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 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
330 struct user_namespace *ns = current_user_ns();
331 const struct cred *old;
336 krgid = make_kgid(ns, rgid);
337 kegid = make_kgid(ns, egid);
339 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
341 if ((egid != (gid_t) -1) && !gid_valid(kegid))
344 new = prepare_creds();
347 old = current_cred();
350 if (rgid != (gid_t) -1) {
351 if (gid_eq(old->gid, krgid) ||
352 gid_eq(old->egid, krgid) ||
353 ns_capable(old->user_ns, CAP_SETGID))
358 if (egid != (gid_t) -1) {
359 if (gid_eq(old->gid, kegid) ||
360 gid_eq(old->egid, kegid) ||
361 gid_eq(old->sgid, kegid) ||
362 ns_capable(old->user_ns, CAP_SETGID))
368 if (rgid != (gid_t) -1 ||
369 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
370 new->sgid = new->egid;
371 new->fsgid = new->egid;
373 return commit_creds(new);
381 * setgid() is implemented like SysV w/ SAVED_IDS
383 * SMP: Same implicit races as above.
385 SYSCALL_DEFINE1(setgid, gid_t, gid)
387 struct user_namespace *ns = current_user_ns();
388 const struct cred *old;
393 kgid = make_kgid(ns, gid);
394 if (!gid_valid(kgid))
397 new = prepare_creds();
400 old = current_cred();
403 if (ns_capable(old->user_ns, CAP_SETGID))
404 new->gid = new->egid = new->sgid = new->fsgid = kgid;
405 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
406 new->egid = new->fsgid = kgid;
410 return commit_creds(new);
418 * change the user struct in a credentials set to match the new UID
420 static int set_user(struct cred *new)
422 struct user_struct *new_user;
424 new_user = alloc_uid(new->uid);
429 * We don't fail in case of NPROC limit excess here because too many
430 * poorly written programs don't check set*uid() return code, assuming
431 * it never fails if called by root. We may still enforce NPROC limit
432 * for programs doing set*uid()+execve() by harmlessly deferring the
433 * failure to the execve() stage.
435 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
436 new_user != INIT_USER)
437 current->flags |= PF_NPROC_EXCEEDED;
439 current->flags &= ~PF_NPROC_EXCEEDED;
442 new->user = new_user;
447 * Unprivileged users may change the real uid to the effective uid
448 * or vice versa. (BSD-style)
450 * If you set the real uid at all, or set the effective uid to a value not
451 * equal to the real uid, then the saved uid is set to the new effective uid.
453 * This makes it possible for a setuid program to completely drop its
454 * privileges, which is often a useful assertion to make when you are doing
455 * a security audit over a program.
457 * The general idea is that a program which uses just setreuid() will be
458 * 100% compatible with BSD. A program which uses just setuid() will be
459 * 100% compatible with POSIX with saved IDs.
461 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
463 struct user_namespace *ns = current_user_ns();
464 const struct cred *old;
469 kruid = make_kuid(ns, ruid);
470 keuid = make_kuid(ns, euid);
472 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
474 if ((euid != (uid_t) -1) && !uid_valid(keuid))
477 new = prepare_creds();
480 old = current_cred();
483 if (ruid != (uid_t) -1) {
485 if (!uid_eq(old->uid, kruid) &&
486 !uid_eq(old->euid, kruid) &&
487 !ns_capable(old->user_ns, CAP_SETUID))
491 if (euid != (uid_t) -1) {
493 if (!uid_eq(old->uid, keuid) &&
494 !uid_eq(old->euid, keuid) &&
495 !uid_eq(old->suid, keuid) &&
496 !ns_capable(old->user_ns, CAP_SETUID))
500 if (!uid_eq(new->uid, old->uid)) {
501 retval = set_user(new);
505 if (ruid != (uid_t) -1 ||
506 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
507 new->suid = new->euid;
508 new->fsuid = new->euid;
510 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
514 return commit_creds(new);
522 * setuid() is implemented like SysV with SAVED_IDS
524 * Note that SAVED_ID's is deficient in that a setuid root program
525 * like sendmail, for example, cannot set its uid to be a normal
526 * user and then switch back, because if you're root, setuid() sets
527 * the saved uid too. If you don't like this, blame the bright people
528 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
529 * will allow a root program to temporarily drop privileges and be able to
530 * regain them by swapping the real and effective uid.
532 SYSCALL_DEFINE1(setuid, uid_t, uid)
534 struct user_namespace *ns = current_user_ns();
535 const struct cred *old;
540 kuid = make_kuid(ns, uid);
541 if (!uid_valid(kuid))
544 new = prepare_creds();
547 old = current_cred();
550 if (ns_capable(old->user_ns, CAP_SETUID)) {
551 new->suid = new->uid = kuid;
552 if (!uid_eq(kuid, old->uid)) {
553 retval = set_user(new);
557 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
561 new->fsuid = new->euid = kuid;
563 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
567 return commit_creds(new);
576 * This function implements a generic ability to update ruid, euid,
577 * and suid. This allows you to implement the 4.4 compatible seteuid().
