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>
45 #include <linux/syscall_user_dispatch.h>
47 #include <linux/compat.h>
48 #include <linux/syscalls.h>
49 #include <linux/kprobes.h>
50 #include <linux/user_namespace.h>
51 #include <linux/time_namespace.h>
52 #include <linux/binfmts.h>
54 #include <linux/sched.h>
55 #include <linux/sched/autogroup.h>
56 #include <linux/sched/loadavg.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/mm.h>
59 #include <linux/sched/coredump.h>
60 #include <linux/sched/task.h>
61 #include <linux/sched/cputime.h>
62 #include <linux/rcupdate.h>
63 #include <linux/uidgid.h>
64 #include <linux/cred.h>
66 #include <linux/nospec.h>
68 #include <linux/kmsg_dump.h>
69 /* Move somewhere else to avoid recompiling? */
70 #include <generated/utsrelease.h>
72 #include <linux/uaccess.h>
74 #include <asm/unistd.h>
78 #ifndef SET_UNALIGN_CTL
79 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
81 #ifndef GET_UNALIGN_CTL
82 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
85 # define SET_FPEMU_CTL(a, b) (-EINVAL)
88 # define GET_FPEMU_CTL(a, b) (-EINVAL)
91 # define SET_FPEXC_CTL(a, b) (-EINVAL)
94 # define GET_FPEXC_CTL(a, b) (-EINVAL)
97 # define GET_ENDIAN(a, b) (-EINVAL)
100 # define SET_ENDIAN(a, b) (-EINVAL)
103 # define GET_TSC_CTL(a) (-EINVAL)
106 # define SET_TSC_CTL(a) (-EINVAL)
109 # define GET_FP_MODE(a) (-EINVAL)
112 # define SET_FP_MODE(a,b) (-EINVAL)
115 # define SVE_SET_VL(a) (-EINVAL)
118 # define SVE_GET_VL() (-EINVAL)
120 #ifndef PAC_RESET_KEYS
121 # define PAC_RESET_KEYS(a, b) (-EINVAL)
123 #ifndef SET_TAGGED_ADDR_CTRL
124 # define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
126 #ifndef GET_TAGGED_ADDR_CTRL
127 # define GET_TAGGED_ADDR_CTRL() (-EINVAL)
131 * this is where the system-wide overflow UID and GID are defined, for
132 * architectures that now have 32-bit UID/GID but didn't in the past
135 int overflowuid = DEFAULT_OVERFLOWUID;
136 int overflowgid = DEFAULT_OVERFLOWGID;
138 EXPORT_SYMBOL(overflowuid);
139 EXPORT_SYMBOL(overflowgid);
142 * the same as above, but for filesystems which can only store a 16-bit
143 * UID and GID. as such, this is needed on all architectures
146 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
147 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
149 EXPORT_SYMBOL(fs_overflowuid);
150 EXPORT_SYMBOL(fs_overflowgid);
153 * Returns true if current's euid is same as p's uid or euid,
154 * or has CAP_SYS_NICE to p's user_ns.
156 * Called with rcu_read_lock, creds are safe
158 static bool set_one_prio_perm(struct task_struct *p)
160 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
162 if (uid_eq(pcred->uid, cred->euid) ||
163 uid_eq(pcred->euid, cred->euid))
165 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
171 * set the priority of a task
172 * - the caller must hold the RCU read lock
174 static int set_one_prio(struct task_struct *p, int niceval, int error)
178 if (!set_one_prio_perm(p)) {
182 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
186 no_nice = security_task_setnice(p, niceval);
193 set_user_nice(p, niceval);
198 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
200 struct task_struct *g, *p;
201 struct user_struct *user;
202 const struct cred *cred = current_cred();
207 if (which > PRIO_USER || which < PRIO_PROCESS)
210 /* normalize: avoid signed division (rounding problems) */
212 if (niceval < MIN_NICE)
214 if (niceval > MAX_NICE)
218 read_lock(&tasklist_lock);
222 p = find_task_by_vpid(who);
226 error = set_one_prio(p, niceval, error);
230 pgrp = find_vpid(who);
232 pgrp = task_pgrp(current);
233 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
234 error = set_one_prio(p, niceval, error);
235 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
238 uid = make_kuid(cred->user_ns, who);
242 else if (!uid_eq(uid, cred->uid)) {
243 user = find_user(uid);
245 goto out_unlock; /* No processes for this user */
247 do_each_thread(g, p) {
248 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
249 error = set_one_prio(p, niceval, error);
250 } while_each_thread(g, p);
251 if (!uid_eq(uid, cred->uid))
252 free_uid(user); /* For find_user() */
256 read_unlock(&tasklist_lock);
263 * Ugh. To avoid negative return values, "getpriority()" will
264 * not return the normal nice-value, but a negated value that
265 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
266 * to stay compatible.
268 SYSCALL_DEFINE2(getpriority, int, which, int, who)
270 struct task_struct *g, *p;
271 struct user_struct *user;
272 const struct cred *cred = current_cred();
273 long niceval, retval = -ESRCH;
277 if (which > PRIO_USER || which < PRIO_PROCESS)
281 read_lock(&tasklist_lock);
285 p = find_task_by_vpid(who);
289 niceval = nice_to_rlimit(task_nice(p));
290 if (niceval > retval)
296 pgrp = find_vpid(who);
298 pgrp = task_pgrp(current);
299 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
300 niceval = nice_to_rlimit(task_nice(p));
301 if (niceval > retval)
303 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
306 uid = make_kuid(cred->user_ns, who);
310 else if (!uid_eq(uid, cred->uid)) {
311 user = find_user(uid);
313 goto out_unlock; /* No processes for this user */
315 do_each_thread(g, p) {
316 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
317 niceval = nice_to_rlimit(task_nice(p));
318 if (niceval > retval)
321 } while_each_thread(g, p);
322 if (!uid_eq(uid, cred->uid))
323 free_uid(user); /* for find_user() */
327 read_unlock(&tasklist_lock);
334 * Unprivileged users may change the real gid to the effective gid
335 * or vice versa. (BSD-style)
337 * If you set the real gid at all, or set the effective gid to a value not
338 * equal to the real gid, then the saved gid is set to the new effective gid.
340 * This makes it possible for a setgid program to completely drop its
341 * privileges, which is often a useful assertion to make when you are doing
342 * a security audit over a program.
344 * The general idea is that a program which uses just setregid() will be
345 * 100% compatible with BSD. A program which uses just setgid() will be
346 * 100% compatible with POSIX with saved IDs.
348 * SMP: There are not races, the GIDs are checked only by filesystem
349 * operations (as far as semantic preservation is concerned).
