_ENDIAN(a,b) (-EINVAL)
71 # define GET_TSC_CTL(a) (-EINVAL)
74 # define SET_TSC_CTL(a) (-EINVAL)
78 * this is where the system-wide overflow UID and GID are defined, for
79 * architectures that now have 32-bit UID/GID but didn't in the past
82 int overflowuid
= DEFAULT_OVERFLOWUID
;
83 int overflowgid
= DEFAULT_OVERFLOWGID
;
86 EXPORT_SYMBOL(overflowuid
);
87 EXPORT_SYMBOL(overflowgid
);
91 * the same as above, but for filesystems which can only store a 16-bit
92 * UID and GID. as such, this is needed on all architectures
95 int fs_overflowuid
= DEFAULT_FS_OVERFLOWUID
;
96 int fs_overflowgid
= DEFAULT_FS_OVERFLOWUID
;
98 EXPORT_SYMBOL(fs_overflowuid
);
99 EXPORT_SYMBOL(fs_overflowgid
);
102 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
107 EXPORT_SYMBOL(cad_pid
);
110 * If set, this is used for preparing the system to power off.
113 void (*pm_power_off_prepare
)(void);
115 static int set_one_prio(struct task_struct
*p
, int niceval
, int error
)
119 if (p
->uid
!= current
->euid
&&
120 p
->euid
!= current
->euid
&& !capable(CAP_SYS_NICE
)) {
124 if (niceval
< task_nice(p
) && !can_nice(p
, niceval
)) {
128 no_nice
= security_task_setnice(p
, niceval
);
135 set_user_nice(p
, niceval
);
140 SYSCALL_DEFINE3(setpriority
, int, which
, int, who
, int, niceval
)
142 struct task_struct
*g
, *p
;
143 struct user_struct
*user
;
147 if (which
> PRIO_USER
|| which
< PRIO_PROCESS
)
150 /* normalize: avoid signed division (rounding problems) */
157 read_lock(&tasklist_lock
);
161 p
= find_task_by_vpid(who
);
165 error
= set_one_prio(p
, niceval
, error
);
169 pgrp
= find_vpid(who
);
171 pgrp
= task_pgrp(current
);
172 do_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
) {
173 error
= set_one_prio(p
, niceval
, error
);
174 } while_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
);
177 user
= current
->user
;
181 if ((who
!= current
->uid
) && !(user
= find_user(who
)))
182 goto out_unlock
; /* No processes for this user */
186 error
= set_one_prio(p
, niceval
, error
);
187 while_each_thread(g
, p
);
188 if (who
!= current
->uid
)
189 free_uid(user
); /* For find_user() */
193 read_unlock(&tasklist_lock
);
199 * Ugh. To avoid negative return values, "getpriority()" will
200 * not return the normal nice-value, but a negated value that
201 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
202 * to stay compatible.
204 SYSCALL_DEFINE2(getpriority
, int, which
, int, who
)
206 struct task_struct
*g
, *p
;
207 struct user_struct
*user
;
208 long niceval
, retval
= -ESRCH
;
211 if (which
> PRIO_USER
|| which
< PRIO_PROCESS
)
214 read_lock(&tasklist_lock
);
218 p
= find_task_by_vpid(who
);
222 niceval
= 20 - task_nice(p
);
223 if (niceval
> retval
)
229 pgrp
= find_vpid(who
);
231 pgrp
= task_pgrp(current
);
232 do_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
) {
233 niceval
= 20 - task_nice(p
);
234 if (niceval
> retval
)
236 } while_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
);
239 user
= current
->user
;
243 if ((who
!= current
->uid
) && !(user
= find_user(who
)))
244 goto out_unlock
; /* No processes for this user */
248 niceval
= 20 - task_nice(p
);
249 if (niceval
> retval
)
252 while_each_thread(g
, p
);
253 if (who
!= current
->uid
)
254 free_uid(user
); /* for find_user() */
258 read_unlock(&tasklist_lock
);
264 * emergency_restart - reboot the system
266 * Without shutting down any hardware or taking any locks
267 * reboot the system. This is called when we know we are in
268 * trouble so this is our best effort to reboot. This is
269 * safe to call in interrupt context.
271 void emergency_restart(void)
273 machine_emergency_restart();
275 EXPORT_SYMBOL_GPL(emergency_restart
);
277 void kernel_restart_prepare(char *cmd
)
279 blocking_notifier_call_chain(&reboot_notifier_list
, SYS_RESTART
, cmd
);
280 system_state
= SYSTEM_RESTART
;
286 * kernel_restart - reboot the system
287 * @cmd: pointer to buffer containing command to execute for restart
290 * Shutdown everything and perform a clean reboot.
291 * This is not safe to call in interrupt context.
293 void kernel_restart(char *cmd
)
295 kernel_restart_prepare(cmd
);
297 printk(KERN_EMERG
"Restarting system.\n");
299 printk(KERN_EMERG
"Restarting system with command '%s'.\n", cmd
);
300 machine_restart(cmd
);
302 EXPORT_SYMBOL_GPL(kernel_restart
);
304 static void kernel_shutdown_prepare(enum system_states state
)
306 blocking_notifier_call_chain(&reboot_notifier_list
,
307 (state
== SYSTEM_HALT
)?SYS_HALT
:SYS_POWER_OFF
, NULL
);
308 system_state
= state
;
312 * kernel_halt - halt the system
314 * Shutdown everything and perform a clean system halt.
316 void kernel_halt(void)
318 kernel_shutdown_prepare(SYSTEM_HALT
);
320 printk(KERN_EMERG
"System halted.\n");
324 EXPORT_SYMBOL_GPL(kernel_halt
);
327 * kernel_power_off - power_off the system
329 * Shutdown everything and perform a clean system power_off.
