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(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(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(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;
638 kruid = make_kuid(ns, ruid);
639 keuid = make_kuid(ns, euid);
640 ksuid = make_kuid(ns, suid);
642 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
645 if ((euid != (uid_t) -1) && !uid_valid(keuid))
648 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
651 new = prepare_creds();
655 old = current_cred();
658 if (!ns_capable_setid(old->user_ns, CAP_SETUID)) {
659 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
660 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
662 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
663 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
665 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
666 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
670 if (ruid != (uid_t) -1) {
672 if (!uid_eq(kruid, old->uid)) {
673 retval = set_user(new);
678 if (euid != (uid_t) -1)
680 if (suid != (uid_t) -1)
682 new->fsuid = new->euid;
684 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
688 return commit_creds(new);
695 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
697 return __sys_setresuid(ruid, euid, suid);
700 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
702 const struct cred *cred = current_cred();
704 uid_t ruid, euid, suid;
706 ruid = from_kuid_munged(cred->user_ns, cred->uid);
707 euid = from_kuid_munged(cred->user_ns, cred->euid);
708 suid = from_kuid_munged(cred->user_ns, cred->suid);
710 retval = put_user(ruid, ruidp);
712 retval = put_user(euid, euidp);
714 return put_user(suid, suidp);
720 * Same as above, but for rgid, egid, sgid.
722 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
724 struct user_namespace *ns = current_user_ns();
725 const struct cred *old;
728 kgid_t krgid, kegid, ksgid;
730 krgid = make_kgid(ns, rgid);
731 kegid = make_kgid(ns, egid);
732 ksgid = make_kgid(ns, sgid);
734 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
736 if ((egid != (gid_t) -1) && !gid_valid(kegid))
738 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
741 new = prepare_creds();
744 old = current_cred();
747 if (!ns_capable(old->user_ns, CAP_SETGID)) {
748 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
749 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
751 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
752 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
754 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
755 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
759 if (rgid != (gid_t) -1)
761 if (egid != (gid_t) -1)
763 if (sgid != (gid_t) -1)
765 new->fsgid = new->egid;
767 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
771 return commit_creds(new);
778 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
780 return __sys_setresgid(rgid, egid, sgid);
783 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
785 const struct cred *cred = current_cred();
787 gid_t rgid, egid, sgid;
789 rgid = from_kgid_munged(cred->user_ns, cred->gid);
790 egid = from_kgid_munged(cred->user_ns, cred->egid);
791 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
793 retval = put_user(rgid, rgidp);
795 retval = put_user(egid, egidp);
797 retval = put_user(sgid, sgidp);
805 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
806 * is used for "access()" and for the NFS daemon (letting nfsd stay at
807 * whatever uid it wants to). It normally shadows "euid", except when
808 * explicitly set by setfsuid() or for access..
810 long __sys_setfsuid(uid_t uid)
812 const struct cred *old;
817 old = current_cred();
818 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
820 kuid = make_kuid(old->user_ns, uid);
821 if (!uid_valid(kuid))
824 new = prepare_creds();
828 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
829 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
830 ns_capable_setid(old->user_ns, CAP_SETUID)) {
831 if (!uid_eq(kuid, old->fsuid)) {
833 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
846 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
848 return __sys_setfsuid(uid);
852 * Samma på svenska..
854 long __sys_setfsgid(gid_t gid)
856 const struct cred *old;
861 old = current_cred();
862 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
864 kgid = make_kgid(old->user_ns, gid);
865 if (!gid_valid(kgid))
868 new = prepare_creds();
872 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
873 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
874 ns_capable(old->user_ns, CAP_SETGID)) {
875 if (!gid_eq(kgid, old->fsgid)) {
877 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
890 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
892 return __sys_setfsgid(gid);
894 #endif /* CONFIG_MULTIUSER */
897 * sys_getpid - return the thread group id of the current process
899 * Note, despite the name, this returns the tgid not the pid. The tgid and
900 * the pid are identical unless CLONE_THREAD was specified on clone() in
901 * which case the tgid is the same in all threads of the same group.
903 * This is SMP safe as current->tgid does not change.
905 SYSCALL_DEFINE0(getpid)
907 return task_tgid_vnr(current);
910 /* Thread ID - the internal kernel "pid" */
911 SYSCALL_DEFINE0(gettid)
913 return task_pid_vnr(current);
917 * Accessing ->real_parent is not SMP-safe, it could
918 * change from under us. However, we can use a stale
919 * value of ->real_parent under rcu_read_lock(), see
920 * release_task()->call_rcu(delayed_put_task_struct).
922 SYSCALL_DEFINE0(getppid)
927 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
933 SYSCALL_DEFINE0(getuid)
935 /* Only we change this so SMP safe */
936 return from_kuid_munged(current_user_ns(), current_uid());
939 SYSCALL_DEFINE0(geteuid)
941 /* Only we change this so SMP safe */
942 return from_kuid_munged(current_user_ns(), current_euid());
945 SYSCALL_DEFINE0(getgid)
947 /* Only we change this so SMP safe */
948 return from_kgid_munged(current_user_ns(), current_gid());
951 SYSCALL_DEFINE0(getegid)
953 /* Only we change this so SMP safe */
954 return from_kgid_munged(current_user_ns(), current_egid());
957 static void do_sys_times(struct tms *tms)
959 u64 tgutime, tgstime, cutime, cstime;
961 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
962 cutime = current->signal->cutime;
963 cstime = current->signal->cstime;
964 tms->tms_utime = nsec_to_clock_t(tgutime);
965 tms->tms_stime = nsec_to_clock_t(tgstime);
966 tms->tms_cutime = nsec_to_clock_t(cutime);
967 tms->tms_cstime = nsec_to_clock_t(cstime);
970 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
976 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
979 force_successful_syscall_return();
980 return (long) jiffies_64_to_clock_t(get_jiffies_64());
984 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
986 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
989 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
993 struct compat_tms tmp;
996 /* Convert our struct tms to the compat version. */
997 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
998 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
999 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1000 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1001 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1004 force_successful_syscall_return();
1005 return compat_jiffies_to_clock_t(jiffies);
1010 * This needs some heavy checking ...
