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/notifier.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.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/gfp.h>
40 #include <linux/syscore_ops.h>
41 #include <linux/version.h>
42 #include <linux/ctype.h>
44 #include <linux/compat.h>
45 #include <linux/syscalls.h>
46 #include <linux/kprobes.h>
47 #include <linux/user_namespace.h>
49 #include <linux/kmsg_dump.h>
50 /* Move somewhere else to avoid recompiling? */
51 #include <generated/utsrelease.h>
53 #include <asm/uaccess.h>
55 #include <asm/unistd.h>
61 #define restart_dbg(format, arg...) \
62 printk("RESTART_DEBUG : " format "\n" , ## arg)
64 #define restart_dbg(format, arg...) do {} while (0)
69 #ifndef SET_UNALIGN_CTL
70 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
72 #ifndef GET_UNALIGN_CTL
73 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
76 # define SET_FPEMU_CTL(a,b) (-EINVAL)
79 # define GET_FPEMU_CTL(a,b) (-EINVAL)
82 # define SET_FPEXC_CTL(a,b) (-EINVAL)
85 # define GET_FPEXC_CTL(a,b) (-EINVAL)
88 # define GET_ENDIAN(a,b) (-EINVAL)
91 # define SET_ENDIAN(a,b) (-EINVAL)
94 # define GET_TSC_CTL(a) (-EINVAL)
97 # define SET_TSC_CTL(a) (-EINVAL)
101 * this is where the system-wide overflow UID and GID are defined, for
102 * architectures that now have 32-bit UID/GID but didn't in the past
105 int overflowuid = DEFAULT_OVERFLOWUID;
106 int overflowgid = DEFAULT_OVERFLOWGID;
109 EXPORT_SYMBOL(overflowuid);
110 EXPORT_SYMBOL(overflowgid);
114 * the same as above, but for filesystems which can only store a 16-bit
115 * UID and GID. as such, this is needed on all architectures
118 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
119 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
121 EXPORT_SYMBOL(fs_overflowuid);
122 EXPORT_SYMBOL(fs_overflowgid);
125 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
130 EXPORT_SYMBOL(cad_pid);
133 * If set, this is used for preparing the system to power off.
136 void (*pm_power_off_prepare)(void);
139 * Returns true if current's euid is same as p's uid or euid,
140 * or has CAP_SYS_NICE to p's user_ns.
142 * Called with rcu_read_lock, creds are safe
144 static bool set_one_prio_perm(struct task_struct *p)
146 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
148 if (pcred->user->user_ns == cred->user->user_ns &&
149 (pcred->uid == cred->euid ||
150 pcred->euid == cred->euid))
152 if (ns_capable(pcred->user->user_ns, CAP_SYS_NICE))
158 * set the priority of a task
159 * - the caller must hold the RCU read lock
161 static int set_one_prio(struct task_struct *p, int niceval, int error)
165 if (!set_one_prio_perm(p)) {
169 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
173 no_nice = security_task_setnice(p, niceval);
180 set_user_nice(p, niceval);
185 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
187 struct task_struct *g, *p;
188 struct user_struct *user;
189 const struct cred *cred = current_cred();
193 if (which > PRIO_USER || which < PRIO_PROCESS)
196 /* normalize: avoid signed division (rounding problems) */
204 read_lock(&tasklist_lock);
208 p = find_task_by_vpid(who);
212 error = set_one_prio(p, niceval, error);
216 pgrp = find_vpid(who);
218 pgrp = task_pgrp(current);
219 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
220 error = set_one_prio(p, niceval, error);
221 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
224 user = (struct user_struct *) cred->user;
227 else if ((who != cred->uid) &&
228 !(user = find_user(who)))
229 goto out_unlock; /* No processes for this user */
231 do_each_thread(g, p) {
232 if (__task_cred(p)->uid == who)
233 error = set_one_prio(p, niceval, error);
234 } while_each_thread(g, p);
235 if (who != cred->uid)
236 free_uid(user); /* For find_user() */
240 read_unlock(&tasklist_lock);
247 * Ugh. To avoid negative return values, "getpriority()" will
248 * not return the normal nice-value, but a negated value that
249 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
250 * to stay compatible.