331 void kernel_power_off(void)
333 kernel_shutdown_prepare(SYSTEM_POWER_OFF
);
334 if (pm_power_off_prepare
)
335 pm_power_off_prepare();
336 disable_nonboot_cpus();
338 printk(KERN_EMERG
"Power down.\n");
341 EXPORT_SYMBOL_GPL(kernel_power_off
);
343 * Reboot system call: for obvious reasons only root may call it,
344 * and even root needs to set up some magic numbers in the registers
345 * so that some mistake won't make this reboot the whole machine.
346 * You can also set the meaning of the ctrl-alt-del-key here.
348 * reboot doesn't sync: do that yourself before calling this.
350 SYSCALL_DEFINE4(reboot
, int, magic1
, int, magic2
, unsigned int, cmd
,
355 /* We only trust the superuser with rebooting the system. */
356 if (!capable(CAP_SYS_BOOT
))
359 /* For safety, we require "magic" arguments. */
360 if (magic1
!= LINUX_REBOOT_MAGIC1
||
361 (magic2
!= LINUX_REBOOT_MAGIC2
&&
362 magic2
!= LINUX_REBOOT_MAGIC2A
&&
363 magic2
!= LINUX_REBOOT_MAGIC2B
&&
364 magic2
!= LINUX_REBOOT_MAGIC2C
))
367 /* Instead of trying to make the power_off code look like
368 * halt when pm_power_off is not set do it the easy way.
370 if ((cmd
== LINUX_REBOOT_CMD_POWER_OFF
) && !pm_power_off
)
371 cmd
= LINUX_REBOOT_CMD_HALT
;
375 case LINUX_REBOOT_CMD_RESTART
:
376 kernel_restart(NULL
);
379 case LINUX_REBOOT_CMD_CAD_ON
:
383 case LINUX_REBOOT_CMD_CAD_OFF
:
387 case LINUX_REBOOT_CMD_HALT
:
393 case LINUX_REBOOT_CMD_POWER_OFF
:
399 case LINUX_REBOOT_CMD_RESTART2
:
400 if (strncpy_from_user(&buffer
[0], arg
, sizeof(buffer
) - 1) < 0) {
404 buffer
[sizeof(buffer
) - 1] = '\0';
406 kernel_restart(buffer
);
410 case LINUX_REBOOT_CMD_KEXEC
:
413 ret
= kernel_kexec();
419 #ifdef CONFIG_HIBERNATION
420 case LINUX_REBOOT_CMD_SW_SUSPEND
:
422 int ret
= hibernate();
436 static void deferred_cad(struct work_struct
*dummy
)
438 kernel_restart(NULL
);
442 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
443 * As it's called within an interrupt, it may NOT sync: the only choice
444 * is whether to reboot at once, or just ignore the ctrl-alt-del.
446 void ctrl_alt_del(void)
448 static DECLARE_WORK(cad_work
, deferred_cad
);
451 schedule_work(&cad_work
);
453 kill_cad_pid(SIGINT
, 1);
457 * Unprivileged users may change the real gid to the effective gid
458 * or vice versa. (BSD-style)
460 * If you set the real gid at all, or set the effective gid to a value not
461 * equal to the real gid, then the saved gid is set to the new effective gid.
463 * This makes it possible for a setgid program to completely drop its
464 * privileges, which is often a useful assertion to make when you are doing
465 * a security audit over a program.
467 * The general idea is that a program which uses just setregid() will be
468 * 100% compatible with BSD. A program which uses just setgid() will be
469 * 100% compatible with POSIX with saved IDs.
471 * SMP: There are not races, the GIDs are checked only by filesystem
472 * operations (as far as semantic preservation is concerned).
474 SYSCALL_DEFINE2(setregid
, gid_t
, rgid
, gid_t
, egid
)
476 int old_rgid
= current
->gid
;
477 int old_egid
= current
->egid
;
478 int new_rgid
= old_rgid
;
479 int new_egid
= old_egid
;
482 retval
= security_task_setgid(rgid
, egid
, (gid_t
)-1, LSM_SETID_RE
);
486 if (rgid
!= (gid_t
) -1) {
487 if ((old_rgid
== rgid
) ||
488 (current
->egid
==rgid
) ||
494 if (egid
!= (gid_t
) -1) {
495 if ((old_rgid
== egid
) ||
496 (current
->egid
== egid
) ||
497 (current
->sgid
== egid
) ||
503 if (new_egid
!= old_egid
) {
504 set_dumpable(current
->mm
, suid_dumpable
);
507 if (rgid
!= (gid_t
) -1 ||
508 (egid
!= (gid_t
) -1 && egid
!= old_rgid
))
509 current
->sgid
= new_egid
;
510 current
->fsgid
= new_egid
;
511 current
->egid
= new_egid
;
512 current
->gid
= new_rgid
;
513 key_fsgid_changed(current
);
514 proc_id_connector(current
, PROC_EVENT_GID
);
519 * setgid() is implemented like SysV w/ SAVED_IDS
521 * SMP: Same implicit races as above.
523 SYSCALL_DEFINE1(setgid
, gid_t
, gid
)
525 int old_egid
= current
->egid
;
528 retval
= security_task_setgid(gid
, (gid_t
)-1, (gid_t
)-1, LSM_SETID_ID
);
532 if (capable(CAP_SETGID
)) {
533 if (old_egid
!= gid
) {
534 set_dumpable(current
->mm
, suid_dumpable
);
537 current
->gid
= current
->egid
= current
->sgid
= current
->fsgid
= gid
;
538 } else if ((gid
== current
->gid
) || (gid
== current
->sgid
)) {
539 if (old_egid
!= gid
) {
540 set_dumpable(current
->mm
, suid_dumpable
);
543 current
->egid
= current
->fsgid
= gid
;
548 key_fsgid_changed(current
);
549 proc_id_connector(current
, PROC_EVENT_GID
);
553 static int set_user(uid_t new_ruid
, int dumpclear
)
555 struct user_struct
*new_user
;
557 new_user
= alloc_uid(current
->nsproxy
->user_ns
, new_ruid
);
561 if (atomic_read(&new_user
->processes
) >=
562 current
->signal
->rlim
[RLIMIT_NPROC
].rlim_cur
&&
563 new_user
!= current
->nsproxy
->user_ns
->root_user
) {
568 switch_uid(new_user
);
571 set_dumpable(current
->mm
, suid_dumpable
);
574 current
->uid
= new_ruid
;
579 * Unprivileged users may change the real uid to the effective uid
580 * or vice versa. (BSD-style)
582 * If you set the real uid at all, or set the effective uid to a value not
583 * equal to the real uid, then the saved uid is set to the new effective uid.
