4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/export.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/workqueue.h>
20 #include <linux/capability.h>
21 #include <linux/device.h>
22 #include <linux/key.h>
23 #include <linux/times.h>
24 #include <linux/posix-timers.h>
25 #include <linux/security.h>
26 #include <linux/dcookies.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
45 #include <linux/compat.h>
46 #include <linux/syscalls.h>
47 #include <linux/kprobes.h>
48 #include <linux/user_namespace.h>
49 #include <linux/binfmts.h>
51 #include <linux/sched.h>
52 #include <linux/sched/autogroup.h>
53 #include <linux/sched/loadavg.h>
54 #include <linux/sched/stat.h>
55 #include <linux/sched/mm.h>
56 #include <linux/sched/coredump.h>
57 #include <linux/rcupdate.h>
58 #include <linux/uidgid.h>
59 #include <linux/cred.h>
61 #include <linux/kmsg_dump.h>
62 /* Move somewhere else to avoid recompiling? */
63 #include <generated/utsrelease.h>
65 #include <linux/uaccess.h>
67 #include <asm/unistd.h>
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)
99 #ifndef MPX_ENABLE_MANAGEMENT
100 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
102 #ifndef MPX_DISABLE_MANAGEMENT
103 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
106 # define GET_FP_MODE(a) (-EINVAL)
109 # define SET_FP_MODE(a,b) (-EINVAL)
113 * this is where the system-wide overflow UID and GID are defined, for
114 * architectures that now have 32-bit UID/GID but didn't in the past
117 int overflowuid
= DEFAULT_OVERFLOWUID
;
118 int overflowgid
= DEFAULT_OVERFLOWGID
;
120 EXPORT_SYMBOL(overflowuid
);
121 EXPORT_SYMBOL(overflowgid
);
124 * the same as above, but for filesystems which can only store a 16-bit
125 * UID and GID. as such, this is needed on all architectures
128 int fs_overflowuid
= DEFAULT_FS_OVERFLOWUID
;
129 int fs_overflowgid
= DEFAULT_FS_OVERFLOWUID
;
131 EXPORT_SYMBOL(fs_overflowuid
);
132 EXPORT_SYMBOL(fs_overflowgid
);
135 * Returns true if current's euid is same as p's uid or euid,
136 * or has CAP_SYS_NICE to p's user_ns.
138 * Called with rcu_read_lock, creds are safe
140 static bool set_one_prio_perm(struct task_struct
*p
)
142 const struct cred
*cred
= current_cred(), *pcred
= __task_cred(p
);
144 if (uid_eq(pcred
->uid
, cred
->euid
) ||
145 uid_eq(pcred
->euid
, cred
->euid
))
147 if (ns_capable(pcred
->user_ns
, CAP_SYS_NICE
))
153 * set the priority of a task
154 * - the caller must hold the RCU read lock
156 static int set_one_prio(struct task_struct
*p
, int niceval
, int error
)
160 if (!set_one_prio_perm(p
)) {
164 if (niceval
< task_nice(p
) && !can_nice(p
, niceval
)) {
168 no_nice
= security_task_setnice(p
, niceval
);
175 set_user_nice(p
, niceval
);
180 SYSCALL_DEFINE3(setpriority
, int, which
, int, who
, int, niceval
)
182 struct task_struct
*g
, *p
;
183 struct user_struct
*user
;
184 const struct cred
*cred
= current_cred();
189 if (which
> PRIO_USER
|| which
< PRIO_PROCESS
)
192 /* normalize: avoid signed division (rounding problems) */
194 if (niceval
< MIN_NICE
)
196 if (niceval
> MAX_NICE
)
200 read_lock(&tasklist_lock
);
204 p
= find_task_by_vpid(who
);
208 error
= set_one_prio(p
, niceval
, error
);
212 pgrp
= find_vpid(who
);
214 pgrp
= task_pgrp(current
);
215 do_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
) {
216 error
= set_one_prio(p
, niceval
, error
);
217 } while_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
);
220 uid
= make_kuid(cred
->user_ns
, who
);
224 else if (!uid_eq(uid
, cred
->uid
)) {
225 user
= find_user(uid
);
227 goto out_unlock
; /* No processes for this user */
229 do_each_thread(g
, p
) {
230 if (uid_eq(task_uid(p
), uid
) && task_pid_vnr(p
))
231 error
= set_one_prio(p
, niceval
, error
);
232 } while_each_thread(g
, p
);
233 if (!uid_eq(uid
, cred
->uid
))
234 free_uid(user
); /* For find_user() */
238 read_unlock(&tasklist_lock
);
245 * Ugh. To avoid negative return values, "getpriority()" will
246 * not return the normal nice-value, but a negated value that
247 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
248 * to stay compatible.
250 SYSCALL_DEFINE2(getpriority
, int, which
, int, who
)
252 struct task_struct
*g
, *p
;
253 struct user_struct
*user
;
254 const struct cred
*cred
= current_cred();
255 long niceval
, retval
= -ESRCH
;
259 if (which
> PRIO_USER
|| which
< PRIO_PROCESS
)
263 read_lock(&tasklist_lock
);
267 p
= find_task_by_vpid(who
);
271 niceval
= nice_to_rlimit(task_nice(p
));
272 if (niceval
> retval
)
278 pgrp
= find_vpid(who
);
280 pgrp
= task_pgrp(current
);
281 do_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
) {
282 niceval
= nice_to_rlimit(task_nice(p
));
283 if (niceval
> retval
)
285 } while_each_pid_thread(pgrp
, PIDTYPE_PGID
, p
);
288 uid
= make_kuid(cred
->user_ns
, who
);
292 else if (!uid_eq(uid
, cred
->uid
)) {
293 user
= find_user(uid
);
295 goto out_unlock
; /* No processes for this user */
297 do_each_thread(g
, p
) {
298 if (uid_eq(task_uid(p
), uid
) && task_pid_vnr(p
)) {
299 niceval
= nice_to_rlimit(task_nice(p
));
300 if (niceval
> retval
)
303 } while_each_thread(g
, p
);
304 if (!uid_eq(uid
, cred
->uid
))
305 free_uid(user
); /* for find_user() */
309 read_unlock(&tasklist_lock
);
316 * Unprivileged users may change the real gid to the effective gid
317 * or vice versa. (BSD-style)
319 * If you set the real gid at all, or set the effective gid to a value not
320 * equal to the real gid, then the saved gid is set to the new effective gid.
322 * This makes it possible for a setgid program to completely drop its
323 * privileges, which is often a useful assertion to make when you are doing
324 * a security audit over a program.
326 * The general idea is that a program which uses just setregid() will be
327 * 100% compatible with BSD. A program which uses just setgid() will be
328 * 100% compatible with POSIX with saved IDs.
330 * SMP: There are not races, the GIDs are checked only by filesystem
331 * operations (as far as semantic preservation is concerned).
333 #ifdef CONFIG_MULTIUSER
334 SYSCALL_DEFINE2(setregid
, gid_t
, rgid
, gid_t
, egid
)
336 struct user_namespace
*ns
= current_user_ns();
337 const struct cred
*old
;
342 krgid
= make_kgid(ns
, rgid
);
343 kegid
= make_kgid(ns
, egid
);
345 if ((rgid
!= (gid_t
) -1) && !gid_valid(krgid
))
347 if ((egid
!= (gid_t
) -1) && !gid_valid(kegid
))
350 new = prepare_creds();
353 old
= current_cred();
356 if (rgid
!= (gid_t
) -1) {
357 if (gid_eq(old
->gid
, krgid
) ||
358 gid_eq(old
->egid
, krgid
) ||
359 ns_capable(old
->user_ns
, CAP_SETGID
))
364 if (egid
!= (gid_t
) -1) {
365 if (gid_eq(old
->gid
, kegid
) ||
366 gid_eq(old
->egid
, kegid
) ||
367 gid_eq(old
->sgid
, kegid
) ||
368 ns_capable(old
->user_ns
, CAP_SETGID
))
374 if (rgid
!= (gid_t
) -1 ||
375 (egid
!= (gid_t
) -1 && !gid_eq(kegid
, old
->gid
)))
376 new->sgid
= new->egid
;
377 new->fsgid
= new->egid
;
379 return commit_creds(new);
387 * setgid() is implemented like SysV w/ SAVED_IDS
389 * SMP: Same implicit races as above.