579 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
581 struct user_namespace *ns = current_user_ns();
582 const struct cred *old;
585 kuid_t kruid, keuid, ksuid;
587 kruid = make_kuid(ns, ruid);
588 keuid = make_kuid(ns, euid);
589 ksuid = make_kuid(ns, suid);
591 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
594 if ((euid != (uid_t) -1) && !uid_valid(keuid))
597 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
600 new = prepare_creds();
604 old = current_cred();
607 if (!ns_capable(old->user_ns, CAP_SETUID)) {
608 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
609 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
611 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
612 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
614 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
615 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
619 if (ruid != (uid_t) -1) {
621 if (!uid_eq(kruid, old->uid)) {
622 retval = set_user(new);
627 if (euid != (uid_t) -1)
629 if (suid != (uid_t) -1)
631 new->fsuid = new->euid;
633 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
637 return commit_creds(new);
644 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
646 const struct cred *cred = current_cred();
648 uid_t ruid, euid, suid;
650 ruid = from_kuid_munged(cred->user_ns, cred->uid);
651 euid = from_kuid_munged(cred->user_ns, cred->euid);
652 suid = from_kuid_munged(cred->user_ns, cred->suid);
654 retval = put_user(ruid, ruidp);
656 retval = put_user(euid, euidp);
658 return put_user(suid, suidp);
664 * Same as above, but for rgid, egid, sgid.
666 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
668 struct user_namespace *ns = current_user_ns();
669 const struct cred *old;
672 kgid_t krgid, kegid, ksgid;
674 krgid = make_kgid(ns, rgid);
675 kegid = make_kgid(ns, egid);
676 ksgid = make_kgid(ns, sgid);
678 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
680 if ((egid != (gid_t) -1) && !gid_valid(kegid))
682 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
685 new = prepare_creds();
688 old = current_cred();
691 if (!ns_capable(old->user_ns, CAP_SETGID)) {
692 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
693 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
695 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
696 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
698 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
699 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
703 if (rgid != (gid_t) -1)
705 if (egid != (gid_t) -1)
707 if (sgid != (gid_t) -1)
709 new->fsgid = new->egid;
711 return commit_creds(new);
718 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
720 const struct cred *cred = current_cred();
722 gid_t rgid, egid, sgid;
724 rgid = from_kgid_munged(cred->user_ns, cred->gid);
725 egid = from_kgid_munged(cred->user_ns, cred->egid);
726 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
728 retval = put_user(rgid, rgidp);
730 retval = put_user(egid, egidp);
732 retval = put_user(sgid, sgidp);
740 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
741 * is used for "access()" and for the NFS daemon (letting nfsd stay at
742 * whatever uid it wants to). It normally shadows "euid", except when
743 * explicitly set by setfsuid() or for access..
745 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
747 const struct cred *old;
752 old = current_cred();
753 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
755 kuid = make_kuid(old->user_ns, uid);
756 if (!uid_valid(kuid))
759 new = prepare_creds();
763 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
764 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
765 ns_capable(old->user_ns, CAP_SETUID)) {
766 if (!uid_eq(kuid, old->fsuid)) {
768 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
782 * Samma på svenska..
784 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
786 const struct cred *old;
791 old = current_cred();
792 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
794 kgid = make_kgid(old->user_ns, gid);
795 if (!gid_valid(kgid))
798 new = prepare_creds();
802 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
803 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
804 ns_capable(old->user_ns, CAP_SETGID)) {
805 if (!gid_eq(kgid, old->fsgid)) {
820 * sys_getpid - return the thread group id of the current process
822 * Note, despite the name, this returns the tgid not the pid. The tgid and
823 * the pid are identical unless CLONE_THREAD was specified on clone() in
824 * which case the tgid is the same in all threads of the same group.
826 * This is SMP safe as current->tgid does not change.
828 SYSCALL_DEFINE0(getpid)
830 return task_tgid_vnr(current);
833 /* Thread ID - the internal kernel "pid" */
834 SYSCALL_DEFINE0(gettid)
836 return task_pid_vnr(current);
840 * Accessing ->real_parent is not SMP-safe, it could
841 * change from under us. However, we can use a stale
842 * value of ->real_parent under rcu_read_lock(), see
843 * release_task()->call_rcu(delayed_put_task_struct).
845 SYSCALL_DEFINE0(getppid)
850 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
856 SYSCALL_DEFINE0(getuid)
858 /* Only we change this so SMP safe */
859 return from_kuid_munged(current_user_ns(), current_uid());
862 SYSCALL_DEFINE0(geteuid)
864 /* Only we change this so SMP safe */
865 return from_kuid_munged(current_user_ns(), current_euid());
868 SYSCALL_DEFINE0(getgid)
870 /* Only we change this so SMP safe */
871 return from_kgid_munged(current_user_ns(), current_gid());
874 SYSCALL_DEFINE0(getegid)
876 /* Only we change this so SMP safe */
877 return from_kgid_munged(current_user_ns(), current_egid());
880 void do_sys_times(struct tms *tms)
882 cputime_t tgutime, tgstime, cutime, cstime;
884 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
885 cutime = current->signal->cutime;
886 cstime = current->signal->cstime;
887 tms->tms_utime = cputime_to_clock_t(tgutime);
888 tms->tms_stime = cputime_to_clock_t(tgstime);
889 tms->tms_cutime = cputime_to_clock_t(cutime);
890 tms->tms_cstime = cputime_to_clock_t(cstime);
893 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
899 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
902 force_successful_syscall_return();
903 return (long) jiffies_64_to_clock_t(get_jiffies_64());
907 * This needs some heavy checking ...
908 * I just haven't the stomach for it. I also don't fully
909 * understand sessions/pgrp etc. Let somebody who does explain it.
911 * OK, I think I have the protection semantics right.... this is really
912 * only important on a multi-user system anyway, to make sure one user
913 * can't send a signal to a process owned by another. -TYT, 12/12/91
915 * !PF_FORKNOEXEC check to conform completely to POSIX.
