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>
38 #include <linux/fs_struct.h>
40 #include <linux/compat.h>
41 #include <linux/syscalls.h>
42 #include <linux/kprobes.h>
43 #include <linux/user_namespace.h>
45 #include <asm/uaccess.h>
47 #include <asm/unistd.h>
49 #ifndef SET_UNALIGN_CTL
50 # define SET_UNALIGN_CTL(a,b) (-EINVAL)
52 #ifndef GET_UNALIGN_CTL
53 # define GET_UNALIGN_CTL(a,b) (-EINVAL)
56 # define SET_FPEMU_CTL(a,b) (-EINVAL)
59 # define GET_FPEMU_CTL(a,b) (-EINVAL)
62 # define SET_FPEXC_CTL(a,b) (-EINVAL)
65 # define GET_FPEXC_CTL(a,b) (-EINVAL)
68 # define GET_ENDIAN(a,b) (-EINVAL)
71 # define SET_ENDIAN(a,b) (-EINVAL)
74 # define GET_TSC_CTL(a) (-EINVAL)
77 # define SET_TSC_CTL(a) (-EINVAL)
81 * this is where the system-wide overflow UID and GID are defined, for
82 * architectures that now have 32-bit UID/GID but didn't in the past
85 int overflowuid = DEFAULT_OVERFLOWUID;
86 int overflowgid = DEFAULT_OVERFLOWGID;
89 EXPORT_SYMBOL(overflowuid);
90 EXPORT_SYMBOL(overflowgid);
94 * the same as above, but for filesystems which can only store a 16-bit
95 * UID and GID. as such, this is needed on all architectures
98 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
99 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
101 EXPORT_SYMBOL(fs_overflowuid);
102 EXPORT_SYMBOL(fs_overflowgid);
105 * this indicates whether you can reboot with ctrl-alt-del: the default is yes
110 EXPORT_SYMBOL(cad_pid);
113 * If set, this is used for preparing the system to power off.
116 void (*pm_power_off_prepare)(void);
119 * set the priority of a task
120 * - the caller must hold the RCU read lock
122 static int set_one_prio(struct task_struct *p, int niceval, int error)
124 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
127 if (pcred->uid != cred->euid &&
128 pcred->euid != cred->euid && !capable(CAP_SYS_NICE)) {
132 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
136 no_nice = security_task_setnice(p, niceval);
143 set_user_nice(p, niceval);
148 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
150 struct task_struct *g, *p;
151 struct user_struct *user;
152 const struct cred *cred = current_cred();
156 if (which > PRIO_USER || which < PRIO_PROCESS)
159 /* normalize: avoid signed division (rounding problems) */
166 read_lock(&tasklist_lock);
170 p = find_task_by_vpid(who);
174 error = set_one_prio(p, niceval, error);
178 pgrp = find_vpid(who);
180 pgrp = task_pgrp(current);
181 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
182 error = set_one_prio(p, niceval, error);
183 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
186 user = (struct user_struct *) cred->user;
189 else if ((who != cred->uid) &&
190 !(user = find_user(who)))
191 goto out_unlock; /* No processes for this user */
194 if (__task_cred(p)->uid == who)
195 error = set_one_prio(p, niceval, error);
196 while_each_thread(g, p);
197 if (who != cred->uid)
198 free_uid(user); /* For find_user() */
202 read_unlock(&tasklist_lock);
208 * Ugh. To avoid negative return values, "getpriority()" will
209 * not return the normal nice-value, but a negated value that
210 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
211 * to stay compatible.
213 SYSCALL_DEFINE2(getpriority, int, which, int, who)
215 struct task_struct *g, *p;
216 struct user_struct *user;
217 const struct cred *cred = current_cred();
218 long niceval, retval = -ESRCH;
221 if (which > PRIO_USER || which < PRIO_PROCESS)
224 read_lock(&tasklist_lock);
228 p = find_task_by_vpid(who);
232 niceval = 20 - task_nice(p);
233 if (niceval > retval)
239 pgrp = find_vpid(who);
241 pgrp = task_pgrp(current);
242 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
243 niceval = 20 - task_nice(p);
244 if (niceval > retval)
246 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
249 user = (struct user_struct *) cred->user;
252 else if ((who != cred->uid) &&
253 !(user = find_user(who)))
254 goto out_unlock; /* No processes for this user */
257 if (__task_cred(p)->uid == who) {
258 niceval = 20 - task_nice(p);
259 if (niceval > retval)
262 while_each_thread(g, p);
263 if (who != cred->uid)
264 free_uid(user); /* for find_user() */
268 read_unlock(&tasklist_lock);
274 * emergency_restart - reboot the system
276 * Without shutting down any hardware or taking any locks
277 * reboot the system. This is called when we know we are in
278 * trouble so this is our best effort to reboot. This is
279 * safe to call in interrupt context.