391 SYSCALL_DEFINE1(setgid
, gid_t
, gid
)
393 struct user_namespace
*ns
= current_user_ns();
394 const struct cred
*old
;
399 kgid
= make_kgid(ns
, gid
);
400 if (!gid_valid(kgid
))
403 new = prepare_creds();
406 old
= current_cred();
409 if (ns_capable(old
->user_ns
, CAP_SETGID
))
410 new->gid
= new->egid
= new->sgid
= new->fsgid
= kgid
;
411 else if (gid_eq(kgid
, old
->gid
) || gid_eq(kgid
, old
->sgid
))
412 new->egid
= new->fsgid
= kgid
;
416 return commit_creds(new);
424 * change the user struct in a credentials set to match the new UID
426 static int set_user(struct cred
*new)
428 struct user_struct
*new_user
;
430 new_user
= alloc_uid(new->uid
);
435 * We don't fail in case of NPROC limit excess here because too many
436 * poorly written programs don't check set*uid() return code, assuming
437 * it never fails if called by root. We may still enforce NPROC limit
438 * for programs doing set*uid()+execve() by harmlessly deferring the
439 * failure to the execve() stage.
441 if (atomic_read(&new_user
->processes
) >= rlimit(RLIMIT_NPROC
) &&
442 new_user
!= INIT_USER
)
443 current
->flags
|= PF_NPROC_EXCEEDED
;
445 current
->flags
&= ~PF_NPROC_EXCEEDED
;
448 new->user
= new_user
;
453 * Unprivileged users may change the real uid to the effective uid
454 * or vice versa. (BSD-style)
456 * If you set the real uid at all, or set the effective uid to a value not
457 * equal to the real uid, then the saved uid is set to the new effective uid.
459 * This makes it possible for a setuid program to completely drop its
460 * privileges, which is often a useful assertion to make when you are doing
461 * a security audit over a program.
463 * The general idea is that a program which uses just setreuid() will be
464 * 100% compatible with BSD. A program which uses just setuid() will be
465 * 100% compatible with POSIX with saved IDs.
467 SYSCALL_DEFINE2(setreuid
, uid_t
, ruid
, uid_t
, euid
)
469 struct user_namespace
*ns
= current_user_ns();
470 const struct cred
*old
;
475 kruid
= make_kuid(ns
, ruid
);
476 keuid
= make_kuid(ns
, euid
);
478 if ((ruid
!= (uid_t
) -1) && !uid_valid(kruid
))
480 if ((euid
!= (uid_t
) -1) && !uid_valid(keuid
))
483 new = prepare_creds();
486 old
= current_cred();
489 if (ruid
!= (uid_t
) -1) {
491 if (!uid_eq(old
->uid
, kruid
) &&
492 !uid_eq(old
->euid
, kruid
) &&
493 !ns_capable(old
->user_ns
, CAP_SETUID
))
497 if (euid
!= (uid_t
) -1) {
499 if (!uid_eq(old
->uid
, keuid
) &&
500 !uid_eq(old
->euid
, keuid
) &&
501 !uid_eq(old
->suid
, keuid
) &&
502 !ns_capable(old
->user_ns
, CAP_SETUID
))
506 if (!uid_eq(new->uid
, old
->uid
)) {
507 retval
= set_user(new);
511 if (ruid
!= (uid_t
) -1 ||
512 (euid
!= (uid_t
) -1 && !uid_eq(keuid
, old
->uid
)))
513 new->suid
= new->euid
;
514 new->fsuid
= new->euid
;
516 retval
= security_task_fix_setuid(new, old
, LSM_SETID_RE
);
520 return commit_creds(new);
528 * setuid() is implemented like SysV with SAVED_IDS
530 * Note that SAVED_ID's is deficient in that a setuid root program
531 * like sendmail, for example, cannot set its uid to be a normal
532 * user and then switch back, because if you're root, setuid() sets
533 * the saved uid too. If you don't like this, blame the bright people
534 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
535 * will allow a root program to temporarily drop privileges and be able to
536 * regain them by swapping the real and effective uid.
538 SYSCALL_DEFINE1(setuid
, uid_t
, uid
)
540 struct user_namespace
*ns
= current_user_ns();
541 const struct cred
*old
;
546 kuid
= make_kuid(ns
, uid
);
547 if (!uid_valid(kuid
))
550 new = prepare_creds();
553 old
= current_cred();
556 if (ns_capable(old
->user_ns
, CAP_SETUID
)) {
557 new->suid
= new->uid
= kuid
;
558 if (!uid_eq(kuid
, old
->uid
)) {
559 retval
= set_user(new);
563 } else if (!uid_eq(kuid
, old
->uid
) && !uid_eq(kuid
, new->suid
)) {
567 new->fsuid
= new->euid
= kuid
;
569 retval
= security_task_fix_setuid(new, old
, LSM_SETID_ID
);
573 return commit_creds(new);
582 * This function implements a generic ability to update ruid, euid,
583 * and suid. This allows you to implement the 4.4 compatible seteuid().
585 SYSCALL_DEFINE3(setresuid
, uid_t
, ruid
, uid_t
, euid
, uid_t
, suid
)
587 struct user_namespace
*ns
= current_user_ns();
588 const struct cred
*old
;
591 kuid_t kruid
, keuid
, ksuid
;
593 kruid
= make_kuid(ns
, ruid
);
594 keuid
= make_kuid(ns
, euid
);
595 ksuid
= make_kuid(ns
, suid
);
597 if ((ruid
!= (uid_t
) -1) && !uid_valid(kruid
))
600 if ((euid
!= (uid_t
) -1) && !uid_valid(keuid
))
603 if ((suid
!= (uid_t
) -1) && !uid_valid(ksuid
))
606 new = prepare_creds();
610 old
= current_cred();
613 if (!ns_capable(old
->user_ns
, CAP_SETUID
)) {
614 if (ruid
!= (uid_t
) -1 && !uid_eq(kruid
, old
->uid
) &&
615 !uid_eq(kruid
, old
->euid
) && !uid_eq(kruid
, old
->suid
))
617 if (euid
!= (uid_t
) -1 && !uid_eq(keuid
, old
->uid
) &&
618 !uid_eq(keuid
, old
->euid
) && !uid_eq(keuid
, old
->suid
))
620 if (suid
!= (uid_t
) -1 && !uid_eq(ksuid
, old
->uid
) &&
621 !uid_eq(ksuid
, old
->euid
) && !uid_eq(ksuid
, old
->suid
))
625 if (ruid
!= (uid_t
) -1) {
627 if (!uid_eq(kruid
, old
->uid
)) {
628 retval
= set_user(new);
633 if (euid
!= (uid_t
) -1)
635 if (suid
!= (uid_t
) -1)
637 new->fsuid
= new->euid
;
639 retval
= security_task_fix_setuid(new, old
, LSM_SETID_RES
);
643 return commit_creds(new);
650 SYSCALL_DEFINE3(getresuid
, uid_t __user
*, ruidp
, uid_t __user
*, euidp
, uid_t __user
*, suidp
)
652 const struct cred
*cred
= current_cred();
654 uid_t ruid
, euid
, suid
;
656 ruid
= from_kuid_munged(cred
->user_ns
, cred
->uid
);
657 euid
= from_kuid_munged(cred
->user_ns
, cred
->euid
);
658 suid
= from_kuid_munged(cred
->user_ns
, cred
->suid
);
660 retval
= put_user(ruid
, ruidp
);
662 retval
= put_user(euid
, euidp
);
664 return put_user(suid
, suidp
);
670 * Same as above, but for rgid, egid, sgid.
672 SYSCALL_DEFINE3(setresgid
, gid_t
, rgid
, gid_t
, egid
, gid_t
, sgid
)
674 struct user_namespace
*ns
= current_user_ns();
675 const struct cred
*old
;
678 kgid_t krgid
, kegid
, ksgid
;
680 krgid
= make_kgid(ns
, rgid
);
681 kegid
= make_kgid(ns
, egid
);
682 ksgid
= make_kgid(ns
, sgid
);
684 if ((rgid
!= (gid_t
) -1) && !gid_valid(krgid
))
686 if ((egid
!= (gid_t
) -1) && !gid_valid(kegid
))
688 if ((sgid
!= (gid_t
) -1) && !gid_valid(ksgid
))
691 new = prepare_creds();
694 old
= current_cred();
697 if (!ns_capable(old
->user_ns
, CAP_SETGID
)) {
698 if (rgid
!= (gid_t
) -1 && !gid_eq(krgid
, old
->gid
) &&
699 !gid_eq(krgid
, old
->egid
) && !gid_eq(krgid
, old
->sgid
))
701 if (egid
!= (gid_t
) -1 && !gid_eq(kegid
, old
->gid
) &&
702 !gid_eq(kegid
, old
->egid
) && !gid_eq(kegid
, old
->sgid
))
704 if (sgid
!= (gid_t
) -1 && !gid_eq(ksgid
, old
->gid
) &&
705 !gid_eq(ksgid
, old
->egid
) && !gid_eq(ksgid
, old
->sgid
))
709 if (rgid
!= (gid_t
) -1)
711 if (egid
!= (gid_t
) -1)
713 if (sgid
!= (gid_t
) -1)
715 new->fsgid
= new->egid
;
717 return commit_creds(new);
724 SYSCALL_DEFINE3(getresgid
, gid_t __user
*, rgidp
, gid_t __user
*, egidp
, gid_t __user
*, sgidp
)
726 const struct cred
*cred
= current_cred();
728 gid_t rgid
, egid
, sgid
;
730 rgid
= from_kgid_munged(cred
->user_ns
, cred
->gid
);
731 egid
= from_kgid_munged(cred
->user_ns
, cred
->egid
);
732 sgid
= from_kgid_munged(cred
->user_ns
, cred
->sgid
);
734 retval
= put_user(rgid
, rgidp
);
736 retval
= put_user(egid
, egidp
);
738 retval
= put_user(sgid
, sgidp
);
746 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
747 * is used for "access()" and for the NFS daemon (letting nfsd stay at
748 * whatever uid it wants to). It normally shadows "euid", except when
749 * explicitly set by setfsuid() or for access..
