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/perf_counter.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/kexec.h>
21 #include <linux/workqueue.h>
22 #include <linux/capability.h>
23 #include <linux/device.h>
24 #include <linux/key.h>
25 #include <linux/times.h>
26 #include <linux/posix-timers.h>
27 #include <linux/security.h>
28 #include <linux/dcookies.h>
29 #include <linux/suspend.h>
30 #include <linux/tty.h>
31 #include <linux/signal.h>
32 #include <linux/cn_proc.h>
33 #include <linux/getcpu.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/seccomp.h>
36 #include <linux/cpu.h>
37 #include <linux/ptrace.h>
39 #include <linux/compat.h>
40 #include <linux/syscalls.h>
41 #include <linux/kprobes.h>
42 #include <linux/user_namespace.h>
44 #include <asm/uaccess.h>
46 #include <asm/unistd.h>
48 #ifndef SET_UNALIGN_CTL
49 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
51 #ifndef GET_UNALIGN_CTL
52 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
55 # define SET_FPEMU_CTL(a,b) (-EINVAL)
58 # define GET_FPEMU_CTL(a,b) (-EINVAL)
61 # define SET_FPEXC_CTL(a,b) (-EINVAL)
64 # define GET_FPEXC_CTL(a,b) (-EINVAL)
67 # define GET_ENDIAN(a,b) (-EINVAL)
70 # define SET_ENDIAN(a,b) (-EINVAL)
73 # define GET_TSC_CTL(a) (-EINVAL)
76 # define SET_TSC_CTL(a) (-EINVAL)
80 * this is where the system-wide overflow UID and GID are defined, for
81 * architectures that now have 32-bit UID/GID but didn't in the past
84 int overflowuid
= DEFAULT_OVERFLOWUID
;
85 int overflowgid
= DEFAULT_OVERFLOWGID
;
88 EXPORT_SYMBOL(overflowuid
);
89 EXPORT_SYMBOL(overflowgid
);
93 * the same as above, but for filesystems which can only store a 16-bit
94 * UID and GID. as such, this is needed on all architectures
97 int fs_overflowuid
= DEFAULT_FS_OVERFLOWUID
;
98 int fs_overflowgid
= DEFAULT_FS_OVERFLOWUID
;
100 EXPORT_SYMBOL(fs_overflowuid
);
101 EXPORT_SYMBOL(fs_overflowgid
);
104 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
109 EXPORT_SYMBOL(cad_pid
);
112 * If set, this is used for preparing the system to power off.
115 void (*pm_power_off_prepare
)(void);
118 * set the priority of a task
119 * - the caller must hold the RCU read lock
121 static int set_one_prio(struct task_struct
*p
, int niceval
, int error
)
123 const struct cred
*cred
= current_cred(), *pcred
= __task_cred(p
);
126 if (pcred
->uid
!= cred
->euid
&&
127 pcred
->euid
!= cred
->euid
&& !capable(CAP_SYS_NICE
)) {
131 if (niceval
< task_nice(p
) && !can_nice(p
, niceval
)) {
135 no_nice
= security_task_setnice(p
, niceval
);
142 set_user_nice(p
, niceval
);
147 SYSCALL_DEFINE3(setpriority
, int, which
, int, who
, int, niceval
)
149 struct task_struct
*g
, *p
;
150 struct user_struct
*user
;
151 const struct cred
*cred
= current_cred();
155 if (which
> PRIO_USER
|| which
< PRIO_PROCESS
)
158 /* normalize: avoid signed division (rounding problems) */
165 read_lock(&tasklist_lock
);
169 p
= find_task_by_vpid(who
);
173 error
= set_one_prio(p
, niceval
, error
);
177 pgrp
= find_vpid(who
);
179 pgrp
= task_pgrp(current
);
180 do_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
) {
181 error
= set_one_prio(p
, niceval
, error
);
182 } while_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
);
185 user
= (struct user_struct
*) cred
->user
;
188 else if ((who
!= cred
->uid
) &&
189 !(user
= find_user(who
)))
190 goto out_unlock
; /* No processes for this user */
193 if (__task_cred(p
)->uid
== who
)
194 error
= set_one_prio(p
, niceval
, error
);
195 while_each_thread(g
, p
);
196 if (who
!= cred
->uid
)
197 free_uid(user
); /* For find_user() */
201 read_unlock(&tasklist_lock
);
207 * Ugh. To avoid negative return values, "getpriority()" will
208 * not return the normal nice-value, but a negated value that
209 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
210 * to stay compatible.
212 SYSCALL_DEFINE2(getpriority
, int, which
, int, who
)
214 struct task_struct
*g
, *p
;
215 struct user_struct
*user
;
216 const struct cred
*cred
= current_cred();
217 long niceval
, retval
= -ESRCH
;
220 if (which
> PRIO_USER
|| which
< PRIO_PROCESS
)
223 read_lock(&tasklist_lock
);
227 p
= find_task_by_vpid(who
);
231 niceval
= 20 - task_nice(p
);
232 if (niceval
> retval
)
238 pgrp
= find_vpid(who
);
240 pgrp
= task_pgrp(current
);
241 do_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
) {
242 niceval
= 20 - task_nice(p
);
243 if (niceval
> retval
)
245 } while_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
);
248 user
= (struct user_struct
*) cred
->user
;
251 else if ((who
!= cred
->uid
) &&
252 !(user
= find_user(who
)))
253 goto out_unlock
; /* No processes for this user */
256 if (__task_cred(p
)->uid
== who
) {
257 niceval
= 20 - task_nice(p
);
258 if (niceval
> retval
)
261 while_each_thread(g
, p
);
262 if (who
!= cred
->uid
)
263 free_uid(user
); /* for find_user() */
267 read_unlock(&tasklist_lock
);
273 * emergency_restart - reboot the system
275 * Without shutting down any hardware or taking any locks
276 * reboot the system. This is called when we know we are in
277 * trouble so this is our best effort to reboot. This is
278 * safe to call in interrupt context.
