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/time_namespace.h>
51 #include <linux/binfmts.h>
53 #include <linux/sched.h>
54 #include <linux/sched/autogroup.h>
55 #include <linux/sched/loadavg.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/mm.h>
58 #include <linux/sched/coredump.h>
59 #include <linux/sched/task.h>
60 #include <linux/sched/cputime.h>
61 #include <linux/rcupdate.h>
62 #include <linux/uidgid.h>
63 #include <linux/cred.h>
65 #include <linux/nospec.h>
67 #include <linux/kmsg_dump.h>
68 /* Move somewhere else to avoid recompiling? */
69 #include <generated/utsrelease.h>
71 #include <linux/uaccess.h>
73 #include <asm/unistd.h>
77 #ifndef SET_UNALIGN_CTL
78 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
80 #ifndef GET_UNALIGN_CTL
81 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
84 # define SET_FPEMU_CTL(a, b) (-EINVAL)
87 # define GET_FPEMU_CTL(a, b) (-EINVAL)
90 # define SET_FPEXC_CTL(a, b) (-EINVAL)
93 # define GET_FPEXC_CTL(a, b) (-EINVAL)
96 # define GET_ENDIAN(a, b) (-EINVAL)
99 # define SET_ENDIAN(a, b) (-EINVAL)
102 # define GET_TSC_CTL(a) (-EINVAL)
105 # define SET_TSC_CTL(a) (-EINVAL)
108 # define GET_FP_MODE(a) (-EINVAL)
111 # define SET_FP_MODE(a,b) (-EINVAL)
114 # define SVE_SET_VL(a) (-EINVAL)
117 # define SVE_GET_VL() (-EINVAL)
119 #ifndef PAC_RESET_KEYS
120 # define PAC_RESET_KEYS(a, b) (-EINVAL)
122 #ifndef SET_TAGGED_ADDR_CTRL
123 # define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
125 #ifndef GET_TAGGED_ADDR_CTRL
126 # define GET_TAGGED_ADDR_CTRL() (-EINVAL)
130 * this is where the system-wide overflow UID and GID are defined, for
131 * architectures that now have 32-bit UID/GID but didn't in the past
134 int overflowuid = DEFAULT_OVERFLOWUID;
135 int overflowgid = DEFAULT_OVERFLOWGID;
137 EXPORT_SYMBOL(overflowuid);
138 EXPORT_SYMBOL(overflowgid);
141 * the same as above, but for filesystems which can only store a 16-bit
142 * UID and GID. as such, this is needed on all architectures
145 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
146 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
148 EXPORT_SYMBOL(fs_overflowuid);
149 EXPORT_SYMBOL(fs_overflowgid);
152 * Returns true if current's euid is same as p's uid or euid,
153 * or has CAP_SYS_NICE to p's user_ns.
155 * Called with rcu_read_lock, creds are safe
157 static bool set_one_prio_perm(struct task_struct *p)
159 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
161 if (uid_eq(pcred->uid, cred->euid) ||
162 uid_eq(pcred->euid, cred->euid))
164 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
170 * set the priority of a task
171 * - the caller must hold the RCU read lock
173 static int set_one_prio(struct task_struct *p, int niceval, int error)
177 if (!set_one_prio_perm(p)) {
181 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
185 no_nice = security_task_setnice(p, niceval);
192 set_user_nice(p, niceval);
197 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
199 struct task_struct *g, *p;
200 struct user_struct *user;
201 const struct cred *cred = current_cred();
206 if (which > PRIO_USER || which < PRIO_PROCESS)
209 /* normalize: avoid signed division (rounding problems) */
211 if (niceval < MIN_NICE)
213 if (niceval > MAX_NICE)
217 read_lock(&tasklist_lock);
221 p = find_task_by_vpid(who);
225 error = set_one_prio(p, niceval, error);
229 pgrp = find_vpid(who);
231 pgrp = task_pgrp(current);
232 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
233 error = set_one_prio(p, niceval, error);
234 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
237 uid = make_kuid(cred->user_ns, who);
241 else if (!uid_eq(uid, cred->uid)) {
242 user = find_user(uid);
244 goto out_unlock; /* No processes for this user */
246 do_each_thread(g, p) {
247 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
248 error = set_one_prio(p, niceval, error);
249 } while_each_thread(g, p);
250 if (!uid_eq(uid, cred->uid))
251 free_uid(user); /* For find_user() */
255 read_unlock(&tasklist_lock);
262 * Ugh. To avoid negative return values, "getpriority()" will
263 * not return the normal nice-value, but a negated value that
264 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
265 * to stay compatible.
267 SYSCALL_DEFINE2(getpriority, int, which, int, who)
269 struct task_struct *g, *p;
270 struct user_struct *user;
271 const struct cred *cred = current_cred();
272 long niceval, retval = -ESRCH;
276 if (which > PRIO_USER || which < PRIO_PROCESS)
280 read_lock(&tasklist_lock);
284 p = find_task_by_vpid(who);
288 niceval = nice_to_rlimit(task_nice(p));
289 if (niceval > retval)
295 pgrp = find_vpid(who);
297 pgrp = task_pgrp(current);
298 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
299 niceval = nice_to_rlimit(task_nice(p));
300 if (niceval > retval)
302 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
305 uid = make_kuid(cred->user_ns, who);
309 else if (!uid_eq(uid, cred->uid)) {
310 user = find_user(uid);
312 goto out_unlock; /* No processes for this user */
314 do_each_thread(g, p) {
315 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
316 niceval = nice_to_rlimit(task_nice(p));
317 if (niceval > retval)
320 } while_each_thread(g, p);
321 if (!uid_eq(uid, cred->uid))
322 free_uid(user); /* for find_user() */
326 read_unlock(&tasklist_lock);
333 * Unprivileged users may change the real gid to the effective gid
334 * or vice versa. (BSD-style)
336 * If you set the real gid at all, or set the effective gid to a value not
337 * equal to the real gid, then the saved gid is set to the new effective gid.
339 * This makes it possible for a setgid program to completely drop its
340 * privileges, which is often a useful assertion to make when you are doing
341 * a security audit over a program.
343 * The general idea is that a program which uses just setregid() will be
344 * 100% compatible with BSD. A program which uses just setgid() will be
345 * 100% compatible with POSIX with saved IDs.
347 * SMP: There are not races, the GIDs are checked only by filesystem
348 * operations (as far as semantic preservation is concerned).
350 #ifdef CONFIG_MULTIUSER
351 long __sys_setregid(gid_t rgid, gid_t egid)
353 struct user_namespace *ns = current_user_ns();
354 const struct cred *old;
359 krgid = make_kgid(ns, rgid);
360 kegid = make_kgid(ns, egid);
362 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
364 if ((egid != (gid_t) -1) && !gid_valid(kegid))
367 new = prepare_creds();
370 old = current_cred();
373 if (rgid != (gid_t) -1) {
374 if (gid_eq(old->gid, krgid) ||
375 gid_eq(old->egid, krgid) ||
376 ns_capable_setid(old->user_ns, CAP_SETGID))
381 if (egid != (gid_t) -1) {
382 if (gid_eq(old->gid, kegid) ||
383 gid_eq(old->egid, kegid) ||
384 gid_eq(old->sgid, kegid) ||
385 ns_capable_setid(old->user_ns, CAP_SETGID))
391 if (rgid != (gid_t) -1 ||
392 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
393 new->sgid = new->egid;
394 new->fsgid = new->egid;
396 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
400 return commit_creds(new);
407 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
409 return __sys_setregid(rgid, egid);
413 * setgid() is implemented like SysV w/ SAVED_IDS
415 * SMP: Same implicit races as above.
