4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/export.h>
9 #include <linux/utsname.h>
10 #include <linux/mman.h>
11 #include <linux/reboot.h>
12 #include <linux/prctl.h>
13 #include <linux/highuid.h>
15 #include <linux/kmod.h>
16 #include <linux/perf_event.h>
17 #include <linux/resource.h>
18 #include <linux/kernel.h>
19 #include <linux/workqueue.h>
20 #include <linux/capability.h>
21 #include <linux/device.h>
22 #include <linux/key.h>
23 #include <linux/times.h>
24 #include <linux/posix-timers.h>
25 #include <linux/security.h>
26 #include <linux/dcookies.h>
27 #include <linux/suspend.h>
28 #include <linux/tty.h>
29 #include <linux/signal.h>
30 #include <linux/cn_proc.h>
31 #include <linux/getcpu.h>
32 #include <linux/task_io_accounting_ops.h>
33 #include <linux/seccomp.h>
34 #include <linux/cpu.h>
35 #include <linux/personality.h>
36 #include <linux/ptrace.h>
37 #include <linux/fs_struct.h>
38 #include <linux/file.h>
39 #include <linux/mount.h>
40 #include <linux/gfp.h>
41 #include <linux/syscore_ops.h>
42 #include <linux/version.h>
43 #include <linux/ctype.h>
45 #include <linux/compat.h>
46 #include <linux/syscalls.h>
47 #include <linux/kprobes.h>
48 #include <linux/user_namespace.h>
49 #include <linux/binfmts.h>
51 #include <linux/sched.h>
52 #include <linux/rcupdate.h>
53 #include <linux/uidgid.h>
54 #include <linux/cred.h>
56 #include <linux/kmsg_dump.h>
57 /* Move somewhere else to avoid recompiling? */
58 #include <generated/utsrelease.h>
60 #include <asm/uaccess.h>
62 #include <asm/unistd.h>
64 #ifndef SET_UNALIGN_CTL
65 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
67 #ifndef GET_UNALIGN_CTL
68 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
71 # define SET_FPEMU_CTL(a, b) (-EINVAL)
74 # define GET_FPEMU_CTL(a, b) (-EINVAL)
77 # define SET_FPEXC_CTL(a, b) (-EINVAL)
80 # define GET_FPEXC_CTL(a, b) (-EINVAL)
83 # define GET_ENDIAN(a, b) (-EINVAL)
86 # define SET_ENDIAN(a, b) (-EINVAL)
89 # define GET_TSC_CTL(a) (-EINVAL)
92 # define SET_TSC_CTL(a) (-EINVAL)
94 #ifndef MPX_ENABLE_MANAGEMENT
95 # define MPX_ENABLE_MANAGEMENT(a) (-EINVAL)
97 #ifndef MPX_DISABLE_MANAGEMENT
98 # define MPX_DISABLE_MANAGEMENT(a) (-EINVAL)
101 # define GET_FP_MODE(a) (-EINVAL)
104 # define SET_FP_MODE(a,b) (-EINVAL)
108 * this is where the system-wide overflow UID and GID are defined, for
109 * architectures that now have 32-bit UID/GID but didn't in the past
112 int overflowuid = DEFAULT_OVERFLOWUID;
113 int overflowgid = DEFAULT_OVERFLOWGID;
115 EXPORT_SYMBOL(overflowuid);
116 EXPORT_SYMBOL(overflowgid);
119 * the same as above, but for filesystems which can only store a 16-bit
120 * UID and GID. as such, this is needed on all architectures
123 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
124 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
126 EXPORT_SYMBOL(fs_overflowuid);
127 EXPORT_SYMBOL(fs_overflowgid);
130 * Returns true if current's euid is same as p's uid or euid,
131 * or has CAP_SYS_NICE to p's user_ns.
133 * Called with rcu_read_lock, creds are safe
135 static bool set_one_prio_perm(struct task_struct *p)
137 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
139 if (uid_eq(pcred->uid, cred->euid) ||
140 uid_eq(pcred->euid, cred->euid))
142 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
148 * set the priority of a task
149 * - the caller must hold the RCU read lock
151 static int set_one_prio(struct task_struct *p, int niceval, int error)
155 if (!set_one_prio_perm(p)) {
159 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
163 no_nice = security_task_setnice(p, niceval);
170 set_user_nice(p, niceval);
175 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
177 struct task_struct *g, *p;
178 struct user_struct *user;
179 const struct cred *cred = current_cred();
184 if (which > PRIO_USER || which < PRIO_PROCESS)
187 /* normalize: avoid signed division (rounding problems) */
189 if (niceval < MIN_NICE)
191 if (niceval > MAX_NICE)
195 read_lock(&tasklist_lock);
199 p = find_task_by_vpid(who);
203 error = set_one_prio(p, niceval, error);
207 pgrp = find_vpid(who);
209 pgrp = task_pgrp(current);
210 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
211 error = set_one_prio(p, niceval, error);
212 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
215 uid = make_kuid(cred->user_ns, who);
219 else if (!uid_eq(uid, cred->uid)) {
220 user = find_user(uid);
222 goto out_unlock; /* No processes for this user */
224 do_each_thread(g, p) {
225 if (uid_eq(task_uid(p), uid))
226 error = set_one_prio(p, niceval, error);
227 } while_each_thread(g, p);
228 if (!uid_eq(uid, cred->uid))
229 free_uid(user); /* For find_user() */
233 read_unlock(&tasklist_lock);
240 * Ugh. To avoid negative return values, "getpriority()" will
241 * not return the normal nice-value, but a negated value that
242 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
243 * to stay compatible.
245 SYSCALL_DEFINE2(getpriority, int, which, int, who)
247 struct task_struct *g, *p;
248 struct user_struct *user;
249 const struct cred *cred = current_cred();
250 long niceval, retval = -ESRCH;
254 if (which > PRIO_USER || which < PRIO_PROCESS)
258 read_lock(&tasklist_lock);
262 p = find_task_by_vpid(who);
266 niceval = nice_to_rlimit(task_nice(p));
267 if (niceval > retval)
273 pgrp = find_vpid(who);
275 pgrp = task_pgrp(current);
276 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
277 niceval = nice_to_rlimit(task_nice(p));
278 if (niceval > retval)
280 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
283 uid = make_kuid(cred->user_ns, who);
287 else if (!uid_eq(uid, cred->uid)) {
288 user = find_user(uid);
290 goto out_unlock; /* No processes for this user */
292 do_each_thread(g, p) {
293 if (uid_eq(task_uid(p), uid)) {
294 niceval = nice_to_rlimit(task_nice(p));
295 if (niceval > retval)
298 } while_each_thread(g, p);
299 if (!uid_eq(uid, cred->uid))
300 free_uid(user); /* for find_user() */
304 read_unlock(&tasklist_lock);
311 * Unprivileged users may change the real gid to the effective gid
312 * or vice versa. (BSD-style)
314 * If you set the real gid at all, or set the effective gid to a value not
315 * equal to the real gid, then the saved gid is set to the new effective gid.