280 void emergency_restart(void)
282 machine_emergency_restart();
284 EXPORT_SYMBOL_GPL(emergency_restart
);
286 void kernel_restart_prepare(char *cmd
)
288 blocking_notifier_call_chain(&reboot_notifier_list
, SYS_RESTART
, cmd
);
289 system_state
= SYSTEM_RESTART
;
295 * kernel_restart - reboot the system
296 * @cmd: pointer to buffer containing command to execute for restart
299 * Shutdown everything and perform a clean reboot.
300 * This is not safe to call in interrupt context.
302 void kernel_restart(char *cmd
)
304 kernel_restart_prepare(cmd
);
306 printk(KERN_EMERG
"Restarting system.\n");
308 printk(KERN_EMERG
"Restarting system with command '%s'.\n", cmd
);
309 machine_restart(cmd
);
311 EXPORT_SYMBOL_GPL(kernel_restart
);
313 static void kernel_shutdown_prepare(enum system_states state
)
315 blocking_notifier_call_chain(&reboot_notifier_list
,
316 (state
== SYSTEM_HALT
)?SYS_HALT
:SYS_POWER_OFF
, NULL
);
317 system_state
= state
;
321 * kernel_halt - halt the system
323 * Shutdown everything and perform a clean system halt.
325 void kernel_halt(void)
327 kernel_shutdown_prepare(SYSTEM_HALT
);
329 printk(KERN_EMERG
"System halted.\n");
333 EXPORT_SYMBOL_GPL(kernel_halt
);
336 * kernel_power_off - power_off the system
338 * Shutdown everything and perform a clean system power_off.
340 void kernel_power_off(void)
342 kernel_shutdown_prepare(SYSTEM_POWER_OFF
);
343 if (pm_power_off_prepare
)
344 pm_power_off_prepare();
345 disable_nonboot_cpus();
347 printk(KERN_EMERG
"Power down.\n");
350 EXPORT_SYMBOL_GPL(kernel_power_off
);
352 * Reboot system call: for obvious reasons only root may call it,
353 * and even root needs to set up some magic numbers in the registers
354 * so that some mistake won't make this reboot the whole machine.
355 * You can also set the meaning of the ctrl-alt-del-key here.
357 * reboot doesn't sync: do that yourself before calling this.
359 SYSCALL_DEFINE4(reboot
, int, magic1
, int, magic2
, unsigned int, cmd
,
364 /* We only trust the superuser with rebooting the system. */
365 if (!capable(CAP_SYS_BOOT
))
368 /* For safety, we require "magic" arguments. */
369 if (magic1
!= LINUX_REBOOT_MAGIC1
||
370 (magic2
!= LINUX_REBOOT_MAGIC2
&&
371 magic2
!= LINUX_REBOOT_MAGIC2A
&&
372 magic2
!= LINUX_REBOOT_MAGIC2B
&&
373 magic2
!= LINUX_REBOOT_MAGIC2C
))
376 /* Instead of trying to make the power_off code look like
377 * halt when pm_power_off is not set do it the easy way.
379 if ((cmd
== LINUX_REBOOT_CMD_POWER_OFF
) && !pm_power_off
)
380 cmd
= LINUX_REBOOT_CMD_HALT
;
384 case LINUX_REBOOT_CMD_RESTART
:
385 kernel_restart(NULL
);
388 case LINUX_REBOOT_CMD_CAD_ON
:
392 case LINUX_REBOOT_CMD_CAD_OFF
:
396 case LINUX_REBOOT_CMD_HALT
:
402 case LINUX_REBOOT_CMD_POWER_OFF
:
408 case LINUX_REBOOT_CMD_RESTART2
:
409 if (strncpy_from_user(&buffer
[0], arg
, sizeof(buffer
) - 1) < 0) {
413 buffer
[sizeof(buffer
) - 1] = '\0';
415 kernel_restart(buffer
);
419 case LINUX_REBOOT_CMD_KEXEC
:
422 ret
= kernel_kexec();
428 #ifdef CONFIG_HIBERNATION
429 case LINUX_REBOOT_CMD_SW_SUSPEND
:
431 int ret
= hibernate();
445 static void deferred_cad(struct work_struct
*dummy
)
447 kernel_restart(NULL
);
451 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
452 * As it's called within an interrupt, it may NOT sync: the only choice
453 * is whether to reboot at once, or just ignore the ctrl-alt-del.
455 void ctrl_alt_del(void)
457 static DECLARE_WORK(cad_work
, deferred_cad
);
460 schedule_work(&cad_work
);
462 kill_cad_pid(SIGINT
, 1);
466 * Unprivileged users may change the real gid to the effective gid
467 * or vice versa. (BSD-style)
469 * If you set the real gid at all, or set the effective gid to a value not
470 * equal to the real gid, then the saved gid is set to the new effective gid.
472 * This makes it possible for a setgid program to completely drop its
473 * privileges, which is often a useful assertion to make when you are doing
474 * a security audit over a program.
476 * The general idea is that a program which uses just setregid() will be
477 * 100% compatible with BSD. A program which uses just setgid() will be
478 * 100% compatible with POSIX with saved IDs.
480 * SMP: There are not races, the GIDs are checked only by filesystem
481 * operations (as far as semantic preservation is concerned).
483 SYSCALL_DEFINE2(setregid
, gid_t
, rgid
, gid_t
, egid
)
485 const struct cred
*old
;
489 new = prepare_creds();
492 old
= current_cred();
494 retval
= security_task_setgid(rgid
, egid
, (gid_t
)-1, LSM_SETID_RE
);
499 if (rgid
!= (gid_t
) -1) {
500 if (old
->gid
== rgid
||
507 if (egid
!= (gid_t
) -1) {
508 if (old
->gid
== egid
||
517 if (rgid
!= (gid_t
) -1 ||
518 (egid
!= (gid_t
) -1 && egid
!= old
->gid
))
519 new->sgid
= new->egid
;
520 new->fsgid
= new->egid
;
522 return commit_creds(new);
530 * setgid() is implemented like SysV w/ SAVED_IDS
532 * SMP: Same implicit races as above.
