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/mempolicy.h>
47 #include <linux/compat.h>
48 #include <linux/syscalls.h>
49 #include <linux/kprobes.h>
50 #include <linux/user_namespace.h>
51 #include <linux/binfmts.h>
53 #include <linux/sched.h>
54 #include <linux/rcupdate.h>
55 #include <linux/uidgid.h>
56 #include <linux/cred.h>
58 #include <linux/kmsg_dump.h>
59 /* Move somewhere else to avoid recompiling? */
60 #include <generated/utsrelease.h>
62 #include <asm/uaccess.h>
64 #include <asm/unistd.h>
66 #ifndef SET_UNALIGN_CTL
67 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
69 #ifndef GET_UNALIGN_CTL
70 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
73 # define SET_FPEMU_CTL(a, b) (-EINVAL)
76 # define GET_FPEMU_CTL(a, b) (-EINVAL)
79 # define SET_FPEXC_CTL(a, b) (-EINVAL)
82 # define GET_FPEXC_CTL(a, b) (-EINVAL)
85 # define GET_ENDIAN(a, b) (-EINVAL)
88 # define SET_ENDIAN(a, b) (-EINVAL)
91 # define GET_TSC_CTL(a) (-EINVAL)
94 # define SET_TSC_CTL(a) (-EINVAL)
96 #ifndef MPX_ENABLE_MANAGEMENT
97 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
99 #ifndef MPX_DISABLE_MANAGEMENT
100 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
103 # define GET_FP_MODE(a) (-EINVAL)
106 # define SET_FP_MODE(a,b) (-EINVAL)
110 * this is where the system-wide overflow UID and GID are defined, for
111 * architectures that now have 32-bit UID/GID but didn't in the past
114 int overflowuid = DEFAULT_OVERFLOWUID;
115 int overflowgid = DEFAULT_OVERFLOWGID;
117 EXPORT_SYMBOL(overflowuid);
118 EXPORT_SYMBOL(overflowgid);
121 * the same as above, but for filesystems which can only store a 16-bit
122 * UID and GID. as such, this is needed on all architectures
125 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
126 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
128 EXPORT_SYMBOL(fs_overflowuid);
129 EXPORT_SYMBOL(fs_overflowgid);
132 * Returns true if current's euid is same as p's uid or euid,
133 * or has CAP_SYS_NICE to p's user_ns.
135 * Called with rcu_read_lock, creds are safe
137 static bool set_one_prio_perm(struct task_struct *p)
139 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
141 if (uid_eq(pcred->uid, cred->euid) ||
142 uid_eq(pcred->euid, cred->euid))
144 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
150 * set the priority of a task
151 * - the caller must hold the RCU read lock
153 static int set_one_prio(struct task_struct *p, int niceval, int error)
157 if (!set_one_prio_perm(p)) {
161 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
165 no_nice = security_task_setnice(p, niceval);
172 set_user_nice(p, niceval);
177 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
179 struct task_struct *g, *p;
180 struct user_struct *user;
181 const struct cred *cred = current_cred();
186 if (which > PRIO_USER || which < PRIO_PROCESS)
189 /* normalize: avoid signed division (rounding problems) */
191 if (niceval < MIN_NICE)
193 if (niceval > MAX_NICE)
197 read_lock(&tasklist_lock);
201 p = find_task_by_vpid(who);
205 error = set_one_prio(p, niceval, error);
209 pgrp = find_vpid(who);
211 pgrp = task_pgrp(current);
212 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
213 error = set_one_prio(p, niceval, error);
214 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
217 uid = make_kuid(cred->user_ns, who);
221 else if (!uid_eq(uid, cred->uid)) {
222 user = find_user(uid);
224 goto out_unlock; /* No processes for this user */
226 do_each_thread(g, p) {
227 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
228 error = set_one_prio(p, niceval, error);
229 } while_each_thread(g, p);
230 if (!uid_eq(uid, cred->uid))
231 free_uid(user); /* For find_user() */
235 read_unlock(&tasklist_lock);
242 * Ugh. To avoid negative return values, "getpriority()" will
243 * not return the normal nice-value, but a negated value that
244 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
245 * to stay compatible.