1011 * I just haven't the stomach for it. I also don't fully
1012 * understand sessions/pgrp etc. Let somebody who does explain it.
1014 * OK, I think I have the protection semantics right.... this is really
1015 * only important on a multi-user system anyway, to make sure one user
1016 * can't send a signal to a process owned by another. -TYT, 12/12/91
1018 * !PF_FORKNOEXEC check to conform completely to POSIX.
1020 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1022 struct task_struct *p;
1023 struct task_struct *group_leader = current->group_leader;
1028 pid = task_pid_vnr(group_leader);
1035 /* From this point forward we keep holding onto the tasklist lock
1036 * so that our parent does not change from under us. -DaveM
1038 write_lock_irq(&tasklist_lock);
1041 p = find_task_by_vpid(pid);
1046 if (!thread_group_leader(p))
1049 if (same_thread_group(p->real_parent, group_leader)) {
1051 if (task_session(p) != task_session(group_leader))
1054 if (!(p->flags & PF_FORKNOEXEC))
1058 if (p != group_leader)
1063 if (p->signal->leader)
1068 struct task_struct *g;
1070 pgrp = find_vpid(pgid);
1071 g = pid_task(pgrp, PIDTYPE_PGID);
1072 if (!g || task_session(g) != task_session(group_leader))
1076 err = security_task_setpgid(p, pgid);
1080 if (task_pgrp(p) != pgrp)
1081 change_pid(p, PIDTYPE_PGID, pgrp);
1085 /* All paths lead to here, thus we are safe. -DaveM */
1086 write_unlock_irq(&tasklist_lock);
1091 static int do_getpgid(pid_t pid)
1093 struct task_struct *p;
1099 grp = task_pgrp(current);
1102 p = find_task_by_vpid(pid);
1109 retval = security_task_getpgid(p);
1113 retval = pid_vnr(grp);
1119 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1121 return do_getpgid(pid);
1124 #ifdef __ARCH_WANT_SYS_GETPGRP
1126 SYSCALL_DEFINE0(getpgrp)
1128 return do_getpgid(0);
1133 SYSCALL_DEFINE1(getsid, pid_t, pid)
1135 struct task_struct *p;
1141 sid = task_session(current);
1144 p = find_task_by_vpid(pid);
1147 sid = task_session(p);
1151 retval = security_task_getsid(p);
1155 retval = pid_vnr(sid);
1161 static void set_special_pids(struct pid *pid)
1163 struct task_struct *curr = current->group_leader;
1165 if (task_session(curr) != pid)
1166 change_pid(curr, PIDTYPE_SID, pid);
1168 if (task_pgrp(curr) != pid)
1169 change_pid(curr, PIDTYPE_PGID, pid);
1172 int ksys_setsid(void)
1174 struct task_struct *group_leader = current->group_leader;
1175 struct pid *sid = task_pid(group_leader);
1176 pid_t session = pid_vnr(sid);
1179 write_lock_irq(&tasklist_lock);
1180 /* Fail if I am already a session leader */
1181 if (group_leader->signal->leader)
1184 /* Fail if a process group id already exists that equals the
1185 * proposed session id.
1187 if (pid_task(sid, PIDTYPE_PGID))
1190 group_leader->signal->leader = 1;
1191 set_special_pids(sid);
1193 proc_clear_tty(group_leader);
1197 write_unlock_irq(&tasklist_lock);
1199 proc_sid_connector(group_leader);
1200 sched_autogroup_create_attach(group_leader);
1205 SYSCALL_DEFINE0(setsid)
1207 return ksys_setsid();
1210 DECLARE_RWSEM(uts_sem);
1212 #ifdef COMPAT_UTS_MACHINE
1213 #define override_architecture(name) \
1214 (personality(current->personality) == PER_LINUX32 && \
1215 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1216 sizeof(COMPAT_UTS_MACHINE)))
1218 #define override_architecture(name) 0
1222 * Work around broken programs that cannot handle "Linux 3.0".
1223 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1224 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1227 static int override_release(char __user *release, size_t len)
1231 if (current->personality & UNAME26) {
1232 const char *rest = UTS_RELEASE;
1233 char buf[65] = { 0 };
1239 if (*rest == '.' && ++ndots >= 3)
1241 if (!isdigit(*rest) && *rest != '.')
