4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/module.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/smp_lock.h>
12 #include <linux/notifier.h>
13 #include <linux/reboot.h>
14 #include <linux/prctl.h>
15 #include <linux/highuid.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/kexec.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.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,
365 /* We only trust the superuser with rebooting the system. */
366 if (!capable(CAP_SYS_BOOT))
369 /* For safety, we require "magic" arguments. */
370 if (magic1 != LINUX_REBOOT_MAGIC1 ||
371 (magic2 != LINUX_REBOOT_MAGIC2 &&
372 magic2 != LINUX_REBOOT_MAGIC2A &&
373 magic2 != LINUX_REBOOT_MAGIC2B &&
374 magic2 != LINUX_REBOOT_MAGIC2C))
377 /* Instead of trying to make the power_off code look like
378 * halt when pm_power_off is not set do it the easy way.
380 if ((cmd == LINUX_REBOOT_CMD_POWER_OFF) && !pm_power_off)
381 cmd = LINUX_REBOOT_CMD_HALT;
385 case LINUX_REBOOT_CMD_RESTART:
386 kernel_restart(NULL);
389 case LINUX_REBOOT_CMD_CAD_ON:
393 case LINUX_REBOOT_CMD_CAD_OFF:
397 case LINUX_REBOOT_CMD_HALT:
401 panic("cannot halt");
403 case LINUX_REBOOT_CMD_POWER_OFF:
409 case LINUX_REBOOT_CMD_RESTART2:
410 if (strncpy_from_user(&buffer[0], arg, sizeof(buffer) - 1) < 0) {
414 buffer[sizeof(buffer) - 1] = '\0';
416 kernel_restart(buffer);
420 case LINUX_REBOOT_CMD_KEXEC:
421 ret = kernel_kexec();
425 #ifdef CONFIG_HIBERNATION
426 case LINUX_REBOOT_CMD_SW_SUSPEND:
439 static void deferred_cad(struct work_struct *dummy)
441 kernel_restart(NULL);
445 * This function gets called by ctrl-alt-del - ie the keyboard interrupt.
446 * As it's called within an interrupt, it may NOT sync: the only choice
447 * is whether to reboot at once, or just ignore the ctrl-alt-del.
449 void ctrl_alt_del(void)
451 static DECLARE_WORK(cad_work, deferred_cad);
454 schedule_work(&cad_work);
456 kill_cad_pid(SIGINT, 1);
460 * Unprivileged users may change the real gid to the effective gid
461 * or vice versa. (BSD-style)
463 * If you set the real gid at all, or set the effective gid to a value not
464 * equal to the real gid, then the saved gid is set to the new effective gid.
466 * This makes it possible for a setgid program to completely drop its
467 * privileges, which is often a useful assertion to make when you are doing
468 * a security audit over a program.
470 * The general idea is that a program which uses just setregid() will be
471 * 100% compatible with BSD. A program which uses just setgid() will be
472 * 100% compatible with POSIX with saved IDs.
474 * SMP: There are not races, the GIDs are checked only by filesystem
475 * operations (as far as semantic preservation is concerned).
477 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
479 const struct cred *old;
483 new = prepare_creds();
486 old = current_cred();
488 retval = security_task_setgid(rgid, egid, (gid_t)-1, LSM_SETID_RE);
493 if (rgid != (gid_t) -1) {
494 if (old->gid == rgid ||
501 if (egid != (gid_t) -1) {
502 if (old->gid == egid ||
511 if (rgid != (gid_t) -1 ||
512 (egid != (gid_t) -1 && egid != old->gid))
513 new->sgid = new->egid;
514 new->fsgid = new->egid;
516 return commit_creds(new);
524 * setgid() is implemented like SysV w/ SAVED_IDS
526 * SMP: Same implicit races as above.
528 SYSCALL_DEFINE1(setgid, gid_t, gid)
530 const struct cred *old;
534 new = prepare_creds();
537 old = current_cred();
539 retval = security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_ID);
544 if (capable(CAP_SETGID))
545 new->gid = new->egid = new->sgid = new->fsgid = gid;
546 else if (gid == old->gid || gid == old->sgid)
547 new->egid = new->fsgid = gid;
551 return commit_creds(new);
559 * change the user struct in a credentials set to match the new UID
561 static int set_user(struct cred *new)
563 struct user_struct *new_user;
565 new_user = alloc_uid(current_user_ns(), new->uid);
569 if (!task_can_switch_user(new_user, current)) {
574 if (atomic_read(&new_user->processes) >=
575 current->signal->rlim[RLIMIT_NPROC].rlim_cur &&
576 new_user != INIT_USER) {
582 new->user = new_user;
587 * Unprivileged users may change the real uid to the effective uid
588 * or vice versa. (BSD-style)
590 * If you set the real uid at all, or set the effective uid to a value not
591 * equal to the real uid, then the saved uid is set to the new effective uid.