247 SYSCALL_DEFINE2(getpriority, int, which, int, who)
249 struct task_struct *g, *p;
250 struct user_struct *user;
251 const struct cred *cred = current_cred();
252 long niceval, retval = -ESRCH;
256 if (which > PRIO_USER || which < PRIO_PROCESS)
260 read_lock(&tasklist_lock);
264 p = find_task_by_vpid(who);
268 niceval = nice_to_rlimit(task_nice(p));
269 if (niceval > retval)
275 pgrp = find_vpid(who);
277 pgrp = task_pgrp(current);
278 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
279 niceval = nice_to_rlimit(task_nice(p));
280 if (niceval > retval)
282 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
285 uid = make_kuid(cred->user_ns, who);
289 else if (!uid_eq(uid, cred->uid)) {
290 user = find_user(uid);
292 goto out_unlock; /* No processes for this user */
294 do_each_thread(g, p) {
295 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
296 niceval = nice_to_rlimit(task_nice(p));
297 if (niceval > retval)
300 } while_each_thread(g, p);
301 if (!uid_eq(uid, cred->uid))
302 free_uid(user); /* for find_user() */
306 read_unlock(&tasklist_lock);
313 * Unprivileged users may change the real gid to the effective gid
314 * or vice versa. (BSD-style)
316 * If you set the real gid at all, or set the effective gid to a value not
317 * equal to the real gid, then the saved gid is set to the new effective gid.
319 * This makes it possible for a setgid program to completely drop its
320 * privileges, which is often a useful assertion to make when you are doing
321 * a security audit over a program.
323 * The general idea is that a program which uses just setregid() will be
324 * 100% compatible with BSD. A program which uses just setgid() will be
325 * 100% compatible with POSIX with saved IDs.
327 * SMP: There are not races, the GIDs are checked only by filesystem
328 * operations (as far as semantic preservation is concerned).
330 #ifdef CONFIG_MULTIUSER
331 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
333 struct user_namespace *ns = current_user_ns();
334 const struct cred *old;
339 krgid = make_kgid(ns, rgid);
340 kegid = make_kgid(ns, egid);
342 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
344 if ((egid != (gid_t) -1) && !gid_valid(kegid))
347 new = prepare_creds();
350 old = current_cred();
353 if (rgid != (gid_t) -1) {
354 if (gid_eq(old->gid, krgid) ||
355 gid_eq(old->egid, krgid) ||
356 ns_capable(old->user_ns, CAP_SETGID))
361 if (egid != (gid_t) -1) {
362 if (gid_eq(old->gid, kegid) ||
363 gid_eq(old->egid, kegid) ||
364 gid_eq(old->sgid, kegid) ||
365 ns_capable(old->user_ns, CAP_SETGID))
371 if (rgid != (gid_t) -1 ||
372 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
373 new->sgid = new->egid;
374 new->fsgid = new->egid;
376 return commit_creds(new);
384 * setgid() is implemented like SysV w/ SAVED_IDS
386 * SMP: Same implicit races as above.
388 SYSCALL_DEFINE1(setgid, gid_t, gid)
390 struct user_namespace *ns = current_user_ns();
391 const struct cred *old;
396 kgid = make_kgid(ns, gid);
397 if (!gid_valid(kgid))
400 new = prepare_creds();
403 old = current_cred();
406 if (ns_capable(old->user_ns, CAP_SETGID))
407 new->gid = new->egid = new->sgid = new->fsgid = kgid;
408 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
409 new->egid = new->fsgid = kgid;
413 return commit_creds(new);
421 * change the user struct in a credentials set to match the new UID
423 static int set_user(struct cred *new)
425 struct user_struct *new_user;
427 new_user = alloc_uid(new->uid);
432 * We don't fail in case of NPROC limit excess here because too many
433 * poorly written programs don't check set*uid() return code, assuming
434 * it never fails if called by root. We may still enforce NPROC limit
435 * for programs doing set*uid()+execve() by harmlessly deferring the
436 * failure to the execve() stage.
438 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
439 new_user != INIT_USER)
440 current->flags |= PF_NPROC_EXCEEDED;
442 current->flags &= ~PF_NPROC_EXCEEDED;
445 new->user = new_user;
450 * Unprivileged users may change the real uid to the effective uid
451 * or vice versa. (BSD-style)
453 * If you set the real uid at all, or set the effective uid to a value not
454 * equal to the real uid, then the saved uid is set to the new effective uid.
456 * This makes it possible for a setuid program to completely drop its
457 * privileges, which is often a useful assertion to make when you are doing
458 * a security audit over a program.
460 * The general idea is that a program which uses just setreuid() will be
461 * 100% compatible with BSD. A program which uses just setuid() will be
462 * 100% compatible with POSIX with saved IDs.