351 #ifdef CONFIG_MULTIUSER
352 long __sys_setregid(gid_t rgid, gid_t egid)
354 struct user_namespace *ns = current_user_ns();
355 const struct cred *old;
360 krgid = make_kgid(ns, rgid);
361 kegid = make_kgid(ns, egid);
363 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
365 if ((egid != (gid_t) -1) && !gid_valid(kegid))
368 new = prepare_creds();
371 old = current_cred();
374 if (rgid != (gid_t) -1) {
375 if (gid_eq(old->gid, krgid) ||
376 gid_eq(old->egid, krgid) ||
377 ns_capable_setid(old->user_ns, CAP_SETGID))
382 if (egid != (gid_t) -1) {
383 if (gid_eq(old->gid, kegid) ||
384 gid_eq(old->egid, kegid) ||
385 gid_eq(old->sgid, kegid) ||
386 ns_capable_setid(old->user_ns, CAP_SETGID))
392 if (rgid != (gid_t) -1 ||
393 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
394 new->sgid = new->egid;
395 new->fsgid = new->egid;
397 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
401 return commit_creds(new);
408 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
410 return __sys_setregid(rgid, egid);
414 * setgid() is implemented like SysV w/ SAVED_IDS
416 * SMP: Same implicit races as above.
418 long __sys_setgid(gid_t gid)
420 struct user_namespace *ns = current_user_ns();
421 const struct cred *old;
426 kgid = make_kgid(ns, gid);
427 if (!gid_valid(kgid))
430 new = prepare_creds();
433 old = current_cred();
436 if (ns_capable_setid(old->user_ns, CAP_SETGID))
437 new->gid = new->egid = new->sgid = new->fsgid = kgid;
438 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
439 new->egid = new->fsgid = kgid;
443 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
447 return commit_creds(new);
454 SYSCALL_DEFINE1(setgid, gid_t, gid)
456 return __sys_setgid(gid);
460 * change the user struct in a credentials set to match the new UID
462 static int set_user(struct cred *new)
464 struct user_struct *new_user;
466 new_user = alloc_uid(new->uid);
471 * We don't fail in case of NPROC limit excess here because too many
472 * poorly written programs don't check set*uid() return code, assuming
473 * it never fails if called by root. We may still enforce NPROC limit
474 * for programs doing set*uid()+execve() by harmlessly deferring the
475 * failure to the execve() stage.
477 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
478 new_user != INIT_USER)
479 current->flags |= PF_NPROC_EXCEEDED;
481 current->flags &= ~PF_NPROC_EXCEEDED;
484 new->user = new_user;
489 * Unprivileged users may change the real uid to the effective uid
490 * or vice versa. (BSD-style)
492 * If you set the real uid at all, or set the effective uid to a value not
493 * equal to the real uid, then the saved uid is set to the new effective uid.
495 * This makes it possible for a setuid program to completely drop its
496 * privileges, which is often a useful assertion to make when you are doing
497 * a security audit over a program.
499 * The general idea is that a program which uses just setreuid() will be
500 * 100% compatible with BSD. A program which uses just setuid() will be
501 * 100% compatible with POSIX with saved IDs.
503 long __sys_setreuid(uid_t ruid, uid_t euid)
505 struct user_namespace *ns = current_user_ns();
506 const struct cred *old;
511 kruid = make_kuid(ns, ruid);
512 keuid = make_kuid(ns, euid);
514 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
516 if ((euid != (uid_t) -1) && !uid_valid(keuid))
519 new = prepare_creds();
522 old = current_cred();
525 if (ruid != (uid_t) -1) {
527 if (!uid_eq(old->uid, kruid) &&
528 !uid_eq(old->euid, kruid) &&
529 !ns_capable_setid(old->user_ns, CAP_SETUID))
533 if (euid != (uid_t) -1) {
535 if (!uid_eq(old->uid, keuid) &&
536 !uid_eq(old->euid, keuid) &&
537 !uid_eq(old->suid, keuid) &&
538 !ns_capable_setid(old->user_ns, CAP_SETUID))
542 if (!uid_eq(new->uid, old->uid)) {
543 retval = set_user(new);
547 if (ruid != (uid_t) -1 ||
548 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
549 new->suid = new->euid;
550 new->fsuid = new->euid;
552 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
556 return commit_creds(new);
563 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
565 return __sys_setreuid(ruid, euid);
569 * setuid() is implemented like SysV with SAVED_IDS
571 * Note that SAVED_ID's is deficient in that a setuid root program
572 * like sendmail, for example, cannot set its uid to be a normal
573 * user and then switch back, because if you're root, setuid() sets
574 * the saved uid too. If you don't like this, blame the bright people
575 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
576 * will allow a root program to temporarily drop privileges and be able to
577 * regain them by swapping the real and effective uid.
579 long __sys_setuid(uid_t uid)
581 struct user_namespace *ns = current_user_ns();
582 const struct cred *old;
587 kuid = make_kuid(ns, uid);
588 if (!uid_valid(kuid))
591 new = prepare_creds();
594 old = current_cred();
597 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
598 new->suid = new->uid = kuid;
599 if (!uid_eq(kuid, old->uid)) {
600 retval = set_user(new);
604 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
608 new->fsuid = new->euid = kuid;
610 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
614 return commit_creds(new);
621 SYSCALL_DEFINE1(setuid, uid_t, uid)
623 return __sys_setuid(uid);
628 * This function implements a generic ability to update ruid, euid,
629 * and suid. This allows you to implement the 4.4 compatible seteuid().
631 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
633 struct user_namespace *ns = current_user_ns();
634 const struct cred *old;
637 kuid_t kruid, keuid, ksuid;
639 kruid = make_kuid(ns, ruid);
640 keuid = make_kuid(ns, euid);
641 ksuid = make_kuid(ns, suid);
643 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
646 if ((euid != (uid_t) -1) && !uid_valid(keuid))
649 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
652 new = prepare_creds();
656 old = current_cred();
659 if (!ns_capable_setid(old->user_ns, CAP_SETUID)) {
660 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
661 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
663 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
664 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
666 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
667 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
671 if (ruid != (uid_t) -1) {
673 if (!uid_eq(kruid, old->uid)) {
674 retval = set_user(new);
679 if (euid != (uid_t) -1)
681 if (suid != (uid_t) -1)
683 new->fsuid = new->euid;
685 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
689 return commit_creds(new);
696 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
698 return __sys_setresuid(ruid, euid, suid);
701 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
703 const struct cred *cred = current_cred();
705 uid_t ruid, euid, suid;
707 ruid = from_kuid_munged(cred->user_ns, cred->uid);
708 euid = from_kuid_munged(cred->user_ns, cred->euid);
709 suid = from_kuid_munged(cred->user_ns, cred->suid);
711 retval = put_user(ruid, ruidp);
713 retval = put_user(euid, euidp);
715 return put_user(suid, suidp);
721 * Same as above, but for rgid, egid, sgid.
