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/kmsg_dump.h>
65 /* Move somewhere else to avoid recompiling? */
66 #include <generated/utsrelease.h>
68 #include <linux/uaccess.h>
70 #include <asm/unistd.h>
72 #ifndef SET_UNALIGN_CTL
73 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
75 #ifndef GET_UNALIGN_CTL
76 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
79 # define SET_FPEMU_CTL(a, b) (-EINVAL)
82 # define GET_FPEMU_CTL(a, b) (-EINVAL)
85 # define SET_FPEXC_CTL(a, b) (-EINVAL)
88 # define GET_FPEXC_CTL(a, b) (-EINVAL)
91 # define GET_ENDIAN(a, b) (-EINVAL)
94 # define SET_ENDIAN(a, b) (-EINVAL)
97 # define GET_TSC_CTL(a) (-EINVAL)
100 # define SET_TSC_CTL(a) (-EINVAL)
102 #ifndef MPX_ENABLE_MANAGEMENT
103 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
105 #ifndef MPX_DISABLE_MANAGEMENT
106 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
109 # define GET_FP_MODE(a) (-EINVAL)
112 # define SET_FP_MODE(a,b) (-EINVAL)
116 * this is where the system-wide overflow UID and GID are defined, for
117 * architectures that now have 32-bit UID/GID but didn't in the past
120 int overflowuid = DEFAULT_OVERFLOWUID;
121 int overflowgid = DEFAULT_OVERFLOWGID;
123 EXPORT_SYMBOL(overflowuid);
124 EXPORT_SYMBOL(overflowgid);
127 * the same as above, but for filesystems which can only store a 16-bit
128 * UID and GID. as such, this is needed on all architectures
131 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
132 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
134 EXPORT_SYMBOL(fs_overflowuid);
135 EXPORT_SYMBOL(fs_overflowgid);
138 * Returns true if current's euid is same as p's uid or euid,
139 * or has CAP_SYS_NICE to p's user_ns.
141 * Called with rcu_read_lock, creds are safe
143 static bool set_one_prio_perm(struct task_struct *p)
145 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
147 if (uid_eq(pcred->uid, cred->euid) ||
148 uid_eq(pcred->euid, cred->euid))
150 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
156 * set the priority of a task
157 * - the caller must hold the RCU read lock
159 static int set_one_prio(struct task_struct *p, int niceval, int error)
163 if (!set_one_prio_perm(p)) {
167 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
171 no_nice = security_task_setnice(p, niceval);
178 set_user_nice(p, niceval);
183 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
185 struct task_struct *g, *p;
186 struct user_struct *user;
187 const struct cred *cred = current_cred();
192 if (which > PRIO_USER || which < PRIO_PROCESS)
195 /* normalize: avoid signed division (rounding problems) */
197 if (niceval < MIN_NICE)
199 if (niceval > MAX_NICE)
203 read_lock(&tasklist_lock);
207 p = find_task_by_vpid(who);
211 error = set_one_prio(p, niceval, error);
215 pgrp = find_vpid(who);
217 pgrp = task_pgrp(current);
218 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
219 error = set_one_prio(p, niceval, error);
220 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
223 uid = make_kuid(cred->user_ns, who);
227 else if (!uid_eq(uid, cred->uid)) {
228 user = find_user(uid);
230 goto out_unlock; /* No processes for this user */
232 do_each_thread(g, p) {
233 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
234 error = set_one_prio(p, niceval, error);
235 } while_each_thread(g, p);
236 if (!uid_eq(uid, cred->uid))
237 free_uid(user); /* For find_user() */
241 read_unlock(&tasklist_lock);
248 * Ugh. To avoid negative return values, "getpriority()" will
249 * not return the normal nice-value, but a negated value that
250 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
251 * to stay compatible.
253 SYSCALL_DEFINE2(getpriority, int, which, int, who)
255 struct task_struct *g, *p;
256 struct user_struct *user;
257 const struct cred *cred = current_cred();
258 long niceval, retval = -ESRCH;
262 if (which > PRIO_USER || which < PRIO_PROCESS)
266 read_lock(&tasklist_lock);
270 p = find_task_by_vpid(who);
274 niceval = nice_to_rlimit(task_nice(p));
275 if (niceval > retval)
281 pgrp = find_vpid(who);
283 pgrp = task_pgrp(current);
284 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
285 niceval = nice_to_rlimit(task_nice(p));
286 if (niceval > retval)
288 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
291 uid = make_kuid(cred->user_ns, who);
295 else if (!uid_eq(uid, cred->uid)) {
296 user = find_user(uid);
298 goto out_unlock; /* No processes for this user */
300 do_each_thread(g, p) {
301 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
302 niceval = nice_to_rlimit(task_nice(p));
303 if (niceval > retval)
306 } while_each_thread(g, p);
307 if (!uid_eq(uid, cred->uid))
308 free_uid(user); /* for find_user() */
312 read_unlock(&tasklist_lock);
319 * Unprivileged users may change the real gid to the effective gid
320 * or vice versa. (BSD-style)
322 * If you set the real gid at all, or set the effective gid to a value not
323 * equal to the real gid, then the saved gid is set to the new effective gid.
325 * This makes it possible for a setgid program to completely drop its
326 * privileges, which is often a useful assertion to make when you are doing
327 * a security audit over a program.
329 * The general idea is that a program which uses just setregid() will be
330 * 100% compatible with BSD. A program which uses just setgid() will be
331 * 100% compatible with POSIX with saved IDs.
333 * SMP: There are not races, the GIDs are checked only by filesystem
334 * operations (as far as semantic preservation is concerned).
336 #ifdef CONFIG_MULTIUSER
337 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
339 struct user_namespace *ns = current_user_ns();
340 const struct cred *old;
345 krgid = make_kgid(ns, rgid);
346 kegid = make_kgid(ns, egid);
348 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
350 if ((egid != (gid_t) -1) && !gid_valid(kegid))
353 new = prepare_creds();
356 old = current_cred();
359 if (rgid != (gid_t) -1) {
360 if (gid_eq(old->gid, krgid) ||
361 gid_eq(old->egid, krgid) ||
362 ns_capable(old->user_ns, CAP_SETGID))
367 if (egid != (gid_t) -1) {
368 if (gid_eq(old->gid, kegid) ||
369 gid_eq(old->egid, kegid) ||
370 gid_eq(old->sgid, kegid) ||
371 ns_capable(old->user_ns, CAP_SETGID))
377 if (rgid != (gid_t) -1 ||
378 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
379 new->sgid = new->egid;
380 new->fsgid = new->egid;
382 return commit_creds(new);
390 * setgid() is implemented like SysV w/ SAVED_IDS
392 * SMP: Same implicit races as above.