464 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
466 struct user_namespace *ns = current_user_ns();
467 const struct cred *old;
472 kruid = make_kuid(ns, ruid);
473 keuid = make_kuid(ns, euid);
475 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
477 if ((euid != (uid_t) -1) && !uid_valid(keuid))
480 new = prepare_creds();
483 old = current_cred();
486 if (ruid != (uid_t) -1) {
488 if (!uid_eq(old->uid, kruid) &&
489 !uid_eq(old->euid, kruid) &&
490 !ns_capable(old->user_ns, CAP_SETUID))
494 if (euid != (uid_t) -1) {
496 if (!uid_eq(old->uid, keuid) &&
497 !uid_eq(old->euid, keuid) &&
498 !uid_eq(old->suid, keuid) &&
499 !ns_capable(old->user_ns, CAP_SETUID))
503 if (!uid_eq(new->uid, old->uid)) {
504 retval = set_user(new);
508 if (ruid != (uid_t) -1 ||
509 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
510 new->suid = new->euid;
511 new->fsuid = new->euid;
513 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
517 return commit_creds(new);
525 * setuid() is implemented like SysV with SAVED_IDS
527 * Note that SAVED_ID's is deficient in that a setuid root program
528 * like sendmail, for example, cannot set its uid to be a normal
529 * user and then switch back, because if you're root, setuid() sets
530 * the saved uid too. If you don't like this, blame the bright people
531 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
532 * will allow a root program to temporarily drop privileges and be able to
533 * regain them by swapping the real and effective uid.
535 SYSCALL_DEFINE1(setuid, uid_t, uid)
537 struct user_namespace *ns = current_user_ns();
538 const struct cred *old;
543 kuid = make_kuid(ns, uid);
544 if (!uid_valid(kuid))
547 new = prepare_creds();
550 old = current_cred();
553 if (ns_capable(old->user_ns, CAP_SETUID)) {
554 new->suid = new->uid = kuid;
555 if (!uid_eq(kuid, old->uid)) {
556 retval = set_user(new);
560 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
564 new->fsuid = new->euid = kuid;
566 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
570 return commit_creds(new);
579 * This function implements a generic ability to update ruid, euid,
580 * and suid. This allows you to implement the 4.4 compatible seteuid().
582 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
584 struct user_namespace *ns = current_user_ns();
585 const struct cred *old;
588 kuid_t kruid, keuid, ksuid;
590 kruid = make_kuid(ns, ruid);
591 keuid = make_kuid(ns, euid);
592 ksuid = make_kuid(ns, suid);
594 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
597 if ((euid != (uid_t) -1) && !uid_valid(keuid))
600 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
603 new = prepare_creds();
607 old = current_cred();
610 if (!ns_capable(old->user_ns, CAP_SETUID)) {
611 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
612 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
614 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
615 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
617 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
618 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
622 if (ruid != (uid_t) -1) {
624 if (!uid_eq(kruid, old->uid)) {
625 retval = set_user(new);
630 if (euid != (uid_t) -1)
632 if (suid != (uid_t) -1)
634 new->fsuid = new->euid;
636 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
640 return commit_creds(new);
647 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
649 const struct cred *cred = current_cred();
651 uid_t ruid, euid, suid;
653 ruid = from_kuid_munged(cred->user_ns, cred->uid);
654 euid = from_kuid_munged(cred->user_ns, cred->euid);
655 suid = from_kuid_munged(cred->user_ns, cred->suid);
657 retval = put_user(ruid, ruidp);
659 retval = put_user(euid, euidp);
661 return put_user(suid, suidp);
667 * Same as above, but for rgid, egid, sgid.
669 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
671 struct user_namespace *ns = current_user_ns();
672 const struct cred *old;
675 kgid_t krgid, kegid, ksgid;
677 krgid = make_kgid(ns, rgid);
678 kegid = make_kgid(ns, egid);
679 ksgid = make_kgid(ns, sgid);
681 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
683 if ((egid != (gid_t) -1) && !gid_valid(kegid))
685 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
688 new = prepare_creds();
691 old = current_cred();
694 if (!ns_capable(old->user_ns, CAP_SETGID)) {
695 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
696 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
698 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
699 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
701 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
702 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
706 if (rgid != (gid_t) -1)
708 if (egid != (gid_t) -1)
710 if (sgid != (gid_t) -1)
712 new->fsgid = new->egid;
714 return commit_creds(new);
721 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
723 const struct cred *cred = current_cred();
725 gid_t rgid, egid, sgid;
727 rgid = from_kgid_munged(cred->user_ns, cred->gid);
728 egid = from_kgid_munged(cred->user_ns, cred->egid);
729 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
731 retval = put_user(rgid, rgidp);
733 retval = put_user(egid, egidp);
735 retval = put_user(sgid, sgidp);
743 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
744 * is used for "access()" and for the NFS daemon (letting nfsd stay at
745 * whatever uid it wants to). It normally shadows "euid", except when
746 * explicitly set by setfsuid() or for access..
