1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/export.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
44 #include <linux/syscall_user_dispatch.h>
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/time_namespace.h>
51 #include <linux/binfmts.h>
53 #include <linux/sched.h>
54 #include <linux/sched/autogroup.h>
55 #include <linux/sched/loadavg.h>
56 #include <linux/sched/stat.h>
57 #include <linux/sched/mm.h>
58 #include <linux/sched/coredump.h>
59 #include <linux/sched/task.h>
60 #include <linux/sched/cputime.h>
61 #include <linux/rcupdate.h>
62 #include <linux/uidgid.h>
63 #include <linux/cred.h>
65 #include <linux/nospec.h>
67 #include <linux/kmsg_dump.h>
68 /* Move somewhere else to avoid recompiling? */
69 #include <generated/utsrelease.h>
71 #include <linux/uaccess.h>
73 #include <asm/unistd.h>
77 #ifndef SET_UNALIGN_CTL
78 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
80 #ifndef GET_UNALIGN_CTL
81 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
84 # define SET_FPEMU_CTL(a, b) (-EINVAL)
87 # define GET_FPEMU_CTL(a, b) (-EINVAL)
90 # define SET_FPEXC_CTL(a, b) (-EINVAL)
93 # define GET_FPEXC_CTL(a, b) (-EINVAL)
96 # define GET_ENDIAN(a, b) (-EINVAL)
99 # define SET_ENDIAN(a, b) (-EINVAL)
102 # define GET_TSC_CTL(a) (-EINVAL)
105 # define SET_TSC_CTL(a) (-EINVAL)
108 # define GET_FP_MODE(a) (-EINVAL)
111 # define SET_FP_MODE(a,b) (-EINVAL)
114 # define SVE_SET_VL(a) (-EINVAL)
117 # define SVE_GET_VL() (-EINVAL)
119 #ifndef PAC_RESET_KEYS
120 # define PAC_RESET_KEYS(a, b) (-EINVAL)
122 #ifndef PAC_SET_ENABLED_KEYS
123 # define PAC_SET_ENABLED_KEYS(a, b, c) (-EINVAL)
125 #ifndef PAC_GET_ENABLED_KEYS
126 # define PAC_GET_ENABLED_KEYS(a) (-EINVAL)
128 #ifndef SET_TAGGED_ADDR_CTRL
129 # define SET_TAGGED_ADDR_CTRL(a) (-EINVAL)
131 #ifndef GET_TAGGED_ADDR_CTRL
132 # define GET_TAGGED_ADDR_CTRL() (-EINVAL)
136 * this is where the system-wide overflow UID and GID are defined, for
137 * architectures that now have 32-bit UID/GID but didn't in the past
140 int overflowuid = DEFAULT_OVERFLOWUID;
141 int overflowgid = DEFAULT_OVERFLOWGID;
143 EXPORT_SYMBOL(overflowuid);
144 EXPORT_SYMBOL(overflowgid);
147 * the same as above, but for filesystems which can only store a 16-bit
148 * UID and GID. as such, this is needed on all architectures
151 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
152 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
154 EXPORT_SYMBOL(fs_overflowuid);
155 EXPORT_SYMBOL(fs_overflowgid);
158 * Returns true if current's euid is same as p's uid or euid,
159 * or has CAP_SYS_NICE to p's user_ns.
161 * Called with rcu_read_lock, creds are safe
163 static bool set_one_prio_perm(struct task_struct *p)
165 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
167 if (uid_eq(pcred->uid, cred->euid) ||
168 uid_eq(pcred->euid, cred->euid))
170 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
176 * set the priority of a task
177 * - the caller must hold the RCU read lock
179 static int set_one_prio(struct task_struct *p, int niceval, int error)
183 if (!set_one_prio_perm(p)) {
187 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
191 no_nice = security_task_setnice(p, niceval);
198 set_user_nice(p, niceval);
203 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
205 struct task_struct *g, *p;
206 struct user_struct *user;
207 const struct cred *cred = current_cred();
212 if (which > PRIO_USER || which < PRIO_PROCESS)
215 /* normalize: avoid signed division (rounding problems) */
217 if (niceval < MIN_NICE)
219 if (niceval > MAX_NICE)
223 read_lock(&tasklist_lock);
227 p = find_task_by_vpid(who);
231 error = set_one_prio(p, niceval, error);
235 pgrp = find_vpid(who);
237 pgrp = task_pgrp(current);
238 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
239 error = set_one_prio(p, niceval, error);
240 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
243 uid = make_kuid(cred->user_ns, who);
247 else if (!uid_eq(uid, cred->uid)) {
248 user = find_user(uid);
250 goto out_unlock; /* No processes for this user */
252 do_each_thread(g, p) {
253 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
254 error = set_one_prio(p, niceval, error);
255 } while_each_thread(g, p);
256 if (!uid_eq(uid, cred->uid))
257 free_uid(user); /* For find_user() */
261 read_unlock(&tasklist_lock);
268 * Ugh. To avoid negative return values, "getpriority()" will
269 * not return the normal nice-value, but a negated value that
270 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
271 * to stay compatible.
273 SYSCALL_DEFINE2(getpriority, int, which, int, who)
275 struct task_struct *g, *p;
276 struct user_struct *user;
277 const struct cred *cred = current_cred();
278 long niceval, retval = -ESRCH;
282 if (which > PRIO_USER || which < PRIO_PROCESS)
286 read_lock(&tasklist_lock);
290 p = find_task_by_vpid(who);
294 niceval = nice_to_rlimit(task_nice(p));
295 if (niceval > retval)
301 pgrp = find_vpid(who);
303 pgrp = task_pgrp(current);
304 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
305 niceval = nice_to_rlimit(task_nice(p));
306 if (niceval > retval)
308 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
311 uid = make_kuid(cred->user_ns, who);
315 else if (!uid_eq(uid, cred->uid)) {
316 user = find_user(uid);
318 goto out_unlock; /* No processes for this user */
320 do_each_thread(g, p) {
321 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
322 niceval = nice_to_rlimit(task_nice(p));
323 if (niceval > retval)
326 } while_each_thread(g, p);
327 if (!uid_eq(uid, cred->uid))
328 free_uid(user); /* for find_user() */
332 read_unlock(&tasklist_lock);
339 * Unprivileged users may change the real gid to the effective gid
340 * or vice versa. (BSD-style)
342 * If you set the real gid at all, or set the effective gid to a value not
343 * equal to the real gid, then the saved gid is set to the new effective gid.
345 * This makes it possible for a setgid program to completely drop its
346 * privileges, which is often a useful assertion to make when you are doing
347 * a security audit over a program.
349 * The general idea is that a program which uses just setregid() will be
350 * 100% compatible with BSD. A program which uses just setgid() will be
351 * 100% compatible with POSIX with saved IDs.
353 * SMP: There are not races, the GIDs are checked only by filesystem
354 * operations (as far as semantic preservation is concerned).