252 SYSCALL_DEFINE2(getpriority, int, which, int, who)
254 struct task_struct *g, *p;
255 struct user_struct *user;
256 const struct cred *cred = current_cred();
257 long niceval, retval = -ESRCH;
260 if (which > PRIO_USER || which < PRIO_PROCESS)
264 read_lock(&tasklist_lock);
268 p = find_task_by_vpid(who);
272 niceval = 20 - task_nice(p);
273 if (niceval > retval)
279 pgrp = find_vpid(who);
281 pgrp = task_pgrp(current);
282 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
283 niceval = 20 - task_nice(p);
284 if (niceval > retval)
286 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
289 user = (struct user_struct *) cred->user;
292 else if ((who != cred->uid) &&
293 !(user = find_user(who)))
294 goto out_unlock; /* No processes for this user */
296 do_each_thread(g, p) {
297 if (__task_cred(p)->uid == who) {
298 niceval = 20 - task_nice(p);
299 if (niceval > retval)
302 } while_each_thread(g, p);
303 if (who != cred->uid)
304 free_uid(user); /* for find_user() */
308 read_unlock(&tasklist_lock);
315 * emergency_restart - reboot the system
317 * Without shutting down any hardware or taking any locks
318 * reboot the system. This is called when we know we are in
319 * trouble so this is our best effort to reboot. This is
320 * safe to call in interrupt context.
322 void emergency_restart(void)
324 kmsg_dump(KMSG_DUMP_EMERG);
325 machine_emergency_restart();
327 EXPORT_SYMBOL_GPL(emergency_restart);
329 void kernel_restart_prepare(char *cmd)
331 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
332 system_state = SYSTEM_RESTART;
333 usermodehelper_disable();
339 * kernel_restart - reboot the system
340 * @cmd: pointer to buffer containing command to execute for restart
343 * Shutdown everything and perform a clean reboot.
344 * This is not safe to call in interrupt context.
346 void kernel_restart(char *cmd)
351 restart_dbg("%s->%d->cmd=%s",__FUNCTION__,__LINE__,cmd);
353 kernel_restart_prepare(cmd);
355 printk(KERN_EMERG "Restarting system.\n");
357 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
358 kmsg_dump(KMSG_DUMP_RESTART);
359 machine_restart(cmd);
361 EXPORT_SYMBOL_GPL(kernel_restart);
363 static void kernel_shutdown_prepare(enum system_states state)
365 blocking_notifier_call_chain(&reboot_notifier_list,
366 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
367 system_state = state;
368 usermodehelper_disable();
372 * kernel_halt - halt the system
374 * Shutdown everything and perform a clean system halt.
376 void kernel_halt(void)
378 kernel_shutdown_prepare(SYSTEM_HALT);
380 printk(KERN_EMERG "System halted.\n");
381 kmsg_dump(KMSG_DUMP_HALT);
385 EXPORT_SYMBOL_GPL(kernel_halt);
388 * kernel_power_off - power_off the system
390 * Shutdown everything and perform a clean system power_off.
392 void kernel_power_off(void)
394 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
395 if (pm_power_off_prepare)
396 pm_power_off_prepare();
397 disable_nonboot_cpus();
399 printk(KERN_EMERG "Power down.\n");
400 kmsg_dump(KMSG_DUMP_POWEROFF);
403 EXPORT_SYMBOL_GPL(kernel_power_off);
405 static DEFINE_MUTEX(reboot_mutex);
408 * Reboot system call: for obvious reasons only root may call it,
409 * and even root needs to set up some magic numbers in the registers
410 * so that some mistake won't make this reboot the whole machine.
411 * You can also set the meaning of the ctrl-alt-del-key here.
413 * reboot doesn't sync: do that yourself before calling this.
415 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
421 /* We only trust the superuser with rebooting the system. */
422 if (!capable(CAP_SYS_BOOT))
425 /* For safety, we require "magic" arguments. */
426 if (magic1 != LINUX_REBOOT_MAGIC1 ||
427 (magic2 != LINUX_REBOOT_MAGIC2 &&
428 magic2 != LINUX_REBOOT_MAGIC2A &&
429 magic2 != LINUX_REBOOT_MAGIC2B &&
430 magic2 != LINUX_REBOOT_MAGIC2C))
433 /* Instead of trying to make the power_off code look like
434 * halt when pm_power_off is not set do it the easy way.
436 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
437 cmd = LINUX_REBOOT_CMD_HALT;
439 mutex_lock(&reboot_mutex);
441 case LINUX_REBOOT_CMD_RESTART:
445 restart_dbg("%s->%d->cmd=%x",__FUNCTION__,__LINE__,cmd);
447 kernel_restart(NULL);
450 case LINUX_REBOOT_CMD_CAD_ON:
454 case LINUX_REBOOT_CMD_CAD_OFF:
458 case LINUX_REBOOT_CMD_HALT:
461 panic("cannot halt");
463 case LINUX_REBOOT_CMD_POWER_OFF:
467 restart_dbg("%s->%d->cmd=%x",__FUNCTION__,__LINE__,cmd);
473 case LINUX_REBOOT_CMD_RESTART2:
474 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
478 buffer[sizeof(buffer) - 1] = '\0';
482 restart_dbg("%s->%d->cmd=%x args=%s",__FUNCTION__,__LINE__,cmd,buffer);
484 kernel_restart(buffer);
488 case LINUX_REBOOT_CMD_KEXEC:
489 ret = kernel_kexec();
493 #ifdef CONFIG_HIBERNATION
494 case LINUX_REBOOT_CMD_SW_SUSPEND:
503 mutex_unlock(&reboot_mutex);
507 static void deferred_cad(struct work_struct *dummy)
509 kernel_restart(NULL);
513 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
514 * As it's called within an interrupt, it may NOT sync: the only choice
515 * is whether to reboot at once, or just ignore the ctrl-alt-del.