394 SYSCALL_DEFINE1(setgid, gid_t, gid)
396 struct user_namespace *ns = current_user_ns();
397 const struct cred *old;
402 kgid = make_kgid(ns, gid);
403 if (!gid_valid(kgid))
406 new = prepare_creds();
409 old = current_cred();
412 if (ns_capable(old->user_ns, CAP_SETGID))
413 new->gid = new->egid = new->sgid = new->fsgid = kgid;
414 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
415 new->egid = new->fsgid = kgid;
419 return commit_creds(new);
427 * change the user struct in a credentials set to match the new UID
429 static int set_user(struct cred *new)
431 struct user_struct *new_user;
433 new_user = alloc_uid(new->uid);
438 * We don't fail in case of NPROC limit excess here because too many
439 * poorly written programs don't check set*uid() return code, assuming
440 * it never fails if called by root. We may still enforce NPROC limit
441 * for programs doing set*uid()+execve() by harmlessly deferring the
442 * failure to the execve() stage.
444 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
445 new_user != INIT_USER)
446 current->flags |= PF_NPROC_EXCEEDED;
448 current->flags &= ~PF_NPROC_EXCEEDED;
451 new->user = new_user;
456 * Unprivileged users may change the real uid to the effective uid
457 * or vice versa. (BSD-style)
459 * If you set the real uid at all, or set the effective uid to a value not
460 * equal to the real uid, then the saved uid is set to the new effective uid.
462 * This makes it possible for a setuid program to completely drop its
463 * privileges, which is often a useful assertion to make when you are doing
464 * a security audit over a program.
466 * The general idea is that a program which uses just setreuid() will be
467 * 100% compatible with BSD. A program which uses just setuid() will be
468 * 100% compatible with POSIX with saved IDs.
470 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
472 struct user_namespace *ns = current_user_ns();
473 const struct cred *old;
478 kruid = make_kuid(ns, ruid);
479 keuid = make_kuid(ns, euid);
481 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
483 if ((euid != (uid_t) -1) && !uid_valid(keuid))
486 new = prepare_creds();
489 old = current_cred();
492 if (ruid != (uid_t) -1) {
494 if (!uid_eq(old->uid, kruid) &&
495 !uid_eq(old->euid, kruid) &&
496 !ns_capable(old->user_ns, CAP_SETUID))
500 if (euid != (uid_t) -1) {
502 if (!uid_eq(old->uid, keuid) &&
503 !uid_eq(old->euid, keuid) &&
504 !uid_eq(old->suid, keuid) &&
505 !ns_capable(old->user_ns, CAP_SETUID))
509 if (!uid_eq(new->uid, old->uid)) {
510 retval = set_user(new);
514 if (ruid != (uid_t) -1 ||
515 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
516 new->suid = new->euid;
517 new->fsuid = new->euid;
519 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
523 return commit_creds(new);
531 * setuid() is implemented like SysV with SAVED_IDS
533 * Note that SAVED_ID's is deficient in that a setuid root program
534 * like sendmail, for example, cannot set its uid to be a normal
535 * user and then switch back, because if you're root, setuid() sets
536 * the saved uid too. If you don't like this, blame the bright people
537 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
538 * will allow a root program to temporarily drop privileges and be able to
539 * regain them by swapping the real and effective uid.
541 SYSCALL_DEFINE1(setuid, uid_t, uid)
543 struct user_namespace *ns = current_user_ns();
544 const struct cred *old;
549 kuid = make_kuid(ns, uid);
550 if (!uid_valid(kuid))
553 new = prepare_creds();
556 old = current_cred();
559 if (ns_capable(old->user_ns, CAP_SETUID)) {
560 new->suid = new->uid = kuid;
561 if (!uid_eq(kuid, old->uid)) {
562 retval = set_user(new);
566 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
570 new->fsuid = new->euid = kuid;
572 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
576 return commit_creds(new);
585 * This function implements a generic ability to update ruid, euid,
586 * and suid. This allows you to implement the 4.4 compatible seteuid().
588 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
590 struct user_namespace *ns = current_user_ns();
591 const struct cred *old;
594 kuid_t kruid, keuid, ksuid;
596 kruid = make_kuid(ns, ruid);
597 keuid = make_kuid(ns, euid);
598 ksuid = make_kuid(ns, suid);
600 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
603 if ((euid != (uid_t) -1) && !uid_valid(keuid))
606 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
609 new = prepare_creds();
613 old = current_cred();
616 if (!ns_capable(old->user_ns, CAP_SETUID)) {
617 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
618 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
620 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
621 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
623 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
624 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
628 if (ruid != (uid_t) -1) {
630 if (!uid_eq(kruid, old->uid)) {
631 retval = set_user(new);
636 if (euid != (uid_t) -1)
638 if (suid != (uid_t) -1)
640 new->fsuid = new->euid;
642 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
646 return commit_creds(new);
653 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
655 const struct cred *cred = current_cred();
657 uid_t ruid, euid, suid;
659 ruid = from_kuid_munged(cred->user_ns, cred->uid);
660 euid = from_kuid_munged(cred->user_ns, cred->euid);
661 suid = from_kuid_munged(cred->user_ns, cred->suid);
663 retval = put_user(ruid, ruidp);
665 retval = put_user(euid, euidp);
667 return put_user(suid, suidp);
673 * Same as above, but for rgid, egid, sgid.
675 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
677 struct user_namespace *ns = current_user_ns();
678 const struct cred *old;
681 kgid_t krgid, kegid, ksgid;
683 krgid = make_kgid(ns, rgid);
684 kegid = make_kgid(ns, egid);
685 ksgid = make_kgid(ns, sgid);
687 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
689 if ((egid != (gid_t) -1) && !gid_valid(kegid))
691 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
694 new = prepare_creds();
697 old = current_cred();
700 if (!ns_capable(old->user_ns, CAP_SETGID)) {
701 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
702 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
704 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
705 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
707 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
708 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
712 if (rgid != (gid_t) -1)
714 if (egid != (gid_t) -1)
716 if (sgid != (gid_t) -1)
718 new->fsgid = new->egid;
720 return commit_creds(new);
727 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
729 const struct cred *cred = current_cred();
731 gid_t rgid, egid, sgid;
733 rgid = from_kgid_munged(cred->user_ns, cred->gid);
734 egid = from_kgid_munged(cred->user_ns, cred->egid);
735 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
737 retval = put_user(rgid, rgidp);
739 retval = put_user(egid, egidp);
741 retval = put_user(sgid, sgidp);
749 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
750 * is used for "access()" and for the NFS daemon (letting nfsd stay at
751 * whatever uid it wants to). It normally shadows "euid", except when
752 * explicitly set by setfsuid() or for access..