356 #ifdef CONFIG_MULTIUSER
357 long __sys_setregid(gid_t rgid, gid_t egid)
359 struct user_namespace *ns = current_user_ns();
360 const struct cred *old;
365 krgid = make_kgid(ns, rgid);
366 kegid = make_kgid(ns, egid);
368 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
370 if ((egid != (gid_t) -1) && !gid_valid(kegid))
373 new = prepare_creds();
376 old = current_cred();
379 if (rgid != (gid_t) -1) {
380 if (gid_eq(old->gid, krgid) ||
381 gid_eq(old->egid, krgid) ||
382 ns_capable_setid(old->user_ns, CAP_SETGID))
387 if (egid != (gid_t) -1) {
388 if (gid_eq(old->gid, kegid) ||
389 gid_eq(old->egid, kegid) ||
390 gid_eq(old->sgid, kegid) ||
391 ns_capable_setid(old->user_ns, CAP_SETGID))
397 if (rgid != (gid_t) -1 ||
398 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
399 new->sgid = new->egid;
400 new->fsgid = new->egid;
402 retval = security_task_fix_setgid(new, old, LSM_SETID_RE);
406 return commit_creds(new);
413 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
415 return __sys_setregid(rgid, egid);
419 * setgid() is implemented like SysV w/ SAVED_IDS
421 * SMP: Same implicit races as above.
423 long __sys_setgid(gid_t gid)
425 struct user_namespace *ns = current_user_ns();
426 const struct cred *old;
431 kgid = make_kgid(ns, gid);
432 if (!gid_valid(kgid))
435 new = prepare_creds();
438 old = current_cred();
441 if (ns_capable_setid(old->user_ns, CAP_SETGID))
442 new->gid = new->egid = new->sgid = new->fsgid = kgid;
443 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
444 new->egid = new->fsgid = kgid;
448 retval = security_task_fix_setgid(new, old, LSM_SETID_ID);
452 return commit_creds(new);
459 SYSCALL_DEFINE1(setgid, gid_t, gid)
461 return __sys_setgid(gid);
465 * change the user struct in a credentials set to match the new UID
467 static int set_user(struct cred *new)
469 struct user_struct *new_user;
471 new_user = alloc_uid(new->uid);
476 * We don't fail in case of NPROC limit excess here because too many
477 * poorly written programs don't check set*uid() return code, assuming
478 * it never fails if called by root. We may still enforce NPROC limit
479 * for programs doing set*uid()+execve() by harmlessly deferring the
480 * failure to the execve() stage.
482 if (is_ucounts_overlimit(new->ucounts, UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC)) &&
483 new_user != INIT_USER)
484 current->flags |= PF_NPROC_EXCEEDED;
486 current->flags &= ~PF_NPROC_EXCEEDED;
489 new->user = new_user;
494 * Unprivileged users may change the real uid to the effective uid
495 * or vice versa. (BSD-style)
497 * If you set the real uid at all, or set the effective uid to a value not
498 * equal to the real uid, then the saved uid is set to the new effective uid.
500 * This makes it possible for a setuid program to completely drop its
501 * privileges, which is often a useful assertion to make when you are doing
502 * a security audit over a program.
504 * The general idea is that a program which uses just setreuid() will be
505 * 100% compatible with BSD. A program which uses just setuid() will be
506 * 100% compatible with POSIX with saved IDs.
508 long __sys_setreuid(uid_t ruid, uid_t euid)
510 struct user_namespace *ns = current_user_ns();
511 const struct cred *old;
516 kruid = make_kuid(ns, ruid);
517 keuid = make_kuid(ns, euid);
519 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
521 if ((euid != (uid_t) -1) && !uid_valid(keuid))
524 new = prepare_creds();
527 old = current_cred();
530 if (ruid != (uid_t) -1) {
532 if (!uid_eq(old->uid, kruid) &&
533 !uid_eq(old->euid, kruid) &&
534 !ns_capable_setid(old->user_ns, CAP_SETUID))
538 if (euid != (uid_t) -1) {
540 if (!uid_eq(old->uid, keuid) &&
541 !uid_eq(old->euid, keuid) &&
542 !uid_eq(old->suid, keuid) &&
543 !ns_capable_setid(old->user_ns, CAP_SETUID))
547 if (!uid_eq(new->uid, old->uid)) {
548 retval = set_user(new);
552 if (ruid != (uid_t) -1 ||
553 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
554 new->suid = new->euid;
555 new->fsuid = new->euid;
557 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
561 retval = set_cred_ucounts(new);
565 return commit_creds(new);
572 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
574 return __sys_setreuid(ruid, euid);
578 * setuid() is implemented like SysV with SAVED_IDS
580 * Note that SAVED_ID's is deficient in that a setuid root program
581 * like sendmail, for example, cannot set its uid to be a normal
582 * user and then switch back, because if you're root, setuid() sets
583 * the saved uid too. If you don't like this, blame the bright people
584 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
585 * will allow a root program to temporarily drop privileges and be able to
586 * regain them by swapping the real and effective uid.
588 long __sys_setuid(uid_t uid)
590 struct user_namespace *ns = current_user_ns();
591 const struct cred *old;
596 kuid = make_kuid(ns, uid);
597 if (!uid_valid(kuid))
600 new = prepare_creds();
603 old = current_cred();
606 if (ns_capable_setid(old->user_ns, CAP_SETUID)) {
607 new->suid = new->uid = kuid;
608 if (!uid_eq(kuid, old->uid)) {
609 retval = set_user(new);
613 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
617 new->fsuid = new->euid = kuid;
619 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
623 retval = set_cred_ucounts(new);
627 return commit_creds(new);
634 SYSCALL_DEFINE1(setuid, uid_t, uid)
636 return __sys_setuid(uid);
641 * This function implements a generic ability to update ruid, euid,
642 * and suid. This allows you to implement the 4.4 compatible seteuid().
644 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
646 struct user_namespace *ns = current_user_ns();
647 const struct cred *old;
650 kuid_t kruid, keuid, ksuid;
652 kruid = make_kuid(ns, ruid);
653 keuid = make_kuid(ns, euid);
654 ksuid = make_kuid(ns, suid);
656 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
659 if ((euid != (uid_t) -1) && !uid_valid(keuid))
662 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
665 new = prepare_creds();
669 old = current_cred();
672 if (!ns_capable_setid(old->user_ns, CAP_SETUID)) {
673 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
674 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
676 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
677 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
679 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
680 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
684 if (ruid != (uid_t) -1) {
686 if (!uid_eq(kruid, old->uid)) {
687 retval = set_user(new);
692 if (euid != (uid_t) -1)
694 if (suid != (uid_t) -1)
696 new->fsuid = new->euid;
698 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
702 retval = set_cred_ucounts(new);
706 return commit_creds(new);
713 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
715 return __sys_setresuid(ruid, euid, suid);
718 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
720 const struct cred *cred = current_cred();
722 uid_t ruid, euid, suid;
724 ruid = from_kuid_munged(cred->user_ns, cred->uid);
725 euid = from_kuid_munged(cred->user_ns, cred->euid);
726 suid = from_kuid_munged(cred->user_ns, cred->suid);
728 retval = put_user(ruid, ruidp);
730 retval = put_user(euid, euidp);
732 return put_user(suid, suidp);
738 * Same as above, but for rgid, egid, sgid.