517 void ctrl_alt_del(void)
519 static DECLARE_WORK(cad_work, deferred_cad);
522 schedule_work(&cad_work);
524 kill_cad_pid(SIGINT, 1);
528 * Unprivileged users may change the real gid to the effective gid
529 * or vice versa. (BSD-style)
531 * If you set the real gid at all, or set the effective gid to a value not
532 * equal to the real gid, then the saved gid is set to the new effective gid.
534 * This makes it possible for a setgid program to completely drop its
535 * privileges, which is often a useful assertion to make when you are doing
536 * a security audit over a program.
538 * The general idea is that a program which uses just setregid() will be
539 * 100% compatible with BSD. A program which uses just setgid() will be
540 * 100% compatible with POSIX with saved IDs.
542 * SMP: There are not races, the GIDs are checked only by filesystem
543 * operations (as far as semantic preservation is concerned).
545 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
547 const struct cred *old;
551 new = prepare_creds();
554 old = current_cred();
557 if (rgid != (gid_t) -1) {
558 if (old->gid == rgid ||
560 nsown_capable(CAP_SETGID))
565 if (egid != (gid_t) -1) {
566 if (old->gid == egid ||
569 nsown_capable(CAP_SETGID))
575 if (rgid != (gid_t) -1 ||
576 (egid != (gid_t) -1 && egid != old->gid))
577 new->sgid = new->egid;
578 new->fsgid = new->egid;
580 return commit_creds(new);
588 * setgid() is implemented like SysV w/ SAVED_IDS
590 * SMP: Same implicit races as above.
592 SYSCALL_DEFINE1(setgid, gid_t, gid)
594 const struct cred *old;
598 new = prepare_creds();
601 old = current_cred();
604 if (nsown_capable(CAP_SETGID))
605 new->gid = new->egid = new->sgid = new->fsgid = gid;
606 else if (gid == old->gid || gid == old->sgid)
607 new->egid = new->fsgid = gid;
611 return commit_creds(new);
619 * change the user struct in a credentials set to match the new UID
621 static int set_user(struct cred *new)
623 struct user_struct *new_user;
625 new_user = alloc_uid(current_user_ns(), new->uid);
629 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
630 new_user != INIT_USER) {
636 new->user = new_user;
641 * Unprivileged users may change the real uid to the effective uid
642 * or vice versa. (BSD-style)
644 * If you set the real uid at all, or set the effective uid to a value not
645 * equal to the real uid, then the saved uid is set to the new effective uid.
647 * This makes it possible for a setuid program to completely drop its
648 * privileges, which is often a useful assertion to make when you are doing
649 * a security audit over a program.
651 * The general idea is that a program which uses just setreuid() will be
652 * 100% compatible with BSD. A program which uses just setuid() will be
653 * 100% compatible with POSIX with saved IDs.
655 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
657 const struct cred *old;
661 new = prepare_creds();
664 old = current_cred();
667 if (ruid != (uid_t) -1) {
669 if (old->uid != ruid &&
671 !nsown_capable(CAP_SETUID))
675 if (euid != (uid_t) -1) {
677 if (old->uid != euid &&
680 !nsown_capable(CAP_SETUID))
684 if (new->uid != old->uid) {
685 retval = set_user(new);
689 if (ruid != (uid_t) -1 ||
690 (euid != (uid_t) -1 && euid != old->uid))
691 new->suid = new->euid;
692 new->fsuid = new->euid;
694 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
698 return commit_creds(new);
706 * setuid() is implemented like SysV with SAVED_IDS
708 * Note that SAVED_ID's is deficient in that a setuid root program
709 * like sendmail, for example, cannot set its uid to be a normal
710 * user and then switch back, because if you're root, setuid() sets
711 * the saved uid too. If you don't like this, blame the bright people
712 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
713 * will allow a root program to temporarily drop privileges and be able to
714 * regain them by swapping the real and effective uid.