417 long __sys_setgid(gid_t gid)
419 struct user_namespace *ns = current_user_ns();
420 const struct cred *old;
425 kgid = make_kgid(ns, gid);
426 if (!gid_valid(kgid))
429 new = prepare_creds();
432 old = current_cred();
435 if (ns_capable_setid(old->user_ns, CAP_SETGID))
436 new->gid = new->egid = new->sgid = new->fsgid = kgid;
437 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
438 new->egid = new->fsgid = kgid;
442 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
446 return commit_creds(new);
453 SYSCALL_DEFINE1(setgid, gid_t, gid)
455 return __sys_setgid(gid);
459 * change the user struct in a credentials set to match the new UID
461 static int set_user(struct cred *new)
463 struct user_struct *new_user;
465 new_user = alloc_uid(new->uid);
470 * We don't fail in case of NPROC limit excess here because too many
471 * poorly written programs don't check set*uid() return code, assuming
472 * it never fails if called by root. We may still enforce NPROC limit
473 * for programs doing set*uid()+execve() by harmlessly deferring the
474 * failure to the execve() stage.
476 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
477 new_user != INIT_USER)
478 current->flags |= PF_NPROC_EXCEEDED;
480 current->flags &= ~PF_NPROC_EXCEEDED;
483 new->user = new_user;
488 * Unprivileged users may change the real uid to the effective uid
489 * or vice versa. (BSD-style)
491 * If you set the real uid at all, or set the effective uid to a value not
492 * equal to the real uid, then the saved uid is set to the new effective uid.
494 * This makes it possible for a setuid program to completely drop its
495 * privileges, which is often a useful assertion to make when you are doing
496 * a security audit over a program.
498 * The general idea is that a program which uses just setreuid() will be
499 * 100% compatible with BSD. A program which uses just setuid() will be
500 * 100% compatible with POSIX with saved IDs.
502 long __sys_setreuid(uid_t ruid, uid_t euid)
504 struct user_namespace *ns = current_user_ns();
505 const struct cred *old;
510 kruid = make_kuid(ns, ruid);
511 keuid = make_kuid(ns, euid);
513 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
515 if ((euid != (uid_t) -1) && !uid_valid(keuid))
518 new = prepare_creds();
521 old = current_cred();
524 if (ruid != (uid_t) -1) {
526 if (!uid_eq(old->uid, kruid) &&
527 !uid_eq(old->euid, kruid) &&
528 !ns_capable_setid(old->user_ns, CAP_SETUID))
532 if (euid != (uid_t) -1) {
534 if (!uid_eq(old->uid, keuid) &&
535 !uid_eq(old->euid, keuid) &&
536 !uid_eq(old->suid, keuid) &&
537 !ns_capable_setid(old->user_ns, CAP_SETUID))
541 if (!uid_eq(new->uid, old->uid)) {
542 retval = set_user(new);
546 if (ruid != (uid_t) -1 ||
547 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
548 new->suid = new->euid;
549 new->fsuid = new->euid;
551 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
555 return commit_creds(new);
562 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
564 return __sys_setreuid(ruid, euid);
568 * setuid() is implemented like SysV with SAVED_IDS
570 * Note that SAVED_ID's is deficient in that a setuid root program
571 * like sendmail, for example, cannot set its uid to be a normal
572 * user and then switch back, because if you're root, setuid() sets
573 * the saved uid too. If you don't like this, blame the bright people
574 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
575 * will allow a root program to temporarily drop privileges and be able to
576 * regain them by swapping the real and effective uid.
578 long __sys_setuid(uid_t uid)
580 struct user_namespace *ns = current_user_ns();
581 const struct cred *old;
586 kuid = make_kuid(ns, uid);
587 if (!uid_valid(kuid))
590 new = prepare_creds();
593 old = current_cred();
596 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
597 new->suid = new->uid = kuid;
598 if (!uid_eq(kuid, old->uid)) {
599 retval = set_user(new);
603 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
607 new->fsuid = new->euid = kuid;
609 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
613 return commit_creds(new);
620 SYSCALL_DEFINE1(setuid, uid_t, uid)
622 return __sys_setuid(uid);
627 * This function implements a generic ability to update ruid, euid,
628 * and suid. This allows you to implement the 4.4 compatible seteuid().
630 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
632 struct user_namespace *ns = current_user_ns();
633 const struct cred *old;
636 kuid_t kruid, keuid, ksuid;
637 bool ruid_new, euid_new, suid_new;
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 old = current_cred();
654 /* check for no-op */
655 if ((ruid == (uid_t) -1 || uid_eq(kruid, old->uid)) &&
656 (euid == (uid_t) -1 || (uid_eq(keuid, old->euid) &&
657 uid_eq(keuid, old->fsuid))) &&
658 (suid == (uid_t) -1 || uid_eq(ksuid, old->suid)))
661 ruid_new = ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
662 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid);
663 euid_new = euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
664 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid);
665 suid_new = suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
666 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid);
667 if ((ruid_new || euid_new || suid_new) &&
668 !ns_capable_setid(old->user_ns, CAP_SETUID))
671 new = prepare_creds();
675 if (ruid != (uid_t) -1) {
677 if (!uid_eq(kruid, old->uid)) {
678 retval = set_user(new);
683 if (euid != (uid_t) -1)
685 if (suid != (uid_t) -1)
687 new->fsuid = new->euid;
689 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
693 return commit_creds(new);
700 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
702 return __sys_setresuid(ruid, euid, suid);
705 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
707 const struct cred *cred = current_cred();
709 uid_t ruid, euid, suid;
711 ruid = from_kuid_munged(cred->user_ns, cred->uid);
712 euid = from_kuid_munged(cred->user_ns, cred->euid);
713 suid = from_kuid_munged(cred->user_ns, cred->suid);
715 retval = put_user(ruid, ruidp);
717 retval = put_user(euid, euidp);
719 return put_user(suid, suidp);
725 * Same as above, but for rgid, egid, sgid.