1245 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1246 copy = clamp_t(size_t, len, 1, sizeof(buf));
1247 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1248 ret = copy_to_user(release, buf, copy + 1);
1253 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1255 struct new_utsname tmp;
1257 down_read(&uts_sem);
1258 memcpy(&tmp, utsname(), sizeof(tmp));
1260 if (copy_to_user(name, &tmp, sizeof(tmp)))
1263 if (override_release(name->release, sizeof(name->release)))
1265 if (override_architecture(name))
1270 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1274 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1276 struct old_utsname tmp;
1281 down_read(&uts_sem);
1282 memcpy(&tmp, utsname(), sizeof(tmp));
1284 if (copy_to_user(name, &tmp, sizeof(tmp)))
1287 if (override_release(name->release, sizeof(name->release)))
1289 if (override_architecture(name))
1294 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1296 struct oldold_utsname tmp;
1301 memset(&tmp, 0, sizeof(tmp));
1303 down_read(&uts_sem);
1304 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1305 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1306 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1307 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1308 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1310 if (copy_to_user(name, &tmp, sizeof(tmp)))
1313 if (override_architecture(name))
1315 if (override_release(name->release, sizeof(name->release)))
1321 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1324 char tmp[__NEW_UTS_LEN];
1326 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1329 if (len < 0 || len > __NEW_UTS_LEN)
1332 if (!copy_from_user(tmp, name, len)) {
1333 struct new_utsname *u;
1335 down_write(&uts_sem);
1337 memcpy(u->nodename, tmp, len);
1338 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1340 uts_proc_notify(UTS_PROC_HOSTNAME);
1346 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1348 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1351 struct new_utsname *u;
1352 char tmp[__NEW_UTS_LEN + 1];
1356 down_read(&uts_sem);
1358 i = 1 + strlen(u->nodename);
1361 memcpy(tmp, u->nodename, i);
1363 if (copy_to_user(name, tmp, i))
1371 * Only setdomainname; getdomainname can be implemented by calling
1374 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1377 char tmp[__NEW_UTS_LEN];
1379 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1381 if (len < 0 || len > __NEW_UTS_LEN)
1385 if (!copy_from_user(tmp, name, len)) {
1386 struct new_utsname *u;
1388 down_write(&uts_sem);
1390 memcpy(u->domainname, tmp, len);
1391 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1393 uts_proc_notify(UTS_PROC_DOMAINNAME);
1399 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1401 struct rlimit value;
1404 ret = do_prlimit(current, resource, NULL, &value);
1406 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1411 #ifdef CONFIG_COMPAT
1413 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1414 struct compat_rlimit __user *, rlim)
1417 struct compat_rlimit r32;
1419 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1422 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1423 r.rlim_cur = RLIM_INFINITY;
1425 r.rlim_cur = r32.rlim_cur;
1426 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1427 r.rlim_max = RLIM_INFINITY;
1429 r.rlim_max = r32.rlim_max;
1430 return do_prlimit(current, resource, &r, NULL);
1433 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1434 struct compat_rlimit __user *, rlim)
1439 ret = do_prlimit(current, resource, NULL, &r);
1441 struct compat_rlimit r32;
1442 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1443 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1445 r32.rlim_cur = r.rlim_cur;
1446 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1447 r32.rlim_max = COMPAT_RLIM_INFINITY;
1449 r32.rlim_max = r.rlim_max;
1451 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1459 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1462 * Back compatibility for getrlimit. Needed for some apps.
1464 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1465 struct rlimit __user *, rlim)
1468 if (resource >= RLIM_NLIMITS)
1471 resource = array_index_nospec(resource, RLIM_NLIMITS);
1472 task_lock(current->group_leader);
1473 x = current->signal->rlim[resource];
1474 task_unlock(current->group_leader);
1475 if (x.rlim_cur > 0x7FFFFFFF)
1476 x.rlim_cur = 0x7FFFFFFF;
1477 if (x.rlim_max > 0x7FFFFFFF)
1478 x.rlim_max = 0x7FFFFFFF;
1479 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1482 #ifdef CONFIG_COMPAT
1483 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1484 struct compat_rlimit __user *, rlim)
1488 if (resource >= RLIM_NLIMITS)
1491 resource = array_index_nospec(resource, RLIM_NLIMITS);
1492 task_lock(current->group_leader);
1493 r = current->signal->rlim[resource];
1494 task_unlock(current->group_leader);
1495 if (r.rlim_cur > 0x7FFFFFFF)
1496 r.rlim_cur = 0x7FFFFFFF;
1497 if (r.rlim_max > 0x7FFFFFFF)
1498 r.rlim_max = 0x7FFFFFFF;
1500 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1501 put_user(r.rlim_max, &rlim->rlim_max))
1509 static inline bool rlim64_is_infinity(__u64 rlim64)
1511 #if BITS_PER_LONG < 64
1512 return rlim64 >= ULONG_MAX;
1514 return rlim64 == RLIM64_INFINITY;
1518 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1520 if (rlim->rlim_cur == RLIM_INFINITY)
1521 rlim64->rlim_cur = RLIM64_INFINITY;
1523 rlim64->rlim_cur = rlim->rlim_cur;
1524 if (rlim->rlim_max == RLIM_INFINITY)
1525 rlim64->rlim_max = RLIM64_INFINITY;
1527 rlim64->rlim_max = rlim->rlim_max;
1530 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1532 if (rlim64_is_infinity(rlim64->rlim_cur))
1533 rlim->rlim_cur = RLIM_INFINITY;
1535 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1536 if (rlim64_is_infinity(rlim64->rlim_max))
1537 rlim->rlim_max = RLIM_INFINITY;
1539 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1542 /* make sure you are allowed to change @tsk limits before calling this */
1543 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1544 struct rlimit *new_rlim, struct rlimit *old_rlim)
1546 struct rlimit *rlim;
1549 if (resource >= RLIM_NLIMITS)
1552 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1554 if (resource == RLIMIT_NOFILE &&
1555 new_rlim->rlim_max > sysctl_nr_open)
1559 /* protect tsk->signal and tsk->sighand from disappearing */
1560 read_lock(&tasklist_lock);
1561 if (!tsk->sighand) {
1566 rlim = tsk->signal->rlim + resource;
1567 task_lock(tsk->group_leader);
1569 /* Keep the capable check against init_user_ns until
1570 cgroups can contain all limits */
1571 if (new_rlim->rlim_max > rlim->rlim_max &&
1572 !capable(CAP_SYS_RESOURCE))
1575 retval = security_task_setrlimit(tsk, resource, new_rlim);
1583 task_unlock(tsk->group_leader);
1586 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1587 * infite. In case of RLIM_INFINITY the posix CPU timer code
1588 * ignores the rlimit.