723 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
725 struct user_namespace *ns = current_user_ns();
726 const struct cred *old;
729 kgid_t krgid, kegid, ksgid;
731 krgid = make_kgid(ns, rgid);
732 kegid = make_kgid(ns, egid);
733 ksgid = make_kgid(ns, sgid);
735 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
737 if ((egid != (gid_t) -1) && !gid_valid(kegid))
739 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
742 new = prepare_creds();
745 old = current_cred();
748 if (!ns_capable_setid(old->user_ns, CAP_SETGID)) {
749 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
750 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
752 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
753 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
755 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
756 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
760 if (rgid != (gid_t) -1)
762 if (egid != (gid_t) -1)
764 if (sgid != (gid_t) -1)
766 new->fsgid = new->egid;
768 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
772 return commit_creds(new);
779 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
781 return __sys_setresgid(rgid, egid, sgid);
784 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
786 const struct cred *cred = current_cred();
788 gid_t rgid, egid, sgid;
790 rgid = from_kgid_munged(cred->user_ns, cred->gid);
791 egid = from_kgid_munged(cred->user_ns, cred->egid);
792 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
794 retval = put_user(rgid, rgidp);
796 retval = put_user(egid, egidp);
798 retval = put_user(sgid, sgidp);
806 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
807 * is used for "access()" and for the NFS daemon (letting nfsd stay at
808 * whatever uid it wants to). It normally shadows "euid", except when
809 * explicitly set by setfsuid() or for access..
811 long __sys_setfsuid(uid_t uid)
813 const struct cred *old;
818 old = current_cred();
819 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
821 kuid = make_kuid(old->user_ns, uid);
822 if (!uid_valid(kuid))
825 new = prepare_creds();
829 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
830 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
831 ns_capable_setid(old->user_ns, CAP_SETUID)) {
832 if (!uid_eq(kuid, old->fsuid)) {
834 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
847 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
849 return __sys_setfsuid(uid);
853 * Samma på svenska..
855 long __sys_setfsgid(gid_t gid)
857 const struct cred *old;
862 old = current_cred();
863 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
865 kgid = make_kgid(old->user_ns, gid);
866 if (!gid_valid(kgid))
869 new = prepare_creds();
873 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
874 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
875 ns_capable_setid(old->user_ns, CAP_SETGID)) {
876 if (!gid_eq(kgid, old->fsgid)) {
878 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
891 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
893 return __sys_setfsgid(gid);
895 #endif /* CONFIG_MULTIUSER */
898 * sys_getpid - return the thread group id of the current process
900 * Note, despite the name, this returns the tgid not the pid. The tgid and
901 * the pid are identical unless CLONE_THREAD was specified on clone() in
902 * which case the tgid is the same in all threads of the same group.
904 * This is SMP safe as current->tgid does not change.
906 SYSCALL_DEFINE0(getpid)
908 return task_tgid_vnr(current);
911 /* Thread ID - the internal kernel "pid" */
912 SYSCALL_DEFINE0(gettid)
914 return task_pid_vnr(current);
918 * Accessing ->real_parent is not SMP-safe, it could
919 * change from under us. However, we can use a stale
920 * value of ->real_parent under rcu_read_lock(), see
921 * release_task()->call_rcu(delayed_put_task_struct).
923 SYSCALL_DEFINE0(getppid)
928 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
934 SYSCALL_DEFINE0(getuid)
936 /* Only we change this so SMP safe */
937 return from_kuid_munged(current_user_ns(), current_uid());
940 SYSCALL_DEFINE0(geteuid)
942 /* Only we change this so SMP safe */
943 return from_kuid_munged(current_user_ns(), current_euid());
946 SYSCALL_DEFINE0(getgid)
948 /* Only we change this so SMP safe */
949 return from_kgid_munged(current_user_ns(), current_gid());
952 SYSCALL_DEFINE0(getegid)
954 /* Only we change this so SMP safe */
955 return from_kgid_munged(current_user_ns(), current_egid());
958 static void do_sys_times(struct tms *tms)
960 u64 tgutime, tgstime, cutime, cstime;
962 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
963 cutime = current->signal->cutime;
964 cstime = current->signal->cstime;
965 tms->tms_utime = nsec_to_clock_t(tgutime);
966 tms->tms_stime = nsec_to_clock_t(tgstime);
967 tms->tms_cutime = nsec_to_clock_t(cutime);
968 tms->tms_cstime = nsec_to_clock_t(cstime);
971 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
977 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
980 force_successful_syscall_return();
981 return (long) jiffies_64_to_clock_t(get_jiffies_64());
985 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
987 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
990 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
994 struct compat_tms tmp;
997 /* Convert our struct tms to the compat version. */
998 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
999 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1000 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1001 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1002 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1005 force_successful_syscall_return();
1006 return compat_jiffies_to_clock_t(jiffies);
1011 * This needs some heavy checking ...
1012 * I just haven't the stomach for it. I also don't fully
1013 * understand sessions/pgrp etc. Let somebody who does explain it.
1015 * OK, I think I have the protection semantics right.... this is really
1016 * only important on a multi-user system anyway, to make sure one user
1017 * can't send a signal to a process owned by another. -TYT, 12/12/91
1019 * !PF_FORKNOEXEC check to conform completely to POSIX.
1021 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1023 struct task_struct *p;
1024 struct task_struct *group_leader = current->group_leader;
1029 pid = task_pid_vnr(group_leader);
1036 /* From this point forward we keep holding onto the tasklist lock
1037 * so that our parent does not change from under us. -DaveM
1039 write_lock_irq(&tasklist_lock);
1042 p = find_task_by_vpid(pid);
1047 if (!thread_group_leader(p))
1050 if (same_thread_group(p->real_parent, group_leader)) {
1052 if (task_session(p) != task_session(group_leader))
1055 if (!(p->flags & PF_FORKNOEXEC))
1059 if (p != group_leader)
1064 if (p->signal->leader)
1069 struct task_struct *g;
1071 pgrp = find_vpid(pgid);
1072 g = pid_task(pgrp, PIDTYPE_PGID);
1073 if (!g || task_session(g) != task_session(group_leader))
1077 err = security_task_setpgid(p, pgid);
1081 if (task_pgrp(p) != pgrp)
1082 change_pid(p, PIDTYPE_PGID, pgrp);
1086 /* All paths lead to here, thus we are safe. -DaveM */
1087 write_unlock_irq(&tasklist_lock);
1092 static int do_getpgid(pid_t pid)
1094 struct task_struct *p;
1100 grp = task_pgrp(current);
1103 p = find_task_by_vpid(pid);
1110 retval = security_task_getpgid(p);
1114 retval = pid_vnr(grp);
1120 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1122 return do_getpgid(pid);
1125 #ifdef __ARCH_WANT_SYS_GETPGRP
1127 SYSCALL_DEFINE0(getpgrp)
1129 return do_getpgid(0);
1134 SYSCALL_DEFINE1(getsid, pid_t, pid)
1136 struct task_struct *p;
1142 sid = task_session(current);
1145 p = find_task_by_vpid(pid);
1148 sid = task_session(p);
1152 retval = security_task_getsid(p);
1156 retval = pid_vnr(sid);
1162 static void set_special_pids(struct pid *pid)
1164 struct task_struct *curr = current->group_leader;
1166 if (task_session(curr) != pid)
1167 change_pid(curr, PIDTYPE_SID, pid);
1169 if (task_pgrp(curr) != pid)
1170 change_pid(curr, PIDTYPE_PGID, pid);
1173 int ksys_setsid(void)
1175 struct task_struct *group_leader = current->group_leader;
1176 struct pid *sid = task_pid(group_leader);
1177 pid_t session = pid_vnr(sid);
1180 write_lock_irq(&tasklist_lock);
1181 /* Fail if I am already a session leader */
1182 if (group_leader->signal->leader)
1185 /* Fail if a process group id already exists that equals the
1186 * proposed session id.