917 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
919 struct task_struct *p;
920 struct task_struct *group_leader = current->group_leader;
925 pid = task_pid_vnr(group_leader);
932 /* From this point forward we keep holding onto the tasklist lock
933 * so that our parent does not change from under us. -DaveM
935 write_lock_irq(&tasklist_lock);
938 p = find_task_by_vpid(pid);
943 if (!thread_group_leader(p))
946 if (same_thread_group(p->real_parent, group_leader)) {
948 if (task_session(p) != task_session(group_leader))
951 if (!(p->flags & PF_FORKNOEXEC))
955 if (p != group_leader)
960 if (p->signal->leader)
965 struct task_struct *g;
967 pgrp = find_vpid(pgid);
968 g = pid_task(pgrp, PIDTYPE_PGID);
969 if (!g || task_session(g) != task_session(group_leader))
973 err = security_task_setpgid(p, pgid);
977 if (task_pgrp(p) != pgrp)
978 change_pid(p, PIDTYPE_PGID, pgrp);
982 /* All paths lead to here, thus we are safe. -DaveM */
983 write_unlock_irq(&tasklist_lock);
988 SYSCALL_DEFINE1(getpgid, pid_t, pid)
990 struct task_struct *p;
996 grp = task_pgrp(current);
999 p = find_task_by_vpid(pid);
1006 retval = security_task_getpgid(p);
1010 retval = pid_vnr(grp);
1016 #ifdef __ARCH_WANT_SYS_GETPGRP
1018 SYSCALL_DEFINE0(getpgrp)
1020 return sys_getpgid(0);
1025 SYSCALL_DEFINE1(getsid, pid_t, pid)
1027 struct task_struct *p;
1033 sid = task_session(current);
1036 p = find_task_by_vpid(pid);
1039 sid = task_session(p);
1043 retval = security_task_getsid(p);
1047 retval = pid_vnr(sid);
1053 static void set_special_pids(struct pid *pid)
1055 struct task_struct *curr = current->group_leader;
1057 if (task_session(curr) != pid)
1058 change_pid(curr, PIDTYPE_SID, pid);
1060 if (task_pgrp(curr) != pid)
1061 change_pid(curr, PIDTYPE_PGID, pid);
1064 SYSCALL_DEFINE0(setsid)
1066 struct task_struct *group_leader = current->group_leader;
1067 struct pid *sid = task_pid(group_leader);
1068 pid_t session = pid_vnr(sid);
1071 write_lock_irq(&tasklist_lock);
1072 /* Fail if I am already a session leader */
1073 if (group_leader->signal->leader)
1076 /* Fail if a process group id already exists that equals the
1077 * proposed session id.
1079 if (pid_task(sid, PIDTYPE_PGID))
1082 group_leader->signal->leader = 1;
1083 set_special_pids(sid);
1085 proc_clear_tty(group_leader);
1089 write_unlock_irq(&tasklist_lock);
1091 proc_sid_connector(group_leader);
1092 sched_autogroup_create_attach(group_leader);
1097 DECLARE_RWSEM(uts_sem);
1099 #ifdef COMPAT_UTS_MACHINE
1100 #define override_architecture(name) \
1101 (personality(current->personality) == PER_LINUX32 && \
1102 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1103 sizeof(COMPAT_UTS_MACHINE)))
1105 #define override_architecture(name) 0
1109 * Work around broken programs that cannot handle "Linux 3.0".
1110 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1111 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1113 static int override_release(char __user *release, size_t len)
1117 if (current->personality & UNAME26) {
1118 const char *rest = UTS_RELEASE;
1119 char buf[65] = { 0 };
1125 if (*rest == '.' && ++ndots >= 3)
1127 if (!isdigit(*rest) && *rest != '.')
1131 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1132 copy = clamp_t(size_t, len, 1, sizeof(buf));
1133 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1134 ret = copy_to_user(release, buf, copy + 1);
1139 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1143 down_read(&uts_sem);
1144 if (copy_to_user(name, utsname(), sizeof *name))
1148 if (!errno && override_release(name->release, sizeof(name->release)))
1150 if (!errno && override_architecture(name))
1155 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1159 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1166 down_read(&uts_sem);
1167 if (copy_to_user(name, utsname(), sizeof(*name)))
1171 if (!error && override_release(name->release, sizeof(name->release)))
1173 if (!error && override_architecture(name))
1178 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1184 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1187 down_read(&uts_sem);
1188 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1190 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1191 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1193 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1194 error |= __copy_to_user(&name->release, &utsname()->release,
1196 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1197 error |= __copy_to_user(&name->version, &utsname()->version,
1199 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1200 error |= __copy_to_user(&name->machine, &utsname()->machine,
1202 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1205 if (!error && override_architecture(name))
1207 if (!error && override_release(name->release, sizeof(name->release)))
1209 return error ? -EFAULT : 0;
1213 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1216 char tmp[__NEW_UTS_LEN];
1218 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1221 if (len < 0 || len > __NEW_UTS_LEN)
1223 down_write(&uts_sem);
1225 if (!copy_from_user(tmp, name, len)) {
1226 struct new_utsname *u = utsname();
1228 memcpy(u->nodename, tmp, len);
1229 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1231 uts_proc_notify(UTS_PROC_HOSTNAME);
1237 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1239 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1242 struct new_utsname *u;
1246 down_read(&uts_sem);
1248 i = 1 + strlen(u->nodename);
1252 if (copy_to_user(name, u->nodename, i))
1261 * Only setdomainname; getdomainname can be implemented by calling
1264 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1267 char tmp[__NEW_UTS_LEN];
1269 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1271 if (len < 0 || len > __NEW_UTS_LEN)
1274 down_write(&uts_sem);
1276 if (!copy_from_user(tmp, name, len)) {
1277 struct new_utsname *u = utsname();
1279 memcpy(u->domainname, tmp, len);
1280 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1282 uts_proc_notify(UTS_PROC_DOMAINNAME);
1288 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1290 struct rlimit value;
1293 ret = do_prlimit(current, resource, NULL, &value);
1295 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1300 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1303 * Back compatibility for getrlimit. Needed for some apps.