727 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
729 struct user_namespace *ns = current_user_ns();
730 const struct cred *old;
733 kgid_t krgid, kegid, ksgid;
734 bool rgid_new, egid_new, sgid_new;
736 krgid = make_kgid(ns, rgid);
737 kegid = make_kgid(ns, egid);
738 ksgid = make_kgid(ns, sgid);
740 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
742 if ((egid != (gid_t) -1) && !gid_valid(kegid))
744 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
747 old = current_cred();
749 /* check for no-op */
750 if ((rgid == (gid_t) -1 || gid_eq(krgid, old->gid)) &&
751 (egid == (gid_t) -1 || (gid_eq(kegid, old->egid) &&
752 gid_eq(kegid, old->fsgid))) &&
753 (sgid == (gid_t) -1 || gid_eq(ksgid, old->sgid)))
756 rgid_new = rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
757 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid);
758 egid_new = egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
759 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid);
760 sgid_new = sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
761 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid);
762 if ((rgid_new || egid_new || sgid_new) &&
763 !ns_capable_setid(old->user_ns, CAP_SETGID))
766 new = prepare_creds();
770 if (rgid != (gid_t) -1)
772 if (egid != (gid_t) -1)
774 if (sgid != (gid_t) -1)
776 new->fsgid = new->egid;
778 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
782 return commit_creds(new);
789 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
791 return __sys_setresgid(rgid, egid, sgid);
794 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
796 const struct cred *cred = current_cred();
798 gid_t rgid, egid, sgid;
800 rgid = from_kgid_munged(cred->user_ns, cred->gid);
801 egid = from_kgid_munged(cred->user_ns, cred->egid);
802 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
804 retval = put_user(rgid, rgidp);
806 retval = put_user(egid, egidp);
808 retval = put_user(sgid, sgidp);
816 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
817 * is used for "access()" and for the NFS daemon (letting nfsd stay at
818 * whatever uid it wants to). It normally shadows "euid", except when
819 * explicitly set by setfsuid() or for access..
821 long __sys_setfsuid(uid_t uid)
823 const struct cred *old;
828 old = current_cred();
829 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
831 kuid = make_kuid(old->user_ns, uid);
832 if (!uid_valid(kuid))
835 new = prepare_creds();
839 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
840 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
841 ns_capable_setid(old->user_ns, CAP_SETUID)) {
842 if (!uid_eq(kuid, old->fsuid)) {
844 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
857 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
859 return __sys_setfsuid(uid);
863 * Samma på svenska..
865 long __sys_setfsgid(gid_t gid)
867 const struct cred *old;
872 old = current_cred();
873 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
875 kgid = make_kgid(old->user_ns, gid);
876 if (!gid_valid(kgid))
879 new = prepare_creds();
883 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
884 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
885 ns_capable_setid(old->user_ns, CAP_SETGID)) {
886 if (!gid_eq(kgid, old->fsgid)) {
888 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
901 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
903 return __sys_setfsgid(gid);
905 #endif /* CONFIG_MULTIUSER */
908 * sys_getpid - return the thread group id of the current process
910 * Note, despite the name, this returns the tgid not the pid. The tgid and
911 * the pid are identical unless CLONE_THREAD was specified on clone() in
912 * which case the tgid is the same in all threads of the same group.
914 * This is SMP safe as current->tgid does not change.
916 SYSCALL_DEFINE0(getpid)
918 return task_tgid_vnr(current);
921 /* Thread ID - the internal kernel "pid" */
922 SYSCALL_DEFINE0(gettid)
924 return task_pid_vnr(current);
928 * Accessing ->real_parent is not SMP-safe, it could
929 * change from under us. However, we can use a stale
930 * value of ->real_parent under rcu_read_lock(), see
931 * release_task()->call_rcu(delayed_put_task_struct).
933 SYSCALL_DEFINE0(getppid)
938 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
944 SYSCALL_DEFINE0(getuid)
946 /* Only we change this so SMP safe */
947 return from_kuid_munged(current_user_ns(), current_uid());
950 SYSCALL_DEFINE0(geteuid)
952 /* Only we change this so SMP safe */
953 return from_kuid_munged(current_user_ns(), current_euid());
956 SYSCALL_DEFINE0(getgid)
958 /* Only we change this so SMP safe */
959 return from_kgid_munged(current_user_ns(), current_gid());
962 SYSCALL_DEFINE0(getegid)
964 /* Only we change this so SMP safe */
965 return from_kgid_munged(current_user_ns(), current_egid());
968 static void do_sys_times(struct tms *tms)
970 u64 tgutime, tgstime, cutime, cstime;
972 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
973 cutime = current->signal->cutime;
974 cstime = current->signal->cstime;
975 tms->tms_utime = nsec_to_clock_t(tgutime);
976 tms->tms_stime = nsec_to_clock_t(tgstime);
977 tms->tms_cutime = nsec_to_clock_t(cutime);
978 tms->tms_cstime = nsec_to_clock_t(cstime);
981 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
987 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
990 force_successful_syscall_return();
991 return (long) jiffies_64_to_clock_t(get_jiffies_64());
995 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
997 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1000 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1004 struct compat_tms tmp;
1007 /* Convert our struct tms to the compat version. */
1008 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1009 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1010 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1011 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1012 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1015 force_successful_syscall_return();
1016 return compat_jiffies_to_clock_t(jiffies);
1021 * This needs some heavy checking ...
1022 * I just haven't the stomach for it. I also don't fully
1023 * understand sessions/pgrp etc. Let somebody who does explain it.
1025 * OK, I think I have the protection semantics right.... this is really
1026 * only important on a multi-user system anyway, to make sure one user
1027 * can't send a signal to a process owned by another. -TYT, 12/12/91
1029 * !PF_FORKNOEXEC check to conform completely to POSIX.
1031 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1033 struct task_struct *p;
1034 struct task_struct *group_leader = current->group_leader;
1039 pid = task_pid_vnr(group_leader);
1046 /* From this point forward we keep holding onto the tasklist lock
1047 * so that our parent does not change from under us. -DaveM
1049 write_lock_irq(&tasklist_lock);
1052 p = find_task_by_vpid(pid);
1057 if (!thread_group_leader(p))
1060 if (same_thread_group(p->real_parent, group_leader)) {
1062 if (task_session(p) != task_session(group_leader))
1065 if (!(p->flags & PF_FORKNOEXEC))
1069 if (p != group_leader)
1074 if (p->signal->leader)
1079 struct task_struct *g;
1081 pgrp = find_vpid(pgid);
1082 g = pid_task(pgrp, PIDTYPE_PGID);
1083 if (!g || task_session(g) != task_session(group_leader))
1087 err = security_task_setpgid(p, pgid);
1091 if (task_pgrp(p) != pgrp)
1092 change_pid(p, PIDTYPE_PGID, pgrp);
1096 /* All paths lead to here, thus we are safe. -DaveM */
1097 write_unlock_irq(&tasklist_lock);
1102 static int do_getpgid(pid_t pid)
1104 struct task_struct *p;
1110 grp = task_pgrp(current);
1113 p = find_task_by_vpid(pid);
1120 retval = security_task_getpgid(p);
1124 retval = pid_vnr(grp);
1130 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1132 return do_getpgid(pid);
1135 #ifdef __ARCH_WANT_SYS_GETPGRP
1137 SYSCALL_DEFINE0(getpgrp)
1139 return do_getpgid(0);
1144 SYSCALL_DEFINE1(getsid, pid_t, pid)
1146 struct task_struct *p;
1152 sid = task_session(current);
1155 p = find_task_by_vpid(pid);
1158 sid = task_session(p);
1162 retval = security_task_getsid(p);
1166 retval = pid_vnr(sid);
1172 static void set_special_pids(struct pid *pid)
1174 struct task_struct *curr = current->group_leader;
1176 if (task_session(curr) != pid)
1177 change_pid(curr, PIDTYPE_SID, pid);
1179 if (task_pgrp(curr) != pid)
1180 change_pid(curr, PIDTYPE_PGID, pid);
1183 int ksys_setsid(void)
1185 struct task_struct *group_leader = current->group_leader;
1186 struct pid *sid = task_pid(group_leader);
1187 pid_t session = pid_vnr(sid);
1190 write_lock_irq(&tasklist_lock);
1191 /* Fail if I am already a session leader */
1192 if (group_leader->signal->leader)
1195 /* Fail if a process group id already exists that equals the
1196 * proposed session id.