740 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
742 struct user_namespace *ns = current_user_ns();
743 const struct cred *old;
746 kgid_t krgid, kegid, ksgid;
748 krgid = make_kgid(ns, rgid);
749 kegid = make_kgid(ns, egid);
750 ksgid = make_kgid(ns, sgid);
752 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
754 if ((egid != (gid_t) -1) && !gid_valid(kegid))
756 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
759 new = prepare_creds();
762 old = current_cred();
765 if (!ns_capable_setid(old->user_ns, CAP_SETGID)) {
766 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
767 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
769 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
770 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
772 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
773 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
777 if (rgid != (gid_t) -1)
779 if (egid != (gid_t) -1)
781 if (sgid != (gid_t) -1)
783 new->fsgid = new->egid;
785 retval = security_task_fix_setgid(new, old, LSM_SETID_RES);
789 return commit_creds(new);
796 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
798 return __sys_setresgid(rgid, egid, sgid);
801 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
803 const struct cred *cred = current_cred();
805 gid_t rgid, egid, sgid;
807 rgid = from_kgid_munged(cred->user_ns, cred->gid);
808 egid = from_kgid_munged(cred->user_ns, cred->egid);
809 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
811 retval = put_user(rgid, rgidp);
813 retval = put_user(egid, egidp);
815 retval = put_user(sgid, sgidp);
823 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
824 * is used for "access()" and for the NFS daemon (letting nfsd stay at
825 * whatever uid it wants to). It normally shadows "euid", except when
826 * explicitly set by setfsuid() or for access..
828 long __sys_setfsuid(uid_t uid)
830 const struct cred *old;
835 old = current_cred();
836 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
838 kuid = make_kuid(old->user_ns, uid);
839 if (!uid_valid(kuid))
842 new = prepare_creds();
846 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
847 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
848 ns_capable_setid(old->user_ns, CAP_SETUID)) {
849 if (!uid_eq(kuid, old->fsuid)) {
851 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
864 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
866 return __sys_setfsuid(uid);
870 * Samma på svenska..
872 long __sys_setfsgid(gid_t gid)
874 const struct cred *old;
879 old = current_cred();
880 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
882 kgid = make_kgid(old->user_ns, gid);
883 if (!gid_valid(kgid))
886 new = prepare_creds();
890 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
891 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
892 ns_capable_setid(old->user_ns, CAP_SETGID)) {
893 if (!gid_eq(kgid, old->fsgid)) {
895 if (security_task_fix_setgid(new,old,LSM_SETID_FS) == 0)
908 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
910 return __sys_setfsgid(gid);
912 #endif /* CONFIG_MULTIUSER */
915 * sys_getpid - return the thread group id of the current process
917 * Note, despite the name, this returns the tgid not the pid. The tgid and
918 * the pid are identical unless CLONE_THREAD was specified on clone() in
919 * which case the tgid is the same in all threads of the same group.
921 * This is SMP safe as current->tgid does not change.
923 SYSCALL_DEFINE0(getpid)
925 return task_tgid_vnr(current);
928 /* Thread ID - the internal kernel "pid" */
929 SYSCALL_DEFINE0(gettid)
931 return task_pid_vnr(current);
935 * Accessing ->real_parent is not SMP-safe, it could
936 * change from under us. However, we can use a stale
937 * value of ->real_parent under rcu_read_lock(), see
938 * release_task()->call_rcu(delayed_put_task_struct).
940 SYSCALL_DEFINE0(getppid)
945 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
951 SYSCALL_DEFINE0(getuid)
953 /* Only we change this so SMP safe */
954 return from_kuid_munged(current_user_ns(), current_uid());
957 SYSCALL_DEFINE0(geteuid)
959 /* Only we change this so SMP safe */
960 return from_kuid_munged(current_user_ns(), current_euid());
963 SYSCALL_DEFINE0(getgid)
965 /* Only we change this so SMP safe */
966 return from_kgid_munged(current_user_ns(), current_gid());
969 SYSCALL_DEFINE0(getegid)
971 /* Only we change this so SMP safe */
972 return from_kgid_munged(current_user_ns(), current_egid());
975 static void do_sys_times(struct tms *tms)
977 u64 tgutime, tgstime, cutime, cstime;
979 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
980 cutime = current->signal->cutime;
981 cstime = current->signal->cstime;
982 tms->tms_utime = nsec_to_clock_t(tgutime);
983 tms->tms_stime = nsec_to_clock_t(tgstime);
984 tms->tms_cutime = nsec_to_clock_t(cutime);
985 tms->tms_cstime = nsec_to_clock_t(cstime);
988 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
994 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
997 force_successful_syscall_return();
998 return (long) jiffies_64_to_clock_t(get_jiffies_64());
1001 #ifdef CONFIG_COMPAT
1002 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
1004 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
1007 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
1011 struct compat_tms tmp;
1014 /* Convert our struct tms to the compat version. */
1015 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
1016 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
1017 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
1018 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
1019 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
1022 force_successful_syscall_return();
1023 return compat_jiffies_to_clock_t(jiffies);
1028 * This needs some heavy checking ...
1029 * I just haven't the stomach for it. I also don't fully
1030 * understand sessions/pgrp etc. Let somebody who does explain it.
1032 * OK, I think I have the protection semantics right.... this is really
1033 * only important on a multi-user system anyway, to make sure one user
1034 * can't send a signal to a process owned by another. -TYT, 12/12/91
1036 * !PF_FORKNOEXEC check to conform completely to POSIX.
1038 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1040 struct task_struct *p;
1041 struct task_struct *group_leader = current->group_leader;
1046 pid = task_pid_vnr(group_leader);
1053 /* From this point forward we keep holding onto the tasklist lock
1054 * so that our parent does not change from under us. -DaveM
1056 write_lock_irq(&tasklist_lock);
1059 p = find_task_by_vpid(pid);
1064 if (!thread_group_leader(p))
1067 if (same_thread_group(p->real_parent, group_leader)) {
1069 if (task_session(p) != task_session(group_leader))
1072 if (!(p->flags & PF_FORKNOEXEC))
1076 if (p != group_leader)
1081 if (p->signal->leader)
1086 struct task_struct *g;
1088 pgrp = find_vpid(pgid);
1089 g = pid_task(pgrp, PIDTYPE_PGID);
1090 if (!g || task_session(g) != task_session(group_leader))
1094 err = security_task_setpgid(p, pgid);
1098 if (task_pgrp(p) != pgrp)
1099 change_pid(p, PIDTYPE_PGID, pgrp);
1103 /* All paths lead to here, thus we are safe. -DaveM */
1104 write_unlock_irq(&tasklist_lock);
1109 static int do_getpgid(pid_t pid)
1111 struct task_struct *p;
1117 grp = task_pgrp(current);
1120 p = find_task_by_vpid(pid);
1127 retval = security_task_getpgid(p);
1131 retval = pid_vnr(grp);
1137 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1139 return do_getpgid(pid);
1142 #ifdef __ARCH_WANT_SYS_GETPGRP
1144 SYSCALL_DEFINE0(getpgrp)
1146 return do_getpgid(0);
1151 SYSCALL_DEFINE1(getsid, pid_t, pid)
1153 struct task_struct *p;
1159 sid = task_session(current);
1162 p = find_task_by_vpid(pid);
1165 sid = task_session(p);
1169 retval = security_task_getsid(p);
1173 retval = pid_vnr(sid);
1179 static void set_special_pids(struct pid *pid)
1181 struct task_struct *curr = current->group_leader;
1183 if (task_session(curr) != pid)
1184 change_pid(curr, PIDTYPE_SID, pid);
1186 if (task_pgrp(curr) != pid)
1187 change_pid(curr, PIDTYPE_PGID, pid);
1190 int ksys_setsid(void)
1192 struct task_struct *group_leader = current->group_leader;
1193 struct pid *sid = task_pid(group_leader);
1194 pid_t session = pid_vnr(sid);
1197 write_lock_irq(&tasklist_lock);
1198 /* Fail if I am already a session leader */
1199 if (group_leader->signal->leader)
1202 /* Fail if a process group id already exists that equals the
1203 * proposed session id.