1188 if (pid_task(sid, PIDTYPE_PGID))
1191 group_leader->signal->leader = 1;
1192 set_special_pids(sid);
1194 proc_clear_tty(group_leader);
1198 write_unlock_irq(&tasklist_lock);
1200 proc_sid_connector(group_leader);
1201 sched_autogroup_create_attach(group_leader);
1206 SYSCALL_DEFINE0(setsid)
1208 return ksys_setsid();
1211 DECLARE_RWSEM(uts_sem);
1213 #ifdef COMPAT_UTS_MACHINE
1214 #define override_architecture(name) \
1215 (personality(current->personality) == PER_LINUX32 && \
1216 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1217 sizeof(COMPAT_UTS_MACHINE)))
1219 #define override_architecture(name) 0
1223 * Work around broken programs that cannot handle "Linux 3.0".
1224 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1225 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1228 static int override_release(char __user *release, size_t len)
1232 if (current->personality & UNAME26) {
1233 const char *rest = UTS_RELEASE;
1234 char buf[65] = { 0 };
1240 if (*rest == '.' && ++ndots >= 3)
1242 if (!isdigit(*rest) && *rest != '.')
1246 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1247 copy = clamp_t(size_t, len, 1, sizeof(buf));
1248 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1249 ret = copy_to_user(release, buf, copy + 1);
1254 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1256 struct new_utsname tmp;
1258 down_read(&uts_sem);
1259 memcpy(&tmp, utsname(), sizeof(tmp));
1261 if (copy_to_user(name, &tmp, sizeof(tmp)))
1264 if (override_release(name->release, sizeof(name->release)))
1266 if (override_architecture(name))
1271 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1275 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1277 struct old_utsname tmp;
1282 down_read(&uts_sem);
1283 memcpy(&tmp, utsname(), sizeof(tmp));
1285 if (copy_to_user(name, &tmp, sizeof(tmp)))
1288 if (override_release(name->release, sizeof(name->release)))
1290 if (override_architecture(name))
1295 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1297 struct oldold_utsname tmp;
1302 memset(&tmp, 0, sizeof(tmp));
1304 down_read(&uts_sem);
1305 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1306 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1307 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1308 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1309 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1311 if (copy_to_user(name, &tmp, sizeof(tmp)))
1314 if (override_architecture(name))
1316 if (override_release(name->release, sizeof(name->release)))
1322 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1325 char tmp[__NEW_UTS_LEN];
1327 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1330 if (len < 0 || len > __NEW_UTS_LEN)
1333 if (!copy_from_user(tmp, name, len)) {
1334 struct new_utsname *u;
1336 down_write(&uts_sem);
1338 memcpy(u->nodename, tmp, len);
1339 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1341 uts_proc_notify(UTS_PROC_HOSTNAME);
1347 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1349 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1352 struct new_utsname *u;
1353 char tmp[__NEW_UTS_LEN + 1];
1357 down_read(&uts_sem);
1359 i = 1 + strlen(u->nodename);
1362 memcpy(tmp, u->nodename, i);
1364 if (copy_to_user(name, tmp, i))
1372 * Only setdomainname; getdomainname can be implemented by calling
1375 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1378 char tmp[__NEW_UTS_LEN];
1380 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1382 if (len < 0 || len > __NEW_UTS_LEN)
1386 if (!copy_from_user(tmp, name, len)) {
1387 struct new_utsname *u;
1389 down_write(&uts_sem);
1391 memcpy(u->domainname, tmp, len);
1392 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1394 uts_proc_notify(UTS_PROC_DOMAINNAME);
1400 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1402 struct rlimit value;
1405 ret = do_prlimit(current, resource, NULL, &value);
1407 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1412 #ifdef CONFIG_COMPAT
1414 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1415 struct compat_rlimit __user *, rlim)
1418 struct compat_rlimit r32;
1420 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1423 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1424 r.rlim_cur = RLIM_INFINITY;
1426 r.rlim_cur = r32.rlim_cur;
1427 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1428 r.rlim_max = RLIM_INFINITY;
1430 r.rlim_max = r32.rlim_max;
1431 return do_prlimit(current, resource, &r, NULL);
1434 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1435 struct compat_rlimit __user *, rlim)
1440 ret = do_prlimit(current, resource, NULL, &r);
1442 struct compat_rlimit r32;
1443 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1444 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1446 r32.rlim_cur = r.rlim_cur;
1447 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1448 r32.rlim_max = COMPAT_RLIM_INFINITY;
1450 r32.rlim_max = r.rlim_max;
1452 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1460 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1463 * Back compatibility for getrlimit. Needed for some apps.
1465 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1466 struct rlimit __user *, rlim)
1469 if (resource >= RLIM_NLIMITS)
1472 resource = array_index_nospec(resource, RLIM_NLIMITS);
1473 task_lock(current->group_leader);
1474 x = current->signal->rlim[resource];
1475 task_unlock(current->group_leader);
1476 if (x.rlim_cur > 0x7FFFFFFF)
1477 x.rlim_cur = 0x7FFFFFFF;
1478 if (x.rlim_max > 0x7FFFFFFF)
1479 x.rlim_max = 0x7FFFFFFF;
1480 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1483 #ifdef CONFIG_COMPAT
1484 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1485 struct compat_rlimit __user *, rlim)
1489 if (resource >= RLIM_NLIMITS)
1492 resource = array_index_nospec(resource, RLIM_NLIMITS);
1493 task_lock(current->group_leader);
1494 r = current->signal->rlim[resource];
1495 task_unlock(current->group_leader);
1496 if (r.rlim_cur > 0x7FFFFFFF)
1497 r.rlim_cur = 0x7FFFFFFF;
1498 if (r.rlim_max > 0x7FFFFFFF)
1499 r.rlim_max = 0x7FFFFFFF;
1501 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1502 put_user(r.rlim_max, &rlim->rlim_max))
1510 static inline bool rlim64_is_infinity(__u64 rlim64)
1512 #if BITS_PER_LONG < 64
1513 return rlim64 >= ULONG_MAX;
1515 return rlim64 == RLIM64_INFINITY;
1519 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1521 if (rlim->rlim_cur == RLIM_INFINITY)
1522 rlim64->rlim_cur = RLIM64_INFINITY;
1524 rlim64->rlim_cur = rlim->rlim_cur;
1525 if (rlim->rlim_max == RLIM_INFINITY)
1526 rlim64->rlim_max = RLIM64_INFINITY;
1528 rlim64->rlim_max = rlim->rlim_max;
1531 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1533 if (rlim64_is_infinity(rlim64->rlim_cur))
1534 rlim->rlim_cur = RLIM_INFINITY;
1536 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1537 if (rlim64_is_infinity(rlim64->rlim_max))
1538 rlim->rlim_max = RLIM_INFINITY;
1540 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1543 /* make sure you are allowed to change @tsk limits before calling this */
1544 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1545 struct rlimit *new_rlim, struct rlimit *old_rlim)
1547 struct rlimit *rlim;
1550 if (resource >= RLIM_NLIMITS)
1553 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1555 if (resource == RLIMIT_NOFILE &&
1556 new_rlim->rlim_max > sysctl_nr_open)
1560 /* protect tsk->signal and tsk->sighand from disappearing */
1561 read_lock(&tasklist_lock);
1562 if (!tsk->sighand) {
1567 rlim = tsk->signal->rlim + resource;
1568 task_lock(tsk->group_leader);
1570 /* Keep the capable check against init_user_ns until
1571 cgroups can contain all limits */
1572 if (new_rlim->rlim_max > rlim->rlim_max &&
1573 !capable(CAP_SYS_RESOURCE))
1576 retval = security_task_setrlimit(tsk, resource, new_rlim);
1584 task_unlock(tsk->group_leader);
1587 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1588 * infite. In case of RLIM_INFINITY the posix CPU timer code
1589 * ignores the rlimit.