585 * This makes it possible for a setuid program to completely drop its
586 * privileges, which is often a useful assertion to make when you are doing
587 * a security audit over a program.
589 * The general idea is that a program which uses just setreuid() will be
590 * 100% compatible with BSD. A program which uses just setuid() will be
591 * 100% compatible with POSIX with saved IDs.
593 SYSCALL_DEFINE2(setreuid
, uid_t
, ruid
, uid_t
, euid
)
595 int old_ruid
, old_euid
, old_suid
, new_ruid
, new_euid
;
598 retval
= security_task_setuid(ruid
, euid
, (uid_t
)-1, LSM_SETID_RE
);
602 new_ruid
= old_ruid
= current
->uid
;
603 new_euid
= old_euid
= current
->euid
;
604 old_suid
= current
->suid
;
606 if (ruid
!= (uid_t
) -1) {
608 if ((old_ruid
!= ruid
) &&
609 (current
->euid
!= ruid
) &&
610 !capable(CAP_SETUID
))
614 if (euid
!= (uid_t
) -1) {
616 if ((old_ruid
!= euid
) &&
617 (current
->euid
!= euid
) &&
618 (current
->suid
!= euid
) &&
619 !capable(CAP_SETUID
))
623 if (new_ruid
!= old_ruid
&& set_user(new_ruid
, new_euid
!= old_euid
) < 0)
626 if (new_euid
!= old_euid
) {
627 set_dumpable(current
->mm
, suid_dumpable
);
630 current
->fsuid
= current
->euid
= new_euid
;
631 if (ruid
!= (uid_t
) -1 ||
632 (euid
!= (uid_t
) -1 && euid
!= old_ruid
))
633 current
->suid
= current
->euid
;
634 current
->fsuid
= current
->euid
;
636 key_fsuid_changed(current
);
637 proc_id_connector(current
, PROC_EVENT_UID
);
639 return security_task_post_setuid(old_ruid
, old_euid
, old_suid
, LSM_SETID_RE
);
645 * setuid() is implemented like SysV with SAVED_IDS
647 * Note that SAVED_ID's is deficient in that a setuid root program
648 * like sendmail, for example, cannot set its uid to be a normal
649 * user and then switch back, because if you're root, setuid() sets
650 * the saved uid too. If you don't like this, blame the bright people
651 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
652 * will allow a root program to temporarily drop privileges and be able to
653 * regain them by swapping the real and effective uid.
655 SYSCALL_DEFINE1(setuid
, uid_t
, uid
)
657 int old_euid
= current
->euid
;
658 int old_ruid
, old_suid
, new_suid
;
661 retval
= security_task_setuid(uid
, (uid_t
)-1, (uid_t
)-1, LSM_SETID_ID
);
665 old_ruid
= current
->uid
;
666 old_suid
= current
->suid
;
669 if (capable(CAP_SETUID
)) {
670 if (uid
!= old_ruid
&& set_user(uid
, old_euid
!= uid
) < 0)
673 } else if ((uid
!= current
->uid
) && (uid
!= new_suid
))
676 if (old_euid
!= uid
) {
677 set_dumpable(current
->mm
, suid_dumpable
);
680 current
->fsuid
= current
->euid
= uid
;
681 current
->suid
= new_suid
;
683 key_fsuid_changed(current
);
684 proc_id_connector(current
, PROC_EVENT_UID
);
686 return security_task_post_setuid(old_ruid
, old_euid
, old_suid
, LSM_SETID_ID
);
691 * This function implements a generic ability to update ruid, euid,
692 * and suid. This allows you to implement the 4.4 compatible seteuid().
694 SYSCALL_DEFINE3(setresuid
, uid_t
, ruid
, uid_t
, euid
, uid_t
, suid
)
696 int old_ruid
= current
->uid
;
697 int old_euid
= current
->euid
;
698 int old_suid
= current
->suid
;
701 retval
= security_task_setuid(ruid
, euid
, suid
, LSM_SETID_RES
);
705 if (!capable(CAP_SETUID
)) {
706 if ((ruid
!= (uid_t
) -1) && (ruid
!= current
->uid
) &&
707 (ruid
!= current
->euid
) && (ruid
!= current
->suid
))
709 if ((euid
!= (uid_t
) -1) && (euid
!= current
->uid
) &&
710 (euid
!= current
->euid
) && (euid
!= current
->suid
))
712 if ((suid
!= (uid_t
) -1) && (suid
!= current
->uid
) &&
713 (suid
!= current
->euid
) && (suid
!= current
->suid
))
716 if (ruid
!= (uid_t
) -1) {
717 if (ruid
!= current
->uid
&& set_user(ruid
, euid
!= current
->euid
) < 0)
720 if (euid
!= (uid_t
) -1) {
721 if (euid
!= current
->euid
) {
722 set_dumpable(current
->mm
, suid_dumpable
);
725 current
->euid
= euid
;
727 current
->fsuid
= current
->euid
;
728 if (suid
!= (uid_t
) -1)
729 current
->suid
= suid
;
731 key_fsuid_changed(current
);
732 proc_id_connector(current
, PROC_EVENT_UID
);
734 return security_task_post_setuid(old_ruid
, old_euid
, old_suid
, LSM_SETID_RES
);
737 SYSCALL_DEFINE3(getresuid
, uid_t __user
*, ruid
, uid_t __user
*, euid
, uid_t __user
*, suid
)
741 if (!(retval
= put_user(current
->uid
, ruid
)) &&
742 !(retval
= put_user(current
->euid
, euid
)))
743 retval
= put_user(current
->suid
, suid
);
749 * Same as above, but for rgid, egid, sgid.