751 SYSCALL_DEFINE1(setfsuid
, uid_t
, uid
)
753 const struct cred
*old
;
758 old
= current_cred();
759 old_fsuid
= from_kuid_munged(old
->user_ns
, old
->fsuid
);
761 kuid
= make_kuid(old
->user_ns
, uid
);
762 if (!uid_valid(kuid
))
765 new = prepare_creds();
769 if (uid_eq(kuid
, old
->uid
) || uid_eq(kuid
, old
->euid
) ||
770 uid_eq(kuid
, old
->suid
) || uid_eq(kuid
, old
->fsuid
) ||
771 ns_capable(old
->user_ns
, CAP_SETUID
)) {
772 if (!uid_eq(kuid
, old
->fsuid
)) {
774 if (security_task_fix_setuid(new, old
, LSM_SETID_FS
) == 0)
788 * Samma på svenska..
790 SYSCALL_DEFINE1(setfsgid
, gid_t
, gid
)
792 const struct cred
*old
;
797 old
= current_cred();
798 old_fsgid
= from_kgid_munged(old
->user_ns
, old
->fsgid
);
800 kgid
= make_kgid(old
->user_ns
, gid
);
801 if (!gid_valid(kgid
))
804 new = prepare_creds();
808 if (gid_eq(kgid
, old
->gid
) || gid_eq(kgid
, old
->egid
) ||
809 gid_eq(kgid
, old
->sgid
) || gid_eq(kgid
, old
->fsgid
) ||
810 ns_capable(old
->user_ns
, CAP_SETGID
)) {
811 if (!gid_eq(kgid
, old
->fsgid
)) {
824 #endif /* CONFIG_MULTIUSER */
827 * sys_getpid - return the thread group id of the current process
829 * Note, despite the name, this returns the tgid not the pid. The tgid and
830 * the pid are identical unless CLONE_THREAD was specified on clone() in
831 * which case the tgid is the same in all threads of the same group.
833 * This is SMP safe as current->tgid does not change.
835 SYSCALL_DEFINE0(getpid
)
837 return task_tgid_vnr(current
);
840 /* Thread ID - the internal kernel "pid" */
841 SYSCALL_DEFINE0(gettid
)
843 return task_pid_vnr(current
);
847 * Accessing ->real_parent is not SMP-safe, it could
848 * change from under us. However, we can use a stale
849 * value of ->real_parent under rcu_read_lock(), see
850 * release_task()->call_rcu(delayed_put_task_struct).
852 SYSCALL_DEFINE0(getppid
)
857 pid
= task_tgid_vnr(rcu_dereference(current
->real_parent
));
863 SYSCALL_DEFINE0(getuid
)
865 /* Only we change this so SMP safe */
866 return from_kuid_munged(current_user_ns(), current_uid());
869 SYSCALL_DEFINE0(geteuid
)
871 /* Only we change this so SMP safe */
872 return from_kuid_munged(current_user_ns(), current_euid());
875 SYSCALL_DEFINE0(getgid
)
877 /* Only we change this so SMP safe */
878 return from_kgid_munged(current_user_ns(), current_gid());
881 SYSCALL_DEFINE0(getegid
)
883 /* Only we change this so SMP safe */
884 return from_kgid_munged(current_user_ns(), current_egid());
887 void do_sys_times(struct tms
*tms
)
889 u64 tgutime
, tgstime
, cutime
, cstime
;
891 thread_group_cputime_adjusted(current
, &tgutime
, &tgstime
);
892 cutime
= current
->signal
->cutime
;
893 cstime
= current
->signal
->cstime
;
894 tms
->tms_utime
= nsec_to_clock_t(tgutime
);
895 tms
->tms_stime
= nsec_to_clock_t(tgstime
);
896 tms
->tms_cutime
= nsec_to_clock_t(cutime
);
897 tms
->tms_cstime
= nsec_to_clock_t(cstime
);
900 SYSCALL_DEFINE1(times
, struct tms __user
*, tbuf
)
906 if (copy_to_user(tbuf
, &tmp
, sizeof(struct tms
)))
909 force_successful_syscall_return();
910 return (long) jiffies_64_to_clock_t(get_jiffies_64());
914 * This needs some heavy checking ...
915 * I just haven't the stomach for it. I also don't fully
916 * understand sessions/pgrp etc. Let somebody who does explain it.
918 * OK, I think I have the protection semantics right.... this is really
919 * only important on a multi-user system anyway, to make sure one user
920 * can't send a signal to a process owned by another. -TYT, 12/12/91
922 * !PF_FORKNOEXEC check to conform completely to POSIX.
924 SYSCALL_DEFINE2(setpgid
, pid_t
, pid
, pid_t
, pgid
)
926 struct task_struct
*p
;
927 struct task_struct
*group_leader
= current
->group_leader
;
932 pid
= task_pid_vnr(group_leader
);
939 /* From this point forward we keep holding onto the tasklist lock
940 * so that our parent does not change from under us. -DaveM
942 write_lock_irq(&tasklist_lock
);
945 p
= find_task_by_vpid(pid
);
950 if (!thread_group_leader(p
))
953 if (same_thread_group(p
->real_parent
, group_leader
)) {
955 if (task_session(p
) != task_session(group_leader
))
958 if (!(p
->flags
& PF_FORKNOEXEC
))
962 if (p
!= group_leader
)
967 if (p
->signal
->leader
)
972 struct task_struct
*g
;
974 pgrp
= find_vpid(pgid
);
975 g
= pid_task(pgrp
, PIDTYPE_PGID
);
976 if (!g
|| task_session(g
) != task_session(group_leader
))
980 err
= security_task_setpgid(p
, pgid
);
984 if (task_pgrp(p
) != pgrp
)
985 change_pid(p
, PIDTYPE_PGID
, pgrp
);
989 /* All paths lead to here, thus we are safe. -DaveM */
990 write_unlock_irq(&tasklist_lock
);
995 SYSCALL_DEFINE1(getpgid
, pid_t
, pid
)
997 struct task_struct
*p
;
1003 grp
= task_pgrp(current
);
1006 p
= find_task_by_vpid(pid
);
1013 retval
= security_task_getpgid(p
);
1017 retval
= pid_vnr(grp
);
1023 #ifdef __ARCH_WANT_SYS_GETPGRP
1025 SYSCALL_DEFINE0(getpgrp
)
1027 return sys_getpgid(0);
1032 SYSCALL_DEFINE1(getsid
, pid_t
, pid
)
1034 struct task_struct
*p
;
1040 sid
= task_session(current
);
1043 p
= find_task_by_vpid(pid
);
1046 sid
= task_session(p
);
1050 retval
= security_task_getsid(p
);
1054 retval
= pid_vnr(sid
);
1060 static void set_special_pids(struct pid
*pid
)
1062 struct task_struct
*curr
= current
->group_leader
;
1064 if (task_session(curr
) != pid
)
1065 change_pid(curr
, PIDTYPE_SID
, pid
);
1067 if (task_pgrp(curr
) != pid
)
1068 change_pid(curr
, PIDTYPE_PGID
, pid
);
1071 SYSCALL_DEFINE0(setsid
)
1073 struct task_struct
*group_leader
= current
->group_leader
;
1074 struct pid
*sid
= task_pid(group_leader
);
1075 pid_t session
= pid_vnr(sid
);
1078 write_lock_irq(&tasklist_lock
);
1079 /* Fail if I am already a session leader */
1080 if (group_leader
->signal
->leader
)
1083 /* Fail if a process group id already exists that equals the
1084 * proposed session id.