593 * This makes it possible for a setuid program to completely drop its
594 * privileges, which is often a useful assertion to make when you are doing
595 * a security audit over a program.
597 * The general idea is that a program which uses just setreuid() will be
598 * 100% compatible with BSD. A program which uses just setuid() will be
599 * 100% compatible with POSIX with saved IDs.
601 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
603 const struct cred *old;
607 new = prepare_creds();
610 old = current_cred();
612 retval = security_task_setuid(ruid, euid, (uid_t)-1, LSM_SETID_RE);
617 if (ruid != (uid_t) -1) {
619 if (old->uid != ruid &&
621 !capable(CAP_SETUID))
625 if (euid != (uid_t) -1) {
627 if (old->uid != euid &&
630 !capable(CAP_SETUID))
634 if (new->uid != old->uid) {
635 retval = set_user(new);
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) {
685 retval = set_user(new);
689 } else if (uid != old->uid && uid != new->suid) {
693 new->fsuid = new->euid = uid;
695 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
699 return commit_creds(new);
708 * This function implements a generic ability to update ruid, euid,
709 * and suid. This allows you to implement the 4.4 compatible seteuid().
711 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
713 const struct cred *old;
717 new = prepare_creds();
721 retval = security_task_setuid(ruid, euid, suid, LSM_SETID_RES);
724 old = current_cred();
727 if (!capable(CAP_SETUID)) {
728 if (ruid != (uid_t) -1 && ruid != old->uid &&
729 ruid != old->euid && ruid != old->suid)
731 if (euid != (uid_t) -1 && euid != old->uid &&
732 euid != old->euid && euid != old->suid)
734 if (suid != (uid_t) -1 && suid != old->uid &&
735 suid != old->euid && suid != old->suid)
739 if (ruid != (uid_t) -1) {
741 if (ruid != old->uid) {
742 retval = set_user(new);
747 if (euid != (uid_t) -1)
749 if (suid != (uid_t) -1)
751 new->fsuid = new->euid;
753 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
757 return commit_creds(new);
764 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruid, uid_t __user *, euid, uid_t __user *, suid)
766 const struct cred *cred = current_cred();
769 if (!(retval = put_user(cred->uid, ruid)) &&
770 !(retval = put_user(cred->euid, euid)))
771 retval = put_user(cred->suid, suid);
777 * Same as above, but for rgid, egid, sgid.
779 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
781 const struct cred *old;
785 new = prepare_creds();
788 old = current_cred();
790 retval = security_task_setgid(rgid, egid, sgid, LSM_SETID_RES);
795 if (!capable(CAP_SETGID)) {
796 if (rgid != (gid_t) -1 && rgid != old->gid &&
797 rgid != old->egid && rgid != old->sgid)
799 if (egid != (gid_t) -1 && egid != old->gid &&
800 egid != old->egid && egid != old->sgid)
802 if (sgid != (gid_t) -1 && sgid != old->gid &&
803 sgid != old->egid && sgid != old->sgid)
807 if (rgid != (gid_t) -1)
809 if (egid != (gid_t) -1)
811 if (sgid != (gid_t) -1)
813 new->fsgid = new->egid;
815 return commit_creds(new);
822 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgid, gid_t __user *, egid, gid_t __user *, sgid)
824 const struct cred *cred = current_cred();
827 if (!(retval = put_user(cred->gid, rgid)) &&
828 !(retval = put_user(cred->egid, egid)))
829 retval = put_user(cred->sgid, sgid);
836 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
837 * is used for "access()" and for the NFS daemon (letting nfsd stay at
838 * whatever uid it wants to). It normally shadows "euid", except when
839 * explicitly set by setfsuid() or for access..