1591 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1592 new_rlim->rlim_cur != RLIM_INFINITY &&
1593 IS_ENABLED(CONFIG_POSIX_TIMERS))
1594 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1596 read_unlock(&tasklist_lock);
1600 /* rcu lock must be held */
1601 static int check_prlimit_permission(struct task_struct *task,
1604 const struct cred *cred = current_cred(), *tcred;
1607 if (current == task)
1610 tcred = __task_cred(task);
1611 id_match = (uid_eq(cred->uid, tcred->euid) &&
1612 uid_eq(cred->uid, tcred->suid) &&
1613 uid_eq(cred->uid, tcred->uid) &&
1614 gid_eq(cred->gid, tcred->egid) &&
1615 gid_eq(cred->gid, tcred->sgid) &&
1616 gid_eq(cred->gid, tcred->gid));
1617 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1620 return security_task_prlimit(cred, tcred, flags);
1623 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1624 const struct rlimit64 __user *, new_rlim,
1625 struct rlimit64 __user *, old_rlim)
1627 struct rlimit64 old64, new64;
1628 struct rlimit old, new;
1629 struct task_struct *tsk;
1630 unsigned int checkflags = 0;
1634 checkflags |= LSM_PRLIMIT_READ;
1637 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1639 rlim64_to_rlim(&new64, &new);
1640 checkflags |= LSM_PRLIMIT_WRITE;
1644 tsk = pid ? find_task_by_vpid(pid) : current;
1649 ret = check_prlimit_permission(tsk, checkflags);
1654 get_task_struct(tsk);
1657 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1658 old_rlim ? &old : NULL);
1660 if (!ret && old_rlim) {
1661 rlim_to_rlim64(&old, &old64);
1662 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1666 put_task_struct(tsk);
1670 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1672 struct rlimit new_rlim;
1674 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1676 return do_prlimit(current, resource, &new_rlim, NULL);
1680 * It would make sense to put struct rusage in the task_struct,
1681 * except that would make the task_struct be *really big*. After
1682 * task_struct gets moved into malloc'ed memory, it would
1683 * make sense to do this. It will make moving the rest of the information
1684 * a lot simpler! (Which we're not doing right now because we're not
1685 * measuring them yet).
1687 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1688 * races with threads incrementing their own counters. But since word
1689 * reads are atomic, we either get new values or old values and we don't
1690 * care which for the sums. We always take the siglock to protect reading
1691 * the c* fields from p->signal from races with exit.c updating those
1692 * fields when reaping, so a sample either gets all the additions of a
1693 * given child after it's reaped, or none so this sample is before reaping.
1696 * We need to take the siglock for CHILDEREN, SELF and BOTH
1697 * for the cases current multithreaded, non-current single threaded
1698 * non-current multithreaded. Thread traversal is now safe with
1700 * Strictly speaking, we donot need to take the siglock if we are current and
1701 * single threaded, as no one else can take our signal_struct away, no one
1702 * else can reap the children to update signal->c* counters, and no one else
1703 * can race with the signal-> fields. If we do not take any lock, the
1704 * signal-> fields could be read out of order while another thread was just
1705 * exiting. So we should place a read memory barrier when we avoid the lock.
1706 * On the writer side, write memory barrier is implied in __exit_signal
1707 * as __exit_signal releases the siglock spinlock after updating the signal->
1708 * fields. But we don't do this yet to keep things simple.
1712 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1714 r->ru_nvcsw += t->nvcsw;
1715 r->ru_nivcsw += t->nivcsw;
1716 r->ru_minflt += t->min_flt;
1717 r->ru_majflt += t->maj_flt;
1718 r->ru_inblock += task_io_get_inblock(t);
1719 r->ru_oublock += task_io_get_oublock(t);
1722 void getrusage(struct task_struct *p, int who, struct rusage *r)
1724 struct task_struct *t;
1725 unsigned long flags;
1726 u64 tgutime, tgstime, utime, stime;
1727 unsigned long maxrss = 0;
1729 memset((char *)r, 0, sizeof (*r));
1732 if (who == RUSAGE_THREAD) {
1733 task_cputime_adjusted(current, &utime, &stime);
1734 accumulate_thread_rusage(p, r);
1735 maxrss = p->signal->maxrss;
1739 if (!lock_task_sighand(p, &flags))
1744 case RUSAGE_CHILDREN:
1745 utime = p->signal->cutime;
1746 stime = p->signal->cstime;
1747 r->ru_nvcsw = p->signal->cnvcsw;
1748 r->ru_nivcsw = p->signal->cnivcsw;
1749 r->ru_minflt = p->signal->cmin_flt;
1750 r->ru_majflt = p->signal->cmaj_flt;
1751 r->ru_inblock = p->signal->cinblock;
1752 r->ru_oublock = p->signal->coublock;
1753 maxrss = p->signal->cmaxrss;
1755 if (who == RUSAGE_CHILDREN)
1760 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1763 r->ru_nvcsw += p->signal->nvcsw;
1764 r->ru_nivcsw += p->signal->nivcsw;
1765 r->ru_minflt += p->signal->min_flt;
1766 r->ru_majflt += p->signal->maj_flt;
1767 r->ru_inblock += p->signal->inblock;
1768 r->ru_oublock += p->signal->oublock;
1769 if (maxrss < p->signal->maxrss)
1770 maxrss = p->signal->maxrss;
1773 accumulate_thread_rusage(t, r);
1774 } while_each_thread(p, t);
1780 unlock_task_sighand(p, &flags);
1783 r->ru_utime = ns_to_kernel_old_timeval(utime);
1784 r->ru_stime = ns_to_kernel_old_timeval(stime);
1786 if (who != RUSAGE_CHILDREN) {
1787 struct mm_struct *mm = get_task_mm(p);
1790 setmax_mm_hiwater_rss(&maxrss, mm);
1794 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1797 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1801 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1802 who != RUSAGE_THREAD)
1805 getrusage(current, who, &r);
1806 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1809 #ifdef CONFIG_COMPAT
1810 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1814 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1815 who != RUSAGE_THREAD)
1818 getrusage(current, who, &r);
1819 return put_compat_rusage(&r, ru);
1823 SYSCALL_DEFINE1(umask, int, mask)
1825 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1829 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1832 struct file *old_exe, *exe_file;
1833 struct inode *inode;
1840 inode = file_inode(exe.file);
1843 * Because the original mm->exe_file points to executable file, make
1844 * sure that this one is executable as well, to avoid breaking an
1848 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1851 err = inode_permission(inode, MAY_EXEC);
1856 * Forbid mm->exe_file change if old file still mapped.