1590 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1591 new_rlim->rlim_cur != RLIM_INFINITY &&
1592 IS_ENABLED(CONFIG_POSIX_TIMERS))
1593 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1595 read_unlock(&tasklist_lock);
1599 /* rcu lock must be held */
1600 static int check_prlimit_permission(struct task_struct *task,
1603 const struct cred *cred = current_cred(), *tcred;
1606 if (current == task)
1609 tcred = __task_cred(task);
1610 id_match = (uid_eq(cred->uid, tcred->euid) &&
1611 uid_eq(cred->uid, tcred->suid) &&
1612 uid_eq(cred->uid, tcred->uid) &&
1613 gid_eq(cred->gid, tcred->egid) &&
1614 gid_eq(cred->gid, tcred->sgid) &&
1615 gid_eq(cred->gid, tcred->gid));
1616 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1619 return security_task_prlimit(cred, tcred, flags);
1622 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1623 const struct rlimit64 __user *, new_rlim,
1624 struct rlimit64 __user *, old_rlim)
1626 struct rlimit64 old64, new64;
1627 struct rlimit old, new;
1628 struct task_struct *tsk;
1629 unsigned int checkflags = 0;
1633 checkflags |= LSM_PRLIMIT_READ;
1636 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1638 rlim64_to_rlim(&new64, &new);
1639 checkflags |= LSM_PRLIMIT_WRITE;
1643 tsk = pid ? find_task_by_vpid(pid) : current;
1648 ret = check_prlimit_permission(tsk, checkflags);
1653 get_task_struct(tsk);
1656 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1657 old_rlim ? &old : NULL);
1659 if (!ret && old_rlim) {
1660 rlim_to_rlim64(&old, &old64);
1661 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1665 put_task_struct(tsk);
1669 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1671 struct rlimit new_rlim;
1673 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1675 return do_prlimit(current, resource, &new_rlim, NULL);
1679 * It would make sense to put struct rusage in the task_struct,
1680 * except that would make the task_struct be *really big*. After
1681 * task_struct gets moved into malloc'ed memory, it would
1682 * make sense to do this. It will make moving the rest of the information
1683 * a lot simpler! (Which we're not doing right now because we're not
1684 * measuring them yet).
1686 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1687 * races with threads incrementing their own counters. But since word
1688 * reads are atomic, we either get new values or old values and we don't
1689 * care which for the sums. We always take the siglock to protect reading
1690 * the c* fields from p->signal from races with exit.c updating those
1691 * fields when reaping, so a sample either gets all the additions of a
1692 * given child after it's reaped, or none so this sample is before reaping.
1695 * We need to take the siglock for CHILDEREN, SELF and BOTH
1696 * for the cases current multithreaded, non-current single threaded
1697 * non-current multithreaded. Thread traversal is now safe with
1699 * Strictly speaking, we donot need to take the siglock if we are current and
1700 * single threaded, as no one else can take our signal_struct away, no one
1701 * else can reap the children to update signal->c* counters, and no one else
1702 * can race with the signal-> fields. If we do not take any lock, the
1703 * signal-> fields could be read out of order while another thread was just
1704 * exiting. So we should place a read memory barrier when we avoid the lock.
1705 * On the writer side, write memory barrier is implied in __exit_signal
1706 * as __exit_signal releases the siglock spinlock after updating the signal->
1707 * fields. But we don't do this yet to keep things simple.