281 void emergency_restart(void)
283 machine_emergency_restart();
285 EXPORT_SYMBOL_GPL(emergency_restart);
287 void kernel_restart_prepare(char *cmd)
289 blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
290 system_state = SYSTEM_RESTART;
296 * kernel_restart - reboot the system
297 * @cmd: pointer to buffer containing command to execute for restart
300 * Shutdown everything and perform a clean reboot.
301 * This is not safe to call in interrupt context.
303 void kernel_restart(char *cmd)
305 kernel_restart_prepare(cmd);
307 printk(KERN_EMERG "Restarting system.\n");
309 printk(KERN_EMERG "Restarting system with command '%s'.\n", cmd);
310 machine_restart(cmd);
312 EXPORT_SYMBOL_GPL(kernel_restart);
314 static void kernel_shutdown_prepare(enum system_states state)
316 blocking_notifier_call_chain(&reboot_notifier_list,
317 (state == SYSTEM_HALT)?SYS_HALT:SYS_POWER_OFF, NULL);
318 system_state = state;
322 * kernel_halt - halt the system
324 * Shutdown everything and perform a clean system halt.
326 void kernel_halt(void)
328 kernel_shutdown_prepare(SYSTEM_HALT);
330 printk(KERN_EMERG "System halted.\n");
334 EXPORT_SYMBOL_GPL(kernel_halt);
337 * kernel_power_off - power_off the system
339 * Shutdown everything and perform a clean system power_off.
341 void kernel_power_off(void)
343 kernel_shutdown_prepare(SYSTEM_POWER_OFF);
344 if (pm_power_off_prepare)
345 pm_power_off_prepare();
346 disable_nonboot_cpus();
348 printk(KERN_EMERG "Power down.\n");
351 EXPORT_SYMBOL_GPL(kernel_power_off);
353 * Reboot system call: for obvious reasons only root may call it,
354 * and even root needs to set up some magic numbers in the registers
355 * so that some mistake won't make this reboot the whole machine.
356 * You can also set the meaning of the ctrl-alt-del-key here.
358 * reboot doesn't sync: do that yourself before calling this.
360 SYSCALL_DEFINE4(reboot, int, magic1, int, magic2, unsigned int, cmd,
366 /* We only trust the superuser with rebooting the system. */
367 if (!capable(CAP_SYS_BOOT))
370 /* For safety, we require "magic" arguments. */
371 if (magic1 != LINUX_REBOOT_MAGIC1 ||
372 (magic2 != LINUX_REBOOT_MAGIC2 &&
373 magic2 != LINUX_REBOOT_MAGIC2A &&
374 magic2 != LINUX_REBOOT_MAGIC2B &&
375 magic2 != LINUX_REBOOT_MAGIC2C))
378 /* Instead of trying to make the power_off code look like
379 * halt when pm_power_off is not set do it the easy way.
381 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
382 cmd = LINUX_REBOOT_CMD_HALT;
386 case LINUX_REBOOT_CMD_RESTART:
387 kernel_restart(NULL);
390 case LINUX_REBOOT_CMD_CAD_ON:
394 case LINUX_REBOOT_CMD_CAD_OFF:
398 case LINUX_REBOOT_CMD_HALT:
402 panic("cannot halt");
404 case LINUX_REBOOT_CMD_POWER_OFF:
410 case LINUX_REBOOT_CMD_RESTART2:
411 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
415 buffer[sizeof(buffer) - 1] = '\0';
417 kernel_restart(buffer);
421 case LINUX_REBOOT_CMD_KEXEC:
422 ret = kernel_kexec();
426 #ifdef CONFIG_HIBERNATION
427 case LINUX_REBOOT_CMD_SW_SUSPEND:
440 static void deferred_cad(struct work_struct *dummy)
442 kernel_restart(NULL);
446 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
447 * As it's called within an interrupt, it may NOT sync: the only choice
448 * is whether to reboot at once, or just ignore the ctrl-alt-del.