1086 if (pid_task(sid
, PIDTYPE_PGID
))
1089 group_leader
->signal
->leader
= 1;
1090 set_special_pids(sid
);
1092 proc_clear_tty(group_leader
);
1096 write_unlock_irq(&tasklist_lock
);
1098 proc_sid_connector(group_leader
);
1099 sched_autogroup_create_attach(group_leader
);
1104 DECLARE_RWSEM(uts_sem
);
1106 #ifdef COMPAT_UTS_MACHINE
1107 #define override_architecture(name) \
1108 (personality(current->personality) == PER_LINUX32 && \
1109 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1110 sizeof(COMPAT_UTS_MACHINE)))
1112 #define override_architecture(name) 0
1116 * Work around broken programs that cannot handle "Linux 3.0".
1117 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1118 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1120 static int override_release(char __user
*release
, size_t len
)
1124 if (current
->personality
& UNAME26
) {
1125 const char *rest
= UTS_RELEASE
;
1126 char buf
[65] = { 0 };
1132 if (*rest
== '.' && ++ndots
>= 3)
1134 if (!isdigit(*rest
) && *rest
!= '.')
1138 v
= ((LINUX_VERSION_CODE
>> 8) & 0xff) + 60;
1139 copy
= clamp_t(size_t, len
, 1, sizeof(buf
));
1140 copy
= scnprintf(buf
, copy
, "2.6.%u%s", v
, rest
);
1141 ret
= copy_to_user(release
, buf
, copy
+ 1);
1146 SYSCALL_DEFINE1(newuname
, struct new_utsname __user
*, name
)
1150 down_read(&uts_sem
);
1151 if (copy_to_user(name
, utsname(), sizeof *name
))
1155 if (!errno
&& override_release(name
->release
, sizeof(name
->release
)))
1157 if (!errno
&& override_architecture(name
))
1162 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1166 SYSCALL_DEFINE1(uname
, struct old_utsname __user
*, name
)
1173 down_read(&uts_sem
);
1174 if (copy_to_user(name
, utsname(), sizeof(*name
)))
1178 if (!error
&& override_release(name
->release
, sizeof(name
->release
)))
1180 if (!error
&& override_architecture(name
))
1185 SYSCALL_DEFINE1(olduname
, struct oldold_utsname __user
*, name
)
1191 if (!access_ok(VERIFY_WRITE
, name
, sizeof(struct oldold_utsname
)))
1194 down_read(&uts_sem
);
1195 error
= __copy_to_user(&name
->sysname
, &utsname()->sysname
,
1197 error
|= __put_user(0, name
->sysname
+ __OLD_UTS_LEN
);
1198 error
|= __copy_to_user(&name
->nodename
, &utsname()->nodename
,
1200 error
|= __put_user(0, name
->nodename
+ __OLD_UTS_LEN
);
1201 error
|= __copy_to_user(&name
->release
, &utsname()->release
,
1203 error
|= __put_user(0, name
->release
+ __OLD_UTS_LEN
);
1204 error
|= __copy_to_user(&name
->version
, &utsname()->version
,
1206 error
|= __put_user(0, name
->version
+ __OLD_UTS_LEN
);
1207 error
|= __copy_to_user(&name
->machine
, &utsname()->machine
,
1209 error
|= __put_user(0, name
->machine
+ __OLD_UTS_LEN
);
1212 if (!error
&& override_architecture(name
))
1214 if (!error
&& override_release(name
->release
, sizeof(name
->release
)))
1216 return error
? -EFAULT
: 0;
1220 SYSCALL_DEFINE2(sethostname
, char __user
*, name
, int, len
)
1223 char tmp
[__NEW_UTS_LEN
];
1225 if (!ns_capable(current
->nsproxy
->uts_ns
->user_ns
, CAP_SYS_ADMIN
))
1228 if (len
< 0 || len
> __NEW_UTS_LEN
)
1230 down_write(&uts_sem
);
1232 if (!copy_from_user(tmp
, name
, len
)) {
1233 struct new_utsname
*u
= utsname();
1235 memcpy(u
->nodename
, tmp
, len
);
1236 memset(u
->nodename
+ len
, 0, sizeof(u
->nodename
) - len
);
1238 uts_proc_notify(UTS_PROC_HOSTNAME
);
1244 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1246 SYSCALL_DEFINE2(gethostname
, char __user
*, name
, int, len
)
1249 struct new_utsname
*u
;
1253 down_read(&uts_sem
);
1255 i
= 1 + strlen(u
->nodename
);
1259 if (copy_to_user(name
, u
->nodename
, i
))
1268 * Only setdomainname; getdomainname can be implemented by calling
1271 SYSCALL_DEFINE2(setdomainname
, char __user
*, name
, int, len
)
1274 char tmp
[__NEW_UTS_LEN
];
1276 if (!ns_capable(current
->nsproxy
->uts_ns
->user_ns
, CAP_SYS_ADMIN
))
1278 if (len
< 0 || len
> __NEW_UTS_LEN
)
1281 down_write(&uts_sem
);
1283 if (!copy_from_user(tmp
, name
, len
)) {
1284 struct new_utsname
*u
= utsname();
1286 memcpy(u
->domainname
, tmp
, len
);
1287 memset(u
->domainname
+ len
, 0, sizeof(u
->domainname
) - len
);
1289 uts_proc_notify(UTS_PROC_DOMAINNAME
);
1295 SYSCALL_DEFINE2(getrlimit
, unsigned int, resource
, struct rlimit __user
*, rlim
)
1297 struct rlimit value
;
1300 ret
= do_prlimit(current
, resource
, NULL
, &value
);
1302 ret
= copy_to_user(rlim
, &value
, sizeof(*rlim
)) ? -EFAULT
: 0;
1307 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1310 * Back compatibility for getrlimit. Needed for some apps.
1312 SYSCALL_DEFINE2(old_getrlimit
, unsigned int, resource
,
1313 struct rlimit __user
*, rlim
)
1316 if (resource
>= RLIM_NLIMITS
)
1319 task_lock(current
->group_leader
);
1320 x
= current
->signal
->rlim
[resource
];
1321 task_unlock(current
->group_leader
);
1322 if (x
.rlim_cur
> 0x7FFFFFFF)
1323 x
.rlim_cur
= 0x7FFFFFFF;
1324 if (x
.rlim_max
> 0x7FFFFFFF)
1325 x
.rlim_max
= 0x7FFFFFFF;
1326 return copy_to_user(rlim
, &x
, sizeof(x
)) ? -EFAULT
: 0;
1331 static inline bool rlim64_is_infinity(__u64 rlim64
)
1333 #if BITS_PER_LONG < 64
1334 return rlim64
>= ULONG_MAX
;
1336 return rlim64
== RLIM64_INFINITY
;
1340 static void rlim_to_rlim64(const struct rlimit
*rlim
, struct rlimit64
*rlim64
)
1342 if (rlim
->rlim_cur
== RLIM_INFINITY
)
1343 rlim64
->rlim_cur
= RLIM64_INFINITY
;
1345 rlim64
->rlim_cur
= rlim
->rlim_cur
;
1346 if (rlim
->rlim_max
== RLIM_INFINITY
)
1347 rlim64
->rlim_max
= RLIM64_INFINITY
;
1349 rlim64
->rlim_max
= rlim
->rlim_max
;
1352 static void rlim64_to_rlim(const struct rlimit64
*rlim64
, struct rlimit
*rlim
)
1354 if (rlim64_is_infinity(rlim64
->rlim_cur
))
1355 rlim
->rlim_cur
= RLIM_INFINITY
;
1357 rlim
->rlim_cur
= (unsigned long)rlim64
->rlim_cur
;
1358 if (rlim64_is_infinity(rlim64
->rlim_max
))
1359 rlim
->rlim_max
= RLIM_INFINITY
;
1361 rlim
->rlim_max
= (unsigned long)rlim64
->rlim_max
;
1364 /* make sure you are allowed to change @tsk limits before calling this */
1365 int do_prlimit(struct task_struct
*tsk
, unsigned int resource
,
1366 struct rlimit
*new_rlim
, struct rlimit
*old_rlim
)
1368 struct rlimit
*rlim
;
1371 if (resource
>= RLIM_NLIMITS
)
1374 if (new_rlim
->rlim_cur
> new_rlim
->rlim_max
)
1376 if (resource
== RLIMIT_NOFILE
&&
1377 new_rlim
->rlim_max
> sysctl_nr_open
)
1381 /* protect tsk->signal and tsk->sighand from disappearing */
1382 read_lock(&tasklist_lock
);
1383 if (!tsk
->sighand
) {
1388 rlim
= tsk
->signal
->rlim
+ resource
;
1389 task_lock(tsk
->group_leader
);
1391 /* Keep the capable check against init_user_ns until
1392 cgroups can contain all limits */
1393 if (new_rlim
->rlim_max
> rlim
->rlim_max
&&
1394 !capable(CAP_SYS_RESOURCE
))
1397 retval
= security_task_setrlimit(tsk
->group_leader
,
1398 resource
, new_rlim
);
1399 if (resource
== RLIMIT_CPU
&& new_rlim
->rlim_cur
== 0) {
1401 * The caller is asking for an immediate RLIMIT_CPU
1402 * expiry. But we use the zero value to mean "it was
1403 * never set". So let's cheat and make it one second
1406 new_rlim
->rlim_cur
= 1;
1415 task_unlock(tsk
->group_leader