534 SYSCALL_DEFINE1(setgid
, gid_t
, gid
)
536 const struct cred
*old
;
540 new = prepare_creds();
543 old
= current_cred();
545 retval
= security_task_setgid(gid
, (gid_t
)-1, (gid_t
)-1, LSM_SETID_ID
);
550 if (capable(CAP_SETGID
))
551 new->gid
= new->egid
= new->sgid
= new->fsgid
= gid
;
552 else if (gid
== old
->gid
|| gid
== old
->sgid
)
553 new->egid
= new->fsgid
= gid
;
557 return commit_creds(new);
565 * change the user struct in a credentials set to match the new UID
567 static int set_user(struct cred
*new)
569 struct user_struct
*new_user
;
571 new_user
= alloc_uid(current_user_ns(), new->uid
);
575 if (atomic_read(&new_user
->processes
) >=
576 current
->signal
->rlim
[RLIMIT_NPROC
].rlim_cur
&&
577 new_user
!= INIT_USER
) {
583 new->user
= new_user
;
588 * Unprivileged users may change the real uid to the effective uid
589 * or vice versa. (BSD-style)
591 * If you set the real uid at all, or set the effective uid to a value not
592 * equal to the real uid, then the saved uid is set to the new effective uid.
594 * This makes it possible for a setuid program to completely drop its
595 * privileges, which is often a useful assertion to make when you are doing
596 * a security audit over a program.
598 * The general idea is that a program which uses just setreuid() will be
599 * 100% compatible with BSD. A program which uses just setuid() will be
600 * 100% compatible with POSIX with saved IDs.
602 SYSCALL_DEFINE2(setreuid
, uid_t
, ruid
, uid_t
, euid
)
604 const struct cred
*old
;
608 new = prepare_creds();
611 old
= current_cred();
613 retval
= security_task_setuid(ruid
, euid
, (uid_t
)-1, LSM_SETID_RE
);
618 if (ruid
!= (uid_t
) -1) {
620 if (old
->uid
!= ruid
&&
622 !capable(CAP_SETUID
))
626 if (euid
!= (uid_t
) -1) {
628 if (old
->uid
!= euid
&&
631 !capable(CAP_SETUID
))
636 if (new->uid
!= old
->uid
&& set_user(new) < 0)
639 if (ruid
!= (uid_t
) -1 ||
640 (euid
!= (uid_t
) -1 && euid
!= old
->uid
))
641 new->suid
= new->euid
;
642 new->fsuid
= new->euid
;
644 retval
= security_task_fix_setuid(new, old
, LSM_SETID_RE
);
648 return commit_creds(new);
656 * setuid() is implemented like SysV with SAVED_IDS
658 * Note that SAVED_ID's is deficient in that a setuid root program
659 * like sendmail, for example, cannot set its uid to be a normal
660 * user and then switch back, because if you're root, setuid() sets
661 * the saved uid too. If you don't like this, blame the bright people
662 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
663 * will allow a root program to temporarily drop privileges and be able to
664 * regain them by swapping the real and effective uid.
666 SYSCALL_DEFINE1(setuid
, uid_t
, uid
)
668 const struct cred
*old
;
672 new = prepare_creds();
675 old
= current_cred();
677 retval
= security_task_setuid(uid
, (uid_t
)-1, (uid_t
)-1, LSM_SETID_ID
);
682 if (capable(CAP_SETUID
)) {
683 new->suid
= new->uid
= uid
;
684 if (uid
!= old
->uid
&& set_user(new) < 0) {
688 } else if (uid
!= old
->uid
&& uid
!= new->suid
) {
692 new->fsuid
= new->euid
= uid
;
694 retval
= security_task_fix_setuid(new, old
, LSM_SETID_ID
);
698 return commit_creds(new);
707 * This function implements a generic ability to update ruid, euid,
708 * and suid. This allows you to implement the 4.4 compatible seteuid().
710 SYSCALL_DEFINE3(setresuid
, uid_t
, ruid
, uid_t
, euid
, uid_t
, suid
)
712 const struct cred
*old
;
716 new = prepare_creds();
720 retval
= security_task_setuid(ruid
, euid
, suid
, LSM_SETID_RES
);
723 old
= current_cred();
726 if (!capable(CAP_SETUID
)) {
727 if (ruid
!= (uid_t
) -1 && ruid
!= old
->uid
&&
728 ruid
!= old
->euid
&& ruid
!= old
->suid
)
730 if (euid
!= (uid_t
) -1 && euid
!= old
->uid
&&
731 euid
!= old
->euid
&& euid
!= old
->suid
)
733 if (suid
!= (uid_t
) -1 && suid
!= old
->uid
&&
734 suid
!= old
->euid
&& suid
!= old
->suid
)
739 if (ruid
!= (uid_t
) -1) {
741 if (ruid
!= old
->uid
&& set_user(new) < 0)
744 if (euid
!= (uid_t
) -1)
746 if (suid
!= (uid_t
) -1)
748 new->fsuid
= new->euid
;
750 retval
= security_task_fix_setuid(new, old
, LSM_SETID_RES
);
754 return commit_creds(new);
761 SYSCALL_DEFINE3(getresuid
, uid_t __user
*, ruid
, uid_t __user
*, euid
, uid_t __user
*, suid
)
763 const struct cred
*cred
= current_cred();
766 if (!(retval
= put_user(cred
->uid
, ruid
)) &&
767 !(retval
= put_user(cred
->euid
, euid
)))
768 retval
= put_user(cred
->suid
, suid
);
774 * Same as above, but for rgid, egid, sgid.
