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/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.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/binfmts.h>
52 #include <linux/sched.h>
53 #include <linux/sched/autogroup.h>
54 #include <linux/sched/loadavg.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/mm.h>
57 #include <linux/sched/coredump.h>
58 #include <linux/sched/task.h>
59 #include <linux/sched/cputime.h>
60 #include <linux/rcupdate.h>
61 #include <linux/uidgid.h>
62 #include <linux/cred.h>
64 #include <linux/nospec.h>
66 #include <linux/kmsg_dump.h>
67 /* Move somewhere else to avoid recompiling? */
68 #include <generated/utsrelease.h>
70 #include <linux/uaccess.h>
72 #include <asm/unistd.h>
76 #ifndef SET_UNALIGN_CTL
77 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
79 #ifndef GET_UNALIGN_CTL
80 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
83 # define SET_FPEMU_CTL(a, b) (-EINVAL)
86 # define GET_FPEMU_CTL(a, b) (-EINVAL)
89 # define SET_FPEXC_CTL(a, b) (-EINVAL)
92 # define GET_FPEXC_CTL(a, b) (-EINVAL)
95 # define GET_ENDIAN(a, b) (-EINVAL)
98 # define SET_ENDIAN(a, b) (-EINVAL)
101 # define GET_TSC_CTL(a) (-EINVAL)
104 # define SET_TSC_CTL(a) (-EINVAL)
106 #ifndef MPX_ENABLE_MANAGEMENT
107 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
109 #ifndef MPX_DISABLE_MANAGEMENT
110 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
113 # define GET_FP_MODE(a) (-EINVAL)
116 # define SET_FP_MODE(a,b) (-EINVAL)
119 # define SVE_SET_VL(a) (-EINVAL)
122 # define SVE_GET_VL() (-EINVAL)
124 #ifndef PAC_RESET_KEYS
125 # define PAC_RESET_KEYS(a, b) (-EINVAL)
129 * this is where the system-wide overflow UID and GID are defined, for
130 * architectures that now have 32-bit UID/GID but didn't in the past
133 int overflowuid = DEFAULT_OVERFLOWUID;
134 int overflowgid = DEFAULT_OVERFLOWGID;
136 EXPORT_SYMBOL(overflowuid);
137 EXPORT_SYMBOL(overflowgid);
140 * the same as above, but for filesystems which can only store a 16-bit
141 * UID and GID. as such, this is needed on all architectures
144 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
145 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
147 EXPORT_SYMBOL(fs_overflowuid);
148 EXPORT_SYMBOL(fs_overflowgid);
151 * Returns true if current's euid is same as p's uid or euid,
152 * or has CAP_SYS_NICE to p's user_ns.
154 * Called with rcu_read_lock, creds are safe
156 static bool set_one_prio_perm(struct task_struct *p)
158 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
160 if (uid_eq(pcred->uid, cred->euid) ||
161 uid_eq(pcred->euid, cred->euid))
163 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
169 * set the priority of a task
170 * - the caller must hold the RCU read lock
172 static int set_one_prio(struct task_struct *p, int niceval, int error)
176 if (!set_one_prio_perm(p)) {
180 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
184 no_nice = security_task_setnice(p, niceval);
191 set_user_nice(p, niceval);
196 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
198 struct task_struct *g, *p;
199 struct user_struct *user;
200 const struct cred *cred = current_cred();
205 if (which > PRIO_USER || which < PRIO_PROCESS)
208 /* normalize: avoid signed division (rounding problems) */
210 if (niceval < MIN_NICE)
212 if (niceval > MAX_NICE)
216 read_lock(&tasklist_lock);
220 p = find_task_by_vpid(who);
224 error = set_one_prio(p, niceval, error);
228 pgrp = find_vpid(who);
230 pgrp = task_pgrp(current);
231 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
232 error = set_one_prio(p, niceval, error);
233 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
236 uid = make_kuid(cred->user_ns, who);
240 else if (!uid_eq(uid, cred->uid)) {
241 user = find_user(uid);
243 goto out_unlock; /* No processes for this user */
245 do_each_thread(g, p) {
246 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
247 error = set_one_prio(p, niceval, error);
248 } while_each_thread(g, p);
249 if (!uid_eq(uid, cred->uid))
250 free_uid(user); /* For find_user() */
254 read_unlock(&tasklist_lock);
261 * Ugh. To avoid negative return values, "getpriority()" will
262 * not return the normal nice-value, but a negated value that
263 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
264 * to stay compatible.
266 SYSCALL_DEFINE2(getpriority, int, which, int, who)
268 struct task_struct *g, *p;
269 struct user_struct *user;
270 const struct cred *cred = current_cred();
271 long niceval, retval = -ESRCH;
275 if (which > PRIO_USER || which < PRIO_PROCESS)
279 read_lock(&tasklist_lock);
283 p = find_task_by_vpid(who);
287 niceval = nice_to_rlimit(task_nice(p));
288 if (niceval > retval)
294 pgrp = find_vpid(who);
296 pgrp = task_pgrp(current);
297 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
298 niceval = nice_to_rlimit(task_nice(p));
299 if (niceval > retval)
301 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
304 uid = make_kuid(cred->user_ns, who);
308 else if (!uid_eq(uid, cred->uid)) {
309 user = find_user(uid);
311 goto out_unlock; /* No processes for this user */
313 do_each_thread(g, p) {
314 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
315 niceval = nice_to_rlimit(task_nice(p));
316 if (niceval > retval)
319 } while_each_thread(g, p);
320 if (!uid_eq(uid, cred->uid))
321 free_uid(user); /* for find_user() */
325 read_unlock(&tasklist_lock);
332 * Unprivileged users may change the real gid to the effective gid
333 * or vice versa. (BSD-style)
335 * If you set the real gid at all, or set the effective gid to a value not
336 * equal to the real gid, then the saved gid is set to the new effective gid.
338 * This makes it possible for a setgid program to completely drop its
339 * privileges, which is often a useful assertion to make when you are doing
340 * a security audit over a program.
342 * The general idea is that a program which uses just setregid() will be
343 * 100% compatible with BSD. A program which uses just setgid() will be
344 * 100% compatible with POSIX with saved IDs.
346 * SMP: There are not races, the GIDs are checked only by filesystem
347 * operations (as far as semantic preservation is concerned).