);
1418 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1419 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1420 * very long-standing error, and fixing it now risks breakage of
1421 * applications, so we live with it
1423 if (!retval
&& new_rlim
&& resource
== RLIMIT_CPU
&&
1424 new_rlim
->rlim_cur
!= RLIM_INFINITY
&&
1425 IS_ENABLED(CONFIG_POSIX_TIMERS
))
1426 update_rlimit_cpu(tsk
, new_rlim
->rlim_cur
);
1428 read_unlock(&tasklist_lock
);
1432 /* rcu lock must be held */
1433 static int check_prlimit_permission(struct task_struct
*task
)
1435 const struct cred
*cred
= current_cred(), *tcred
;
1437 if (current
== task
)
1440 tcred
= __task_cred(task
);
1441 if (uid_eq(cred
->uid
, tcred
->euid
) &&
1442 uid_eq(cred
->uid
, tcred
->suid
) &&
1443 uid_eq(cred
->uid
, tcred
->uid
) &&
1444 gid_eq(cred
->gid
, tcred
->egid
) &&
1445 gid_eq(cred
->gid
, tcred
->sgid
) &&
1446 gid_eq(cred
->gid
, tcred
->gid
))
1448 if (ns_capable(tcred
->user_ns
, CAP_SYS_RESOURCE
))
1454 SYSCALL_DEFINE4(prlimit64
, pid_t
, pid
, unsigned int, resource
,
1455 const struct rlimit64 __user
*, new_rlim
,
1456 struct rlimit64 __user
*, old_rlim
)
1458 struct rlimit64 old64
, new64
;
1459 struct rlimit old
, new;
1460 struct task_struct
*tsk
;
1464 if (copy_from_user(&new64
, new_rlim
, sizeof(new64
)))
1466 rlim64_to_rlim(&new64
, &new);
1470 tsk
= pid
? find_task_by_vpid(pid
) : current
;
1475 ret
= check_prlimit_permission(tsk
);
1480 get_task_struct(tsk
);
1483 ret
= do_prlimit(tsk
, resource
, new_rlim
? &new : NULL
,
1484 old_rlim
? &old
: NULL
);
1486 if (!ret
&& old_rlim
) {
1487 rlim_to_rlim64(&old
, &old64
);
1488 if (copy_to_user(old_rlim
, &old64
, sizeof(old64
)))
1492 put_task_struct(tsk
);
1496 SYSCALL_DEFINE2(setrlimit
, unsigned int, resource
, struct rlimit __user
*, rlim
)
1498 struct rlimit new_rlim
;
1500 if (copy_from_user(&new_rlim
, rlim
, sizeof(*rlim
)))
1502 return do_prlimit(current
, resource
, &new_rlim
, NULL
);
1506 * It would make sense to put struct rusage in the task_struct,
1507 * except that would make the task_struct be *really big*. After
1508 * task_struct gets moved into malloc'ed memory, it would
1509 * make sense to do this. It will make moving the rest of the information
1510 * a lot simpler! (Which we're not doing right now because we're not
1511 * measuring them yet).
1513 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1514 * races with threads incrementing their own counters. But since word
1515 * reads are atomic, we either get new values or old values and we don't
1516 * care which for the sums. We always take the siglock to protect reading
1517 * the c* fields from p->signal from races with exit.c updating those
1518 * fields when reaping, so a sample either gets all the additions of a
1519 * given child after it's reaped, or none so this sample is before reaping.
1522 * We need to take the siglock for CHILDEREN, SELF and BOTH
1523 * for the cases current multithreaded, non-current single threaded
1524 * non-current multithreaded. Thread traversal is now safe with
1526 * Strictly speaking, we donot need to take the siglock if we are current and
1527 * single threaded, as no one else can take our signal_struct away, no one
1528 * else can reap the children to update signal->c* counters, and no one else
1529 * can race with the signal-> fields. If we do not take any lock, the
1530 * signal-> fields could be read out of order while another thread was just
1531 * exiting. So we should place a read memory barrier when we avoid the lock.
1532 * On the writer side, write memory barrier is implied in __exit_signal
1533 * as __exit_signal releases the siglock spinlock after updating the signal->
1534 * fields. But we don't do this yet to keep things simple.
1538 static void accumulate_thread_rusage(struct task_struct
*t
, struct rusage
*r
)
1540 r
->ru_nvcsw
+= t
->nvcsw
;
1541 r
->ru_nivcsw
+= t
->nivcsw
;
1542 r
->ru_minflt
+= t
->min_flt
;
1543 r
->ru_majflt
+= t
->maj_flt
;
1544 r
->ru_inblock
+= task_io_get_inblock(t
);
1545 r
->ru_oublock
+= task_io_get_oublock(t
);
1548 static void k_getrusage(struct task_struct
*p
, int who
, struct rusage
*r
)
1550 struct task_struct
*t
;
1551 unsigned long flags
;
1552 u64 tgutime
, tgstime
, utime
, stime
;
1553 unsigned long maxrss
= 0;
1555 memset((char *)r
, 0, sizeof (*r
));
1558 if (who
== RUSAGE_THREAD
) {
1559 task_cputime_adjusted(current
, &utime
, &stime
);
1560 accumulate_thread_rusage(p
, r
);
1561 maxrss
= p
->signal
->maxrss
;
1565 if (!lock_task_sighand(p
, &flags
))
1570 case RUSAGE_CHILDREN
:
1571 utime
= p
->signal
->cutime
;
1572 stime
= p
->signal
->cstime
;
1573 r
->ru_nvcsw
= p
->signal
->cnvcsw
;
1574 r
->ru_nivcsw
= p
->signal
->cnivcsw
;
1575 r
->ru_minflt
= p
->signal
->cmin_flt
;
1576 r
->ru_majflt
= p
->signal
->cmaj_flt
;
1577 r
->ru_inblock
= p
->signal
->cinblock
;
1578 r
->ru_oublock
= p
->signal
->coublock
;
1579 maxrss
= p
->signal
->cmaxrss
;
1581 if (who
== RUSAGE_CHILDREN
)
1585 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1588 r
->ru_nvcsw
+= p
->signal
->nvcsw
;
1589 r
->ru_nivcsw
+= p
->signal
->nivcsw
;
1590 r
->ru_minflt
+= p
->signal
->min_flt
;
1591 r
->ru_majflt
+= p
->signal
->maj_flt
;
1592 r
->ru_inblock
+= p
->signal
->inblock
;
1593 r
->ru_oublock
+= p
->signal
->oublock
;
1594 if (maxrss
< p
->signal
->maxrss
)
1595 maxrss
= p
->signal
->maxrss
;
1598 accumulate_thread_rusage(t
, r
);
1599 } while_each_thread(p
, t
);
1605 unlock_task_sighand(p
, &flags
);
1608 r
->ru_utime
= ns_to_timeval(utime
);
1609 r
->ru_stime
= ns_to_timeval(stime
);
1611 if (who
!= RUSAGE_CHILDREN
) {
1612 struct mm_struct
*mm
= get_task_mm(p
);
1615 setmax_mm_hiwater_rss(&maxrss
, mm
);
1619 r
->ru_maxrss
= maxrss
* (PAGE_SIZE
/ 1024); /* convert pages to KBs */
1622 int getrusage(struct task_struct
*p
, int who
, struct rusage __user
*ru
)
1626 k_getrusage(p
, who
, &r
);
1627 return copy_to_user(ru
, &r
, sizeof(r
)) ? -EFAULT
: 0;
1630 SYSCALL_DEFINE2(getrusage
, int, who
, struct rusage __user
*, ru
)
1632 if (who
!= RUSAGE_SELF
&& who
!= RUSAGE_CHILDREN
&&
1633 who
!= RUSAGE_THREAD
)
1635 return getrusage(current
, who
, ru
);
1638 #ifdef CONFIG_COMPAT
1639 COMPAT_SYSCALL_DEFINE2(getrusage
, int, who
, struct compat_rusage __user
*, ru
)
1643 if (who
!= RUSAGE_SELF
&& who
!= RUSAGE_CHILDREN
&&
1644 who
!= RUSAGE_THREAD
)
1647 k_getrusage(current
, who
, &r
);
1648 return put_compat_rusage(&r
, ru
);
1652 SYSCALL_DEFINE1(umask
, int, mask
)
1654 mask
= xchg(¤t
->fs
->umask
, mask
& S_IRWXUGO
);
1658 static int prctl_set_mm_exe_file(struct mm_struct
*mm
, unsigned int fd
)
1661 struct file
*old_exe
, *exe_file
;
1662 struct inode
*inode
;
1669 inode
= file_inode(exe
.file
);
1672 * Because the original mm->exe_file points to executable file, make
1673 * sure that this one is executable as well, to avoid breaking an
1677 if (!S_ISREG(inode
->i_mode
) || path_noexec(&exe
.file
->f_path
))
1680 err
= inode_permission(inode
, MAY_EXEC
);
1685 * Forbid mm->exe_file change if old file still mapped.