1205 if (pid_task(sid, PIDTYPE_PGID))
1208 group_leader->signal->leader = 1;
1209 set_special_pids(sid);
1211 proc_clear_tty(group_leader);
1215 write_unlock_irq(&tasklist_lock);
1217 proc_sid_connector(group_leader);
1218 sched_autogroup_create_attach(group_leader);
1223 SYSCALL_DEFINE0(setsid)
1225 return ksys_setsid();
1228 DECLARE_RWSEM(uts_sem);
1230 #ifdef COMPAT_UTS_MACHINE
1231 #define override_architecture(name) \
1232 (personality(current->personality) == PER_LINUX32 && \
1233 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1234 sizeof(COMPAT_UTS_MACHINE)))
1236 #define override_architecture(name) 0
1240 * Work around broken programs that cannot handle "Linux 3.0".
1241 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1242 * And we map 4.x and later versions to 2.6.60+x, so 4.0/5.0/6.0/... would be
1245 static int override_release(char __user *release, size_t len)
1249 if (current->personality & UNAME26) {
1250 const char *rest = UTS_RELEASE;
1251 char buf[65] = { 0 };
1257 if (*rest == '.' && ++ndots >= 3)
1259 if (!isdigit(*rest) && *rest != '.')
1263 v = LINUX_VERSION_PATCHLEVEL + 60;
1264 copy = clamp_t(size_t, len, 1, sizeof(buf));
1265 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1266 ret = copy_to_user(release, buf, copy + 1);
1271 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1273 struct new_utsname tmp;
1275 down_read(&uts_sem);
1276 memcpy(&tmp, utsname(), sizeof(tmp));
1278 if (copy_to_user(name, &tmp, sizeof(tmp)))
1281 if (override_release(name->release, sizeof(name->release)))
1283 if (override_architecture(name))
1288 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1292 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1294 struct old_utsname tmp;
1299 down_read(&uts_sem);
1300 memcpy(&tmp, utsname(), sizeof(tmp));
1302 if (copy_to_user(name, &tmp, sizeof(tmp)))
1305 if (override_release(name->release, sizeof(name->release)))
1307 if (override_architecture(name))
1312 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1314 struct oldold_utsname tmp;
1319 memset(&tmp, 0, sizeof(tmp));
1321 down_read(&uts_sem);
1322 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1323 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1324 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1325 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1326 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1328 if (copy_to_user(name, &tmp, sizeof(tmp)))
1331 if (override_architecture(name))
1333 if (override_release(name->release, sizeof(name->release)))
1339 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1342 char tmp[__NEW_UTS_LEN];
1344 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1347 if (len < 0 || len > __NEW_UTS_LEN)
1350 if (!copy_from_user(tmp, name, len)) {
1351 struct new_utsname *u;
1353 down_write(&uts_sem);
1355 memcpy(u->nodename, tmp, len);
1356 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1358 uts_proc_notify(UTS_PROC_HOSTNAME);
1364 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1366 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1369 struct new_utsname *u;
1370 char tmp[__NEW_UTS_LEN + 1];
1374 down_read(&uts_sem);
1376 i = 1 + strlen(u->nodename);
1379 memcpy(tmp, u->nodename, i);
1381 if (copy_to_user(name, tmp, i))
1389 * Only setdomainname; getdomainname can be implemented by calling
1392 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1395 char tmp[__NEW_UTS_LEN];
1397 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1399 if (len < 0 || len > __NEW_UTS_LEN)
1403 if (!copy_from_user(tmp, name, len)) {
1404 struct new_utsname *u;
1406 down_write(&uts_sem);
1408 memcpy(u->domainname, tmp, len);
1409 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1411 uts_proc_notify(UTS_PROC_DOMAINNAME);
1417 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1419 struct rlimit value;
1422 ret = do_prlimit(current, resource, NULL, &value);
1424 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1429 #ifdef CONFIG_COMPAT
1431 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1432 struct compat_rlimit __user *, rlim)
1435 struct compat_rlimit r32;
1437 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1440 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1441 r.rlim_cur = RLIM_INFINITY;
1443 r.rlim_cur = r32.rlim_cur;
1444 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1445 r.rlim_max = RLIM_INFINITY;
1447 r.rlim_max = r32.rlim_max;
1448 return do_prlimit(current, resource, &r, NULL);
1451 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1452 struct compat_rlimit __user *, rlim)
1457 ret = do_prlimit(current, resource, NULL, &r);
1459 struct compat_rlimit r32;
1460 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1461 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1463 r32.rlim_cur = r.rlim_cur;
1464 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1465 r32.rlim_max = COMPAT_RLIM_INFINITY;
1467 r32.rlim_max = r.rlim_max;
1469 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1477 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1480 * Back compatibility for getrlimit. Needed for some apps.
1482 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1483 struct rlimit __user *, rlim)
1486 if (resource >= RLIM_NLIMITS)
1489 resource = array_index_nospec(resource, RLIM_NLIMITS);
1490 task_lock(current->group_leader);
1491 x = current->signal->rlim[resource];
1492 task_unlock(current->group_leader);
1493 if (x.rlim_cur > 0x7FFFFFFF)
1494 x.rlim_cur = 0x7FFFFFFF;
1495 if (x.rlim_max > 0x7FFFFFFF)
1496 x.rlim_max = 0x7FFFFFFF;
1497 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1500 #ifdef CONFIG_COMPAT
1501 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1502 struct compat_rlimit __user *, rlim)
1506 if (resource >= RLIM_NLIMITS)
1509 resource = array_index_nospec(resource, RLIM_NLIMITS);
1510 task_lock(current->group_leader);
1511 r = current->signal->rlim[resource];
1512 task_unlock(current->group_leader);
1513 if (r.rlim_cur > 0x7FFFFFFF)
1514 r.rlim_cur = 0x7FFFFFFF;
1515 if (r.rlim_max > 0x7FFFFFFF)
1516 r.rlim_max = 0x7FFFFFFF;
1518 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1519 put_user(r.rlim_max, &rlim->rlim_max))
1527 static inline bool rlim64_is_infinity(__u64 rlim64)
1529 #if BITS_PER_LONG < 64
1530 return rlim64 >= ULONG_MAX;
1532 return rlim64 == RLIM64_INFINITY;
1536 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1538 if (rlim->rlim_cur == RLIM_INFINITY)
1539 rlim64->rlim_cur = RLIM64_INFINITY;
1541 rlim64->rlim_cur = rlim->rlim_cur;
1542 if (rlim->rlim_max == RLIM_INFINITY)
1543 rlim64->rlim_max = RLIM64_INFINITY;
1545 rlim64->rlim_max = rlim->rlim_max;
1548 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1550 if (rlim64_is_infinity(rlim64->rlim_cur))
1551 rlim->rlim_cur = RLIM_INFINITY;
1553 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1554 if (rlim64_is_infinity(rlim64->rlim_max))
1555 rlim->rlim_max = RLIM_INFINITY;
1557 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1560 /* make sure you are allowed to change @tsk limits before calling this */
1561 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1562 struct rlimit *new_rlim, struct rlimit *old_rlim)
1564 struct rlimit *rlim;
1567 if (resource >= RLIM_NLIMITS)
1570 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1572 if (resource == RLIMIT_NOFILE &&
1573 new_rlim->rlim_max > sysctl_nr_open)
1577 /* protect tsk->signal and tsk->sighand from disappearing */
1578 read_lock(&tasklist_lock);
1579 if (!tsk->sighand) {
1584 rlim = tsk->signal->rlim + resource;
1585 task_lock(tsk->group_leader);
1587 /* Keep the capable check against init_user_ns until
1588 cgroups can contain all limits */
1589 if (new_rlim->rlim_max > rlim->rlim_max &&
1590 !capable(CAP_SYS_RESOURCE))
1593 retval = security_task_setrlimit(tsk, resource, new_rlim);
1601 task_unlock(tsk->group_leader);
1604 * RLIMIT_CPU handling. Arm the posix CPU timer if the limit is not
1605 * infinite. In case of RLIM_INFINITY the posix CPU timer code
1606 * ignores the rlimit.