349 #ifdef CONFIG_MULTIUSER
350 long __sys_setregid(gid_t rgid, gid_t egid)
352 struct user_namespace *ns = current_user_ns();
353 const struct cred *old;
358 krgid = make_kgid(ns, rgid);
359 kegid = make_kgid(ns, egid);
361 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
363 if ((egid != (gid_t) -1) && !gid_valid(kegid))
366 new = prepare_creds();
369 old = current_cred();
372 if (rgid != (gid_t) -1) {
373 if (gid_eq(old->gid, krgid) ||
374 gid_eq(old->egid, krgid) ||
375 ns_capable(old->user_ns, CAP_SETGID))
380 if (egid != (gid_t) -1) {
381 if (gid_eq(old->gid, kegid) ||
382 gid_eq(old->egid, kegid) ||
383 gid_eq(old->sgid, kegid) ||
384 ns_capable(old->user_ns, CAP_SETGID))
390 if (rgid != (gid_t) -1 ||
391 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
392 new->sgid = new->egid;
393 new->fsgid = new->egid;
395 return commit_creds(new);
402 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
404 return __sys_setregid(rgid, egid);
408 * setgid() is implemented like SysV w/ SAVED_IDS
410 * SMP: Same implicit races as above.
412 long __sys_setgid(gid_t gid)
414 struct user_namespace *ns = current_user_ns();
415 const struct cred *old;
420 kgid = make_kgid(ns, gid);
421 if (!gid_valid(kgid))
424 new = prepare_creds();
427 old = current_cred();
430 if (ns_capable(old->user_ns, CAP_SETGID))
431 new->gid = new->egid = new->sgid = new->fsgid = kgid;
432 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
433 new->egid = new->fsgid = kgid;
437 return commit_creds(new);
444 SYSCALL_DEFINE1(setgid, gid_t, gid)
446 return __sys_setgid(gid);
450 * change the user struct in a credentials set to match the new UID
452 static int set_user(struct cred *new)
454 struct user_struct *new_user;
456 new_user = alloc_uid(new->uid);
461 * We don't fail in case of NPROC limit excess here because too many
462 * poorly written programs don't check set*uid() return code, assuming
463 * it never fails if called by root. We may still enforce NPROC limit
464 * for programs doing set*uid()+execve() by harmlessly deferring the
465 * failure to the execve() stage.
467 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
468 new_user != INIT_USER)
469 current->flags |= PF_NPROC_EXCEEDED;
471 current->flags &= ~PF_NPROC_EXCEEDED;
474 new->user = new_user;
479 * Unprivileged users may change the real uid to the effective uid
480 * or vice versa. (BSD-style)
482 * If you set the real uid at all, or set the effective uid to a value not
483 * equal to the real uid, then the saved uid is set to the new effective uid.
485 * This makes it possible for a setuid program to completely drop its
486 * privileges, which is often a useful assertion to make when you are doing
487 * a security audit over a program.
489 * The general idea is that a program which uses just setreuid() will be
490 * 100% compatible with BSD. A program which uses just setuid() will be
491 * 100% compatible with POSIX with saved IDs.
493 long __sys_setreuid(uid_t ruid, uid_t euid)
495 struct user_namespace *ns = current_user_ns();
496 const struct cred *old;
501 kruid = make_kuid(ns, ruid);
502 keuid = make_kuid(ns, euid);
504 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
506 if ((euid != (uid_t) -1) && !uid_valid(keuid))
509 new = prepare_creds();
512 old = current_cred();
515 if (ruid != (uid_t) -1) {
517 if (!uid_eq(old->uid, kruid) &&
518 !uid_eq(old->euid, kruid) &&
519 !ns_capable_setid(old->user_ns, CAP_SETUID))
523 if (euid != (uid_t) -1) {
525 if (!uid_eq(old->uid, keuid) &&
526 !uid_eq(old->euid, keuid) &&
527 !uid_eq(old->suid, keuid) &&
528 !ns_capable_setid(old->user_ns, CAP_SETUID))
532 if (!uid_eq(new->uid, old->uid)) {
533 retval = set_user(new);
537 if (ruid != (uid_t) -1 ||
538 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
539 new->suid = new->euid;
540 new->fsuid = new->euid;
542 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
546 return commit_creds(new);
553 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
555 return __sys_setreuid(ruid, euid);
559 * setuid() is implemented like SysV with SAVED_IDS
561 * Note that SAVED_ID's is deficient in that a setuid root program
562 * like sendmail, for example, cannot set its uid to be a normal
563 * user and then switch back, because if you're root, setuid() sets
564 * the saved uid too. If you don't like this, blame the bright people
565 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
566 * will allow a root program to temporarily drop privileges and be able to
567 * regain them by swapping the real and effective uid.
569 long __sys_setuid(uid_t uid)
571 struct user_namespace *ns = current_user_ns();
572 const struct cred *old;
577 kuid = make_kuid(ns, uid);
578 if (!uid_valid(kuid))
581 new = prepare_creds();
584 old = current_cred();
587 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
588 new->suid = new->uid = kuid;
589 if (!uid_eq(kuid, old->uid)) {
590 retval = set_user(new);
594 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
598 new->fsuid = new->euid = kuid;
600 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
604 return commit_creds(new);
611 SYSCALL_DEFINE1(setuid, uid_t, uid)
613 return __sys_setuid(uid);
618 * This function implements a generic ability to update ruid, euid,
619 * and suid. This allows you to implement the 4.4 compatible seteuid().
621 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
623 struct user_namespace *ns = current_user_ns();
624 const struct cred *old;
627 kuid_t kruid, keuid, ksuid;
629 kruid = make_kuid(ns, ruid);
630 keuid = make_kuid(ns, euid);
631 ksuid = make_kuid(ns, suid);
633 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
636 if ((euid != (uid_t) -1) && !uid_valid(keuid))
639 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
642 new = prepare_creds();
646 old = current_cred();
649 if (!ns_capable_setid(old->user_ns, CAP_SETUID)) {
650 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
651 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
653 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
654 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
656 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
657 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
661 if (ruid != (uid_t) -1) {
663 if (!uid_eq(kruid, old->uid)) {
664 retval = set_user(new);
669 if (euid != (uid_t) -1)
671 if (suid != (uid_t) -1)
673 new->fsuid = new->euid;
675 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
679 return commit_creds(new);
686 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
688 return __sys_setresuid(ruid, euid, suid);
691 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
693 const struct cred *cred = current_cred();
695 uid_t ruid, euid, suid;
697 ruid = from_kuid_munged(cred->user_ns, cred->uid);
698 euid = from_kuid_munged(cred->user_ns, cred->euid);
699 suid = from_kuid_munged(cred->user_ns, cred->suid);
701 retval = put_user(ruid, ruidp);
703 retval = put_user(euid, euidp);
705 return put_user(suid, suidp);
711 * Same as above, but for rgid, egid, sgid.