1687 exe_file
= get_mm_exe_file(mm
);
1690 struct vm_area_struct
*vma
;
1692 down_read(&mm
->mmap_sem
);
1693 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
1696 if (path_equal(&vma
->vm_file
->f_path
,
1701 up_read(&mm
->mmap_sem
);
1706 /* set the new file, lockless */
1708 old_exe
= xchg(&mm
->exe_file
, exe
.file
);
1715 up_read(&mm
->mmap_sem
);
1721 * WARNING: we don't require any capability here so be very careful
1722 * in what is allowed for modification from userspace.
1724 static int validate_prctl_map(struct prctl_mm_map
*prctl_map
)
1726 unsigned long mmap_max_addr
= TASK_SIZE
;
1727 struct mm_struct
*mm
= current
->mm
;
1728 int error
= -EINVAL
, i
;
1730 static const unsigned char offsets
[] = {
1731 offsetof(struct prctl_mm_map
, start_code
),
1732 offsetof(struct prctl_mm_map
, end_code
),
1733 offsetof(struct prctl_mm_map
, start_data
),
1734 offsetof(struct prctl_mm_map
, end_data
),
1735 offsetof(struct prctl_mm_map
, start_brk
),
1736 offsetof(struct prctl_mm_map
, brk
),
1737 offsetof(struct prctl_mm_map
, start_stack
),
1738 offsetof(struct prctl_mm_map
, arg_start
),
1739 offsetof(struct prctl_mm_map
, arg_end
),
1740 offsetof(struct prctl_mm_map
, env_start
),
1741 offsetof(struct prctl_mm_map
, env_end
),
1745 * Make sure the members are not somewhere outside
1746 * of allowed address space.
1748 for (i
= 0; i
< ARRAY_SIZE(offsets
); i
++) {
1749 u64 val
= *(u64
*)((char *)prctl_map
+ offsets
[i
]);
1751 if ((unsigned long)val
>= mmap_max_addr
||
1752 (unsigned long)val
< mmap_min_addr
)
1757 * Make sure the pairs are ordered.
1759 #define __prctl_check_order(__m1, __op, __m2) \
1760 ((unsigned long)prctl_map->__m1 __op \
1761 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1762 error
= __prctl_check_order(start_code
, <, end_code
);
1763 error
|= __prctl_check_order(start_data
, <, end_data
);
1764 error
|= __prctl_check_order(start_brk
, <=, brk
);
1765 error
|= __prctl_check_order(arg_start
, <=, arg_end
);
1766 error
|= __prctl_check_order(env_start
, <=, env_end
);
1769 #undef __prctl_check_order
1774 * @brk should be after @end_data in traditional maps.
1776 if (prctl_map
->start_brk
<= prctl_map
->end_data
||
1777 prctl_map
->brk
<= prctl_map
->end_data
)
1781 * Neither we should allow to override limits if they set.
1783 if (check_data_rlimit(rlimit(RLIMIT_DATA
), prctl_map
->brk
,
1784 prctl_map
->start_brk
, prctl_map
->end_data
,
1785 prctl_map
->start_data
))
1789 * Someone is trying to cheat the auxv vector.
1791 if (prctl_map
->auxv_size
) {
1792 if (!prctl_map
->auxv
|| prctl_map
->auxv_size
> sizeof(mm
->saved_auxv
))
1797 * Finally, make sure the caller has the rights to
1798 * change /proc/pid/exe link: only local root should
1801 if (prctl_map
->exe_fd
!= (u32
)-1) {
1802 struct user_namespace
*ns
= current_user_ns();
1803 const struct cred
*cred
= current_cred();
1805 if (!uid_eq(cred
->uid
, make_kuid(ns
, 0)) ||
1806 !gid_eq(cred
->gid
, make_kgid(ns
, 0)))
1815 #ifdef CONFIG_CHECKPOINT_RESTORE
1816 static int prctl_set_mm_map(int opt
, const void __user
*addr
, unsigned long data_size
)
1818 struct prctl_mm_map prctl_map
= { .exe_fd
= (u32
)-1, };
1819 unsigned long user_auxv
[AT_VECTOR_SIZE
];
1820 struct mm_struct
*mm
= current
->mm
;
1823 BUILD_BUG_ON(sizeof(user_auxv
) != sizeof(mm
->saved_auxv
));
1824 BUILD_BUG_ON(sizeof(struct prctl_mm_map
) > 256);
1826 if (opt
== PR_SET_MM_MAP_SIZE
)
1827 return put_user((unsigned int)sizeof(prctl_map
),
1828 (unsigned int __user
*)addr
);
1830 if (data_size
!= sizeof(prctl_map
))
1833 if (copy_from_user(&prctl_map
, addr
, sizeof(prctl_map
)))
1836 error
= validate_prctl_map(&prctl_map
);
1840 if (prctl_map
.auxv_size
) {
1841 memset(user_auxv
, 0, sizeof(user_auxv
));
1842 if (copy_from_user(user_auxv
,
1843 (const void __user
*)prctl_map
.auxv
,
1844 prctl_map
.auxv_size
))
1847 /* Last entry must be AT_NULL as specification requires */
1848 user_auxv
[AT_VECTOR_SIZE
- 2] = AT_NULL
;
1849 user_auxv
[AT_VECTOR_SIZE
- 1] = AT_NULL
;
1852 if (prctl_map
.exe_fd
!= (u32
)-1) {
1853 error
= prctl_set_mm_exe_file(mm
, prctl_map
.exe_fd
);
1858 down_write(&mm
->mmap_sem
);
1861 * We don't validate if these members are pointing to
1862 * real present VMAs because application may have correspond
1863 * VMAs already unmapped and kernel uses these members for statistics
1864 * output in procfs mostly, except
1866 * - @start_brk/@brk which are used in do_brk but kernel lookups
1867 * for VMAs when updating these memvers so anything wrong written
1868 * here cause kernel to swear at userspace program but won't lead
1869 * to any problem in kernel itself
1872 mm
->start_code
= prctl_map
.start_code
;
1873 mm
->end_code
= prctl_map
.end_code
;
1874 mm
->start_data
= prctl_map
.start_data
;
1875 mm
->end_data
= prctl_map
.end_data
;
1876 mm
->start_brk
= prctl_map
.start_brk
;
1877 mm
->brk
= prctl_map
.brk
;
1878 mm
->start_stack
= prctl_map
.start_stack
;
1879 mm
->arg_start
= prctl_map
.arg_start
;
1880 mm
->arg_end
= prctl_map
.arg_end
;
1881 mm
->env_start
= prctl_map
.env_start
;
1882 mm
->env_end
= prctl_map
.env_end
;
1885 * Note this update of @saved_auxv is lockless thus
1886 * if someone reads this member in procfs while we're
1887 * updating -- it may get partly updated results. It's
1888 * known and acceptable trade off: we leave it as is to
1889 * not introduce additional locks here making the kernel
1892 if (prctl_map
.auxv_size
)
1893 memcpy(mm
->saved_auxv
, user_auxv
, sizeof(user_auxv
));
1895 up_write(&mm
->mmap_sem
);
1898 #endif /* CONFIG_CHECKPOINT_RESTORE */
1900 static int prctl_set_auxv(struct mm_struct
*mm
, unsigned long addr
,
1904 * This doesn't move the auxiliary vector itself since it's pinned to
1905 * mm_struct, but it permits filling the vector with new values. It's
1906 * up to the caller to provide sane values here, otherwise userspace
1907 * tools which use this vector might be unhappy.