748 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
750 const struct cred *old;
755 old = current_cred();
756 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
758 kuid = make_kuid(old->user_ns, uid);
759 if (!uid_valid(kuid))
762 new = prepare_creds();
766 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
767 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
768 ns_capable(old->user_ns, CAP_SETUID)) {
769 if (!uid_eq(kuid, old->fsuid)) {
771 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
785 * Samma på svenska..
787 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
789 const struct cred *old;
794 old = current_cred();
795 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
797 kgid = make_kgid(old->user_ns, gid);
798 if (!gid_valid(kgid))
801 new = prepare_creds();
805 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
806 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
807 ns_capable(old->user_ns, CAP_SETGID)) {
808 if (!gid_eq(kgid, old->fsgid)) {
821 #endif /* CONFIG_MULTIUSER */
824 * sys_getpid - return the thread group id of the current process
826 * Note, despite the name, this returns the tgid not the pid. The tgid and
827 * the pid are identical unless CLONE_THREAD was specified on clone() in
828 * which case the tgid is the same in all threads of the same group.
830 * This is SMP safe as current->tgid does not change.
832 SYSCALL_DEFINE0(getpid)
834 return task_tgid_vnr(current);
837 /* Thread ID - the internal kernel "pid" */
838 SYSCALL_DEFINE0(gettid)
840 return task_pid_vnr(current);
844 * Accessing ->real_parent is not SMP-safe, it could
845 * change from under us. However, we can use a stale
846 * value of ->real_parent under rcu_read_lock(), see
847 * release_task()->call_rcu(delayed_put_task_struct).
849 SYSCALL_DEFINE0(getppid)
854 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
860 SYSCALL_DEFINE0(getuid)
862 /* Only we change this so SMP safe */
863 return from_kuid_munged(current_user_ns(), current_uid());
866 SYSCALL_DEFINE0(geteuid)
868 /* Only we change this so SMP safe */
869 return from_kuid_munged(current_user_ns(), current_euid());
872 SYSCALL_DEFINE0(getgid)
874 /* Only we change this so SMP safe */
875 return from_kgid_munged(current_user_ns(), current_gid());
878 SYSCALL_DEFINE0(getegid)
880 /* Only we change this so SMP safe */
881 return from_kgid_munged(current_user_ns(), current_egid());
884 void do_sys_times(struct tms *tms)
886 cputime_t tgutime, tgstime, cutime, cstime;
888 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
889 cutime = current->signal->cutime;
890 cstime = current->signal->cstime;
891 tms->tms_utime = cputime_to_clock_t(tgutime);
892 tms->tms_stime = cputime_to_clock_t(tgstime);
893 tms->tms_cutime = cputime_to_clock_t(cutime);
894 tms->tms_cstime = cputime_to_clock_t(cstime);
897 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
903 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
906 force_successful_syscall_return();
907 return (long) jiffies_64_to_clock_t(get_jiffies_64());
911 * This needs some heavy checking ...
912 * I just haven't the stomach for it. I also don't fully
913 * understand sessions/pgrp etc. Let somebody who does explain it.
915 * OK, I think I have the protection semantics right.... this is really
916 * only important on a multi-user system anyway, to make sure one user
917 * can't send a signal to a process owned by another. -TYT, 12/12/91
919 * !PF_FORKNOEXEC check to conform completely to POSIX.