841 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
843 const struct cred *old;
847 new = prepare_creds();
849 return current_fsuid();
850 old = current_cred();
851 old_fsuid = old->fsuid;
853 if (security_task_setuid(uid, (uid_t)-1, (uid_t)-1, LSM_SETID_FS) < 0)
856 if (uid == old->uid || uid == old->euid ||
857 uid == old->suid || uid == old->fsuid ||
858 capable(CAP_SETUID)) {
859 if (uid != old_fsuid) {
861 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
876 * Samma på svenska..
878 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
880 const struct cred *old;
884 new = prepare_creds();
886 return current_fsgid();
887 old = current_cred();
888 old_fsgid = old->fsgid;
890 if (security_task_setgid(gid, (gid_t)-1, (gid_t)-1, LSM_SETID_FS))
893 if (gid == old->gid || gid == old->egid ||
894 gid == old->sgid || gid == old->fsgid ||
895 capable(CAP_SETGID)) {
896 if (gid != old_fsgid) {
911 void do_sys_times(struct tms *tms)
913 struct task_cputime cputime;
914 cputime_t cutime, cstime;
916 thread_group_cputime(current, &cputime);
917 spin_lock_irq(¤t->sighand->siglock);
918 cutime = current->signal->cutime;
919 cstime = current->signal->cstime;
920 spin_unlock_irq(¤t->sighand->siglock);
921 tms->tms_utime = cputime_to_clock_t(cputime.utime);
922 tms->tms_stime = cputime_to_clock_t(cputime.stime);
923 tms->tms_cutime = cputime_to_clock_t(cutime);
924 tms->tms_cstime = cputime_to_clock_t(cstime);
927 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
933 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
936 force_successful_syscall_return();
937 return (long) jiffies_64_to_clock_t(get_jiffies_64());
941 * This needs some heavy checking ...
942 * I just haven't the stomach for it. I also don't fully
943 * understand sessions/pgrp etc. Let somebody who does explain it.
945 * OK, I think I have the protection semantics right.... this is really
946 * only important on a multi-user system anyway, to make sure one user
947 * can't send a signal to a process owned by another. -TYT, 12/12/91
949 * Auch. Had to add the 'did_exec' flag to conform completely to POSIX.
952 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
954 struct task_struct *p;
955 struct task_struct *group_leader = current->group_leader;
960 pid = task_pid_vnr(group_leader);
966 /* From this point forward we keep holding onto the tasklist lock
967 * so that our parent does not change from under us. -DaveM
969 write_lock_irq(&tasklist_lock);
972 p = find_task_by_vpid(pid);
977 if (!thread_group_leader(p))
980 if (same_thread_group(p->real_parent, group_leader)) {
982 if (task_session(p) != task_session(group_leader))
989 if (p != group_leader)
994 if (p->signal->leader)
999 struct task_struct *g;
1001 pgrp = find_vpid(pgid);
1002 g = pid_task(pgrp, PIDTYPE_PGID);
1003 if (!g || task_session(g) != task_session(group_leader))
1007 err = security_task_setpgid(p, pgid);
1011 if (task_pgrp(p) != pgrp)
1012 change_pid(p, PIDTYPE_PGID, pgrp);
1016 /* All paths lead to here, thus we are safe. -DaveM */
1017 write_unlock_irq(&tasklist_lock);
1021 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1023 struct task_struct *p;
1029 grp = task_pgrp(current);
1032 p = find_task_by_vpid(pid);
1039 retval = security_task_getpgid(p);
1043 retval = pid_vnr(grp);
1049 #ifdef __ARCH_WANT_SYS_GETPGRP
1051 SYSCALL_DEFINE0(getpgrp)
1053 return sys_getpgid(0);
1058 SYSCALL_DEFINE1(getsid, pid_t, pid)
1060 struct task_struct *p;
1066 sid = task_session(current);
1069 p = find_task_by_vpid(pid);
1072 sid = task_session(p);
1076 retval = security_task_getsid(p);
1080 retval = pid_vnr(sid);
1086 SYSCALL_DEFINE0(setsid)
1088 struct task_struct *group_leader = current->group_leader;
1089 struct pid *sid = task_pid(group_leader);
1090 pid_t session = pid_vnr(sid);
1093 write_lock_irq(&tasklist_lock);
1094 /* Fail if I am already a session leader */
1095 if (group_leader->signal->leader)
1098 /* Fail if a process group id already exists that equals the
1099 * proposed session id.