713 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
715 struct user_namespace *ns = current_user_ns();
716 const struct cred *old;
719 kgid_t krgid, kegid, ksgid;
721 krgid = make_kgid(ns, rgid);
722 kegid = make_kgid(ns, egid);
723 ksgid = make_kgid(ns, sgid);
725 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
727 if ((egid != (gid_t) -1) && !gid_valid(kegid))
729 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
732 new = prepare_creds();
735 old = current_cred();
738 if (!ns_capable(old->user_ns, CAP_SETGID)) {
739 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
740 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
742 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
743 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
745 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
746 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
750 if (rgid != (gid_t) -1)
752 if (egid != (gid_t) -1)
754 if (sgid != (gid_t) -1)
756 new->fsgid = new->egid;
758 return commit_creds(new);
765 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
767 return __sys_setresgid(rgid, egid, sgid);
770 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
772 const struct cred *cred = current_cred();
774 gid_t rgid, egid, sgid;
776 rgid = from_kgid_munged(cred->user_ns, cred->gid);
777 egid = from_kgid_munged(cred->user_ns, cred->egid);
778 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
780 retval = put_user(rgid, rgidp);
782 retval = put_user(egid, egidp);
784 retval = put_user(sgid, sgidp);
792 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
793 * is used for "access()" and for the NFS daemon (letting nfsd stay at
794 * whatever uid it wants to). It normally shadows "euid", except when
795 * explicitly set by setfsuid() or for access..
797 long __sys_setfsuid(uid_t uid)
799 const struct cred *old;
804 old = current_cred();
805 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
807 kuid = make_kuid(old->user_ns, uid);
808 if (!uid_valid(kuid))
811 new = prepare_creds();
815 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
816 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
817 ns_capable_setid(old->user_ns, CAP_SETUID)) {
818 if (!uid_eq(kuid, old->fsuid)) {
820 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
833 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
835 return __sys_setfsuid(uid);
839 * Samma på svenska..
841 long __sys_setfsgid(gid_t gid)
843 const struct cred *old;
848 old = current_cred();
849 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
851 kgid = make_kgid(old->user_ns, gid);
852 if (!gid_valid(kgid))
855 new = prepare_creds();
859 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
860 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
861 ns_capable(old->user_ns, CAP_SETGID)) {
862 if (!gid_eq(kgid, old->fsgid)) {
876 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
878 return __sys_setfsgid(gid);
880 #endif /* CONFIG_MULTIUSER */
883 * sys_getpid - return the thread group id of the current process
885 * Note, despite the name, this returns the tgid not the pid. The tgid and
886 * the pid are identical unless CLONE_THREAD was specified on clone() in
887 * which case the tgid is the same in all threads of the same group.
889 * This is SMP safe as current->tgid does not change.
891 SYSCALL_DEFINE0(getpid)
893 return task_tgid_vnr(current);
896 /* Thread ID - the internal kernel "pid" */
897 SYSCALL_DEFINE0(gettid)
899 return task_pid_vnr(current);
903 * Accessing ->real_parent is not SMP-safe, it could
904 * change from under us. However, we can use a stale
905 * value of ->real_parent under rcu_read_lock(), see
906 * release_task()->call_rcu(delayed_put_task_struct).
908 SYSCALL_DEFINE0(getppid)
913 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
919 SYSCALL_DEFINE0(getuid)
921 /* Only we change this so SMP safe */
922 return from_kuid_munged(current_user_ns(), current_uid());
925 SYSCALL_DEFINE0(geteuid)
927 /* Only we change this so SMP safe */
928 return from_kuid_munged(current_user_ns(), current_euid());
931 SYSCALL_DEFINE0(getgid)
933 /* Only we change this so SMP safe */
934 return from_kgid_munged(current_user_ns(), current_gid());
937 SYSCALL_DEFINE0(getegid)
939 /* Only we change this so SMP safe */
940 return from_kgid_munged(current_user_ns(), current_egid());
943 static void do_sys_times(struct tms *tms)
945 u64 tgutime, tgstime, cutime, cstime;
947 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
948 cutime = current->signal->cutime;
949 cstime = current->signal->cstime;
950 tms->tms_utime = nsec_to_clock_t(tgutime);
951 tms->tms_stime = nsec_to_clock_t(tgstime);
952 tms->tms_cutime = nsec_to_clock_t(cutime);
953 tms->tms_cstime = nsec_to_clock_t(cstime);
956 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
962 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
965 force_successful_syscall_return();
966 return (long) jiffies_64_to_clock_t(get_jiffies_64());
970 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
972 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
975 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
979 struct compat_tms tmp;
982 /* Convert our struct tms to the compat version. */
983 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
984 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
985 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
986 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
987 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
990 force_successful_syscall_return();
991 return compat_jiffies_to_clock_t(jiffies);
996 * This needs some heavy checking ...
997 * I just haven't the stomach for it. I also don't fully
998 * understand sessions/pgrp etc. Let somebody who does explain it.
1000 * OK, I think I have the protection semantics right.... this is really
1001 * only important on a multi-user system anyway, to make sure one user
1002 * can't send a signal to a process owned by another. -TYT, 12/12/91
1004 * !PF_FORKNOEXEC check to conform completely to POSIX.
1006 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1008 struct task_struct *p;
1009 struct task_struct *group_leader = current->group_leader;
1014 pid = task_pid_vnr(group_leader);
1021 /* From this point forward we keep holding onto the tasklist lock
1022 * so that our parent does not change from under us. -DaveM
1024 write_lock_irq(&tasklist_lock);
1027 p = find_task_by_vpid(pid);
1032 if (!thread_group_leader(p))
1035 if (same_thread_group(p->real_parent, group_leader)) {
1037 if (task_session(p) != task_session(group_leader))
1040 if (!(p->flags & PF_FORKNOEXEC))
1044 if (p != group_leader)
1049 if (p->signal->leader)
1054 struct task_struct *g;
1056 pgrp = find_vpid(pgid);
1057 g = pid_task(pgrp, PIDTYPE_PGID);
1058 if (!g || task_session(g) != task_session(group_leader))
1062 err = security_task_setpgid(p, pgid);
1066 if (task_pgrp(p) != pgrp)
1067 change_pid(p, PIDTYPE_PGID, pgrp);
1071 /* All paths lead to here, thus we are safe. -DaveM */
1072 write_unlock_irq(&tasklist_lock);
1077 static int do_getpgid(pid_t pid)
1079 struct task_struct *p;
1085 grp = task_pgrp(current);
1088 p = find_task_by_vpid(pid);
1095 retval = security_task_getpgid(p);
1099 retval = pid_vnr(grp);
1105 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1107 return do_getpgid(pid);
1110 #ifdef __ARCH_WANT_SYS_GETPGRP
1112 SYSCALL_DEFINE0(getpgrp)
1114 return do_getpgid(0);
1119 SYSCALL_DEFINE1(getsid, pid_t, pid)
1121 struct task_struct *p;
1127 sid = task_session(current);
1130 p = find_task_by_vpid(pid);
1133 sid = task_session(p);
1137 retval = security_task_getsid(p);
1141 retval = pid_vnr(sid);
1147 static void set_special_pids(struct pid *pid)
1149 struct task_struct *curr = current->group_leader;
1151 if (task_session(curr) != pid)
1152 change_pid(curr, PIDTYPE_SID, pid);
1154 if (task_pgrp(curr) != pid)
1155 change_pid(curr, PIDTYPE_PGID, pid);
1158 int ksys_setsid(void)
1160 struct task_struct *group_leader = current->group_leader;
1161 struct pid *sid = task_pid(group_leader);
1162 pid_t session = pid_vnr(sid);
1165 write_lock_irq(&tasklist_lock);
1166 /* Fail if I am already a session leader */
1167 if (group_leader->signal->leader)
1170 /* Fail if a process group id already exists that equals the
1171 * proposed session id.