751 SYSCALL_DEFINE3(setresgid
, gid_t
, rgid
, gid_t
, egid
, gid_t
, sgid
)
755 retval
= security_task_setgid(rgid
, egid
, sgid
, LSM_SETID_RES
);
759 if (!capable(CAP_SETGID
)) {
760 if ((rgid
!= (gid_t
) -1) && (rgid
!= current
->gid
) &&
761 (rgid
!= current
->egid
) && (rgid
!= current
->sgid
))
763 if ((egid
!= (gid_t
) -1) && (egid
!= current
->gid
) &&
764 (egid
!= current
->egid
) && (egid
!= current
->sgid
))
766 if ((sgid
!= (gid_t
) -1) && (sgid
!= current
->gid
) &&
767 (sgid
!= current
->egid
) && (sgid
!= current
->sgid
))
770 if (egid
!= (gid_t
) -1) {
771 if (egid
!= current
->egid
) {
772 set_dumpable(current
->mm
, suid_dumpable
);
775 current
->egid
= egid
;
777 current
->fsgid
= current
->egid
;
778 if (rgid
!= (gid_t
) -1)
780 if (sgid
!= (gid_t
) -1)
781 current
->sgid
= sgid
;
783 key_fsgid_changed(current
);
784 proc_id_connector(current
, PROC_EVENT_GID
);
788 SYSCALL_DEFINE3(getresgid
, gid_t __user
*, rgid
, gid_t __user
*, egid
, gid_t __user
*, sgid
)
792 if (!(retval
= put_user(current
->gid
, rgid
)) &&
793 !(retval
= put_user(current
->egid
, egid
)))
794 retval
= put_user(current
->sgid
, sgid
);
801 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
802 * is used for "access()" and for the NFS daemon (letting nfsd stay at
803 * whatever uid it wants to). It normally shadows "euid", except when
804 * explicitly set by setfsuid() or for access..
806 SYSCALL_DEFINE1(setfsuid
, uid_t
, uid
)
810 old_fsuid
= current
->fsuid
;
811 if (security_task_setuid(uid
, (uid_t
)-1, (uid_t
)-1, LSM_SETID_FS
))
814 if (uid
== current
->uid
|| uid
== current
->euid
||
815 uid
== current
->suid
|| uid
== current
->fsuid
||
816 capable(CAP_SETUID
)) {
817 if (uid
!= old_fsuid
) {
818 set_dumpable(current
->mm
, suid_dumpable
);
821 current
->fsuid
= uid
;
824 key_fsuid_changed(current
);
825 proc_id_connector(current
, PROC_EVENT_UID
);
827 security_task_post_setuid(old_fsuid
, (uid_t
)-1, (uid_t
)-1, LSM_SETID_FS
);
833 * Samma på svenska..
835 SYSCALL_DEFINE1(setfsgid
, gid_t
, gid
)
839 old_fsgid
= current
->fsgid
;
840 if (security_task_setgid(gid
, (gid_t
)-1, (gid_t
)-1, LSM_SETID_FS
))
843 if (gid
== current
->gid
|| gid
== current
->egid
||
844 gid
== current
->sgid
|| gid
== current
->fsgid
||
845 capable(CAP_SETGID
)) {
846 if (gid
!= old_fsgid
) {
847 set_dumpable(current
->mm
, suid_dumpable
);
850 current
->fsgid
= gid
;
851 key_fsgid_changed(current
);
852 proc_id_connector(current
, PROC_EVENT_GID
);
857 SYSCALL_DEFINE1(times
, struct tms __user
*, tbuf
)
860 * In the SMP world we might just be unlucky and have one of
861 * the times increment as we use it. Since the value is an
862 * atomically safe type this is just fine. Conceptually its
863 * as if the syscall took an instant longer to occur.
867 struct task_struct
*tsk
= current
;
868 struct task_struct
*t
;
869 cputime_t utime
, stime
, cutime
, cstime
;
871 spin_lock_irq(&tsk
->sighand
->siglock
);
872 utime
= tsk
->signal
->utime
;
873 stime
= tsk
->signal
->stime
;
876 utime
= cputime_add(utime
, t
->utime
);
877 stime
= cputime_add(stime
, t
->stime
);
881 cutime
= tsk
->signal
->cutime
;
882 cstime
= tsk
->signal
->cstime
;
883 spin_unlock_irq(&tsk
->sighand
->siglock
);
885 tmp
.tms_utime
= cputime_to_clock_t(utime
);
886 tmp
.tms_stime
= cputime_to_clock_t(stime
);
887 tmp
.tms_cutime
= cputime_to_clock_t(cutime
);
888 tmp
.tms_cstime
= cputime_to_clock_t(cstime
);
889 if (copy_to_user(tbuf
, &tmp
, sizeof(struct tms
)))
892 return (long) jiffies_64_to_clock_t(get_jiffies_64());
896 * This needs some heavy checking ...
897 * I just haven't the stomach for it. I also don't fully
898 * understand sessions/pgrp etc. Let somebody who does explain it.