1909 unsigned long user_auxv
[AT_VECTOR_SIZE
];
1911 if (len
> sizeof(user_auxv
))
1914 if (copy_from_user(user_auxv
, (const void __user
*)addr
, len
))
1917 /* Make sure the last entry is always AT_NULL */
1918 user_auxv
[AT_VECTOR_SIZE
- 2] = 0;
1919 user_auxv
[AT_VECTOR_SIZE
- 1] = 0;
1921 BUILD_BUG_ON(sizeof(user_auxv
) != sizeof(mm
->saved_auxv
));
1924 memcpy(mm
->saved_auxv
, user_auxv
, len
);
1925 task_unlock(current
);
1930 static int prctl_set_mm(int opt
, unsigned long addr
,
1931 unsigned long arg4
, unsigned long arg5
)
1933 struct mm_struct
*mm
= current
->mm
;
1934 struct prctl_mm_map prctl_map
;
1935 struct vm_area_struct
*vma
;
1938 if (arg5
|| (arg4
&& (opt
!= PR_SET_MM_AUXV
&&
1939 opt
!= PR_SET_MM_MAP
&&
1940 opt
!= PR_SET_MM_MAP_SIZE
)))
1943 #ifdef CONFIG_CHECKPOINT_RESTORE
1944 if (opt
== PR_SET_MM_MAP
|| opt
== PR_SET_MM_MAP_SIZE
)
1945 return prctl_set_mm_map(opt
, (const void __user
*)addr
, arg4
);
1948 if (!capable(CAP_SYS_RESOURCE
))
1951 if (opt
== PR_SET_MM_EXE_FILE
)
1952 return prctl_set_mm_exe_file(mm
, (unsigned int)addr
);
1954 if (opt
== PR_SET_MM_AUXV
)
1955 return prctl_set_auxv(mm
, addr
, arg4
);
1957 if (addr
>= TASK_SIZE
|| addr
< mmap_min_addr
)
1962 down_write(&mm
->mmap_sem
);
1963 vma
= find_vma(mm
, addr
);
1965 prctl_map
.start_code
= mm
->start_code
;
1966 prctl_map
.end_code
= mm
->end_code
;
1967 prctl_map
.start_data
= mm
->start_data
;
1968 prctl_map
.end_data
= mm
->end_data
;
1969 prctl_map
.start_brk
= mm
->start_brk
;
1970 prctl_map
.brk
= mm
->brk
;
1971 prctl_map
.start_stack
= mm
->start_stack
;
1972 prctl_map
.arg_start
= mm
->arg_start
;
1973 prctl_map
.arg_end
= mm
->arg_end
;
1974 prctl_map
.env_start
= mm
->env_start
;
1975 prctl_map
.env_end
= mm
->env_end
;
1976 prctl_map
.auxv
= NULL
;
1977 prctl_map
.auxv_size
= 0;
1978 prctl_map
.exe_fd
= -1;
1981 case PR_SET_MM_START_CODE
:
1982 prctl_map
.start_code
= addr
;
1984 case PR_SET_MM_END_CODE
:
1985 prctl_map
.end_code
= addr
;
1987 case PR_SET_MM_START_DATA
:
1988 prctl_map
.start_data
= addr
;
1990 case PR_SET_MM_END_DATA
:
1991 prctl_map
.end_data
= addr
;
1993 case PR_SET_MM_START_STACK
:
1994 prctl_map
.start_stack
= addr
;
1996 case PR_SET_MM_START_BRK
:
1997 prctl_map
.start_brk
= addr
;
2000 prctl_map
.brk
= addr
;
2002 case PR_SET_MM_ARG_START
:
2003 prctl_map
.arg_start
= addr
;
2005 case PR_SET_MM_ARG_END
:
2006 prctl_map
.arg_end
= addr
;
2008 case PR_SET_MM_ENV_START
:
2009 prctl_map
.env_start
= addr
;
2011 case PR_SET_MM_ENV_END
:
2012 prctl_map
.env_end
= addr
;
2018 error
= validate_prctl_map(&prctl_map
);
2024 * If command line arguments and environment
2025 * are placed somewhere else on stack, we can
2026 * set them up here, ARG_START/END to setup
2027 * command line argumets and ENV_START/END
2030 case PR_SET_MM_START_STACK
:
2031 case PR_SET_MM_ARG_START
:
2032 case PR_SET_MM_ARG_END
:
2033 case PR_SET_MM_ENV_START
:
2034 case PR_SET_MM_ENV_END
:
2041 mm
->start_code
= prctl_map
.start_code
;
2042 mm
->end_code
= prctl_map
.end_code
;
2043 mm
->start_data
= prctl_map
.start_data
;
2044 mm
->end_data
= prctl_map
.end_data
;
2045 mm
->start_brk
= prctl_map
.start_brk
;
2046 mm
->brk
= prctl_map
.brk
;
2047 mm
->start_stack
= prctl_map
.start_stack
;
2048 mm
->arg_start
= prctl_map
.arg_start
;
2049 mm
->arg_end
= prctl_map
.arg_end
;
2050 mm
->env_start
= prctl_map
.env_start
;
2051 mm
->env_end
= prctl_map
.env_end
;
2055 up_write(&mm
->mmap_sem
);
2059 #ifdef CONFIG_CHECKPOINT_RESTORE
2060 static int prctl_get_tid_address(struct task_struct
*me
, int __user
**tid_addr
)
2062 return put_user(me
->clear_child_tid
, tid_addr
);
2065 static int prctl_get_tid_address(struct task_struct
*me
, int __user
**tid_addr
)
2071 static int propagate_has_child_subreaper(struct task_struct
*p
, void *data
)
2074 * If task has has_child_subreaper - all its decendants
2075 * already have these flag too and new decendants will
2076 * inherit it on fork, skip them.
2078 * If we've found child_reaper - skip descendants in
2079 * it's subtree as they will never get out pidns.
2081 if (p
->signal
->has_child_subreaper
||
2082 is_child_reaper(task_pid(p
)))
2085 p
->signal
->has_child_subreaper
= 1;
2089 SYSCALL_DEFINE5(prctl
, int, option
, unsigned long, arg2
, unsigned long, arg3
,
2090 unsigned long, arg4
, unsigned long, arg5
)
2092 struct task_struct
*me
= current
;
2093 unsigned char comm
[sizeof(me
->comm
)];
2096 error
= security_task_prctl(option
, arg2
, arg3
, arg4
, arg5
);
2097 if (error
!= -ENOSYS
)
2102 case PR_SET_PDEATHSIG
:
2103 if (!valid_signal(arg2
)) {
2107 me
->pdeath_signal
= arg2
;
2109 case PR_GET_PDEATHSIG
:
2110 error
= put_user(me
->pdeath_signal
, (int __user
*)arg2
);
2112 case PR_GET_DUMPABLE
:
2113 error
= get_dumpable(me
->mm
);
2115 case PR_SET_DUMPABLE
:
2116 if (arg2
!= SUID_DUMP_DISABLE
&& arg2
!= SUID_DUMP_USER
) {
2120 set_dumpable(me
->mm
, arg2
);
2123 case PR_SET_UNALIGN
:
2124 error
= SET_UNALIGN_CTL(me
, arg2
);
2126 case PR_GET_UNALIGN
:
2127 error
= GET_UNALIGN_CTL(me
, arg2
);
2130 error
= SET_FPEMU_CTL(me
, arg2
);
2133 error
= GET_FPEMU_CTL(me
, arg2
);
2136 error
= SET_FPEXC_CTL(me
, arg2
);
2139 error
= GET_FPEXC_CTL(me
, arg2
);
2142 error
= PR_TIMING_STATISTICAL
;
2145 if (arg2
!= PR_TIMING_STATISTICAL
)
2149 comm
[sizeof(me
->comm
) - 1] = 0;
2150 if (strncpy_from_user(comm
, (char __user
*)arg2
,
2151 sizeof(me
->comm
) - 1) < 0)
2153 set_task_comm(me
, comm
);
2154 proc_comm_connector(me
);
2157 get_task_comm(comm
, me
);
2158 if (copy_to_user((char __user
*)arg2
, comm
, sizeof(comm
)))
2162 error
= GET_ENDIAN(me
, arg2
);
2165 error
= SET_ENDIAN(me
, arg2
);
2167 case PR_GET_SECCOMP
:
2168 error
= prctl_get_seccomp();
2170 case PR_SET_SECCOMP
:
2171 error
= prctl_set_seccomp(arg2
, (char __user
*)arg3
);
2174 error
= GET_TSC_CTL(arg2
);
2177 error
= SET_TSC_CTL(arg2
);
2179 case PR_TASK_PERF_EVENTS_DISABLE
:
2180 error
= perf_event_task_disable();
2182 case PR_TASK_PERF_EVENTS_ENABLE
:
2183 error
= perf_event_task_enable();
2185 case PR_GET_TIMERSLACK
:
2186 if (current
->timer_slack_ns
> ULONG_MAX
)
2189 error
= current
->timer_slack_ns
;
2191 case PR_SET_TIMERSLACK
:
2193 current
->timer_slack_ns
=
2194 current
->default_timer_slack_ns
;
2196 current
->timer_slack_ns
= arg2
;
2202 case PR_MCE_KILL_CLEAR
:
2205 current
->flags
&= ~PF_MCE_PROCESS
;
2207 case PR_MCE_KILL_SET
:
2208 current
->flags
|= PF_MCE_PROCESS
;
2209 if (arg3
== PR_MCE_KILL_EARLY
)
2210 current
->flags
|= PF_MCE_EARLY
;
2211 else if (arg3
== PR_MCE_KILL_LATE
)
2212 current
->flags
&= ~PF_MCE_EARLY
;
2213 else if (arg3
== PR_MCE_KILL_DEFAULT
)
2215 ~(PF_MCE_EARLY
|PF_MCE_PROCESS
);
2223 case PR_MCE_KILL_GET
:
2224 if (arg2
| arg3
| arg4
| arg5
)
2226 if (current
->flags
& PF_MCE_PROCESS
)
2227 error
= (current
->flags
& PF_MCE_EARLY
) ?
