1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/export.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/binfmts.h>
52 #include <linux/sched.h>
53 #include <linux/sched/autogroup.h>
54 #include <linux/sched/loadavg.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/mm.h>
57 #include <linux/sched/coredump.h>
58 #include <linux/sched/task.h>
59 #include <linux/sched/cputime.h>
60 #include <linux/rcupdate.h>
61 #include <linux/uidgid.h>
62 #include <linux/cred.h>
64 #include <linux/nospec.h>
66 #include <linux/kmsg_dump.h>
67 /* Move somewhere else to avoid recompiling? */
68 #include <generated/utsrelease.h>
70 #include <linux/uaccess.h>
72 #include <asm/unistd.h>
74 #ifndef SET_UNALIGN_CTL
75 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
77 #ifndef GET_UNALIGN_CTL
78 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
81 # define SET_FPEMU_CTL(a, b) (-EINVAL)
84 # define GET_FPEMU_CTL(a, b) (-EINVAL)
87 # define SET_FPEXC_CTL(a, b) (-EINVAL)
90 # define GET_FPEXC_CTL(a, b) (-EINVAL)
93 # define GET_ENDIAN(a, b) (-EINVAL)
96 # define SET_ENDIAN(a, b) (-EINVAL)
99 # define GET_TSC_CTL(a) (-EINVAL)
102 # define SET_TSC_CTL(a) (-EINVAL)
104 #ifndef MPX_ENABLE_MANAGEMENT
105 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
107 #ifndef MPX_DISABLE_MANAGEMENT
108 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
111 # define GET_FP_MODE(a) (-EINVAL)
114 # define SET_FP_MODE(a,b) (-EINVAL)
118 * this is where the system-wide overflow UID and GID are defined, for
119 * architectures that now have 32-bit UID/GID but didn't in the past
122 int overflowuid = DEFAULT_OVERFLOWUID;
123 int overflowgid = DEFAULT_OVERFLOWGID;
125 EXPORT_SYMBOL(overflowuid);
126 EXPORT_SYMBOL(overflowgid);
129 * the same as above, but for filesystems which can only store a 16-bit
130 * UID and GID. as such, this is needed on all architectures
133 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
134 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
136 EXPORT_SYMBOL(fs_overflowuid);
137 EXPORT_SYMBOL(fs_overflowgid);
140 * Returns true if current's euid is same as p's uid or euid,
141 * or has CAP_SYS_NICE to p's user_ns.
143 * Called with rcu_read_lock, creds are safe
145 static bool set_one_prio_perm(struct task_struct *p)
147 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
149 if (uid_eq(pcred->uid, cred->euid) ||
150 uid_eq(pcred->euid, cred->euid))
152 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
158 * set the priority of a task
159 * - the caller must hold the RCU read lock
161 static int set_one_prio(struct task_struct *p, int niceval, int error)
165 if (!set_one_prio_perm(p)) {
169 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
173 no_nice = security_task_setnice(p, niceval);
180 set_user_nice(p, niceval);
185 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
187 struct task_struct *g, *p;
188 struct user_struct *user;
189 const struct cred *cred = current_cred();
194 if (which > PRIO_USER || which < PRIO_PROCESS)
197 /* normalize: avoid signed division (rounding problems) */
199 if (niceval < MIN_NICE)
201 if (niceval > MAX_NICE)
205 read_lock(&tasklist_lock);
209 p = find_task_by_vpid(who);
213 error = set_one_prio(p, niceval, error);
217 pgrp = find_vpid(who);
219 pgrp = task_pgrp(current);
220 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
221 error = set_one_prio(p, niceval, error);
222 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
225 uid = make_kuid(cred->user_ns, who);
229 else if (!uid_eq(uid, cred->uid)) {
230 user = find_user(uid);
232 goto out_unlock; /* No processes for this user */
234 do_each_thread(g, p) {
235 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
236 error = set_one_prio(p, niceval, error);
237 } while_each_thread(g, p);
238 if (!uid_eq(uid, cred->uid))
239 free_uid(user); /* For find_user() */
243 read_unlock(&tasklist_lock);
250 * Ugh. To avoid negative return values, "getpriority()" will
251 * not return the normal nice-value, but a negated value that
252 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
253 * to stay compatible.
255 SYSCALL_DEFINE2(getpriority, int, which, int, who)
257 struct task_struct *g, *p;
258 struct user_struct *user;
259 const struct cred *cred = current_cred();
260 long niceval, retval = -ESRCH;
264 if (which > PRIO_USER || which < PRIO_PROCESS)
268 read_lock(&tasklist_lock);
272 p = find_task_by_vpid(who);
276 niceval = nice_to_rlimit(task_nice(p));
277 if (niceval > retval)
283 pgrp = find_vpid(who);
285 pgrp = task_pgrp(current);
286 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
287 niceval = nice_to_rlimit(task_nice(p));
288 if (niceval > retval)
290 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
293 uid = make_kuid(cred->user_ns, who);
297 else if (!uid_eq(uid, cred->uid)) {
298 user = find_user(uid);
300 goto out_unlock; /* No processes for this user */
302 do_each_thread(g, p) {
303 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
304 niceval = nice_to_rlimit(task_nice(p));
305 if (niceval > retval)
308 } while_each_thread(g, p);
309 if (!uid_eq(uid, cred->uid))
310 free_uid(user); /* for find_user() */
314 read_unlock(&tasklist_lock);
321 * Unprivileged users may change the real gid to the effective gid
322 * or vice versa. (BSD-style)
324 * If you set the real gid at all, or set the effective gid to a value not
325 * equal to the real gid, then the saved gid is set to the new effective gid.
327 * This makes it possible for a setgid program to completely drop its
328 * privileges, which is often a useful assertion to make when you are doing
329 * a security audit over a program.
331 * The general idea is that a program which uses just setregid() will be
332 * 100% compatible with BSD. A program which uses just setgid() will be
333 * 100% compatible with POSIX with saved IDs.
335 * SMP: There are not races, the GIDs are checked only by filesystem
336 * operations (as far as semantic preservation is concerned).
338 #ifdef CONFIG_MULTIUSER
339 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
341 struct user_namespace *ns = current_user_ns();
342 const struct cred *old;
347 krgid = make_kgid(ns, rgid);
348 kegid = make_kgid(ns, egid);
350 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
352 if ((egid != (gid_t) -1) && !gid_valid(kegid))
355 new = prepare_creds();
358 old = current_cred();
361 if (rgid != (gid_t) -1) {
362 if (gid_eq(old->gid, krgid) ||
363 gid_eq(old->egid, krgid) ||
364 ns_capable(old->user_ns, CAP_SETGID))
369 if (egid != (gid_t) -1) {
370 if (gid_eq(old->gid, kegid) ||
371 gid_eq(old->egid, kegid) ||
372 gid_eq(old->sgid, kegid) ||
373 ns_capable(old->user_ns, CAP_SETGID))
379 if (rgid != (gid_t) -1 ||
380 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
381 new->sgid = new->egid;
382 new->fsgid = new->egid;
384 return commit_creds(new);
392 * setgid() is implemented like SysV w/ SAVED_IDS
394 * SMP: Same implicit races as above.