1198 if (pid_task(sid, PIDTYPE_PGID))
1201 group_leader->signal->leader = 1;
1202 set_special_pids(sid);
1204 proc_clear_tty(group_leader);
1208 write_unlock_irq(&tasklist_lock);
1210 proc_sid_connector(group_leader);
1211 sched_autogroup_create_attach(group_leader);
1216 SYSCALL_DEFINE0(setsid)
1218 return ksys_setsid();
1221 DECLARE_RWSEM(uts_sem);
1223 #ifdef COMPAT_UTS_MACHINE
1224 #define override_architecture(name) \
1225 (personality(current->personality) == PER_LINUX32 && \
1226 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1227 sizeof(COMPAT_UTS_MACHINE)))
1229 #define override_architecture(name) 0
1233 * Work around broken programs that cannot handle "Linux 3.0".
1234 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1235 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1238 static int override_release(char __user *release, size_t len)
1242 if (current->personality & UNAME26) {
1243 const char *rest = UTS_RELEASE;
1244 char buf[65] = { 0 };
1250 if (*rest == '.' && ++ndots >= 3)
1252 if (!isdigit(*rest) && *rest != '.')
1256 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1257 copy = clamp_t(size_t, len, 1, sizeof(buf));
1258 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1259 ret = copy_to_user(release, buf, copy + 1);
1264 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1266 struct new_utsname tmp;
1268 down_read(&uts_sem);
1269 memcpy(&tmp, utsname(), sizeof(tmp));
1271 if (copy_to_user(name, &tmp, sizeof(tmp)))
1274 if (override_release(name->release, sizeof(name->release)))
1276 if (override_architecture(name))
1281 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1285 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1287 struct old_utsname tmp;
1292 down_read(&uts_sem);
1293 memcpy(&tmp, utsname(), sizeof(tmp));
1295 if (copy_to_user(name, &tmp, sizeof(tmp)))
1298 if (override_release(name->release, sizeof(name->release)))
1300 if (override_architecture(name))
1305 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1307 struct oldold_utsname tmp;
1312 memset(&tmp, 0, sizeof(tmp));
1314 down_read(&uts_sem);
1315 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1316 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1317 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1318 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1319 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1321 if (copy_to_user(name, &tmp, sizeof(tmp)))
1324 if (override_architecture(name))
1326 if (override_release(name->release, sizeof(name->release)))
1332 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1335 char tmp[__NEW_UTS_LEN];
1337 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1340 if (len < 0 || len > __NEW_UTS_LEN)
1343 if (!copy_from_user(tmp, name, len)) {
1344 struct new_utsname *u;
1346 down_write(&uts_sem);
1348 memcpy(u->nodename, tmp, len);
1349 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1351 uts_proc_notify(UTS_PROC_HOSTNAME);
1357 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1359 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1362 struct new_utsname *u;
1363 char tmp[__NEW_UTS_LEN + 1];
1367 down_read(&uts_sem);
1369 i = 1 + strlen(u->nodename);
1372 memcpy(tmp, u->nodename, i);
1374 if (copy_to_user(name, tmp, i))
1382 * Only setdomainname; getdomainname can be implemented by calling
1385 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1388 char tmp[__NEW_UTS_LEN];
1390 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1392 if (len < 0 || len > __NEW_UTS_LEN)
1396 if (!copy_from_user(tmp, name, len)) {
1397 struct new_utsname *u;
1399 down_write(&uts_sem);
1401 memcpy(u->domainname, tmp, len);
1402 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1404 uts_proc_notify(UTS_PROC_DOMAINNAME);
1410 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1412 struct rlimit value;
1415 ret = do_prlimit(current, resource, NULL, &value);
1417 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1422 #ifdef CONFIG_COMPAT
1424 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1425 struct compat_rlimit __user *, rlim)
1428 struct compat_rlimit r32;
1430 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1433 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1434 r.rlim_cur = RLIM_INFINITY;
1436 r.rlim_cur = r32.rlim_cur;
1437 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1438 r.rlim_max = RLIM_INFINITY;
1440 r.rlim_max = r32.rlim_max;
1441 return do_prlimit(current, resource, &r, NULL);
1444 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1445 struct compat_rlimit __user *, rlim)
1450 ret = do_prlimit(current, resource, NULL, &r);
1452 struct compat_rlimit r32;
1453 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1454 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1456 r32.rlim_cur = r.rlim_cur;
1457 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1458 r32.rlim_max = COMPAT_RLIM_INFINITY;
1460 r32.rlim_max = r.rlim_max;
1462 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1470 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1473 * Back compatibility for getrlimit. Needed for some apps.