1173 if (pid_task(sid, PIDTYPE_PGID))
1176 group_leader->signal->leader = 1;
1177 set_special_pids(sid);
1179 proc_clear_tty(group_leader);
1183 write_unlock_irq(&tasklist_lock);
1185 proc_sid_connector(group_leader);
1186 sched_autogroup_create_attach(group_leader);
1191 SYSCALL_DEFINE0(setsid)
1193 return ksys_setsid();
1196 DECLARE_RWSEM(uts_sem);
1198 #ifdef COMPAT_UTS_MACHINE
1199 #define override_architecture(name) \
1200 (personality(current->personality) == PER_LINUX32 && \
1201 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1202 sizeof(COMPAT_UTS_MACHINE)))
1204 #define override_architecture(name) 0
1208 * Work around broken programs that cannot handle "Linux 3.0".
1209 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1210 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1213 static int override_release(char __user *release, size_t len)
1217 if (current->personality & UNAME26) {
1218 const char *rest = UTS_RELEASE;
1219 char buf[65] = { 0 };
1225 if (*rest == '.' && ++ndots >= 3)
1227 if (!isdigit(*rest) && *rest != '.')
1231 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1232 copy = clamp_t(size_t, len, 1, sizeof(buf));
1233 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1234 ret = copy_to_user(release, buf, copy + 1);
1239 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1241 struct new_utsname tmp;
1243 down_read(&uts_sem);
1244 memcpy(&tmp, utsname(), sizeof(tmp));
1246 if (copy_to_user(name, &tmp, sizeof(tmp)))
1249 if (override_release(name->release, sizeof(name->release)))
1251 if (override_architecture(name))
1256 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1260 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1262 struct old_utsname tmp;
1267 down_read(&uts_sem);
1268 memcpy(&tmp, utsname(), sizeof(tmp));
1270 if (copy_to_user(name, &tmp, sizeof(tmp)))
1273 if (override_release(name->release, sizeof(name->release)))
1275 if (override_architecture(name))
1280 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1282 struct oldold_utsname tmp = {};
1287 down_read(&uts_sem);
1288 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1289 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1290 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1291 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1292 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1294 if (copy_to_user(name, &tmp, sizeof(tmp)))
1297 if (override_architecture(name))
1299 if (override_release(name->release, sizeof(name->release)))
1305 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1308 char tmp[__NEW_UTS_LEN];
1310 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1313 if (len < 0 || len > __NEW_UTS_LEN)
1316 if (!copy_from_user(tmp, name, len)) {
1317 struct new_utsname *u;
1319 down_write(&uts_sem);
1321 memcpy(u->nodename, tmp, len);
1322 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1324 uts_proc_notify(UTS_PROC_HOSTNAME);
1330 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1332 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1335 struct new_utsname *u;
1336 char tmp[__NEW_UTS_LEN + 1];
1340 down_read(&uts_sem);
1342 i = 1 + strlen(u->nodename);
1345 memcpy(tmp, u->nodename, i);
1347 if (copy_to_user(name, tmp, i))
1355 * Only setdomainname; getdomainname can be implemented by calling
1358 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1361 char tmp[__NEW_UTS_LEN];
1363 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1365 if (len < 0 || len > __NEW_UTS_LEN)
1369 if (!copy_from_user(tmp, name, len)) {
1370 struct new_utsname *u;
1372 down_write(&uts_sem);
1374 memcpy(u->domainname, tmp, len);
1375 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1377 uts_proc_notify(UTS_PROC_DOMAINNAME);
1383 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1385 struct rlimit value;
1388 ret = do_prlimit(current, resource, NULL, &value);
1390 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1395 #ifdef CONFIG_COMPAT
1397 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1398 struct compat_rlimit __user *, rlim)
1401 struct compat_rlimit r32;
1403 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1406 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1407 r.rlim_cur = RLIM_INFINITY;
1409 r.rlim_cur = r32.rlim_cur;
1410 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1411 r.rlim_max = RLIM_INFINITY;
1413 r.rlim_max = r32.rlim_max;
1414 return do_prlimit(current, resource, &r, NULL);
1417 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1418 struct compat_rlimit __user *, rlim)
1423 ret = do_prlimit(current, resource, NULL, &r);
1425 struct compat_rlimit r32;
1426 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1427 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1429 r32.rlim_cur = r.rlim_cur;
1430 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1431 r32.rlim_max = COMPAT_RLIM_INFINITY;
1433 r32.rlim_max = r.rlim_max;
1435 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1443 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1446 * Back compatibility for getrlimit. Needed for some apps.