450 void ctrl_alt_del(void)
452 static DECLARE_WORK(cad_work, deferred_cad);
455 schedule_work(&cad_work);
457 kill_cad_pid(SIGINT, 1);
461 * Unprivileged users may change the real gid to the effective gid
462 * or vice versa. (BSD-style)
464 * If you set the real gid at all, or set the effective gid to a value not
465 * equal to the real gid, then the saved gid is set to the new effective gid.
467 * This makes it possible for a setgid program to completely drop its
468 * privileges, which is often a useful assertion to make when you are doing
469 * a security audit over a program.
471 * The general idea is that a program which uses just setregid() will be
472 * 100% compatible with BSD. A program which uses just setgid() will be
473 * 100% compatible with POSIX with saved IDs.
475 * SMP: There are not races, the GIDs are checked only by filesystem
476 * operations (as far as semantic preservation is concerned).
478 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
480 const struct cred *old;
484 new = prepare_creds();
487 old = current_cred();
489 retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
494 if (rgid != (gid_t) -1) {
495 if (old->gid == rgid ||
502 if (egid != (gid_t) -1) {
503 if (old->gid == egid ||
512 if (rgid != (gid_t) -1 ||
513 (egid != (gid_t) -1 && egid != old->gid))
514 new->sgid = new->egid;
515 new->fsgid = new->egid;
517 return commit_creds(new);
525 * setgid() is implemented like SysV w/ SAVED_IDS
527 * SMP: Same implicit races as above.
529 SYSCALL_DEFINE1(setgid, gid_t, gid)
531 const struct cred *old;
535 new = prepare_creds();
538 old = current_cred();
540 retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
545 if (capable(CAP_SETGID))
546 new->gid = new->egid = new->sgid = new->fsgid = gid;
547 else if (gid == old->gid || gid == old->sgid)
548 new->egid = new->fsgid = gid;
552 return commit_creds(new);
560 * change the user struct in a credentials set to match the new UID
562 static int set_user(struct cred *new)
564 struct user_struct *new_user;
566 new_user = alloc_uid(current_user_ns(), new->uid);
570 if (!task_can_switch_user(new_user, current)) {
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))
635 if (new->uid != old->uid) {
636 retval = set_user(new);
640 if (ruid != (uid_t) -1 ||
641 (euid != (uid_t) -1 && euid != old->uid))
642 new->suid = new->euid;
643 new->fsuid = new->euid;
645 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
649 return commit_creds(new);
657 * setuid() is implemented like SysV with SAVED_IDS
659 * Note that SAVED_ID's is deficient in that a setuid root program
660 * like sendmail, for example, cannot set its uid to be a normal
661 * user and then switch back, because if you're root, setuid() sets
662 * the saved uid too. If you don't like this, blame the bright people
663 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
664 * will allow a root program to temporarily drop privileges and be able to
665 * regain them by swapping the real and effective uid.
667 SYSCALL_DEFINE1(setuid, uid_t, uid)
669 const struct cred *old;
673 new = prepare_creds();
676 old = current_cred();
678 retval = security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_ID);
683 if (capable(CAP_SETUID)) {
684 new->suid = new->uid = uid;
685 if (uid != old->uid) {
686 retval = set_user(new);
690 } else if (uid != old->uid && uid != new->suid) {
694 new->fsuid = new->euid = uid;
696 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
700 return commit_creds(new);
709 * This function implements a generic ability to update ruid, euid,
710 * and suid. This allows you to implement the 4.4 compatible seteuid().
712 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
714 const struct cred *old;
718 new = prepare_creds();
722 retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
725 old = current_cred();
728 if (!capable(CAP_SETUID)) {
729 if (ruid != (uid_t) -1 && ruid != old->uid &&
730 ruid != old->euid && ruid != old->suid)
732 if (euid != (uid_t) -1 && euid != old->uid &&
733 euid != old->euid && euid != old->suid)
735 if (suid != (uid_t) -1 && suid != old->uid &&
736 suid != old->euid && suid != old->suid)
740 if (ruid != (uid_t) -1) {
742 if (ruid != old->uid) {
743 retval = set_user(new);
748 if (euid != (uid_t) -1)
750 if (suid != (uid_t) -1)
752 new->fsuid = new->euid;
754 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
758 return commit_creds(new);
765 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
767 const struct cred *cred = current_cred();
770 if (!(retval = put_user(cred->uid, ruid)) &&
771 !(retval = put_user(cred->euid, euid)))
772 retval = put_user(cred->suid, suid);
778 * Same as above, but for rgid, egid, sgid.