1475 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1476 struct rlimit __user *, rlim)
1479 if (resource >= RLIM_NLIMITS)
1482 resource = array_index_nospec(resource, RLIM_NLIMITS);
1483 task_lock(current->group_leader);
1484 x = current->signal->rlim[resource];
1485 task_unlock(current->group_leader);
1486 if (x.rlim_cur > 0x7FFFFFFF)
1487 x.rlim_cur = 0x7FFFFFFF;
1488 if (x.rlim_max > 0x7FFFFFFF)
1489 x.rlim_max = 0x7FFFFFFF;
1490 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1493 #ifdef CONFIG_COMPAT
1494 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1495 struct compat_rlimit __user *, rlim)
1499 if (resource >= RLIM_NLIMITS)
1502 resource = array_index_nospec(resource, RLIM_NLIMITS);
1503 task_lock(current->group_leader);
1504 r = current->signal->rlim[resource];
1505 task_unlock(current->group_leader);
1506 if (r.rlim_cur > 0x7FFFFFFF)
1507 r.rlim_cur = 0x7FFFFFFF;
1508 if (r.rlim_max > 0x7FFFFFFF)
1509 r.rlim_max = 0x7FFFFFFF;
1511 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1512 put_user(r.rlim_max, &rlim->rlim_max))
1520 static inline bool rlim64_is_infinity(__u64 rlim64)
1522 #if BITS_PER_LONG < 64
1523 return rlim64 >= ULONG_MAX;
1525 return rlim64 == RLIM64_INFINITY;
1529 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1531 if (rlim->rlim_cur == RLIM_INFINITY)
1532 rlim64->rlim_cur = RLIM64_INFINITY;
1534 rlim64->rlim_cur = rlim->rlim_cur;
1535 if (rlim->rlim_max == RLIM_INFINITY)
1536 rlim64->rlim_max = RLIM64_INFINITY;
1538 rlim64->rlim_max = rlim->rlim_max;
1541 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1543 if (rlim64_is_infinity(rlim64->rlim_cur))
1544 rlim->rlim_cur = RLIM_INFINITY;
1546 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1547 if (rlim64_is_infinity(rlim64->rlim_max))
1548 rlim->rlim_max = RLIM_INFINITY;
1550 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1553 /* make sure you are allowed to change @tsk limits before calling this */
1554 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1555 struct rlimit *new_rlim, struct rlimit *old_rlim)
1557 struct rlimit *rlim;
1560 if (resource >= RLIM_NLIMITS)
1562 resource = array_index_nospec(resource, RLIM_NLIMITS);
1565 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1567 if (resource == RLIMIT_NOFILE &&
1568 new_rlim->rlim_max > sysctl_nr_open)
1572 /* protect tsk->signal and tsk->sighand from disappearing */
1573 read_lock(&tasklist_lock);
1574 if (!tsk->sighand) {
1579 rlim = tsk->signal->rlim + resource;
1580 task_lock(tsk->group_leader);
1582 /* Keep the capable check against init_user_ns until
1583 cgroups can contain all limits */
1584 if (new_rlim->rlim_max > rlim->rlim_max &&
1585 !capable(CAP_SYS_RESOURCE))
1588 retval = security_task_setrlimit(tsk, resource, new_rlim);
1596 task_unlock(tsk->group_leader);
1599 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1600 * infite. In case of RLIM_INFINITY the posix CPU timer code
1601 * ignores the rlimit.
1603 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1604 new_rlim->rlim_cur != RLIM_INFINITY &&
1605 IS_ENABLED(CONFIG_POSIX_TIMERS))
1606 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1608 read_unlock(&tasklist_lock);
1612 /* rcu lock must be held */
1613 static int check_prlimit_permission(struct task_struct *task,
1616 const struct cred *cred = current_cred(), *tcred;
1619 if (current == task)
1622 tcred = __task_cred(task);
1623 id_match = (uid_eq(cred->uid, tcred->euid) &&
1624 uid_eq(cred->uid, tcred->suid) &&
1625 uid_eq(cred->uid, tcred->uid) &&
1626 gid_eq(cred->gid, tcred->egid) &&
1627 gid_eq(cred->gid, tcred->sgid) &&
1628 gid_eq(cred->gid, tcred->gid));
1629 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1632 return security_task_prlimit(cred, tcred, flags);
1635 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1636 const struct rlimit64 __user *, new_rlim,
1637 struct rlimit64 __user *, old_rlim)
1639 struct rlimit64 old64, new64;
1640 struct rlimit old, new;
1641 struct task_struct *tsk;
1642 unsigned int checkflags = 0;
1646 checkflags |= LSM_PRLIMIT_READ;
1649 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1651 rlim64_to_rlim(&new64, &new);
1652 checkflags |= LSM_PRLIMIT_WRITE;
1656 tsk = pid ? find_task_by_vpid(pid) : current;
1661 ret = check_prlimit_permission(tsk, checkflags);
1666 get_task_struct(tsk);
1669 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1670 old_rlim ? &old : NULL);
1672 if (!ret && old_rlim) {
1673 rlim_to_rlim64(&old, &old64);
1674 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1678 put_task_struct(tsk);
1682 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1684 struct rlimit new_rlim;
1686 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1688 return do_prlimit(current, resource, &new_rlim, NULL);
1692 * It would make sense to put struct rusage in the task_struct,
1693 * except that would make the task_struct be *really big*. After
1694 * task_struct gets moved into malloc'ed memory, it would
1695 * make sense to do this. It will make moving the rest of the information
1696 * a lot simpler! (Which we're not doing right now because we're not
1697 * measuring them yet).
1699 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1700 * races with threads incrementing their own counters. But since word
1701 * reads are atomic, we either get new values or old values and we don't
1702 * care which for the sums. We always take the siglock to protect reading
1703 * the c* fields from p->signal from races with exit.c updating those
1704 * fields when reaping, so a sample either gets all the additions of a
1705 * given child after it's reaped, or none so this sample is before reaping.
1708 * We need to take the siglock for CHILDEREN, SELF and BOTH
1709 * for the cases current multithreaded, non-current single threaded
1710 * non-current multithreaded. Thread traversal is now safe with
1712 * Strictly speaking, we donot need to take the siglock if we are current and
1713 * single threaded, as no one else can take our signal_struct away, no one
1714 * else can reap the children to update signal->c* counters, and no one else
1715 * can race with the signal-> fields. If we do not take any lock, the
1716 * signal-> fields could be read out of order while another thread was just
1717 * exiting. So we should place a read memory barrier when we avoid the lock.
1718 * On the writer side, write memory barrier is implied in __exit_signal
1719 * as __exit_signal releases the siglock spinlock after updating the signal->
1720 * fields. But we don't do this yet to keep things simple.
1724 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1726 r->ru_nvcsw += t->nvcsw;
1727 r->ru_nivcsw += t->nivcsw;
1728 r->ru_minflt += t->min_flt;
1729 r->ru_majflt += t->maj_flt;
1730 r->ru_inblock += task_io_get_inblock(t);
1731 r->ru_oublock += task_io_get_oublock(t);
1734 void getrusage(struct task_struct *p, int who, struct rusage *r)
1736 struct task_struct *t;
1737 unsigned long flags;
1738 u64 tgutime, tgstime, utime, stime;
1739 unsigned long maxrss;
1740 struct mm_struct *mm;
1741 struct signal_struct *sig = p->signal;
1742 unsigned int seq = 0;
1745 memset(r, 0, sizeof(*r));
1749 if (who == RUSAGE_THREAD) {
1750 task_cputime_adjusted(current, &utime, &stime);
1751 accumulate_thread_rusage(p, r);
1752 maxrss = sig->maxrss;
1756 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
1760 case RUSAGE_CHILDREN:
1761 utime = sig->cutime;
1762 stime = sig->cstime;
1763 r->ru_nvcsw = sig->cnvcsw;
1764 r->ru_nivcsw = sig->cnivcsw;
1765 r->ru_minflt = sig->cmin_flt;
1766 r->ru_majflt = sig->cmaj_flt;
1767 r->ru_inblock = sig->cinblock;
1768 r->ru_oublock = sig->coublock;
1769 maxrss = sig->cmaxrss;
1771 if (who == RUSAGE_CHILDREN)
1776 r->ru_nvcsw += sig->nvcsw;
1777 r->ru_nivcsw += sig->nivcsw;
1778 r->ru_minflt += sig->min_flt;
1779 r->ru_majflt += sig->maj_flt;
1780 r->ru_inblock += sig->inblock;
1781 r->ru_oublock += sig->oublock;
1782 if (maxrss < sig->maxrss)
1783 maxrss = sig->maxrss;
1786 __for_each_thread(sig, t)
1787 accumulate_thread_rusage(t, r);
1796 if (need_seqretry(&sig->stats_lock, seq)) {
1800 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
1802 if (who == RUSAGE_CHILDREN)
1805 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1810 mm = get_task_mm(p);
1812 setmax_mm_hiwater_rss(&maxrss, mm);
1817 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1818 r->ru_utime = ns_to_kernel_old_timeval(utime);
1819 r->ru_stime = ns_to_kernel_old_timeval(stime);
1822 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1826 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1827 who != RUSAGE_THREAD)
1830 getrusage(current, who, &r);
1831 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1834 #ifdef CONFIG_COMPAT
1835 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1839 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1840 who != RUSAGE_THREAD)
1843 getrusage(current, who, &r);
1844 return put_compat_rusage(&r, ru);
1848 SYSCALL_DEFINE1(umask, int, mask)
1850 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1854 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1857 struct file *old_exe, *exe_file;
1858 struct inode *inode;
1865 inode = file_inode(exe.file);
1868 * Because the original mm->exe_file points to executable file, make
1869 * sure that this one is executable as well, to avoid breaking an
1873 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1876 err = inode_permission(inode, MAY_EXEC);
1881 * Forbid mm->exe_file change if old file still mapped.