776 SYSCALL_DEFINE3(setresgid
, gid_t
, rgid
, gid_t
, egid
, gid_t
, sgid
)
778 const struct cred
*old
;
782 new = prepare_creds();
785 old
= current_cred();
787 retval
= security_task_setgid(rgid
, egid
, sgid
, LSM_SETID_RES
);
792 if (!capable(CAP_SETGID
)) {
793 if (rgid
!= (gid_t
) -1 && rgid
!= old
->gid
&&
794 rgid
!= old
->egid
&& rgid
!= old
->sgid
)
796 if (egid
!= (gid_t
) -1 && egid
!= old
->gid
&&
797 egid
!= old
->egid
&& egid
!= old
->sgid
)
799 if (sgid
!= (gid_t
) -1 && sgid
!= old
->gid
&&
800 sgid
!= old
->egid
&& sgid
!= old
->sgid
)
804 if (rgid
!= (gid_t
) -1)
806 if (egid
!= (gid_t
) -1)
808 if (sgid
!= (gid_t
) -1)
810 new->fsgid
= new->egid
;
812 return commit_creds(new);
819 SYSCALL_DEFINE3(getresgid
, gid_t __user
*, rgid
, gid_t __user
*, egid
, gid_t __user
*, sgid
)
821 const struct cred
*cred
= current_cred();
824 if (!(retval
= put_user(cred
->gid
, rgid
)) &&
825 !(retval
= put_user(cred
->egid
, egid
)))
826 retval
= put_user(cred
->sgid
, sgid
);
833 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
834 * is used for "access()" and for the NFS daemon (letting nfsd stay at
835 * whatever uid it wants to). It normally shadows "euid", except when
836 * explicitly set by setfsuid() or for access..
838 SYSCALL_DEFINE1(setfsuid
, uid_t
, uid
)
840 const struct cred
*old
;
844 new = prepare_creds();
846 return current_fsuid();
847 old
= current_cred();
848 old_fsuid
= old
->fsuid
;
850 if (security_task_setuid(uid
, (uid_t
)-1, (uid_t
)-1, LSM_SETID_FS
) < 0)
853 if (uid
== old
->uid
|| uid
== old
->euid
||
854 uid
== old
->suid
|| uid
== old
->fsuid
||
855 capable(CAP_SETUID
)) {
856 if (uid
!= old_fsuid
) {
858 if (security_task_fix_setuid(new, old
, LSM_SETID_FS
) == 0)
873 * Samma på svenska..
875 SYSCALL_DEFINE1(setfsgid
, gid_t
, gid
)
877 const struct cred
*old
;
881 new = prepare_creds();
883 return current_fsgid();
884 old
= current_cred();
885 old_fsgid
= old
->fsgid
;
887 if (security_task_setgid(gid
, (gid_t
)-1, (gid_t
)-1, LSM_SETID_FS
))
890 if (gid
== old
->gid
|| gid
== old
->egid
||
891 gid
== old
->sgid
|| gid
== old
->fsgid
||
892 capable(CAP_SETGID
)) {
893 if (gid
!= old_fsgid
) {
908 void do_sys_times(struct tms
*tms
)
910 struct task_cputime cputime
;
911 cputime_t cutime
, cstime
;
913 thread_group_cputime(current
, &cputime
);
914 spin_lock_irq(¤t
->sighand
->siglock
);
915 cutime
= current
->signal
->cutime
;
916 cstime
= current
->signal
->cstime
;
917 spin_unlock_irq(¤t
->sighand
->siglock
);
918 tms
->tms_utime
= cputime_to_clock_t(cputime
.utime
);
919 tms
->tms_stime
= cputime_to_clock_t(cputime
.stime
);
920 tms
->tms_cutime
= cputime_to_clock_t(cutime
);
921 tms
->tms_cstime
= cputime_to_clock_t(cstime
);
924 SYSCALL_DEFINE1(times
, struct tms __user
*, tbuf
)
930 if (copy_to_user(tbuf
, &tmp
, sizeof(struct tms
)))
933 force_successful_syscall_return();
934 return (long) jiffies_64_to_clock_t(get_jiffies_64());
938 * This needs some heavy checking ...
939 * I just haven't the stomach for it. I also don't fully
940 * understand sessions/pgrp etc. Let somebody who does explain it.