1711 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1713 r->ru_nvcsw += t->nvcsw;
1714 r->ru_nivcsw += t->nivcsw;
1715 r->ru_minflt += t->min_flt;
1716 r->ru_majflt += t->maj_flt;
1717 r->ru_inblock += task_io_get_inblock(t);
1718 r->ru_oublock += task_io_get_oublock(t);
1721 void getrusage(struct task_struct *p, int who, struct rusage *r)
1723 struct task_struct *t;
1724 unsigned long flags;
1725 u64 tgutime, tgstime, utime, stime;
1726 unsigned long maxrss = 0;
1728 memset((char *)r, 0, sizeof (*r));
1731 if (who == RUSAGE_THREAD) {
1732 task_cputime_adjusted(current, &utime, &stime);
1733 accumulate_thread_rusage(p, r);
1734 maxrss = p->signal->maxrss;
1738 if (!lock_task_sighand(p, &flags))
1743 case RUSAGE_CHILDREN:
1744 utime = p->signal->cutime;
1745 stime = p->signal->cstime;
1746 r->ru_nvcsw = p->signal->cnvcsw;
1747 r->ru_nivcsw = p->signal->cnivcsw;
1748 r->ru_minflt = p->signal->cmin_flt;
1749 r->ru_majflt = p->signal->cmaj_flt;
1750 r->ru_inblock = p->signal->cinblock;
1751 r->ru_oublock = p->signal->coublock;
1752 maxrss = p->signal->cmaxrss;
1754 if (who == RUSAGE_CHILDREN)
1759 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1762 r->ru_nvcsw += p->signal->nvcsw;
1763 r->ru_nivcsw += p->signal->nivcsw;
1764 r->ru_minflt += p->signal->min_flt;
1765 r->ru_majflt += p->signal->maj_flt;
1766 r->ru_inblock += p->signal->inblock;
1767 r->ru_oublock += p->signal->oublock;
1768 if (maxrss < p->signal->maxrss)
1769 maxrss = p->signal->maxrss;
1772 accumulate_thread_rusage(t, r);
1773 } while_each_thread(p, t);
1779 unlock_task_sighand(p, &flags);
1782 r->ru_utime = ns_to_kernel_old_timeval(utime);
1783 r->ru_stime = ns_to_kernel_old_timeval(stime);
1785 if (who != RUSAGE_CHILDREN) {
1786 struct mm_struct *mm = get_task_mm(p);
1789 setmax_mm_hiwater_rss(&maxrss, mm);
1793 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1796 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1800 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1801 who != RUSAGE_THREAD)
1804 getrusage(current, who, &r);
1805 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1808 #ifdef CONFIG_COMPAT
1809 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1813 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1814 who != RUSAGE_THREAD)
1817 getrusage(current, who, &r);
1818 return put_compat_rusage(&r, ru);
1822 SYSCALL_DEFINE1(umask, int, mask)
1824 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1828 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1831 struct file *old_exe, *exe_file;
1832 struct inode *inode;
1839 inode = file_inode(exe.file);
1842 * Because the original mm->exe_file points to executable file, make
1843 * sure that this one is executable as well, to avoid breaking an
1847 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1850 err = inode_permission(inode, MAY_EXEC);
1855 * Forbid mm->exe_file change if old file still mapped.
1857 exe_file = get_mm_exe_file(mm);
1860 struct vm_area_struct *vma;
1863 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1866 if (path_equal(&vma->vm_file->f_path,
1871 mmap_read_unlock(mm);
1876 /* set the new file, lockless */
1878 old_exe = xchg(&mm->exe_file, exe.file);
1885 mmap_read_unlock(mm);
1891 * Check arithmetic relations of passed addresses.
1893 * WARNING: we don't require any capability here so be very careful
1894 * in what is allowed for modification from userspace.
1896 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1898 unsigned long mmap_max_addr = TASK_SIZE;
1899 int error = -EINVAL, i;
1901 static const unsigned char offsets[] = {
1902 offsetof(struct prctl_mm_map, start_code),
1903 offsetof(struct prctl_mm_map, end_code),
1904 offsetof(struct prctl_mm_map, start_data),
1905 offsetof(struct prctl_mm_map, end_data),
1906 offsetof(struct prctl_mm_map, start_brk),
1907 offsetof(struct prctl_mm_map, brk),
1908 offsetof(struct prctl_mm_map, start_stack),
1909 offsetof(struct prctl_mm_map, arg_start),
1910 offsetof(struct prctl_mm_map, arg_end),
1911 offsetof(struct prctl_mm_map, env_start),
1912 offsetof(struct prctl_mm_map, env_end),
1916 * Make sure the members are not somewhere outside
1917 * of allowed address space.
1919 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1920 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1922 if ((unsigned long)val >= mmap_max_addr ||
1923 (unsigned long)val < mmap_min_addr)
1928 * Make sure the pairs are ordered.
1930 #define __prctl_check_order(__m1, __op, __m2) \
1931 ((unsigned long)prctl_map->__m1 __op \
1932 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1933 error = __prctl_check_order(start_code, <, end_code);
1934 error |= __prctl_check_order(start_data,<=, end_data);
1935 error |= __prctl_check_order(start_brk, <=, brk);
1936 error |= __prctl_check_order(arg_start, <=, arg_end);
1937 error |= __prctl_check_order(env_start, <=, env_end);
1940 #undef __prctl_check_order
1945 * @brk should be after @end_data in traditional maps.
1947 if (prctl_map->start_brk <= prctl_map->end_data ||
1948 prctl_map->brk <= prctl_map->end_data)
1952 * Neither we should allow to override limits if they set.
1954 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1955 prctl_map->start_brk, prctl_map->end_data,
1956 prctl_map->start_data))
1964 #ifdef CONFIG_CHECKPOINT_RESTORE
1965 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1967 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1968 unsigned long user_auxv[AT_VECTOR_SIZE];
1969 struct mm_struct *mm = current->mm;
1972 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1973 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1975 if (opt == PR_SET_MM_MAP_SIZE)
1976 return put_user((unsigned int)sizeof(prctl_map),
1977 (unsigned int __user *)addr);
1979 if (data_size != sizeof(prctl_map))
1982 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1985 error = validate_prctl_map_addr(&prctl_map);
1989 if (prctl_map.auxv_size) {
1991 * Someone is trying to cheat the auxv vector.
1993 if (!prctl_map.auxv ||
1994 prctl_map.auxv_size > sizeof(mm->saved_auxv))
1997 memset(user_auxv, 0, sizeof(user_auxv));
1998 if (copy_from_user(user_auxv,
1999 (const void __user *)prctl_map.auxv,
2000 prctl_map.auxv_size))
2003 /* Last entry must be AT_NULL as specification requires */
2004 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2005 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2008 if (prctl_map.exe_fd != (u32)-1) {
2010 * Check if the current user is checkpoint/restore capable.
2011 * At the time of this writing, it checks for CAP_SYS_ADMIN
2012 * or CAP_CHECKPOINT_RESTORE.