317 * This makes it possible for a setgid program to completely drop its
318 * privileges, which is often a useful assertion to make when you are doing
319 * a security audit over a program.
321 * The general idea is that a program which uses just setregid() will be
322 * 100% compatible with BSD. A program which uses just setgid() will be
323 * 100% compatible with POSIX with saved IDs.
325 * SMP: There are not races, the GIDs are checked only by filesystem
326 * operations (as far as semantic preservation is concerned).
328 #ifdef CONFIG_MULTIUSER
329 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
331 struct user_namespace *ns = current_user_ns();
332 const struct cred *old;
337 krgid = make_kgid(ns, rgid);
338 kegid = make_kgid(ns, egid);
340 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
342 if ((egid != (gid_t) -1) && !gid_valid(kegid))
345 new = prepare_creds();
348 old = current_cred();
351 if (rgid != (gid_t) -1) {
352 if (gid_eq(old->gid, krgid) ||
353 gid_eq(old->egid, krgid) ||
354 ns_capable(old->user_ns, CAP_SETGID))
359 if (egid != (gid_t) -1) {
360 if (gid_eq(old->gid, kegid) ||
361 gid_eq(old->egid, kegid) ||
362 gid_eq(old->sgid, kegid) ||
363 ns_capable(old->user_ns, CAP_SETGID))
369 if (rgid != (gid_t) -1 ||
370 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
371 new->sgid = new->egid;
372 new->fsgid = new->egid;
374 return commit_creds(new);
382 * setgid() is implemented like SysV w/ SAVED_IDS
384 * SMP: Same implicit races as above.
386 SYSCALL_DEFINE1(setgid, gid_t, gid)
388 struct user_namespace *ns = current_user_ns();
389 const struct cred *old;
394 kgid = make_kgid(ns, gid);
395 if (!gid_valid(kgid))
398 new = prepare_creds();
401 old = current_cred();
404 if (ns_capable(old->user_ns, CAP_SETGID))
405 new->gid = new->egid = new->sgid = new->fsgid = kgid;
406 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
407 new->egid = new->fsgid = kgid;
411 return commit_creds(new);
419 * change the user struct in a credentials set to match the new UID
421 static int set_user(struct cred *new)
423 struct user_struct *new_user;
425 new_user = alloc_uid(new->uid);
430 * We don't fail in case of NPROC limit excess here because too many
431 * poorly written programs don't check set*uid() return code, assuming
432 * it never fails if called by root. We may still enforce NPROC limit
433 * for programs doing set*uid()+execve() by harmlessly deferring the
434 * failure to the execve() stage.
436 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
437 new_user != INIT_USER)
438 current->flags |= PF_NPROC_EXCEEDED;
440 current->flags &= ~PF_NPROC_EXCEEDED;
443 new->user = new_user;
448 * Unprivileged users may change the real uid to the effective uid
449 * or vice versa. (BSD-style)
451 * If you set the real uid at all, or set the effective uid to a value not
452 * equal to the real uid, then the saved uid is set to the new effective uid.
454 * This makes it possible for a setuid program to completely drop its
455 * privileges, which is often a useful assertion to make when you are doing
456 * a security audit over a program.
458 * The general idea is that a program which uses just setreuid() will be
459 * 100% compatible with BSD. A program which uses just setuid() will be
460 * 100% compatible with POSIX with saved IDs.
462 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
464 struct user_namespace *ns = current_user_ns();
465 const struct cred *old;
470 kruid = make_kuid(ns, ruid);
471 keuid = make_kuid(ns, euid);
473 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
475 if ((euid != (uid_t) -1) && !uid_valid(keuid))
478 new = prepare_creds();
481 old = current_cred();
484 if (ruid != (uid_t) -1) {
486 if (!uid_eq(old->uid, kruid) &&
487 !uid_eq(old->euid, kruid) &&
488 !ns_capable(old->user_ns, CAP_SETUID))
492 if (euid != (uid_t) -1) {
494 if (!uid_eq(old->uid, keuid) &&
495 !uid_eq(old->euid, keuid) &&
496 !uid_eq(old->suid, keuid) &&
497 !ns_capable(old->user_ns, CAP_SETUID))
501 if (!uid_eq(new->uid, old->uid)) {
502 retval = set_user(new);
506 if (ruid != (uid_t) -1 ||
507 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
508 new->suid = new->euid;
509 new->fsuid = new->euid;
511 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
515 return commit_creds(new);
523 * setuid() is implemented like SysV with SAVED_IDS
525 * Note that SAVED_ID's is deficient in that a setuid root program
526 * like sendmail, for example, cannot set its uid to be a normal
527 * user and then switch back, because if you're root, setuid() sets
528 * the saved uid too. If you don't like this, blame the bright people
529 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
530 * will allow a root program to temporarily drop privileges and be able to
531 * regain them by swapping the real and effective uid.
533 SYSCALL_DEFINE1(setuid, uid_t, uid)
535 struct user_namespace *ns = current_user_ns();
536 const struct cred *old;
541 kuid = make_kuid(ns, uid);
542 if (!uid_valid(kuid))
545 new = prepare_creds();
548 old = current_cred();
551 if (ns_capable(old->user_ns, CAP_SETUID)) {
552 new->suid = new->uid = kuid;
553 if (!uid_eq(kuid, old->uid)) {
554 retval = set_user(new);
558 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
562 new->fsuid = new->euid = kuid;
564 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
568 return commit_creds(new);
577 * This function implements a generic ability to update ruid, euid,
578 * and suid. This allows you to implement the 4.4 compatible seteuid().
