1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/export.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.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/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>
44 #include <linux/syscall_user_dispatch.h>
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/time_namespace.h>
51 #include <linux/binfmts.h>
53 #include <linux/sched.h>
54 #include <linux/sched/autogroup.h>
55 #include <linux/sched/loadavg.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/mm.h>
58 #include <linux/sched/coredump.h>
59 #include <linux/sched/task.h>
60 #include <linux/sched/cputime.h>
61 #include <linux/rcupdate.h>
62 #include <linux/uidgid.h>
63 #include <linux/cred.h>
65 #include <linux/nospec.h>
67 #include <linux/kmsg_dump.h>
68 /* Move somewhere else to avoid recompiling? */
69 #include <generated/utsrelease.h>
71 #include <linux/uaccess.h>
73 #include <asm/unistd.h>
77 #ifndef SET_UNALIGN_CTL
78 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
80 #ifndef GET_UNALIGN_CTL
81 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
84 # define SET_FPEMU_CTL(a, b) (-EINVAL)
87 # define GET_FPEMU_CTL(a, b) (-EINVAL)
90 # define SET_FPEXC_CTL(a, b) (-EINVAL)
93 # define GET_FPEXC_CTL(a, b) (-EINVAL)
96 # define GET_ENDIAN(a, b) (-EINVAL)
99 # define SET_ENDIAN(a, b) (-EINVAL)
102 # define GET_TSC_CTL(a) (-EINVAL)
105 # define SET_TSC_CTL(a) (-EINVAL)
108 # define GET_FP_MODE(a) (-EINVAL)
111 # define SET_FP_MODE(a,b) (-EINVAL)
114 # define SVE_SET_VL(a) (-EINVAL)
117 # define SVE_GET_VL() (-EINVAL)
119 #ifndef PAC_RESET_KEYS
120 # define PAC_RESET_KEYS(a, b) (-EINVAL)
122 #ifndef PAC_SET_ENABLED_KEYS
123 # define PAC_SET_ENABLED_KEYS(a, b, c) (-EINVAL)
125 #ifndef PAC_GET_ENABLED_KEYS
126 # define PAC_GET_ENABLED_KEYS(a) (-EINVAL)
128 #ifndef SET_TAGGED_ADDR_CTRL
129 # define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
131 #ifndef GET_TAGGED_ADDR_CTRL
132 # define GET_TAGGED_ADDR_CTRL() (-EINVAL)
136 * this is where the system-wide overflow UID and GID are defined, for
137 * architectures that now have 32-bit UID/GID but didn't in the past
140 int overflowuid = DEFAULT_OVERFLOWUID;
141 int overflowgid = DEFAULT_OVERFLOWGID;
143 EXPORT_SYMBOL(overflowuid);
144 EXPORT_SYMBOL(overflowgid);
147 * the same as above, but for filesystems which can only store a 16-bit
148 * UID and GID. as such, this is needed on all architectures
151 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
152 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
154 EXPORT_SYMBOL(fs_overflowuid);
155 EXPORT_SYMBOL(fs_overflowgid);
158 * Returns true if current's euid is same as p's uid or euid,
159 * or has CAP_SYS_NICE to p's user_ns.
161 * Called with rcu_read_lock, creds are safe
163 static bool set_one_prio_perm(struct task_struct *p)
165 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
167 if (uid_eq(pcred->uid, cred->euid) ||
168 uid_eq(pcred->euid, cred->euid))
170 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
176 * set the priority of a task
177 * - the caller must hold the RCU read lock
179 static int set_one_prio(struct task_struct *p, int niceval, int error)
183 if (!set_one_prio_perm(p)) {
187 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
191 no_nice = security_task_setnice(p, niceval);
198 set_user_nice(p, niceval);
203 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
205 struct task_struct *g, *p;
206 struct user_struct *user;
207 const struct cred *cred = current_cred();
212 if (which > PRIO_USER || which < PRIO_PROCESS)
215 /* normalize: avoid signed division (rounding problems) */
217 if (niceval < MIN_NICE)
219 if (niceval > MAX_NICE)
223 read_lock(&tasklist_lock);
227 p = find_task_by_vpid(who);
231 error = set_one_prio(p, niceval, error);
235 pgrp = find_vpid(who);
237 pgrp = task_pgrp(current);
238 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
239 error = set_one_prio(p, niceval, error);
240 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
243 uid = make_kuid(cred->user_ns, who);
247 else if (!uid_eq(uid, cred->uid)) {
248 user = find_user(uid);
250 goto out_unlock; /* No processes for this user */
252 do_each_thread(g, p) {
253 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
254 error = set_one_prio(p, niceval, error);
255 } while_each_thread(g, p);
256 if (!uid_eq(uid, cred->uid))
257 free_uid(user); /* For find_user() */
261 read_unlock(&tasklist_lock);
268 * Ugh. To avoid negative return values, "getpriority()" will
269 * not return the normal nice-value, but a negated value that
270 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
271 * to stay compatible.
273 SYSCALL_DEFINE2(getpriority, int, which, int, who)
275 struct task_struct *g, *p;
276 struct user_struct *user;
277 const struct cred *cred = current_cred();
278 long niceval, retval = -ESRCH;
282 if (which > PRIO_USER || which < PRIO_PROCESS)
286 read_lock(&tasklist_lock);
290 p = find_task_by_vpid(who);
294 niceval = nice_to_rlimit(task_nice(p));
295 if (niceval > retval)
301 pgrp = find_vpid(who);
303 pgrp = task_pgrp(current);
304 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
305 niceval = nice_to_rlimit(task_nice(p));
306 if (niceval > retval)
308 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
311 uid = make_kuid(cred->user_ns, who);
315 else if (!uid_eq(uid, cred->uid)) {
316 user = find_user(uid);
318 goto out_unlock; /* No processes for this user */
320 do_each_thread(g, p) {
321 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
322 niceval = nice_to_rlimit(task_nice(p));
323 if (niceval > retval)
326 } while_each_thread(g, p);
327 if (!uid_eq(uid, cred->uid))
328 free_uid(user); /* for find_user() */
332 read_unlock(&tasklist_lock);
339 * Unprivileged users may change the real gid to the effective gid
340 * or vice versa. (BSD-style)
342 * If you set the real gid at all, or set the effective gid to a value not
343 * equal to the real gid, then the saved gid is set to the new effective gid.
345 * This makes it possible for a setgid program to completely drop its
346 * privileges, which is often a useful assertion to make when you are doing
347 * a security audit over a program.
349 * The general idea is that a program which uses just setregid() will be
350 * 100% compatible with BSD. A program which uses just setgid() will be
351 * 100% compatible with POSIX with saved IDs.
353 * SMP: There are not races, the GIDs are checked only by filesystem
354 * operations (as far as semantic preservation is concerned).