780 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
782 const struct cred *old;
786 new = prepare_creds();
789 old = current_cred();
791 retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
796 if (!capable(CAP_SETGID)) {
797 if (rgid != (gid_t) -1 && rgid != old->gid &&
798 rgid != old->egid && rgid != old->sgid)
800 if (egid != (gid_t) -1 && egid != old->gid &&
801 egid != old->egid && egid != old->sgid)
803 if (sgid != (gid_t) -1 && sgid != old->gid &&
804 sgid != old->egid && sgid != old->sgid)
808 if (rgid != (gid_t) -1)
810 if (egid != (gid_t) -1)
812 if (sgid != (gid_t) -1)
814 new->fsgid = new->egid;
816 return commit_creds(new);
823 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
825 const struct cred *cred = current_cred();
828 if (!(retval = put_user(cred->gid, rgid)) &&
829 !(retval = put_user(cred->egid, egid)))
830 retval = put_user(cred->sgid, sgid);
837 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
838 * is used for "access()" and for the NFS daemon (letting nfsd stay at
839 * whatever uid it wants to). It normally shadows "euid", except when
840 * explicitly set by setfsuid() or for access..
842 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
844 const struct cred *old;
848 new = prepare_creds();
850 return current_fsuid();
851 old = current_cred();
852 old_fsuid = old->fsuid;
854 if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
857 if (uid == old->uid || uid == old->euid ||
858 uid == old->suid || uid == old->fsuid ||
859 capable(CAP_SETUID)) {
860 if (uid != old_fsuid) {
862 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
877 * Samma på svenska..
879 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
881 const struct cred *old;
885 new = prepare_creds();
887 return current_fsgid();
888 old = current_cred();
889 old_fsgid = old->fsgid;
891 if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
894 if (gid == old->gid || gid == old->egid ||
895 gid == old->sgid || gid == old->fsgid ||
896 capable(CAP_SETGID)) {
897 if (gid != old_fsgid) {
912 void do_sys_times(struct tms *tms)
914 struct task_cputime cputime;
915 cputime_t cutime, cstime;
917 thread_group_cputime(current, &cputime);
918 spin_lock_irq(¤t->sighand->siglock);
919 cutime = current->signal->cutime;
920 cstime = current->signal->cstime;
921 spin_unlock_irq(¤t->sighand->siglock);
922 tms->tms_utime = cputime_to_clock_t(cputime.utime);
923 tms->tms_stime = cputime_to_clock_t(cputime.stime);
924 tms->tms_cutime = cputime_to_clock_t(cutime);
925 tms->tms_cstime = cputime_to_clock_t(cstime);
928 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
934 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
937 force_successful_syscall_return();
938 return (long) jiffies_64_to_clock_t(get_jiffies_64());
942 * This needs some heavy checking ...
943 * I just haven't the stomach for it. I also don't fully
944 * understand sessions/pgrp etc. Let somebody who does explain it.
946 * OK, I think I have the protection semantics right.... this is really
947 * only important on a multi-user system anyway, to make sure one user
948 * can't send a signal to a process owned by another. -TYT, 12/12/91
950 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
953 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
955 struct task_struct *p;
956 struct task_struct *group_leader = current->group_leader;
961 pid = task_pid_vnr(group_leader);
967 /* From this point forward we keep holding onto the tasklist lock
968 * so that our parent does not change from under us. -DaveM
970 write_lock_irq(&tasklist_lock);
973 p = find_task_by_vpid(pid);
978 if (!thread_group_leader(p))
981 if (same_thread_group(p->real_parent, group_leader)) {
983 if (task_session(p) != task_session(group_leader))
990 if (p != group_leader)
995 if (p->signal->leader)
1000 struct task_struct *g;
1002 pgrp = find_vpid(pgid);
1003 g = pid_task(pgrp, PIDTYPE_PGID);
1004 if (!g || task_session(g) != task_session(group_leader))
1008 err = security_task_setpgid(p, pgid);
1012 if (task_pgrp(p) != pgrp)
1013 change_pid(p, PIDTYPE_PGID, pgrp);
1017 /* All paths lead to here, thus we are safe. -DaveM */
1018 write_unlock_irq(&tasklist_lock);
1022 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1024 struct task_struct *p;
1030 grp = task_pgrp(current);
1033 p = find_task_by_vpid(pid);
1040 retval = security_task_getpgid(p);
1044 retval = pid_vnr(grp);
1050 #ifdef __ARCH_WANT_SYS_GETPGRP
1052 SYSCALL_DEFINE0(getpgrp)
1054 return sys_getpgid(0);
1059 SYSCALL_DEFINE1(getsid, pid_t, pid)
1061 struct task_struct *p;
1067 sid = task_session(current);
1070 p = find_task_by_vpid(pid);
1073 sid = task_session(p);
1077 retval = security_task_getsid(p);
1081 retval = pid_vnr(sid);
1087 SYSCALL_DEFINE0(setsid)
1089 struct task_struct *group_leader = current->group_leader;
1090 struct pid *sid = task_pid(group_leader);
1091 pid_t session = pid_vnr(sid);
1094 write_lock_irq(&tasklist_lock);
1095 /* Fail if I am already a session leader */
1096 if (group_leader->signal->leader)
1099 /* Fail if a process group id already exists that equals the
1100 * proposed session id.