396 SYSCALL_DEFINE1(setgid, gid_t, gid)
398 struct user_namespace *ns = current_user_ns();
399 const struct cred *old;
404 kgid = make_kgid(ns, gid);
405 if (!gid_valid(kgid))
408 new = prepare_creds();
411 old = current_cred();
414 if (ns_capable(old->user_ns, CAP_SETGID))
415 new->gid = new->egid = new->sgid = new->fsgid = kgid;
416 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
417 new->egid = new->fsgid = kgid;
421 return commit_creds(new);
429 * change the user struct in a credentials set to match the new UID
431 static int set_user(struct cred *new)
433 struct user_struct *new_user;
435 new_user = alloc_uid(new->uid);
440 * We don't fail in case of NPROC limit excess here because too many
441 * poorly written programs don't check set*uid() return code, assuming
442 * it never fails if called by root. We may still enforce NPROC limit
443 * for programs doing set*uid()+execve() by harmlessly deferring the
444 * failure to the execve() stage.
446 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
447 new_user != INIT_USER)
448 current->flags |= PF_NPROC_EXCEEDED;
450 current->flags &= ~PF_NPROC_EXCEEDED;
453 new->user = new_user;
458 * Unprivileged users may change the real uid to the effective uid
459 * or vice versa. (BSD-style)
461 * If you set the real uid at all, or set the effective uid to a value not
462 * equal to the real uid, then the saved uid is set to the new effective uid.
464 * This makes it possible for a setuid program to completely drop its
465 * privileges, which is often a useful assertion to make when you are doing
466 * a security audit over a program.
468 * The general idea is that a program which uses just setreuid() will be
469 * 100% compatible with BSD. A program which uses just setuid() will be
470 * 100% compatible with POSIX with saved IDs.
472 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
474 struct user_namespace *ns = current_user_ns();
475 const struct cred *old;
480 kruid = make_kuid(ns, ruid);
481 keuid = make_kuid(ns, euid);
483 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
485 if ((euid != (uid_t) -1) && !uid_valid(keuid))
488 new = prepare_creds();
491 old = current_cred();
494 if (ruid != (uid_t) -1) {
496 if (!uid_eq(old->uid, kruid) &&
497 !uid_eq(old->euid, kruid) &&
498 !ns_capable(old->user_ns, CAP_SETUID))
502 if (euid != (uid_t) -1) {
504 if (!uid_eq(old->uid, keuid) &&
505 !uid_eq(old->euid, keuid) &&
506 !uid_eq(old->suid, keuid) &&
507 !ns_capable(old->user_ns, CAP_SETUID))
511 if (!uid_eq(new->uid, old->uid)) {
512 retval = set_user(new);
516 if (ruid != (uid_t) -1 ||
517 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
518 new->suid = new->euid;
519 new->fsuid = new->euid;
521 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
525 return commit_creds(new);
533 * setuid() is implemented like SysV with SAVED_IDS
535 * Note that SAVED_ID's is deficient in that a setuid root program
536 * like sendmail, for example, cannot set its uid to be a normal
537 * user and then switch back, because if you're root, setuid() sets
538 * the saved uid too. If you don't like this, blame the bright people
539 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
540 * will allow a root program to temporarily drop privileges and be able to
541 * regain them by swapping the real and effective uid.
543 SYSCALL_DEFINE1(setuid, uid_t, uid)
545 struct user_namespace *ns = current_user_ns();
546 const struct cred *old;
551 kuid = make_kuid(ns, uid);
552 if (!uid_valid(kuid))
555 new = prepare_creds();
558 old = current_cred();
561 if (ns_capable(old->user_ns, CAP_SETUID)) {
562 new->suid = new->uid = kuid;
563 if (!uid_eq(kuid, old->uid)) {
564 retval = set_user(new);
568 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
572 new->fsuid = new->euid = kuid;
574 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
578 return commit_creds(new);
587 * This function implements a generic ability to update ruid, euid,
588 * and suid. This allows you to implement the 4.4 compatible seteuid().
590 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
592 struct user_namespace *ns = current_user_ns();
593 const struct cred *old;
596 kuid_t kruid, keuid, ksuid;
598 kruid = make_kuid(ns, ruid);
599 keuid = make_kuid(ns, euid);
600 ksuid = make_kuid(ns, suid);
602 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
605 if ((euid != (uid_t) -1) && !uid_valid(keuid))
608 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
611 new = prepare_creds();
615 old = current_cred();
618 if (!ns_capable(old->user_ns, CAP_SETUID)) {
619 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
620 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
622 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
623 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
625 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
626 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
630 if (ruid != (uid_t) -1) {
632 if (!uid_eq(kruid, old->uid)) {
633 retval = set_user(new);
638 if (euid != (uid_t) -1)
640 if (suid != (uid_t) -1)
642 new->fsuid = new->euid;
644 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
648 return commit_creds(new);
655 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
657 const struct cred *cred = current_cred();
659 uid_t ruid, euid, suid;
661 ruid = from_kuid_munged(cred->user_ns, cred->uid);
662 euid = from_kuid_munged(cred->user_ns, cred->euid);
663 suid = from_kuid_munged(cred->user_ns, cred->suid);
665 retval = put_user(ruid, ruidp);
667 retval = put_user(euid, euidp);
669 return put_user(suid, suidp);
675 * Same as above, but for rgid, egid, sgid.
677 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
679 struct user_namespace *ns = current_user_ns();
680 const struct cred *old;
683 kgid_t krgid, kegid, ksgid;
685 krgid = make_kgid(ns, rgid);
686 kegid = make_kgid(ns, egid);
687 ksgid = make_kgid(ns, sgid);
689 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
691 if ((egid != (gid_t) -1) && !gid_valid(kegid))
693 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
696 new = prepare_creds();
699 old = current_cred();
702 if (!ns_capable(old->user_ns, CAP_SETGID)) {
703 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
704 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
706 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
707 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
709 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
710 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
714 if (rgid != (gid_t) -1)
716 if (egid != (gid_t) -1)
718 if (sgid != (gid_t) -1)
720 new->fsgid = new->egid;
722 return commit_creds(new);
729 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
731 const struct cred *cred = current_cred();
733 gid_t rgid, egid, sgid;
735 rgid = from_kgid_munged(cred->user_ns, cred->gid);
736 egid = from_kgid_munged(cred->user_ns, cred->egid);
737 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
739 retval = put_user(rgid, rgidp);
741 retval = put_user(egid, egidp);
743 retval = put_user(sgid, sgidp);
751 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
752 * is used for "access()" and for the NFS daemon (letting nfsd stay at
753 * whatever uid it wants to). It normally shadows "euid", except when
754 * explicitly set by setfsuid() or for access..