942 * OK, I think I have the protection semantics right.... this is really
943 * only important on a multi-user system anyway, to make sure one user
944 * can't send a signal to a process owned by another. -TYT, 12/12/91
946 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
949 SYSCALL_DEFINE2(setpgid
, pid_t
, pid
, pid_t
, pgid
)
951 struct task_struct
*p
;
952 struct task_struct
*group_leader
= current
->group_leader
;
957 pid
= task_pid_vnr(group_leader
);
963 /* From this point forward we keep holding onto the tasklist lock
964 * so that our parent does not change from under us. -DaveM
966 write_lock_irq(&tasklist_lock
);
969 p
= find_task_by_vpid(pid
);
974 if (!thread_group_leader(p
))
977 if (same_thread_group(p
->real_parent
, group_leader
)) {
979 if (task_session(p
) != task_session(group_leader
))
986 if (p
!= group_leader
)
991 if (p
->signal
->leader
)
996 struct task_struct
*g
;
998 pgrp
= find_vpid(pgid
);
999 g
= pid_task(pgrp
, PIDTYPE_PGID
);
1000 if (!g
|| task_session(g
) != task_session(group_leader
))
1004 err
= security_task_setpgid(p
, pgid
);
1008 if (task_pgrp(p
) != pgrp
) {
1009 change_pid(p
, PIDTYPE_PGID
, pgrp
);
1010 set_task_pgrp(p
, pid_nr(pgrp
));
1015 /* All paths lead to here, thus we are safe. -DaveM */
1016 write_unlock_irq(&tasklist_lock
);
1020 SYSCALL_DEFINE1(getpgid
, pid_t
, pid
)
1022 struct task_struct
*p
;
1028 grp
= task_pgrp(current
);
1031 p
= find_task_by_vpid(pid
);
1038 retval
= security_task_getpgid(p
);
1042 retval
= pid_vnr(grp
);
1048 #ifdef __ARCH_WANT_SYS_GETPGRP
1050 SYSCALL_DEFINE0(getpgrp
)
1052 return sys_getpgid(0);
1057 SYSCALL_DEFINE1(getsid
, pid_t
, pid
)
1059 struct task_struct
*p
;
1065 sid
= task_session(current
);
1068 p
= find_task_by_vpid(pid
);
1071 sid
= task_session(p
);
1075 retval
= security_task_getsid(p
);
1079 retval
= pid_vnr(sid
);
1085 SYSCALL_DEFINE0(setsid
)
1087 struct task_struct
*group_leader
= current
->group_leader
;
1088 struct pid
*sid
= task_pid(group_leader
);
1089 pid_t session
= pid_vnr(sid
);
1092 write_lock_irq(&tasklist_lock
);
1093 /* Fail if I am already a session leader */
1094 if (group_leader
->signal
->leader
)
1097 /* Fail if a process group id already exists that equals the
1098 * proposed session id.
1100 if (pid_task(sid
, PIDTYPE_PGID
))
1103 group_leader
->signal
->leader
= 1;
1104 __set_special_pids(sid
);
1106 proc_clear_tty(group_leader
);
1110 write_unlock_irq(&tasklist_lock
);
1115 * Supplementary group IDs
1118 /* init to 2 - one for init_task, one to ensure it is never freed */
1119 struct group_info init_groups
= { .usage
= ATOMIC_INIT(2) };
1121 struct group_info
*groups_alloc(int gidsetsize
)
1123 struct group_info
*group_info
;
1127 nblocks
= (gidsetsize
+ NGROUPS_PER_BLOCK
- 1) / NGROUPS_PER_BLOCK
;
1128 /* Make sure we always allocate at least one indirect block pointer */
1129 nblocks
= nblocks
? : 1;
1130 group_info
= kmalloc(sizeof(*group_info
) + nblocks
*sizeof(gid_t
*), GFP_USER
);
1133 group_info
->ngroups
= gidsetsize
;
1134 group_info
->nblocks
= nblocks
;
1135 atomic_set(&group_info
->usage
, 1);
1137 if (gidsetsize
<= NGROUPS_SMALL
)
1138 group_info
->blocks
[0] = group_info
->small_block
;
1140 for (i
= 0; i
< nblocks
; i
++) {
1142 b
= (void *)__get_free_page(GFP_USER
);
1144 goto out_undo_partial_alloc
;
1145 group_info
->blocks
[i
] = b
;
1150 out_undo_partial_alloc
:
1152 free_page((unsigned long)group_info
->blocks
[i
]);
1158 EXPORT_SYMBOL(groups_alloc
);
1160 void groups_free(struct group_info
*group_info
)
1162 if (group_info
->blocks
[0] != group_info
->small_block
) {
1164 for (i
= 0; i
< group_info
->nblocks
; i
++)
1165 free_page((unsigned long)group_info
->blocks
[i
]);
1170 EXPORT_SYMBOL(groups_free
);
1172 /* export the group_info to a user-space array */
1173 static int groups_to_user(gid_t __user
*grouplist
,
1174 const struct group_info
*group_info
)
1177 unsigned int count
= group_info
->ngroups
;
1179 for (i
= 0; i
< group_info
->nblocks
; i
++) {
1180 unsigned int cp_count
= min(NGROUPS_PER_BLOCK
, count
);
1181 unsigned int len
= cp_count
* sizeof(*grouplist
);
1183 if (copy_to_user(grouplist
, group_info
->blocks
[i
], len
))
1186 grouplist
+= NGROUPS_PER_BLOCK
;
1192 /* fill a group_info from a user-space array - it must be allocated already */
1193 static int groups_from_user(struct group_info
*group_info
,
1194 gid_t __user
*grouplist
)
1197 unsigned int count
= group_info
->ngroups
;
1199 for (i
= 0; i
< group_info
->nblocks
; i
++) {
1200 unsigned int cp_count
= min(NGROUPS_PER_BLOCK
, count
);
1201 unsigned int len
= cp_count
* sizeof(*grouplist
);
1203 if (copy_from_user(group_info
->blocks
[i
], grouplist
, len
))
1206 grouplist
+= NGROUPS_PER_BLOCK
;
1212 /* a simple Shell sort */
1213 static void groups_sort(struct group_info
*group_info
)
1215 int base
, max
, stride
;
1216 int gidsetsize
= group_info
->ngroups
;
1218 for (stride
= 1; stride
< gidsetsize
; stride
= 3 * stride
+ 1)
1223 max
= gidsetsize
- stride
;
1224 for (base
= 0; base
< max
; base
++) {
1226 int right
= left
+ stride
;
1227 gid_t tmp
= GROUP_AT(group_info
, right
);
1229 while (left
>= 0 && GROUP_AT(group_info
, left
) > tmp
) {
1230 GROUP_AT(group_info
, right
) =
1231 GROUP_AT(group_info
, left
);
1235 GROUP_AT(group_info
, right
) = tmp
;
1241 /* a simple bsearch */
1242 int groups_search(const struct group_info
*group_info
, gid_t grp
)
1244 unsigned int left
, right
;
1250 right
= group_info
->ngroups
;
1251 while (left
< right
) {
1252 unsigned int mid
= (left
+right
)/2;
1253 int cmp
= grp
- GROUP_AT(group_info
, mid
);
1265 * set_groups - Change a group subscription in a set of credentials
1266 * @new: The newly prepared set of credentials to alter
1267 * @group_info: The group list to install
1269 * Validate a group subscription and, if valid, insert it into a set
1272 int set_groups(struct cred
*new, struct group_info
*group_info
)
1276 retval
= security_task_setgroups(group_info
);
1280 put_group_info(new->group_info
);
1281 groups_sort(group_info
);
1282 get_group_info(group_info
);
1283 new->group_info
= group_info
;
1287 EXPORT_SYMBOL(set_groups
);
1290 * set_current_groups - Change current's group subscription
1291 * @group_info: The group list to impose
1293 * Validate a group subscription and, if valid, impose it upon current's task
1296 int set_current_groups(struct group_info
*group_info
)
1301 new = prepare_creds();
1305 ret
= set_groups(new, group_info
);
1311 return commit_creds(new);
1314 EXPORT_SYMBOL(set_current_groups
);
1316 SYSCALL_DEFINE2(getgroups
, int, gidsetsize
, gid_t __user
*, grouplist
)
1318 const struct cred
*cred
= current_cred();
1324 /* no need to grab task_lock here; it cannot change */
1325 i
= cred
->group_info
->ngroups
;
1327 if (i
> gidsetsize
) {
1331 if (groups_to_user(grouplist
, cred
->group_info
)) {
1341 * SMP: Our groups are copy-on-write. We can set them safely
1342 * without another task interfering.