580 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
582 struct user_namespace *ns = current_user_ns();
583 const struct cred *old;
586 kuid_t kruid, keuid, ksuid;
588 kruid = make_kuid(ns, ruid);
589 keuid = make_kuid(ns, euid);
590 ksuid = make_kuid(ns, suid);
592 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
595 if ((euid != (uid_t) -1) && !uid_valid(keuid))
598 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
601 new = prepare_creds();
605 old = current_cred();
608 if (!ns_capable(old->user_ns, CAP_SETUID)) {
609 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
610 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
612 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
613 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
615 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
616 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
620 if (ruid != (uid_t) -1) {
622 if (!uid_eq(kruid, old->uid)) {
623 retval = set_user(new);
628 if (euid != (uid_t) -1)
630 if (suid != (uid_t) -1)
632 new->fsuid = new->euid;
634 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
638 return commit_creds(new);
645 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
647 const struct cred *cred = current_cred();
649 uid_t ruid, euid, suid;
651 ruid = from_kuid_munged(cred->user_ns, cred->uid);
652 euid = from_kuid_munged(cred->user_ns, cred->euid);
653 suid = from_kuid_munged(cred->user_ns, cred->suid);
655 retval = put_user(ruid, ruidp);
657 retval = put_user(euid, euidp);
659 return put_user(suid, suidp);
665 * Same as above, but for rgid, egid, sgid.
667 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
669 struct user_namespace *ns = current_user_ns();
670 const struct cred *old;
673 kgid_t krgid, kegid, ksgid;
675 krgid = make_kgid(ns, rgid);
676 kegid = make_kgid(ns, egid);
677 ksgid = make_kgid(ns, sgid);
679 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
681 if ((egid != (gid_t) -1) && !gid_valid(kegid))
683 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
686 new = prepare_creds();
689 old = current_cred();
692 if (!ns_capable(old->user_ns, CAP_SETGID)) {
693 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
694 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
696 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
697 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
699 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
700 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
704 if (rgid != (gid_t) -1)
706 if (egid != (gid_t) -1)
708 if (sgid != (gid_t) -1)
710 new->fsgid = new->egid;
712 return commit_creds(new);
719 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
721 const struct cred *cred = current_cred();
723 gid_t rgid, egid, sgid;
725 rgid = from_kgid_munged(cred->user_ns, cred->gid);
726 egid = from_kgid_munged(cred->user_ns, cred->egid);
727 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
729 retval = put_user(rgid, rgidp);
731 retval = put_user(egid, egidp);
733 retval = put_user(sgid, sgidp);
741 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
742 * is used for "access()" and for the NFS daemon (letting nfsd stay at
743 * whatever uid it wants to). It normally shadows "euid", except when
744 * explicitly set by setfsuid() or for access..
746 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
748 const struct cred *old;
753 old = current_cred();
754 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
756 kuid = make_kuid(old->user_ns, uid);
757 if (!uid_valid(kuid))
760 new = prepare_creds();
764 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
765 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
766 ns_capable(old->user_ns, CAP_SETUID)) {
767 if (!uid_eq(kuid, old->fsuid)) {
769 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
783 * Samma på svenska..
785 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
787 const struct cred *old;
792 old = current_cred();
793 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
795 kgid = make_kgid(old->user_ns, gid);
796 if (!gid_valid(kgid))
799 new = prepare_creds();
803 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
804 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
805 ns_capable(old->user_ns, CAP_SETGID)) {
806 if (!gid_eq(kgid, old->fsgid)) {
819 #endif /* CONFIG_MULTIUSER */
822 * sys_getpid - return the thread group id of the current process
824 * Note, despite the name, this returns the tgid not the pid. The tgid and
825 * the pid are identical unless CLONE_THREAD was specified on clone() in
826 * which case the tgid is the same in all threads of the same group.
828 * This is SMP safe as current->tgid does not change.
830 SYSCALL_DEFINE0(getpid)
832 return task_tgid_vnr(current);
835 /* Thread ID - the internal kernel "pid" */
836 SYSCALL_DEFINE0(gettid)
838 return task_pid_vnr(current);
842 * Accessing ->real_parent is not SMP-safe, it could
843 * change from under us. However, we can use a stale
844 * value of ->real_parent under rcu_read_lock(), see
845 * release_task()->call_rcu(delayed_put_task_struct).
847 SYSCALL_DEFINE0(getppid)
852 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
858 SYSCALL_DEFINE0(getuid)
860 /* Only we change this so SMP safe */
861 return from_kuid_munged(current_user_ns(), current_uid());
864 SYSCALL_DEFINE0(geteuid)
866 /* Only we change this so SMP safe */
867 return from_kuid_munged(current_user_ns(), current_euid());
870 SYSCALL_DEFINE0(getgid)
872 /* Only we change this so SMP safe */
873 return from_kgid_munged(current_user_ns(), current_gid());
876 SYSCALL_DEFINE0(getegid)
878 /* Only we change this so SMP safe */
879 return from_kgid_munged(current_user_ns(), current_egid());
882 void do_sys_times(struct tms *tms)
884 cputime_t tgutime, tgstime, cutime, cstime;
886 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
887 cutime = current->signal->cutime;
888 cstime = current->signal->cstime;
889 tms->tms_utime = cputime_to_clock_t(tgutime);
890 tms->tms_stime = cputime_to_clock_t(tgstime);
891 tms->tms_cutime = cputime_to_clock_t(cutime);
892 tms->tms_cstime = cputime_to_clock_t(cstime);
895 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
901 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
904 force_successful_syscall_return();
905 return (long) jiffies_64_to_clock_t(get_jiffies_64());
909 * This needs some heavy checking ...
910 * I just haven't the stomach for it. I also don't fully
911 * understand sessions/pgrp etc. Let somebody who does explain it.
913 * OK, I think I have the protection semantics right.... this is really
914 * only important on a multi-user system anyway, to make sure one user
915 * can't send a signal to a process owned by another. -TYT, 12/12/91
917 * !PF_FORKNOEXEC check to conform completely to POSIX.