2013 * Note that a user with access to ptrace can masquerade an
2014 * arbitrary program as any executable, even setuid ones.
2015 * This may have implications in the tomoyo subsystem.
2017 if (!checkpoint_restore_ns_capable(current_user_ns()))
2020 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2026 * arg_lock protects concurent updates but we still need mmap_lock for
2027 * read to exclude races with sys_brk.
2032 * We don't validate if these members are pointing to
2033 * real present VMAs because application may have correspond
2034 * VMAs already unmapped and kernel uses these members for statistics
2035 * output in procfs mostly, except
2037 * - @start_brk/@brk which are used in do_brk but kernel lookups
2038 * for VMAs when updating these memvers so anything wrong written
2039 * here cause kernel to swear at userspace program but won't lead
2040 * to any problem in kernel itself
2043 spin_lock(&mm->arg_lock);
2044 mm->start_code = prctl_map.start_code;
2045 mm->end_code = prctl_map.end_code;
2046 mm->start_data = prctl_map.start_data;
2047 mm->end_data = prctl_map.end_data;
2048 mm->start_brk = prctl_map.start_brk;
2049 mm->brk = prctl_map.brk;
2050 mm->start_stack = prctl_map.start_stack;
2051 mm->arg_start = prctl_map.arg_start;
2052 mm->arg_end = prctl_map.arg_end;
2053 mm->env_start = prctl_map.env_start;
2054 mm->env_end = prctl_map.env_end;
2055 spin_unlock(&mm->arg_lock);
2058 * Note this update of @saved_auxv is lockless thus
2059 * if someone reads this member in procfs while we're
2060 * updating -- it may get partly updated results. It's
2061 * known and acceptable trade off: we leave it as is to
2062 * not introduce additional locks here making the kernel
2065 if (prctl_map.auxv_size)
2066 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2068 mmap_read_unlock(mm);
2071 #endif /* CONFIG_CHECKPOINT_RESTORE */
2073 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2077 * This doesn't move the auxiliary vector itself since it's pinned to
2078 * mm_struct, but it permits filling the vector with new values. It's
2079 * up to the caller to provide sane values here, otherwise userspace
2080 * tools which use this vector might be unhappy.
2082 unsigned long user_auxv[AT_VECTOR_SIZE];
2084 if (len > sizeof(user_auxv))
2087 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2090 /* Make sure the last entry is always AT_NULL */
2091 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2092 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2094 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2097 memcpy(mm->saved_auxv, user_auxv, len);
2098 task_unlock(current);
2103 static int prctl_set_mm(int opt, unsigned long addr,
2104 unsigned long arg4, unsigned long arg5)
2106 struct mm_struct *mm = current->mm;
2107 struct prctl_mm_map prctl_map = {
2112 struct vm_area_struct *vma;
2115 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2116 opt != PR_SET_MM_MAP &&
2117 opt != PR_SET_MM_MAP_SIZE)))
2120 #ifdef CONFIG_CHECKPOINT_RESTORE
2121 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2122 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2125 if (!capable(CAP_SYS_RESOURCE))
2128 if (opt == PR_SET_MM_EXE_FILE)
2129 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2131 if (opt == PR_SET_MM_AUXV)
2132 return prctl_set_auxv(mm, addr, arg4);
2134 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2140 * arg_lock protects concurent updates of arg boundaries, we need
2141 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2145 vma = find_vma(mm, addr);
2147 spin_lock(&mm->arg_lock);
2148 prctl_map.start_code = mm->start_code;
2149 prctl_map.end_code = mm->end_code;
2150 prctl_map.start_data = mm->start_data;
2151 prctl_map.end_data = mm->end_data;
2152 prctl_map.start_brk = mm->start_brk;
2153 prctl_map.brk = mm->brk;
2154 prctl_map.start_stack = mm->start_stack;
2155 prctl_map.arg_start = mm->arg_start;
2156 prctl_map.arg_end = mm->arg_end;
2157 prctl_map.env_start = mm->env_start;
2158 prctl_map.env_end = mm->env_end;
2161 case PR_SET_MM_START_CODE:
2162 prctl_map.start_code = addr;
2164 case PR_SET_MM_END_CODE:
2165 prctl_map.end_code = addr;
2167 case PR_SET_MM_START_DATA:
2168 prctl_map.start_data = addr;
2170 case PR_SET_MM_END_DATA:
2171 prctl_map.end_data = addr;
2173 case PR_SET_MM_START_STACK:
2174 prctl_map.start_stack = addr;
2176 case PR_SET_MM_START_BRK:
2177 prctl_map.start_brk = addr;
2180 prctl_map.brk = addr;
2182 case PR_SET_MM_ARG_START:
2183 prctl_map.arg_start = addr;
2185 case PR_SET_MM_ARG_END:
2186 prctl_map.arg_end = addr;
2188 case PR_SET_MM_ENV_START:
2189 prctl_map.env_start = addr;
2191 case PR_SET_MM_ENV_END:
2192 prctl_map.env_end = addr;
2198 error = validate_prctl_map_addr(&prctl_map);
2204 * If command line arguments and environment
2205 * are placed somewhere else on stack, we can
2206 * set them up here, ARG_START/END to setup
2207 * command line argumets and ENV_START/END
2210 case PR_SET_MM_START_STACK:
2211 case PR_SET_MM_ARG_START:
2212 case PR_SET_MM_ARG_END:
2213 case PR_SET_MM_ENV_START:
2214 case PR_SET_MM_ENV_END:
2221 mm->start_code = prctl_map.start_code;
2222 mm->end_code = prctl_map.end_code;
2223 mm->start_data = prctl_map.start_data;
2224 mm->end_data = prctl_map.end_data;
2225 mm->start_brk = prctl_map.start_brk;
2226 mm->brk = prctl_map.brk;
2227 mm->start_stack = prctl_map.start_stack;
2228 mm->arg_start = prctl_map.arg_start;
2229 mm->arg_end = prctl_map.arg_end;
2230 mm->env_start = prctl_map.env_start;
2231 mm->env_end = prctl_map.env_end;
2235 spin_unlock(&mm->arg_lock);
2236 mmap_read_unlock(mm);
2240 #ifdef CONFIG_CHECKPOINT_RESTORE
2241 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2243 return put_user(me->clear_child_tid, tid_addr);
2246 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2252 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2255 * If task has has_child_subreaper - all its decendants
2256 * already have these flag too and new decendants will
2257 * inherit it on fork, skip them.