1101 if (pid_task(sid, PIDTYPE_PGID))
1104 group_leader->signal->leader = 1;
1105 __set_special_pids(sid);
1107 proc_clear_tty(group_leader);
1111 write_unlock_irq(&tasklist_lock);
1116 * Supplementary group IDs
1119 /* init to 2 - one for init_task, one to ensure it is never freed */
1120 struct group_info init_groups = { .usage = ATOMIC_INIT(2) };
1122 struct group_info *groups_alloc(int gidsetsize)
1124 struct group_info *group_info;
1128 nblocks = (gidsetsize + NGROUPS_PER_BLOCK - 1) / NGROUPS_PER_BLOCK;
1129 /* Make sure we always allocate at least one indirect block pointer */
1130 nblocks = nblocks ? : 1;
1131 group_info = kmalloc(sizeof(*group_info) + nblocks*sizeof(gid_t *), GFP_USER);
1134 group_info->ngroups = gidsetsize;
1135 group_info->nblocks = nblocks;
1136 atomic_set(&group_info->usage, 1);
1138 if (gidsetsize <= NGROUPS_SMALL)
1139 group_info->blocks[0] = group_info->small_block;
1141 for (i = 0; i < nblocks; i++) {
1143 b = (void *)__get_free_page(GFP_USER);
1145 goto out_undo_partial_alloc;
1146 group_info->blocks[i] = b;
1151 out_undo_partial_alloc:
1153 free_page((unsigned long)group_info->blocks[i]);
1159 EXPORT_SYMBOL(groups_alloc);
1161 void groups_free(struct group_info *group_info)
1163 if (group_info->blocks[0] != group_info->small_block) {
1165 for (i = 0; i < group_info->nblocks; i++)
1166 free_page((unsigned long)group_info->blocks[i]);
1171 EXPORT_SYMBOL(groups_free);
1173 /* export the group_info to a user-space array */
1174 static int groups_to_user(gid_t __user *grouplist,
1175 const struct group_info *group_info)
1178 unsigned int count = group_info->ngroups;
1180 for (i = 0; i < group_info->nblocks; i++) {
1181 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1182 unsigned int len = cp_count * sizeof(*grouplist);
1184 if (copy_to_user(grouplist, group_info->blocks[i], len))
1187 grouplist += NGROUPS_PER_BLOCK;
1193 /* fill a group_info from a user-space array - it must be allocated already */
1194 static int groups_from_user(struct group_info *group_info,
1195 gid_t __user *grouplist)
1198 unsigned int count = group_info->ngroups;
1200 for (i = 0; i < group_info->nblocks; i++) {
1201 unsigned int cp_count = min(NGROUPS_PER_BLOCK, count);
1202 unsigned int len = cp_count * sizeof(*grouplist);
1204 if (copy_from_user(group_info->blocks[i], grouplist, len))
1207 grouplist += NGROUPS_PER_BLOCK;
1213 /* a simple Shell sort */
1214 static void groups_sort(struct group_info *group_info)
1216 int base, max, stride;
1217 int gidsetsize = group_info->ngroups;
1219 for (stride = 1; stride < gidsetsize; stride = 3 * stride + 1)
1224 max = gidsetsize - stride;
1225 for (base = 0; base < max; base++) {
1227 int right = left + stride;
1228 gid_t tmp = GROUP_AT(group_info, right);
1230 while (left >= 0 && GROUP_AT(group_info, left) > tmp) {
1231 GROUP_AT(group_info, right) =
1232 GROUP_AT(group_info, left);
1236 GROUP_AT(group_info, right) = tmp;
1242 /* a simple bsearch */
1243 int groups_search(const struct group_info *group_info, gid_t grp)
1245 unsigned int left, right;
1251 right = group_info->ngroups;
1252 while (left < right) {
1253 unsigned int mid = (left+right)/2;
1254 int cmp = grp - GROUP_AT(group_info, mid);
1266 * set_groups - Change a group subscription in a set of credentials
1267 * @new: The newly prepared set of credentials to alter
1268 * @group_info: The group list to install
1270 * Validate a group subscription and, if valid, insert it into a set
1273 int set_groups(struct cred *new, struct group_info *group_info)
1277 retval = security_task_setgroups(group_info);
1281 put_group_info(new->group_info);
1282 groups_sort(group_info);
1283 get_group_info(group_info);
1284 new->group_info = group_info;
1288 EXPORT_SYMBOL(set_groups);
1291 * set_current_groups - Change current's group subscription
1292 * @group_info: The group list to impose
1294 * Validate a group subscription and, if valid, impose it upon current's task
1297 int set_current_groups(struct group_info *group_info)
1302 new = prepare_creds();
1306 ret = set_groups(new, group_info);
1312 return commit_creds(new);
1315 EXPORT_SYMBOL(set_current_groups);
1317 SYSCALL_DEFINE2(getgroups, int, gidsetsize, gid_t __user *, grouplist)
1319 const struct cred *cred = current_cred();
1325 /* no need to grab task_lock here; it cannot change */
1326 i = cred->group_info->ngroups;
1328 if (i > gidsetsize) {
1332 if (groups_to_user(grouplist, cred->group_info)) {
1342 * SMP: Our groups are copy-on-write. We can set them safely
1343 * without another task interfering.