919 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
921 struct task_struct *p;
922 struct task_struct *group_leader = current->group_leader;
927 pid = task_pid_vnr(group_leader);
934 /* From this point forward we keep holding onto the tasklist lock
935 * so that our parent does not change from under us. -DaveM
937 write_lock_irq(&tasklist_lock);
940 p = find_task_by_vpid(pid);
945 if (!thread_group_leader(p))
948 if (same_thread_group(p->real_parent, group_leader)) {
950 if (task_session(p) != task_session(group_leader))
953 if (!(p->flags & PF_FORKNOEXEC))
957 if (p != group_leader)
962 if (p->signal->leader)
967 struct task_struct *g;
969 pgrp = find_vpid(pgid);
970 g = pid_task(pgrp, PIDTYPE_PGID);
971 if (!g || task_session(g) != task_session(group_leader))
975 err = security_task_setpgid(p, pgid);
979 if (task_pgrp(p) != pgrp)
980 change_pid(p, PIDTYPE_PGID, pgrp);
984 /* All paths lead to here, thus we are safe. -DaveM */
985 write_unlock_irq(&tasklist_lock);
990 SYSCALL_DEFINE1(getpgid, pid_t, pid)
992 struct task_struct *p;
998 grp = task_pgrp(current);
1001 p = find_task_by_vpid(pid);
1008 retval = security_task_getpgid(p);
1012 retval = pid_vnr(grp);
1018 #ifdef __ARCH_WANT_SYS_GETPGRP
1020 SYSCALL_DEFINE0(getpgrp)
1022 return sys_getpgid(0);
1027 SYSCALL_DEFINE1(getsid, pid_t, pid)
1029 struct task_struct *p;
1035 sid = task_session(current);
1038 p = find_task_by_vpid(pid);
1041 sid = task_session(p);
1045 retval = security_task_getsid(p);
1049 retval = pid_vnr(sid);
1055 static void set_special_pids(struct pid *pid)
1057 struct task_struct *curr = current->group_leader;
1059 if (task_session(curr) != pid)
1060 change_pid(curr, PIDTYPE_SID, pid);
1062 if (task_pgrp(curr) != pid)
1063 change_pid(curr, PIDTYPE_PGID, pid);
1066 SYSCALL_DEFINE0(setsid)
1068 struct task_struct *group_leader = current->group_leader;
1069 struct pid *sid = task_pid(group_leader);
1070 pid_t session = pid_vnr(sid);
1073 write_lock_irq(&tasklist_lock);
1074 /* Fail if I am already a session leader */
1075 if (group_leader->signal->leader)
1078 /* Fail if a process group id already exists that equals the
1079 * proposed session id.
1081 if (pid_task(sid, PIDTYPE_PGID))
1084 group_leader->signal->leader = 1;
1085 set_special_pids(sid);
1087 proc_clear_tty(group_leader);
1091 write_unlock_irq(&tasklist_lock);
1093 proc_sid_connector(group_leader);
1094 sched_autogroup_create_attach(group_leader);
1099 DECLARE_RWSEM(uts_sem);
1101 #ifdef COMPAT_UTS_MACHINE
1102 #define override_architecture(name) \
1103 (personality(current->personality) == PER_LINUX32 && \
1104 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1105 sizeof(COMPAT_UTS_MACHINE)))
1107 #define override_architecture(name) 0
1111 * Work around broken programs that cannot handle "Linux 3.0".
1112 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1113 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1115 static int override_release(char __user *release, size_t len)
1119 if (current->personality & UNAME26) {
1120 const char *rest = UTS_RELEASE;
1121 char buf[65] = { 0 };
1127 if (*rest == '.' && ++ndots >= 3)
1129 if (!isdigit(*rest) && *rest != '.')
1133 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1134 copy = clamp_t(size_t, len, 1, sizeof(buf));
1135 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1136 ret = copy_to_user(release, buf, copy + 1);
1141 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1145 down_read(&uts_sem);
1146 if (copy_to_user(name, utsname(), sizeof *name))
1150 if (!errno && override_release(name->release, sizeof(name->release)))
1152 if (!errno && override_architecture(name))
1157 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1161 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1168 down_read(&uts_sem);
1169 if (copy_to_user(name, utsname(), sizeof(*name)))
1173 if (!error && override_release(name->release, sizeof(name->release)))
1175 if (!error && override_architecture(name))
1180 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1186 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1189 down_read(&uts_sem);
1190 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1192 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1193 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1195 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1196 error |= __copy_to_user(&name->release, &utsname()->release,
1198 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1199 error |= __copy_to_user(&name->version, &utsname()->version,
1201 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1202 error |= __copy_to_user(&name->machine, &utsname()->machine,
1204 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1207 if (!error && override_architecture(name))
1209 if (!error && override_release(name->release, sizeof(name->release)))
1211 return error ? -EFAULT : 0;
1215 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1218 char tmp[__NEW_UTS_LEN];
1220 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1223 if (len < 0 || len > __NEW_UTS_LEN)
1225 down_write(&uts_sem);
1227 if (!copy_from_user(tmp, name, len)) {
1228 struct new_utsname *u = utsname();
1230 memcpy(u->nodename, tmp, len);
1231 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1233 uts_proc_notify(UTS_PROC_HOSTNAME);
1239 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1241 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1244 struct new_utsname *u;
1248 down_read(&uts_sem);
1250 i = 1 + strlen(u->nodename);
1254 if (copy_to_user(name, u->nodename, i))
1263 * Only setdomainname; getdomainname can be implemented by calling
1266 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1269 char tmp[__NEW_UTS_LEN];
1271 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1273 if (len < 0 || len > __NEW_UTS_LEN)
1276 down_write(&uts_sem);
1278 if (!copy_from_user(tmp, name, len)) {
1279 struct new_utsname *u = utsname();
1281 memcpy(u->domainname, tmp, len);
1282 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1284 uts_proc_notify(UTS_PROC_DOMAINNAME);
1290 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1292 struct rlimit value;
1295 ret = do_prlimit(current, resource, NULL, &value);
1297 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1302 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1305 * Back compatibility for getrlimit. Needed for some apps.