756 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
758 const struct cred *old;
763 old = current_cred();
764 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
766 kuid = make_kuid(old->user_ns, uid);
767 if (!uid_valid(kuid))
770 new = prepare_creds();
774 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
775 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
776 ns_capable(old->user_ns, CAP_SETUID)) {
777 if (!uid_eq(kuid, old->fsuid)) {
779 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
793 * Samma på svenska..
795 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
797 const struct cred *old;
802 old = current_cred();
803 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
805 kgid = make_kgid(old->user_ns, gid);
806 if (!gid_valid(kgid))
809 new = prepare_creds();
813 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
814 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
815 ns_capable(old->user_ns, CAP_SETGID)) {
816 if (!gid_eq(kgid, old->fsgid)) {
829 #endif /* CONFIG_MULTIUSER */
832 * sys_getpid - return the thread group id of the current process
834 * Note, despite the name, this returns the tgid not the pid. The tgid and
835 * the pid are identical unless CLONE_THREAD was specified on clone() in
836 * which case the tgid is the same in all threads of the same group.
838 * This is SMP safe as current->tgid does not change.
840 SYSCALL_DEFINE0(getpid)
842 return task_tgid_vnr(current);
845 /* Thread ID - the internal kernel "pid" */
846 SYSCALL_DEFINE0(gettid)
848 return task_pid_vnr(current);
852 * Accessing ->real_parent is not SMP-safe, it could
853 * change from under us. However, we can use a stale
854 * value of ->real_parent under rcu_read_lock(), see
855 * release_task()->call_rcu(delayed_put_task_struct).
857 SYSCALL_DEFINE0(getppid)
862 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
868 SYSCALL_DEFINE0(getuid)
870 /* Only we change this so SMP safe */
871 return from_kuid_munged(current_user_ns(), current_uid());
874 SYSCALL_DEFINE0(geteuid)
876 /* Only we change this so SMP safe */
877 return from_kuid_munged(current_user_ns(), current_euid());
880 SYSCALL_DEFINE0(getgid)
882 /* Only we change this so SMP safe */
883 return from_kgid_munged(current_user_ns(), current_gid());
886 SYSCALL_DEFINE0(getegid)
888 /* Only we change this so SMP safe */
889 return from_kgid_munged(current_user_ns(), current_egid());
892 static void do_sys_times(struct tms *tms)
894 u64 tgutime, tgstime, cutime, cstime;
896 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
897 cutime = current->signal->cutime;
898 cstime = current->signal->cstime;
899 tms->tms_utime = nsec_to_clock_t(tgutime);
900 tms->tms_stime = nsec_to_clock_t(tgstime);
901 tms->tms_cutime = nsec_to_clock_t(cutime);
902 tms->tms_cstime = nsec_to_clock_t(cstime);
905 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
911 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
914 force_successful_syscall_return();
915 return (long) jiffies_64_to_clock_t(get_jiffies_64());
919 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
921 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
924 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
928 struct compat_tms tmp;
931 /* Convert our struct tms to the compat version. */
932 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
933 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
934 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
935 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
936 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
939 force_successful_syscall_return();
940 return compat_jiffies_to_clock_t(jiffies);
945 * This needs some heavy checking ...
946 * I just haven't the stomach for it. I also don't fully
947 * understand sessions/pgrp etc. Let somebody who does explain it.
949 * OK, I think I have the protection semantics right.... this is really
950 * only important on a multi-user system anyway, to make sure one user
951 * can't send a signal to a process owned by another. -TYT, 12/12/91
953 * !PF_FORKNOEXEC check to conform completely to POSIX.
955 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
957 struct task_struct *p;
958 struct task_struct *group_leader = current->group_leader;
963 pid = task_pid_vnr(group_leader);
970 /* From this point forward we keep holding onto the tasklist lock
971 * so that our parent does not change from under us. -DaveM
973 write_lock_irq(&tasklist_lock);
976 p = find_task_by_vpid(pid);
981 if (!thread_group_leader(p))
984 if (same_thread_group(p->real_parent, group_leader)) {
986 if (task_session(p) != task_session(group_leader))
989 if (!(p->flags & PF_FORKNOEXEC))
993 if (p != group_leader)
998 if (p->signal->leader)
1003 struct task_struct *g;
1005 pgrp = find_vpid(pgid);
1006 g = pid_task(pgrp, PIDTYPE_PGID);
1007 if (!g || task_session(g) != task_session(group_leader))
1011 err = security_task_setpgid(p, pgid);
1015 if (task_pgrp(p) != pgrp)
1016 change_pid(p, PIDTYPE_PGID, pgrp);
1020 /* All paths lead to here, thus we are safe. -DaveM */
1021 write_unlock_irq(&tasklist_lock);
1026 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1028 struct task_struct *p;
1034 grp = task_pgrp(current);
1037 p = find_task_by_vpid(pid);
1044 retval = security_task_getpgid(p);
1048 retval = pid_vnr(grp);
1054 #ifdef __ARCH_WANT_SYS_GETPGRP
1056 SYSCALL_DEFINE0(getpgrp)
1058 return sys_getpgid(0);
1063 SYSCALL_DEFINE1(getsid, pid_t, pid)
1065 struct task_struct *p;
1071 sid = task_session(current);
1074 p = find_task_by_vpid(pid);
1077 sid = task_session(p);
1081 retval = security_task_getsid(p);
1085 retval = pid_vnr(sid);
1091 static void set_special_pids(struct pid *pid)
1093 struct task_struct *curr = current->group_leader;
1095 if (task_session(curr) != pid)
1096 change_pid(curr, PIDTYPE_SID, pid);
1098 if (task_pgrp(curr) != pid)
1099 change_pid(curr, PIDTYPE_PGID, pid);
1102 SYSCALL_DEFINE0(setsid)
1104 struct task_struct *group_leader = current->group_leader;
1105 struct pid *sid = task_pid(group_leader);
1106 pid_t session = pid_vnr(sid);
1109 write_lock_irq(&tasklist_lock);
1110 /* Fail if I am already a session leader */
1111 if (group_leader->signal->leader)
1114 /* Fail if a process group id already exists that equals the
1115 * proposed session id.