1448 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1449 struct rlimit __user *, rlim)
1452 if (resource >= RLIM_NLIMITS)
1455 resource = array_index_nospec(resource, RLIM_NLIMITS);
1456 task_lock(current->group_leader);
1457 x = current->signal->rlim[resource];
1458 task_unlock(current->group_leader);
1459 if (x.rlim_cur > 0x7FFFFFFF)
1460 x.rlim_cur = 0x7FFFFFFF;
1461 if (x.rlim_max > 0x7FFFFFFF)
1462 x.rlim_max = 0x7FFFFFFF;
1463 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1466 #ifdef CONFIG_COMPAT
1467 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1468 struct compat_rlimit __user *, rlim)
1472 if (resource >= RLIM_NLIMITS)
1475 resource = array_index_nospec(resource, RLIM_NLIMITS);
1476 task_lock(current->group_leader);
1477 r = current->signal->rlim[resource];
1478 task_unlock(current->group_leader);
1479 if (r.rlim_cur > 0x7FFFFFFF)
1480 r.rlim_cur = 0x7FFFFFFF;
1481 if (r.rlim_max > 0x7FFFFFFF)
1482 r.rlim_max = 0x7FFFFFFF;
1484 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1485 put_user(r.rlim_max, &rlim->rlim_max))
1493 static inline bool rlim64_is_infinity(__u64 rlim64)
1495 #if BITS_PER_LONG < 64
1496 return rlim64 >= ULONG_MAX;
1498 return rlim64 == RLIM64_INFINITY;
1502 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1504 if (rlim->rlim_cur == RLIM_INFINITY)
1505 rlim64->rlim_cur = RLIM64_INFINITY;
1507 rlim64->rlim_cur = rlim->rlim_cur;
1508 if (rlim->rlim_max == RLIM_INFINITY)
1509 rlim64->rlim_max = RLIM64_INFINITY;
1511 rlim64->rlim_max = rlim->rlim_max;
1514 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1516 if (rlim64_is_infinity(rlim64->rlim_cur))
1517 rlim->rlim_cur = RLIM_INFINITY;
1519 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1520 if (rlim64_is_infinity(rlim64->rlim_max))
1521 rlim->rlim_max = RLIM_INFINITY;
1523 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1526 /* make sure you are allowed to change @tsk limits before calling this */
1527 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1528 struct rlimit *new_rlim, struct rlimit *old_rlim)
1530 struct rlimit *rlim;
1533 if (resource >= RLIM_NLIMITS)
1536 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1538 if (resource == RLIMIT_NOFILE &&
1539 new_rlim->rlim_max > sysctl_nr_open)
1543 /* protect tsk->signal and tsk->sighand from disappearing */
1544 read_lock(&tasklist_lock);
1545 if (!tsk->sighand) {
1550 rlim = tsk->signal->rlim + resource;
1551 task_lock(tsk->group_leader);
1553 /* Keep the capable check against init_user_ns until
1554 cgroups can contain all limits */
1555 if (new_rlim->rlim_max > rlim->rlim_max &&
1556 !capable(CAP_SYS_RESOURCE))
1559 retval = security_task_setrlimit(tsk, resource, new_rlim);
1560 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1562 * The caller is asking for an immediate RLIMIT_CPU
1563 * expiry. But we use the zero value to mean "it was
1564 * never set". So let's cheat and make it one second
1567 new_rlim->rlim_cur = 1;
1576 task_unlock(tsk->group_leader);
1579 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1580 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1581 * very long-standing error, and fixing it now risks breakage of
1582 * applications, so we live with it
1584 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1585 new_rlim->rlim_cur != RLIM_INFINITY &&
1586 IS_ENABLED(CONFIG_POSIX_TIMERS))
1587 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1589 read_unlock(&tasklist_lock);
1593 /* rcu lock must be held */
1594 static int check_prlimit_permission(struct task_struct *task,
1597 const struct cred *cred = current_cred(), *tcred;
1600 if (current == task)
1603 tcred = __task_cred(task);
1604 id_match = (uid_eq(cred->uid, tcred->euid) &&
1605 uid_eq(cred->uid, tcred->suid) &&
1606 uid_eq(cred->uid, tcred->uid) &&
1607 gid_eq(cred->gid, tcred->egid) &&
1608 gid_eq(cred->gid, tcred->sgid) &&
1609 gid_eq(cred->gid, tcred->gid));
1610 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1613 return security_task_prlimit(cred, tcred, flags);
1616 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1617 const struct rlimit64 __user *, new_rlim,
1618 struct rlimit64 __user *, old_rlim)
1620 struct rlimit64 old64, new64;
1621 struct rlimit old, new;
1622 struct task_struct *tsk;
1623 unsigned int checkflags = 0;
1627 checkflags |= LSM_PRLIMIT_READ;
1630 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1632 rlim64_to_rlim(&new64, &new);
1633 checkflags |= LSM_PRLIMIT_WRITE;
1637 tsk = pid ? find_task_by_vpid(pid) : current;
1642 ret = check_prlimit_permission(tsk, checkflags);
1647 get_task_struct(tsk);
1650 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1651 old_rlim ? &old : NULL);
1653 if (!ret && old_rlim) {
1654 rlim_to_rlim64(&old, &old64);
1655 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1659 put_task_struct(tsk);
1663 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1665 struct rlimit new_rlim;
1667 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1669 return do_prlimit(current, resource, &new_rlim, NULL);
1673 * It would make sense to put struct rusage in the task_struct,
1674 * except that would make the task_struct be *really big*. After
1675 * task_struct gets moved into malloc'ed memory, it would
1676 * make sense to do this. It will make moving the rest of the information
1677 * a lot simpler! (Which we're not doing right now because we're not
1678 * measuring them yet).
1680 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1681 * races with threads incrementing their own counters. But since word
1682 * reads are atomic, we either get new values or old values and we don't
1683 * care which for the sums. We always take the siglock to protect reading
1684 * the c* fields from p->signal from races with exit.c updating those
1685 * fields when reaping, so a sample either gets all the additions of a
1686 * given child after it's reaped, or none so this sample is before reaping.
1689 * We need to take the siglock for CHILDEREN, SELF and BOTH
1690 * for the cases current multithreaded, non-current single threaded
1691 * non-current multithreaded. Thread traversal is now safe with
1693 * Strictly speaking, we donot need to take the siglock if we are current and
1694 * single threaded, as no one else can take our signal_struct away, no one
1695 * else can reap the children to update signal->c* counters, and no one else
1696 * can race with the signal-> fields. If we do not take any lock, the
1697 * signal-> fields could be read out of order while another thread was just
1698 * exiting. So we should place a read memory barrier when we avoid the lock.
1699 * On the writer side, write memory barrier is implied in __exit_signal
1700 * as __exit_signal releases the siglock spinlock after updating the signal->
1701 * fields. But we don't do this yet to keep things simple.
1705 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1707 r->ru_nvcsw += t->nvcsw;
1708 r->ru_nivcsw += t->nivcsw;
1709 r->ru_minflt += t->min_flt;
1710 r->ru_majflt += t->maj_flt;
1711 r->ru_inblock += task_io_get_inblock(t);
1712 r->ru_oublock += task_io_get_oublock(t);
1715 void getrusage(struct task_struct *p, int who, struct rusage *r)
1717 struct task_struct *t;
1718 unsigned long flags;
1719 u64 tgutime, tgstime, utime, stime;
1720 unsigned long maxrss = 0;
1722 memset((char *)r, 0, sizeof (*r));
1725 if (who == RUSAGE_THREAD) {
1726 task_cputime_adjusted(current, &utime, &stime);
1727 accumulate_thread_rusage(p, r);
1728 maxrss = p->signal->maxrss;
1732 if (!lock_task_sighand(p, &flags))
1737 case RUSAGE_CHILDREN:
1738 utime = p->signal->cutime;
1739 stime = p->signal->cstime;
1740 r->ru_nvcsw = p->signal->cnvcsw;
1741 r->ru_nivcsw = p->signal->cnivcsw;
1742 r->ru_minflt = p->signal->cmin_flt;
1743 r->ru_majflt = p->signal->cmaj_flt;
1744 r->ru_inblock = p->signal->cinblock;
1745 r->ru_oublock = p->signal->coublock;
1746 maxrss = p->signal->cmaxrss;
1748 if (who == RUSAGE_CHILDREN)
1753 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1756 r->ru_nvcsw += p->signal->nvcsw;
1757 r->ru_nivcsw += p->signal->nivcsw;
1758 r->ru_minflt += p->signal->min_flt;
1759 r->ru_majflt += p->signal->maj_flt;
1760 r->ru_inblock += p->signal->inblock;
1761 r->ru_oublock += p->signal->oublock;
1762 if (maxrss < p->signal->maxrss)
1763 maxrss = p->signal->maxrss;
1766 accumulate_thread_rusage(t, r);
1767 } while_each_thread(p, t);
1773 unlock_task_sighand(p, &flags);
1776 r->ru_utime = ns_to_timeval(utime);
1777 r->ru_stime = ns_to_timeval(stime);
1779 if (who != RUSAGE_CHILDREN) {
1780 struct mm_struct *mm = get_task_mm(p);
1783 setmax_mm_hiwater_rss(&maxrss, mm);
1787 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1790 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1794 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1795 who != RUSAGE_THREAD)
1798 getrusage(current, who, &r);
1799 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1802 #ifdef CONFIG_COMPAT
1803 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1807 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1808 who != RUSAGE_THREAD)
1811 getrusage(current, who, &r);
1812 return put_compat_rusage(&r, ru);
1816 SYSCALL_DEFINE1(umask, int, mask)
1818 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1822 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1825 struct file *old_exe, *exe_file;
1826 struct inode *inode;
1833 inode = file_inode(exe.file);
1836 * Because the original mm->exe_file points to executable file, make
1837 * sure that this one is executable as well, to avoid breaking an
1841 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1844 err = inode_permission(inode, MAY_EXEC);
1849 * Forbid mm->exe_file change if old file still mapped.