1307 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1308 struct rlimit __user *, rlim)
1311 if (resource >= RLIM_NLIMITS)
1314 task_lock(current->group_leader);
1315 x = current->signal->rlim[resource];
1316 task_unlock(current->group_leader);
1317 if (x.rlim_cur > 0x7FFFFFFF)
1318 x.rlim_cur = 0x7FFFFFFF;
1319 if (x.rlim_max > 0x7FFFFFFF)
1320 x.rlim_max = 0x7FFFFFFF;
1321 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1326 static inline bool rlim64_is_infinity(__u64 rlim64)
1328 #if BITS_PER_LONG < 64
1329 return rlim64 >= ULONG_MAX;
1331 return rlim64 == RLIM64_INFINITY;
1335 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1337 if (rlim->rlim_cur == RLIM_INFINITY)
1338 rlim64->rlim_cur = RLIM64_INFINITY;
1340 rlim64->rlim_cur = rlim->rlim_cur;
1341 if (rlim->rlim_max == RLIM_INFINITY)
1342 rlim64->rlim_max = RLIM64_INFINITY;
1344 rlim64->rlim_max = rlim->rlim_max;
1347 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1349 if (rlim64_is_infinity(rlim64->rlim_cur))
1350 rlim->rlim_cur = RLIM_INFINITY;
1352 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1353 if (rlim64_is_infinity(rlim64->rlim_max))
1354 rlim->rlim_max = RLIM_INFINITY;
1356 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1359 /* make sure you are allowed to change @tsk limits before calling this */
1360 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1361 struct rlimit *new_rlim, struct rlimit *old_rlim)
1363 struct rlimit *rlim;
1366 if (resource >= RLIM_NLIMITS)
1369 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1371 if (resource == RLIMIT_NOFILE &&
1372 new_rlim->rlim_max > sysctl_nr_open)
1376 /* protect tsk->signal and tsk->sighand from disappearing */
1377 read_lock(&tasklist_lock);
1378 if (!tsk->sighand) {
1383 rlim = tsk->signal->rlim + resource;
1384 task_lock(tsk->group_leader);
1386 /* Keep the capable check against init_user_ns until
1387 cgroups can contain all limits */
1388 if (new_rlim->rlim_max > rlim->rlim_max &&
1389 !capable(CAP_SYS_RESOURCE))
1392 retval = security_task_setrlimit(tsk->group_leader,
1393 resource, new_rlim);
1394 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1396 * The caller is asking for an immediate RLIMIT_CPU
1397 * expiry. But we use the zero value to mean "it was
1398 * never set". So let's cheat and make it one second
1401 new_rlim->rlim_cur = 1;
1410 task_unlock(tsk->group_leader);
1413 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1414 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1415 * very long-standing error, and fixing it now risks breakage of
1416 * applications, so we live with it
1418 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1419 new_rlim->rlim_cur != RLIM_INFINITY)
1420 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1422 read_unlock(&tasklist_lock);
1426 /* rcu lock must be held */
1427 static int check_prlimit_permission(struct task_struct *task)
1429 const struct cred *cred = current_cred(), *tcred;
1431 if (current == task)
1434 tcred = __task_cred(task);
1435 if (uid_eq(cred->uid, tcred->euid) &&
1436 uid_eq(cred->uid, tcred->suid) &&
1437 uid_eq(cred->uid, tcred->uid) &&
1438 gid_eq(cred->gid, tcred->egid) &&
1439 gid_eq(cred->gid, tcred->sgid) &&
1440 gid_eq(cred->gid, tcred->gid))
1442 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1448 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1449 const struct rlimit64 __user *, new_rlim,
1450 struct rlimit64 __user *, old_rlim)
1452 struct rlimit64 old64, new64;
1453 struct rlimit old, new;
1454 struct task_struct *tsk;
1458 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1460 rlim64_to_rlim(&new64, &new);
1464 tsk = pid ? find_task_by_vpid(pid) : current;
1469 ret = check_prlimit_permission(tsk);
1474 get_task_struct(tsk);
1477 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1478 old_rlim ? &old : NULL);
1480 if (!ret && old_rlim) {
1481 rlim_to_rlim64(&old, &old64);
1482 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1486 put_task_struct(tsk);
1490 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1492 struct rlimit new_rlim;
1494 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1496 return do_prlimit(current, resource, &new_rlim, NULL);
1500 * It would make sense to put struct rusage in the task_struct,
1501 * except that would make the task_struct be *really big*. After
1502 * task_struct gets moved into malloc'ed memory, it would
1503 * make sense to do this. It will make moving the rest of the information
1504 * a lot simpler! (Which we're not doing right now because we're not
1505 * measuring them yet).
1507 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1508 * races with threads incrementing their own counters. But since word
1509 * reads are atomic, we either get new values or old values and we don't
1510 * care which for the sums. We always take the siglock to protect reading
1511 * the c* fields from p->signal from races with exit.c updating those
1512 * fields when reaping, so a sample either gets all the additions of a
1513 * given child after it's reaped, or none so this sample is before reaping.
1516 * We need to take the siglock for CHILDEREN, SELF and BOTH
1517 * for the cases current multithreaded, non-current single threaded
1518 * non-current multithreaded. Thread traversal is now safe with
1520 * Strictly speaking, we donot need to take the siglock if we are current and
1521 * single threaded, as no one else can take our signal_struct away, no one
1522 * else can reap the children to update signal->c* counters, and no one else
1523 * can race with the signal-> fields. If we do not take any lock, the
1524 * signal-> fields could be read out of order while another thread was just
1525 * exiting. So we should place a read memory barrier when we avoid the lock.
1526 * On the writer side, write memory barrier is implied in __exit_signal
1527 * as __exit_signal releases the siglock spinlock after updating the signal->
1528 * fields. But we don't do this yet to keep things simple.