754 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
756 const struct cred *old;
761 old = current_cred();
762 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
764 kuid = make_kuid(old->user_ns, uid);
765 if (!uid_valid(kuid))
768 new = prepare_creds();
772 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
773 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
774 ns_capable(old->user_ns, CAP_SETUID)) {
775 if (!uid_eq(kuid, old->fsuid)) {
777 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
791 * Samma på svenska..
793 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
795 const struct cred *old;
800 old = current_cred();
801 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
803 kgid = make_kgid(old->user_ns, gid);
804 if (!gid_valid(kgid))
807 new = prepare_creds();
811 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
812 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
813 ns_capable(old->user_ns, CAP_SETGID)) {
814 if (!gid_eq(kgid, old->fsgid)) {
827 #endif /* CONFIG_MULTIUSER */
830 * sys_getpid - return the thread group id of the current process
832 * Note, despite the name, this returns the tgid not the pid. The tgid and
833 * the pid are identical unless CLONE_THREAD was specified on clone() in
834 * which case the tgid is the same in all threads of the same group.
836 * This is SMP safe as current->tgid does not change.
838 SYSCALL_DEFINE0(getpid)
840 return task_tgid_vnr(current);
843 /* Thread ID - the internal kernel "pid" */
844 SYSCALL_DEFINE0(gettid)
846 return task_pid_vnr(current);
850 * Accessing ->real_parent is not SMP-safe, it could
851 * change from under us. However, we can use a stale
852 * value of ->real_parent under rcu_read_lock(), see
853 * release_task()->call_rcu(delayed_put_task_struct).
855 SYSCALL_DEFINE0(getppid)
860 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
866 SYSCALL_DEFINE0(getuid)
868 /* Only we change this so SMP safe */
869 return from_kuid_munged(current_user_ns(), current_uid());
872 SYSCALL_DEFINE0(geteuid)
874 /* Only we change this so SMP safe */
875 return from_kuid_munged(current_user_ns(), current_euid());
878 SYSCALL_DEFINE0(getgid)
880 /* Only we change this so SMP safe */
881 return from_kgid_munged(current_user_ns(), current_gid());
884 SYSCALL_DEFINE0(getegid)
886 /* Only we change this so SMP safe */
887 return from_kgid_munged(current_user_ns(), current_egid());
890 static void do_sys_times(struct tms *tms)
892 u64 tgutime, tgstime, cutime, cstime;
894 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
895 cutime = current->signal->cutime;
896 cstime = current->signal->cstime;
897 tms->tms_utime = nsec_to_clock_t(tgutime);
898 tms->tms_stime = nsec_to_clock_t(tgstime);
899 tms->tms_cutime = nsec_to_clock_t(cutime);
900 tms->tms_cstime = nsec_to_clock_t(cstime);
903 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
909 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
912 force_successful_syscall_return();
913 return (long) jiffies_64_to_clock_t(get_jiffies_64());
917 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
919 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
922 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
926 struct compat_tms tmp;
929 /* Convert our struct tms to the compat version. */
930 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
931 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
932 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
933 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
934 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
937 force_successful_syscall_return();
938 return compat_jiffies_to_clock_t(jiffies);
943 * This needs some heavy checking ...
944 * I just haven't the stomach for it. I also don't fully
945 * understand sessions/pgrp etc. Let somebody who does explain it.
947 * OK, I think I have the protection semantics right.... this is really
948 * only important on a multi-user system anyway, to make sure one user
949 * can't send a signal to a process owned by another. -TYT, 12/12/91
951 * !PF_FORKNOEXEC check to conform completely to POSIX.
953 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
955 struct task_struct *p;
956 struct task_struct *group_leader = current->group_leader;
961 pid = task_pid_vnr(group_leader);
968 /* From this point forward we keep holding onto the tasklist lock
969 * so that our parent does not change from under us. -DaveM
971 write_lock_irq(&tasklist_lock);
974 p = find_task_by_vpid(pid);
979 if (!thread_group_leader(p))
982 if (same_thread_group(p->real_parent, group_leader)) {
984 if (task_session(p) != task_session(group_leader))
987 if (!(p->flags & PF_FORKNOEXEC))
991 if (p != group_leader)
996 if (p->signal->leader)
1001 struct task_struct *g;
1003 pgrp = find_vpid(pgid);
1004 g = pid_task(pgrp, PIDTYPE_PGID);
1005 if (!g || task_session(g) != task_session(group_leader))
1009 err = security_task_setpgid(p, pgid);
1013 if (task_pgrp(p) != pgrp)
1014 change_pid(p, PIDTYPE_PGID, pgrp);
1018 /* All paths lead to here, thus we are safe. -DaveM */
1019 write_unlock_irq(&tasklist_lock);
1024 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1026 struct task_struct *p;
1032 grp = task_pgrp(current);
1035 p = find_task_by_vpid(pid);
1042 retval = security_task_getpgid(p);
1046 retval = pid_vnr(grp);
1052 #ifdef __ARCH_WANT_SYS_GETPGRP
1054 SYSCALL_DEFINE0(getpgrp)
1056 return sys_getpgid(0);
1061 SYSCALL_DEFINE1(getsid, pid_t, pid)
1063 struct task_struct *p;
1069 sid = task_session(current);
1072 p = find_task_by_vpid(pid);
1075 sid = task_session(p);
1079 retval = security_task_getsid(p);
1083 retval = pid_vnr(sid);
1089 static void set_special_pids(struct pid *pid)
1091 struct task_struct *curr = current->group_leader;
1093 if (task_session(curr) != pid)
1094 change_pid(curr, PIDTYPE_SID, pid);
1096 if (task_pgrp(curr) != pid)
1097 change_pid(curr, PIDTYPE_PGID, pid);
1100 SYSCALL_DEFINE0(setsid)
1102 struct task_struct *group_leader = current->group_leader;
1103 struct pid *sid = task_pid(group_leader);
1104 pid_t session = pid_vnr(sid);
1107 write_lock_irq(&tasklist_lock);
1108 /* Fail if I am already a session leader */
1109 if (group_leader->signal->leader)
1112 /* Fail if a process group id already exists that equals the
1113 * proposed session id.