1345 SYSCALL_DEFINE2(setgroups
, int, gidsetsize
, gid_t __user
*, grouplist
)
1347 struct group_info
*group_info
;
1350 if (!capable(CAP_SETGID
))
1352 if ((unsigned)gidsetsize
> NGROUPS_MAX
)
1355 group_info
= groups_alloc(gidsetsize
);
1358 retval
= groups_from_user(group_info
, grouplist
);
1360 put_group_info(group_info
);
1364 retval
= set_current_groups(group_info
);
1365 put_group_info(group_info
);
1371 * Check whether we're fsgid/egid or in the supplemental group..
1373 int in_group_p(gid_t grp
)
1375 const struct cred
*cred
= current_cred();
1378 if (grp
!= cred
->fsgid
)
1379 retval
= groups_search(cred
->group_info
, grp
);
1383 EXPORT_SYMBOL(in_group_p
);
1385 int in_egroup_p(gid_t grp
)
1387 const struct cred
*cred
= current_cred();
1390 if (grp
!= cred
->egid
)
1391 retval
= groups_search(cred
->group_info
, grp
);
1395 EXPORT_SYMBOL(in_egroup_p
);
1397 DECLARE_RWSEM(uts_sem
);
1399 SYSCALL_DEFINE1(newuname
, struct new_utsname __user
*, name
)
1403 down_read(&uts_sem
);
1404 if (copy_to_user(name
, utsname(), sizeof *name
))
1410 SYSCALL_DEFINE2(sethostname
, char __user
*, name
, int, len
)
1413 char tmp
[__NEW_UTS_LEN
];
1415 if (!capable(CAP_SYS_ADMIN
))
1417 if (len
< 0 || len
> __NEW_UTS_LEN
)
1419 down_write(&uts_sem
);
1421 if (!copy_from_user(tmp
, name
, len
)) {
1422 struct new_utsname
*u
= utsname();
1424 memcpy(u
->nodename
, tmp
, len
);
1425 memset(u
->nodename
+ len
, 0, sizeof(u
->nodename
) - len
);
1432 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1434 SYSCALL_DEFINE2(gethostname
, char __user
*, name
, int, len
)
1437 struct new_utsname
*u
;
1441 down_read(&uts_sem
);
1443 i
= 1 + strlen(u
->nodename
);
1447 if (copy_to_user(name
, u
->nodename
, i
))
1456 * Only setdomainname; getdomainname can be implemented by calling
1459 SYSCALL_DEFINE2(setdomainname
, char __user
*, name
, int, len
)
1462 char tmp
[__NEW_UTS_LEN
];
1464 if (!capable(CAP_SYS_ADMIN
))
1466 if (len
< 0 || len
> __NEW_UTS_LEN
)
1469 down_write(&uts_sem
);
1471 if (!copy_from_user(tmp
, name
, len
)) {
1472 struct new_utsname
*u
= utsname();
1474 memcpy(u
->domainname
, tmp
, len
);
1475 memset(u
->domainname
+ len
, 0, sizeof(u
->domainname
) - len
);
1482 SYSCALL_DEFINE2(getrlimit
, unsigned int, resource
, struct rlimit __user
*, rlim
)
1484 if (resource
>= RLIM_NLIMITS
)
1487 struct rlimit value
;
1488 task_lock(current
->group_leader
);
1489 value
= current
->signal
->rlim
[resource
];
1490 task_unlock(current
->group_leader
);
1491 return copy_to_user(rlim
, &value
, sizeof(*rlim
)) ? -EFAULT
: 0;
1495 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1498 * Back compatibility for getrlimit. Needed for some apps.