921 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
923 struct task_struct *p;
924 struct task_struct *group_leader = current->group_leader;
929 pid = task_pid_vnr(group_leader);
936 /* From this point forward we keep holding onto the tasklist lock
937 * so that our parent does not change from under us. -DaveM
939 write_lock_irq(&tasklist_lock);
942 p = find_task_by_vpid(pid);
947 if (!thread_group_leader(p))
950 if (same_thread_group(p->real_parent, group_leader)) {
952 if (task_session(p) != task_session(group_leader))
955 if (!(p->flags & PF_FORKNOEXEC))
959 if (p != group_leader)
964 if (p->signal->leader)
969 struct task_struct *g;
971 pgrp = find_vpid(pgid);
972 g = pid_task(pgrp, PIDTYPE_PGID);
973 if (!g || task_session(g) != task_session(group_leader))
977 err = security_task_setpgid(p, pgid);
981 if (task_pgrp(p) != pgrp)
982 change_pid(p, PIDTYPE_PGID, pgrp);
986 /* All paths lead to here, thus we are safe. -DaveM */
987 write_unlock_irq(&tasklist_lock);
992 SYSCALL_DEFINE1(getpgid, pid_t, pid)
994 struct task_struct *p;
1000 grp = task_pgrp(current);
1003 p = find_task_by_vpid(pid);
1010 retval = security_task_getpgid(p);
1014 retval = pid_vnr(grp);
1020 #ifdef __ARCH_WANT_SYS_GETPGRP
1022 SYSCALL_DEFINE0(getpgrp)
1024 return sys_getpgid(0);
1029 SYSCALL_DEFINE1(getsid, pid_t, pid)
1031 struct task_struct *p;
1037 sid = task_session(current);
1040 p = find_task_by_vpid(pid);
1043 sid = task_session(p);
1047 retval = security_task_getsid(p);
1051 retval = pid_vnr(sid);
1057 static void set_special_pids(struct pid *pid)
1059 struct task_struct *curr = current->group_leader;
1061 if (task_session(curr) != pid)
1062 change_pid(curr, PIDTYPE_SID, pid);
1064 if (task_pgrp(curr) != pid)
1065 change_pid(curr, PIDTYPE_PGID, pid);
1068 SYSCALL_DEFINE0(setsid)
1070 struct task_struct *group_leader = current->group_leader;
1071 struct pid *sid = task_pid(group_leader);
1072 pid_t session = pid_vnr(sid);
1075 write_lock_irq(&tasklist_lock);
1076 /* Fail if I am already a session leader */
1077 if (group_leader->signal->leader)
1080 /* Fail if a process group id already exists that equals the
1081 * proposed session id.
1083 if (pid_task(sid, PIDTYPE_PGID))
1086 group_leader->signal->leader = 1;
1087 set_special_pids(sid);
1089 proc_clear_tty(group_leader);
1093 write_unlock_irq(&tasklist_lock);
1095 proc_sid_connector(group_leader);
1096 sched_autogroup_create_attach(group_leader);
1101 DECLARE_RWSEM(uts_sem);
1103 #ifdef COMPAT_UTS_MACHINE
1104 #define override_architecture(name) \
1105 (personality(current->personality) == PER_LINUX32 && \
1106 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1107 sizeof(COMPAT_UTS_MACHINE)))
1109 #define override_architecture(name) 0
1113 * Work around broken programs that cannot handle "Linux 3.0".
1114 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1115 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1117 static int override_release(char __user *release, size_t len)
1121 if (current->personality & UNAME26) {
1122 const char *rest = UTS_RELEASE;
1123 char buf[65] = { 0 };
1129 if (*rest == '.' && ++ndots >= 3)
1131 if (!isdigit(*rest) && *rest != '.')