2259 * If we've found child_reaper - skip descendants in
2260 * it's subtree as they will never get out pidns.
2262 if (p->signal->has_child_subreaper ||
2263 is_child_reaper(task_pid(p)))
2266 p->signal->has_child_subreaper = 1;
2270 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2275 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2281 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2283 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2284 unsigned long, arg4, unsigned long, arg5)
2286 struct task_struct *me = current;
2287 unsigned char comm[sizeof(me->comm)];
2290 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2291 if (error != -ENOSYS)
2296 case PR_SET_PDEATHSIG:
2297 if (!valid_signal(arg2)) {
2301 me->pdeath_signal = arg2;
2303 case PR_GET_PDEATHSIG:
2304 error = put_user(me->pdeath_signal, (int __user *)arg2);
2306 case PR_GET_DUMPABLE:
2307 error = get_dumpable(me->mm);
2309 case PR_SET_DUMPABLE:
2310 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2314 set_dumpable(me->mm, arg2);
2317 case PR_SET_UNALIGN:
2318 error = SET_UNALIGN_CTL(me, arg2);
2320 case PR_GET_UNALIGN:
2321 error = GET_UNALIGN_CTL(me, arg2);
2324 error = SET_FPEMU_CTL(me, arg2);
2327 error = GET_FPEMU_CTL(me, arg2);
2330 error = SET_FPEXC_CTL(me, arg2);
2333 error = GET_FPEXC_CTL(me, arg2);
2336 error = PR_TIMING_STATISTICAL;
2339 if (arg2 != PR_TIMING_STATISTICAL)
2343 comm[sizeof(me->comm) - 1] = 0;
2344 if (strncpy_from_user(comm, (char __user *)arg2,
2345 sizeof(me->comm) - 1) < 0)
2347 set_task_comm(me, comm);
2348 proc_comm_connector(me);
2351 get_task_comm(comm, me);
2352 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2356 error = GET_ENDIAN(me, arg2);
2359 error = SET_ENDIAN(me, arg2);
2361 case PR_GET_SECCOMP:
2362 error = prctl_get_seccomp();
2364 case PR_SET_SECCOMP:
2365 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2368 error = GET_TSC_CTL(arg2);
2371 error = SET_TSC_CTL(arg2);
2373 case PR_TASK_PERF_EVENTS_DISABLE:
2374 error = perf_event_task_disable();
2376 case PR_TASK_PERF_EVENTS_ENABLE:
2377 error = perf_event_task_enable();
2379 case PR_GET_TIMERSLACK:
2380 if (current->timer_slack_ns > ULONG_MAX)
2383 error = current->timer_slack_ns;
2385 case PR_SET_TIMERSLACK:
2387 current->timer_slack_ns =
2388 current->default_timer_slack_ns;
2390 current->timer_slack_ns = arg2;
2396 case PR_MCE_KILL_CLEAR:
2399 current->flags &= ~PF_MCE_PROCESS;
2401 case PR_MCE_KILL_SET:
2402 current->flags |= PF_MCE_PROCESS;
2403 if (arg3 == PR_MCE_KILL_EARLY)
2404 current->flags |= PF_MCE_EARLY;
2405 else if (arg3 == PR_MCE_KILL_LATE)
2406 current->flags &= ~PF_MCE_EARLY;
2407 else if (arg3 == PR_MCE_KILL_DEFAULT)
2409 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2417 case PR_MCE_KILL_GET:
2418 if (arg2 | arg3 | arg4 | arg5)
2420 if (current->flags & PF_MCE_PROCESS)
2421 error = (current->flags & PF_MCE_EARLY) ?