1115 if (pid_task(sid, PIDTYPE_PGID))
1118 group_leader->signal->leader = 1;
1119 set_special_pids(sid);
1121 proc_clear_tty(group_leader);
1125 write_unlock_irq(&tasklist_lock);
1127 proc_sid_connector(group_leader);
1128 sched_autogroup_create_attach(group_leader);
1133 DECLARE_RWSEM(uts_sem);
1135 #ifdef COMPAT_UTS_MACHINE
1136 #define override_architecture(name) \
1137 (personality(current->personality) == PER_LINUX32 && \
1138 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1139 sizeof(COMPAT_UTS_MACHINE)))
1141 #define override_architecture(name) 0
1145 * Work around broken programs that cannot handle "Linux 3.0".
1146 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1147 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1149 static int override_release(char __user *release, size_t len)
1153 if (current->personality & UNAME26) {
1154 const char *rest = UTS_RELEASE;
1155 char buf[65] = { 0 };
1161 if (*rest == '.' && ++ndots >= 3)
1163 if (!isdigit(*rest) && *rest != '.')
1167 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1168 copy = clamp_t(size_t, len, 1, sizeof(buf));
1169 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1170 ret = copy_to_user(release, buf, copy + 1);
1175 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1179 down_read(&uts_sem);
1180 if (copy_to_user(name, utsname(), sizeof *name))
1184 if (!errno && override_release(name->release, sizeof(name->release)))
1186 if (!errno && override_architecture(name))
1191 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1195 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1202 down_read(&uts_sem);
1203 if (copy_to_user(name, utsname(), sizeof(*name)))
1207 if (!error && override_release(name->release, sizeof(name->release)))
1209 if (!error && override_architecture(name))
1214 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1220 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1223 down_read(&uts_sem);
1224 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1226 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1227 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1229 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1230 error |= __copy_to_user(&name->release, &utsname()->release,
1232 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1233 error |= __copy_to_user(&name->version, &utsname()->version,
1235 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1236 error |= __copy_to_user(&name->machine, &utsname()->machine,
1238 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1241 if (!error && override_architecture(name))
1243 if (!error && override_release(name->release, sizeof(name->release)))
1245 return error ? -EFAULT : 0;
1249 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1252 char tmp[__NEW_UTS_LEN];
1254 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1257 if (len < 0 || len > __NEW_UTS_LEN)
1259 down_write(&uts_sem);
1261 if (!copy_from_user(tmp, name, len)) {
1262 struct new_utsname *u = utsname();
1264 memcpy(u->nodename, tmp, len);
1265 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1267 uts_proc_notify(UTS_PROC_HOSTNAME);
1273 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1275 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1278 struct new_utsname *u;
1282 down_read(&uts_sem);
1284 i = 1 + strlen(u->nodename);
1288 if (copy_to_user(name, u->nodename, i))
1297 * Only setdomainname; getdomainname can be implemented by calling
1300 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1303 char tmp[__NEW_UTS_LEN];
1305 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1307 if (len < 0 || len > __NEW_UTS_LEN)
1310 down_write(&uts_sem);
1312 if (!copy_from_user(tmp, name, len)) {
1313 struct new_utsname *u = utsname();
1315 memcpy(u->domainname, tmp, len);
1316 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1318 uts_proc_notify(UTS_PROC_DOMAINNAME);
1324 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1326 struct rlimit value;
1329 ret = do_prlimit(current, resource, NULL, &value);
1331 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1336 #ifdef CONFIG_COMPAT
1338 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1339 struct compat_rlimit __user *, rlim)
1342 struct compat_rlimit r32;
1344 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1347 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1348 r.rlim_cur = RLIM_INFINITY;
1350 r.rlim_cur = r32.rlim_cur;
1351 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1352 r.rlim_max = RLIM_INFINITY;
1354 r.rlim_max = r32.rlim_max;
1355 return do_prlimit(current, resource, &r, NULL);
1358 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1359 struct compat_rlimit __user *, rlim)
1364 ret = do_prlimit(current, resource, NULL, &r);
1366 struct compat_rlimit r32;
1367 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1368 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1370 r32.rlim_cur = r.rlim_cur;
1371 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1372 r32.rlim_max = COMPAT_RLIM_INFINITY;
1374 r32.rlim_max = r.rlim_max;
1376 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1384 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1387 * Back compatibility for getrlimit. Needed for some apps.
1389 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1390 struct rlimit __user *, rlim)
1393 if (resource >= RLIM_NLIMITS)
1396 task_lock(current->group_leader);
1397 x = current->signal->rlim[resource];
1398 task_unlock(current->group_leader);
1399 if (x.rlim_cur > 0x7FFFFFFF)
1400 x.rlim_cur = 0x7FFFFFFF;
1401 if (x.rlim_max > 0x7FFFFFFF)
1402 x.rlim_max = 0x7FFFFFFF;
1403 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1406 #ifdef CONFIG_COMPAT
1407 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1408 struct compat_rlimit __user *, rlim)
1412 if (resource >= RLIM_NLIMITS)
1415 task_lock(current->group_leader);
1416 r = current->signal->rlim[resource];
1417 task_unlock(current->group_leader);
1418 if (r.rlim_cur > 0x7FFFFFFF)
1419 r.rlim_cur = 0x7FFFFFFF;
1420 if (r.rlim_max > 0x7FFFFFFF)
1421 r.rlim_max = 0x7FFFFFFF;
1423 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1424 put_user(r.rlim_max, &rlim->rlim_max))
1432 static inline bool rlim64_is_infinity(__u64 rlim64)
1434 #if BITS_PER_LONG < 64
1435 return rlim64 >= ULONG_MAX;
1437 return rlim64 == RLIM64_INFINITY;
1441 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1443 if (rlim->rlim_cur == RLIM_INFINITY)
1444 rlim64->rlim_cur = RLIM64_INFINITY;
1446 rlim64->rlim_cur = rlim->rlim_cur;
1447 if (rlim->rlim_max == RLIM_INFINITY)
1448 rlim64->rlim_max = RLIM64_INFINITY;
1450 rlim64->rlim_max = rlim->rlim_max;
1453 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1455 if (rlim64_is_infinity(rlim64->rlim_cur))
1456 rlim->rlim_cur = RLIM_INFINITY;
1458 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1459 if (rlim64_is_infinity(rlim64->rlim_max))
1460 rlim->rlim_max = RLIM_INFINITY;
1462 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1465 /* make sure you are allowed to change @tsk limits before calling this */
1466 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1467 struct rlimit *new_rlim, struct rlimit *old_rlim)
1469 struct rlimit *rlim;
1472 if (resource >= RLIM_NLIMITS)
1475 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1477 if (resource == RLIMIT_NOFILE &&
1478 new_rlim->rlim_max > sysctl_nr_open)
1482 /* protect tsk->signal and tsk->sighand from disappearing */