1135 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1136 copy = clamp_t(size_t, len, 1, sizeof(buf));
1137 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1138 ret = copy_to_user(release, buf, copy + 1);
1143 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1147 down_read(&uts_sem);
1148 if (copy_to_user(name, utsname(), sizeof *name))
1152 if (!errno && override_release(name->release, sizeof(name->release)))
1154 if (!errno && override_architecture(name))
1159 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1163 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1170 down_read(&uts_sem);
1171 if (copy_to_user(name, utsname(), sizeof(*name)))
1175 if (!error && override_release(name->release, sizeof(name->release)))
1177 if (!error && override_architecture(name))
1182 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1188 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1191 down_read(&uts_sem);
1192 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1194 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1195 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1197 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1198 error |= __copy_to_user(&name->release, &utsname()->release,
1200 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1201 error |= __copy_to_user(&name->version, &utsname()->version,
1203 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1204 error |= __copy_to_user(&name->machine, &utsname()->machine,
1206 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1209 if (!error && override_architecture(name))
1211 if (!error && override_release(name->release, sizeof(name->release)))
1213 return error ? -EFAULT : 0;
1217 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1220 char tmp[__NEW_UTS_LEN];
1222 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1225 if (len < 0 || len > __NEW_UTS_LEN)
1227 down_write(&uts_sem);
1229 if (!copy_from_user(tmp, name, len)) {
1230 struct new_utsname *u = utsname();
1232 memcpy(u->nodename, tmp, len);
1233 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1235 uts_proc_notify(UTS_PROC_HOSTNAME);
1241 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1243 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1246 struct new_utsname *u;
1250 down_read(&uts_sem);
1252 i = 1 + strlen(u->nodename);
1256 if (copy_to_user(name, u->nodename, i))
1265 * Only setdomainname; getdomainname can be implemented by calling
1268 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1271 char tmp[__NEW_UTS_LEN];
1273 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1275 if (len < 0 || len > __NEW_UTS_LEN)
1278 down_write(&uts_sem);
1280 if (!copy_from_user(tmp, name, len)) {
1281 struct new_utsname *u = utsname();
1283 memcpy(u->domainname, tmp, len);
1284 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1286 uts_proc_notify(UTS_PROC_DOMAINNAME);
1292 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1294 struct rlimit value;
1297 ret = do_prlimit(current, resource, NULL, &value);
1299 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1304 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1307 * Back compatibility for getrlimit. Needed for some apps.
1309 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1310 struct rlimit __user *, rlim)
1313 if (resource >= RLIM_NLIMITS)
1316 task_lock(current->group_leader);
1317 x = current->signal->rlim[resource];
1318 task_unlock(current->group_leader);
1319 if (x.rlim_cur > 0x7FFFFFFF)
1320 x.rlim_cur = 0x7FFFFFFF;
1321 if (x.rlim_max > 0x7FFFFFFF)
1322 x.rlim_max = 0x7FFFFFFF;
1323 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1328 static inline bool rlim64_is_infinity(__u64 rlim64)
1330 #if BITS_PER_LONG < 64
1331 return rlim64 >= ULONG_MAX;
1333 return rlim64 == RLIM64_INFINITY;
1337 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1339 if (rlim->rlim_cur == RLIM_INFINITY)
1340 rlim64->rlim_cur = RLIM64_INFINITY;
1342 rlim64->rlim_cur = rlim->rlim_cur;
1343 if (rlim->rlim_max == RLIM_INFINITY)
1344 rlim64->rlim_max = RLIM64_INFINITY;
1346 rlim64->rlim_max = rlim->rlim_max;
1349 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1351 if (rlim64_is_infinity(rlim64->rlim_cur))
1352 rlim->rlim_cur = RLIM_INFINITY;
1354 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1355 if (rlim64_is_infinity(rlim64->rlim_max))
1356 rlim->rlim_max = RLIM_INFINITY;
1358 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1361 /* make sure you are allowed to change @tsk limits before calling this */
1362 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1363 struct rlimit *new_rlim, struct rlimit *old_rlim)
1365 struct rlimit *rlim;
1368 if (resource >= RLIM_NLIMITS)
1371 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1373 if (resource == RLIMIT_NOFILE &&
1374 new_rlim->rlim_max > sysctl_nr_open)
1378 /* protect tsk->signal and tsk->sighand from disappearing */
1379 read_lock(&tasklist_lock);
1380 if (!tsk->sighand) {
1385 rlim = tsk->signal->rlim + resource;