1305 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1306 struct rlimit __user *, rlim)
1309 if (resource >= RLIM_NLIMITS)
1312 task_lock(current->group_leader);
1313 x = current->signal->rlim[resource];
1314 task_unlock(current->group_leader);
1315 if (x.rlim_cur > 0x7FFFFFFF)
1316 x.rlim_cur = 0x7FFFFFFF;
1317 if (x.rlim_max > 0x7FFFFFFF)
1318 x.rlim_max = 0x7FFFFFFF;
1319 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1324 static inline bool rlim64_is_infinity(__u64 rlim64)
1326 #if BITS_PER_LONG < 64
1327 return rlim64 >= ULONG_MAX;
1329 return rlim64 == RLIM64_INFINITY;
1333 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1335 if (rlim->rlim_cur == RLIM_INFINITY)
1336 rlim64->rlim_cur = RLIM64_INFINITY;
1338 rlim64->rlim_cur = rlim->rlim_cur;
1339 if (rlim->rlim_max == RLIM_INFINITY)
1340 rlim64->rlim_max = RLIM64_INFINITY;
1342 rlim64->rlim_max = rlim->rlim_max;
1345 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1347 if (rlim64_is_infinity(rlim64->rlim_cur))
1348 rlim->rlim_cur = RLIM_INFINITY;
1350 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1351 if (rlim64_is_infinity(rlim64->rlim_max))
1352 rlim->rlim_max = RLIM_INFINITY;
1354 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1357 /* make sure you are allowed to change @tsk limits before calling this */
1358 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1359 struct rlimit *new_rlim, struct rlimit *old_rlim)
1361 struct rlimit *rlim;
1364 if (resource >= RLIM_NLIMITS)
1367 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1369 if (resource == RLIMIT_NOFILE &&
1370 new_rlim->rlim_max > sysctl_nr_open)
1374 /* protect tsk->signal and tsk->sighand from disappearing */
1375 read_lock(&tasklist_lock);
1376 if (!tsk->sighand) {
1381 rlim = tsk->signal->rlim + resource;
1382 task_lock(tsk->group_leader);
1384 /* Keep the capable check against init_user_ns until
1385 cgroups can contain all limits */
1386 if (new_rlim->rlim_max > rlim->rlim_max &&
1387 !capable(CAP_SYS_RESOURCE))
1390 retval = security_task_setrlimit(tsk->group_leader,
1391 resource, new_rlim);
1392 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1394 * The caller is asking for an immediate RLIMIT_CPU
1395 * expiry. But we use the zero value to mean "it was
1396 * never set". So let's cheat and make it one second
1399 new_rlim->rlim_cur = 1;
1408 task_unlock(tsk->group_leader);
1411 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1412 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1413 * very long-standing error, and fixing it now risks breakage of
1414 * applications, so we live with it
1416 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1417 new_rlim->rlim_cur != RLIM_INFINITY)
1418 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1420 read_unlock(&tasklist_lock);
1424 /* rcu lock must be held */
1425 static int check_prlimit_permission(struct task_struct *task)
1427 const struct cred *cred = current_cred(), *tcred;
1429 if (current == task)
1432 tcred = __task_cred(task);
1433 if (uid_eq(cred->uid, tcred->euid) &&
1434 uid_eq(cred->uid, tcred->suid) &&
1435 uid_eq(cred->uid, tcred->uid) &&
1436 gid_eq(cred->gid, tcred->egid) &&
1437 gid_eq(cred->gid, tcred->sgid) &&
1438 gid_eq(cred->gid, tcred->gid))
1440 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1446 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1447 const struct rlimit64 __user *, new_rlim,
1448 struct rlimit64 __user *, old_rlim)
1450 struct rlimit64 old64, new64;
1451 struct rlimit old, new;
1452 struct task_struct *tsk;
1456 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1458 rlim64_to_rlim(&new64, &new);
1462 tsk = pid ? find_task_by_vpid(pid) : current;
1467 ret = check_prlimit_permission(tsk);
1472 get_task_struct(tsk);
1475 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1476 old_rlim ? &old : NULL);
1478 if (!ret && old_rlim) {
1479 rlim_to_rlim64(&old, &old64);
1480 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1484 put_task_struct(tsk);
1488 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1490 struct rlimit new_rlim;
1492 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1494 return do_prlimit(current, resource, &new_rlim, NULL);
1498 * It would make sense to put struct rusage in the task_struct,
1499 * except that would make the task_struct be *really big*. After
1500 * task_struct gets moved into malloc'ed memory, it would
1501 * make sense to do this. It will make moving the rest of the information
1502 * a lot simpler! (Which we're not doing right now because we're not
1503 * measuring them yet).