1102 if (pid_task(sid, PIDTYPE_PGID))
1105 group_leader->signal->leader = 1;
1106 __set_special_pids(sid);
1108 proc_clear_tty(group_leader);
1112 write_unlock_irq(&tasklist_lock);
1117 * Supplementary group IDs
1120 /* init to 2 - one for init_task, one to ensure it is never freed */
1121 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1123 struct group_info *groups_alloc(int gidsetsize)
1125 struct group_info *group_info;
1129 nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1130 /* Make sure we always allocate at least one indirect block pointer */
1131 nblocks = nblocks ? : 1;
1132 group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1135 group_info->ngroups = gidsetsize;
1136 group_info->nblocks = nblocks;
1137 atomic_set(&group_info->usage, 1);
1139 if (gidsetsize <= NGROUPS_SMALL)
1140 group_info->blocks[0] = group_info->small_block;
1142 for (i = 0; i < nblocks; i++) {
1144 b = (void *)__get_free_page(GFP_USER);
1146 goto out_undo_partial_alloc;
1147 group_info->blocks[i] = b;
1152 out_undo_partial_alloc:
1154 free_page((unsigned long)group_info->blocks[i]);
1160 EXPORT_SYMBOL(groups_alloc);
1162 void groups_free(struct group_info *group_info)
1164 if (group_info->blocks[0] != group_info->small_block) {
1166 for (i = 0; i < group_info->nblocks; i++)
1167 free_page((unsigned long)group_info->blocks[i]);
1172 EXPORT_SYMBOL(groups_free);
1174 /* export the group_info to a user-space array */
1175 static int groups_to_user(gid_t __user *grouplist,
1176 const struct group_info *group_info)
1179 unsigned int count = group_info->ngroups;
1181 for (i = 0; i < group_info->nblocks; i++) {
1182 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1183 unsigned int len = cp_count * sizeof(*grouplist);
1185 if (copy_to_user(grouplist, group_info->blocks[i], len))
1188 grouplist += NGROUPS_PER_BLOCK;
1194 /* fill a group_info from a user-space array - it must be allocated already */
1195 static int groups_from_user(struct group_info *group_info,
1196 gid_t __user *grouplist)
1199 unsigned int count = group_info->ngroups;
1201 for (i = 0; i < group_info->nblocks; i++) {
1202 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1203 unsigned int len = cp_count * sizeof(*grouplist);
1205 if (copy_from_user(group_info->blocks[i], grouplist, len))
1208 grouplist += NGROUPS_PER_BLOCK;
1214 /* a simple Shell sort */
1215 static void groups_sort(struct group_info *group_info)
1217 int base, max, stride;
1218 int gidsetsize = group_info->ngroups;
1220 for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1225 max = gidsetsize - stride;
1226 for (base = 0; base < max; base++) {
1228 int right = left + stride;
1229 gid_t tmp = GROUP_AT(group_info, right);
1231 while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1232 GROUP_AT(group_info, right) =
1233 GROUP_AT(group_info, left);
1237 GROUP_AT(group_info, right) = tmp;
1243 /* a simple bsearch */
1244 int groups_search(const struct group_info *group_info, gid_t grp)
1246 unsigned int left, right;
1252 right = group_info->ngroups;
1253 while (left < right) {
1254 unsigned int mid = (left+right)/2;
1255 int cmp = grp - GROUP_AT(group_info, mid);
1267 * set_groups - Change a group subscription in a set of credentials
1268 * @new: The newly prepared set of credentials to alter
1269 * @group_info: The group list to install
1271 * Validate a group subscription and, if valid, insert it into a set
1274 int set_groups(struct cred *new, struct group_info *group_info)