1386 task_lock(tsk->group_leader);
1388 /* Keep the capable check against init_user_ns until
1389 cgroups can contain all limits */
1390 if (new_rlim->rlim_max > rlim->rlim_max &&
1391 !capable(CAP_SYS_RESOURCE))
1394 retval = security_task_setrlimit(tsk->group_leader,
1395 resource, new_rlim);
1396 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1398 * The caller is asking for an immediate RLIMIT_CPU
1399 * expiry. But we use the zero value to mean "it was
1400 * never set". So let's cheat and make it one second
1403 new_rlim->rlim_cur = 1;
1412 task_unlock(tsk->group_leader);
1415 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1416 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1417 * very long-standing error, and fixing it now risks breakage of
1418 * applications, so we live with it
1420 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1421 new_rlim->rlim_cur != RLIM_INFINITY)
1422 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1424 read_unlock(&tasklist_lock);
1428 /* rcu lock must be held */
1429 static int check_prlimit_permission(struct task_struct *task)
1431 const struct cred *cred = current_cred(), *tcred;
1433 if (current == task)
1436 tcred = __task_cred(task);
1437 if (uid_eq(cred->uid, tcred->euid) &&
1438 uid_eq(cred->uid, tcred->suid) &&
1439 uid_eq(cred->uid, tcred->uid) &&
1440 gid_eq(cred->gid, tcred->egid) &&
1441 gid_eq(cred->gid, tcred->sgid) &&
1442 gid_eq(cred->gid, tcred->gid))
1444 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1450 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1451 const struct rlimit64 __user *, new_rlim,
1452 struct rlimit64 __user *, old_rlim)
1454 struct rlimit64 old64, new64;
1455 struct rlimit old, new;
1456 struct task_struct *tsk;
1460 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1462 rlim64_to_rlim(&new64, &new);
1466 tsk = pid ? find_task_by_vpid(pid) : current;
1471 ret = check_prlimit_permission(tsk);
1476 get_task_struct(tsk);
1479 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1480 old_rlim ? &old : NULL);
1482 if (!ret && old_rlim) {
1483 rlim_to_rlim64(&old, &old64);
1484 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1488 put_task_struct(tsk);
1492 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1494 struct rlimit new_rlim;
1496 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1498 return do_prlimit(current, resource, &new_rlim, NULL);
1502 * It would make sense to put struct rusage in the task_struct,
1503 * except that would make the task_struct be *really big*. After
1504 * task_struct gets moved into malloc'ed memory, it would
1505 * make sense to do this. It will make moving the rest of the information
1506 * a lot simpler! (Which we're not doing right now because we're not
1507 * measuring them yet).
1509 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1510 * races with threads incrementing their own counters. But since word
1511 * reads are atomic, we either get new values or old values and we don't
1512 * care which for the sums. We always take the siglock to protect reading
1513 * the c* fields from p->signal from races with exit.c updating those
1514 * fields when reaping, so a sample either gets all the additions of a
1515 * given child after it's reaped, or none so this sample is before reaping.
1518 * We need to take the siglock for CHILDEREN, SELF and BOTH
1519 * for the cases current multithreaded, non-current single threaded
1520 * non-current multithreaded. Thread traversal is now safe with
1522 * Strictly speaking, we donot need to take the siglock if we are current and
1523 * single threaded, as no one else can take our signal_struct away, no one
1524 * else can reap the children to update signal->c* counters, and no one else
1525 * can race with the signal-> fields. If we do not take any lock, the
1526 * signal-> fields could be read out of order while another thread was just
1527 * exiting. So we should place a read memory barrier when we avoid the lock.
1528 * On the writer side, write memory barrier is implied in __exit_signal
1529 * as __exit_signal releases the siglock spinlock after updating the signal->
1530 * fields. But we don't do this yet to keep things simple.
1534 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1536 r->ru_nvcsw += t->nvcsw;
1537 r->ru_nivcsw += t->nivcsw;
1538 r->ru_minflt += t->min_flt;
1539 r->ru_majflt += t->maj_flt;
1540 r->ru_inblock += task_io_get_inblock(t);
1541 r->ru_oublock += task_io_get_oublock(t);
1544 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1546 struct task_struct *t;
1547 unsigned long flags;
1548 cputime_t tgutime, tgstime, utime, stime;
1549 unsigned long maxrss = 0;
1551 memset((char *)r, 0, sizeof (*r));
1554 if (who == RUSAGE_THREAD) {
1555 task_cputime_adjusted(current, &utime, &stime);
1556 accumulate_thread_rusage(p, r);
1557 maxrss = p->signal->maxrss;
1561 if (!lock_task_sighand(p, &flags))
1566 case RUSAGE_CHILDREN:
1567 utime = p->signal->cutime;
1568 stime = p->signal->cstime;
1569 r->ru_nvcsw = p->signal->cnvcsw;