1608 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1609 new_rlim->rlim_cur != RLIM_INFINITY &&
1610 IS_ENABLED(CONFIG_POSIX_TIMERS))
1611 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1613 read_unlock(&tasklist_lock);
1617 /* rcu lock must be held */
1618 static int check_prlimit_permission(struct task_struct *task,
1621 const struct cred *cred = current_cred(), *tcred;
1624 if (current == task)
1627 tcred = __task_cred(task);
1628 id_match = (uid_eq(cred->uid, tcred->euid) &&
1629 uid_eq(cred->uid, tcred->suid) &&
1630 uid_eq(cred->uid, tcred->uid) &&
1631 gid_eq(cred->gid, tcred->egid) &&
1632 gid_eq(cred->gid, tcred->sgid) &&
1633 gid_eq(cred->gid, tcred->gid));
1634 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1637 return security_task_prlimit(cred, tcred, flags);
1640 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1641 const struct rlimit64 __user *, new_rlim,
1642 struct rlimit64 __user *, old_rlim)
1644 struct rlimit64 old64, new64;
1645 struct rlimit old, new;
1646 struct task_struct *tsk;
1647 unsigned int checkflags = 0;
1651 checkflags |= LSM_PRLIMIT_READ;
1654 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1656 rlim64_to_rlim(&new64, &new);
1657 checkflags |= LSM_PRLIMIT_WRITE;
1661 tsk = pid ? find_task_by_vpid(pid) : current;
1666 ret = check_prlimit_permission(tsk, checkflags);
1671 get_task_struct(tsk);
1674 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1675 old_rlim ? &old : NULL);
1677 if (!ret && old_rlim) {
1678 rlim_to_rlim64(&old, &old64);
1679 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1683 put_task_struct(tsk);
1687 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1689 struct rlimit new_rlim;
1691 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1693 return do_prlimit(current, resource, &new_rlim, NULL);
1697 * It would make sense to put struct rusage in the task_struct,
1698 * except that would make the task_struct be *really big*. After
1699 * task_struct gets moved into malloc'ed memory, it would
1700 * make sense to do this. It will make moving the rest of the information
1701 * a lot simpler! (Which we're not doing right now because we're not
1702 * measuring them yet).
1704 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1705 * races with threads incrementing their own counters. But since word
1706 * reads are atomic, we either get new values or old values and we don't
1707 * care which for the sums. We always take the siglock to protect reading
1708 * the c* fields from p->signal from races with exit.c updating those
1709 * fields when reaping, so a sample either gets all the additions of a
1710 * given child after it's reaped, or none so this sample is before reaping.
1713 * We need to take the siglock for CHILDEREN, SELF and BOTH
1714 * for the cases current multithreaded, non-current single threaded
1715 * non-current multithreaded. Thread traversal is now safe with
1717 * Strictly speaking, we donot need to take the siglock if we are current and
1718 * single threaded, as no one else can take our signal_struct away, no one
1719 * else can reap the children to update signal->c* counters, and no one else
1720 * can race with the signal-> fields. If we do not take any lock, the
1721 * signal-> fields could be read out of order while another thread was just
1722 * exiting. So we should place a read memory barrier when we avoid the lock.
1723 * On the writer side, write memory barrier is implied in __exit_signal
1724 * as __exit_signal releases the siglock spinlock after updating the signal->
1725 * fields. But we don't do this yet to keep things simple.
1729 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1731 r->ru_nvcsw += t->nvcsw;
1732 r->ru_nivcsw += t->nivcsw;
1733 r->ru_minflt += t->min_flt;
1734 r->ru_majflt += t->maj_flt;
1735 r->ru_inblock += task_io_get_inblock(t);
1736 r->ru_oublock += task_io_get_oublock(t);
1739 void getrusage(struct task_struct *p, int who, struct rusage *r)
1741 struct task_struct *t;
1742 unsigned long flags;
1743 u64 tgutime, tgstime, utime, stime;
1744 unsigned long maxrss = 0;
1746 memset((char *)r, 0, sizeof (*r));
1749 if (who == RUSAGE_THREAD) {
1750 task_cputime_adjusted(current, &utime, &stime);
1751 accumulate_thread_rusage(p, r);
1752 maxrss = p->signal->maxrss;
1756 if (!lock_task_sighand(p, &flags))
1761 case RUSAGE_CHILDREN:
1762 utime = p->signal->cutime;
1763 stime = p->signal->cstime;
1764 r->ru_nvcsw = p->signal->cnvcsw;
1765 r->ru_nivcsw = p->signal->cnivcsw;
1766 r->ru_minflt = p->signal->cmin_flt;
1767 r->ru_majflt = p->signal->cmaj_flt;
1768 r->ru_inblock = p->signal->cinblock;
1769 r->ru_oublock = p->signal->coublock;
1770 maxrss = p->signal->cmaxrss;
1772 if (who == RUSAGE_CHILDREN)
1777 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1780 r->ru_nvcsw += p->signal->nvcsw;
1781 r->ru_nivcsw += p->signal->nivcsw;
1782 r->ru_minflt += p->signal->min_flt;
1783 r->ru_majflt += p->signal->maj_flt;
1784 r->ru_inblock += p->signal->inblock;
1785 r->ru_oublock += p->signal->oublock;
1786 if (maxrss < p->signal->maxrss)
1787 maxrss = p->signal->maxrss;
1790 accumulate_thread_rusage(t, r);
1791 } while_each_thread(p, t);
1797 unlock_task_sighand(p, &flags);
1800 r->ru_utime = ns_to_kernel_old_timeval(utime);
1801 r->ru_stime = ns_to_kernel_old_timeval(stime);
1803 if (who != RUSAGE_CHILDREN) {
1804 struct mm_struct *mm = get_task_mm(p);
1807 setmax_mm_hiwater_rss(&maxrss, mm);
1811 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1814 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1818 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1819 who != RUSAGE_THREAD)
1822 getrusage(current, who, &r);
1823 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1826 #ifdef CONFIG_COMPAT
1827 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1831 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1832 who != RUSAGE_THREAD)
1835 getrusage(current, who, &r);
1836 return put_compat_rusage(&r, ru);
1840 SYSCALL_DEFINE1(umask, int, mask)
1842 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1846 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1849 struct file *old_exe, *exe_file;
1850 struct inode *inode;
1857 inode = file_inode(exe.file);
1860 * Because the original mm->exe_file points to executable file, make
1861 * sure that this one is executable as well, to avoid breaking an
1865 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1868 err = file_permission(exe.file, MAY_EXEC);
1873 * Forbid mm->exe_file change if old file still mapped.