2422 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2424 error = PR_MCE_KILL_DEFAULT;
2427 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2429 case PR_GET_TID_ADDRESS:
2430 error = prctl_get_tid_address(me, (int __user **)arg2);
2432 case PR_SET_CHILD_SUBREAPER:
2433 me->signal->is_child_subreaper = !!arg2;
2437 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2439 case PR_GET_CHILD_SUBREAPER:
2440 error = put_user(me->signal->is_child_subreaper,
2441 (int __user *)arg2);
2443 case PR_SET_NO_NEW_PRIVS:
2444 if (arg2 != 1 || arg3 || arg4 || arg5)
2447 task_set_no_new_privs(current);
2449 case PR_GET_NO_NEW_PRIVS:
2450 if (arg2 || arg3 || arg4 || arg5)
2452 return task_no_new_privs(current) ? 1 : 0;
2453 case PR_GET_THP_DISABLE:
2454 if (arg2 || arg3 || arg4 || arg5)
2456 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2458 case PR_SET_THP_DISABLE:
2459 if (arg3 || arg4 || arg5)
2461 if (mmap_write_lock_killable(me->mm))
2464 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2466 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2467 mmap_write_unlock(me->mm);
2469 case PR_MPX_ENABLE_MANAGEMENT:
2470 case PR_MPX_DISABLE_MANAGEMENT:
2471 /* No longer implemented: */
2473 case PR_SET_FP_MODE:
2474 error = SET_FP_MODE(me, arg2);
2476 case PR_GET_FP_MODE:
2477 error = GET_FP_MODE(me);
2480 error = SVE_SET_VL(arg2);
2483 error = SVE_GET_VL();
2485 case PR_GET_SPECULATION_CTRL:
2486 if (arg3 || arg4 || arg5)
2488 error = arch_prctl_spec_ctrl_get(me, arg2);
2490 case PR_SET_SPECULATION_CTRL:
2493 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2495 case PR_PAC_RESET_KEYS:
2496 if (arg3 || arg4 || arg5)
2498 error = PAC_RESET_KEYS(me, arg2);
2500 case PR_SET_TAGGED_ADDR_CTRL:
2501 if (arg3 || arg4 || arg5)
2503 error = SET_TAGGED_ADDR_CTRL(arg2);
2505 case PR_GET_TAGGED_ADDR_CTRL:
2506 if (arg2 || arg3 || arg4 || arg5)
2508 error = GET_TAGGED_ADDR_CTRL();
2510 case PR_SET_IO_FLUSHER:
2511 if (!capable(CAP_SYS_RESOURCE))
2514 if (arg3 || arg4 || arg5)
2518 current->flags |= PR_IO_FLUSHER;
2520 current->flags &= ~PR_IO_FLUSHER;
2524 case PR_GET_IO_FLUSHER:
2525 if (!capable(CAP_SYS_RESOURCE))
2528 if (arg2 || arg3 || arg4 || arg5)
2531 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2540 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2541 struct getcpu_cache __user *, unused)
2544 int cpu = raw_smp_processor_id();
2547 err |= put_user(cpu, cpup);
2549 err |= put_user(cpu_to_node(cpu), nodep);
2550 return err ? -EFAULT : 0;
2554 * do_sysinfo - fill in sysinfo struct
2555 * @info: pointer to buffer to fill
2557 static int do_sysinfo(struct sysinfo *info)
2559 unsigned long mem_total, sav_total;
2560 unsigned int mem_unit, bitcount;
2561 struct timespec64 tp;
2563 memset(info, 0, sizeof(struct sysinfo));
2565 ktime_get_boottime_ts64(&tp);
2566 timens_add_boottime(&tp);
2567 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2569 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2571 info->procs = nr_threads;
2577 * If the sum of all the available memory (i.e. ram + swap)
2578 * is less than can be stored in a 32 bit unsigned long then
2579 * we can be binary compatible with 2.2.x kernels. If not,
2580 * well, in that case 2.2.x was broken anyways...
2582 * -Erik Andersen <andersee@debian.org>
2585 mem_total = info->totalram + info->totalswap;
2586 if (mem_total < info->totalram || mem_total < info->totalswap)
2589 mem_unit = info->mem_unit;
2590 while (mem_unit > 1) {
2593 sav_total = mem_total;
2595 if (mem_total < sav_total)
2600 * If mem_total did not overflow, multiply all memory values by
2601 * info->mem_unit and set it to 1. This leaves things compatible
2602 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2607 info->totalram <<= bitcount;
2608 info->freeram <<= bitcount;
2609 info->sharedram <<= bitcount;
2610 info->bufferram <<= bitcount;
2611 info->totalswap <<= bitcount;
2612 info->freeswap <<= bitcount;
2613 info->totalhigh <<= bitcount;
2614 info->freehigh <<= bitcount;
2620 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2626 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2632 #ifdef CONFIG_COMPAT
2633 struct compat_sysinfo {
2647 char _f[20-2*sizeof(u32)-sizeof(int)];
2650 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2653 struct compat_sysinfo s_32;
2657 /* Check to see if any memory value is too large for 32-bit and scale
2660 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2663 while (s.mem_unit < PAGE_SIZE) {
2668 s.totalram >>= bitcount;
2669 s.freeram >>= bitcount;
2670 s.sharedram >>= bitcount;
2671 s.bufferram >>= bitcount;
2672 s.totalswap >>= bitcount;
2673 s.freeswap >>= bitcount;
2674 s.totalhigh >>= bitcount;
2675 s.freehigh >>= bitcount;
2678 memset(&s_32, 0, sizeof(s_32));
2679 s_32.uptime = s.uptime;
2680 s_32.loads[0] = s.loads[0];
2681 s_32.loads[1] = s.loads[1];
2682 s_32.loads[2] = s.loads[2];
2683 s_32.totalram = s.totalram;
2684 s_32.freeram = s.freeram;
2685 s_32.sharedram = s.sharedram;
2686 s_32.bufferram = s.bufferram;
2687 s_32.totalswap = s.totalswap;
2688 s_32.freeswap = s.freeswap;
2689 s_32.procs = s.procs;
2690 s_32.totalhigh = s.totalhigh;
2691 s_32.freehigh = s.freehigh;
2692 s_32.mem_unit = s.mem_unit;
2693 if (copy_to_user(info, &s_32, sizeof(s_32)))
2697 #endif /* CONFIG_COMPAT */