1858 exe_file = get_mm_exe_file(mm);
1861 struct vm_area_struct *vma;
1864 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1867 if (path_equal(&vma->vm_file->f_path,
1872 mmap_read_unlock(mm);
1877 /* set the new file, lockless */
1879 old_exe = xchg(&mm->exe_file, exe.file);
1886 mmap_read_unlock(mm);
1892 * Check arithmetic relations of passed addresses.
1894 * WARNING: we don't require any capability here so be very careful
1895 * in what is allowed for modification from userspace.
1897 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1899 unsigned long mmap_max_addr = TASK_SIZE;
1900 int error = -EINVAL, i;
1902 static const unsigned char offsets[] = {
1903 offsetof(struct prctl_mm_map, start_code),
1904 offsetof(struct prctl_mm_map, end_code),
1905 offsetof(struct prctl_mm_map, start_data),
1906 offsetof(struct prctl_mm_map, end_data),
1907 offsetof(struct prctl_mm_map, start_brk),
1908 offsetof(struct prctl_mm_map, brk),
1909 offsetof(struct prctl_mm_map, start_stack),
1910 offsetof(struct prctl_mm_map, arg_start),
1911 offsetof(struct prctl_mm_map, arg_end),
1912 offsetof(struct prctl_mm_map, env_start),
1913 offsetof(struct prctl_mm_map, env_end),
1917 * Make sure the members are not somewhere outside
1918 * of allowed address space.
1920 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1921 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1923 if ((unsigned long)val >= mmap_max_addr ||
1924 (unsigned long)val < mmap_min_addr)
1929 * Make sure the pairs are ordered.
1931 #define __prctl_check_order(__m1, __op, __m2) \
1932 ((unsigned long)prctl_map->__m1 __op \
1933 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1934 error = __prctl_check_order(start_code, <, end_code);
1935 error |= __prctl_check_order(start_data,<=, end_data);
1936 error |= __prctl_check_order(start_brk, <=, brk);
1937 error |= __prctl_check_order(arg_start, <=, arg_end);
1938 error |= __prctl_check_order(env_start, <=, env_end);
1941 #undef __prctl_check_order
1946 * @brk should be after @end_data in traditional maps.
1948 if (prctl_map->start_brk <= prctl_map->end_data ||
1949 prctl_map->brk <= prctl_map->end_data)
1953 * Neither we should allow to override limits if they set.
1955 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1956 prctl_map->start_brk, prctl_map->end_data,
1957 prctl_map->start_data))
1965 #ifdef CONFIG_CHECKPOINT_RESTORE
1966 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1968 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1969 unsigned long user_auxv[AT_VECTOR_SIZE];
1970 struct mm_struct *mm = current->mm;
1973 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1974 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1976 if (opt == PR_SET_MM_MAP_SIZE)
1977 return put_user((unsigned int)sizeof(prctl_map),
1978 (unsigned int __user *)addr);
1980 if (data_size != sizeof(prctl_map))
1983 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1986 error = validate_prctl_map_addr(&prctl_map);
1990 if (prctl_map.auxv_size) {
1992 * Someone is trying to cheat the auxv vector.
1994 if (!prctl_map.auxv ||
1995 prctl_map.auxv_size > sizeof(mm->saved_auxv))
1998 memset(user_auxv, 0, sizeof(user_auxv));
1999 if (copy_from_user(user_auxv,
2000 (const void __user *)prctl_map.auxv,
2001 prctl_map.auxv_size))
2004 /* Last entry must be AT_NULL as specification requires */
2005 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2006 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2009 if (prctl_map.exe_fd != (u32)-1) {
2011 * Check if the current user is checkpoint/restore capable.
2012 * At the time of this writing, it checks for CAP_SYS_ADMIN
2013 * or CAP_CHECKPOINT_RESTORE.
2014 * Note that a user with access to ptrace can masquerade an
2015 * arbitrary program as any executable, even setuid ones.
2016 * This may have implications in the tomoyo subsystem.
2018 if (!checkpoint_restore_ns_capable(current_user_ns()))
2021 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2027 * arg_lock protects concurent updates but we still need mmap_lock for
2028 * read to exclude races with sys_brk.
2033 * We don't validate if these members are pointing to
2034 * real present VMAs because application may have correspond
2035 * VMAs already unmapped and kernel uses these members for statistics
2036 * output in procfs mostly, except
2038 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2039 * for VMAs when updating these memvers so anything wrong written
2040 * here cause kernel to swear at userspace program but won't lead
2041 * to any problem in kernel itself
2044 spin_lock(&mm->arg_lock);
2045 mm->start_code = prctl_map.start_code;
2046 mm->end_code = prctl_map.end_code;
2047 mm->start_data = prctl_map.start_data;
2048 mm->end_data = prctl_map.end_data;
2049 mm->start_brk = prctl_map.start_brk;
2050 mm->brk = prctl_map.brk;
2051 mm->start_stack = prctl_map.start_stack;
2052 mm->arg_start = prctl_map.arg_start;
2053 mm->arg_end = prctl_map.arg_end;
2054 mm->env_start = prctl_map.env_start;
2055 mm->env_end = prctl_map.env_end;
2056 spin_unlock(&mm->arg_lock);
2059 * Note this update of @saved_auxv is lockless thus
2060 * if someone reads this member in procfs while we're
2061 * updating -- it may get partly updated results. It's
2062 * known and acceptable trade off: we leave it as is to
2063 * not introduce additional locks here making the kernel
2066 if (prctl_map.auxv_size)
2067 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2069 mmap_read_unlock(mm);
2072 #endif /* CONFIG_CHECKPOINT_RESTORE */
2074 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2078 * This doesn't move the auxiliary vector itself since it's pinned to
2079 * mm_struct, but it permits filling the vector with new values. It's
2080 * up to the caller to provide sane values here, otherwise userspace
2081 * tools which use this vector might be unhappy.