356 #ifdef CONFIG_MULTIUSER
357 long __sys_setregid(gid_t rgid, gid_t egid)
359 struct user_namespace *ns = current_user_ns();
360 const struct cred *old;
365 krgid = make_kgid(ns, rgid);
366 kegid = make_kgid(ns, egid);
368 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
370 if ((egid != (gid_t) -1) && !gid_valid(kegid))
373 new = prepare_creds();
376 old = current_cred();
379 if (rgid != (gid_t) -1) {
380 if (gid_eq(old->gid, krgid) ||
381 gid_eq(old->egid, krgid) ||
382 ns_capable_setid(old->user_ns, CAP_SETGID))
387 if (egid != (gid_t) -1) {
388 if (gid_eq(old->gid, kegid) ||
389 gid_eq(old->egid, kegid) ||
390 gid_eq(old->sgid, kegid) ||
391 ns_capable_setid(old->user_ns, CAP_SETGID))
397 if (rgid != (gid_t) -1 ||
398 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
399 new->sgid = new->egid;
400 new->fsgid = new->egid;
402 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
406 return commit_creds(new);
413 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
415 return __sys_setregid(rgid, egid);
419 * setgid() is implemented like SysV w/ SAVED_IDS
421 * SMP: Same implicit races as above.
423 long __sys_setgid(gid_t gid)
425 struct user_namespace *ns = current_user_ns();
426 const struct cred *old;
431 kgid = make_kgid(ns, gid);
432 if (!gid_valid(kgid))
435 new = prepare_creds();
438 old = current_cred();
441 if (ns_capable_setid(old->user_ns, CAP_SETGID))
442 new->gid = new->egid = new->sgid = new->fsgid = kgid;
443 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
444 new->egid = new->fsgid = kgid;
448 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
452 return commit_creds(new);
459 SYSCALL_DEFINE1(setgid, gid_t, gid)
461 return __sys_setgid(gid);
465 * change the user struct in a credentials set to match the new UID
467 static int set_user(struct cred *new)
469 struct user_struct *new_user;
471 new_user = alloc_uid(new->uid);
476 new->user = new_user;
480 static void flag_nproc_exceeded(struct cred *new)
482 if (new->ucounts == current_ucounts())
486 * We don't fail in case of NPROC limit excess here because too many
487 * poorly written programs don't check set*uid() return code, assuming
488 * it never fails if called by root. We may still enforce NPROC limit
489 * for programs doing set*uid()+execve() by harmlessly deferring the
490 * failure to the execve() stage.
492 if (is_ucounts_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
493 new->user != INIT_USER)
494 current->flags |= PF_NPROC_EXCEEDED;
496 current->flags &= ~PF_NPROC_EXCEEDED;
500 * Unprivileged users may change the real uid to the effective uid
501 * or vice versa. (BSD-style)
503 * If you set the real uid at all, or set the effective uid to a value not
504 * equal to the real uid, then the saved uid is set to the new effective uid.
506 * This makes it possible for a setuid program to completely drop its
507 * privileges, which is often a useful assertion to make when you are doing
508 * a security audit over a program.
510 * The general idea is that a program which uses just setreuid() will be
511 * 100% compatible with BSD. A program which uses just setuid() will be
512 * 100% compatible with POSIX with saved IDs.
514 long __sys_setreuid(uid_t ruid, uid_t euid)
516 struct user_namespace *ns = current_user_ns();
517 const struct cred *old;
522 kruid = make_kuid(ns, ruid);
523 keuid = make_kuid(ns, euid);
525 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
527 if ((euid != (uid_t) -1) && !uid_valid(keuid))
530 new = prepare_creds();
533 old = current_cred();
536 if (ruid != (uid_t) -1) {
538 if (!uid_eq(old->uid, kruid) &&
539 !uid_eq(old->euid, kruid) &&
540 !ns_capable_setid(old->user_ns, CAP_SETUID))
544 if (euid != (uid_t) -1) {
546 if (!uid_eq(old->uid, keuid) &&
547 !uid_eq(old->euid, keuid) &&
548 !uid_eq(old->suid, keuid) &&
549 !ns_capable_setid(old->user_ns, CAP_SETUID))
553 if (!uid_eq(new->uid, old->uid)) {
554 retval = set_user(new);
558 if (ruid != (uid_t) -1 ||
559 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
560 new->suid = new->euid;
561 new->fsuid = new->euid;
563 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
567 retval = set_cred_ucounts(new);
571 flag_nproc_exceeded(new);
572 return commit_creds(new);
579 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
581 return __sys_setreuid(ruid, euid);
585 * setuid() is implemented like SysV with SAVED_IDS
587 * Note that SAVED_ID's is deficient in that a setuid root program
588 * like sendmail, for example, cannot set its uid to be a normal
589 * user and then switch back, because if you're root, setuid() sets
590 * the saved uid too. If you don't like this, blame the bright people
591 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
592 * will allow a root program to temporarily drop privileges and be able to
593 * regain them by swapping the real and effective uid.
595 long __sys_setuid(uid_t uid)
597 struct user_namespace *ns = current_user_ns();
598 const struct cred *old;
603 kuid = make_kuid(ns, uid);
604 if (!uid_valid(kuid))
607 new = prepare_creds();
610 old = current_cred();
613 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
614 new->suid = new->uid = kuid;
615 if (!uid_eq(kuid, old->uid)) {
616 retval = set_user(new);
620 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
624 new->fsuid = new->euid = kuid;
626 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
630 retval = set_cred_ucounts(new);
634 flag_nproc_exceeded(new);
635 return commit_creds(new);
642 SYSCALL_DEFINE1(setuid, uid_t, uid)
644 return __sys_setuid(uid);
649 * This function implements a generic ability to update ruid, euid,
650 * and suid. This allows you to implement the 4.4 compatible seteuid().
652 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
654 struct user_namespace *ns = current_user_ns();
655 const struct cred *old;
658 kuid_t kruid, keuid, ksuid;
659 bool ruid_new, euid_new, suid_new;
661 kruid = make_kuid(ns, ruid);
662 keuid = make_kuid(ns, euid);
663 ksuid = make_kuid(ns, suid);
665 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
668 if ((euid != (uid_t) -1) && !uid_valid(keuid))
671 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
674 old = current_cred();
676 /* check for no-op */
677 if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
678 (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
679 uid_eq(keuid, old->fsuid))) &&
680 (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
683 ruid_new = ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
684 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
685 euid_new = euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
686 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
687 suid_new = suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
688 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
689 if ((ruid_new || euid_new || suid_new) &&
690 !ns_capable_setid(old->user_ns, CAP_SETUID))
693 new = prepare_creds();
697 if (ruid != (uid_t) -1) {
699 if (!uid_eq(kruid, old->uid)) {
700 retval = set_user(new);
705 if (euid != (uid_t) -1)
707 if (suid != (uid_t) -1)
709 new->fsuid = new->euid;
711 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
715 retval = set_cred_ucounts(new);
719 flag_nproc_exceeded(new);
720 return commit_creds(new);
727 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
729 return __sys_setresuid(ruid, euid, suid);
732 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
734 const struct cred *cred = current_cred();
736 uid_t ruid, euid, suid;
738 ruid = from_kuid_munged(cred->user_ns, cred->uid);
739 euid = from_kuid_munged(cred->user_ns, cred->euid);
740 suid = from_kuid_munged(cred->user_ns, cred->suid);
742 retval = put_user(ruid, ruidp);
744 retval = put_user(euid, euidp);
746 return put_user(suid, suidp);
752 * Same as above, but for rgid, egid, sgid.