1117 if (pid_task(sid, PIDTYPE_PGID))
1120 group_leader->signal->leader = 1;
1121 set_special_pids(sid);
1123 proc_clear_tty(group_leader);
1127 write_unlock_irq(&tasklist_lock);
1129 proc_sid_connector(group_leader);
1130 sched_autogroup_create_attach(group_leader);
1135 DECLARE_RWSEM(uts_sem);
1137 #ifdef COMPAT_UTS_MACHINE
1138 #define override_architecture(name) \
1139 (personality(current->personality) == PER_LINUX32 && \
1140 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1141 sizeof(COMPAT_UTS_MACHINE)))
1143 #define override_architecture(name) 0
1147 * Work around broken programs that cannot handle "Linux 3.0".
1148 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1149 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1151 static int override_release(char __user *release, size_t len)
1155 if (current->personality & UNAME26) {
1156 const char *rest = UTS_RELEASE;
1157 char buf[65] = { 0 };
1163 if (*rest == '.' && ++ndots >= 3)
1165 if (!isdigit(*rest) && *rest != '.')
1169 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1170 copy = clamp_t(size_t, len, 1, sizeof(buf));
1171 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1172 ret = copy_to_user(release, buf, copy + 1);
1177 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1181 down_read(&uts_sem);
1182 if (copy_to_user(name, utsname(), sizeof *name))
1186 if (!errno && override_release(name->release, sizeof(name->release)))
1188 if (!errno && override_architecture(name))
1193 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1197 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1204 down_read(&uts_sem);
1205 if (copy_to_user(name, utsname(), sizeof(*name)))
1209 if (!error && override_release(name->release, sizeof(name->release)))
1211 if (!error && override_architecture(name))
1216 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1222 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1225 down_read(&uts_sem);
1226 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1228 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1229 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1231 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1232 error |= __copy_to_user(&name->release, &utsname()->release,
1234 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1235 error |= __copy_to_user(&name->version, &utsname()->version,
1237 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1238 error |= __copy_to_user(&name->machine, &utsname()->machine,
1240 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1243 if (!error && override_architecture(name))
1245 if (!error && override_release(name->release, sizeof(name->release)))
1247 return error ? -EFAULT : 0;
1251 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1254 char tmp[__NEW_UTS_LEN];
1256 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1259 if (len < 0 || len > __NEW_UTS_LEN)
1261 down_write(&uts_sem);
1263 if (!copy_from_user(tmp, name, len)) {
1264 struct new_utsname *u = utsname();
1266 memcpy(u->nodename, tmp, len);
1267 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1269 uts_proc_notify(UTS_PROC_HOSTNAME);
1275 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1277 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1280 struct new_utsname *u;
1284 down_read(&uts_sem);
1286 i = 1 + strlen(u->nodename);
1290 if (copy_to_user(name, u->nodename, i))
1299 * Only setdomainname; getdomainname can be implemented by calling
1302 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1305 char tmp[__NEW_UTS_LEN];
1307 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1309 if (len < 0 || len > __NEW_UTS_LEN)
1312 down_write(&uts_sem);
1314 if (!copy_from_user(tmp, name, len)) {
1315 struct new_utsname *u = utsname();
1317 memcpy(u->domainname, tmp, len);
1318 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1320 uts_proc_notify(UTS_PROC_DOMAINNAME);
1326 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1328 struct rlimit value;
1331 ret = do_prlimit(current, resource, NULL, &value);
1333 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1338 #ifdef CONFIG_COMPAT
1340 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1341 struct compat_rlimit __user *, rlim)
1344 struct compat_rlimit r32;
1346 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1349 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1350 r.rlim_cur = RLIM_INFINITY;
1352 r.rlim_cur = r32.rlim_cur;
1353 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1354 r.rlim_max = RLIM_INFINITY;
1356 r.rlim_max = r32.rlim_max;
1357 return do_prlimit(current, resource, &r, NULL);
1360 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1361 struct compat_rlimit __user *, rlim)
1366 ret = do_prlimit(current, resource, NULL, &r);
1368 struct compat_rlimit r32;
1369 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1370 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1372 r32.rlim_cur = r.rlim_cur;
1373 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1374 r32.rlim_max = COMPAT_RLIM_INFINITY;
1376 r32.rlim_max = r.rlim_max;
1378 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1386 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1389 * Back compatibility for getrlimit. Needed for some apps.
1391 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1392 struct rlimit __user *, rlim)
1395 if (resource >= RLIM_NLIMITS)
1398 resource = array_index_nospec(resource, RLIM_NLIMITS);
1399 task_lock(current->group_leader);
1400 x = current->signal->rlim[resource];
1401 task_unlock(current->group_leader);
1402 if (x.rlim_cur > 0x7FFFFFFF)
1403 x.rlim_cur = 0x7FFFFFFF;
1404 if (x.rlim_max > 0x7FFFFFFF)
1405 x.rlim_max = 0x7FFFFFFF;
1406 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1409 #ifdef CONFIG_COMPAT
1410 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1411 struct compat_rlimit __user *, rlim)
1415 if (resource >= RLIM_NLIMITS)
1418 resource = array_index_nospec(resource, RLIM_NLIMITS);
1419 task_lock(current->group_leader);
1420 r = current->signal->rlim[resource];
1421 task_unlock(current->group_leader);
1422 if (r.rlim_cur > 0x7FFFFFFF)
1423 r.rlim_cur = 0x7FFFFFFF;
1424 if (r.rlim_max > 0x7FFFFFFF)
1425 r.rlim_max = 0x7FFFFFFF;
1427 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1428 put_user(r.rlim_max, &rlim->rlim_max))
1436 static inline bool rlim64_is_infinity(__u64 rlim64)
1438 #if BITS_PER_LONG < 64
1439 return rlim64 >= ULONG_MAX;
1441 return rlim64 == RLIM64_INFINITY;
1445 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1447 if (rlim->rlim_cur == RLIM_INFINITY)
1448 rlim64->rlim_cur = RLIM64_INFINITY;
1450 rlim64->rlim_cur = rlim->rlim_cur;
1451 if (rlim->rlim_max == RLIM_INFINITY)
1452 rlim64->rlim_max = RLIM64_INFINITY;
1454 rlim64->rlim_max = rlim->rlim_max;
1457 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1459 if (rlim64_is_infinity(rlim64->rlim_cur))
1460 rlim->rlim_cur = RLIM_INFINITY;
1462 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1463 if (rlim64_is_infinity(rlim64->rlim_max))
1464 rlim->rlim_max = RLIM_INFINITY;
1466 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1469 /* make sure you are allowed to change @tsk limits before calling this */
1470 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1471 struct rlimit *new_rlim, struct rlimit *old_rlim)
1473 struct rlimit *rlim;
1476 if (resource >= RLIM_NLIMITS)
1479 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1481 if (resource == RLIMIT_NOFILE &&