1505 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1506 * races with threads incrementing their own counters. But since word
1507 * reads are atomic, we either get new values or old values and we don't
1508 * care which for the sums. We always take the siglock to protect reading
1509 * the c* fields from p->signal from races with exit.c updating those
1510 * fields when reaping, so a sample either gets all the additions of a
1511 * given child after it's reaped, or none so this sample is before reaping.
1514 * We need to take the siglock for CHILDEREN, SELF and BOTH
1515 * for the cases current multithreaded, non-current single threaded
1516 * non-current multithreaded. Thread traversal is now safe with
1518 * Strictly speaking, we donot need to take the siglock if we are current and
1519 * single threaded, as no one else can take our signal_struct away, no one
1520 * else can reap the children to update signal->c* counters, and no one else
1521 * can race with the signal-> fields. If we do not take any lock, the
1522 * signal-> fields could be read out of order while another thread was just
1523 * exiting. So we should place a read memory barrier when we avoid the lock.
1524 * On the writer side, write memory barrier is implied in __exit_signal
1525 * as __exit_signal releases the siglock spinlock after updating the signal->
1526 * fields. But we don't do this yet to keep things simple.
1530 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1532 r->ru_nvcsw += t->nvcsw;
1533 r->ru_nivcsw += t->nivcsw;
1534 r->ru_minflt += t->min_flt;
1535 r->ru_majflt += t->maj_flt;
1536 r->ru_inblock += task_io_get_inblock(t);
1537 r->ru_oublock += task_io_get_oublock(t);
1540 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1542 struct task_struct *t;
1543 unsigned long flags;
1544 cputime_t tgutime, tgstime, utime, stime;
1545 unsigned long maxrss = 0;
1547 memset((char *)r, 0, sizeof (*r));
1550 if (who == RUSAGE_THREAD) {
1551 task_cputime_adjusted(current, &utime, &stime);
1552 accumulate_thread_rusage(p, r);
1553 maxrss = p->signal->maxrss;
1557 if (!lock_task_sighand(p, &flags))
1562 case RUSAGE_CHILDREN:
1563 utime = p->signal->cutime;
1564 stime = p->signal->cstime;
1565 r->ru_nvcsw = p->signal->cnvcsw;
1566 r->ru_nivcsw = p->signal->cnivcsw;
1567 r->ru_minflt = p->signal->cmin_flt;
1568 r->ru_majflt = p->signal->cmaj_flt;
1569 r->ru_inblock = p->signal->cinblock;
1570 r->ru_oublock = p->signal->coublock;
1571 maxrss = p->signal->cmaxrss;
1573 if (who == RUSAGE_CHILDREN)
1577 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1580 r->ru_nvcsw += p->signal->nvcsw;
1581 r->ru_nivcsw += p->signal->nivcsw;
1582 r->ru_minflt += p->signal->min_flt;
1583 r->ru_majflt += p->signal->maj_flt;
1584 r->ru_inblock += p->signal->inblock;
1585 r->ru_oublock += p->signal->oublock;
1586 if (maxrss < p->signal->maxrss)
1587 maxrss = p->signal->maxrss;
1590 accumulate_thread_rusage(t, r);
1591 } while_each_thread(p, t);
1597 unlock_task_sighand(p, &flags);
1600 cputime_to_timeval(utime, &r->ru_utime);
1601 cputime_to_timeval(stime, &r->ru_stime);
1603 if (who != RUSAGE_CHILDREN) {
1604 struct mm_struct *mm = get_task_mm(p);
1607 setmax_mm_hiwater_rss(&maxrss, mm);
1611 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1614 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1618 k_getrusage(p, who, &r);
1619 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1622 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1624 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1625 who != RUSAGE_THREAD)
1627 return getrusage(current, who, ru);
1630 #ifdef CONFIG_COMPAT
1631 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1635 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1636 who != RUSAGE_THREAD)
1639 k_getrusage(current, who, &r);
1640 return put_compat_rusage(&r, ru);
1644 SYSCALL_DEFINE1(umask, int, mask)
1646 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1650 static int prctl_set_mm_exe_file_locked(struct mm_struct *mm, unsigned int fd)
1653 struct inode *inode;
1656 VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
1662 inode = file_inode(exe.file);
1665 * Because the original mm->exe_file points to executable file, make
1666 * sure that this one is executable as well, to avoid breaking an
1670 if (!S_ISREG(inode->i_mode) ||
1671 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1674 err = inode_permission(inode, MAY_EXEC);
1679 * Forbid mm->exe_file change if old file still mapped.
1683 struct vm_area_struct *vma;
1685 for (vma = mm->mmap; vma; vma = vma->vm_next)
1687 path_equal(&vma->vm_file->f_path,
1688 &mm->exe_file->f_path))
1693 * The symlink can be changed only once, just to disallow arbitrary
1694 * transitions malicious software might bring in. This means one
1695 * could make a snapshot over all processes running and monitor
1696 * /proc/pid/exe changes to notice unusual activity if needed.
1699 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1703 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1709 #ifdef CONFIG_CHECKPOINT_RESTORE
1711 * WARNING: we don't require any capability here so be very careful
1712 * in what is allowed for modification from userspace.