754 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
756 struct user_namespace *ns = current_user_ns();
757 const struct cred *old;
760 kgid_t krgid, kegid, ksgid;
761 bool rgid_new, egid_new, sgid_new;
763 krgid = make_kgid(ns, rgid);
764 kegid = make_kgid(ns, egid);
765 ksgid = make_kgid(ns, sgid);
767 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
769 if ((egid != (gid_t) -1) && !gid_valid(kegid))
771 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
774 old = current_cred();
776 /* check for no-op */
777 if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
778 (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
779 gid_eq(kegid, old->fsgid))) &&
780 (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
783 rgid_new = rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
784 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
785 egid_new = egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
786 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
787 sgid_new = sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
788 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
789 if ((rgid_new || egid_new || sgid_new) &&
790 !ns_capable_setid(old->user_ns, CAP_SETGID))
793 new = prepare_creds();
797 if (rgid != (gid_t) -1)
799 if (egid != (gid_t) -1)
801 if (sgid != (gid_t) -1)
803 new->fsgid = new->egid;
805 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
809 return commit_creds(new);
816 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
818 return __sys_setresgid(rgid, egid, sgid);
821 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
823 const struct cred *cred = current_cred();
825 gid_t rgid, egid, sgid;
827 rgid = from_kgid_munged(cred->user_ns, cred->gid);
828 egid = from_kgid_munged(cred->user_ns, cred->egid);
829 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
831 retval = put_user(rgid, rgidp);
833 retval = put_user(egid, egidp);
835 retval = put_user(sgid, sgidp);
843 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
844 * is used for "access()" and for the NFS daemon (letting nfsd stay at
845 * whatever uid it wants to). It normally shadows "euid", except when
846 * explicitly set by setfsuid() or for access..
848 long __sys_setfsuid(uid_t uid)
850 const struct cred *old;
855 old = current_cred();
856 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
858 kuid = make_kuid(old->user_ns, uid);
859 if (!uid_valid(kuid))
862 new = prepare_creds();
866 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
867 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
868 ns_capable_setid(old->user_ns, CAP_SETUID)) {
869 if (!uid_eq(kuid, old->fsuid)) {
871 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
884 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
886 return __sys_setfsuid(uid);
890 * Samma på svenska..
892 long __sys_setfsgid(gid_t gid)
894 const struct cred *old;
899 old = current_cred();
900 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
902 kgid = make_kgid(old->user_ns, gid);
903 if (!gid_valid(kgid))
906 new = prepare_creds();
910 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
911 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
912 ns_capable_setid(old->user_ns, CAP_SETGID)) {
913 if (!gid_eq(kgid, old->fsgid)) {
915 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
928 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
930 return __sys_setfsgid(gid);
932 #endif /* CONFIG_MULTIUSER */
935 * sys_getpid - return the thread group id of the current process
937 * Note, despite the name, this returns the tgid not the pid. The tgid and
938 * the pid are identical unless CLONE_THREAD was specified on clone() in
939 * which case the tgid is the same in all threads of the same group.
941 * This is SMP safe as current->tgid does not change.
943 SYSCALL_DEFINE0(getpid)
945 return task_tgid_vnr(current);
948 /* Thread ID - the internal kernel "pid" */
949 SYSCALL_DEFINE0(gettid)
951 return task_pid_vnr(current);
955 * Accessing ->real_parent is not SMP-safe, it could
956 * change from under us. However, we can use a stale
957 * value of ->real_parent under rcu_read_lock(), see
958 * release_task()->call_rcu(delayed_put_task_struct).
960 SYSCALL_DEFINE0(getppid)
965 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
971 SYSCALL_DEFINE0(getuid)
973 /* Only we change this so SMP safe */
974 return from_kuid_munged(current_user_ns(), current_uid());
977 SYSCALL_DEFINE0(geteuid)
979 /* Only we change this so SMP safe */
980 return from_kuid_munged(current_user_ns(), current_euid());
983 SYSCALL_DEFINE0(getgid)
985 /* Only we change this so SMP safe */
986 return from_kgid_munged(current_user_ns(), current_gid());
989 SYSCALL_DEFINE0(getegid)
991 /* Only we change this so SMP safe */
992 return from_kgid_munged(current_user_ns(), current_egid());
995 static void do_sys_times(struct tms *tms)
997 u64 tgutime, tgstime, cutime, cstime;
999 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
1000 cutime = current->signal->cutime;
1001 cstime = current->signal->cstime;
1002 tms->tms_utime = nsec_to_clock_t(tgutime);
1003 tms->tms_stime = nsec_to_clock_t(tgstime);
1004 tms->tms_cutime = nsec_to_clock_t(cutime);
1005 tms->tms_cstime = nsec_to_clock_t(cstime);
1008 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
1014 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
1017 force_successful_syscall_return();
1018 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1021 #ifdef CONFIG_COMPAT
1022 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1024 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1027 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1031 struct compat_tms tmp;
1034 /* Convert our struct tms to the compat version. */
1035 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1036 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1037 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1038 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1039 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1042 force_successful_syscall_return();
1043 return compat_jiffies_to_clock_t(jiffies);
1048 * This needs some heavy checking ...
1049 * I just haven't the stomach for it. I also don't fully
1050 * understand sessions/pgrp etc. Let somebody who does explain it.
1052 * OK, I think I have the protection semantics right.... this is really
1053 * only important on a multi-user system anyway, to make sure one user
1054 * can't send a signal to a process owned by another. -TYT, 12/12/91
1056 * !PF_FORKNOEXEC check to conform completely to POSIX.