1483 read_lock(&tasklist_lock);
1484 if (!tsk->sighand) {
1489 rlim = tsk->signal->rlim + resource;
1490 task_lock(tsk->group_leader);
1492 /* Keep the capable check against init_user_ns until
1493 cgroups can contain all limits */
1494 if (new_rlim->rlim_max > rlim->rlim_max &&
1495 !capable(CAP_SYS_RESOURCE))
1498 retval = security_task_setrlimit(tsk, resource, new_rlim);
1499 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1501 * The caller is asking for an immediate RLIMIT_CPU
1502 * expiry. But we use the zero value to mean "it was
1503 * never set". So let's cheat and make it one second
1506 new_rlim->rlim_cur = 1;
1515 task_unlock(tsk->group_leader);
1518 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1519 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1520 * very long-standing error, and fixing it now risks breakage of
1521 * applications, so we live with it
1523 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1524 new_rlim->rlim_cur != RLIM_INFINITY &&
1525 IS_ENABLED(CONFIG_POSIX_TIMERS))
1526 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1528 read_unlock(&tasklist_lock);
1532 /* rcu lock must be held */
1533 static int check_prlimit_permission(struct task_struct *task,
1536 const struct cred *cred = current_cred(), *tcred;
1539 if (current == task)
1542 tcred = __task_cred(task);
1543 id_match = (uid_eq(cred->uid, tcred->euid) &&
1544 uid_eq(cred->uid, tcred->suid) &&
1545 uid_eq(cred->uid, tcred->uid) &&
1546 gid_eq(cred->gid, tcred->egid) &&
1547 gid_eq(cred->gid, tcred->sgid) &&
1548 gid_eq(cred->gid, tcred->gid));
1549 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1552 return security_task_prlimit(cred, tcred, flags);
1555 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1556 const struct rlimit64 __user *, new_rlim,
1557 struct rlimit64 __user *, old_rlim)
1559 struct rlimit64 old64, new64;
1560 struct rlimit old, new;
1561 struct task_struct *tsk;
1562 unsigned int checkflags = 0;
1566 checkflags |= LSM_PRLIMIT_READ;
1569 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1571 rlim64_to_rlim(&new64, &new);
1572 checkflags |= LSM_PRLIMIT_WRITE;
1576 tsk = pid ? find_task_by_vpid(pid) : current;
1581 ret = check_prlimit_permission(tsk, checkflags);
1586 get_task_struct(tsk);
1589 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1590 old_rlim ? &old : NULL);
1592 if (!ret && old_rlim) {
1593 rlim_to_rlim64(&old, &old64);
1594 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1598 put_task_struct(tsk);
1602 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1604 struct rlimit new_rlim;
1606 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1608 return do_prlimit(current, resource, &new_rlim, NULL);
1612 * It would make sense to put struct rusage in the task_struct,
1613 * except that would make the task_struct be *really big*. After
1614 * task_struct gets moved into malloc'ed memory, it would
1615 * make sense to do this. It will make moving the rest of the information
1616 * a lot simpler! (Which we're not doing right now because we're not
1617 * measuring them yet).
1619 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1620 * races with threads incrementing their own counters. But since word
1621 * reads are atomic, we either get new values or old values and we don't
1622 * care which for the sums. We always take the siglock to protect reading
1623 * the c* fields from p->signal from races with exit.c updating those
1624 * fields when reaping, so a sample either gets all the additions of a
1625 * given child after it's reaped, or none so this sample is before reaping.
1628 * We need to take the siglock for CHILDEREN, SELF and BOTH
1629 * for the cases current multithreaded, non-current single threaded
1630 * non-current multithreaded. Thread traversal is now safe with
1632 * Strictly speaking, we donot need to take the siglock if we are current and
1633 * single threaded, as no one else can take our signal_struct away, no one
1634 * else can reap the children to update signal->c* counters, and no one else
1635 * can race with the signal-> fields. If we do not take any lock, the
1636 * signal-> fields could be read out of order while another thread was just
1637 * exiting. So we should place a read memory barrier when we avoid the lock.
1638 * On the writer side, write memory barrier is implied in __exit_signal
1639 * as __exit_signal releases the siglock spinlock after updating the signal->
1640 * fields. But we don't do this yet to keep things simple.
1644 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1646 r->ru_nvcsw += t->nvcsw;
1647 r->ru_nivcsw += t->nivcsw;
1648 r->ru_minflt += t->min_flt;
1649 r->ru_majflt += t->maj_flt;
1650 r->ru_inblock += task_io_get_inblock(t);
1651 r->ru_oublock += task_io_get_oublock(t);
1654 void getrusage(struct task_struct *p, int who, struct rusage *r)
1656 struct task_struct *t;
1657 unsigned long flags;
1658 u64 tgutime, tgstime, utime, stime;
1659 unsigned long maxrss = 0;
1661 memset((char *)r, 0, sizeof (*r));
1664 if (who == RUSAGE_THREAD) {
1665 task_cputime_adjusted(current, &utime, &stime);
1666 accumulate_thread_rusage(p, r);
1667 maxrss = p->signal->maxrss;
1671 if (!lock_task_sighand(p, &flags))
1676 case RUSAGE_CHILDREN:
1677 utime = p->signal->cutime;
1678 stime = p->signal->cstime;
1679 r->ru_nvcsw = p->signal->cnvcsw;
1680 r->ru_nivcsw = p->signal->cnivcsw;
1681 r->ru_minflt = p->signal->cmin_flt;
1682 r->ru_majflt = p->signal->cmaj_flt;
1683 r->ru_inblock = p->signal->cinblock;
1684 r->ru_oublock = p->signal->coublock;
1685 maxrss = p->signal->cmaxrss;
1687 if (who == RUSAGE_CHILDREN)
1691 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1694 r->ru_nvcsw += p->signal->nvcsw;
1695 r->ru_nivcsw += p->signal->nivcsw;
1696 r->ru_minflt += p->signal->min_flt;
1697 r->ru_majflt += p->signal->maj_flt;
1698 r->ru_inblock += p->signal->inblock;
1699 r->ru_oublock += p->signal->oublock;
1700 if (maxrss < p->signal->maxrss)
1701 maxrss = p->signal->maxrss;
1704 accumulate_thread_rusage(t, r);
1705 } while_each_thread(p, t);
1711 unlock_task_sighand(p, &flags);
1714 r->ru_utime = ns_to_timeval(utime);
1715 r->ru_stime = ns_to_timeval(stime);
1717 if (who != RUSAGE_CHILDREN) {
1718 struct mm_struct *mm = get_task_mm(p);
1721 setmax_mm_hiwater_rss(&maxrss, mm);
1725 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1728 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1732 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1733 who != RUSAGE_THREAD)
1736 getrusage(current, who, &r);
1737 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1740 #ifdef CONFIG_COMPAT
1741 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1745 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1746 who != RUSAGE_THREAD)
1749 getrusage(current, who, &r);
1750 return put_compat_rusage(&r, ru);
1754 SYSCALL_DEFINE1(umask, int, mask)
1756 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1760 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1763 struct file *old_exe, *exe_file;
1764 struct inode *inode;
1771 inode = file_inode(exe.file);
1774 * Because the original mm->exe_file points to executable file, make
1775 * sure that this one is executable as well, to avoid breaking an
1779 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1782 err = inode_permission(inode, MAY_EXEC);
1787 * Forbid mm->exe_file change if old file still mapped.