1501 SYSCALL_DEFINE2(old_getrlimit
, unsigned int, resource
,
1502 struct rlimit __user
*, rlim
)
1505 if (resource
>= RLIM_NLIMITS
)
1508 task_lock(current
->group_leader
);
1509 x
= current
->signal
->rlim
[resource
];
1510 task_unlock(current
->group_leader
);
1511 if (x
.rlim_cur
> 0x7FFFFFFF)
1512 x
.rlim_cur
= 0x7FFFFFFF;
1513 if (x
.rlim_max
> 0x7FFFFFFF)
1514 x
.rlim_max
= 0x7FFFFFFF;
1515 return copy_to_user(rlim
, &x
, sizeof(x
))?-EFAULT
:0;
1520 SYSCALL_DEFINE2(setrlimit
, unsigned int, resource
, struct rlimit __user
*, rlim
)
1522 struct rlimit new_rlim
, *old_rlim
;
1525 if (resource
>= RLIM_NLIMITS
)
1527 if (copy_from_user(&new_rlim
, rlim
, sizeof(*rlim
)))
1529 old_rlim
= current
->signal
->rlim
+ resource
;
1530 if ((new_rlim
.rlim_max
> old_rlim
->rlim_max
) &&
1531 !capable(CAP_SYS_RESOURCE
))
1534 if (resource
== RLIMIT_NOFILE
) {
1535 if (new_rlim
.rlim_max
== RLIM_INFINITY
)
1536 new_rlim
.rlim_max
= sysctl_nr_open
;
1537 if (new_rlim
.rlim_cur
== RLIM_INFINITY
)
1538 new_rlim
.rlim_cur
= sysctl_nr_open
;
1539 if (new_rlim
.rlim_max
> sysctl_nr_open
)
1543 if (new_rlim
.rlim_cur
> new_rlim
.rlim_max
)
1546 retval
= security_task_setrlimit(resource
, &new_rlim
);
1550 if (resource
== RLIMIT_CPU
&& new_rlim
.rlim_cur
== 0) {
1552 * The caller is asking for an immediate RLIMIT_CPU
1553 * expiry. But we use the zero value to mean "it was
1554 * never set". So let's cheat and make it one second
1557 new_rlim
.rlim_cur
= 1;
1560 task_lock(current
->group_leader
);
1561 *old_rlim
= new_rlim
;
1562 task_unlock(current
->group_leader
);
1564 if (resource
!= RLIMIT_CPU
)
1568 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1569 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1570 * very long-standing error, and fixing it now risks breakage of
1571 * applications, so we live with it
1573 if (new_rlim
.rlim_cur
== RLIM_INFINITY
)
1576 update_rlimit_cpu(new_rlim
.rlim_cur
);
1582 * It would make sense to put struct rusage in the task_struct,
1583 * except that would make the task_struct be *really big*. After
1584 * task_struct gets moved into malloc'ed memory, it would
1585 * make sense to do this. It will make moving the rest of the information
1586 * a lot simpler! (Which we're not doing right now because we're not
1587 * measuring them yet).
1589 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1590 * races with threads incrementing their own counters. But since word
1591 * reads are atomic, we either get new values or old values and we don't
1592 * care which for the sums. We always take the siglock to protect reading
1593 * the c* fields from p->signal from races with exit.c updating those
1594 * fields when reaping, so a sample either gets all the additions of a
1595 * given child after it's reaped, or none so this sample is before reaping.
1598 * We need to take the siglock for CHILDEREN, SELF and BOTH
1599 * for the cases current multithreaded, non-current single threaded
1600 * non-current multithreaded. Thread traversal is now safe with
1602 * Strictly speaking, we donot need to take the siglock if we are current and
1603 * single threaded, as no one else can take our signal_struct away, no one
1604 * else can reap the children to update signal->c* counters, and no one else
1605 * can race with the signal-> fields. If we do not take any lock, the
1606 * signal-> fields could be read out of order while another thread was just
1607 * exiting. So we should place a read memory barrier when we avoid the lock.
1608 * On the writer side, write memory barrier is implied in __exit_signal
1609 * as __exit_signal releases the siglock spinlock after updating the signal->
1610 * fields. But we don't do this yet to keep things simple.
1614 static void accumulate_thread_rusage(struct task_struct
*t
, struct rusage
*r
)
1616 r
->ru_nvcsw
+= t
->nvcsw
;
1617 r
->ru_nivcsw
+= t
->nivcsw
;
1618 r
->ru_minflt
+= t
->min_flt
;
1619 r
->ru_majflt
+= t
->maj_flt
;
1620 r
->ru_inblock
+= task_io_get_inblock(t
);
1621 r
->ru_oublock
+= task_io_get_oublock(t
);
1624 static void k_getrusage(struct task_struct
*p
, int who
, struct rusage
*r
)
1626 struct task_struct
*t
;
1627 unsigned long flags
;
1628 cputime_t utime
, stime
;
1629 struct task_cputime cputime
;
1631 memset((char *) r
, 0, sizeof *r
);
1632 utime
= stime
= cputime_zero
;
1634 if (who
== RUSAGE_THREAD
) {
1635 utime
= task_utime(current
);
1636 stime
= task_stime(current
);
1637 accumulate_thread_rusage(p
, r
);
1641 if (!lock_task_sighand(p
, &flags
))
1646 case RUSAGE_CHILDREN
:
1647 utime
= p
->signal
->cutime
;
1648 stime
= p
->signal
->cstime
;
1649 r
->ru_nvcsw
= p
->signal
->cnvcsw
;
1650 r
->ru_nivcsw
= p
->signal
->cnivcsw
;