1058 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1060 struct task_struct *p;
1061 struct task_struct *group_leader = current->group_leader;
1066 pid = task_pid_vnr(group_leader);
1073 /* From this point forward we keep holding onto the tasklist lock
1074 * so that our parent does not change from under us. -DaveM
1076 write_lock_irq(&tasklist_lock);
1079 p = find_task_by_vpid(pid);
1084 if (!thread_group_leader(p))
1087 if (same_thread_group(p->real_parent, group_leader)) {
1089 if (task_session(p) != task_session(group_leader))
1092 if (!(p->flags & PF_FORKNOEXEC))
1096 if (p != group_leader)
1101 if (p->signal->leader)
1106 struct task_struct *g;
1108 pgrp = find_vpid(pgid);
1109 g = pid_task(pgrp, PIDTYPE_PGID);
1110 if (!g || task_session(g) != task_session(group_leader))
1114 err = security_task_setpgid(p, pgid);
1118 if (task_pgrp(p) != pgrp)
1119 change_pid(p, PIDTYPE_PGID, pgrp);
1123 /* All paths lead to here, thus we are safe. -DaveM */
1124 write_unlock_irq(&tasklist_lock);
1129 static int do_getpgid(pid_t pid)
1131 struct task_struct *p;
1137 grp = task_pgrp(current);
1140 p = find_task_by_vpid(pid);
1147 retval = security_task_getpgid(p);
1151 retval = pid_vnr(grp);
1157 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1159 return do_getpgid(pid);
1162 #ifdef __ARCH_WANT_SYS_GETPGRP
1164 SYSCALL_DEFINE0(getpgrp)
1166 return do_getpgid(0);
1171 SYSCALL_DEFINE1(getsid, pid_t, pid)
1173 struct task_struct *p;
1179 sid = task_session(current);
1182 p = find_task_by_vpid(pid);
1185 sid = task_session(p);
1189 retval = security_task_getsid(p);
1193 retval = pid_vnr(sid);
1199 static void set_special_pids(struct pid *pid)
1201 struct task_struct *curr = current->group_leader;
1203 if (task_session(curr) != pid)
1204 change_pid(curr, PIDTYPE_SID, pid);
1206 if (task_pgrp(curr) != pid)
1207 change_pid(curr, PIDTYPE_PGID, pid);
1210 int ksys_setsid(void)
1212 struct task_struct *group_leader = current->group_leader;
1213 struct pid *sid = task_pid(group_leader);
1214 pid_t session = pid_vnr(sid);
1217 write_lock_irq(&tasklist_lock);
1218 /* Fail if I am already a session leader */
1219 if (group_leader->signal->leader)
1222 /* Fail if a process group id already exists that equals the
1223 * proposed session id.
1225 if (pid_task(sid, PIDTYPE_PGID))
1228 group_leader->signal->leader = 1;
1229 set_special_pids(sid);
1231 proc_clear_tty(group_leader);
1235 write_unlock_irq(&tasklist_lock);
1237 proc_sid_connector(group_leader);
1238 sched_autogroup_create_attach(group_leader);
1243 SYSCALL_DEFINE0(setsid)
1245 return ksys_setsid();
1248 DECLARE_RWSEM(uts_sem);
1250 #ifdef COMPAT_UTS_MACHINE
1251 #define override_architecture(name) \
1252 (personality(current->personality) == PER_LINUX32 && \
1253 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1254 sizeof(COMPAT_UTS_MACHINE)))
1256 #define override_architecture(name) 0
1260 * Work around broken programs that cannot handle "Linux 3.0".
1261 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1262 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1265 static int override_release(char __user *release, size_t len)
1269 if (current->personality & UNAME26) {
1270 const char *rest = UTS_RELEASE;
1271 char buf[65] = { 0 };
1277 if (*rest == '.' && ++ndots >= 3)
1279 if (!isdigit(*rest) && *rest != '.')
1283 v = LINUX_VERSION_PATCHLEVEL + 60;
1284 copy = clamp_t(size_t, len, 1, sizeof(buf));
1285 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1286 ret = copy_to_user(release, buf, copy + 1);
1291 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1293 struct new_utsname tmp;
1295 down_read(&uts_sem);
1296 memcpy(&tmp, utsname(), sizeof(tmp));
1298 if (copy_to_user(name, &tmp, sizeof(tmp)))
1301 if (override_release(name->release, sizeof(name->release)))
1303 if (override_architecture(name))
1308 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1312 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1314 struct old_utsname tmp;
1319 down_read(&uts_sem);
1320 memcpy(&tmp, utsname(), sizeof(tmp));
1322 if (copy_to_user(name, &tmp, sizeof(tmp)))
1325 if (override_release(name->release, sizeof(name->release)))
1327 if (override_architecture(name))
1332 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1334 struct oldold_utsname tmp;
1339 memset(&tmp, 0, sizeof(tmp));
1341 down_read(&uts_sem);
1342 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1343 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1344 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1345 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1346 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1348 if (copy_to_user(name, &tmp, sizeof(tmp)))
1351 if (override_architecture(name))
1353 if (override_release(name->release, sizeof(name->release)))
1359 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1362 char tmp[__NEW_UTS_LEN];
1364 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1367 if (len < 0 || len > __NEW_UTS_LEN)
1370 if (!copy_from_user(tmp, name, len)) {
1371 struct new_utsname *u;
1373 down_write(&uts_sem);
1375 memcpy(u->nodename, tmp, len);
1376 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1378 uts_proc_notify(UTS_PROC_HOSTNAME);
1384 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1386 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1389 struct new_utsname *u;
1390 char tmp[__NEW_UTS_LEN + 1];
1394 down_read(&uts_sem);
1396 i = 1 + strlen(u->nodename);
1399 memcpy(tmp, u->nodename, i);
1401 if (copy_to_user(name, tmp, i))
1409 * Only setdomainname; getdomainname can be implemented by calling
1412 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1415 char tmp[__NEW_UTS_LEN];
1417 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1419 if (len < 0 || len > __NEW_UTS_LEN)
1423 if (!copy_from_user(tmp, name, len)) {
1424 struct new_utsname *u;
1426 down_write(&uts_sem);
1428 memcpy(u->domainname, tmp, len);
1429 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1431 uts_proc_notify(UTS_PROC_DOMAINNAME);
1437 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1439 struct rlimit value;
1442 ret = do_prlimit(current, resource, NULL, &value);
1444 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1449 #ifdef CONFIG_COMPAT
1451 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1452 struct compat_rlimit __user *, rlim)
1455 struct compat_rlimit r32;
1457 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1460 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1461 r.rlim_cur = RLIM_INFINITY;
1463 r.rlim_cur = r32.rlim_cur;
1464 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1465 r.rlim_max = RLIM_INFINITY;
1467 r.rlim_max = r32.rlim_max;
1468 return do_prlimit(current, resource, &r, NULL);
1471 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1472 struct compat_rlimit __user *, rlim)
1477 ret = do_prlimit(current, resource, NULL, &r);
1479 struct compat_rlimit r32;
1480 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1481 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1483 r32.rlim_cur = r.rlim_cur;
1484 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1485 r32.rlim_max = COMPAT_RLIM_INFINITY;
1487 r32.rlim_max = r.rlim_max;
1489 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1497 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1500 * 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 resource = array_index_nospec(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;
1520 #ifdef CONFIG_COMPAT
1521 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1522 struct compat_rlimit __user *, rlim)
1526 if (resource >= RLIM_NLIMITS)
1529 resource = array_index_nospec(resource, RLIM_NLIMITS);
1530 task_lock(current->group_leader);
1531 r = current->signal->rlim[resource];
1532 task_unlock(current->group_leader);
1533 if (r.rlim_cur > 0x7FFFFFFF)
1534 r.rlim_cur = 0x7FFFFFFF;
1535 if (r.rlim_max > 0x7FFFFFFF)
1536 r.rlim_max = 0x7FFFFFFF;
1538 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1539 put_user(r.rlim_max, &rlim->rlim_max))
1547 static inline bool rlim64_is_infinity(__u64 rlim64)
1549 #if BITS_PER_LONG < 64
1550 return rlim64 >= ULONG_MAX;
1552 return rlim64 == RLIM64_INFINITY;
1556 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1558 if (rlim->rlim_cur == RLIM_INFINITY)
1559 rlim64->rlim_cur = RLIM64_INFINITY;
1561 rlim64->rlim_cur = rlim->rlim_cur;
1562 if (rlim->rlim_max == RLIM_INFINITY)
1563 rlim64->rlim_max = RLIM64_INFINITY;
1565 rlim64->rlim_max = rlim->rlim_max;
1568 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1570 if (rlim64_is_infinity(rlim64->rlim_cur))
1571 rlim->rlim_cur = RLIM_INFINITY;
1573 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1574 if (rlim64_is_infinity(rlim64->rlim_max))
1575 rlim->rlim_max = RLIM_INFINITY;
1577 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1580 /* make sure you are allowed to change @tsk limits before calling this */
1581 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1582 struct rlimit *new_rlim, struct rlimit *old_rlim)
1584 struct rlimit *rlim;
1587 if (resource >= RLIM_NLIMITS)
1589 resource = array_index_nospec(resource, RLIM_NLIMITS);
1592 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1594 if (resource == RLIMIT_NOFILE &&
1595 new_rlim->rlim_max > sysctl_nr_open)
1599 /* protect tsk->signal and tsk->sighand from disappearing */
1600 read_lock(&tasklist_lock);
1601 if (!tsk->sighand) {
1606 rlim = tsk->signal->rlim + resource;
1607 task_lock(tsk->group_leader);
1609 /* Keep the capable check against init_user_ns until
1610 cgroups can contain all limits */
1611 if (new_rlim->rlim_max > rlim->rlim_max &&
1612 !capable(CAP_SYS_RESOURCE))
1615 retval = security_task_setrlimit(tsk, resource, new_rlim);
1623 task_unlock(tsk->group_leader);
1626 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1627 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1628 * ignores the rlimit.