1532 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1534 r->ru_nvcsw += t->nvcsw;
1535 r->ru_nivcsw += t->nivcsw;
1536 r->ru_minflt += t->min_flt;
1537 r->ru_majflt += t->maj_flt;
1538 r->ru_inblock += task_io_get_inblock(t);
1539 r->ru_oublock += task_io_get_oublock(t);
1542 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1544 struct task_struct *t;
1545 unsigned long flags;
1546 cputime_t tgutime, tgstime, utime, stime;
1547 unsigned long maxrss = 0;
1549 memset((char *)r, 0, sizeof (*r));
1552 if (who == RUSAGE_THREAD) {
1553 task_cputime_adjusted(current, &utime, &stime);
1554 accumulate_thread_rusage(p, r);
1555 maxrss = p->signal->maxrss;
1559 if (!lock_task_sighand(p, &flags))
1564 case RUSAGE_CHILDREN:
1565 utime = p->signal->cutime;
1566 stime = p->signal->cstime;
1567 r->ru_nvcsw = p->signal->cnvcsw;
1568 r->ru_nivcsw = p->signal->cnivcsw;
1569 r->ru_minflt = p->signal->cmin_flt;
1570 r->ru_majflt = p->signal->cmaj_flt;
1571 r->ru_inblock = p->signal->cinblock;
1572 r->ru_oublock = p->signal->coublock;
1573 maxrss = p->signal->cmaxrss;
1575 if (who == RUSAGE_CHILDREN)
1579 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1582 r->ru_nvcsw += p->signal->nvcsw;
1583 r->ru_nivcsw += p->signal->nivcsw;
1584 r->ru_minflt += p->signal->min_flt;
1585 r->ru_majflt += p->signal->maj_flt;
1586 r->ru_inblock += p->signal->inblock;
1587 r->ru_oublock += p->signal->oublock;
1588 if (maxrss < p->signal->maxrss)
1589 maxrss = p->signal->maxrss;
1592 accumulate_thread_rusage(t, r);
1593 } while_each_thread(p, t);
1599 unlock_task_sighand(p, &flags);
1602 cputime_to_timeval(utime, &r->ru_utime);
1603 cputime_to_timeval(stime, &r->ru_stime);
1605 if (who != RUSAGE_CHILDREN) {
1606 struct mm_struct *mm = get_task_mm(p);
1609 setmax_mm_hiwater_rss(&maxrss, mm);
1613 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1616 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1620 k_getrusage(p, who, &r);
1621 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1624 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1626 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1627 who != RUSAGE_THREAD)
1629 return getrusage(current, who, ru);
1632 #ifdef CONFIG_COMPAT
1633 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1637 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1638 who != RUSAGE_THREAD)
1641 k_getrusage(current, who, &r);
1642 return put_compat_rusage(&r, ru);
1646 SYSCALL_DEFINE1(umask, int, mask)
1648 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1652 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1655 struct file *old_exe, *exe_file;
1656 struct inode *inode;
1663 inode = file_inode(exe.file);
1666 * Because the original mm->exe_file points to executable file, make
1667 * sure that this one is executable as well, to avoid breaking an
1671 if (!S_ISREG(inode->i_mode) ||
1672 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1675 err = inode_permission(inode, MAY_EXEC);
1680 * Forbid mm->exe_file change if old file still mapped.
1682 exe_file = get_mm_exe_file(mm);
1685 struct vm_area_struct *vma;
1687 down_read(&mm->mmap_sem);
1688 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1691 if (path_equal(&vma->vm_file->f_path,
1696 up_read(&mm->mmap_sem);
1701 * The symlink can be changed only once, just to disallow arbitrary
1702 * transitions malicious software might bring in. This means one
1703 * could make a snapshot over all processes running and monitor
1704 * /proc/pid/exe changes to notice unusual activity if needed.
1707 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1711 /* set the new file, lockless */
1713 old_exe = xchg(&mm->exe_file, exe.file);
1720 up_read(&mm->mmap_sem);
1725 #ifdef CONFIG_CHECKPOINT_RESTORE
1727 * WARNING: we don't require any capability here so be very careful
1728 * in what is allowed for modification from userspace.
1730 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1732 unsigned long mmap_max_addr = TASK_SIZE;
1733 struct mm_struct *mm = current->mm;
1734 int error = -EINVAL, i;
1736 static const unsigned char offsets[] = {
1737 offsetof(struct prctl_mm_map, start_code),
1738 offsetof(struct prctl_mm_map, end_code),
1739 offsetof(struct prctl_mm_map, start_data),
1740 offsetof(struct prctl_mm_map, end_data),
1741 offsetof(struct prctl_mm_map, start_brk),
1742 offsetof(struct prctl_mm_map, brk),
1743 offsetof(struct prctl_mm_map, start_stack),
1744 offsetof(struct prctl_mm_map, arg_start),
1745 offsetof(struct prctl_mm_map, arg_end),
1746 offsetof(struct prctl_mm_map, env_start),
1747 offsetof(struct prctl_mm_map, env_end),
1751 * Make sure the members are not somewhere outside
1752 * of allowed address space.
1754 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1755 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1757 if ((unsigned long)val >= mmap_max_addr ||
1758 (unsigned long)val < mmap_min_addr)
1763 * Make sure the pairs are ordered.
1765 #define __prctl_check_order(__m1, __op, __m2) \
1766 ((unsigned long)prctl_map->__m1 __op \
1767 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1768 error = __prctl_check_order(start_code, <, end_code);
1769 error |= __prctl_check_order(start_data, <, end_data);
1770 error |= __prctl_check_order(start_brk, <=, brk);
1771 error |= __prctl_check_order(arg_start, <=, arg_end);
1772 error |= __prctl_check_order(env_start, <=, env_end);
1775 #undef __prctl_check_order
1780 * @brk should be after @end_data in traditional maps.
1782 if (prctl_map->start_brk <= prctl_map->end_data ||
1783 prctl_map->brk <= prctl_map->end_data)
1787 * Neither we should allow to override limits if they set.
1789 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1790 prctl_map->start_brk, prctl_map->end_data,
1791 prctl_map->start_data))
1795 * Someone is trying to cheat the auxv vector.