1278 retval = security_task_setgroups(group_info);
1282 put_group_info(new->group_info);
1283 groups_sort(group_info);
1284 get_group_info(group_info);
1285 new->group_info = group_info;
1289 EXPORT_SYMBOL(set_groups);
1292 * set_current_groups - Change current's group subscription
1293 * @group_info: The group list to impose
1295 * Validate a group subscription and, if valid, impose it upon current's task
1298 int set_current_groups(struct group_info *group_info)
1303 new = prepare_creds();
1307 ret = set_groups(new, group_info);
1313 return commit_creds(new);
1316 EXPORT_SYMBOL(set_current_groups);
1318 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1320 const struct cred *cred = current_cred();
1326 /* no need to grab task_lock here; it cannot change */
1327 i = cred->group_info->ngroups;
1329 if (i > gidsetsize) {
1333 if (groups_to_user(grouplist, cred->group_info)) {
1343 * SMP: Our groups are copy-on-write. We can set them safely
1344 * without another task interfering.
1347 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1349 struct group_info *group_info;
1352 if (!capable(CAP_SETGID))
1354 if ((unsigned)gidsetsize > NGROUPS_MAX)
1357 group_info = groups_alloc(gidsetsize);
1360 retval = groups_from_user(group_info, grouplist);
1362 put_group_info(group_info);
1366 retval = set_current_groups(group_info);
1367 put_group_info(group_info);
1373 * Check whether we're fsgid/egid or in the supplemental group..
1375 int in_group_p(gid_t grp)
1377 const struct cred *cred = current_cred();
1380 if (grp != cred->fsgid)
1381 retval = groups_search(cred->group_info, grp);
1385 EXPORT_SYMBOL(in_group_p);
1387 int in_egroup_p(gid_t grp)
1389 const struct cred *cred = current_cred();
1392 if (grp != cred->egid)
1393 retval = groups_search(cred->group_info, grp);
1397 EXPORT_SYMBOL(in_egroup_p);
1399 DECLARE_RWSEM(uts_sem);
1401 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1405 down_read(&uts_sem);
1406 if (copy_to_user(name, utsname(), sizeof *name))
1412 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1415 char tmp[__NEW_UTS_LEN];
1417 if (!capable(CAP_SYS_ADMIN))
1419 if (len < 0 || len > __NEW_UTS_LEN)
1421 down_write(&uts_sem);
1423 if (!copy_from_user(tmp, name, len)) {
1424 struct new_utsname *u = utsname();
1426 memcpy(u->nodename, tmp, len);
1427 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1434 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1436 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1439 struct new_utsname *u;
1443 down_read(&uts_sem);
1445 i = 1 + strlen(u->nodename);
1449 if (copy_to_user(name, u->nodename, i))
1458 * Only setdomainname; getdomainname can be implemented by calling
1461 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1464 char tmp[__NEW_UTS_LEN];
1466 if (!capable(CAP_SYS_ADMIN))
1468 if (len < 0 || len > __NEW_UTS_LEN)
1471 down_write(&uts_sem);
1473 if (!copy_from_user(tmp, name, len)) {
1474 struct new_utsname *u = utsname();
1476 memcpy(u->domainname, tmp, len);
1477 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1484 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1486 if (resource >= RLIM_NLIMITS)
1489 struct rlimit value;
1490 task_lock(current->group_leader);
1491 value = current->signal->rlim[resource];
1492 task_unlock(current->group_leader);
1493 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1497 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1500 * Back compatibility for getrlimit. Needed for some apps.