1630 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1631 new_rlim->rlim_cur != RLIM_INFINITY &&
1632 IS_ENABLED(CONFIG_POSIX_TIMERS))
1633 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1635 read_unlock(&tasklist_lock);
1639 /* rcu lock must be held */
1640 static int check_prlimit_permission(struct task_struct *task,
1643 const struct cred *cred = current_cred(), *tcred;
1646 if (current == task)
1649 tcred = __task_cred(task);
1650 id_match = (uid_eq(cred->uid, tcred->euid) &&
1651 uid_eq(cred->uid, tcred->suid) &&
1652 uid_eq(cred->uid, tcred->uid) &&
1653 gid_eq(cred->gid, tcred->egid) &&
1654 gid_eq(cred->gid, tcred->sgid) &&
1655 gid_eq(cred->gid, tcred->gid));
1656 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1659 return security_task_prlimit(cred, tcred, flags);
1662 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1663 const struct rlimit64 __user *, new_rlim,
1664 struct rlimit64 __user *, old_rlim)
1666 struct rlimit64 old64, new64;
1667 struct rlimit old, new;
1668 struct task_struct *tsk;
1669 unsigned int checkflags = 0;
1673 checkflags |= LSM_PRLIMIT_READ;
1676 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1678 rlim64_to_rlim(&new64, &new);
1679 checkflags |= LSM_PRLIMIT_WRITE;
1683 tsk = pid ? find_task_by_vpid(pid) : current;
1688 ret = check_prlimit_permission(tsk, checkflags);
1693 get_task_struct(tsk);
1696 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1697 old_rlim ? &old : NULL);
1699 if (!ret && old_rlim) {
1700 rlim_to_rlim64(&old, &old64);
1701 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1705 put_task_struct(tsk);
1709 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1711 struct rlimit new_rlim;
1713 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1715 return do_prlimit(current, resource, &new_rlim, NULL);
1719 * It would make sense to put struct rusage in the task_struct,
1720 * except that would make the task_struct be *really big*. After
1721 * task_struct gets moved into malloc'ed memory, it would
1722 * make sense to do this. It will make moving the rest of the information
1723 * a lot simpler! (Which we're not doing right now because we're not
1724 * measuring them yet).
1726 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1727 * races with threads incrementing their own counters. But since word
1728 * reads are atomic, we either get new values or old values and we don't
1729 * care which for the sums. We always take the siglock to protect reading
1730 * the c* fields from p->signal from races with exit.c updating those
1731 * fields when reaping, so a sample either gets all the additions of a
1732 * given child after it's reaped, or none so this sample is before reaping.
1735 * We need to take the siglock for CHILDEREN, SELF and BOTH
1736 * for the cases current multithreaded, non-current single threaded
1737 * non-current multithreaded. Thread traversal is now safe with
1739 * Strictly speaking, we donot need to take the siglock if we are current and
1740 * single threaded, as no one else can take our signal_struct away, no one
1741 * else can reap the children to update signal->c* counters, and no one else
1742 * can race with the signal-> fields. If we do not take any lock, the
1743 * signal-> fields could be read out of order while another thread was just
1744 * exiting. So we should place a read memory barrier when we avoid the lock.
1745 * On the writer side, write memory barrier is implied in __exit_signal
1746 * as __exit_signal releases the siglock spinlock after updating the signal->
1747 * fields. But we don't do this yet to keep things simple.
1751 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1753 r->ru_nvcsw += t->nvcsw;
1754 r->ru_nivcsw += t->nivcsw;
1755 r->ru_minflt += t->min_flt;
1756 r->ru_majflt += t->maj_flt;
1757 r->ru_inblock += task_io_get_inblock(t);
1758 r->ru_oublock += task_io_get_oublock(t);
1761 void getrusage(struct task_struct *p, int who, struct rusage *r)
1763 struct task_struct *t;
1764 unsigned long flags;
1765 u64 tgutime, tgstime, utime, stime;
1766 unsigned long maxrss = 0;
1768 memset((char *)r, 0, sizeof (*r));
1771 if (who == RUSAGE_THREAD) {
1772 task_cputime_adjusted(current, &utime, &stime);
1773 accumulate_thread_rusage(p, r);
1774 maxrss = p->signal->maxrss;
1778 if (!lock_task_sighand(p, &flags))
1783 case RUSAGE_CHILDREN:
1784 utime = p->signal->cutime;
1785 stime = p->signal->cstime;
1786 r->ru_nvcsw = p->signal->cnvcsw;
1787 r->ru_nivcsw = p->signal->cnivcsw;
1788 r->ru_minflt = p->signal->cmin_flt;
1789 r->ru_majflt = p->signal->cmaj_flt;
1790 r->ru_inblock = p->signal->cinblock;
1791 r->ru_oublock = p->signal->coublock;
1792 maxrss = p->signal->cmaxrss;
1794 if (who == RUSAGE_CHILDREN)
1799 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1802 r->ru_nvcsw += p->signal->nvcsw;
1803 r->ru_nivcsw += p->signal->nivcsw;
1804 r->ru_minflt += p->signal->min_flt;
1805 r->ru_majflt += p->signal->maj_flt;
1806 r->ru_inblock += p->signal->inblock;
1807 r->ru_oublock += p->signal->oublock;
1808 if (maxrss < p->signal->maxrss)
1809 maxrss = p->signal->maxrss;
1812 accumulate_thread_rusage(t, r);
1813 } while_each_thread(p, t);
1819 unlock_task_sighand(p, &flags);
1822 r->ru_utime = ns_to_kernel_old_timeval(utime);
1823 r->ru_stime = ns_to_kernel_old_timeval(stime);
1825 if (who != RUSAGE_CHILDREN) {
1826 struct mm_struct *mm = get_task_mm(p);
1829 setmax_mm_hiwater_rss(&maxrss, mm);
1833 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1836 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1840 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1841 who != RUSAGE_THREAD)
1844 getrusage(current, who, &r);
1845 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1848 #ifdef CONFIG_COMPAT
1849 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1853 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1854 who != RUSAGE_THREAD)
1857 getrusage(current, who, &r);
1858 return put_compat_rusage(&r, ru);
1862 SYSCALL_DEFINE1(umask, int, mask)
1864 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1868 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1871 struct inode *inode;
1878 inode = file_inode(exe.file);
1881 * Because the original mm->exe_file points to executable file, make
1882 * sure that this one is executable as well, to avoid breaking an
1886 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1889 err = file_permission(exe.file, MAY_EXEC);
1893 err = replace_mm_exe_file(mm, exe.file);
1900 * Check arithmetic relations of passed addresses.