1651 r
->ru_minflt
= p
->signal
->cmin_flt
;
1652 r
->ru_majflt
= p
->signal
->cmaj_flt
;
1653 r
->ru_inblock
= p
->signal
->cinblock
;
1654 r
->ru_oublock
= p
->signal
->coublock
;
1656 if (who
== RUSAGE_CHILDREN
)
1660 thread_group_cputime(p
, &cputime
);
1661 utime
= cputime_add(utime
, cputime
.utime
);
1662 stime
= cputime_add(stime
, cputime
.stime
);
1663 r
->ru_nvcsw
+= p
->signal
->nvcsw
;
1664 r
->ru_nivcsw
+= p
->signal
->nivcsw
;
1665 r
->ru_minflt
+= p
->signal
->min_flt
;
1666 r
->ru_majflt
+= p
->signal
->maj_flt
;
1667 r
->ru_inblock
+= p
->signal
->inblock
;
1668 r
->ru_oublock
+= p
->signal
->oublock
;
1671 accumulate_thread_rusage(t
, r
);
1679 unlock_task_sighand(p
, &flags
);
1682 cputime_to_timeval(utime
, &r
->ru_utime
);
1683 cputime_to_timeval(stime
, &r
->ru_stime
);
1686 int getrusage(struct task_struct
*p
, int who
, struct rusage __user
*ru
)
1689 k_getrusage(p
, who
, &r
);
1690 return copy_to_user(ru
, &r
, sizeof(r
)) ? -EFAULT
: 0;
1693 SYSCALL_DEFINE2(getrusage
, int, who
, struct rusage __user
*, ru
)
1695 if (who
!= RUSAGE_SELF
&& who
!= RUSAGE_CHILDREN
&&
1696 who
!= RUSAGE_THREAD
)
1698 return getrusage(current
, who
, ru
);
1701 SYSCALL_DEFINE1(umask
, int, mask
)
1703 mask
= xchg(¤t
->fs
->umask
, mask
& S_IRWXUGO
);
1707 SYSCALL_DEFINE5(prctl
, int, option
, unsigned long, arg2
, unsigned long, arg3
,
1708 unsigned long, arg4
, unsigned long, arg5
)
1710 struct task_struct
*me
= current
;
1711 unsigned char comm
[sizeof(me
->comm
)];
1714 error
= security_task_prctl(option
, arg2
, arg3
, arg4
, arg5
);
1715 if (error
!= -ENOSYS
)
1720 case PR_SET_PDEATHSIG
:
1721 if (!valid_signal(arg2
)) {
1725 me
->pdeath_signal
= arg2
;
1728 case PR_GET_PDEATHSIG
:
1729 error
= put_user(me
->pdeath_signal
, (int __user
*)arg2
);
1731 case PR_GET_DUMPABLE
:
1732 error
= get_dumpable(me
->mm
);
1734 case PR_SET_DUMPABLE
:
1735 if (arg2
< 0 || arg2
> 1) {
1739 set_dumpable(me
->mm
, arg2
);
1743 case PR_SET_UNALIGN
:
1744 error
= SET_UNALIGN_CTL(me
, arg2
);
1746 case PR_GET_UNALIGN
:
1747 error
= GET_UNALIGN_CTL(me
, arg2
);
1750 error
= SET_FPEMU_CTL(me
, arg2
);
1753 error
= GET_FPEMU_CTL(me
, arg2
);
1756 error
= SET_FPEXC_CTL(me
, arg2
);
1759 error
= GET_FPEXC_CTL(me
, arg2
);
1762 error
= PR_TIMING_STATISTICAL
;
1765 if (arg2
!= PR_TIMING_STATISTICAL
)
1772 comm
[sizeof(me
->comm
)-1] = 0;
1773 if (strncpy_from_user(comm
, (char __user
*)arg2
,
1774 sizeof(me
->comm
) - 1) < 0)
1776 set_task_comm(me
, comm
);
1779 get_task_comm(comm
, me
);
1780 if (copy_to_user((char __user
*)arg2
, comm
,
1785 error
= GET_ENDIAN(me
, arg2
);
1788 error
= SET_ENDIAN(me
, arg2
);
1791 case PR_GET_SECCOMP
:
1792 error
= prctl_get_seccomp();
1794 case PR_SET_SECCOMP
:
1795 error
= prctl_set_seccomp(arg2
);
1798 error
= GET_TSC_CTL(arg2
);
1801 error
= SET_TSC_CTL(arg2
);
1803 case PR_TASK_PERF_COUNTERS_DISABLE
:
1804 error
= perf_counter_task_disable();
1806 case PR_TASK_PERF_COUNTERS_ENABLE
:
1807 error
= perf_counter_task_enable();
1809 case PR_GET_TIMERSLACK
:
1810 error
= current
->timer_slack_ns
;
1812 case PR_SET_TIMERSLACK
:
1814 current
->timer_slack_ns
=
1815 current
->default_timer_slack_ns
;
1817 current
->timer_slack_ns
= arg2
;
1827 SYSCALL_DEFINE3(getcpu
, unsigned __user
*, cpup
, unsigned __user
*, nodep
,
1828 struct getcpu_cache __user
*, unused
)
1831 int cpu
= raw_smp_processor_id();
1833 err
|= put_user(cpu
, cpup
);
1835 err
|= put_user(cpu_to_node(cpu
), nodep
);
1836 return err
? -EFAULT
: 0;
1839 char poweroff_cmd
[POWEROFF_CMD_PATH_LEN
] = "/sbin/poweroff";
1841 static void argv_cleanup(char **argv
, char **envp
)
1847 * orderly_poweroff - Trigger an orderly system poweroff
1848 * @force: force poweroff if command execution fails
1850 * This may be called from any context to trigger a system shutdown.
1851 * If the orderly shutdown fails, it will force an immediate shutdown.
1853 int orderly_poweroff(bool force
)
1856 char **argv
= argv_split(GFP_ATOMIC
, poweroff_cmd
, &argc
);
1857 static char *envp
[] = {
1859 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1863 struct subprocess_info
*info
;
1866 printk(KERN_WARNING
"%s failed to allocate memory for \"%s\"\n",
1867 __func__
, poweroff_cmd
);
1871 info
= call_usermodehelper_setup(argv
[0], argv
, envp
, GFP_ATOMIC
);
1877 call_usermodehelper_setcleanup(info
, argv_cleanup
);
1879 ret
= call_usermodehelper_exec(info
, UMH_NO_WAIT
);
1883 printk(KERN_WARNING
"Failed to start orderly shutdown: "
1884 "forcing the issue\n");
1886 /* I guess this should try to kick off some daemon to
1887 sync and poweroff asap. Or not even bother syncing
1888 if we're doing an emergency shutdown? */
1895 EXPORT_SYMBOL_GPL(orderly_poweroff
);