1482 new_rlim->rlim_max > sysctl_nr_open)
1486 /* protect tsk->signal and tsk->sighand from disappearing */
1487 read_lock(&tasklist_lock);
1488 if (!tsk->sighand) {
1493 rlim = tsk->signal->rlim + resource;
1494 task_lock(tsk->group_leader);
1496 /* Keep the capable check against init_user_ns until
1497 cgroups can contain all limits */
1498 if (new_rlim->rlim_max > rlim->rlim_max &&
1499 !capable(CAP_SYS_RESOURCE))
1502 retval = security_task_setrlimit(tsk, resource, new_rlim);
1503 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1505 * The caller is asking for an immediate RLIMIT_CPU
1506 * expiry. But we use the zero value to mean "it was
1507 * never set". So let's cheat and make it one second
1510 new_rlim->rlim_cur = 1;
1519 task_unlock(tsk->group_leader);
1522 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1523 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1524 * very long-standing error, and fixing it now risks breakage of
1525 * applications, so we live with it
1527 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1528 new_rlim->rlim_cur != RLIM_INFINITY &&
1529 IS_ENABLED(CONFIG_POSIX_TIMERS))
1530 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1532 read_unlock(&tasklist_lock);
1536 /* rcu lock must be held */
1537 static int check_prlimit_permission(struct task_struct *task,
1540 const struct cred *cred = current_cred(), *tcred;
1543 if (current == task)
1546 tcred = __task_cred(task);
1547 id_match = (uid_eq(cred->uid, tcred->euid) &&
1548 uid_eq(cred->uid, tcred->suid) &&
1549 uid_eq(cred->uid, tcred->uid) &&
1550 gid_eq(cred->gid, tcred->egid) &&
1551 gid_eq(cred->gid, tcred->sgid) &&
1552 gid_eq(cred->gid, tcred->gid));
1553 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1556 return security_task_prlimit(cred, tcred, flags);
1559 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1560 const struct rlimit64 __user *, new_rlim,
1561 struct rlimit64 __user *, old_rlim)
1563 struct rlimit64 old64, new64;
1564 struct rlimit old, new;
1565 struct task_struct *tsk;
1566 unsigned int checkflags = 0;
1570 checkflags |= LSM_PRLIMIT_READ;
1573 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1575 rlim64_to_rlim(&new64, &new);
1576 checkflags |= LSM_PRLIMIT_WRITE;
1580 tsk = pid ? find_task_by_vpid(pid) : current;
1585 ret = check_prlimit_permission(tsk, checkflags);
1590 get_task_struct(tsk);
1593 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1594 old_rlim ? &old : NULL);
1596 if (!ret && old_rlim) {
1597 rlim_to_rlim64(&old, &old64);
1598 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1602 put_task_struct(tsk);
1606 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1608 struct rlimit new_rlim;
1610 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1612 return do_prlimit(current, resource, &new_rlim, NULL);
1616 * It would make sense to put struct rusage in the task_struct,
1617 * except that would make the task_struct be *really big*. After
1618 * task_struct gets moved into malloc'ed memory, it would
1619 * make sense to do this. It will make moving the rest of the information
1620 * a lot simpler! (Which we're not doing right now because we're not
1621 * measuring them yet).
1623 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1624 * races with threads incrementing their own counters. But since word
1625 * reads are atomic, we either get new values or old values and we don't
1626 * care which for the sums. We always take the siglock to protect reading
1627 * the c* fields from p->signal from races with exit.c updating those
1628 * fields when reaping, so a sample either gets all the additions of a
1629 * given child after it's reaped, or none so this sample is before reaping.
1632 * We need to take the siglock for CHILDEREN, SELF and BOTH
1633 * for the cases current multithreaded, non-current single threaded
1634 * non-current multithreaded. Thread traversal is now safe with
1636 * Strictly speaking, we donot need to take the siglock if we are current and
1637 * single threaded, as no one else can take our signal_struct away, no one
1638 * else can reap the children to update signal->c* counters, and no one else
1639 * can race with the signal-> fields. If we do not take any lock, the
1640 * signal-> fields could be read out of order while another thread was just
1641 * exiting. So we should place a read memory barrier when we avoid the lock.
1642 * On the writer side, write memory barrier is implied in __exit_signal
1643 * as __exit_signal releases the siglock spinlock after updating the signal->
1644 * fields. But we don't do this yet to keep things simple.
1648 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1650 r->ru_nvcsw += t->nvcsw;
1651 r->ru_nivcsw += t->nivcsw;
1652 r->ru_minflt += t->min_flt;
1653 r->ru_majflt += t->maj_flt;
1654 r->ru_inblock += task_io_get_inblock(t);
1655 r->ru_oublock += task_io_get_oublock(t);
1658 void getrusage(struct task_struct *p, int who, struct rusage *r)
1660 struct task_struct *t;
1661 unsigned long flags;
1662 u64 tgutime, tgstime, utime, stime;
1663 unsigned long maxrss = 0;
1665 memset((char *)r, 0, sizeof (*r));
1668 if (who == RUSAGE_THREAD) {
1669 task_cputime_adjusted(current, &utime, &stime);
1670 accumulate_thread_rusage(p, r);
1671 maxrss = p->signal->maxrss;
1675 if (!lock_task_sighand(p, &flags))
1680 case RUSAGE_CHILDREN:
1681 utime = p->signal->cutime;
1682 stime = p->signal->cstime;
1683 r->ru_nvcsw = p->signal->cnvcsw;
1684 r->ru_nivcsw = p->signal->cnivcsw;
1685 r->ru_minflt = p->signal->cmin_flt;
1686 r->ru_majflt = p->signal->cmaj_flt;
1687 r->ru_inblock = p->signal->cinblock;
1688 r->ru_oublock = p->signal->coublock;
1689 maxrss = p->signal->cmaxrss;
1691 if (who == RUSAGE_CHILDREN)
1695 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1698 r->ru_nvcsw += p->signal->nvcsw;
1699 r->ru_nivcsw += p->signal->nivcsw;
1700 r->ru_minflt += p->signal->min_flt;
1701 r->ru_majflt += p->signal->maj_flt;
1702 r->ru_inblock += p->signal->inblock;
1703 r->ru_oublock += p->signal->oublock;
1704 if (maxrss < p->signal->maxrss)
1705 maxrss = p->signal->maxrss;
1708 accumulate_thread_rusage(t, r);
1709 } while_each_thread(p, t);
1715 unlock_task_sighand(p, &flags);
1718 r->ru_utime = ns_to_timeval(utime);
1719 r->ru_stime = ns_to_timeval(stime);
1721 if (who != RUSAGE_CHILDREN) {
1722 struct mm_struct *mm = get_task_mm(p);
1725 setmax_mm_hiwater_rss(&maxrss, mm);
1729 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1732 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1736 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1737 who != RUSAGE_THREAD)
1740 getrusage(current, who, &r);
1741 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1744 #ifdef CONFIG_COMPAT
1745 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1749 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1750 who != RUSAGE_THREAD)
1753 getrusage(current, who, &r);
1754 return put_compat_rusage(&r, ru);
1758 SYSCALL_DEFINE1(umask, int, mask)
1760 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1764 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1767 struct file *old_exe, *exe_file;
1768 struct inode *inode;
1775 inode = file_inode(exe.file);
1778 * Because the original mm->exe_file points to executable file, make
1779 * sure that this one is executable as well, to avoid breaking an
1783 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1786 err = inode_permission(inode, MAY_EXEC);
1791 * Forbid mm->exe_file change if old file still mapped.