1902 * WARNING: we don't require any capability here so be very careful
1903 * in what is allowed for modification from userspace.
1905 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1907 unsigned long mmap_max_addr = TASK_SIZE;
1908 int error = -EINVAL, i;
1910 static const unsigned char offsets[] = {
1911 offsetof(struct prctl_mm_map, start_code),
1912 offsetof(struct prctl_mm_map, end_code),
1913 offsetof(struct prctl_mm_map, start_data),
1914 offsetof(struct prctl_mm_map, end_data),
1915 offsetof(struct prctl_mm_map, start_brk),
1916 offsetof(struct prctl_mm_map, brk),
1917 offsetof(struct prctl_mm_map, start_stack),
1918 offsetof(struct prctl_mm_map, arg_start),
1919 offsetof(struct prctl_mm_map, arg_end),
1920 offsetof(struct prctl_mm_map, env_start),
1921 offsetof(struct prctl_mm_map, env_end),
1925 * Make sure the members are not somewhere outside
1926 * of allowed address space.
1928 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1929 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1931 if ((unsigned long)val >= mmap_max_addr ||
1932 (unsigned long)val < mmap_min_addr)
1937 * Make sure the pairs are ordered.
1939 #define __prctl_check_order(__m1, __op, __m2) \
1940 ((unsigned long)prctl_map->__m1 __op \
1941 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1942 error = __prctl_check_order(start_code, <, end_code);
1943 error |= __prctl_check_order(start_data,<=, end_data);
1944 error |= __prctl_check_order(start_brk, <=, brk);
1945 error |= __prctl_check_order(arg_start, <=, arg_end);
1946 error |= __prctl_check_order(env_start, <=, env_end);
1949 #undef __prctl_check_order
1954 * Neither we should allow to override limits if they set.
1956 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1957 prctl_map->start_brk, prctl_map->end_data,
1958 prctl_map->start_data))
1966 #ifdef CONFIG_CHECKPOINT_RESTORE
1967 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1969 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1970 unsigned long user_auxv[AT_VECTOR_SIZE];
1971 struct mm_struct *mm = current->mm;
1974 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1975 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1977 if (opt == PR_SET_MM_MAP_SIZE)
1978 return put_user((unsigned int)sizeof(prctl_map),
1979 (unsigned int __user *)addr);
1981 if (data_size != sizeof(prctl_map))
1984 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1987 error = validate_prctl_map_addr(&prctl_map);
1991 if (prctl_map.auxv_size) {
1993 * Someone is trying to cheat the auxv vector.
1995 if (!prctl_map.auxv ||
1996 prctl_map.auxv_size > sizeof(mm->saved_auxv))
1999 memset(user_auxv, 0, sizeof(user_auxv));
2000 if (copy_from_user(user_auxv,
2001 (const void __user *)prctl_map.auxv,
2002 prctl_map.auxv_size))
2005 /* Last entry must be AT_NULL as specification requires */
2006 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2007 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2010 if (prctl_map.exe_fd != (u32)-1) {
2012 * Check if the current user is checkpoint/restore capable.
2013 * At the time of this writing, it checks for CAP_SYS_ADMIN
2014 * or CAP_CHECKPOINT_RESTORE.
2015 * Note that a user with access to ptrace can masquerade an
2016 * arbitrary program as any executable, even setuid ones.
2017 * This may have implications in the tomoyo subsystem.
2019 if (!checkpoint_restore_ns_capable(current_user_ns()))
2022 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2028 * arg_lock protects concurrent updates but we still need mmap_lock for
2029 * read to exclude races with sys_brk.
2034 * We don't validate if these members are pointing to
2035 * real present VMAs because application may have correspond
2036 * VMAs already unmapped and kernel uses these members for statistics
2037 * output in procfs mostly, except
2039 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2040 * for VMAs when updating these members so anything wrong written
2041 * here cause kernel to swear at userspace program but won't lead
2042 * to any problem in kernel itself
2045 spin_lock(&mm->arg_lock);
2046 mm->start_code = prctl_map.start_code;
2047 mm->end_code = prctl_map.end_code;
2048 mm->start_data = prctl_map.start_data;
2049 mm->end_data = prctl_map.end_data;
2050 mm->start_brk = prctl_map.start_brk;
2051 mm->brk = prctl_map.brk;
2052 mm->start_stack = prctl_map.start_stack;
2053 mm->arg_start = prctl_map.arg_start;
2054 mm->arg_end = prctl_map.arg_end;
2055 mm->env_start = prctl_map.env_start;
2056 mm->env_end = prctl_map.env_end;
2057 spin_unlock(&mm->arg_lock);
2060 * Note this update of @saved_auxv is lockless thus
2061 * if someone reads this member in procfs while we're
2062 * updating -- it may get partly updated results. It's
2063 * known and acceptable trade off: we leave it as is to
2064 * not introduce additional locks here making the kernel
2067 if (prctl_map.auxv_size)
2068 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2070 mmap_read_unlock(mm);
2073 #endif /* CONFIG_CHECKPOINT_RESTORE */
2075 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2079 * This doesn't move the auxiliary vector itself since it's pinned to
2080 * mm_struct, but it permits filling the vector with new values. It's
2081 * up to the caller to provide sane values here, otherwise userspace
2082 * tools which use this vector might be unhappy.