716 SYSCALL_DEFINE1(setuid, uid_t, uid)
718 const struct cred *old;
722 new = prepare_creds();
725 old = current_cred();
728 if (nsown_capable(CAP_SETUID)) {
729 new->suid = new->uid = uid;
730 if (uid != old->uid) {
731 retval = set_user(new);
735 } else if (uid != old->uid && uid != new->suid) {
739 new->fsuid = new->euid = uid;
741 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
745 return commit_creds(new);
754 * This function implements a generic ability to update ruid, euid,
755 * and suid. This allows you to implement the 4.4 compatible seteuid().
757 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
759 const struct cred *old;
763 new = prepare_creds();
767 old = current_cred();
770 if (!nsown_capable(CAP_SETUID)) {
771 if (ruid != (uid_t) -1 && ruid != old->uid &&
772 ruid != old->euid && ruid != old->suid)
774 if (euid != (uid_t) -1 && euid != old->uid &&
775 euid != old->euid && euid != old->suid)
777 if (suid != (uid_t) -1 && suid != old->uid &&
778 suid != old->euid && suid != old->suid)
782 if (ruid != (uid_t) -1) {
784 if (ruid != old->uid) {
785 retval = set_user(new);
790 if (euid != (uid_t) -1)
792 if (suid != (uid_t) -1)
794 new->fsuid = new->euid;
796 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
800 return commit_creds(new);
807 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
809 const struct cred *cred = current_cred();
812 if (!(retval = put_user(cred->uid, ruid)) &&
813 !(retval = put_user(cred->euid, euid)))
814 retval = put_user(cred->suid, suid);
820 * Same as above, but for rgid, egid, sgid.
822 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
824 const struct cred *old;
828 new = prepare_creds();
831 old = current_cred();
834 if (!nsown_capable(CAP_SETGID)) {
835 if (rgid != (gid_t) -1 && rgid != old->gid &&
836 rgid != old->egid && rgid != old->sgid)
838 if (egid != (gid_t) -1 && egid != old->gid &&
839 egid != old->egid && egid != old->sgid)
841 if (sgid != (gid_t) -1 && sgid != old->gid &&
842 sgid != old->egid && sgid != old->sgid)
846 if (rgid != (gid_t) -1)
848 if (egid != (gid_t) -1)
850 if (sgid != (gid_t) -1)
852 new->fsgid = new->egid;
854 return commit_creds(new);
861 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
863 const struct cred *cred = current_cred();
866 if (!(retval = put_user(cred->gid, rgid)) &&
867 !(retval = put_user(cred->egid, egid)))
868 retval = put_user(cred->sgid, sgid);
875 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
876 * is used for "access()" and for the NFS daemon (letting nfsd stay at
877 * whatever uid it wants to). It normally shadows "euid", except when
878 * explicitly set by setfsuid() or for access..
880 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
882 const struct cred *old;
886 new = prepare_creds();
888 return current_fsuid();
889 old = current_cred();
890 old_fsuid = old->fsuid;
892 if (uid == old->uid || uid == old->euid ||
893 uid == old->suid || uid == old->fsuid ||
894 nsown_capable(CAP_SETUID)) {
895 if (uid != old_fsuid) {
897 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
911 * Samma på svenska..
913 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
915 const struct cred *old;
919 new = prepare_creds();
921 return current_fsgid();
922 old = current_cred();
923 old_fsgid = old->fsgid;
925 if (gid == old->gid || gid == old->egid ||
926 gid == old->sgid || gid == old->fsgid ||
927 nsown_capable(CAP_SETGID)) {
928 if (gid != old_fsgid) {
942 void do_sys_times(struct tms *tms)
944 cputime_t tgutime, tgstime, cutime, cstime;
946 spin_lock_irq(¤t->sighand->siglock);
947 thread_group_times(current, &tgutime, &tgstime);
948 cutime = current->signal->cutime;
949 cstime = current->signal->cstime;
950 spin_unlock_irq(¤t->sighand->siglock);
951 tms->tms_utime = cputime_to_clock_t(tgutime);
952 tms->tms_stime = cputime_to_clock_t(tgstime);
953 tms->tms_cutime = cputime_to_clock_t(cutime);
954 tms->tms_cstime = cputime_to_clock_t(cstime);
957 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
963 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
966 force_successful_syscall_return();
967 return (long) jiffies_64_to_clock_t(get_jiffies_64());
971 * This needs some heavy checking ...
972 * I just haven't the stomach for it. I also don't fully
973 * understand sessions/pgrp etc. Let somebody who does explain it.