1346 SYSCALL_DEFINE2(setgroups, int, gidsetsize, gid_t __user *, grouplist)
1348 struct group_info *group_info;
1351 if (!capable(CAP_SETGID))
1353 if ((unsigned)gidsetsize > NGROUPS_MAX)
1356 group_info = groups_alloc(gidsetsize);
1359 retval = groups_from_user(group_info, grouplist);
1361 put_group_info(group_info);
1365 retval = set_current_groups(group_info);
1366 put_group_info(group_info);
1372 * Check whether we're fsgid/egid or in the supplemental group..
1374 int in_group_p(gid_t grp)
1376 const struct cred *cred = current_cred();
1379 if (grp != cred->fsgid)
1380 retval = groups_search(cred->group_info, grp);
1384 EXPORT_SYMBOL(in_group_p);
1386 int in_egroup_p(gid_t grp)
1388 const struct cred *cred = current_cred();
1391 if (grp != cred->egid)
1392 retval = groups_search(cred->group_info, grp);
1396 EXPORT_SYMBOL(in_egroup_p);
1398 DECLARE_RWSEM(uts_sem);
1400 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1404 down_read(&uts_sem);
1405 if (copy_to_user(name, utsname(), sizeof *name))
1411 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1414 char tmp[__NEW_UTS_LEN];
1416 if (!capable(CAP_SYS_ADMIN))
1418 if (len < 0 || len > __NEW_UTS_LEN)
1420 down_write(&uts_sem);
1422 if (!copy_from_user(tmp, name, len)) {
1423 struct new_utsname *u = utsname();
1425 memcpy(u->nodename, tmp, len);
1426 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1433 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1435 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1438 struct new_utsname *u;
1442 down_read(&uts_sem);
1444 i = 1 + strlen(u->nodename);
1448 if (copy_to_user(name, u->nodename, i))
1457 * Only setdomainname; getdomainname can be implemented by calling
1460 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1463 char tmp[__NEW_UTS_LEN];
1465 if (!capable(CAP_SYS_ADMIN))
1467 if (len < 0 || len > __NEW_UTS_LEN)
1470 down_write(&uts_sem);
1472 if (!copy_from_user(tmp, name, len)) {
1473 struct new_utsname *u = utsname();
1475 memcpy(u->domainname, tmp, len);
1476 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1483 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1485 if (resource >= RLIM_NLIMITS)
1488 struct rlimit value;
1489 task_lock(current->group_leader);
1490 value = current->signal->rlim[resource];
1491 task_unlock(current->group_leader);
1492 return copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1496 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1499 * Back compatibility for getrlimit. Needed for some apps.