2084 unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2086 if (len > sizeof(user_auxv))
2089 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2092 /* Make sure the last entry is always AT_NULL */
2093 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2094 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2096 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2099 memcpy(mm->saved_auxv, user_auxv, len);
2100 task_unlock(current);
2105 static int prctl_set_mm(int opt, unsigned long addr,
2106 unsigned long arg4, unsigned long arg5)
2108 struct mm_struct *mm = current->mm;
2109 struct prctl_mm_map prctl_map = {
2114 struct vm_area_struct *vma;
2117 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2118 opt != PR_SET_MM_MAP &&
2119 opt != PR_SET_MM_MAP_SIZE)))
2122 #ifdef CONFIG_CHECKPOINT_RESTORE
2123 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2124 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2127 if (!capable(CAP_SYS_RESOURCE))
2130 if (opt == PR_SET_MM_EXE_FILE)
2131 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2133 if (opt == PR_SET_MM_AUXV)
2134 return prctl_set_auxv(mm, addr, arg4);
2136 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2142 * arg_lock protects concurrent updates of arg boundaries, we need
2143 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2147 vma = find_vma(mm, addr);
2149 spin_lock(&mm->arg_lock);
2150 prctl_map.start_code = mm->start_code;
2151 prctl_map.end_code = mm->end_code;
2152 prctl_map.start_data = mm->start_data;
2153 prctl_map.end_data = mm->end_data;
2154 prctl_map.start_brk = mm->start_brk;
2155 prctl_map.brk = mm->brk;
2156 prctl_map.start_stack = mm->start_stack;
2157 prctl_map.arg_start = mm->arg_start;
2158 prctl_map.arg_end = mm->arg_end;
2159 prctl_map.env_start = mm->env_start;
2160 prctl_map.env_end = mm->env_end;
2163 case PR_SET_MM_START_CODE:
2164 prctl_map.start_code = addr;
2166 case PR_SET_MM_END_CODE:
2167 prctl_map.end_code = addr;
2169 case PR_SET_MM_START_DATA:
2170 prctl_map.start_data = addr;
2172 case PR_SET_MM_END_DATA:
2173 prctl_map.end_data = addr;
2175 case PR_SET_MM_START_STACK:
2176 prctl_map.start_stack = addr;
2178 case PR_SET_MM_START_BRK:
2179 prctl_map.start_brk = addr;
2182 prctl_map.brk = addr;
2184 case PR_SET_MM_ARG_START:
2185 prctl_map.arg_start = addr;
2187 case PR_SET_MM_ARG_END:
2188 prctl_map.arg_end = addr;
2190 case PR_SET_MM_ENV_START:
2191 prctl_map.env_start = addr;
2193 case PR_SET_MM_ENV_END:
2194 prctl_map.env_end = addr;
2200 error = validate_prctl_map_addr(&prctl_map);
2206 * If command line arguments and environment
2207 * are placed somewhere else on stack, we can
2208 * set them up here, ARG_START/END to setup
2209 * command line arguments and ENV_START/END
2212 case PR_SET_MM_START_STACK:
2213 case PR_SET_MM_ARG_START:
2214 case PR_SET_MM_ARG_END:
2215 case PR_SET_MM_ENV_START:
2216 case PR_SET_MM_ENV_END:
2223 mm->start_code = prctl_map.start_code;
2224 mm->end_code = prctl_map.end_code;
2225 mm->start_data = prctl_map.start_data;
2226 mm->end_data = prctl_map.end_data;
2227 mm->start_brk = prctl_map.start_brk;
2228 mm->brk = prctl_map.brk;
2229 mm->start_stack = prctl_map.start_stack;
2230 mm->arg_start = prctl_map.arg_start;
2231 mm->arg_end = prctl_map.arg_end;
2232 mm->env_start = prctl_map.env_start;
2233 mm->env_end = prctl_map.env_end;
2237 spin_unlock(&mm->arg_lock);
2238 mmap_read_unlock(mm);
2242 #ifdef CONFIG_CHECKPOINT_RESTORE
2243 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2245 return put_user(me->clear_child_tid, tid_addr);
2248 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2254 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2257 * If task has has_child_subreaper - all its descendants
2258 * already have these flag too and new descendants will
2259 * inherit it on fork, skip them.
2261 * If we've found child_reaper - skip descendants in
2262 * it's subtree as they will never get out pidns.
2264 if (p->signal->has_child_subreaper ||
2265 is_child_reaper(task_pid(p)))
2268 p->signal->has_child_subreaper = 1;
2272 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2277 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2283 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2285 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2286 unsigned long, arg4, unsigned long, arg5)
2288 struct task_struct *me = current;
2289 unsigned char comm[sizeof(me->comm)];
2292 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2293 if (error != -ENOSYS)
2298 case PR_SET_PDEATHSIG:
2299 if (!valid_signal(arg2)) {
2303 me->pdeath_signal = arg2;
2305 case PR_GET_PDEATHSIG:
2306 error = put_user(me->pdeath_signal, (int __user *)arg2);
2308 case PR_GET_DUMPABLE:
2309 error = get_dumpable(me->mm);
2311 case PR_SET_DUMPABLE:
2312 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2316 set_dumpable(me->mm, arg2);
2319 case PR_SET_UNALIGN:
2320 error = SET_UNALIGN_CTL(me, arg2);
2322 case PR_GET_UNALIGN:
2323 error = GET_UNALIGN_CTL(me, arg2);
2326 error = SET_FPEMU_CTL(me, arg2);
2329 error = GET_FPEMU_CTL(me, arg2);
2332 error = SET_FPEXC_CTL(me, arg2);
2335 error = GET_FPEXC_CTL(me, arg2);
2338 error = PR_TIMING_STATISTICAL;
2341 if (arg2 != PR_TIMING_STATISTICAL)
2345 comm[sizeof(me->comm) - 1] = 0;
2346 if (strncpy_from_user(comm, (char __user *)arg2,
2347 sizeof(me->comm) - 1) < 0)
2349 set_task_comm(me, comm);
2350 proc_comm_connector(me);
2353 get_task_comm(comm, me);
2354 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2358 error = GET_ENDIAN(me, arg2);
2361 error = SET_ENDIAN(me, arg2);
2363 case PR_GET_SECCOMP:
2364 error = prctl_get_seccomp();
2366 case PR_SET_SECCOMP:
2367 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2370 error = GET_TSC_CTL(arg2);
2373 error = SET_TSC_CTL(arg2);
2375 case PR_TASK_PERF_EVENTS_DISABLE:
2376 error = perf_event_task_disable();
2378 case PR_TASK_PERF_EVENTS_ENABLE:
2379 error = perf_event_task_enable();
2381 case PR_GET_TIMERSLACK:
2382 if (current->timer_slack_ns > ULONG_MAX)
2385 error = current->timer_slack_ns;
2387 case PR_SET_TIMERSLACK:
2389 current->timer_slack_ns =
2390 current->default_timer_slack_ns;
2392 current->timer_slack_ns = arg2;
2398 case PR_MCE_KILL_CLEAR:
2401 current->flags &= ~PF_MCE_PROCESS;
2403 case PR_MCE_KILL_SET:
2404 current->flags |= PF_MCE_PROCESS;
2405 if (arg3 == PR_MCE_KILL_EARLY)
2406 current->flags |= PF_MCE_EARLY;
2407 else if (arg3 == PR_MCE_KILL_LATE)
2408 current->flags &= ~PF_MCE_EARLY;
2409 else if (arg3 == PR_MCE_KILL_DEFAULT)
2411 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2419 case PR_MCE_KILL_GET:
2420 if (arg2 | arg3 | arg4 | arg5)
2422 if (current->flags & PF_MCE_PROCESS)
2423 error = (current->flags & PF_MCE_EARLY) ?