2083 unsigned long user_auxv[AT_VECTOR_SIZE];
2085 if (len > sizeof(user_auxv))
2088 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2091 /* Make sure the last entry is always AT_NULL */
2092 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2093 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2095 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2098 memcpy(mm->saved_auxv, user_auxv, len);
2099 task_unlock(current);
2104 static int prctl_set_mm(int opt, unsigned long addr,
2105 unsigned long arg4, unsigned long arg5)
2107 struct mm_struct *mm = current->mm;
2108 struct prctl_mm_map prctl_map = {
2113 struct vm_area_struct *vma;
2116 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2117 opt != PR_SET_MM_MAP &&
2118 opt != PR_SET_MM_MAP_SIZE)))
2121 #ifdef CONFIG_CHECKPOINT_RESTORE
2122 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2123 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2126 if (!capable(CAP_SYS_RESOURCE))
2129 if (opt == PR_SET_MM_EXE_FILE)
2130 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2132 if (opt == PR_SET_MM_AUXV)
2133 return prctl_set_auxv(mm, addr, arg4);
2135 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2141 * arg_lock protects concurent updates of arg boundaries, we need
2142 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2146 vma = find_vma(mm, addr);
2148 spin_lock(&mm->arg_lock);
2149 prctl_map.start_code = mm->start_code;
2150 prctl_map.end_code = mm->end_code;
2151 prctl_map.start_data = mm->start_data;
2152 prctl_map.end_data = mm->end_data;
2153 prctl_map.start_brk = mm->start_brk;
2154 prctl_map.brk = mm->brk;
2155 prctl_map.start_stack = mm->start_stack;
2156 prctl_map.arg_start = mm->arg_start;
2157 prctl_map.arg_end = mm->arg_end;
2158 prctl_map.env_start = mm->env_start;
2159 prctl_map.env_end = mm->env_end;
2162 case PR_SET_MM_START_CODE:
2163 prctl_map.start_code = addr;
2165 case PR_SET_MM_END_CODE:
2166 prctl_map.end_code = addr;
2168 case PR_SET_MM_START_DATA:
2169 prctl_map.start_data = addr;
2171 case PR_SET_MM_END_DATA:
2172 prctl_map.end_data = addr;
2174 case PR_SET_MM_START_STACK:
2175 prctl_map.start_stack = addr;
2177 case PR_SET_MM_START_BRK:
2178 prctl_map.start_brk = addr;
2181 prctl_map.brk = addr;
2183 case PR_SET_MM_ARG_START:
2184 prctl_map.arg_start = addr;
2186 case PR_SET_MM_ARG_END:
2187 prctl_map.arg_end = addr;
2189 case PR_SET_MM_ENV_START:
2190 prctl_map.env_start = addr;
2192 case PR_SET_MM_ENV_END:
2193 prctl_map.env_end = addr;
2199 error = validate_prctl_map_addr(&prctl_map);
2205 * If command line arguments and environment
2206 * are placed somewhere else on stack, we can
2207 * set them up here, ARG_START/END to setup
2208 * command line argumets and ENV_START/END
2211 case PR_SET_MM_START_STACK:
2212 case PR_SET_MM_ARG_START:
2213 case PR_SET_MM_ARG_END:
2214 case PR_SET_MM_ENV_START:
2215 case PR_SET_MM_ENV_END:
2222 mm->start_code = prctl_map.start_code;
2223 mm->end_code = prctl_map.end_code;
2224 mm->start_data = prctl_map.start_data;
2225 mm->end_data = prctl_map.end_data;
2226 mm->start_brk = prctl_map.start_brk;
2227 mm->brk = prctl_map.brk;
2228 mm->start_stack = prctl_map.start_stack;
2229 mm->arg_start = prctl_map.arg_start;
2230 mm->arg_end = prctl_map.arg_end;
2231 mm->env_start = prctl_map.env_start;
2232 mm->env_end = prctl_map.env_end;
2236 spin_unlock(&mm->arg_lock);
2237 mmap_read_unlock(mm);
2241 #ifdef CONFIG_CHECKPOINT_RESTORE
2242 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2244 return put_user(me->clear_child_tid, tid_addr);
2247 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2253 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2256 * If task has has_child_subreaper - all its decendants
2257 * already have these flag too and new decendants will
2258 * inherit it on fork, skip them.
2260 * If we've found child_reaper - skip descendants in
2261 * it's subtree as they will never get out pidns.
2263 if (p->signal->has_child_subreaper ||
2264 is_child_reaper(task_pid(p)))
2267 p->signal->has_child_subreaper = 1;
2271 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2276 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2282 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2284 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2285 unsigned long, arg4, unsigned long, arg5)
2287 struct task_struct *me = current;
2288 unsigned char comm[sizeof(me->comm)];
2291 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2292 if (error != -ENOSYS)
2297 case PR_SET_PDEATHSIG:
2298 if (!valid_signal(arg2)) {
2302 me->pdeath_signal = arg2;
2304 case PR_GET_PDEATHSIG:
2305 error = put_user(me->pdeath_signal, (int __user *)arg2);
2307 case PR_GET_DUMPABLE:
2308 error = get_dumpable(me->mm);
2310 case PR_SET_DUMPABLE:
2311 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2315 set_dumpable(me->mm, arg2);
2318 case PR_SET_UNALIGN:
2319 error = SET_UNALIGN_CTL(me, arg2);
2321 case PR_GET_UNALIGN:
2322 error = GET_UNALIGN_CTL(me, arg2);
2325 error = SET_FPEMU_CTL(me, arg2);
2328 error = GET_FPEMU_CTL(me, arg2);
2331 error = SET_FPEXC_CTL(me, arg2);
2334 error = GET_FPEXC_CTL(me, arg2);
2337 error = PR_TIMING_STATISTICAL;
2340 if (arg2 != PR_TIMING_STATISTICAL)
2344 comm[sizeof(me->comm) - 1] = 0;
2345 if (strncpy_from_user(comm, (char __user *)arg2,
2346 sizeof(me->comm) - 1) < 0)
2348 set_task_comm(me, comm);
2349 proc_comm_connector(me);
2352 get_task_comm(comm, me);
2353 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2357 error = GET_ENDIAN(me, arg2);
2360 error = SET_ENDIAN(me, arg2);
2362 case PR_GET_SECCOMP:
2363 error = prctl_get_seccomp();
2365 case PR_SET_SECCOMP:
2366 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2369 error = GET_TSC_CTL(arg2);
2372 error = SET_TSC_CTL(arg2);
2374 case PR_TASK_PERF_EVENTS_DISABLE:
2375 error = perf_event_task_disable();
2377 case PR_TASK_PERF_EVENTS_ENABLE:
2378 error = perf_event_task_enable();
2380 case PR_GET_TIMERSLACK:
2381 if (current->timer_slack_ns > ULONG_MAX)
2384 error = current->timer_slack_ns;
2386 case PR_SET_TIMERSLACK:
2388 current->timer_slack_ns =
2389 current->default_timer_slack_ns;
2391 current->timer_slack_ns = arg2;
2397 case PR_MCE_KILL_CLEAR:
2400 current->flags &= ~PF_MCE_PROCESS;
2402 case PR_MCE_KILL_SET:
2403 current->flags |= PF_MCE_PROCESS;
2404 if (arg3 == PR_MCE_KILL_EARLY)
2405 current->flags |= PF_MCE_EARLY;
2406 else if (arg3 == PR_MCE_KILL_LATE)
2407 current->flags &= ~PF_MCE_EARLY;
2408 else if (arg3 == PR_MCE_KILL_DEFAULT)
2410 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2418 case PR_MCE_KILL_GET:
2419 if (arg2 | arg3 | arg4 | arg5)
2421 if (current->flags & PF_MCE_PROCESS)
2422 error = (current->flags & PF_MCE_EARLY) ?