1502 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1503 struct rlimit __user *, rlim)
1506 if (resource >= RLIM_NLIMITS)
1509 task_lock(current->group_leader);
1510 x = current->signal->rlim[resource];
1511 task_unlock(current->group_leader);
1512 if (x.rlim_cur > 0x7FFFFFFF)
1513 x.rlim_cur = 0x7FFFFFFF;
1514 if (x.rlim_max > 0x7FFFFFFF)
1515 x.rlim_max = 0x7FFFFFFF;
1516 return copy_to_user(rlim, &x, sizeof(x))?-EFAULT:0;
1521 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1523 struct rlimit new_rlim, *old_rlim;
1526 if (resource >= RLIM_NLIMITS)
1528 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1530 if (new_rlim.rlim_cur > new_rlim.rlim_max)
1532 old_rlim = current->signal->rlim + resource;
1533 if ((new_rlim.rlim_max > old_rlim->rlim_max) &&
1534 !capable(CAP_SYS_RESOURCE))
1536 if (resource == RLIMIT_NOFILE && new_rlim.rlim_max > sysctl_nr_open)
1539 retval = security_task_setrlimit(resource, &new_rlim);
1543 if (resource == RLIMIT_CPU && new_rlim.rlim_cur == 0) {
1545 * The caller is asking for an immediate RLIMIT_CPU
1546 * expiry. But we use the zero value to mean "it was
1547 * never set". So let's cheat and make it one second
1550 new_rlim.rlim_cur = 1;
1553 task_lock(current->group_leader);
1554 *old_rlim = new_rlim;
1555 task_unlock(current->group_leader);
1557 if (resource != RLIMIT_CPU)
1561 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1562 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1563 * very long-standing error, and fixing it now risks breakage of
1564 * applications, so we live with it
1566 if (new_rlim.rlim_cur == RLIM_INFINITY)
1569 update_rlimit_cpu(new_rlim.rlim_cur);
1575 * It would make sense to put struct rusage in the task_struct,
1576 * except that would make the task_struct be *really big*. After
1577 * task_struct gets moved into malloc'ed memory, it would
1578 * make sense to do this. It will make moving the rest of the information
1579 * a lot simpler! (Which we're not doing right now because we're not
1580 * measuring them yet).
1582 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1583 * races with threads incrementing their own counters. But since word
1584 * reads are atomic, we either get new values or old values and we don't
1585 * care which for the sums. We always take the siglock to protect reading
1586 * the c* fields from p->signal from races with exit.c updating those
1587 * fields when reaping, so a sample either gets all the additions of a
1588 * given child after it's reaped, or none so this sample is before reaping.
1591 * We need to take the siglock for CHILDEREN, SELF and BOTH
1592 * for the cases current multithreaded, non-current single threaded
1593 * non-current multithreaded. Thread traversal is now safe with
1595 * Strictly speaking, we donot need to take the siglock if we are current and
1596 * single threaded, as no one else can take our signal_struct away, no one
1597 * else can reap the children to update signal->c* counters, and no one else
1598 * can race with the signal-> fields. If we do not take any lock, the
1599 * signal-> fields could be read out of order while another thread was just
1600 * exiting. So we should place a read memory barrier when we avoid the lock.
1601 * On the writer side, write memory barrier is implied in __exit_signal
1602 * as __exit_signal releases the siglock spinlock after updating the signal->
1603 * fields. But we don't do this yet to keep things simple.
1607 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1609 r->ru_nvcsw += t->nvcsw;
1610 r->ru_nivcsw += t->nivcsw;
1611 r->ru_minflt += t->min_flt;
1612 r->ru_majflt += t->maj_flt;
1613 r->ru_inblock += task_io_get_inblock(t);
1614 r->ru_oublock += task_io_get_oublock(t);
1617 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1619 struct task_struct *t;
1620 unsigned long flags;
1621 cputime_t utime, stime;
1622 struct task_cputime cputime;
1624 memset((char *) r, 0, sizeof *r);
1625 utime = stime = cputime_zero;
1627 if (who == RUSAGE_THREAD) {
1628 utime = task_utime(current);
1629 stime = task_stime(current);
1630 accumulate_thread_rusage(p, r);
1634 if (!lock_task_sighand(p, &flags))
1639 case RUSAGE_CHILDREN:
1640 utime = p->signal->cutime;
1641 stime = p->signal->cstime;
1642 r->ru_nvcsw = p->signal->cnvcsw;