975 * OK, I think I have the protection semantics right.... this is really
976 * only important on a multi-user system anyway, to make sure one user
977 * can't send a signal to a process owned by another. -TYT, 12/12/91
979 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
982 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
984 struct task_struct *p;
985 struct task_struct *group_leader = current->group_leader;
990 pid = task_pid_vnr(group_leader);
997 /* From this point forward we keep holding onto the tasklist lock
998 * so that our parent does not change from under us. -DaveM
1000 write_lock_irq(&tasklist_lock);
1003 p = find_task_by_vpid(pid);
1008 if (!thread_group_leader(p))
1011 if (same_thread_group(p->real_parent, group_leader)) {
1013 if (task_session(p) != task_session(group_leader))
1020 if (p != group_leader)
1025 if (p->signal->leader)
1030 struct task_struct *g;
1032 pgrp = find_vpid(pgid);
1033 g = pid_task(pgrp, PIDTYPE_PGID);
1034 if (!g || task_session(g) != task_session(group_leader))
1038 err = security_task_setpgid(p, pgid);
1042 if (task_pgrp(p) != pgrp)
1043 change_pid(p, PIDTYPE_PGID, pgrp);
1047 /* All paths lead to here, thus we are safe. -DaveM */
1048 write_unlock_irq(&tasklist_lock);
1053 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1055 struct task_struct *p;
1061 grp = task_pgrp(current);
1064 p = find_task_by_vpid(pid);
1071 retval = security_task_getpgid(p);
1075 retval = pid_vnr(grp);
1081 #ifdef __ARCH_WANT_SYS_GETPGRP
1083 SYSCALL_DEFINE0(getpgrp)
1085 return sys_getpgid(0);
1090 SYSCALL_DEFINE1(getsid, pid_t, pid)
1092 struct task_struct *p;
1098 sid = task_session(current);
1101 p = find_task_by_vpid(pid);
1104 sid = task_session(p);
1108 retval = security_task_getsid(p);
1112 retval = pid_vnr(sid);
1118 SYSCALL_DEFINE0(setsid)
1120 struct task_struct *group_leader = current->group_leader;
1121 struct pid *sid = task_pid(group_leader);
1122 pid_t session = pid_vnr(sid);
1125 write_lock_irq(&tasklist_lock);
1126 /* Fail if I am already a session leader */
1127 if (group_leader->signal->leader)
1130 /* Fail if a process group id already exists that equals the
1131 * proposed session id.
1133 if (pid_task(sid, PIDTYPE_PGID))
1136 group_leader->signal->leader = 1;
1137 __set_special_pids(sid);
1139 proc_clear_tty(group_leader);
1143 write_unlock_irq(&tasklist_lock);
1145 proc_sid_connector(group_leader);
1146 sched_autogroup_create_attach(group_leader);
1151 DECLARE_RWSEM(uts_sem);
1153 #ifdef COMPAT_UTS_MACHINE
1154 #define override_architecture(name) \
1155 (personality(current->personality) == PER_LINUX32 && \
1156 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1157 sizeof(COMPAT_UTS_MACHINE)))
1159 #define override_architecture(name) 0
1163 * Work around broken programs that cannot handle "Linux 3.0".
1164 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1166 static int override_release(char __user *release, int len)
1171 if (current->personality & UNAME26) {
1172 char *rest = UTS_RELEASE;
1177 if (*rest == '.' && ++ndots >= 3)
1179 if (!isdigit(*rest) && *rest != '.')
1183 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 40;
1184 snprintf(buf, len, "2.6.%u%s", v, rest);
1185 ret = copy_to_user(release, buf, len);
1190 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1194 down_read(&uts_sem);
1195 if (copy_to_user(name, utsname(), sizeof *name))
1199 if (!errno && override_release(name->release, sizeof(name->release)))
1201 if (!errno && override_architecture(name))
1206 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1210 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1217 down_read(&uts_sem);
1218 if (copy_to_user(name, utsname(), sizeof(*name)))
1222 if (!error && override_release(name->release, sizeof(name->release)))
1224 if (!error && override_architecture(name))
1229 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1235 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1238 down_read(&uts_sem);
1239 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1241 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1242 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1244 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1245 error |= __copy_to_user(&name->release, &utsname()->release,
1247 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1248 error |= __copy_to_user(&name->version, &utsname()->version,
1250 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1251 error |= __copy_to_user(&name->machine, &utsname()->machine,
1253 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1256 if (!error && override_architecture(name))
1258 if (!error && override_release(name->release, sizeof(name->release)))
1260 return error ? -EFAULT : 0;
1264 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1267 char tmp[__NEW_UTS_LEN];
1269 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1272 if (len < 0 || len > __NEW_UTS_LEN)
1274 down_write(&uts_sem);
1276 if (!copy_from_user(tmp, name, len)) {
1277 struct new_utsname *u = utsname();
1279 memcpy(u->nodename, tmp, len);
1280 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1287 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1289 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1292 struct new_utsname *u;
1296 down_read(&uts_sem);
1298 i = 1 + strlen(u->nodename);
1302 if (copy_to_user(name, u->nodename, i))
1311 * Only setdomainname; getdomainname can be implemented by calling
1314 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1317 char tmp[__NEW_UTS_LEN];
1319 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1321 if (len < 0 || len > __NEW_UTS_LEN)
1324 down_write(&uts_sem);
1326 if (!copy_from_user(tmp, name, len)) {
1327 struct new_utsname *u = utsname();
1329 memcpy(u->domainname, tmp, len);
1330 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1337 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1339 struct rlimit value;
1342 ret = do_prlimit(current, resource, NULL, &value);
1344 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1349 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1352 * Back compatibility for getrlimit. Needed for some apps.