1789 exe_file = get_mm_exe_file(mm);
1792 struct vm_area_struct *vma;
1794 down_read(&mm->mmap_sem);
1795 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1798 if (path_equal(&vma->vm_file->f_path,
1803 up_read(&mm->mmap_sem);
1808 /* set the new file, lockless */
1810 old_exe = xchg(&mm->exe_file, exe.file);
1817 up_read(&mm->mmap_sem);
1823 * WARNING: we don't require any capability here so be very careful
1824 * in what is allowed for modification from userspace.
1826 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1828 unsigned long mmap_max_addr = TASK_SIZE;
1829 struct mm_struct *mm = current->mm;
1830 int error = -EINVAL, i;
1832 static const unsigned char offsets[] = {
1833 offsetof(struct prctl_mm_map, start_code),
1834 offsetof(struct prctl_mm_map, end_code),
1835 offsetof(struct prctl_mm_map, start_data),
1836 offsetof(struct prctl_mm_map, end_data),
1837 offsetof(struct prctl_mm_map, start_brk),
1838 offsetof(struct prctl_mm_map, brk),
1839 offsetof(struct prctl_mm_map, start_stack),
1840 offsetof(struct prctl_mm_map, arg_start),
1841 offsetof(struct prctl_mm_map, arg_end),
1842 offsetof(struct prctl_mm_map, env_start),
1843 offsetof(struct prctl_mm_map, env_end),
1847 * Make sure the members are not somewhere outside
1848 * of allowed address space.
1850 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1851 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1853 if ((unsigned long)val >= mmap_max_addr ||
1854 (unsigned long)val < mmap_min_addr)
1859 * Make sure the pairs are ordered.
1861 #define __prctl_check_order(__m1, __op, __m2) \
1862 ((unsigned long)prctl_map->__m1 __op \
1863 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1864 error = __prctl_check_order(start_code, <, end_code);
1865 error |= __prctl_check_order(start_data, <, end_data);
1866 error |= __prctl_check_order(start_brk, <=, brk);
1867 error |= __prctl_check_order(arg_start, <=, arg_end);
1868 error |= __prctl_check_order(env_start, <=, env_end);
1871 #undef __prctl_check_order
1876 * @brk should be after @end_data in traditional maps.
1878 if (prctl_map->start_brk <= prctl_map->end_data ||
1879 prctl_map->brk <= prctl_map->end_data)
1883 * Neither we should allow to override limits if they set.
1885 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1886 prctl_map->start_brk, prctl_map->end_data,
1887 prctl_map->start_data))
1891 * Someone is trying to cheat the auxv vector.
1893 if (prctl_map->auxv_size) {
1894 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1899 * Finally, make sure the caller has the rights to
1900 * change /proc/pid/exe link: only local sys admin should
1903 if (prctl_map->exe_fd != (u32)-1) {
1904 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1913 #ifdef CONFIG_CHECKPOINT_RESTORE
1914 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1916 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1917 unsigned long user_auxv[AT_VECTOR_SIZE];
1918 struct mm_struct *mm = current->mm;
1921 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1922 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1924 if (opt == PR_SET_MM_MAP_SIZE)
1925 return put_user((unsigned int)sizeof(prctl_map),
1926 (unsigned int __user *)addr);
1928 if (data_size != sizeof(prctl_map))
1931 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1934 error = validate_prctl_map(&prctl_map);
1938 if (prctl_map.auxv_size) {
1939 memset(user_auxv, 0, sizeof(user_auxv));
1940 if (copy_from_user(user_auxv,
1941 (const void __user *)prctl_map.auxv,
1942 prctl_map.auxv_size))
1945 /* Last entry must be AT_NULL as specification requires */
1946 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1947 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1950 if (prctl_map.exe_fd != (u32)-1) {
1951 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1956 down_write(&mm->mmap_sem);
1959 * We don't validate if these members are pointing to
1960 * real present VMAs because application may have correspond
1961 * VMAs already unmapped and kernel uses these members for statistics
1962 * output in procfs mostly, except
1964 * - @start_brk/@brk which are used in do_brk but kernel lookups
1965 * for VMAs when updating these memvers so anything wrong written
1966 * here cause kernel to swear at userspace program but won't lead
1967 * to any problem in kernel itself
1970 mm->start_code = prctl_map.start_code;
1971 mm->end_code = prctl_map.end_code;
1972 mm->start_data = prctl_map.start_data;
1973 mm->end_data = prctl_map.end_data;
1974 mm->start_brk = prctl_map.start_brk;
1975 mm->brk = prctl_map.brk;
1976 mm->start_stack = prctl_map.start_stack;
1977 mm->arg_start = prctl_map.arg_start;
1978 mm->arg_end = prctl_map.arg_end;
1979 mm->env_start = prctl_map.env_start;
1980 mm->env_end = prctl_map.env_end;
1983 * Note this update of @saved_auxv is lockless thus
1984 * if someone reads this member in procfs while we're
1985 * updating -- it may get partly updated results. It's
1986 * known and acceptable trade off: we leave it as is to
1987 * not introduce additional locks here making the kernel
1990 if (prctl_map.auxv_size)
1991 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1993 up_write(&mm->mmap_sem);
1996 #endif /* CONFIG_CHECKPOINT_RESTORE */
1998 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2002 * This doesn't move the auxiliary vector itself since it's pinned to
2003 * mm_struct, but it permits filling the vector with new values. It's
2004 * up to the caller to provide sane values here, otherwise userspace
2005 * tools which use this vector might be unhappy.