1883 exe_file = get_mm_exe_file(mm);
1886 struct vm_area_struct *vma;
1889 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1892 if (path_equal(&vma->vm_file->f_path,
1897 mmap_read_unlock(mm);
1902 /* set the new file, lockless */
1904 old_exe = xchg(&mm->exe_file, exe.file);
1911 mmap_read_unlock(mm);
1917 * Check arithmetic relations of passed addresses.
1919 * WARNING: we don't require any capability here so be very careful
1920 * in what is allowed for modification from userspace.
1922 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1924 unsigned long mmap_max_addr = TASK_SIZE;
1925 int error = -EINVAL, i;
1927 static const unsigned char offsets[] = {
1928 offsetof(struct prctl_mm_map, start_code),
1929 offsetof(struct prctl_mm_map, end_code),
1930 offsetof(struct prctl_mm_map, start_data),
1931 offsetof(struct prctl_mm_map, end_data),
1932 offsetof(struct prctl_mm_map, start_brk),
1933 offsetof(struct prctl_mm_map, brk),
1934 offsetof(struct prctl_mm_map, start_stack),
1935 offsetof(struct prctl_mm_map, arg_start),
1936 offsetof(struct prctl_mm_map, arg_end),
1937 offsetof(struct prctl_mm_map, env_start),
1938 offsetof(struct prctl_mm_map, env_end),
1942 * Make sure the members are not somewhere outside
1943 * of allowed address space.
1945 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1946 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1948 if ((unsigned long)val >= mmap_max_addr ||
1949 (unsigned long)val < mmap_min_addr)
1954 * Make sure the pairs are ordered.
1956 #define __prctl_check_order(__m1, __op, __m2) \
1957 ((unsigned long)prctl_map->__m1 __op \
1958 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1959 error = __prctl_check_order(start_code, <, end_code);
1960 error |= __prctl_check_order(start_data,<=, end_data);
1961 error |= __prctl_check_order(start_brk, <=, brk);
1962 error |= __prctl_check_order(arg_start, <=, arg_end);
1963 error |= __prctl_check_order(env_start, <=, env_end);
1966 #undef __prctl_check_order
1971 * Neither we should allow to override limits if they set.
1973 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1974 prctl_map->start_brk, prctl_map->end_data,
1975 prctl_map->start_data))
1983 #ifdef CONFIG_CHECKPOINT_RESTORE
1984 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1986 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1987 unsigned long user_auxv[AT_VECTOR_SIZE];
1988 struct mm_struct *mm = current->mm;
1991 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1992 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1994 if (opt == PR_SET_MM_MAP_SIZE)
1995 return put_user((unsigned int)sizeof(prctl_map),
1996 (unsigned int __user *)addr);
1998 if (data_size != sizeof(prctl_map))
2001 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2004 error = validate_prctl_map_addr(&prctl_map);
2008 if (prctl_map.auxv_size) {
2010 * Someone is trying to cheat the auxv vector.
2012 if (!prctl_map.auxv ||
2013 prctl_map.auxv_size > sizeof(mm->saved_auxv))
2016 memset(user_auxv, 0, sizeof(user_auxv));
2017 if (copy_from_user(user_auxv,
2018 (const void __user *)prctl_map.auxv,
2019 prctl_map.auxv_size))
2022 /* Last entry must be AT_NULL as specification requires */
2023 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2024 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2027 if (prctl_map.exe_fd != (u32)-1) {
2029 * Check if the current user is checkpoint/restore capable.
2030 * At the time of this writing, it checks for CAP_SYS_ADMIN
2031 * or CAP_CHECKPOINT_RESTORE.
2032 * Note that a user with access to ptrace can masquerade an
2033 * arbitrary program as any executable, even setuid ones.
2034 * This may have implications in the tomoyo subsystem.
2036 if (!checkpoint_restore_ns_capable(current_user_ns()))
2039 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2045 * arg_lock protects concurent updates but we still need mmap_lock for
2046 * read to exclude races with sys_brk.
2051 * We don't validate if these members are pointing to
2052 * real present VMAs because application may have correspond
2053 * VMAs already unmapped and kernel uses these members for statistics
2054 * output in procfs mostly, except
2056 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2057 * for VMAs when updating these memvers so anything wrong written
2058 * here cause kernel to swear at userspace program but won't lead
2059 * to any problem in kernel itself
2062 spin_lock(&mm->arg_lock);
2063 mm->start_code = prctl_map.start_code;
2064 mm->end_code = prctl_map.end_code;
2065 mm->start_data = prctl_map.start_data;
2066 mm->end_data = prctl_map.end_data;
2067 mm->start_brk = prctl_map.start_brk;
2068 mm->brk = prctl_map.brk;
2069 mm->start_stack = prctl_map.start_stack;
2070 mm->arg_start = prctl_map.arg_start;
2071 mm->arg_end = prctl_map.arg_end;
2072 mm->env_start = prctl_map.env_start;
2073 mm->env_end = prctl_map.env_end;
2074 spin_unlock(&mm->arg_lock);
2077 * Note this update of @saved_auxv is lockless thus
2078 * if someone reads this member in procfs while we're
2079 * updating -- it may get partly updated results. It's
2080 * known and acceptable trade off: we leave it as is to
2081 * not introduce additional locks here making the kernel
2084 if (prctl_map.auxv_size)
2085 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2087 mmap_read_unlock(mm);
2090 #endif /* CONFIG_CHECKPOINT_RESTORE */
2092 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2096 * This doesn't move the auxiliary vector itself since it's pinned to
2097 * mm_struct, but it permits filling the vector with new values. It's
2098 * up to the caller to provide sane values here, otherwise userspace
2099 * tools which use this vector might be unhappy.