1714 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1716 unsigned long mmap_max_addr = TASK_SIZE;
1717 struct mm_struct *mm = current->mm;
1718 int error = -EINVAL, i;
1720 static const unsigned char offsets[] = {
1721 offsetof(struct prctl_mm_map, start_code),
1722 offsetof(struct prctl_mm_map, end_code),
1723 offsetof(struct prctl_mm_map, start_data),
1724 offsetof(struct prctl_mm_map, end_data),
1725 offsetof(struct prctl_mm_map, start_brk),
1726 offsetof(struct prctl_mm_map, brk),
1727 offsetof(struct prctl_mm_map, start_stack),
1728 offsetof(struct prctl_mm_map, arg_start),
1729 offsetof(struct prctl_mm_map, arg_end),
1730 offsetof(struct prctl_mm_map, env_start),
1731 offsetof(struct prctl_mm_map, env_end),
1735 * Make sure the members are not somewhere outside
1736 * of allowed address space.
1738 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1739 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1741 if ((unsigned long)val >= mmap_max_addr ||
1742 (unsigned long)val < mmap_min_addr)
1747 * Make sure the pairs are ordered.
1749 #define __prctl_check_order(__m1, __op, __m2) \
1750 ((unsigned long)prctl_map->__m1 __op \
1751 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1752 error = __prctl_check_order(start_code, <, end_code);
1753 error |= __prctl_check_order(start_data, <, end_data);
1754 error |= __prctl_check_order(start_brk, <=, brk);
1755 error |= __prctl_check_order(arg_start, <=, arg_end);
1756 error |= __prctl_check_order(env_start, <=, env_end);
1759 #undef __prctl_check_order
1764 * @brk should be after @end_data in traditional maps.
1766 if (prctl_map->start_brk <= prctl_map->end_data ||
1767 prctl_map->brk <= prctl_map->end_data)
1771 * Neither we should allow to override limits if they set.
1773 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1774 prctl_map->start_brk, prctl_map->end_data,
1775 prctl_map->start_data))
1779 * Someone is trying to cheat the auxv vector.
1781 if (prctl_map->auxv_size) {
1782 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1787 * Finally, make sure the caller has the rights to
1788 * change /proc/pid/exe link: only local root should
1791 if (prctl_map->exe_fd != (u32)-1) {
1792 struct user_namespace *ns = current_user_ns();
1793 const struct cred *cred = current_cred();
1795 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1796 !gid_eq(cred->gid, make_kgid(ns, 0)))
1805 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1807 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1808 unsigned long user_auxv[AT_VECTOR_SIZE];
1809 struct mm_struct *mm = current->mm;
1812 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1813 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1815 if (opt == PR_SET_MM_MAP_SIZE)
1816 return put_user((unsigned int)sizeof(prctl_map),
1817 (unsigned int __user *)addr);
1819 if (data_size != sizeof(prctl_map))
1822 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1825 error = validate_prctl_map(&prctl_map);
1829 if (prctl_map.auxv_size) {
1830 memset(user_auxv, 0, sizeof(user_auxv));
1831 if (copy_from_user(user_auxv,
1832 (const void __user *)prctl_map.auxv,
1833 prctl_map.auxv_size))
1836 /* Last entry must be AT_NULL as specification requires */
1837 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1838 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1841 down_write(&mm->mmap_sem);
1842 if (prctl_map.exe_fd != (u32)-1)
1843 error = prctl_set_mm_exe_file_locked(mm, prctl_map.exe_fd);
1844 downgrade_write(&mm->mmap_sem);
1849 * We don't validate if these members are pointing to
1850 * real present VMAs because application may have correspond
1851 * VMAs already unmapped and kernel uses these members for statistics
1852 * output in procfs mostly, except
1854 * - @start_brk/@brk which are used in do_brk but kernel lookups
1855 * for VMAs when updating these memvers so anything wrong written
1856 * here cause kernel to swear at userspace program but won't lead
1857 * to any problem in kernel itself
1860 mm->start_code = prctl_map.start_code;
1861 mm->end_code = prctl_map.end_code;
1862 mm->start_data = prctl_map.start_data;
1863 mm->end_data = prctl_map.end_data;
1864 mm->start_brk = prctl_map.start_brk;
1865 mm->brk = prctl_map.brk;
1866 mm->start_stack = prctl_map.start_stack;
1867 mm->arg_start = prctl_map.arg_start;
1868 mm->arg_end = prctl_map.arg_end;
1869 mm->env_start = prctl_map.env_start;
1870 mm->env_end = prctl_map.env_end;
1873 * Note this update of @saved_auxv is lockless thus
1874 * if someone reads this member in procfs while we're
1875 * updating -- it may get partly updated results. It's
1876 * known and acceptable trade off: we leave it as is to
1877 * not introduce additional locks here making the kernel
1880 if (prctl_map.auxv_size)
1881 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1885 up_read(&mm->mmap_sem);
1888 #endif /* CONFIG_CHECKPOINT_RESTORE */
1890 static int prctl_set_mm(int opt, unsigned long addr,
1891 unsigned long arg4, unsigned long arg5)
1893 struct mm_struct *mm = current->mm;
1894 struct vm_area_struct *vma;
1897 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1898 opt != PR_SET_MM_MAP &&