1793 exe_file = get_mm_exe_file(mm);
1796 struct vm_area_struct *vma;
1798 down_read(&mm->mmap_sem);
1799 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1802 if (path_equal(&vma->vm_file->f_path,
1807 up_read(&mm->mmap_sem);
1812 /* set the new file, lockless */
1814 old_exe = xchg(&mm->exe_file, exe.file);
1821 up_read(&mm->mmap_sem);
1827 * WARNING: we don't require any capability here so be very careful
1828 * in what is allowed for modification from userspace.
1830 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1832 unsigned long mmap_max_addr = TASK_SIZE;
1833 struct mm_struct *mm = current->mm;
1834 int error = -EINVAL, i;
1836 static const unsigned char offsets[] = {
1837 offsetof(struct prctl_mm_map, start_code),
1838 offsetof(struct prctl_mm_map, end_code),
1839 offsetof(struct prctl_mm_map, start_data),
1840 offsetof(struct prctl_mm_map, end_data),
1841 offsetof(struct prctl_mm_map, start_brk),
1842 offsetof(struct prctl_mm_map, brk),
1843 offsetof(struct prctl_mm_map, start_stack),
1844 offsetof(struct prctl_mm_map, arg_start),
1845 offsetof(struct prctl_mm_map, arg_end),
1846 offsetof(struct prctl_mm_map, env_start),
1847 offsetof(struct prctl_mm_map, env_end),
1851 * Make sure the members are not somewhere outside
1852 * of allowed address space.
1854 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1855 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1857 if ((unsigned long)val >= mmap_max_addr ||
1858 (unsigned long)val < mmap_min_addr)
1863 * Make sure the pairs are ordered.
1865 #define __prctl_check_order(__m1, __op, __m2) \
1866 ((unsigned long)prctl_map->__m1 __op \
1867 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1868 error = __prctl_check_order(start_code, <, end_code);
1869 error |= __prctl_check_order(start_data, <, end_data);
1870 error |= __prctl_check_order(start_brk, <=, brk);
1871 error |= __prctl_check_order(arg_start, <=, arg_end);
1872 error |= __prctl_check_order(env_start, <=, env_end);
1875 #undef __prctl_check_order
1880 * @brk should be after @end_data in traditional maps.
1882 if (prctl_map->start_brk <= prctl_map->end_data ||
1883 prctl_map->brk <= prctl_map->end_data)
1887 * Neither we should allow to override limits if they set.
1889 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1890 prctl_map->start_brk, prctl_map->end_data,
1891 prctl_map->start_data))
1895 * Someone is trying to cheat the auxv vector.
1897 if (prctl_map->auxv_size) {
1898 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1903 * Finally, make sure the caller has the rights to
1904 * change /proc/pid/exe link: only local sys admin should
1907 if (prctl_map->exe_fd != (u32)-1) {
1908 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1917 #ifdef CONFIG_CHECKPOINT_RESTORE
1918 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1920 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1921 unsigned long user_auxv[AT_VECTOR_SIZE];
1922 struct mm_struct *mm = current->mm;
1925 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1926 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1928 if (opt == PR_SET_MM_MAP_SIZE)
1929 return put_user((unsigned int)sizeof(prctl_map),
1930 (unsigned int __user *)addr);
1932 if (data_size != sizeof(prctl_map))
1935 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1938 error = validate_prctl_map(&prctl_map);
1942 if (prctl_map.auxv_size) {
1943 memset(user_auxv, 0, sizeof(user_auxv));
1944 if (copy_from_user(user_auxv,
1945 (const void __user *)prctl_map.auxv,
1946 prctl_map.auxv_size))
1949 /* Last entry must be AT_NULL as specification requires */
1950 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1951 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1954 if (prctl_map.exe_fd != (u32)-1) {
1955 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1960 down_write(&mm->mmap_sem);
1963 * We don't validate if these members are pointing to
1964 * real present VMAs because application may have correspond
1965 * VMAs already unmapped and kernel uses these members for statistics
1966 * output in procfs mostly, except
1968 * - @start_brk/@brk which are used in do_brk but kernel lookups
1969 * for VMAs when updating these memvers so anything wrong written
1970 * here cause kernel to swear at userspace program but won't lead
1971 * to any problem in kernel itself
1974 mm->start_code = prctl_map.start_code;
1975 mm->end_code = prctl_map.end_code;
1976 mm->start_data = prctl_map.start_data;
1977 mm->end_data = prctl_map.end_data;
1978 mm->start_brk = prctl_map.start_brk;
1979 mm->brk = prctl_map.brk;
1980 mm->start_stack = prctl_map.start_stack;
1981 mm->arg_start = prctl_map.arg_start;
1982 mm->arg_end = prctl_map.arg_end;
1983 mm->env_start = prctl_map.env_start;
1984 mm->env_end = prctl_map.env_end;
1987 * Note this update of @saved_auxv is lockless thus
1988 * if someone reads this member in procfs while we're
1989 * updating -- it may get partly updated results. It's
1990 * known and acceptable trade off: we leave it as is to
1991 * not introduce additional locks here making the kernel
1994 if (prctl_map.auxv_size)
1995 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1997 up_write(&mm->mmap_sem);
2000 #endif /* CONFIG_CHECKPOINT_RESTORE */
2002 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2006 * This doesn't move the auxiliary vector itself since it's pinned to
2007 * mm_struct, but it permits filling the vector with new values. It's
2008 * up to the caller to provide sane values here, otherwise userspace
2009 * tools which use this vector might be unhappy.