1851 exe_file = get_mm_exe_file(mm);
1854 struct vm_area_struct *vma;
1856 down_read(&mm->mmap_sem);
1857 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1860 if (path_equal(&vma->vm_file->f_path,
1865 up_read(&mm->mmap_sem);
1870 /* set the new file, lockless */
1872 old_exe = xchg(&mm->exe_file, exe.file);
1879 up_read(&mm->mmap_sem);
1885 * Check arithmetic relations of passed addresses.
1887 * WARNING: we don't require any capability here so be very careful
1888 * in what is allowed for modification from userspace.
1890 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1892 unsigned long mmap_max_addr = TASK_SIZE;
1893 int error = -EINVAL, i;
1895 static const unsigned char offsets[] = {
1896 offsetof(struct prctl_mm_map, start_code),
1897 offsetof(struct prctl_mm_map, end_code),
1898 offsetof(struct prctl_mm_map, start_data),
1899 offsetof(struct prctl_mm_map, end_data),
1900 offsetof(struct prctl_mm_map, start_brk),
1901 offsetof(struct prctl_mm_map, brk),
1902 offsetof(struct prctl_mm_map, start_stack),
1903 offsetof(struct prctl_mm_map, arg_start),
1904 offsetof(struct prctl_mm_map, arg_end),
1905 offsetof(struct prctl_mm_map, env_start),
1906 offsetof(struct prctl_mm_map, env_end),
1910 * Make sure the members are not somewhere outside
1911 * of allowed address space.
1913 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1914 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1916 if ((unsigned long)val >= mmap_max_addr ||
1917 (unsigned long)val < mmap_min_addr)
1922 * Make sure the pairs are ordered.
1924 #define __prctl_check_order(__m1, __op, __m2) \
1925 ((unsigned long)prctl_map->__m1 __op \
1926 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1927 error = __prctl_check_order(start_code, <, end_code);
1928 error |= __prctl_check_order(start_data,<=, end_data);
1929 error |= __prctl_check_order(start_brk, <=, brk);
1930 error |= __prctl_check_order(arg_start, <=, arg_end);
1931 error |= __prctl_check_order(env_start, <=, env_end);
1934 #undef __prctl_check_order
1939 * @brk should be after @end_data in traditional maps.
1941 if (prctl_map->start_brk <= prctl_map->end_data ||
1942 prctl_map->brk <= prctl_map->end_data)
1946 * Neither we should allow to override limits if they set.
1948 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1949 prctl_map->start_brk, prctl_map->end_data,
1950 prctl_map->start_data))
1958 #ifdef CONFIG_CHECKPOINT_RESTORE
1959 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1961 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1962 unsigned long user_auxv[AT_VECTOR_SIZE];
1963 struct mm_struct *mm = current->mm;
1966 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1967 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1969 if (opt == PR_SET_MM_MAP_SIZE)
1970 return put_user((unsigned int)sizeof(prctl_map),
1971 (unsigned int __user *)addr);
1973 if (data_size != sizeof(prctl_map))
1976 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1979 error = validate_prctl_map_addr(&prctl_map);
1983 if (prctl_map.auxv_size) {
1985 * Someone is trying to cheat the auxv vector.
1987 if (!prctl_map.auxv ||
1988 prctl_map.auxv_size > sizeof(mm->saved_auxv))
1991 memset(user_auxv, 0, sizeof(user_auxv));
1992 if (copy_from_user(user_auxv,
1993 (const void __user *)prctl_map.auxv,
1994 prctl_map.auxv_size))
1997 /* Last entry must be AT_NULL as specification requires */
1998 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1999 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2002 if (prctl_map.exe_fd != (u32)-1) {
2004 * Make sure the caller has the rights to
2005 * change /proc/pid/exe link: only local sys admin should
2008 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
2011 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2017 * arg_lock protects concurent updates but we still need mmap_sem for
2018 * read to exclude races with sys_brk.
2020 down_read(&mm->mmap_sem);
2023 * We don't validate if these members are pointing to
2024 * real present VMAs because application may have correspond
2025 * VMAs already unmapped and kernel uses these members for statistics
2026 * output in procfs mostly, except
2028 * - @start_brk/@brk which are used in do_brk but kernel lookups
2029 * for VMAs when updating these memvers so anything wrong written
2030 * here cause kernel to swear at userspace program but won't lead
2031 * to any problem in kernel itself
2034 spin_lock(&mm->arg_lock);
2035 mm->start_code = prctl_map.start_code;
2036 mm->end_code = prctl_map.end_code;
2037 mm->start_data = prctl_map.start_data;
2038 mm->end_data = prctl_map.end_data;
2039 mm->start_brk = prctl_map.start_brk;
2040 mm->brk = prctl_map.brk;
2041 mm->start_stack = prctl_map.start_stack;
2042 mm->arg_start = prctl_map.arg_start;
2043 mm->arg_end = prctl_map.arg_end;
2044 mm->env_start = prctl_map.env_start;
2045 mm->env_end = prctl_map.env_end;
2046 spin_unlock(&mm->arg_lock);
2049 * Note this update of @saved_auxv is lockless thus
2050 * if someone reads this member in procfs while we're
2051 * updating -- it may get partly updated results. It's
2052 * known and acceptable trade off: we leave it as is to
2053 * not introduce additional locks here making the kernel
2056 if (prctl_map.auxv_size)
2057 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2059 up_read(&mm->mmap_sem);
2062 #endif /* CONFIG_CHECKPOINT_RESTORE */
2064 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2068 * This doesn't move the auxiliary vector itself since it's pinned to
2069 * mm_struct, but it permits filling the vector with new values. It's
2070 * up to the caller to provide sane values here, otherwise userspace
2071 * tools which use this vector might be unhappy.