1875 exe_file = get_mm_exe_file(mm);
1878 struct vm_area_struct *vma;
1881 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1884 if (path_equal(&vma->vm_file->f_path,
1889 mmap_read_unlock(mm);
1894 /* set the new file, lockless */
1896 old_exe = xchg(&mm->exe_file, exe.file);
1903 mmap_read_unlock(mm);
1909 * Check arithmetic relations of passed addresses.
1911 * WARNING: we don't require any capability here so be very careful
1912 * in what is allowed for modification from userspace.
1914 static int validate_prctl_map_addr(struct prctl_mm_map *prctl_map)
1916 unsigned long mmap_max_addr = TASK_SIZE;
1917 int error = -EINVAL, i;
1919 static const unsigned char offsets[] = {
1920 offsetof(struct prctl_mm_map, start_code),
1921 offsetof(struct prctl_mm_map, end_code),
1922 offsetof(struct prctl_mm_map, start_data),
1923 offsetof(struct prctl_mm_map, end_data),
1924 offsetof(struct prctl_mm_map, start_brk),
1925 offsetof(struct prctl_mm_map, brk),
1926 offsetof(struct prctl_mm_map, start_stack),
1927 offsetof(struct prctl_mm_map, arg_start),
1928 offsetof(struct prctl_mm_map, arg_end),
1929 offsetof(struct prctl_mm_map, env_start),
1930 offsetof(struct prctl_mm_map, env_end),
1934 * Make sure the members are not somewhere outside
1935 * of allowed address space.
1937 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1938 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1940 if ((unsigned long)val >= mmap_max_addr ||
1941 (unsigned long)val < mmap_min_addr)
1946 * Make sure the pairs are ordered.
1948 #define __prctl_check_order(__m1, __op, __m2) \
1949 ((unsigned long)prctl_map->__m1 __op \
1950 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1951 error = __prctl_check_order(start_code, <, end_code);
1952 error |= __prctl_check_order(start_data,<=, end_data);
1953 error |= __prctl_check_order(start_brk, <=, brk);
1954 error |= __prctl_check_order(arg_start, <=, arg_end);
1955 error |= __prctl_check_order(env_start, <=, env_end);
1958 #undef __prctl_check_order
1963 * @brk should be after @end_data in traditional maps.
1965 if (prctl_map->start_brk <= prctl_map->end_data ||
1966 prctl_map->brk <= prctl_map->end_data)
1970 * Neither we should allow to override limits if they set.
1972 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1973 prctl_map->start_brk, prctl_map->end_data,
1974 prctl_map->start_data))
1982 #ifdef CONFIG_CHECKPOINT_RESTORE
1983 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1985 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1986 unsigned long user_auxv[AT_VECTOR_SIZE];
1987 struct mm_struct *mm = current->mm;
1990 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1991 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1993 if (opt == PR_SET_MM_MAP_SIZE)
1994 return put_user((unsigned int)sizeof(prctl_map),
1995 (unsigned int __user *)addr);
1997 if (data_size != sizeof(prctl_map))
2000 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
2003 error = validate_prctl_map_addr(&prctl_map);
2007 if (prctl_map.auxv_size) {
2009 * Someone is trying to cheat the auxv vector.
2011 if (!prctl_map.auxv ||
2012 prctl_map.auxv_size > sizeof(mm->saved_auxv))
2015 memset(user_auxv, 0, sizeof(user_auxv));
2016 if (copy_from_user(user_auxv,
2017 (const void __user *)prctl_map.auxv,
2018 prctl_map.auxv_size))
2021 /* Last entry must be AT_NULL as specification requires */
2022 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
2023 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2026 if (prctl_map.exe_fd != (u32)-1) {
2028 * Check if the current user is checkpoint/restore capable.
2029 * At the time of this writing, it checks for CAP_SYS_ADMIN
2030 * or CAP_CHECKPOINT_RESTORE.
2031 * Note that a user with access to ptrace can masquerade an
2032 * arbitrary program as any executable, even setuid ones.
2033 * This may have implications in the tomoyo subsystem.
2035 if (!checkpoint_restore_ns_capable(current_user_ns()))
2038 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2044 * arg_lock protects concurrent updates but we still need mmap_lock for
2045 * read to exclude races with sys_brk.
2050 * We don't validate if these members are pointing to
2051 * real present VMAs because application may have correspond
2052 * VMAs already unmapped and kernel uses these members for statistics
2053 * output in procfs mostly, except
2055 * - @start_brk/@brk which are used in do_brk_flags but kernel lookups
2056 * for VMAs when updating these members so anything wrong written
2057 * here cause kernel to swear at userspace program but won't lead
2058 * to any problem in kernel itself
2061 spin_lock(&mm->arg_lock);
2062 mm->start_code = prctl_map.start_code;
2063 mm->end_code = prctl_map.end_code;
2064 mm->start_data = prctl_map.start_data;
2065 mm->end_data = prctl_map.end_data;
2066 mm->start_brk = prctl_map.start_brk;
2067 mm->brk = prctl_map.brk;
2068 mm->start_stack = prctl_map.start_stack;
2069 mm->arg_start = prctl_map.arg_start;
2070 mm->arg_end = prctl_map.arg_end;
2071 mm->env_start = prctl_map.env_start;
2072 mm->env_end = prctl_map.env_end;
2073 spin_unlock(&mm->arg_lock);
2076 * Note this update of @saved_auxv is lockless thus
2077 * if someone reads this member in procfs while we're
2078 * updating -- it may get partly updated results. It's
2079 * known and acceptable trade off: we leave it as is to
2080 * not introduce additional locks here making the kernel
2083 if (prctl_map.auxv_size)
2084 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2086 mmap_read_unlock(mm);
2089 #endif /* CONFIG_CHECKPOINT_RESTORE */
2091 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2095 * This doesn't move the auxiliary vector itself since it's pinned to
2096 * mm_struct, but it permits filling the vector with new values. It's
2097 * up to the caller to provide sane values here, otherwise userspace
2098 * tools which use this vector might be unhappy.