2423 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2425 error = PR_MCE_KILL_DEFAULT;
2428 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2430 case PR_GET_TID_ADDRESS:
2431 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2433 case PR_SET_CHILD_SUBREAPER:
2434 me->signal->is_child_subreaper = !!arg2;
2438 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2440 case PR_GET_CHILD_SUBREAPER:
2441 error = put_user(me->signal->is_child_subreaper,
2442 (int __user *)arg2);
2444 case PR_SET_NO_NEW_PRIVS:
2445 if (arg2 != 1 || arg3 || arg4 || arg5)
2448 task_set_no_new_privs(current);
2450 case PR_GET_NO_NEW_PRIVS:
2451 if (arg2 || arg3 || arg4 || arg5)
2453 return task_no_new_privs(current) ? 1 : 0;
2454 case PR_GET_THP_DISABLE:
2455 if (arg2 || arg3 || arg4 || arg5)
2457 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2459 case PR_SET_THP_DISABLE:
2460 if (arg3 || arg4 || arg5)
2462 if (mmap_write_lock_killable(me->mm))
2465 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2467 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2468 mmap_write_unlock(me->mm);
2470 case PR_MPX_ENABLE_MANAGEMENT:
2471 case PR_MPX_DISABLE_MANAGEMENT:
2472 /* No longer implemented: */
2474 case PR_SET_FP_MODE:
2475 error = SET_FP_MODE(me, arg2);
2477 case PR_GET_FP_MODE:
2478 error = GET_FP_MODE(me);
2481 error = SVE_SET_VL(arg2);
2484 error = SVE_GET_VL();
2486 case PR_GET_SPECULATION_CTRL:
2487 if (arg3 || arg4 || arg5)
2489 error = arch_prctl_spec_ctrl_get(me, arg2);
2491 case PR_SET_SPECULATION_CTRL:
2494 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2496 case PR_PAC_RESET_KEYS:
2497 if (arg3 || arg4 || arg5)
2499 error = PAC_RESET_KEYS(me, arg2);
2501 case PR_SET_TAGGED_ADDR_CTRL:
2502 if (arg3 || arg4 || arg5)
2504 error = SET_TAGGED_ADDR_CTRL(arg2);
2506 case PR_GET_TAGGED_ADDR_CTRL:
2507 if (arg2 || arg3 || arg4 || arg5)
2509 error = GET_TAGGED_ADDR_CTRL();
2511 case PR_SET_IO_FLUSHER:
2512 if (!capable(CAP_SYS_RESOURCE))
2515 if (arg3 || arg4 || arg5)
2519 current->flags |= PR_IO_FLUSHER;
2521 current->flags &= ~PR_IO_FLUSHER;
2525 case PR_GET_IO_FLUSHER:
2526 if (!capable(CAP_SYS_RESOURCE))
2529 if (arg2 || arg3 || arg4 || arg5)
2532 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2534 case PR_SET_SYSCALL_USER_DISPATCH:
2535 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2536 (char __user *) arg5);
2545 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2546 struct getcpu_cache __user *, unused)
2549 int cpu = raw_smp_processor_id();
2552 err |= put_user(cpu, cpup);
2554 err |= put_user(cpu_to_node(cpu), nodep);
2555 return err ? -EFAULT : 0;
2559 * do_sysinfo - fill in sysinfo struct
2560 * @info: pointer to buffer to fill
2562 static int do_sysinfo(struct sysinfo *info)
2564 unsigned long mem_total, sav_total;
2565 unsigned int mem_unit, bitcount;
2566 struct timespec64 tp;
2568 memset(info, 0, sizeof(struct sysinfo));
2570 ktime_get_boottime_ts64(&tp);
2571 timens_add_boottime(&tp);
2572 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2574 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2576 info->procs = nr_threads;
2582 * If the sum of all the available memory (i.e. ram + swap)
2583 * is less than can be stored in a 32 bit unsigned long then
2584 * we can be binary compatible with 2.2.x kernels. If not,
2585 * well, in that case 2.2.x was broken anyways...
2587 * -Erik Andersen <andersee@debian.org>
2590 mem_total = info->totalram + info->totalswap;
2591 if (mem_total < info->totalram || mem_total < info->totalswap)
2594 mem_unit = info->mem_unit;
2595 while (mem_unit > 1) {
2598 sav_total = mem_total;
2600 if (mem_total < sav_total)
2605 * If mem_total did not overflow, multiply all memory values by
2606 * info->mem_unit and set it to 1. This leaves things compatible
2607 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2612 info->totalram <<= bitcount;
2613 info->freeram <<= bitcount;
2614 info->sharedram <<= bitcount;
2615 info->bufferram <<= bitcount;
2616 info->totalswap <<= bitcount;
2617 info->freeswap <<= bitcount;
2618 info->totalhigh <<= bitcount;
2619 info->freehigh <<= bitcount;
2625 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2631 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2637 #ifdef CONFIG_COMPAT
2638 struct compat_sysinfo {
2652 char _f[20-2*sizeof(u32)-sizeof(int)];
2655 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2658 struct compat_sysinfo s_32;
2662 /* Check to see if any memory value is too large for 32-bit and scale
2665 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2668 while (s.mem_unit < PAGE_SIZE) {
2673 s.totalram >>= bitcount;
2674 s.freeram >>= bitcount;
2675 s.sharedram >>= bitcount;
2676 s.bufferram >>= bitcount;
2677 s.totalswap >>= bitcount;
2678 s.freeswap >>= bitcount;
2679 s.totalhigh >>= bitcount;
2680 s.freehigh >>= bitcount;
2683 memset(&s_32, 0, sizeof(s_32));
2684 s_32.uptime = s.uptime;
2685 s_32.loads[0] = s.loads[0];
2686 s_32.loads[1] = s.loads[1];
2687 s_32.loads[2] = s.loads[2];
2688 s_32.totalram = s.totalram;
2689 s_32.freeram = s.freeram;
2690 s_32.sharedram = s.sharedram;
2691 s_32.bufferram = s.bufferram;
2692 s_32.totalswap = s.totalswap;
2693 s_32.freeswap = s.freeswap;
2694 s_32.procs = s.procs;
2695 s_32.totalhigh = s.totalhigh;
2696 s_32.freehigh = s.freehigh;
2697 s_32.mem_unit = s.mem_unit;
2698 if (copy_to_user(info, &s_32, sizeof(s_32)))
2702 #endif /* CONFIG_COMPAT */