2007 unsigned long user_auxv[AT_VECTOR_SIZE];
2009 if (len > sizeof(user_auxv))
2012 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2015 /* Make sure the last entry is always AT_NULL */
2016 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2017 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2019 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2022 memcpy(mm->saved_auxv, user_auxv, len);
2023 task_unlock(current);
2028 static int prctl_set_mm(int opt, unsigned long addr,
2029 unsigned long arg4, unsigned long arg5)
2031 struct mm_struct *mm = current->mm;
2032 struct prctl_mm_map prctl_map;
2033 struct vm_area_struct *vma;
2036 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2037 opt != PR_SET_MM_MAP &&
2038 opt != PR_SET_MM_MAP_SIZE)))
2041 #ifdef CONFIG_CHECKPOINT_RESTORE
2042 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2043 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2046 if (!capable(CAP_SYS_RESOURCE))
2049 if (opt == PR_SET_MM_EXE_FILE)
2050 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2052 if (opt == PR_SET_MM_AUXV)
2053 return prctl_set_auxv(mm, addr, arg4);
2055 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2060 down_write(&mm->mmap_sem);
2061 vma = find_vma(mm, addr);
2063 prctl_map.start_code = mm->start_code;
2064 prctl_map.end_code = mm->end_code;
2065 prctl_map.start_data = mm->start_data;
2066 prctl_map.end_data = mm->end_data;
2067 prctl_map.start_brk = mm->start_brk;
2068 prctl_map.brk = mm->brk;
2069 prctl_map.start_stack = mm->start_stack;
2070 prctl_map.arg_start = mm->arg_start;
2071 prctl_map.arg_end = mm->arg_end;
2072 prctl_map.env_start = mm->env_start;
2073 prctl_map.env_end = mm->env_end;
2074 prctl_map.auxv = NULL;
2075 prctl_map.auxv_size = 0;
2076 prctl_map.exe_fd = -1;
2079 case PR_SET_MM_START_CODE:
2080 prctl_map.start_code = addr;
2082 case PR_SET_MM_END_CODE:
2083 prctl_map.end_code = addr;
2085 case PR_SET_MM_START_DATA:
2086 prctl_map.start_data = addr;
2088 case PR_SET_MM_END_DATA:
2089 prctl_map.end_data = addr;
2091 case PR_SET_MM_START_STACK:
2092 prctl_map.start_stack = addr;
2094 case PR_SET_MM_START_BRK:
2095 prctl_map.start_brk = addr;
2098 prctl_map.brk = addr;
2100 case PR_SET_MM_ARG_START:
2101 prctl_map.arg_start = addr;
2103 case PR_SET_MM_ARG_END:
2104 prctl_map.arg_end = addr;
2106 case PR_SET_MM_ENV_START:
2107 prctl_map.env_start = addr;
2109 case PR_SET_MM_ENV_END:
2110 prctl_map.env_end = addr;
2116 error = validate_prctl_map(&prctl_map);
2122 * If command line arguments and environment
2123 * are placed somewhere else on stack, we can
2124 * set them up here, ARG_START/END to setup
2125 * command line argumets and ENV_START/END
2128 case PR_SET_MM_START_STACK:
2129 case PR_SET_MM_ARG_START:
2130 case PR_SET_MM_ARG_END:
2131 case PR_SET_MM_ENV_START:
2132 case PR_SET_MM_ENV_END:
2139 mm->start_code = prctl_map.start_code;
2140 mm->end_code = prctl_map.end_code;
2141 mm->start_data = prctl_map.start_data;
2142 mm->end_data = prctl_map.end_data;
2143 mm->start_brk = prctl_map.start_brk;
2144 mm->brk = prctl_map.brk;
2145 mm->start_stack = prctl_map.start_stack;
2146 mm->arg_start = prctl_map.arg_start;
2147 mm->arg_end = prctl_map.arg_end;
2148 mm->env_start = prctl_map.env_start;
2149 mm->env_end = prctl_map.env_end;
2153 up_write(&mm->mmap_sem);
2157 #ifdef CONFIG_CHECKPOINT_RESTORE
2158 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2160 return put_user(me->clear_child_tid, tid_addr);
2163 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2169 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2172 * If task has has_child_subreaper - all its decendants
2173 * already have these flag too and new decendants will
2174 * inherit it on fork, skip them.
2176 * If we've found child_reaper - skip descendants in
2177 * it's subtree as they will never get out pidns.
2179 if (p->signal->has_child_subreaper ||
2180 is_child_reaper(task_pid(p)))
2183 p->signal->has_child_subreaper = 1;
2187 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2188 unsigned long, arg4, unsigned long, arg5)
2190 struct task_struct *me = current;
2191 unsigned char comm[sizeof(me->comm)];
2194 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2195 if (error != -ENOSYS)
2200 case PR_SET_PDEATHSIG:
2201 if (!valid_signal(arg2)) {
2205 me->pdeath_signal = arg2;
2207 case PR_GET_PDEATHSIG:
2208 error = put_user(me->pdeath_signal, (int __user *)arg2);
2210 case PR_GET_DUMPABLE:
2211 error = get_dumpable(me->mm);
2213 case PR_SET_DUMPABLE:
2214 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2218 set_dumpable(me->mm, arg2);
2221 case PR_SET_UNALIGN:
2222 error = SET_UNALIGN_CTL(me, arg2);
2224 case PR_GET_UNALIGN:
2225 error = GET_UNALIGN_CTL(me, arg2);
2228 error = SET_FPEMU_CTL(me, arg2);
2231 error = GET_FPEMU_CTL(me, arg2);
2234 error = SET_FPEXC_CTL(me, arg2);
2237 error = GET_FPEXC_CTL(me, arg2);
2240 error = PR_TIMING_STATISTICAL;
2243 if (arg2 != PR_TIMING_STATISTICAL)
2247 comm[sizeof(me->comm) - 1] = 0;
2248 if (strncpy_from_user(comm, (char __user *)arg2,
2249 sizeof(me->comm) - 1) < 0)
2251 set_task_comm(me, comm);
2252 proc_comm_connector(me);
2255 get_task_comm(comm, me);
2256 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2260 error = GET_ENDIAN(me, arg2);
2263 error = SET_ENDIAN(me, arg2);
2265 case PR_GET_SECCOMP:
2266 error = prctl_get_seccomp();
2268 case PR_SET_SECCOMP:
2269 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2272 error = GET_TSC_CTL(arg2);
2275 error = SET_TSC_CTL(arg2);
2277 case PR_TASK_PERF_EVENTS_DISABLE:
2278 error = perf_event_task_disable();
2280 case PR_TASK_PERF_EVENTS_ENABLE:
2281 error = perf_event_task_enable();
2283 case PR_GET_TIMERSLACK:
2284 if (current->timer_slack_ns > ULONG_MAX)
2287 error = current->timer_slack_ns;
2289 case PR_SET_TIMERSLACK:
2291 current->timer_slack_ns =
2292 current->default_timer_slack_ns;
2294 current->timer_slack_ns = arg2;
2300 case PR_MCE_KILL_CLEAR:
2303 current->flags &= ~PF_MCE_PROCESS;
2305 case PR_MCE_KILL_SET:
2306 current->flags |= PF_MCE_PROCESS;
2307 if (arg3 == PR_MCE_KILL_EARLY)
2308 current->flags |= PF_MCE_EARLY;
2309 else if (arg3 == PR_MCE_KILL_LATE)
2310 current->flags &= ~PF_MCE_EARLY;
2311 else if (arg3 == PR_MCE_KILL_DEFAULT)
2313 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2321 case PR_MCE_KILL_GET:
2322 if (arg2 | arg3 | arg4 | arg5)
2324 if (current->flags & PF_MCE_PROCESS)
2325 error = (current->flags & PF_MCE_EARLY) ?