900 * OK, I think I have the protection semantics right.... this is really
901 * only important on a multi-user system anyway, to make sure one user
902 * can't send a signal to a process owned by another. -TYT, 12/12/91
904 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
907 SYSCALL_DEFINE2(setpgid
, pid_t
, pid
, pid_t
, pgid
)
909 struct task_struct
*p
;
910 struct task_struct
*group_leader
= current
->group_leader
;
915 pid
= task_pid_vnr(group_leader
);
921 /* From this point forward we keep holding onto the tasklist lock
922 * so that our parent does not change from under us. -DaveM
924 write_lock_irq(&tasklist_lock
);
927 p
= find_task_by_vpid(pid
);
932 if (!thread_group_leader(p
))
935 if (same_thread_group(p
->real_parent
, group_leader
)) {
937 if (task_session(p
) != task_session(group_leader
))
944 if (p
!= group_leader
)
949 if (p
->signal
->leader
)
954 struct task_struct
*g
;
956 pgrp
= find_vpid(pgid
);
957 g
= pid_task(pgrp
, PIDTYPE_PGID
);
958 if (!g
|| task_session(g
) != task_session(group_leader
))
962 err
= security_task_setpgid(p
, pgid
);
966 if (task_pgrp(p
) != pgrp
) {
967 change_pid(p
, PIDTYPE_PGID
, pgrp
);
968 set_task_pgrp(p
, pid_nr(pgrp
));
973 /* All paths lead to here, thus we are safe. -DaveM */
974 write_unlock_irq(&tasklist_lock
);
978 SYSCALL_DEFINE1(getpgid
, pid_t
, pid
)
980 struct task_struct
*p
;
986 grp
= task_pgrp(current
);
989 p
= find_task_by_vpid(pid
);
996 retval
= security_task_getpgid(p
);
1000 retval
= pid_vnr(grp
);
1006 #ifdef __ARCH_WANT_SYS_GETPGRP
1008 SYSCALL_DEFINE0(getpgrp
)
1010 return sys_getpgid(0);
1015 SYSCALL_DEFINE1(getsid
, pid_t
, pid
)
1017 struct task_struct
*p
;
1023 sid
= task_session(current
);
1026 p
= find_task_by_vpid(pid
);
1029 sid
= task_session(p
);
1033 retval
= security_task_getsid(p
);
1037 retval
= pid_vnr(sid
);
1043 SYSCALL_DEFINE0(setsid
)
1045 struct task_struct
*group_leader
= current
->group_leader
;
1046 struct pid
*sid
= task_pid(group_leader
);
1047 pid_t session
= pid_vnr(sid
);
1050 write_lock_irq(&tasklist_lock
);
1051 /* Fail if I am already a session leader */
1052 if (group_leader
->signal
->leader
)
1055 /* Fail if a process group id already exists that equals the
1056 * proposed session id.
1058 if (pid_task(sid
, PIDTYPE_PGID
))
1061 group_leader
->signal
->leader
= 1;
1062 __set_special_pids(sid
);
1064 spin_lock(&group_leader
->sighand
->siglock
);
1065 group_leader
->signal
->tty
= NULL
;
1066 spin_unlock(&group_leader
->sighand
->siglock
);
1070 write_unlock_irq(&tasklist_lock
);
1075 * Supplementary group IDs
1078 /* init to 2 - one for init_task, one to ensure it is never freed */
1079 struct group_info init_groups
= { .usage
= ATOMIC_INIT(2) };
1081 struct group_info
*groups_alloc(int gidsetsize
)
1083 struct group_info
*group_info
;
1087 nblocks
= (gidsetsize
+ NGROUPS_PER_BLOCK
- 1) / NGROUPS_PER_BLOCK
;
1088 /* Make sure we always allocate at least one indirect block pointer */
1089 nblocks
= nblocks
? : 1;
1090 group_info
= kmalloc(sizeof(*group_info
) + nblocks
*sizeof(gid_t
*), GFP_USER
);
1093 group_info
->ngroups
= gidsetsize
;
1094 group_info
->nblocks
= nblocks
;
1095 atomic_set(&group_info
->usage
, 1);
1097 if (gidsetsize
<= NGROUPS_SMALL
)
1098 group_info
->blocks
[0] = group_info
->small_block
;
1100 for (i
= 0; i
< nblocks
; i
++) {
1102 b
= (void *)__get_free_page(GFP_USER
);
1104 goto out_undo_partial_alloc
;
1105 group_info
->blocks
[i
] = b
;
1110 out_undo_partial_alloc
:
1112 free_page((unsigned long)group_info
->blocks
[i
]);
1118 EXPORT_SYMBOL(groups_alloc
);
1120 void groups_free(struct group_info
*group_info
)
1122 if (group_info
->blocks
[0] != group_info
->small_block
) {
1124 for (i
= 0; i
< group_info
->nblocks
; i
++)
1125 free_page((unsigned long)group_info
->blocks
[i
]);
1130 EXPORT_SYMBOL(groups_free
);
1132 /* export the group_info to a user-space array */
1133 static int groups_to_user(gid_t __user
*grouplist
,
1134 struct group_info
*group_info
)
1137 unsigned int count
= group_info
->ngroups
;
1139 for (i
= 0; i
< group_info
->nblocks
; i
++) {
1140 unsigned int cp_count
= min(NGROUPS_PER_BLOCK
, count
);
1141 unsigned int len
= cp_count
* sizeof(*grouplist
);
1143 if (copy_to_user(grouplist
, group_info
->blocks
[i
], len
))
1146 grouplist
+= NGROUPS_PER_BLOCK
;
1152 /* fill a group_info from a user-space array - it must be allocated already */
1153 static int groups_from_user(struct group_info
*group_info
,
1154 gid_t __user
*grouplist
)
1157 unsigned int count