2100 unsigned long user_auxv[AT_VECTOR_SIZE] = {};
2102 if (len > sizeof(user_auxv))
2105 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2108 /* Make sure the last entry is always AT_NULL */
2109 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2110 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2112 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2115 memcpy(mm->saved_auxv, user_auxv, len);
2116 task_unlock(current);
2121 static int prctl_set_mm(int opt, unsigned long addr,
2122 unsigned long arg4, unsigned long arg5)
2124 struct mm_struct *mm = current->mm;
2125 struct prctl_mm_map prctl_map = {
2130 struct vm_area_struct *vma;
2133 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2134 opt != PR_SET_MM_MAP &&
2135 opt != PR_SET_MM_MAP_SIZE)))
2138 #ifdef CONFIG_CHECKPOINT_RESTORE
2139 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2140 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2143 if (!capable(CAP_SYS_RESOURCE))
2146 if (opt == PR_SET_MM_EXE_FILE)
2147 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2149 if (opt == PR_SET_MM_AUXV)
2150 return prctl_set_auxv(mm, addr, arg4);
2152 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2158 * arg_lock protects concurrent updates of arg boundaries, we need
2159 * mmap_lock for a) concurrent sys_brk, b) finding VMA for addr
2163 vma = find_vma(mm, addr);
2165 spin_lock(&mm->arg_lock);
2166 prctl_map.start_code = mm->start_code;
2167 prctl_map.end_code = mm->end_code;
2168 prctl_map.start_data = mm->start_data;
2169 prctl_map.end_data = mm->end_data;
2170 prctl_map.start_brk = mm->start_brk;
2171 prctl_map.brk = mm->brk;
2172 prctl_map.start_stack = mm->start_stack;
2173 prctl_map.arg_start = mm->arg_start;
2174 prctl_map.arg_end = mm->arg_end;
2175 prctl_map.env_start = mm->env_start;
2176 prctl_map.env_end = mm->env_end;
2179 case PR_SET_MM_START_CODE:
2180 prctl_map.start_code = addr;
2182 case PR_SET_MM_END_CODE:
2183 prctl_map.end_code = addr;
2185 case PR_SET_MM_START_DATA:
2186 prctl_map.start_data = addr;
2188 case PR_SET_MM_END_DATA:
2189 prctl_map.end_data = addr;
2191 case PR_SET_MM_START_STACK:
2192 prctl_map.start_stack = addr;
2194 case PR_SET_MM_START_BRK:
2195 prctl_map.start_brk = addr;
2198 prctl_map.brk = addr;
2200 case PR_SET_MM_ARG_START:
2201 prctl_map.arg_start = addr;
2203 case PR_SET_MM_ARG_END:
2204 prctl_map.arg_end = addr;
2206 case PR_SET_MM_ENV_START:
2207 prctl_map.env_start = addr;
2209 case PR_SET_MM_ENV_END:
2210 prctl_map.env_end = addr;
2216 error = validate_prctl_map_addr(&prctl_map);
2222 * If command line arguments and environment
2223 * are placed somewhere else on stack, we can
2224 * set them up here, ARG_START/END to setup
2225 * command line arguments and ENV_START/END
2228 case PR_SET_MM_START_STACK:
2229 case PR_SET_MM_ARG_START:
2230 case PR_SET_MM_ARG_END:
2231 case PR_SET_MM_ENV_START:
2232 case PR_SET_MM_ENV_END:
2239 mm->start_code = prctl_map.start_code;
2240 mm->end_code = prctl_map.end_code;
2241 mm->start_data = prctl_map.start_data;
2242 mm->end_data = prctl_map.end_data;
2243 mm->start_brk = prctl_map.start_brk;
2244 mm->brk = prctl_map.brk;
2245 mm->start_stack = prctl_map.start_stack;
2246 mm->arg_start = prctl_map.arg_start;
2247 mm->arg_end = prctl_map.arg_end;
2248 mm->env_start = prctl_map.env_start;
2249 mm->env_end = prctl_map.env_end;
2253 spin_unlock(&mm->arg_lock);
2254 mmap_read_unlock(mm);
2258 #ifdef CONFIG_CHECKPOINT_RESTORE
2259 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2261 return put_user(me->clear_child_tid, tid_addr);
2264 static int prctl_get_tid_address(struct task_struct *me, int __user * __user *tid_addr)
2270 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2273 * If task has has_child_subreaper - all its descendants
2274 * already have these flag too and new descendants will
2275 * inherit it on fork, skip them.
2277 * If we've found child_reaper - skip descendants in
2278 * it's subtree as they will never get out pidns.
2280 if (p->signal->has_child_subreaper ||
2281 is_child_reaper(task_pid(p)))
2284 p->signal->has_child_subreaper = 1;
2288 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2293 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2299 #define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE)
2301 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2302 unsigned long, arg4, unsigned long, arg5)
2304 struct task_struct *me = current;
2305 unsigned char comm[sizeof(me->comm)];
2308 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2309 if (error != -ENOSYS)
2314 case PR_SET_PDEATHSIG:
2315 if (!valid_signal(arg2)) {
2319 me->pdeath_signal = arg2;
2321 case PR_GET_PDEATHSIG:
2322 error = put_user(me->pdeath_signal, (int __user *)arg2);
2324 case PR_GET_DUMPABLE:
2325 error = get_dumpable(me->mm);
2327 case PR_SET_DUMPABLE:
2328 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2332 set_dumpable(me->mm, arg2);
2335 case PR_SET_UNALIGN:
2336 error = SET_UNALIGN_CTL(me, arg2);
2338 case PR_GET_UNALIGN:
2339 error = GET_UNALIGN_CTL(me, arg2);
2342 error = SET_FPEMU_CTL(me, arg2);
2345 error = GET_FPEMU_CTL(me, arg2);
2348 error = SET_FPEXC_CTL(me, arg2);
2351 error = GET_FPEXC_CTL(me, arg2);
2354 error = PR_TIMING_STATISTICAL;
2357 if (arg2 != PR_TIMING_STATISTICAL)
2361 comm[sizeof(me->comm) - 1] = 0;
2362 if (strncpy_from_user(comm, (char __user *)arg2,
2363 sizeof(me->comm) - 1) < 0)
2365 set_task_comm(me, comm);
2366 proc_comm_connector(me);
2369 get_task_comm(comm, me);
2370 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2374 error = GET_ENDIAN(me, arg2);
2377 error = SET_ENDIAN(me, arg2);
2379 case PR_GET_SECCOMP:
2380 error = prctl_get_seccomp();
2382 case PR_SET_SECCOMP:
2383 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2386 error = GET_TSC_CTL(arg2);
2389 error = SET_TSC_CTL(arg2);
2391 case PR_TASK_PERF_EVENTS_DISABLE:
2392 error = perf_event_task_disable();
2394 case PR_TASK_PERF_EVENTS_ENABLE:
2395 error = perf_event_task_enable();
2397 case PR_GET_TIMERSLACK:
2398 if (current->timer_slack_ns > ULONG_MAX)
2401 error = current->timer_slack_ns;
2403 case PR_SET_TIMERSLACK:
2405 current->timer_slack_ns =
2406 current->default_timer_slack_ns;
2408 current->timer_slack_ns = arg2;
2414 case PR_MCE_KILL_CLEAR:
2417 current->flags &= ~PF_MCE_PROCESS;
2419 case PR_MCE_KILL_SET:
2420 current->flags |= PF_MCE_PROCESS;
2421 if (arg3 == PR_MCE_KILL_EARLY)
2422 current->flags |= PF_MCE_EARLY;
2423 else if (arg3 == PR_MCE_KILL_LATE)
2424 current->flags &= ~PF_MCE_EARLY;
2425 else if (arg3 == PR_MCE_KILL_DEFAULT)
2427 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2435 case PR_MCE_KILL_GET:
2436 if (arg2 | arg3 | arg4 | arg5)
2438 if (current->flags & PF_MCE_PROCESS)
2439 error = (current->flags & PF_MCE_EARLY) ?