2011 unsigned long user_auxv[AT_VECTOR_SIZE];
2013 if (len > sizeof(user_auxv))
2016 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2019 /* Make sure the last entry is always AT_NULL */
2020 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2021 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2023 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2026 memcpy(mm->saved_auxv, user_auxv, len);
2027 task_unlock(current);
2032 static int prctl_set_mm(int opt, unsigned long addr,
2033 unsigned long arg4, unsigned long arg5)
2035 struct mm_struct *mm = current->mm;
2036 struct prctl_mm_map prctl_map;
2037 struct vm_area_struct *vma;
2040 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2041 opt != PR_SET_MM_MAP &&
2042 opt != PR_SET_MM_MAP_SIZE)))
2045 #ifdef CONFIG_CHECKPOINT_RESTORE
2046 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2047 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2050 if (!capable(CAP_SYS_RESOURCE))
2053 if (opt == PR_SET_MM_EXE_FILE)
2054 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2056 if (opt == PR_SET_MM_AUXV)
2057 return prctl_set_auxv(mm, addr, arg4);
2059 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2064 down_write(&mm->mmap_sem);
2065 vma = find_vma(mm, addr);
2067 prctl_map.start_code = mm->start_code;
2068 prctl_map.end_code = mm->end_code;
2069 prctl_map.start_data = mm->start_data;
2070 prctl_map.end_data = mm->end_data;
2071 prctl_map.start_brk = mm->start_brk;
2072 prctl_map.brk = mm->brk;
2073 prctl_map.start_stack = mm->start_stack;
2074 prctl_map.arg_start = mm->arg_start;
2075 prctl_map.arg_end = mm->arg_end;
2076 prctl_map.env_start = mm->env_start;
2077 prctl_map.env_end = mm->env_end;
2078 prctl_map.auxv = NULL;
2079 prctl_map.auxv_size = 0;
2080 prctl_map.exe_fd = -1;
2083 case PR_SET_MM_START_CODE:
2084 prctl_map.start_code = addr;
2086 case PR_SET_MM_END_CODE:
2087 prctl_map.end_code = addr;
2089 case PR_SET_MM_START_DATA:
2090 prctl_map.start_data = addr;
2092 case PR_SET_MM_END_DATA:
2093 prctl_map.end_data = addr;
2095 case PR_SET_MM_START_STACK:
2096 prctl_map.start_stack = addr;
2098 case PR_SET_MM_START_BRK:
2099 prctl_map.start_brk = addr;
2102 prctl_map.brk = addr;
2104 case PR_SET_MM_ARG_START:
2105 prctl_map.arg_start = addr;
2107 case PR_SET_MM_ARG_END:
2108 prctl_map.arg_end = addr;
2110 case PR_SET_MM_ENV_START:
2111 prctl_map.env_start = addr;
2113 case PR_SET_MM_ENV_END:
2114 prctl_map.env_end = addr;
2120 error = validate_prctl_map(&prctl_map);
2126 * If command line arguments and environment
2127 * are placed somewhere else on stack, we can
2128 * set them up here, ARG_START/END to setup
2129 * command line argumets and ENV_START/END
2132 case PR_SET_MM_START_STACK:
2133 case PR_SET_MM_ARG_START:
2134 case PR_SET_MM_ARG_END:
2135 case PR_SET_MM_ENV_START:
2136 case PR_SET_MM_ENV_END:
2143 mm->start_code = prctl_map.start_code;
2144 mm->end_code = prctl_map.end_code;
2145 mm->start_data = prctl_map.start_data;
2146 mm->end_data = prctl_map.end_data;
2147 mm->start_brk = prctl_map.start_brk;
2148 mm->brk = prctl_map.brk;
2149 mm->start_stack = prctl_map.start_stack;
2150 mm->arg_start = prctl_map.arg_start;
2151 mm->arg_end = prctl_map.arg_end;
2152 mm->env_start = prctl_map.env_start;
2153 mm->env_end = prctl_map.env_end;
2157 up_write(&mm->mmap_sem);
2161 #ifdef CONFIG_CHECKPOINT_RESTORE
2162 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2164 return put_user(me->clear_child_tid, tid_addr);
2167 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2173 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2176 * If task has has_child_subreaper - all its decendants
2177 * already have these flag too and new decendants will
2178 * inherit it on fork, skip them.
2180 * If we've found child_reaper - skip descendants in
2181 * it's subtree as they will never get out pidns.
2183 if (p->signal->has_child_subreaper ||
2184 is_child_reaper(task_pid(p)))
2187 p->signal->has_child_subreaper = 1;
2191 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2196 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2202 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2203 unsigned long, arg4, unsigned long, arg5)
2205 struct task_struct *me = current;
2206 unsigned char comm[sizeof(me->comm)];
2209 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2210 if (error != -ENOSYS)
2215 case PR_SET_PDEATHSIG:
2216 if (!valid_signal(arg2)) {
2220 me->pdeath_signal = arg2;
2222 case PR_GET_PDEATHSIG:
2223 error = put_user(me->pdeath_signal, (int __user *)arg2);
2225 case PR_GET_DUMPABLE:
2226 error = get_dumpable(me->mm);
2228 case PR_SET_DUMPABLE:
2229 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2233 set_dumpable(me->mm, arg2);
2236 case PR_SET_UNALIGN:
2237 error = SET_UNALIGN_CTL(me, arg2);
2239 case PR_GET_UNALIGN:
2240 error = GET_UNALIGN_CTL(me, arg2);
2243 error = SET_FPEMU_CTL(me, arg2);
2246 error = GET_FPEMU_CTL(me, arg2);
2249 error = SET_FPEXC_CTL(me, arg2);
2252 error = GET_FPEXC_CTL(me, arg2);
2255 error = PR_TIMING_STATISTICAL;
2258 if (arg2 != PR_TIMING_STATISTICAL)
2262 comm[sizeof(me->comm) - 1] = 0;
2263 if (strncpy_from_user(comm, (char __user *)arg2,
2264 sizeof(me->comm) - 1) < 0)
2266 set_task_comm(me, comm);
2267 proc_comm_connector(me);
2270 get_task_comm(comm, me);
2271 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2275 error = GET_ENDIAN(me, arg2);
2278 error = SET_ENDIAN(me, arg2);
2280 case PR_GET_SECCOMP:
2281 error = prctl_get_seccomp();
2283 case PR_SET_SECCOMP:
2284 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2287 error = GET_TSC_CTL(arg2);
2290 error = SET_TSC_CTL(arg2);
2292 case PR_TASK_PERF_EVENTS_DISABLE:
2293 error = perf_event_task_disable();
2295 case PR_TASK_PERF_EVENTS_ENABLE:
2296 error = perf_event_task_enable();
2298 case PR_GET_TIMERSLACK:
2299 if (current->timer_slack_ns > ULONG_MAX)
2302 error = current->timer_slack_ns;
2304 case PR_SET_TIMERSLACK:
2306 current->timer_slack_ns =
2307 current->default_timer_slack_ns;
2309 current->timer_slack_ns = arg2;
2315 case PR_MCE_KILL_CLEAR:
2318 current->flags &= ~PF_MCE_PROCESS;
2320 case PR_MCE_KILL_SET:
2321 current->flags |= PF_MCE_PROCESS;
2322 if (arg3 == PR_MCE_KILL_EARLY)
2323 current->flags |= PF_MCE_EARLY;
2324 else if (arg3 == PR_MCE_KILL_LATE)
2325 current->flags &= ~PF_MCE_EARLY;
2326 else if (arg3 == PR_MCE_KILL_DEFAULT)
2328 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2336 case PR_MCE_KILL_GET:
2337 if (arg2 | arg3 | arg4 | arg5)
2339 if (current->flags & PF_MCE_PROCESS)
2340 error = (current->flags & PF_MCE_EARLY) ?