= group_info
->ngroups
;
1159 for (i
= 0; i
< group_info
->nblocks
; i
++) {
1160 unsigned int cp_count
= min(NGROUPS_PER_BLOCK
, count
);
1161 unsigned int len
= cp_count
* sizeof(*grouplist
);
1163 if (copy_from_user(group_info
->blocks
[i
], grouplist
, len
))
1166 grouplist
+= NGROUPS_PER_BLOCK
;
1172 /* a simple Shell sort */
1173 static void groups_sort(struct group_info
*group_info
)
1175 int base
, max
, stride
;
1176 int gidsetsize
= group_info
->ngroups
;
1178 for (stride
= 1; stride
< gidsetsize
; stride
= 3 * stride
+ 1)
1183 max
= gidsetsize
- stride
;
1184 for (base
= 0; base
< max
; base
++) {
1186 int right
= left
+ stride
;
1187 gid_t tmp
= GROUP_AT(group_info
, right
);
1189 while (left
>= 0 && GROUP_AT(group_info
, left
) > tmp
) {
1190 GROUP_AT(group_info
, right
) =
1191 GROUP_AT(group_info
, left
);
1195 GROUP_AT(group_info
, right
) = tmp
;
1201 /* a simple bsearch */
1202 int groups_search(struct group_info
*group_info
, gid_t grp
)
1204 unsigned int left
, right
;
1210 right
= group_info
->ngroups
;
1211 while (left
< right
) {
1212 unsigned int mid
= (left
+right
)/2;
1213 int cmp
= grp
- GROUP_AT(group_info
, mid
);
1224 /* validate and set current->group_info */
1225 int set_current_groups(struct group_info
*group_info
)
1228 struct group_info
*old_info
;
1230 retval
= security_task_setgroups(group_info
);
1234 groups_sort(group_info
);
1235 get_group_info(group_info
);
1238 old_info
= current
->group_info
;
1239 current
->group_info
= group_info
;
1240 task_unlock(current
);
1242 put_group_info(old_info
);
1247 EXPORT_SYMBOL(set_current_groups
);
1249 SYSCALL_DEFINE2(getgroups
, int, gidsetsize
, gid_t __user
*, grouplist
)
1254 * SMP: Nobody else can change our grouplist. Thus we are
1261 /* no need to grab task_lock here; it cannot change */
1262 i
= current
->group_info
->ngroups
;
1264 if (i
> gidsetsize
) {
1268 if (groups_to_user(grouplist
, current
->group_info
)) {
1278 * SMP: Our groups are copy-on-write. We can set them safely
1279 * without another task interfering.
1282 SYSCALL_DEFINE2(setgroups
, int, gidsetsize
, gid_t __user
*, grouplist
)
1284 struct group_info
*group_info
;
1287 if (!capable(CAP_SETGID
))
1289 if ((unsigned)gidsetsize
> NGROUPS_MAX
)
1292 group_info
= groups_alloc(gidsetsize
);
1295 retval
= groups_from_user(group_info
, grouplist
);
1297 put_group_info(group_info
);
1301 retval
= set_current_groups(group_info
);
1302 put_group_info(group_info
);
1308 * Check whether we're fsgid/egid or in the supplemental group..
1310 int in_group_p(gid_t grp
)
1313 if (grp
!= current
->fsgid
)
1314 retval
= groups_search(current
->group_info
, grp
);
1318 EXPORT_SYMBOL(in_group_p
);
1320 int in_egroup_p(gid_t grp
)
1323 if (grp
!= current
->egid
)
1324 retval
= groups_search(current
->group_info
, grp
);
1328 EXPORT_SYMBOL(in_egroup_p
);
1330 DECLARE_RWSEM(uts_sem
);
1332 SYSCALL_DEFINE1(newuname
, struct new_utsname __user
*, name
)
1336 down_read(&uts_sem
);
1337 if (copy_to_user(name
, utsname(), sizeof *name
))
1343 SYSCALL_DEFINE2(sethostname
, char __user
*, name
, int, len
)
1346 char tmp
[__NEW_UTS_LEN
];
1348 if (!capable(CAP_SYS_ADMIN
))
1350 if (len
< 0 || len
> __NEW_UTS_LEN
)
1352 down_write(&uts_sem
);
1354 if (!copy_from_user(tmp
, name
, len
)) {
1355 memcpy(utsname()->nodename
, tmp
, len
);
1356 utsname()->nodename
[len
] = 0;
1363 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1365 SYSCALL_DEFINE2(gethostname
, char __user
*, name
, int, len
)
1371 down_read(&uts_sem
);
1372 i
= 1 + strlen(utsname()->nodename
);
1376 if (copy_to_user(name
, utsname()->nodename
, i
))
1385 * Only setdomainname; getdomainname can be implemented by calling
1388 SYSCALL_DEFINE2(setdomainname
, char __user
*, name
, int, len
)
1391 char tmp
[__NEW_UTS_LEN
];
1393 if (!capable(CAP_SYS_ADMIN
))
1395 if (len
< 0 || len
> __NEW_UTS_LEN
)
1398 down_write(&uts_sem
);
1400 if (!copy_from_user(tmp
, name
, len
)) {
1401 memcpy(utsname()->domainname
, tmp
, len
);
1402 utsname()->domainname
[len
] = 0;
1409 SYSCALL_DEFINE2(getrlimit
, unsigned int, resource
, struct rlimit __user
*, rlim
)
1411 if (resource
>= RLIM_NLIMITS
)
1414 struct rlimit value
;
1415 task_lock(current
->group_leader
);
1416 value
= current
->signal
->rlim
[resource
];
1417 task_unlock(current
->group_leader
);
1418 return copy_to_user(rlim
, &value
, sizeof(*rlim
)) ? -EFAULT
: 0;
1422 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1425 * Back compatibility for getrlimit. Needed for some apps.