2101 unsigned long user_auxv[AT_VECTOR_SIZE];
2103 if (len > sizeof(user_auxv))
2106 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2109 /* Make sure the last entry is always AT_NULL */
2110 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2111 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2113 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2116 memcpy(mm->saved_auxv, user_auxv, len);
2117 task_unlock(current);
2122 static int prctl_set_mm(int opt, unsigned long addr,
2123 unsigned long arg4, unsigned long arg5)
2125 struct mm_struct *mm = current->mm;
2126 struct prctl_mm_map prctl_map = {
2131 struct vm_area_struct *vma;
2134 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2135 opt != PR_SET_MM_MAP &&
2136 opt != PR_SET_MM_MAP_SIZE)))
2139 #ifdef CONFIG_CHECKPOINT_RESTORE
2140 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2141 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2144 if (!capable(CAP_SYS_RESOURCE))
2147 if (opt == PR_SET_MM_EXE_FILE)
2148 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2150 if (opt == PR_SET_MM_AUXV)
2151 return prctl_set_auxv(mm, addr, arg4);
2153 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2159 * arg_lock protects concurent updates of arg boundaries, we need
2160 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2164 vma = find_vma(mm, addr);
2166 spin_lock(&mm->arg_lock);
2167 prctl_map.start_code = mm->start_code;
2168 prctl_map.end_code = mm->end_code;
2169 prctl_map.start_data = mm->start_data;
2170 prctl_map.end_data = mm->end_data;
2171 prctl_map.start_brk = mm->start_brk;
2172 prctl_map.brk = mm->brk;
2173 prctl_map.start_stack = mm->start_stack;
2174 prctl_map.arg_start = mm->arg_start;
2175 prctl_map.arg_end = mm->arg_end;
2176 prctl_map.env_start = mm->env_start;
2177 prctl_map.env_end = mm->env_end;
2180 case PR_SET_MM_START_CODE:
2181 prctl_map.start_code = addr;
2183 case PR_SET_MM_END_CODE:
2184 prctl_map.end_code = addr;
2186 case PR_SET_MM_START_DATA:
2187 prctl_map.start_data = addr;
2189 case PR_SET_MM_END_DATA:
2190 prctl_map.end_data = addr;
2192 case PR_SET_MM_START_STACK:
2193 prctl_map.start_stack = addr;
2195 case PR_SET_MM_START_BRK:
2196 prctl_map.start_brk = addr;
2199 prctl_map.brk = addr;
2201 case PR_SET_MM_ARG_START:
2202 prctl_map.arg_start = addr;
2204 case PR_SET_MM_ARG_END:
2205 prctl_map.arg_end = addr;
2207 case PR_SET_MM_ENV_START:
2208 prctl_map.env_start = addr;
2210 case PR_SET_MM_ENV_END:
2211 prctl_map.env_end = addr;
2217 error = validate_prctl_map_addr(&prctl_map);
2223 * If command line arguments and environment
2224 * are placed somewhere else on stack, we can
2225 * set them up here, ARG_START/END to setup
2226 * command line argumets and ENV_START/END
2229 case PR_SET_MM_START_STACK:
2230 case PR_SET_MM_ARG_START:
2231 case PR_SET_MM_ARG_END:
2232 case PR_SET_MM_ENV_START:
2233 case PR_SET_MM_ENV_END:
2240 mm->start_code = prctl_map.start_code;
2241 mm->end_code = prctl_map.end_code;
2242 mm->start_data = prctl_map.start_data;
2243 mm->end_data = prctl_map.end_data;
2244 mm->start_brk = prctl_map.start_brk;
2245 mm->brk = prctl_map.brk;
2246 mm->start_stack = prctl_map.start_stack;
2247 mm->arg_start = prctl_map.arg_start;
2248 mm->arg_end = prctl_map.arg_end;
2249 mm->env_start = prctl_map.env_start;
2250 mm->env_end = prctl_map.env_end;
2254 spin_unlock(&mm->arg_lock);
2255 mmap_read_unlock(mm);
2259 #ifdef CONFIG_CHECKPOINT_RESTORE
2260 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2262 return put_user(me->clear_child_tid, tid_addr);
2265 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2271 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2274 * If task has has_child_subreaper - all its decendants
2275 * already have these flag too and new decendants will
2276 * inherit it on fork, skip them.
2278 * If we've found child_reaper - skip descendants in
2279 * it's subtree as they will never get out pidns.
2281 if (p->signal->has_child_subreaper ||
2282 is_child_reaper(task_pid(p)))
2285 p->signal->has_child_subreaper = 1;
2289 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2294 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2300 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2302 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2303 unsigned long, arg4, unsigned long, arg5)
2305 struct task_struct *me = current;
2306 unsigned char comm[sizeof(me->comm)];
2309 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2310 if (error != -ENOSYS)
2315 case PR_SET_PDEATHSIG:
2316 if (!valid_signal(arg2)) {
2320 me->pdeath_signal = arg2;
2322 case PR_GET_PDEATHSIG:
2323 error = put_user(me->pdeath_signal, (int __user *)arg2);
2325 case PR_GET_DUMPABLE:
2326 error = get_dumpable(me->mm);
2328 case PR_SET_DUMPABLE:
2329 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2333 set_dumpable(me->mm, arg2);
2336 case PR_SET_UNALIGN:
2337 error = SET_UNALIGN_CTL(me, arg2);
2339 case PR_GET_UNALIGN:
2340 error = GET_UNALIGN_CTL(me, arg2);
2343 error = SET_FPEMU_CTL(me, arg2);
2346 error = GET_FPEMU_CTL(me, arg2);
2349 error = SET_FPEXC_CTL(me, arg2);
2352 error = GET_FPEXC_CTL(me, arg2);
2355 error = PR_TIMING_STATISTICAL;
2358 if (arg2 != PR_TIMING_STATISTICAL)
2362 comm[sizeof(me->comm) - 1] = 0;
2363 if (strncpy_from_user(comm, (char __user *)arg2,
2364 sizeof(me->comm) - 1) < 0)
2366 set_task_comm(me, comm);
2367 proc_comm_connector(me);
2370 get_task_comm(comm, me);
2371 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2375 error = GET_ENDIAN(me, arg2);
2378 error = SET_ENDIAN(me, arg2);
2380 case PR_GET_SECCOMP:
2381 error = prctl_get_seccomp();
2383 case PR_SET_SECCOMP:
2384 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2387 error = GET_TSC_CTL(arg2);
2390 error = SET_TSC_CTL(arg2);
2392 case PR_TASK_PERF_EVENTS_DISABLE:
2393 error = perf_event_task_disable();
2395 case PR_TASK_PERF_EVENTS_ENABLE:
2396 error = perf_event_task_enable();
2398 case PR_GET_TIMERSLACK:
2399 if (current->timer_slack_ns > ULONG_MAX)
2402 error = current->timer_slack_ns;
2404 case PR_SET_TIMERSLACK:
2406 current->timer_slack_ns =
2407 current->default_timer_slack_ns;
2409 current->timer_slack_ns = arg2;
2415 case PR_MCE_KILL_CLEAR:
2418 current->flags &= ~PF_MCE_PROCESS;
2420 case PR_MCE_KILL_SET:
2421 current->flags |= PF_MCE_PROCESS;
2422 if (arg3 == PR_MCE_KILL_EARLY)
2423 current->flags |= PF_MCE_EARLY;
2424 else if (arg3 == PR_MCE_KILL_LATE)
2425 current->flags &= ~PF_MCE_EARLY;
2426 else if (arg3 == PR_MCE_KILL_DEFAULT)
2428 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2436 case PR_MCE_KILL_GET:
2437 if (arg2 | arg3 | arg4 | arg5)
2439 if (current->flags & PF_MCE_PROCESS)
2440 error = (current->flags & PF_MCE_EARLY) ?