1355 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1356 struct rlimit __user *, rlim)
1359 if (resource >= RLIM_NLIMITS)
1362 task_lock(current->group_leader);
1363 x = current->signal->rlim[resource];
1364 task_unlock(current->group_leader);
1365 if (x.rlim_cur > 0x7FFFFFFF)
1366 x.rlim_cur = 0x7FFFFFFF;
1367 if (x.rlim_max > 0x7FFFFFFF)
1368 x.rlim_max = 0x7FFFFFFF;
1369 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1374 static inline bool rlim64_is_infinity(__u64 rlim64)
1376 #if BITS_PER_LONG < 64
1377 return rlim64 >= ULONG_MAX;
1379 return rlim64 == RLIM64_INFINITY;
1383 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1385 if (rlim->rlim_cur == RLIM_INFINITY)
1386 rlim64->rlim_cur = RLIM64_INFINITY;
1388 rlim64->rlim_cur = rlim->rlim_cur;
1389 if (rlim->rlim_max == RLIM_INFINITY)
1390 rlim64->rlim_max = RLIM64_INFINITY;
1392 rlim64->rlim_max = rlim->rlim_max;
1395 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1397 if (rlim64_is_infinity(rlim64->rlim_cur))
1398 rlim->rlim_cur = RLIM_INFINITY;
1400 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1401 if (rlim64_is_infinity(rlim64->rlim_max))
1402 rlim->rlim_max = RLIM_INFINITY;
1404 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1407 /* make sure you are allowed to change @tsk limits before calling this */
1408 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1409 struct rlimit *new_rlim, struct rlimit *old_rlim)
1411 struct rlimit *rlim;
1414 if (resource >= RLIM_NLIMITS)
1417 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1419 if (resource == RLIMIT_NOFILE &&
1420 new_rlim->rlim_max > sysctl_nr_open)
1424 /* protect tsk->signal and tsk->sighand from disappearing */
1425 read_lock(&tasklist_lock);
1426 if (!tsk->sighand) {
1431 rlim = tsk->signal->rlim + resource;
1432 task_lock(tsk->group_leader);
1434 /* Keep the capable check against init_user_ns until
1435 cgroups can contain all limits */
1436 if (new_rlim->rlim_max > rlim->rlim_max &&
1437 !capable(CAP_SYS_RESOURCE))
1440 retval = security_task_setrlimit(tsk->group_leader,
1441 resource, new_rlim);
1442 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1444 * The caller is asking for an immediate RLIMIT_CPU
1445 * expiry. But we use the zero value to mean "it was
1446 * never set". So let's cheat and make it one second
1449 new_rlim->rlim_cur = 1;
1458 task_unlock(tsk->group_leader);
1461 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1462 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1463 * very long-standing error, and fixing it now risks breakage of
1464 * applications, so we live with it
1466 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1467 new_rlim->rlim_cur != RLIM_INFINITY)
1468 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1470 read_unlock(&tasklist_lock);
1474 /* rcu lock must be held */
1475 static int check_prlimit_permission(struct task_struct *task)
1477 const struct cred *cred = current_cred(), *tcred;
1479 if (current == task)
1482 tcred = __task_cred(task);
1483 if (cred->user->user_ns == tcred->user->user_ns &&
1484 (cred->uid == tcred->euid &&
1485 cred->uid == tcred->suid &&
1486 cred->uid == tcred->uid &&
1487 cred->gid == tcred->egid &&
1488 cred->gid == tcred->sgid &&
1489 cred->gid == tcred->gid))
1491 if (ns_capable(tcred->user->user_ns, CAP_SYS_RESOURCE))
1497 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1498 const struct rlimit64 __user *, new_rlim,
1499 struct rlimit64 __user *, old_rlim)
1501 struct rlimit64 old64, new64;
1502 struct rlimit old, new;
1503 struct task_struct *tsk;
1507 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1509 rlim64_to_rlim(&new64, &new);
1513 tsk = pid ? find_task_by_vpid(pid) : current;
1518 ret = check_prlimit_permission(tsk);
1523 get_task_struct(tsk);
1526 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1527 old_rlim ? &old : NULL);
1529 if (!ret && old_rlim) {
1530 rlim_to_rlim64(&old, &old64);
1531 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1535 put_task_struct(tsk);
1539 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1541 struct rlimit new_rlim;
1543 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1545 return do_prlimit(current, resource, &new_rlim, NULL);
1549 * It would make sense to put struct rusage in the task_struct,
1550 * except that would make the task_struct be *really big*. After
1551 * task_struct gets moved into malloc'ed memory, it would
1552 * make sense to do this. It will make moving the rest of the information
1553 * a lot simpler! (Which we're not doing right now because we're not
1554 * measuring them yet).