2228 PR_MCE_KILL_EARLY
: PR_MCE_KILL_LATE
;
2230 error
= PR_MCE_KILL_DEFAULT
;
2233 error
= prctl_set_mm(arg2
, arg3
, arg4
, arg5
);
2235 case PR_GET_TID_ADDRESS
:
2236 error
= prctl_get_tid_address(me
, (int __user
**)arg2
);
2238 case PR_SET_CHILD_SUBREAPER
:
2239 me
->signal
->is_child_subreaper
= !!arg2
;
2243 walk_process_tree(me
, propagate_has_child_subreaper
, NULL
);
2245 case PR_GET_CHILD_SUBREAPER
:
2246 error
= put_user(me
->signal
->is_child_subreaper
,
2247 (int __user
*)arg2
);
2249 case PR_SET_NO_NEW_PRIVS
:
2250 if (arg2
!= 1 || arg3
|| arg4
|| arg5
)
2253 task_set_no_new_privs(current
);
2255 case PR_GET_NO_NEW_PRIVS
:
2256 if (arg2
|| arg3
|| arg4
|| arg5
)
2258 return task_no_new_privs(current
) ? 1 : 0;
2259 case PR_GET_THP_DISABLE
:
2260 if (arg2
|| arg3
|| arg4
|| arg5
)
2262 error
= !!(me
->mm
->def_flags
& VM_NOHUGEPAGE
);
2264 case PR_SET_THP_DISABLE
:
2265 if (arg3
|| arg4
|| arg5
)
2267 if (down_write_killable(&me
->mm
->mmap_sem
))
2270 me
->mm
->def_flags
|= VM_NOHUGEPAGE
;
2272 me
->mm
->def_flags
&= ~VM_NOHUGEPAGE
;
2273 up_write(&me
->mm
->mmap_sem
);
2275 case PR_MPX_ENABLE_MANAGEMENT
:
2276 if (arg2
|| arg3
|| arg4
|| arg5
)
2278 error
= MPX_ENABLE_MANAGEMENT();
2280 case PR_MPX_DISABLE_MANAGEMENT
:
2281 if (arg2
|| arg3
|| arg4
|| arg5
)
2283 error
= MPX_DISABLE_MANAGEMENT();
2285 case PR_SET_FP_MODE
:
2286 error
= SET_FP_MODE(me
, arg2
);
2288 case PR_GET_FP_MODE
:
2289 error
= GET_FP_MODE(me
);
2298 SYSCALL_DEFINE3(getcpu
, unsigned __user
*, cpup
, unsigned __user
*, nodep
,
2299 struct getcpu_cache __user
*, unused
)
2302 int cpu
= raw_smp_processor_id();
2305 err
|= put_user(cpu
, cpup
);
2307 err
|= put_user(cpu_to_node(cpu
), nodep
);
2308 return err
? -EFAULT
: 0;
2312 * do_sysinfo - fill in sysinfo struct
2313 * @info: pointer to buffer to fill
2315 static int do_sysinfo(struct sysinfo
*info
)
2317 unsigned long mem_total
, sav_total
;
2318 unsigned int mem_unit
, bitcount
;
2321 memset(info
, 0, sizeof(struct sysinfo
));
2323 get_monotonic_boottime(&tp
);
2324 info
->uptime
= tp
.tv_sec
+ (tp
.tv_nsec
? 1 : 0);
2326 get_avenrun(info
->loads
, 0, SI_LOAD_SHIFT
- FSHIFT
);
2328 info
->procs
= nr_threads
;
2334 * If the sum of all the available memory (i.e. ram + swap)
2335 * is less than can be stored in a 32 bit unsigned long then
2336 * we can be binary compatible with 2.2.x kernels. If not,
2337 * well, in that case 2.2.x was broken anyways...
2339 * -Erik Andersen <andersee@debian.org>
2342 mem_total
= info
->totalram
+ info
->totalswap
;
2343 if (mem_total
< info
->totalram
|| mem_total
< info
->totalswap
)
2346 mem_unit
= info
->mem_unit
;
2347 while (mem_unit
> 1) {
2350 sav_total
= mem_total
;
2352 if (mem_total
< sav_total
)
2357 * If mem_total did not overflow, multiply all memory values by
2358 * info->mem_unit and set it to 1. This leaves things compatible
2359 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2364 info
->totalram
<<= bitcount
;
2365 info
->freeram
<<= bitcount
;
2366 info
->sharedram
<<= bitcount
;
2367 info
->bufferram
<<= bitcount
;
2368 info
->totalswap
<<= bitcount
;
2369 info
->freeswap
<<= bitcount
;
2370 info
->totalhigh
<<= bitcount
;
2371 info
->freehigh
<<= bitcount
;
2377 SYSCALL_DEFINE1(sysinfo
, struct sysinfo __user
*, info
)
2383 if (copy_to_user(info
, &val
, sizeof(struct sysinfo
)))
2389 #ifdef CONFIG_COMPAT
2390 struct compat_sysinfo
{
2404 char _f
[20-2*sizeof(u32
)-sizeof(int)];
2407 COMPAT_SYSCALL_DEFINE1(sysinfo
, struct compat_sysinfo __user
*, info
)
2413 /* Check to see if any memory value is too large for 32-bit and scale
2416 if (upper_32_bits(s
.totalram
) || upper_32_bits(s
.totalswap
)) {
2419 while (s
.mem_unit
< PAGE_SIZE
) {
2424 s
.totalram
>>= bitcount
;
2425 s
.freeram
>>= bitcount
;
2426 s
.sharedram
>>= bitcount
;
2427 s
.bufferram
>>= bitcount
;
2428 s
.totalswap
>>= bitcount
;
2429 s
.freeswap
>>= bitcount
;
2430 s
.totalhigh
>>= bitcount
;
2431 s
.freehigh
>>= bitcount
;
2434 if (!access_ok(VERIFY_WRITE
, info
, sizeof(struct compat_sysinfo
)) ||
2435 __put_user(s
.uptime
, &info
->uptime
) ||
2436 __put_user(s
.loads
[0], &info
->loads
[0]) ||
2437 __put_user(s
.loads
[1], &info
->loads
[1]) ||
2438 __put_user(s
.loads
[2], &info
->loads
[2]) ||
2439 __put_user(s
.totalram
, &info
->totalram
) ||
2440 __put_user(s
.freeram
, &info
->freeram
) ||
2441 __put_user(s
.sharedram
, &info
->sharedram
) ||
2442 __put_user(s
.bufferram
, &info
->bufferram
) ||
2443 __put_user(s
.totalswap
, &info
->totalswap
) ||
2444 __put_user(s
.freeswap
, &info
->freeswap
) ||
2445 __put_user(s
.procs
, &info
->procs
) ||
2446 __put_user(s
.totalhigh
, &info
->totalhigh
) ||
2447 __put_user(s
.freehigh
, &info
->freehigh
) ||
2448 __put_user(s
.mem_unit
, &info
->mem_unit
))
2453 #endif /* CONFIG_COMPAT */