1643 r->ru_nivcsw = p->signal->cnivcsw;
1644 r->ru_minflt = p->signal->cmin_flt;
1645 r->ru_majflt = p->signal->cmaj_flt;
1646 r->ru_inblock = p->signal->cinblock;
1647 r->ru_oublock = p->signal->coublock;
1649 if (who == RUSAGE_CHILDREN)
1653 thread_group_cputime(p, &cputime);
1654 utime = cputime_add(utime, cputime.utime);
1655 stime = cputime_add(stime, cputime.stime);
1656 r->ru_nvcsw += p->signal->nvcsw;
1657 r->ru_nivcsw += p->signal->nivcsw;
1658 r->ru_minflt += p->signal->min_flt;
1659 r->ru_majflt += p->signal->maj_flt;
1660 r->ru_inblock += p->signal->inblock;
1661 r->ru_oublock += p->signal->oublock;
1664 accumulate_thread_rusage(t, r);
1672 unlock_task_sighand(p, &flags);
1675 cputime_to_timeval(utime, &r->ru_utime);
1676 cputime_to_timeval(stime, &r->ru_stime);
1679 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1682 k_getrusage(p, who, &r);
1683 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1686 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1688 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1689 who != RUSAGE_THREAD)
1691 return getrusage(current, who, ru);
1694 SYSCALL_DEFINE1(umask, int, mask)
1696 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1700 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
1701 unsigned long, arg4, unsigned long, arg5)
1703 struct task_struct *me = current;
1704 unsigned char comm[sizeof(me->comm)];
1707 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
1708 if (error != -ENOSYS)
1713 case PR_SET_PDEATHSIG:
1714 if (!valid_signal(arg2)) {
1718 me->pdeath_signal = arg2;
1721 case PR_GET_PDEATHSIG:
1722 error = put_user(me->pdeath_signal, (int __user *)arg2);
1724 case PR_GET_DUMPABLE:
1725 error = get_dumpable(me->mm);
1727 case PR_SET_DUMPABLE:
1728 if (arg2 < 0 || arg2 > 1) {
1732 set_dumpable(me->mm, arg2);
1736 case PR_SET_UNALIGN:
1737 error = SET_UNALIGN_CTL(me, arg2);
1739 case PR_GET_UNALIGN:
1740 error = GET_UNALIGN_CTL(me, arg2);
1743 error = SET_FPEMU_CTL(me, arg2);
1746 error = GET_FPEMU_CTL(me, arg2);
1749 error = SET_FPEXC_CTL(me, arg2);
1752 error = GET_FPEXC_CTL(me, arg2);
1755 error = PR_TIMING_STATISTICAL;
1758 if (arg2 != PR_TIMING_STATISTICAL)
1765 comm[sizeof(me->comm)-1] = 0;
1766 if (strncpy_from_user(comm, (char __user *)arg2,
1767 sizeof(me->comm) - 1) < 0)
1769 set_task_comm(me, comm);
1772 get_task_comm(comm, me);
1773 if (copy_to_user((char __user *)arg2, comm,
1778 error = GET_ENDIAN(me, arg2);
1781 error = SET_ENDIAN(me, arg2);
1784 case PR_GET_SECCOMP:
1785 error = prctl_get_seccomp();
1787 case PR_SET_SECCOMP:
1788 error = prctl_set_seccomp(arg2);
1791 error = GET_TSC_CTL(arg2);
1794 error = SET_TSC_CTL(arg2);
1796 case PR_GET_TIMERSLACK:
1797 error = current->timer_slack_ns;
1799 case PR_SET_TIMERSLACK:
1801 current->timer_slack_ns =
1802 current->default_timer_slack_ns;
1804 current->timer_slack_ns = arg2;
1814 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
1815 struct getcpu_cache __user *, unused)
1818 int cpu = raw_smp_processor_id();
1820 err |= put_user(cpu, cpup);
1822 err |= put_user(cpu_to_node(cpu), nodep);
1823 return err ? -EFAULT : 0;
1826 char poweroff_cmd[POWEROFF_CMD_PATH_LEN] = "/sbin/poweroff";
1828 static void argv_cleanup(char **argv, char **envp)
1834 * orderly_poweroff - Trigger an orderly system poweroff
1835 * @force: force poweroff if command execution fails
1837 * This may be called from any context to trigger a system shutdown.
1838 * If the orderly shutdown fails, it will force an immediate shutdown.
1840 int orderly_poweroff(bool force)
1843 char **argv = argv_split(GFP_ATOMIC, poweroff_cmd, &argc);
1844 static char *envp[] = {
1846 "PATH=/sbin:/bin:/usr/sbin:/usr/bin",
1850 struct subprocess_info *info;
1853 printk(KERN_WARNING "%s failed to allocate memory for \"%s\"\n",
1854 __func__, poweroff_cmd);
1858 info = call_usermodehelper_setup(argv[0], argv, envp, GFP_ATOMIC);
1864 call_usermodehelper_setcleanup(info, argv_cleanup);
1866 ret = call_usermodehelper_exec(info, UMH_NO_WAIT);
1870 printk(KERN_WARNING "Failed to start orderly shutdown: "
1871 "forcing the issue\n");
1873 /* I guess this should try to kick off some daemon to
1874 sync and poweroff asap. Or not even bother syncing
1875 if we're doing an emergency shutdown? */
1882 EXPORT_SYMBOL_GPL(orderly_poweroff);