2424 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2426 error = PR_MCE_KILL_DEFAULT;
2429 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2431 case PR_GET_TID_ADDRESS:
2432 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2434 case PR_SET_CHILD_SUBREAPER:
2435 me->signal->is_child_subreaper = !!arg2;
2439 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2441 case PR_GET_CHILD_SUBREAPER:
2442 error = put_user(me->signal->is_child_subreaper,
2443 (int __user *)arg2);
2445 case PR_SET_NO_NEW_PRIVS:
2446 if (arg2 != 1 || arg3 || arg4 || arg5)
2449 task_set_no_new_privs(current);
2451 case PR_GET_NO_NEW_PRIVS:
2452 if (arg2 || arg3 || arg4 || arg5)
2454 return task_no_new_privs(current) ? 1 : 0;
2455 case PR_GET_THP_DISABLE:
2456 if (arg2 || arg3 || arg4 || arg5)
2458 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2460 case PR_SET_THP_DISABLE:
2461 if (arg3 || arg4 || arg5)
2463 if (mmap_write_lock_killable(me->mm))
2466 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2468 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2469 mmap_write_unlock(me->mm);
2471 case PR_MPX_ENABLE_MANAGEMENT:
2472 case PR_MPX_DISABLE_MANAGEMENT:
2473 /* No longer implemented: */
2475 case PR_SET_FP_MODE:
2476 error = SET_FP_MODE(me, arg2);
2478 case PR_GET_FP_MODE:
2479 error = GET_FP_MODE(me);
2482 error = SVE_SET_VL(arg2);
2485 error = SVE_GET_VL();
2487 case PR_GET_SPECULATION_CTRL:
2488 if (arg3 || arg4 || arg5)
2490 error = arch_prctl_spec_ctrl_get(me, arg2);
2492 case PR_SET_SPECULATION_CTRL:
2495 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2497 case PR_PAC_RESET_KEYS:
2498 if (arg3 || arg4 || arg5)
2500 error = PAC_RESET_KEYS(me, arg2);
2502 case PR_PAC_SET_ENABLED_KEYS:
2505 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2507 case PR_PAC_GET_ENABLED_KEYS:
2508 if (arg2 || arg3 || arg4 || arg5)
2510 error = PAC_GET_ENABLED_KEYS(me);
2512 case PR_SET_TAGGED_ADDR_CTRL:
2513 if (arg3 || arg4 || arg5)
2515 error = SET_TAGGED_ADDR_CTRL(arg2);
2517 case PR_GET_TAGGED_ADDR_CTRL:
2518 if (arg2 || arg3 || arg4 || arg5)
2520 error = GET_TAGGED_ADDR_CTRL();
2522 case PR_SET_IO_FLUSHER:
2523 if (!capable(CAP_SYS_RESOURCE))
2526 if (arg3 || arg4 || arg5)
2530 current->flags |= PR_IO_FLUSHER;
2532 current->flags &= ~PR_IO_FLUSHER;
2536 case PR_GET_IO_FLUSHER:
2537 if (!capable(CAP_SYS_RESOURCE))
2540 if (arg2 || arg3 || arg4 || arg5)
2543 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2545 case PR_SET_SYSCALL_USER_DISPATCH:
2546 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2547 (char __user *) arg5);
2549 #ifdef CONFIG_SCHED_CORE
2551 error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2561 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2562 struct getcpu_cache __user *, unused)
2565 int cpu = raw_smp_processor_id();
2568 err |= put_user(cpu, cpup);
2570 err |= put_user(cpu_to_node(cpu), nodep);
2571 return err ? -EFAULT : 0;
2575 * do_sysinfo - fill in sysinfo struct
2576 * @info: pointer to buffer to fill
2578 static int do_sysinfo(struct sysinfo *info)
2580 unsigned long mem_total, sav_total;
2581 unsigned int mem_unit, bitcount;
2582 struct timespec64 tp;
2584 memset(info, 0, sizeof(struct sysinfo));
2586 ktime_get_boottime_ts64(&tp);
2587 timens_add_boottime(&tp);
2588 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2590 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2592 info->procs = nr_threads;
2598 * If the sum of all the available memory (i.e. ram + swap)
2599 * is less than can be stored in a 32 bit unsigned long then
2600 * we can be binary compatible with 2.2.x kernels. If not,
2601 * well, in that case 2.2.x was broken anyways...
2603 * -Erik Andersen <andersee@debian.org>
2606 mem_total = info->totalram + info->totalswap;
2607 if (mem_total < info->totalram || mem_total < info->totalswap)
2610 mem_unit = info->mem_unit;
2611 while (mem_unit > 1) {
2614 sav_total = mem_total;
2616 if (mem_total < sav_total)
2621 * If mem_total did not overflow, multiply all memory values by
2622 * info->mem_unit and set it to 1. This leaves things compatible
2623 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2628 info->totalram <<= bitcount;
2629 info->freeram <<= bitcount;
2630 info->sharedram <<= bitcount;
2631 info->bufferram <<= bitcount;
2632 info->totalswap <<= bitcount;
2633 info->freeswap <<= bitcount;
2634 info->totalhigh <<= bitcount;
2635 info->freehigh <<= bitcount;
2641 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2647 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2653 #ifdef CONFIG_COMPAT
2654 struct compat_sysinfo {
2668 char _f[20-2*sizeof(u32)-sizeof(int)];
2671 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2674 struct compat_sysinfo s_32;
2678 /* Check to see if any memory value is too large for 32-bit and scale
2681 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2684 while (s.mem_unit < PAGE_SIZE) {
2689 s.totalram >>= bitcount;
2690 s.freeram >>= bitcount;
2691 s.sharedram >>= bitcount;
2692 s.bufferram >>= bitcount;
2693 s.totalswap >>= bitcount;
2694 s.freeswap >>= bitcount;
2695 s.totalhigh >>= bitcount;
2696 s.freehigh >>= bitcount;
2699 memset(&s_32, 0, sizeof(s_32));
2700 s_32.uptime = s.uptime;
2701 s_32.loads[0] = s.loads[0];
2702 s_32.loads[1] = s.loads[1];
2703 s_32.loads[2] = s.loads[2];
2704 s_32.totalram = s.totalram;
2705 s_32.freeram = s.freeram;
2706 s_32.sharedram = s.sharedram;
2707 s_32.bufferram = s.bufferram;
2708 s_32.totalswap = s.totalswap;
2709 s_32.freeswap = s.freeswap;
2710 s_32.procs = s.procs;
2711 s_32.totalhigh = s.totalhigh;
2712 s_32.freehigh = s.freehigh;
2713 s_32.mem_unit = s.mem_unit;
2714 if (copy_to_user(info, &s_32, sizeof(s_32)))
2718 #endif /* CONFIG_COMPAT */