2341 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2343 error = PR_MCE_KILL_DEFAULT;
2346 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2348 case PR_GET_TID_ADDRESS:
2349 error = prctl_get_tid_address(me, (int __user **)arg2);
2351 case PR_SET_CHILD_SUBREAPER:
2352 me->signal->is_child_subreaper = !!arg2;
2356 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2358 case PR_GET_CHILD_SUBREAPER:
2359 error = put_user(me->signal->is_child_subreaper,
2360 (int __user *)arg2);
2362 case PR_SET_NO_NEW_PRIVS:
2363 if (arg2 != 1 || arg3 || arg4 || arg5)
2366 task_set_no_new_privs(current);
2368 case PR_GET_NO_NEW_PRIVS:
2369 if (arg2 || arg3 || arg4 || arg5)
2371 return task_no_new_privs(current) ? 1 : 0;
2372 case PR_GET_THP_DISABLE:
2373 if (arg2 || arg3 || arg4 || arg5)
2375 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2377 case PR_SET_THP_DISABLE:
2378 if (arg3 || arg4 || arg5)
2380 if (down_write_killable(&me->mm->mmap_sem))
2383 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2385 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2386 up_write(&me->mm->mmap_sem);
2388 case PR_MPX_ENABLE_MANAGEMENT:
2389 if (arg2 || arg3 || arg4 || arg5)
2391 error = MPX_ENABLE_MANAGEMENT();
2393 case PR_MPX_DISABLE_MANAGEMENT:
2394 if (arg2 || arg3 || arg4 || arg5)
2396 error = MPX_DISABLE_MANAGEMENT();
2398 case PR_SET_FP_MODE:
2399 error = SET_FP_MODE(me, arg2);
2401 case PR_GET_FP_MODE:
2402 error = GET_FP_MODE(me);
2404 case PR_GET_SPECULATION_CTRL:
2405 if (arg3 || arg4 || arg5)
2407 error = arch_prctl_spec_ctrl_get(me, arg2);
2409 case PR_SET_SPECULATION_CTRL:
2412 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2421 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2422 struct getcpu_cache __user *, unused)
2425 int cpu = raw_smp_processor_id();
2428 err |= put_user(cpu, cpup);
2430 err |= put_user(cpu_to_node(cpu), nodep);
2431 return err ? -EFAULT : 0;
2435 * do_sysinfo - fill in sysinfo struct
2436 * @info: pointer to buffer to fill
2438 static int do_sysinfo(struct sysinfo *info)
2440 unsigned long mem_total, sav_total;
2441 unsigned int mem_unit, bitcount;
2444 memset(info, 0, sizeof(struct sysinfo));
2446 get_monotonic_boottime(&tp);
2447 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2449 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2451 info->procs = nr_threads;
2457 * If the sum of all the available memory (i.e. ram + swap)
2458 * is less than can be stored in a 32 bit unsigned long then
2459 * we can be binary compatible with 2.2.x kernels. If not,
2460 * well, in that case 2.2.x was broken anyways...
2462 * -Erik Andersen <andersee@debian.org>
2465 mem_total = info->totalram + info->totalswap;
2466 if (mem_total < info->totalram || mem_total < info->totalswap)
2469 mem_unit = info->mem_unit;
2470 while (mem_unit > 1) {
2473 sav_total = mem_total;
2475 if (mem_total < sav_total)
2480 * If mem_total did not overflow, multiply all memory values by
2481 * info->mem_unit and set it to 1. This leaves things compatible
2482 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2487 info->totalram <<= bitcount;
2488 info->freeram <<= bitcount;
2489 info->sharedram <<= bitcount;
2490 info->bufferram <<= bitcount;
2491 info->totalswap <<= bitcount;
2492 info->freeswap <<= bitcount;
2493 info->totalhigh <<= bitcount;
2494 info->freehigh <<= bitcount;
2500 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2506 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2512 #ifdef CONFIG_COMPAT
2513 struct compat_sysinfo {
2527 char _f[20-2*sizeof(u32)-sizeof(int)];
2530 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2536 /* Check to see if any memory value is too large for 32-bit and scale
2539 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2542 while (s.mem_unit < PAGE_SIZE) {
2547 s.totalram >>= bitcount;
2548 s.freeram >>= bitcount;
2549 s.sharedram >>= bitcount;
2550 s.bufferram >>= bitcount;
2551 s.totalswap >>= bitcount;
2552 s.freeswap >>= bitcount;
2553 s.totalhigh >>= bitcount;
2554 s.freehigh >>= bitcount;
2557 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2558 __put_user(s.uptime, &info->uptime) ||
2559 __put_user(s.loads[0], &info->loads[0]) ||
2560 __put_user(s.loads[1], &info->loads[1]) ||
2561 __put_user(s.loads[2], &info->loads[2]) ||
2562 __put_user(s.totalram, &info->totalram) ||
2563 __put_user(s.freeram, &info->freeram) ||
2564 __put_user(s.sharedram, &info->sharedram) ||
2565 __put_user(s.bufferram, &info->bufferram) ||
2566 __put_user(s.totalswap, &info->totalswap) ||
2567 __put_user(s.freeswap, &info->freeswap) ||
2568 __put_user(s.procs, &info->procs) ||
2569 __put_user(s.totalhigh, &info->totalhigh) ||
2570 __put_user(s.freehigh, &info->freehigh) ||
2571 __put_user(s.mem_unit, &info->mem_unit))
2576 #endif /* CONFIG_COMPAT */