1556 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1557 * races with threads incrementing their own counters. But since word
1558 * reads are atomic, we either get new values or old values and we don't
1559 * care which for the sums. We always take the siglock to protect reading
1560 * the c* fields from p->signal from races with exit.c updating those
1561 * fields when reaping, so a sample either gets all the additions of a
1562 * given child after it's reaped, or none so this sample is before reaping.
1565 * We need to take the siglock for CHILDEREN, SELF and BOTH
1566 * for the cases current multithreaded, non-current single threaded
1567 * non-current multithreaded. Thread traversal is now safe with
1569 * Strictly speaking, we donot need to take the siglock if we are current and
1570 * single threaded, as no one else can take our signal_struct away, no one
1571 * else can reap the children to update signal->c* counters, and no one else
1572 * can race with the signal-> fields. If we do not take any lock, the
1573 * signal-> fields could be read out of order while another thread was just
1574 * exiting. So we should place a read memory barrier when we avoid the lock.
1575 * On the writer side, write memory barrier is implied in __exit_signal
1576 * as __exit_signal releases the siglock spinlock after updating the signal->
1577 * fields. But we don't do this yet to keep things simple.
1581 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1583 r->ru_nvcsw += t->nvcsw;
1584 r->ru_nivcsw += t->nivcsw;
1585 r->ru_minflt += t->min_flt;
1586 r->ru_majflt += t->maj_flt;
1587 r->ru_inblock += task_io_get_inblock(t);
1588 r->ru_oublock += task_io_get_oublock(t);
1591 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1593 struct task_struct *t;
1594 unsigned long flags;
1595 cputime_t tgutime, tgstime, utime, stime;
1596 unsigned long maxrss = 0;
1598 memset((char *) r, 0, sizeof *r);
1599 utime = stime = cputime_zero;
1601 if (who == RUSAGE_THREAD) {
1602 task_times(current, &utime, &stime);
1603 accumulate_thread_rusage(p, r);
1604 maxrss = p->signal->maxrss;
1608 if (!lock_task_sighand(p, &flags))
1613 case RUSAGE_CHILDREN:
1614 utime = p->signal->cutime;
1615 stime = p->signal->cstime;
1616 r->ru_nvcsw = p->signal->cnvcsw;
1617 r->ru_nivcsw = p->signal->cnivcsw;
1618 r->ru_minflt = p->signal->cmin_flt;
1619 r->ru_majflt = p->signal->cmaj_flt;
1620 r->ru_inblock = p->signal->cinblock;
1621 r->ru_oublock = p->signal->coublock;
1622 maxrss = p->signal->cmaxrss;
1624 if (who == RUSAGE_CHILDREN)
1628 thread_group_times(p, &tgutime, &tgstime);
1629 utime = cputime_add(utime, tgutime);
1630 stime = cputime_add(stime, tgstime);
1631 r->ru_nvcsw += p->signal->nvcsw;
1632 r->ru_nivcsw += p->signal->nivcsw;
1633 r->ru_minflt += p->signal->min_flt;
1634 r->ru_majflt += p->signal->maj_flt;
1635 r->ru_inblock += p->signal->inblock;
1636 r->ru_oublock += p->signal->oublock;
1637 if (maxrss < p->signal->maxrss)
1638 maxrss = p->signal->maxrss;
1641 accumulate_thread_rusage(t, r);
1649 unlock_task_sighand(p, &flags);
1652 cputime_to_timeval(utime, &r->ru_utime);
1653 cputime_to_timeval(stime, &r->ru_stime);
1655 if (who != RUSAGE_CHILDREN) {
1656 struct mm_struct *mm = get_task_mm(p);
1658 setmax_mm_hiwater_rss(&maxrss, mm);
1662 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1665 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1668 k_getrusage(p, who, &r);
1669 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1672 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1674 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1675 who != RUSAGE_THREAD)
1677 return getrusage(current, who, ru);
1680 SYSCALL_DEFINE1(umask, int, mask)
1682 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1686 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1687 unsigned long, arg4, unsigned long, arg5)