2073 unsigned long user_auxv[AT_VECTOR_SIZE];
2075 if (len > sizeof(user_auxv))
2078 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2081 /* Make sure the last entry is always AT_NULL */
2082 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2083 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2085 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2088 memcpy(mm->saved_auxv, user_auxv, len);
2089 task_unlock(current);
2094 static int prctl_set_mm(int opt, unsigned long addr,
2095 unsigned long arg4, unsigned long arg5)
2097 struct mm_struct *mm = current->mm;
2098 struct prctl_mm_map prctl_map = {
2103 struct vm_area_struct *vma;
2106 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2107 opt != PR_SET_MM_MAP &&
2108 opt != PR_SET_MM_MAP_SIZE)))
2111 #ifdef CONFIG_CHECKPOINT_RESTORE
2112 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2113 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2116 if (!capable(CAP_SYS_RESOURCE))
2119 if (opt == PR_SET_MM_EXE_FILE)
2120 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2122 if (opt == PR_SET_MM_AUXV)
2123 return prctl_set_auxv(mm, addr, arg4);
2125 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2131 * arg_lock protects concurent updates of arg boundaries, we need
2132 * mmap_sem for a) concurrent sys_brk, b) finding VMA for addr
2135 down_read(&mm->mmap_sem);
2136 vma = find_vma(mm, addr);
2138 spin_lock(&mm->arg_lock);
2139 prctl_map.start_code = mm->start_code;
2140 prctl_map.end_code = mm->end_code;
2141 prctl_map.start_data = mm->start_data;
2142 prctl_map.end_data = mm->end_data;
2143 prctl_map.start_brk = mm->start_brk;
2144 prctl_map.brk = mm->brk;
2145 prctl_map.start_stack = mm->start_stack;
2146 prctl_map.arg_start = mm->arg_start;
2147 prctl_map.arg_end = mm->arg_end;
2148 prctl_map.env_start = mm->env_start;
2149 prctl_map.env_end = mm->env_end;
2152 case PR_SET_MM_START_CODE:
2153 prctl_map.start_code = addr;
2155 case PR_SET_MM_END_CODE:
2156 prctl_map.end_code = addr;
2158 case PR_SET_MM_START_DATA:
2159 prctl_map.start_data = addr;
2161 case PR_SET_MM_END_DATA:
2162 prctl_map.end_data = addr;
2164 case PR_SET_MM_START_STACK:
2165 prctl_map.start_stack = addr;
2167 case PR_SET_MM_START_BRK:
2168 prctl_map.start_brk = addr;
2171 prctl_map.brk = addr;
2173 case PR_SET_MM_ARG_START:
2174 prctl_map.arg_start = addr;
2176 case PR_SET_MM_ARG_END:
2177 prctl_map.arg_end = addr;
2179 case PR_SET_MM_ENV_START:
2180 prctl_map.env_start = addr;
2182 case PR_SET_MM_ENV_END:
2183 prctl_map.env_end = addr;
2189 error = validate_prctl_map_addr(&prctl_map);
2195 * If command line arguments and environment
2196 * are placed somewhere else on stack, we can
2197 * set them up here, ARG_START/END to setup
2198 * command line argumets and ENV_START/END
2201 case PR_SET_MM_START_STACK:
2202 case PR_SET_MM_ARG_START:
2203 case PR_SET_MM_ARG_END:
2204 case PR_SET_MM_ENV_START:
2205 case PR_SET_MM_ENV_END:
2212 mm->start_code = prctl_map.start_code;
2213 mm->end_code = prctl_map.end_code;
2214 mm->start_data = prctl_map.start_data;
2215 mm->end_data = prctl_map.end_data;
2216 mm->start_brk = prctl_map.start_brk;
2217 mm->brk = prctl_map.brk;
2218 mm->start_stack = prctl_map.start_stack;
2219 mm->arg_start = prctl_map.arg_start;
2220 mm->arg_end = prctl_map.arg_end;
2221 mm->env_start = prctl_map.env_start;
2222 mm->env_end = prctl_map.env_end;
2226 spin_unlock(&mm->arg_lock);
2227 up_read(&mm->mmap_sem);
2231 #ifdef CONFIG_CHECKPOINT_RESTORE
2232 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2234 return put_user(me->clear_child_tid, tid_addr);
2237 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2243 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2246 * If task has has_child_subreaper - all its decendants
2247 * already have these flag too and new decendants will
2248 * inherit it on fork, skip them.
2250 * If we've found child_reaper - skip descendants in
2251 * it's subtree as they will never get out pidns.
2253 if (p->signal->has_child_subreaper ||
2254 is_child_reaper(task_pid(p)))
2257 p->signal->has_child_subreaper = 1;
2261 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2266 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2272 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2273 unsigned long, arg4, unsigned long, arg5)
2275 struct task_struct *me = current;
2276 unsigned char comm[sizeof(me->comm)];
2279 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2280 if (error != -ENOSYS)
2285 case PR_SET_PDEATHSIG:
2286 if (!valid_signal(arg2)) {
2290 me->pdeath_signal = arg2;
2292 case PR_GET_PDEATHSIG:
2293 error = put_user(me->pdeath_signal, (int __user *)arg2);
2295 case PR_GET_DUMPABLE:
2296 error = get_dumpable(me->mm);
2298 case PR_SET_DUMPABLE:
2299 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2303 set_dumpable(me->mm, arg2);
2306 case PR_SET_UNALIGN:
2307 error = SET_UNALIGN_CTL(me, arg2);
2309 case PR_GET_UNALIGN:
2310 error = GET_UNALIGN_CTL(me, arg2);
2313 error = SET_FPEMU_CTL(me, arg2);
2316 error = GET_FPEMU_CTL(me, arg2);
2319 error = SET_FPEXC_CTL(me, arg2);
2322 error = GET_FPEXC_CTL(me, arg2);
2325 error = PR_TIMING_STATISTICAL;
2328 if (arg2 != PR_TIMING_STATISTICAL)
2332 comm[sizeof(me->comm) - 1] = 0;
2333 if (strncpy_from_user(comm, (char __user *)arg2,
2334 sizeof(me->comm) - 1) < 0)
2336 set_task_comm(me, comm);
2337 proc_comm_connector(me);
2340 get_task_comm(comm, me);
2341 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2345 error = GET_ENDIAN(me, arg2);
2348 error = SET_ENDIAN(me, arg2);
2350 case PR_GET_SECCOMP:
2351 error = prctl_get_seccomp();
2353 case PR_SET_SECCOMP:
2354 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2357 error = GET_TSC_CTL(arg2);
2360 error = SET_TSC_CTL(arg2);
2362 case PR_TASK_PERF_EVENTS_DISABLE:
2363 error = perf_event_task_disable();
2365 case PR_TASK_PERF_EVENTS_ENABLE:
2366 error = perf_event_task_enable();
2368 case PR_GET_TIMERSLACK:
2369 if (current->timer_slack_ns > ULONG_MAX)
2372 error = current->timer_slack_ns;
2374 case PR_SET_TIMERSLACK:
2376 current->timer_slack_ns =
2377 current->default_timer_slack_ns;
2379 current->timer_slack_ns = arg2;
2385 case PR_MCE_KILL_CLEAR:
2388 current->flags &= ~PF_MCE_PROCESS;
2390 case PR_MCE_KILL_SET:
2391 current->flags |= PF_MCE_PROCESS;
2392 if (arg3 == PR_MCE_KILL_EARLY)
2393 current->flags |= PF_MCE_EARLY;
2394 else if (arg3 == PR_MCE_KILL_LATE)
2395 current->flags &= ~PF_MCE_EARLY;
2396 else if (arg3 == PR_MCE_KILL_DEFAULT)
2398 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2406 case PR_MCE_KILL_GET:
2407 if (arg2 | arg3 | arg4 | arg5)
2409 if (current->flags & PF_MCE_PROCESS)
2410 error = (current->flags & PF_MCE_EARLY) ?