2440 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2442 error = PR_MCE_KILL_DEFAULT;
2445 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2447 case PR_GET_TID_ADDRESS:
2448 error = prctl_get_tid_address(me, (int __user * __user *)arg2);
2450 case PR_SET_CHILD_SUBREAPER:
2451 me->signal->is_child_subreaper = !!arg2;
2455 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2457 case PR_GET_CHILD_SUBREAPER:
2458 error = put_user(me->signal->is_child_subreaper,
2459 (int __user *)arg2);
2461 case PR_SET_NO_NEW_PRIVS:
2462 if (arg2 != 1 || arg3 || arg4 || arg5)
2465 task_set_no_new_privs(current);
2467 case PR_GET_NO_NEW_PRIVS:
2468 if (arg2 || arg3 || arg4 || arg5)
2470 return task_no_new_privs(current) ? 1 : 0;
2471 case PR_GET_THP_DISABLE:
2472 if (arg2 || arg3 || arg4 || arg5)
2474 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2476 case PR_SET_THP_DISABLE:
2477 if (arg3 || arg4 || arg5)
2479 if (mmap_write_lock_killable(me->mm))
2482 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2484 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2485 mmap_write_unlock(me->mm);
2487 case PR_MPX_ENABLE_MANAGEMENT:
2488 case PR_MPX_DISABLE_MANAGEMENT:
2489 /* No longer implemented: */
2491 case PR_SET_FP_MODE:
2492 error = SET_FP_MODE(me, arg2);
2494 case PR_GET_FP_MODE:
2495 error = GET_FP_MODE(me);
2498 error = SVE_SET_VL(arg2);
2501 error = SVE_GET_VL();
2503 case PR_GET_SPECULATION_CTRL:
2504 if (arg3 || arg4 || arg5)
2506 error = arch_prctl_spec_ctrl_get(me, arg2);
2508 case PR_SET_SPECULATION_CTRL:
2511 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2513 case PR_PAC_RESET_KEYS:
2514 if (arg3 || arg4 || arg5)
2516 error = PAC_RESET_KEYS(me, arg2);
2518 case PR_PAC_SET_ENABLED_KEYS:
2521 error = PAC_SET_ENABLED_KEYS(me, arg2, arg3);
2523 case PR_PAC_GET_ENABLED_KEYS:
2524 if (arg2 || arg3 || arg4 || arg5)
2526 error = PAC_GET_ENABLED_KEYS(me);
2528 case PR_SET_TAGGED_ADDR_CTRL:
2529 if (arg3 || arg4 || arg5)
2531 error = SET_TAGGED_ADDR_CTRL(arg2);
2533 case PR_GET_TAGGED_ADDR_CTRL:
2534 if (arg2 || arg3 || arg4 || arg5)
2536 error = GET_TAGGED_ADDR_CTRL();
2538 case PR_SET_IO_FLUSHER:
2539 if (!capable(CAP_SYS_RESOURCE))
2542 if (arg3 || arg4 || arg5)
2546 current->flags |= PR_IO_FLUSHER;
2548 current->flags &= ~PR_IO_FLUSHER;
2552 case PR_GET_IO_FLUSHER:
2553 if (!capable(CAP_SYS_RESOURCE))
2556 if (arg2 || arg3 || arg4 || arg5)
2559 error = (current->flags & PR_IO_FLUSHER) == PR_IO_FLUSHER;
2561 case PR_SET_SYSCALL_USER_DISPATCH:
2562 error = set_syscall_user_dispatch(arg2, arg3, arg4,
2563 (char __user *) arg5);
2565 #ifdef CONFIG_SCHED_CORE
2567 error = sched_core_share_pid(arg2, arg3, arg4, arg5);
2577 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2578 struct getcpu_cache __user *, unused)
2581 int cpu = raw_smp_processor_id();
2584 err |= put_user(cpu, cpup);
2586 err |= put_user(cpu_to_node(cpu), nodep);
2587 return err ? -EFAULT : 0;
2591 * do_sysinfo - fill in sysinfo struct
2592 * @info: pointer to buffer to fill
2594 static int do_sysinfo(struct sysinfo *info)
2596 unsigned long mem_total, sav_total;
2597 unsigned int mem_unit, bitcount;
2598 struct timespec64 tp;
2600 memset(info, 0, sizeof(struct sysinfo));
2602 ktime_get_boottime_ts64(&tp);
2603 timens_add_boottime(&tp);
2604 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2606 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2608 info->procs = nr_threads;
2614 * If the sum of all the available memory (i.e. ram + swap)
2615 * is less than can be stored in a 32 bit unsigned long then
2616 * we can be binary compatible with 2.2.x kernels. If not,
2617 * well, in that case 2.2.x was broken anyways...
2619 * -Erik Andersen <andersee@debian.org>
2622 mem_total = info->totalram + info->totalswap;
2623 if (mem_total < info->totalram || mem_total < info->totalswap)
2626 mem_unit = info->mem_unit;
2627 while (mem_unit > 1) {
2630 sav_total = mem_total;
2632 if (mem_total < sav_total)
2637 * If mem_total did not overflow, multiply all memory values by
2638 * info->mem_unit and set it to 1. This leaves things compatible
2639 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2644 info->totalram <<= bitcount;
2645 info->freeram <<= bitcount;
2646 info->sharedram <<= bitcount;
2647 info->bufferram <<= bitcount;
2648 info->totalswap <<= bitcount;
2649 info->freeswap <<= bitcount;
2650 info->totalhigh <<= bitcount;
2651 info->freehigh <<= bitcount;
2657 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2663 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2669 #ifdef CONFIG_COMPAT
2670 struct compat_sysinfo {
2684 char _f[20-2*sizeof(u32)-sizeof(int)];
2687 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2690 struct compat_sysinfo s_32;
2694 /* Check to see if any memory value is too large for 32-bit and scale
2697 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2700 while (s.mem_unit < PAGE_SIZE) {
2705 s.totalram >>= bitcount;
2706 s.freeram >>= bitcount;
2707 s.sharedram >>= bitcount;
2708 s.bufferram >>= bitcount;
2709 s.totalswap >>= bitcount;
2710 s.freeswap >>= bitcount;
2711 s.totalhigh >>= bitcount;
2712 s.freehigh >>= bitcount;
2715 memset(&s_32, 0, sizeof(s_32));
2716 s_32.uptime = s.uptime;
2717 s_32.loads[0] = s.loads[0];
2718 s_32.loads[1] = s.loads[1];
2719 s_32.loads[2] = s.loads[2];
2720 s_32.totalram = s.totalram;
2721 s_32.freeram = s.freeram;
2722 s_32.sharedram = s.sharedram;
2723 s_32.bufferram = s.bufferram;
2724 s_32.totalswap = s.totalswap;
2725 s_32.freeswap = s.freeswap;
2726 s_32.procs = s.procs;
2727 s_32.totalhigh = s.totalhigh;
2728 s_32.freehigh = s.freehigh;
2729 s_32.mem_unit = s.mem_unit;
2730 if (copy_to_user(info, &s_32, sizeof(s_32)))
2734 #endif /* CONFIG_COMPAT */