1428 SYSCALL_DEFINE2(old_getrlimit
, unsigned int, resource
,
1429 struct rlimit __user
*, rlim
)
1432 if (resource
>= RLIM_NLIMITS
)
1435 task_lock(current
->group_leader
);
1436 x
= current
->signal
->rlim
[resource
];
1437 task_unlock(current
->group_leader
);
1438 if (x
.rlim_cur
> 0x7FFFFFFF)
1439 x
.rlim_cur
= 0x7FFFFFFF;
1440 if (x
.rlim_max
> 0x7FFFFFFF)
1441 x
.rlim_max
= 0x7FFFFFFF;
1442 return copy_to_user(rlim
, &x
, sizeof(x
))?-EFAULT
:0;
1447 SYSCALL_DEFINE2(setrlimit
, unsigned int, resource
, struct rlimit __user
*, rlim
)
1449 struct rlimit new_rlim
, *old_rlim
;
1450 unsigned long it_prof_secs
;
1453 if (resource
>= RLIM_NLIMITS
)
1455 if (copy_from_user(&new_rlim
, rlim
, sizeof(*rlim
)))
1457 if (new_rlim
.rlim_cur
> new_rlim
.rlim_max
)
1459 old_rlim
= current
->signal
->rlim
+ resource
;
1460 if ((new_rlim
.rlim_max
> old_rlim
->rlim_max
) &&
1461 !capable(CAP_SYS_RESOURCE
))
1463 if (resource
== RLIMIT_NOFILE
&& new_rlim
.rlim_max
> sysctl_nr_open
)
1466 retval
= security_task_setrlimit(resource
, &new_rlim
);
1470 if (resource
== RLIMIT_CPU
&& new_rlim
.rlim_cur
== 0) {
1472 * The caller is asking for an immediate RLIMIT_CPU
1473 * expiry. But we use the zero value to mean "it was
1474 * never set". So let's cheat and make it one second
1477 new_rlim
.rlim_cur
= 1;
1480 task_lock(current
->group_leader
);
1481 *old_rlim
= new_rlim
;
1482 task_unlock(current
->group_leader
);
1484 if (resource
!= RLIMIT_CPU
)
1488 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1489 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1490 * very long-standing error, and fixing it now risks breakage of
1491 * applications, so we live with it
1493 if (new_rlim
.rlim_cur
== RLIM_INFINITY
)
1496 it_prof_secs
= cputime_to_secs(current
->signal
->it_prof_expires
);
1497 if (it_prof_secs
== 0 || new_rlim
.rlim_cur
<= it_prof_secs
) {
1498 unsigned long rlim_cur
= new_rlim
.rlim_cur
;
1501 cputime
= secs_to_cputime(rlim_cur
);
1502 read_lock(&tasklist_lock
);
1503 spin_lock_irq(¤t
->sighand
->siglock
);
1504 set_process_cpu_timer(current
, CPUCLOCK_PROF
, &cputime
, NULL
);
1505 spin_unlock_irq(¤t
->sighand
->siglock
);
1506 read_unlock(&tasklist_lock
);
1513 * It would make sense to put struct rusage in the task_struct,
1514 * except that would make the task_struct be *really big*. After
1515 * task_struct gets moved into malloc'ed memory, it would
1516 * make sense to do this. It will make moving the rest of the information
1517 * a lot simpler! (Which we're not doing right now because we're not
1518 * measuring them yet).
1520 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1521 * races with threads incrementing their own counters. But since word
1522 * reads are atomic, we either get new values or old values and we don't
1523 * care which for the sums. We always take the siglock to protect reading
1524 * the c* fields from p->signal from races with exit.c updating those
1525 * fields when reaping, so a sample either gets all the additions of a
1526 * given child after it's reaped, or none so this sample is before reaping.
1529 * We need to take the siglock for CHILDEREN, SELF and BOTH
1530 * for the cases current multithreaded, non-current single threaded
1531 * non-current multithreaded. Thread traversal is now safe with
1533 * Strictly speaking, we donot need to take the siglock if we are current and
1534 * single threaded, as no one else can take our signal_struct away, no one
1535 * else can reap the children to update signal->c* counters, and no one else
1536 * can race with the signal-> fields. If we do not take any lock, the
1537 * signal-> fields could be read out of order while another thread was just
1538 * exiting. So we should place a read memory barrier when we avoid the lock.
1539 * On the writer side, write memory barrier is implied in __exit_signal
1540 * as __exit_signal releases the siglock spinlock after updating the signal->
1541 * fields. But we don't do this yet to keep things simple.
1545 static void accumulate_thread_rusage(struct task_struct
*t
, struct rusage
*r
,
1546 cputime_t
*utimep
, cputime_t
*stimep
)
1548 *utimep
= cputime_add(*utimep
, t
->utime
);
1549 *stimep
= cputime_add(*stimep
, t
->stime
);
1550 r
->ru_nvcsw
+= t
->nvcsw
;
1551 r
->ru_nivcsw
+= t
->nivcsw
;
1552 r
->ru_minflt
+= t
->min_flt
;
1553 r
->ru_majflt
+= t
->maj_flt
;
1554 r
->ru_inblock
+= task_io_get_inblock(t
);
1555 r
->ru_oublock
+= task_io_get_oublock(t
);
1558 static void k_getrusage(struct task_struct
*p
, int who
, struct rusage
*r
)
1560 struct task_struct
*t
;
1561 unsigned long flags
;
1562 cputime_t utime
, stime
;
1564 memset((char *) r
, 0, sizeof *r
);
1565 utime
= stime
= cputime_zero
;
1567 if (who
== RUSAGE_THREAD
) {