2411 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2413 error = PR_MCE_KILL_DEFAULT;
2416 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2418 case PR_GET_TID_ADDRESS:
2419 error = prctl_get_tid_address(me, (int __user **)arg2);
2421 case PR_SET_CHILD_SUBREAPER:
2422 me->signal->is_child_subreaper = !!arg2;
2426 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2428 case PR_GET_CHILD_SUBREAPER:
2429 error = put_user(me->signal->is_child_subreaper,
2430 (int __user *)arg2);
2432 case PR_SET_NO_NEW_PRIVS:
2433 if (arg2 != 1 || arg3 || arg4 || arg5)
2436 task_set_no_new_privs(current);
2438 case PR_GET_NO_NEW_PRIVS:
2439 if (arg2 || arg3 || arg4 || arg5)
2441 return task_no_new_privs(current) ? 1 : 0;
2442 case PR_GET_THP_DISABLE:
2443 if (arg2 || arg3 || arg4 || arg5)
2445 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2447 case PR_SET_THP_DISABLE:
2448 if (arg3 || arg4 || arg5)
2450 if (down_write_killable(&me->mm->mmap_sem))
2453 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2455 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2456 up_write(&me->mm->mmap_sem);
2458 case PR_MPX_ENABLE_MANAGEMENT:
2459 if (arg2 || arg3 || arg4 || arg5)
2461 error = MPX_ENABLE_MANAGEMENT();
2463 case PR_MPX_DISABLE_MANAGEMENT:
2464 if (arg2 || arg3 || arg4 || arg5)
2466 error = MPX_DISABLE_MANAGEMENT();
2468 case PR_SET_FP_MODE:
2469 error = SET_FP_MODE(me, arg2);
2471 case PR_GET_FP_MODE:
2472 error = GET_FP_MODE(me);
2475 error = SVE_SET_VL(arg2);
2478 error = SVE_GET_VL();
2480 case PR_GET_SPECULATION_CTRL:
2481 if (arg3 || arg4 || arg5)
2483 error = arch_prctl_spec_ctrl_get(me, arg2);
2485 case PR_SET_SPECULATION_CTRL:
2488 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2490 case PR_PAC_RESET_KEYS:
2491 if (arg3 || arg4 || arg5)
2493 error = PAC_RESET_KEYS(me, arg2);
2502 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2503 struct getcpu_cache __user *, unused)
2506 int cpu = raw_smp_processor_id();
2509 err |= put_user(cpu, cpup);
2511 err |= put_user(cpu_to_node(cpu), nodep);
2512 return err ? -EFAULT : 0;
2516 * do_sysinfo - fill in sysinfo struct
2517 * @info: pointer to buffer to fill
2519 static int do_sysinfo(struct sysinfo *info)
2521 unsigned long mem_total, sav_total;
2522 unsigned int mem_unit, bitcount;
2523 struct timespec64 tp;
2525 memset(info, 0, sizeof(struct sysinfo));
2527 ktime_get_boottime_ts64(&tp);
2528 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2530 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2532 info->procs = nr_threads;
2538 * If the sum of all the available memory (i.e. ram + swap)
2539 * is less than can be stored in a 32 bit unsigned long then
2540 * we can be binary compatible with 2.2.x kernels. If not,
2541 * well, in that case 2.2.x was broken anyways...
2543 * -Erik Andersen <andersee@debian.org>
2546 mem_total = info->totalram + info->totalswap;
2547 if (mem_total < info->totalram || mem_total < info->totalswap)
2550 mem_unit = info->mem_unit;
2551 while (mem_unit > 1) {
2554 sav_total = mem_total;
2556 if (mem_total < sav_total)
2561 * If mem_total did not overflow, multiply all memory values by
2562 * info->mem_unit and set it to 1. This leaves things compatible
2563 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2568 info->totalram <<= bitcount;
2569 info->freeram <<= bitcount;
2570 info->sharedram <<= bitcount;
2571 info->bufferram <<= bitcount;
2572 info->totalswap <<= bitcount;
2573 info->freeswap <<= bitcount;
2574 info->totalhigh <<= bitcount;
2575 info->freehigh <<= bitcount;
2581 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2587 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2593 #ifdef CONFIG_COMPAT
2594 struct compat_sysinfo {
2608 char _f[20-2*sizeof(u32)-sizeof(int)];
2611 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2617 /* Check to see if any memory value is too large for 32-bit and scale
2620 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2623 while (s.mem_unit < PAGE_SIZE) {
2628 s.totalram >>= bitcount;
2629 s.freeram >>= bitcount;
2630 s.sharedram >>= bitcount;
2631 s.bufferram >>= bitcount;
2632 s.totalswap >>= bitcount;
2633 s.freeswap >>= bitcount;
2634 s.totalhigh >>= bitcount;
2635 s.freehigh >>= bitcount;
2638 if (!access_ok(info, sizeof(struct compat_sysinfo)) ||
2639 __put_user(s.uptime, &info->uptime) ||
2640 __put_user(s.loads[0], &info->loads[0]) ||
2641 __put_user(s.loads[1], &info->loads[1]) ||
2642 __put_user(s.loads[2], &info->loads[2]) ||
2643 __put_user(s.totalram, &info->totalram) ||
2644 __put_user(s.freeram, &info->freeram) ||
2645 __put_user(s.sharedram, &info->sharedram) ||
2646 __put_user(s.bufferram, &info->bufferram) ||
2647 __put_user(s.totalswap, &info->totalswap) ||
2648 __put_user(s.freeswap, &info->freeswap) ||
2649 __put_user(s.procs, &info->procs) ||
2650 __put_user(s.totalhigh, &info->totalhigh) ||
2651 __put_user(s.freehigh, &info->freehigh) ||
2652 __put_user(s.mem_unit, &info->mem_unit))
2657 #endif /* CONFIG_COMPAT */