1503 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1504 struct rlimit __user *, rlim)
1507 if (resource >= RLIM_NLIMITS)
1510 task_lock(current->group_leader);
1511 x = current->signal->rlim[resource];
1512 task_unlock(current->group_leader);
1513 if (x.rlim_cur > 0x7FFFFFFF)
1514 x.rlim_cur = 0x7FFFFFFF;
1515 if (x.rlim_max > 0x7FFFFFFF)
1516 x.rlim_max = 0x7FFFFFFF;
1517 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1522 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1524 struct rlimit new_rlim, *old_rlim;
1527 if (resource >= RLIM_NLIMITS)
1529 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1531 if (new_rlim.rlim_cur > new_rlim.rlim_max)
1533 old_rlim = current->signal->rlim + resource;
1534 if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1535 !capable(CAP_SYS_RESOURCE))
1537 if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1540 retval = security_task_setrlimit(resource, &new_rlim);
1544 if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1546 * The caller is asking for an immediate RLIMIT_CPU
1547 * expiry. But we use the zero value to mean "it was
1548 * never set". So let's cheat and make it one second
1551 new_rlim.rlim_cur = 1;
1554 task_lock(current->group_leader);
1555 *old_rlim = new_rlim;
1556 task_unlock(current->group_leader);
1558 if (resource != RLIMIT_CPU)
1562 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1563 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1564 * very long-standing error, and fixing it now risks breakage of
1565 * applications, so we live with it
1567 if (new_rlim.rlim_cur == RLIM_INFINITY)
1570 update_rlimit_cpu(new_rlim.rlim_cur);
1576 * It would make sense to put struct rusage in the task_struct,
1577 * except that would make the task_struct be *really big*. After
1578 * task_struct gets moved into malloc'ed memory, it would
1579 * make sense to do this. It will make moving the rest of the information
1580 * a lot simpler! (Which we're not doing right now because we're not
1581 * measuring them yet).
1583 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1584 * races with threads incrementing their own counters. But since word
1585 * reads are atomic, we either get new values or old values and we don't
1586 * care which for the sums. We always take the siglock to protect reading
1587 * the c* fields from p->signal from races with exit.c updating those
1588 * fields when reaping, so a sample either gets all the additions of a
1589 * given child after it's reaped, or none so this sample is before reaping.
1592 * We need to take the siglock for CHILDEREN, SELF and BOTH
1593 * for the cases current multithreaded, non-current single threaded
1594 * non-current multithreaded. Thread traversal is now safe with
1596 * Strictly speaking, we donot need to take the siglock if we are current and
1597 * single threaded, as no one else can take our signal_struct away, no one
1598 * else can reap the children to update signal->c* counters, and no one else
1599 * can race with the signal-> fields. If we do not take any lock, the
1600 * signal-> fields could be read out of order while another thread was just
1601 * exiting. So we should place a read memory barrier when we avoid the lock.
1602 * On the writer side, write memory barrier is implied in __exit_signal
1603 * as __exit_signal releases the siglock spinlock after updating the signal->
1604 * fields. But we don't do this yet to keep things simple.
1608 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1610 r->ru_nvcsw += t->nvcsw;
1611 r->ru_nivcsw += t->nivcsw;
1612 r->ru_minflt += t->min_flt;
1613 r->ru_majflt += t->maj_flt;
1614 r->ru_inblock += task_io_get_inblock(t);
1615 r->ru_oublock += task_io_get_oublock(t);
1618 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1620 struct task_struct *t;
1621 unsigned long flags;
1622 cputime_t utime, stime;
1623 struct task_cputime cputime;
1625 memset((char *) r, 0, sizeof *r);
1626 utime = stime = cputime_zero;
1628 if (who == RUSAGE_THREAD) {
1629 utime = task_utime(current);
1630 stime = task_stime(current);
1631 accumulate_thread_rusage(p, r);
1635 if (!lock_task_sighand(p, &flags))
1640 case RUSAGE_CHILDREN:
1641 utime = p->signal->cutime;
1642 stime = p->signal->cstime;
1643 r->ru_nvcsw = p->signal->cnvcsw;
1644 r->ru_nivcsw = p->signal->cnivcsw;
1645 r->ru_minflt = p->signal->cmin_flt;