1568 accumulate_thread_rusage(p
, r
, &utime
, &stime
);
1572 if (!lock_task_sighand(p
, &flags
))
1577 case RUSAGE_CHILDREN
:
1578 utime
= p
->signal
->cutime
;
1579 stime
= p
->signal
->cstime
;
1580 r
->ru_nvcsw
= p
->signal
->cnvcsw
;
1581 r
->ru_nivcsw
= p
->signal
->cnivcsw
;
1582 r
->ru_minflt
= p
->signal
->cmin_flt
;
1583 r
->ru_majflt
= p
->signal
->cmaj_flt
;
1584 r
->ru_inblock
= p
->signal
->cinblock
;
1585 r
->ru_oublock
= p
->signal
->coublock
;
1587 if (who
== RUSAGE_CHILDREN
)
1591 utime
= cputime_add(utime
, p
->signal
->utime
);
1592 stime
= cputime_add(stime
, p
->signal
->stime
);
1593 r
->ru_nvcsw
+= p
->signal
->nvcsw
;
1594 r
->ru_nivcsw
+= p
->signal
->nivcsw
;
1595 r
->ru_minflt
+= p
->signal
->min_flt
;
1596 r
->ru_majflt
+= p
->signal
->maj_flt
;
1597 r
->ru_inblock
+= p
->signal
->inblock
;
1598 r
->ru_oublock
+= p
->signal
->oublock
;
1601 accumulate_thread_rusage(t
, r
, &utime
, &stime
);
1609 unlock_task_sighand(p
, &flags
);
1612 cputime_to_timeval(utime
, &r
->ru_utime
);
1613 cputime_to_timeval(stime
, &r
->ru_stime
);
1616 int getrusage(struct task_struct
*p
, int who
, struct rusage __user
*ru
)
1619 k_getrusage(p
, who
, &r
);
1620 return copy_to_user(ru
, &r
, sizeof(r
)) ? -EFAULT
: 0;
1623 SYSCALL_DEFINE2(getrusage
, int, who
, struct rusage __user
*, ru
)
1625 if (who
!= RUSAGE_SELF
&& who
!= RUSAGE_CHILDREN
&&
1626 who
!= RUSAGE_THREAD
)
1628 return getrusage(current
, who
, ru
);
1631 SYSCALL_DEFINE1(umask
, int, mask
)
1633 mask
= xchg(¤t
->fs
->umask
, mask
& S_IRWXUGO
);
1637 SYSCALL_DEFINE5(prctl
, int, option
, unsigned long, arg2
, unsigned long, arg3
,
1638 unsigned long, arg4
, unsigned long, arg5
)
1642 if (security_task_prctl(option
, arg2
, arg3
, arg4
, arg5
, &error
))
1646 case PR_SET_PDEATHSIG
:
1647 if (!valid_signal(arg2
)) {
1651 current
->pdeath_signal
= arg2
;
1653 case PR_GET_PDEATHSIG
:
1654 error
= put_user(current
->pdeath_signal
, (int __user
*)arg2
);
1656 case PR_GET_DUMPABLE
:
1657 error
= get_dumpable(current
->mm
);
1659 case PR_SET_DUMPABLE
:
1660 if (arg2
< 0 || arg2
> 1) {
1664 set_dumpable(current
->mm
, arg2
);
1667 case PR_SET_UNALIGN
:
1668 error
= SET_UNALIGN_CTL(current
, arg2
);
1670 case PR_GET_UNALIGN
:
1671 error
= GET_UNALIGN_CTL(current
, arg2
);
1674 error
= SET_FPEMU_CTL(current
, arg2
);
1677 error
= GET_FPEMU_CTL(current
, arg2
);
1680 error
= SET_FPEXC_CTL(current
, arg2
);
1683 error
= GET_FPEXC_CTL(current
, arg2
);
1686 error
= PR_TIMING_STATISTICAL
;
1689 if (arg2
!= PR_TIMING_STATISTICAL
)
1694 struct task_struct
*me
= current
;
1695 unsigned char ncomm
[sizeof(me
->comm
)];
1697 ncomm
[sizeof(me
->comm
)-1] = 0;
1698 if (strncpy_from_user(ncomm
, (char __user
*)arg2
,
1699 sizeof(me
->comm
)-1) < 0)
1701 set_task_comm(me
, ncomm
);
1705 struct task_struct
*me
= current
;
1706 unsigned char tcomm
[sizeof(me
->comm
)];
1708 get_task_comm(tcomm
, me
);
1709 if (copy_to_user((char __user
*)arg2
, tcomm
, sizeof(tcomm
)))
1714 error
= GET_ENDIAN(current
, arg2
);
1717 error
= SET_ENDIAN(current
, arg2
);
1720 case PR_GET_SECCOMP
:
1721 error
= prctl_get_seccomp();
1723 case PR_SET_SECCOMP
:
1724 error
= prctl_set_seccomp(arg2
);
1727 error
= GET_TSC_CTL(arg2
);
1730 error
= SET_TSC_CTL(arg2
);
1739 SYSCALL_DEFINE3(getcpu
, unsigned __user
*, cpup
, unsigned __user
*, nodep
,
1740 struct getcpu_cache __user
*, unused
)
1743 int cpu
= raw_smp_processor_id();
1745 err
|= put_user(cpu
, cpup
);
1747 err
|= put_user(cpu_to_node(cpu
), nodep
);
1748 return err
? -EFAULT
: 0;
1751 char poweroff_cmd
[POWEROFF_CMD_PATH_LEN
] = "/sbin/poweroff";
1753 static void argv_cleanup(char **argv
, char **envp
)
1759 * orderly_poweroff - Trigger an orderly system poweroff
1760 * @force: force poweroff if command execution fails
1762 * This may be called from any context to trigger a system shutdown.
1763 * If the orderly shutdown fails, it will force an immediate shutdown.
1765 int orderly_poweroff(bool force
)
1768 char **argv
= argv_split(GFP_ATOMIC
, poweroff_cmd
, &argc
);
1769 static char *envp
[] = {
1771 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1775 struct subprocess_info
*info
;
1778 printk(KERN_WARNING
"%s failed to allocate memory for \"%s\"\n",
1779 __func__
, poweroff_cmd
);
1783 info
= call_usermodehelper_setup(argv
[0], argv
, envp
, GFP_ATOMIC
);
1789 call_usermodehelper_setcleanup(info
, argv_cleanup
);
1791 ret
= call_usermodehelper_exec(info
, UMH_NO_WAIT
);
1795 printk(KERN_WARNING
"Failed to start orderly shutdown: "
1796 "forcing the issue\n");
1798 /* I guess this should try to kick off some daemon to
1799 sync and poweroff asap. Or not even bother syncing
1800 if we're doing an emergency shutdown? */
1807 EXPORT_SYMBOL_GPL(orderly_poweroff
);