1797 if (prctl_map->auxv_size) {
1798 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1803 * Finally, make sure the caller has the rights to
1804 * change /proc/pid/exe link: only local root should
1807 if (prctl_map->exe_fd != (u32)-1) {
1808 struct user_namespace *ns = current_user_ns();
1809 const struct cred *cred = current_cred();
1811 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1812 !gid_eq(cred->gid, make_kgid(ns, 0)))
1821 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1823 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1824 unsigned long user_auxv[AT_VECTOR_SIZE];
1825 struct mm_struct *mm = current->mm;
1828 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1829 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1831 if (opt == PR_SET_MM_MAP_SIZE)
1832 return put_user((unsigned int)sizeof(prctl_map),
1833 (unsigned int __user *)addr);
1835 if (data_size != sizeof(prctl_map))
1838 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1841 error = validate_prctl_map(&prctl_map);
1845 if (prctl_map.auxv_size) {
1846 memset(user_auxv, 0, sizeof(user_auxv));
1847 if (copy_from_user(user_auxv,
1848 (const void __user *)prctl_map.auxv,
1849 prctl_map.auxv_size))
1852 /* Last entry must be AT_NULL as specification requires */
1853 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1854 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1857 if (prctl_map.exe_fd != (u32)-1)
1858 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1859 down_read(&mm->mmap_sem);
1864 * We don't validate if these members are pointing to
1865 * real present VMAs because application may have correspond
1866 * VMAs already unmapped and kernel uses these members for statistics
1867 * output in procfs mostly, except
1869 * - @start_brk/@brk which are used in do_brk but kernel lookups
1870 * for VMAs when updating these memvers so anything wrong written
1871 * here cause kernel to swear at userspace program but won't lead
1872 * to any problem in kernel itself
1875 mm->start_code = prctl_map.start_code;
1876 mm->end_code = prctl_map.end_code;
1877 mm->start_data = prctl_map.start_data;
1878 mm->end_data = prctl_map.end_data;
1879 mm->start_brk = prctl_map.start_brk;
1880 mm->brk = prctl_map.brk;
1881 mm->start_stack = prctl_map.start_stack;
1882 mm->arg_start = prctl_map.arg_start;
1883 mm->arg_end = prctl_map.arg_end;
1884 mm->env_start = prctl_map.env_start;
1885 mm->env_end = prctl_map.env_end;
1888 * Note this update of @saved_auxv is lockless thus
1889 * if someone reads this member in procfs while we're
1890 * updating -- it may get partly updated results. It's
1891 * known and acceptable trade off: we leave it as is to
1892 * not introduce additional locks here making the kernel
1895 if (prctl_map.auxv_size)
1896 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1900 up_read(&mm->mmap_sem);
1903 #endif /* CONFIG_CHECKPOINT_RESTORE */
1905 static int prctl_set_mm(int opt, unsigned long addr,
1906 unsigned long arg4, unsigned long arg5)
1908 struct mm_struct *mm = current->mm;
1909 struct vm_area_struct *vma;
1912 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1913 opt != PR_SET_MM_MAP &&
1914 opt != PR_SET_MM_MAP_SIZE)))
1917 #ifdef CONFIG_CHECKPOINT_RESTORE
1918 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1919 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1922 if (!capable(CAP_SYS_RESOURCE))
1925 if (opt == PR_SET_MM_EXE_FILE)
1926 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1928 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1933 down_read(&mm->mmap_sem);
1934 vma = find_vma(mm, addr);
1937 case PR_SET_MM_START_CODE:
1938 mm->start_code = addr;
1940 case PR_SET_MM_END_CODE:
1941 mm->end_code = addr;
1943 case PR_SET_MM_START_DATA:
1944 mm->start_data = addr;
1946 case PR_SET_MM_END_DATA:
1947 mm->end_data = addr;
1950 case PR_SET_MM_START_BRK:
1951 if (addr <= mm->end_data)
1954 if (check_data_rlimit(rlimit(RLIMIT_DATA), mm->brk, addr,
1955 mm->end_data, mm->start_data))
1958 mm->start_brk = addr;
1962 if (addr <= mm->end_data)
1965 if (check_data_rlimit(rlimit(RLIMIT_DATA), addr, mm->start_brk,
1966 mm->end_data, mm->start_data))
1973 * If command line arguments and environment
1974 * are placed somewhere else on stack, we can
1975 * set them up here, ARG_START/END to setup
1976 * command line argumets and ENV_START/END
1979 case PR_SET_MM_START_STACK:
1980 case PR_SET_MM_ARG_START:
1981 case PR_SET_MM_ARG_END:
1982 case PR_SET_MM_ENV_START:
1983 case PR_SET_MM_ENV_END:
1988 if (opt == PR_SET_MM_START_STACK)
1989 mm->start_stack = addr;
1990 else if (opt == PR_SET_MM_ARG_START)
1991 mm->arg_start = addr;
1992 else if (opt == PR_SET_MM_ARG_END)
1994 else if (opt == PR_SET_MM_ENV_START)
1995 mm->env_start = addr;
1996 else if (opt == PR_SET_MM_ENV_END)
2001 * This doesn't move auxiliary vector itself
2002 * since it's pinned to mm_struct, but allow
2003 * to fill vector with new values. It's up
2004 * to a caller to provide sane values here
2005 * otherwise user space tools which use this
2006 * vector might be unhappy.
2008 case PR_SET_MM_AUXV: {
2009 unsigned long user_auxv[AT_VECTOR_SIZE];
2011 if (arg4 > sizeof(user_auxv))
2013 up_read(&mm->mmap_sem);
2015 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
2018 /* Make sure the last entry is always AT_NULL */
2019 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2020 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2022 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2025 memcpy(mm->saved_auxv, user_auxv, arg4);
2026 task_unlock(current);
2036 up_read(&mm->mmap_sem);
2040 #ifdef CONFIG_CHECKPOINT_RESTORE
2041 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2043 return put_user(me->clear_child_tid, tid_addr);
2046 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2052 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2053 unsigned long, arg4, unsigned long, arg5)
2055 struct task_struct *me = current;
2056 unsigned char comm[sizeof(me->comm)];
2059 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2060 if (error != -ENOSYS)
2065 case PR_SET_PDEATHSIG:
2066 if (!valid_signal(arg2)) {
2070 me->pdeath_signal = arg2;
2072 case PR_GET_PDEATHSIG:
2073 error = put_user(me->pdeath_signal, (int __user *)arg2);
2075 case PR_GET_DUMPABLE:
2076 error = get_dumpable(me->mm);
2078 case PR_SET_DUMPABLE:
2079 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2083 set_dumpable(me->mm, arg2);
2086 case PR_SET_UNALIGN:
2087 error = SET_UNALIGN_CTL(me, arg2);
2089 case PR_GET_UNALIGN:
2090 error = GET_UNALIGN_CTL(me, arg2);
2093 error = SET_FPEMU_CTL(me, arg2);
2096 error = GET_FPEMU_CTL(me, arg2);
2099 error = SET_FPEXC_CTL(me, arg2);
2102 error = GET_FPEXC_CTL(me, arg2);
2105 error = PR_TIMING_STATISTICAL;
2108 if (arg2 != PR_TIMING_STATISTICAL)
2112 comm[sizeof(me->comm) - 1] = 0;
2113 if (strncpy_from_user(comm, (char __user *)arg2,
2114 sizeof(me->comm) - 1) < 0)
2116 set_task_comm(me, comm);
2117 proc_comm_connector(me);
2120 get_task_comm(comm, me);
2121 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2125 error = GET_ENDIAN(me, arg2);
2128 error = SET_ENDIAN(me, arg2);
2130 case PR_GET_SECCOMP:
2131 error = prctl_get_seccomp();
2133 case PR_SET_SECCOMP:
2134 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2137 error = GET_TSC_CTL(arg2);
2140 error = SET_TSC_CTL(arg2);
2142 case PR_TASK_PERF_EVENTS_DISABLE:
2143 error = perf_event_task_disable();
2145 case PR_TASK_PERF_EVENTS_ENABLE:
2146 error = perf_event_task_enable();
2148 case PR_GET_TIMERSLACK:
2149 error = current->timer_slack_ns;
2151 case PR_SET_TIMERSLACK:
2153 current->timer_slack_ns =
2154 current->default_timer_slack_ns;
2156 current->timer_slack_ns = arg2;
2162 case PR_MCE_KILL_CLEAR:
2165 current->flags &= ~PF_MCE_PROCESS;
2167 case PR_MCE_KILL_SET:
2168 current->flags |= PF_MCE_PROCESS;
2169 if (arg3 == PR_MCE_KILL_EARLY)
2170 current->flags |= PF_MCE_EARLY;
2171 else if (arg3 == PR_MCE_KILL_LATE)
2172 current->flags &= ~PF_MCE_EARLY;
2173 else if (arg3 == PR_MCE_KILL_DEFAULT)
2175 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2183 case PR_MCE_KILL_GET:
2184 if (arg2 | arg3 | arg4 | arg5)
2186 if (current->flags & PF_MCE_PROCESS)
2187 error = (current->flags & PF_MCE_EARLY) ?