2326 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2328 error = PR_MCE_KILL_DEFAULT;
2331 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2333 case PR_GET_TID_ADDRESS:
2334 error = prctl_get_tid_address(me, (int __user **)arg2);
2336 case PR_SET_CHILD_SUBREAPER:
2337 me->signal->is_child_subreaper = !!arg2;
2341 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2343 case PR_GET_CHILD_SUBREAPER:
2344 error = put_user(me->signal->is_child_subreaper,
2345 (int __user *)arg2);
2347 case PR_SET_NO_NEW_PRIVS:
2348 if (arg2 != 1 || arg3 || arg4 || arg5)
2351 task_set_no_new_privs(current);
2353 case PR_GET_NO_NEW_PRIVS:
2354 if (arg2 || arg3 || arg4 || arg5)
2356 return task_no_new_privs(current) ? 1 : 0;
2357 case PR_GET_THP_DISABLE:
2358 if (arg2 || arg3 || arg4 || arg5)
2360 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2362 case PR_SET_THP_DISABLE:
2363 if (arg3 || arg4 || arg5)
2365 if (down_write_killable(&me->mm->mmap_sem))
2368 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2370 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2371 up_write(&me->mm->mmap_sem);
2373 case PR_MPX_ENABLE_MANAGEMENT:
2374 if (arg2 || arg3 || arg4 || arg5)
2376 error = MPX_ENABLE_MANAGEMENT();
2378 case PR_MPX_DISABLE_MANAGEMENT:
2379 if (arg2 || arg3 || arg4 || arg5)
2381 error = MPX_DISABLE_MANAGEMENT();
2383 case PR_SET_FP_MODE:
2384 error = SET_FP_MODE(me, arg2);
2386 case PR_GET_FP_MODE:
2387 error = GET_FP_MODE(me);
2396 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2397 struct getcpu_cache __user *, unused)
2400 int cpu = raw_smp_processor_id();
2403 err |= put_user(cpu, cpup);
2405 err |= put_user(cpu_to_node(cpu), nodep);
2406 return err ? -EFAULT : 0;
2410 * do_sysinfo - fill in sysinfo struct
2411 * @info: pointer to buffer to fill
2413 static int do_sysinfo(struct sysinfo *info)
2415 unsigned long mem_total, sav_total;
2416 unsigned int mem_unit, bitcount;
2419 memset(info, 0, sizeof(struct sysinfo));
2421 get_monotonic_boottime(&tp);
2422 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2424 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2426 info->procs = nr_threads;
2432 * If the sum of all the available memory (i.e. ram + swap)
2433 * is less than can be stored in a 32 bit unsigned long then
2434 * we can be binary compatible with 2.2.x kernels. If not,
2435 * well, in that case 2.2.x was broken anyways...
2437 * -Erik Andersen <andersee@debian.org>
2440 mem_total = info->totalram + info->totalswap;
2441 if (mem_total < info->totalram || mem_total < info->totalswap)
2444 mem_unit = info->mem_unit;
2445 while (mem_unit > 1) {
2448 sav_total = mem_total;
2450 if (mem_total < sav_total)
2455 * If mem_total did not overflow, multiply all memory values by
2456 * info->mem_unit and set it to 1. This leaves things compatible
2457 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2462 info->totalram <<= bitcount;
2463 info->freeram <<= bitcount;
2464 info->sharedram <<= bitcount;
2465 info->bufferram <<= bitcount;
2466 info->totalswap <<= bitcount;
2467 info->freeswap <<= bitcount;
2468 info->totalhigh <<= bitcount;
2469 info->freehigh <<= bitcount;
2475 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2481 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2487 #ifdef CONFIG_COMPAT
2488 struct compat_sysinfo {
2502 char _f[20-2*sizeof(u32)-sizeof(int)];
2505 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2511 /* Check to see if any memory value is too large for 32-bit and scale
2514 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2517 while (s.mem_unit < PAGE_SIZE) {
2522 s.totalram >>= bitcount;
2523 s.freeram >>= bitcount;
2524 s.sharedram >>= bitcount;
2525 s.bufferram >>= bitcount;
2526 s.totalswap >>= bitcount;
2527 s.freeswap >>= bitcount;
2528 s.totalhigh >>= bitcount;
2529 s.freehigh >>= bitcount;
2532 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2533 __put_user(s.uptime, &info->uptime) ||
2534 __put_user(s.loads[0], &info->loads[0]) ||
2535 __put_user(s.loads[1], &info->loads[1]) ||
2536 __put_user(s.loads[2], &info->loads[2]) ||
2537 __put_user(s.totalram, &info->totalram) ||
2538 __put_user(s.freeram, &info->freeram) ||
2539 __put_user(s.sharedram, &info->sharedram) ||
2540 __put_user(s.bufferram, &info->bufferram) ||
2541 __put_user(s.totalswap, &info->totalswap) ||
2542 __put_user(s.freeswap, &info->freeswap) ||
2543 __put_user(s.procs, &info->procs) ||
2544 __put_user(s.totalhigh, &info->totalhigh) ||
2545 __put_user(s.freehigh, &info->freehigh) ||
2546 __put_user(s.mem_unit, &info->mem_unit))
2551 #endif /* CONFIG_COMPAT */