1689 struct task_struct *me = current;
1690 unsigned char comm[sizeof(me->comm)];
1693 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1694 if (error != -ENOSYS)
1699 case PR_SET_PDEATHSIG:
1700 if (!valid_signal(arg2)) {
1704 me->pdeath_signal = arg2;
1707 case PR_GET_PDEATHSIG:
1708 error = put_user(me->pdeath_signal, (int __user *)arg2);
1710 case PR_GET_DUMPABLE:
1711 error = get_dumpable(me->mm);
1713 case PR_SET_DUMPABLE:
1714 if (arg2 < 0 || arg2 > 1) {
1718 set_dumpable(me->mm, arg2);
1722 case PR_SET_UNALIGN:
1723 error = SET_UNALIGN_CTL(me, arg2);
1725 case PR_GET_UNALIGN:
1726 error = GET_UNALIGN_CTL(me, arg2);
1729 error = SET_FPEMU_CTL(me, arg2);
1732 error = GET_FPEMU_CTL(me, arg2);
1735 error = SET_FPEXC_CTL(me, arg2);
1738 error = GET_FPEXC_CTL(me, arg2);
1741 error = PR_TIMING_STATISTICAL;
1744 if (arg2 != PR_TIMING_STATISTICAL)
1751 comm[sizeof(me->comm)-1] = 0;
1752 if (strncpy_from_user(comm, (char __user *)arg2,
1753 sizeof(me->comm) - 1) < 0)
1755 set_task_comm(me, comm);
1758 get_task_comm(comm, me);
1759 if (copy_to_user((char __user *)arg2, comm,
1764 error = GET_ENDIAN(me, arg2);
1767 error = SET_ENDIAN(me, arg2);
1770 case PR_GET_SECCOMP:
1771 error = prctl_get_seccomp();
1773 case PR_SET_SECCOMP:
1774 error = prctl_set_seccomp(arg2);
1777 error = GET_TSC_CTL(arg2);
1780 error = SET_TSC_CTL(arg2);
1782 case PR_TASK_PERF_EVENTS_DISABLE:
1783 error = perf_event_task_disable();
1785 case PR_TASK_PERF_EVENTS_ENABLE:
1786 error = perf_event_task_enable();
1788 case PR_GET_TIMERSLACK:
1789 error = current->timer_slack_ns;
1791 case PR_SET_TIMERSLACK:
1793 current->timer_slack_ns =
1794 current->default_timer_slack_ns;
1796 current->timer_slack_ns = arg2;
1803 case PR_MCE_KILL_CLEAR:
1806 current->flags &= ~PF_MCE_PROCESS;
1808 case PR_MCE_KILL_SET:
1809 current->flags |= PF_MCE_PROCESS;
1810 if (arg3 == PR_MCE_KILL_EARLY)
1811 current->flags |= PF_MCE_EARLY;
1812 else if (arg3 == PR_MCE_KILL_LATE)
1813 current->flags &= ~PF_MCE_EARLY;
1814 else if (arg3 == PR_MCE_KILL_DEFAULT)
1816 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
1825 case PR_MCE_KILL_GET:
1826 if (arg2 | arg3 | arg4 | arg5)
1828 if (current->flags & PF_MCE_PROCESS)
1829 error = (current->flags & PF_MCE_EARLY) ?
1830 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
1832 error = PR_MCE_KILL_DEFAULT;
1841 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1842 struct getcpu_cache __user *, unused)
1845 int cpu = raw_smp_processor_id();
1847 err |= put_user(cpu, cpup);
1849 err |= put_user(cpu_to_node(cpu), nodep);
1850 return err ? -EFAULT : 0;
1853 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1855 static void argv_cleanup(struct subprocess_info *info)
1857 argv_free(info->argv);
1861 * orderly_poweroff - Trigger an orderly system poweroff
1862 * @force: force poweroff if command execution fails
1864 * This may be called from any context to trigger a system shutdown.
1865 * If the orderly shutdown fails, it will force an immediate shutdown.
1867 int orderly_poweroff(bool force)
1870 char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1871 static char *envp[] = {
1873 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1877 struct subprocess_info *info;
1880 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1881 __func__, poweroff_cmd);
1885 info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1891 call_usermodehelper_setfns(info, NULL, argv_cleanup, NULL);
1893 ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1897 printk(KERN_WARNING "Failed to start orderly shutdown: "
1898 "forcing the issue\n");
1900 /* I guess this should try to kick off some daemon to
1901 sync and poweroff asap. Or not even bother syncing
1902 if we're doing an emergency shutdown? */
1909 EXPORT_SYMBOL_GPL(orderly_poweroff);