1899 opt != PR_SET_MM_MAP_SIZE)))
1902 #ifdef CONFIG_CHECKPOINT_RESTORE
1903 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1904 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1907 if (!capable(CAP_SYS_RESOURCE))
1910 if (opt == PR_SET_MM_EXE_FILE) {
1911 down_write(&mm->mmap_sem);
1912 error = prctl_set_mm_exe_file_locked(mm, (unsigned int)addr);
1913 up_write(&mm->mmap_sem);
1917 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1922 down_read(&mm->mmap_sem);
1923 vma = find_vma(mm, addr);
1926 case PR_SET_MM_START_CODE:
1927 mm->start_code = addr;
1929 case PR_SET_MM_END_CODE:
1930 mm->end_code = addr;
1932 case PR_SET_MM_START_DATA:
1933 mm->start_data = addr;
1935 case PR_SET_MM_END_DATA:
1936 mm->end_data = addr;
1939 case PR_SET_MM_START_BRK:
1940 if (addr <= mm->end_data)
1943 if (check_data_rlimit(rlimit(RLIMIT_DATA), mm->brk, addr,
1944 mm->end_data, mm->start_data))
1947 mm->start_brk = addr;
1951 if (addr <= mm->end_data)
1954 if (check_data_rlimit(rlimit(RLIMIT_DATA), addr, mm->start_brk,
1955 mm->end_data, mm->start_data))
1962 * If command line arguments and environment
1963 * are placed somewhere else on stack, we can
1964 * set them up here, ARG_START/END to setup
1965 * command line argumets and ENV_START/END
1968 case PR_SET_MM_START_STACK:
1969 case PR_SET_MM_ARG_START:
1970 case PR_SET_MM_ARG_END:
1971 case PR_SET_MM_ENV_START:
1972 case PR_SET_MM_ENV_END:
1977 if (opt == PR_SET_MM_START_STACK)
1978 mm->start_stack = addr;
1979 else if (opt == PR_SET_MM_ARG_START)
1980 mm->arg_start = addr;
1981 else if (opt == PR_SET_MM_ARG_END)
1983 else if (opt == PR_SET_MM_ENV_START)
1984 mm->env_start = addr;
1985 else if (opt == PR_SET_MM_ENV_END)
1990 * This doesn't move auxiliary vector itself
1991 * since it's pinned to mm_struct, but allow
1992 * to fill vector with new values. It's up
1993 * to a caller to provide sane values here
1994 * otherwise user space tools which use this
1995 * vector might be unhappy.
1997 case PR_SET_MM_AUXV: {
1998 unsigned long user_auxv[AT_VECTOR_SIZE];
2000 if (arg4 > sizeof(user_auxv))
2002 up_read(&mm->mmap_sem);
2004 if (copy_from_user(user_auxv, (const void __user *)addr, arg4))
2007 /* Make sure the last entry is always AT_NULL */
2008 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2009 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2011 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2014 memcpy(mm->saved_auxv, user_auxv, arg4);
2015 task_unlock(current);
2025 up_read(&mm->mmap_sem);
2029 #ifdef CONFIG_CHECKPOINT_RESTORE
2030 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2032 return put_user(me->clear_child_tid, tid_addr);
2035 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2041 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2042 unsigned long, arg4, unsigned long, arg5)
2044 struct task_struct *me = current;
2045 unsigned char comm[sizeof(me->comm)];
2048 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2049 if (error != -ENOSYS)
2054 case PR_SET_PDEATHSIG:
2055 if (!valid_signal(arg2)) {
2059 me->pdeath_signal = arg2;
2061 case PR_GET_PDEATHSIG:
2062 error = put_user(me->pdeath_signal, (int __user *)arg2);
2064 case PR_GET_DUMPABLE:
2065 error = get_dumpable(me->mm);
2067 case PR_SET_DUMPABLE:
2068 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2072 set_dumpable(me->mm, arg2);
2075 case PR_SET_UNALIGN:
2076 error = SET_UNALIGN_CTL(me, arg2);
2078 case PR_GET_UNALIGN:
2079 error = GET_UNALIGN_CTL(me, arg2);
2082 error = SET_FPEMU_CTL(me, arg2);
2085 error = GET_FPEMU_CTL(me, arg2);
2088 error = SET_FPEXC_CTL(me, arg2);
2091 error = GET_FPEXC_CTL(me, arg2);
2094 error = PR_TIMING_STATISTICAL;
2097 if (arg2 != PR_TIMING_STATISTICAL)
2101 comm[sizeof(me->comm) - 1] = 0;
2102 if (strncpy_from_user(comm, (char __user *)arg2,
2103 sizeof(me->comm) - 1) < 0)
2105 set_task_comm(me, comm);
2106 proc_comm_connector(me);
2109 get_task_comm(comm, me);
2110 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2114 error = GET_ENDIAN(me, arg2);
2117 error = SET_ENDIAN(me, arg2);
2119 case PR_GET_SECCOMP:
2120 error = prctl_get_seccomp();
2122 case PR_SET_SECCOMP:
2123 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2126 error = GET_TSC_CTL(arg2);
2129 error = SET_TSC_CTL(arg2);
2131 case PR_TASK_PERF_EVENTS_DISABLE:
2132 error = perf_event_task_disable();
2134 case PR_TASK_PERF_EVENTS_ENABLE:
2135 error = perf_event_task_enable();
2137 case PR_GET_TIMERSLACK:
2138 error = current->timer_slack_ns;
2140 case PR_SET_TIMERSLACK:
2142 current->timer_slack_ns =
2143 current->default_timer_slack_ns;
2145 current->timer_slack_ns = arg2;
2151 case PR_MCE_KILL_CLEAR:
2154 current->flags &= ~PF_MCE_PROCESS;
2156 case PR_MCE_KILL_SET:
2157 current->flags |= PF_MCE_PROCESS;
2158 if (arg3 == PR_MCE_KILL_EARLY)
2159 current->flags |= PF_MCE_EARLY;
2160 else if (arg3 == PR_MCE_KILL_LATE)
2161 current->flags &= ~PF_MCE_EARLY;
2162 else if (arg3 == PR_MCE_KILL_DEFAULT)
2164 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2172 case PR_MCE_KILL_GET:
2173 if (arg2 | arg3 | arg4 | arg5)
2175 if (current->flags & PF_MCE_PROCESS)
2176 error = (current->flags & PF_MCE_EARLY) ?