2188 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2190 error = PR_MCE_KILL_DEFAULT;
2193 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2195 case PR_GET_TID_ADDRESS:
2196 error = prctl_get_tid_address(me, (int __user **)arg2);
2198 case PR_SET_CHILD_SUBREAPER:
2199 me->signal->is_child_subreaper = !!arg2;
2201 case PR_GET_CHILD_SUBREAPER:
2202 error = put_user(me->signal->is_child_subreaper,
2203 (int __user *)arg2);
2205 case PR_SET_NO_NEW_PRIVS:
2206 if (arg2 != 1 || arg3 || arg4 || arg5)
2209 task_set_no_new_privs(current);
2211 case PR_GET_NO_NEW_PRIVS:
2212 if (arg2 || arg3 || arg4 || arg5)
2214 return task_no_new_privs(current) ? 1 : 0;
2215 case PR_GET_THP_DISABLE:
2216 if (arg2 || arg3 || arg4 || arg5)
2218 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2220 case PR_SET_THP_DISABLE:
2221 if (arg3 || arg4 || arg5)
2223 down_write(&me->mm->mmap_sem);
2225 me->mm->def_flags |= VM_NOHUGEPAGE;
2227 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2228 up_write(&me->mm->mmap_sem);
2230 case PR_MPX_ENABLE_MANAGEMENT:
2231 if (arg2 || arg3 || arg4 || arg5)
2233 error = MPX_ENABLE_MANAGEMENT(me);
2235 case PR_MPX_DISABLE_MANAGEMENT:
2236 if (arg2 || arg3 || arg4 || arg5)
2238 error = MPX_DISABLE_MANAGEMENT(me);
2240 case PR_SET_FP_MODE:
2241 error = SET_FP_MODE(me, arg2);
2243 case PR_GET_FP_MODE:
2244 error = GET_FP_MODE(me);
2253 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2254 struct getcpu_cache __user *, unused)
2257 int cpu = raw_smp_processor_id();
2260 err |= put_user(cpu, cpup);
2262 err |= put_user(cpu_to_node(cpu), nodep);
2263 return err ? -EFAULT : 0;
2267 * do_sysinfo - fill in sysinfo struct
2268 * @info: pointer to buffer to fill
2270 static int do_sysinfo(struct sysinfo *info)
2272 unsigned long mem_total, sav_total;
2273 unsigned int mem_unit, bitcount;
2276 memset(info, 0, sizeof(struct sysinfo));
2278 get_monotonic_boottime(&tp);
2279 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2281 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2283 info->procs = nr_threads;
2289 * If the sum of all the available memory (i.e. ram + swap)
2290 * is less than can be stored in a 32 bit unsigned long then
2291 * we can be binary compatible with 2.2.x kernels. If not,
2292 * well, in that case 2.2.x was broken anyways...
2294 * -Erik Andersen <andersee@debian.org>
2297 mem_total = info->totalram + info->totalswap;
2298 if (mem_total < info->totalram || mem_total < info->totalswap)
2301 mem_unit = info->mem_unit;
2302 while (mem_unit > 1) {
2305 sav_total = mem_total;
2307 if (mem_total < sav_total)
2312 * If mem_total did not overflow, multiply all memory values by
2313 * info->mem_unit and set it to 1. This leaves things compatible
2314 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2319 info->totalram <<= bitcount;
2320 info->freeram <<= bitcount;
2321 info->sharedram <<= bitcount;
2322 info->bufferram <<= bitcount;
2323 info->totalswap <<= bitcount;
2324 info->freeswap <<= bitcount;
2325 info->totalhigh <<= bitcount;
2326 info->freehigh <<= bitcount;
2332 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2338 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2344 #ifdef CONFIG_COMPAT
2345 struct compat_sysinfo {
2359 char _f[20-2*sizeof(u32)-sizeof(int)];
2362 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2368 /* Check to see if any memory value is too large for 32-bit and scale
2371 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2374 while (s.mem_unit < PAGE_SIZE) {
2379 s.totalram >>= bitcount;
2380 s.freeram >>= bitcount;
2381 s.sharedram >>= bitcount;
2382 s.bufferram >>= bitcount;
2383 s.totalswap >>= bitcount;
2384 s.freeswap >>= bitcount;
2385 s.totalhigh >>= bitcount;
2386 s.freehigh >>= bitcount;
2389 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2390 __put_user(s.uptime, &info->uptime) ||
2391 __put_user(s.loads[0], &info->loads[0]) ||
2392 __put_user(s.loads[1], &info->loads[1]) ||
2393 __put_user(s.loads[2], &info->loads[2]) ||
2394 __put_user(s.totalram, &info->totalram) ||
2395 __put_user(s.freeram, &info->freeram) ||
2396 __put_user(s.sharedram, &info->sharedram) ||
2397 __put_user(s.bufferram, &info->bufferram) ||
2398 __put_user(s.totalswap, &info->totalswap) ||
2399 __put_user(s.freeswap, &info->freeswap) ||
2400 __put_user(s.procs, &info->procs) ||
2401 __put_user(s.totalhigh, &info->totalhigh) ||
2402 __put_user(s.freehigh, &info->freehigh) ||
2403 __put_user(s.mem_unit, &info->mem_unit))
2408 #endif /* CONFIG_COMPAT */