2441 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2443 error = PR_MCE_KILL_DEFAULT;
2446 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2448 case PR_GET_TID_ADDRESS:
2449 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2451 case PR_SET_CHILD_SUBREAPER:
2452 me->signal->is_child_subreaper = !!arg2;
2456 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2458 case PR_GET_CHILD_SUBREAPER:
2459 error = put_user(me->signal->is_child_subreaper,
2460 (int __user *)arg2);
2462 case PR_SET_NO_NEW_PRIVS:
2463 if (arg2 != 1 || arg3 || arg4 || arg5)
2466 task_set_no_new_privs(current);
2468 case PR_GET_NO_NEW_PRIVS:
2469 if (arg2 || arg3 || arg4 || arg5)
2471 return task_no_new_privs(current) ? 1 : 0;
2472 case PR_GET_THP_DISABLE:
2473 if (arg2 || arg3 || arg4 || arg5)
2475 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2477 case PR_SET_THP_DISABLE:
2478 if (arg3 || arg4 || arg5)
2480 if (mmap_write_lock_killable(me->mm))
2483 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2485 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2486 mmap_write_unlock(me->mm);
2488 case PR_MPX_ENABLE_MANAGEMENT:
2489 case PR_MPX_DISABLE_MANAGEMENT:
2490 /* No longer implemented: */
2492 case PR_SET_FP_MODE:
2493 error = SET_FP_MODE(me, arg2);
2495 case PR_GET_FP_MODE:
2496 error = GET_FP_MODE(me);
2499 error = SVE_SET_VL(arg2);
2502 error = SVE_GET_VL();
2504 case PR_GET_SPECULATION_CTRL:
2505 if (arg3 || arg4 || arg5)
2507 error = arch_prctl_spec_ctrl_get(me, arg2);
2509 case PR_SET_SPECULATION_CTRL:
2512 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2514 case PR_PAC_RESET_KEYS:
2515 if (arg3 || arg4 || arg5)
2517 error = PAC_RESET_KEYS(me, arg2);
2519 case PR_SET_TAGGED_ADDR_CTRL:
2520 if (arg3 || arg4 || arg5)
2522 error = SET_TAGGED_ADDR_CTRL(arg2);
2524 case PR_GET_TAGGED_ADDR_CTRL:
2525 if (arg2 || arg3 || arg4 || arg5)
2527 error = GET_TAGGED_ADDR_CTRL();
2529 case PR_SET_IO_FLUSHER:
2530 if (!capable(CAP_SYS_RESOURCE))
2533 if (arg3 || arg4 || arg5)
2537 current->flags |= PR_IO_FLUSHER;
2539 current->flags &= ~PR_IO_FLUSHER;
2543 case PR_GET_IO_FLUSHER:
2544 if (!capable(CAP_SYS_RESOURCE))
2547 if (arg2 || arg3 || arg4 || arg5)
2550 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2559 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2560 struct getcpu_cache __user *, unused)
2563 int cpu = raw_smp_processor_id();
2566 err |= put_user(cpu, cpup);
2568 err |= put_user(cpu_to_node(cpu), nodep);
2569 return err ? -EFAULT : 0;
2573 * do_sysinfo - fill in sysinfo struct
2574 * @info: pointer to buffer to fill
2576 static int do_sysinfo(struct sysinfo *info)
2578 unsigned long mem_total, sav_total;
2579 unsigned int mem_unit, bitcount;
2580 struct timespec64 tp;
2582 memset(info, 0, sizeof(struct sysinfo));
2584 ktime_get_boottime_ts64(&tp);
2585 timens_add_boottime(&tp);
2586 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2588 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2590 info->procs = nr_threads;
2596 * If the sum of all the available memory (i.e. ram + swap)
2597 * is less than can be stored in a 32 bit unsigned long then
2598 * we can be binary compatible with 2.2.x kernels. If not,
2599 * well, in that case 2.2.x was broken anyways...
2601 * -Erik Andersen <andersee@debian.org>
2604 mem_total = info->totalram + info->totalswap;
2605 if (mem_total < info->totalram || mem_total < info->totalswap)
2608 mem_unit = info->mem_unit;
2609 while (mem_unit > 1) {
2612 sav_total = mem_total;
2614 if (mem_total < sav_total)
2619 * If mem_total did not overflow, multiply all memory values by
2620 * info->mem_unit and set it to 1. This leaves things compatible
2621 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2626 info->totalram <<= bitcount;
2627 info->freeram <<= bitcount;
2628 info->sharedram <<= bitcount;
2629 info->bufferram <<= bitcount;
2630 info->totalswap <<= bitcount;
2631 info->freeswap <<= bitcount;
2632 info->totalhigh <<= bitcount;
2633 info->freehigh <<= bitcount;
2639 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2645 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2651 #ifdef CONFIG_COMPAT
2652 struct compat_sysinfo {
2666 char _f[20-2*sizeof(u32)-sizeof(int)];
2669 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2672 struct compat_sysinfo s_32;
2676 /* Check to see if any memory value is too large for 32-bit and scale
2679 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2682 while (s.mem_unit < PAGE_SIZE) {
2687 s.totalram >>= bitcount;
2688 s.freeram >>= bitcount;
2689 s.sharedram >>= bitcount;
2690 s.bufferram >>= bitcount;
2691 s.totalswap >>= bitcount;
2692 s.freeswap >>= bitcount;
2693 s.totalhigh >>= bitcount;
2694 s.freehigh >>= bitcount;
2697 memset(&s_32, 0, sizeof(s_32));
2698 s_32.uptime = s.uptime;
2699 s_32.loads[0] = s.loads[0];
2700 s_32.loads[1] = s.loads[1];
2701 s_32.loads[2] = s.loads[2];
2702 s_32.totalram = s.totalram;
2703 s_32.freeram = s.freeram;
2704 s_32.sharedram = s.sharedram;
2705 s_32.bufferram = s.bufferram;
2706 s_32.totalswap = s.totalswap;
2707 s_32.freeswap = s.freeswap;
2708 s_32.procs = s.procs;
2709 s_32.totalhigh = s.totalhigh;
2710 s_32.freehigh = s.freehigh;
2711 s_32.mem_unit = s.mem_unit;
2712 if (copy_to_user(info, &s_32, sizeof(s_32)))
2716 #endif /* CONFIG_COMPAT */