2177 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2179 error = PR_MCE_KILL_DEFAULT;
2182 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2184 case PR_GET_TID_ADDRESS:
2185 error = prctl_get_tid_address(me, (int __user **)arg2);
2187 case PR_SET_CHILD_SUBREAPER:
2188 me->signal->is_child_subreaper = !!arg2;
2190 case PR_GET_CHILD_SUBREAPER:
2191 error = put_user(me->signal->is_child_subreaper,
2192 (int __user *)arg2);
2194 case PR_SET_NO_NEW_PRIVS:
2195 if (arg2 != 1 || arg3 || arg4 || arg5)
2198 task_set_no_new_privs(current);
2200 case PR_GET_NO_NEW_PRIVS:
2201 if (arg2 || arg3 || arg4 || arg5)
2203 return task_no_new_privs(current) ? 1 : 0;
2204 case PR_GET_THP_DISABLE:
2205 if (arg2 || arg3 || arg4 || arg5)
2207 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2209 case PR_SET_THP_DISABLE:
2210 if (arg3 || arg4 || arg5)
2212 down_write(&me->mm->mmap_sem);
2214 me->mm->def_flags |= VM_NOHUGEPAGE;
2216 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2217 up_write(&me->mm->mmap_sem);
2219 case PR_MPX_ENABLE_MANAGEMENT:
2220 if (arg2 || arg3 || arg4 || arg5)
2222 error = MPX_ENABLE_MANAGEMENT(me);
2224 case PR_MPX_DISABLE_MANAGEMENT:
2225 if (arg2 || arg3 || arg4 || arg5)
2227 error = MPX_DISABLE_MANAGEMENT(me);
2229 case PR_SET_FP_MODE:
2230 error = SET_FP_MODE(me, arg2);
2232 case PR_GET_FP_MODE:
2233 error = GET_FP_MODE(me);
2242 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2243 struct getcpu_cache __user *, unused)
2246 int cpu = raw_smp_processor_id();
2249 err |= put_user(cpu, cpup);
2251 err |= put_user(cpu_to_node(cpu), nodep);
2252 return err ? -EFAULT : 0;
2256 * do_sysinfo - fill in sysinfo struct
2257 * @info: pointer to buffer to fill
2259 static int do_sysinfo(struct sysinfo *info)
2261 unsigned long mem_total, sav_total;
2262 unsigned int mem_unit, bitcount;
2265 memset(info, 0, sizeof(struct sysinfo));
2267 get_monotonic_boottime(&tp);
2268 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2270 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2272 info->procs = nr_threads;
2278 * If the sum of all the available memory (i.e. ram + swap)
2279 * is less than can be stored in a 32 bit unsigned long then
2280 * we can be binary compatible with 2.2.x kernels. If not,
2281 * well, in that case 2.2.x was broken anyways...
2283 * -Erik Andersen <andersee@debian.org>
2286 mem_total = info->totalram + info->totalswap;
2287 if (mem_total < info->totalram || mem_total < info->totalswap)
2290 mem_unit = info->mem_unit;
2291 while (mem_unit > 1) {
2294 sav_total = mem_total;
2296 if (mem_total < sav_total)
2301 * If mem_total did not overflow, multiply all memory values by
2302 * info->mem_unit and set it to 1. This leaves things compatible
2303 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2308 info->totalram <<= bitcount;
2309 info->freeram <<= bitcount;
2310 info->sharedram <<= bitcount;
2311 info->bufferram <<= bitcount;
2312 info->totalswap <<= bitcount;
2313 info->freeswap <<= bitcount;
2314 info->totalhigh <<= bitcount;
2315 info->freehigh <<= bitcount;
2321 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2327 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2333 #ifdef CONFIG_COMPAT
2334 struct compat_sysinfo {
2348 char _f[20-2*sizeof(u32)-sizeof(int)];
2351 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2357 /* Check to see if any memory value is too large for 32-bit and scale
2360 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2363 while (s.mem_unit < PAGE_SIZE) {
2368 s.totalram >>= bitcount;
2369 s.freeram >>= bitcount;
2370 s.sharedram >>= bitcount;
2371 s.bufferram >>= bitcount;
2372 s.totalswap >>= bitcount;
2373 s.freeswap >>= bitcount;
2374 s.totalhigh >>= bitcount;
2375 s.freehigh >>= bitcount;
2378 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2379 __put_user(s.uptime, &info->uptime) ||
2380 __put_user(s.loads[0], &info->loads[0]) ||
2381 __put_user(s.loads[1], &info->loads[1]) ||
2382 __put_user(s.loads[2], &info->loads[2]) ||
2383 __put_user(s.totalram, &info->totalram) ||
2384 __put_user(s.freeram, &info->freeram) ||
2385 __put_user(s.sharedram, &info->sharedram) ||
2386 __put_user(s.bufferram, &info->bufferram) ||
2387 __put_user(s.totalswap, &info->totalswap) ||
2388 __put_user(s.freeswap, &info->freeswap) ||
2389 __put_user(s.procs, &info->procs) ||
2390 __put_user(s.totalhigh, &info->totalhigh) ||
2391 __put_user(s.freehigh, &info->freehigh) ||
2392 __put_user(s.mem_unit, &info->mem_unit))
2397 #endif /* CONFIG_COMPAT */