1646 r->ru_majflt = p->signal->cmaj_flt;
1647 r->ru_inblock = p->signal->cinblock;
1648 r->ru_oublock = p->signal->coublock;
1650 if (who == RUSAGE_CHILDREN)
1654 thread_group_cputime(p, &cputime);
1655 utime = cputime_add(utime, cputime.utime);
1656 stime = cputime_add(stime, cputime.stime);
1657 r->ru_nvcsw += p->signal->nvcsw;
1658 r->ru_nivcsw += p->signal->nivcsw;
1659 r->ru_minflt += p->signal->min_flt;
1660 r->ru_majflt += p->signal->maj_flt;
1661 r->ru_inblock += p->signal->inblock;
1662 r->ru_oublock += p->signal->oublock;
1665 accumulate_thread_rusage(t, r);
1673 unlock_task_sighand(p, &flags);
1676 cputime_to_timeval(utime, &r->ru_utime);
1677 cputime_to_timeval(stime, &r->ru_stime);
1680 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1683 k_getrusage(p, who, &r);
1684 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1687 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1689 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1690 who != RUSAGE_THREAD)
1692 return getrusage(current, who, ru);
1695 SYSCALL_DEFINE1(umask, int, mask)
1697 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1701 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1702 unsigned long, arg4, unsigned long, arg5)
1704 struct task_struct *me = current;
1705 unsigned char comm[sizeof(me->comm)];
1708 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1709 if (error != -ENOSYS)
1714 case PR_SET_PDEATHSIG:
1715 if (!valid_signal(arg2)) {
1719 me->pdeath_signal = arg2;
1722 case PR_GET_PDEATHSIG:
1723 error = put_user(me->pdeath_signal, (int __user *)arg2);
1725 case PR_GET_DUMPABLE:
1726 error = get_dumpable(me->mm);
1728 case PR_SET_DUMPABLE:
1729 if (arg2 < 0 || arg2 > 1) {
1733 set_dumpable(me->mm, arg2);
1737 case PR_SET_UNALIGN:
1738 error = SET_UNALIGN_CTL(me, arg2);
1740 case PR_GET_UNALIGN:
1741 error = GET_UNALIGN_CTL(me, arg2);
1744 error = SET_FPEMU_CTL(me, arg2);
1747 error = GET_FPEMU_CTL(me, arg2);
1750 error = SET_FPEXC_CTL(me, arg2);
1753 error = GET_FPEXC_CTL(me, arg2);
1756 error = PR_TIMING_STATISTICAL;
1759 if (arg2 != PR_TIMING_STATISTICAL)
1766 comm[sizeof(me->comm)-1] = 0;
1767 if (strncpy_from_user(comm, (char __user *)arg2,
1768 sizeof(me->comm) - 1) < 0)
1770 set_task_comm(me, comm);
1773 get_task_comm(comm, me);
1774 if (copy_to_user((char __user *)arg2, comm,
1779 error = GET_ENDIAN(me, arg2);
1782 error = SET_ENDIAN(me, arg2);
1785 case PR_GET_SECCOMP:
1786 error = prctl_get_seccomp();
1788 case PR_SET_SECCOMP:
1789 error = prctl_set_seccomp(arg2);
1792 error = GET_TSC_CTL(arg2);
1795 error = SET_TSC_CTL(arg2);
1797 case PR_TASK_PERF_COUNTERS_DISABLE:
1798 error = perf_counter_task_disable();
1800 case PR_TASK_PERF_COUNTERS_ENABLE:
1801 error = perf_counter_task_enable();
1803 case PR_GET_TIMERSLACK:
1804 error = current->timer_slack_ns;
1806 case PR_SET_TIMERSLACK:
1808 current->timer_slack_ns =
1809 current->default_timer_slack_ns;
1811 current->timer_slack_ns = arg2;
1821 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1822 struct getcpu_cache __user *, unused)
1825 int cpu = raw_smp_processor_id();
1827 err |= put_user(cpu, cpup);
1829 err |= put_user(cpu_to_node(cpu), nodep);
1830 return err ? -EFAULT : 0;
1833 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1835 static void argv_cleanup(char **argv, char **envp)
1841 * orderly_poweroff - Trigger an orderly system poweroff
1842 * @force: force poweroff if command execution fails
1844 * This may be called from any context to trigger a system shutdown.
1845 * If the orderly shutdown fails, it will force an immediate shutdown.
1847 int orderly_poweroff(bool force)
1850 char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1851 static char *envp[] = {
1853 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1857 struct subprocess_info *info;
1860 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1861 __func__, poweroff_cmd);
1865 info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1871 call_usermodehelper_setcleanup(info, argv_cleanup);
1873 ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1877 printk(KERN_WARNING "Failed to start orderly shutdown: "
1878 "forcing the issue\n");
1880 /* I guess this should try to kick off some daemon to
1881 sync and poweroff asap. Or not even bother syncing
1882 if we're doing an emergency shutdown? */
1889 EXPORT_SYMBOL_GPL(orderly_poweroff);