1570 r->ru_nivcsw = p->signal->cnivcsw;
1571 r->ru_minflt = p->signal->cmin_flt;
1572 r->ru_majflt = p->signal->cmaj_flt;
1573 r->ru_inblock = p->signal->cinblock;
1574 r->ru_oublock = p->signal->coublock;
1575 maxrss = p->signal->cmaxrss;
1577 if (who == RUSAGE_CHILDREN)
1581 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1584 r->ru_nvcsw += p->signal->nvcsw;
1585 r->ru_nivcsw += p->signal->nivcsw;
1586 r->ru_minflt += p->signal->min_flt;
1587 r->ru_majflt += p->signal->maj_flt;
1588 r->ru_inblock += p->signal->inblock;
1589 r->ru_oublock += p->signal->oublock;
1590 if (maxrss < p->signal->maxrss)
1591 maxrss = p->signal->maxrss;
1594 accumulate_thread_rusage(t, r);
1595 } while_each_thread(p, t);
1601 unlock_task_sighand(p, &flags);
1604 cputime_to_timeval(utime, &r->ru_utime);
1605 cputime_to_timeval(stime, &r->ru_stime);
1607 if (who != RUSAGE_CHILDREN) {
1608 struct mm_struct *mm = get_task_mm(p);
1611 setmax_mm_hiwater_rss(&maxrss, mm);
1615 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1618 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1622 k_getrusage(p, who, &r);
1623 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1626 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1628 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1629 who != RUSAGE_THREAD)
1631 return getrusage(current, who, ru);
1634 #ifdef CONFIG_COMPAT
1635 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1639 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1640 who != RUSAGE_THREAD)
1643 k_getrusage(current, who, &r);
1644 return put_compat_rusage(&r, ru);
1648 SYSCALL_DEFINE1(umask, int, mask)
1650 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1654 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1657 struct file *old_exe, *exe_file;
1658 struct inode *inode;
1665 inode = file_inode(exe.file);
1668 * Because the original mm->exe_file points to executable file, make
1669 * sure that this one is executable as well, to avoid breaking an
1673 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1676 err = inode_permission(inode, MAY_EXEC);
1681 * Forbid mm->exe_file change if old file still mapped.
1683 exe_file = get_mm_exe_file(mm);
1686 struct vm_area_struct *vma;
1688 down_read(&mm->mmap_sem);
1689 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1692 if (path_equal(&vma->vm_file->f_path,
1697 up_read(&mm->mmap_sem);
1702 * The symlink can be changed only once, just to disallow arbitrary
1703 * transitions malicious software might bring in. This means one
1704 * could make a snapshot over all processes running and monitor
1705 * /proc/pid/exe changes to notice unusual activity if needed.
1708 if (test_and_set_bit(MMF_EXE_FILE_CHANGED, &mm->flags))
1712 /* set the new file, lockless */
1714 old_exe = xchg(&mm->exe_file, exe.file);
1721 up_read(&mm->mmap_sem);
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 #ifdef CONFIG_CHECKPOINT_RESTORE
1822 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1824 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1825 unsigned long user_auxv[AT_VECTOR_SIZE];
1826 struct mm_struct *mm = current->mm;
1829 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1830 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1832 if (opt == PR_SET_MM_MAP_SIZE)
1833 return put_user((unsigned int)sizeof(prctl_map),
1834 (unsigned int __user *)addr);
1836 if (data_size != sizeof(prctl_map))
1839 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1842 error = validate_prctl_map(&prctl_map);
1846 if (prctl_map.auxv_size) {
1847 memset(user_auxv, 0, sizeof(user_auxv));
1848 if (copy_from_user(user_auxv,
1849 (const void __user *)prctl_map.auxv,
1850 prctl_map.auxv_size))
1853 /* Last entry must be AT_NULL as specification requires */
1854 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1855 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1858 if (prctl_map.exe_fd != (u32)-1) {
1859 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1864 down_write(&mm->mmap_sem);
1867 * We don't validate if these members are pointing to
1868 * real present VMAs because application may have correspond
1869 * VMAs already unmapped and kernel uses these members for statistics
1870 * output in procfs mostly, except
1872 * - @start_brk/@brk which are used in do_brk but kernel lookups
1873 * for VMAs when updating these memvers so anything wrong written
1874 * here cause kernel to swear at userspace program but won't lead
1875 * to any problem in kernel itself
1878 mm->start_code = prctl_map.start_code;
1879 mm->end_code = prctl_map.end_code;
1880 mm->start_data = prctl_map.start_data;
1881 mm->end_data = prctl_map.end_data;
1882 mm->start_brk = prctl_map.start_brk;
1883 mm->brk = prctl_map.brk;
1884 mm->start_stack = prctl_map.start_stack;
1885 mm->arg_start = prctl_map.arg_start;
1886 mm->arg_end = prctl_map.arg_end;
1887 mm->env_start = prctl_map.env_start;
1888 mm->env_end = prctl_map.env_end;
1891 * Note this update of @saved_auxv is lockless thus
1892 * if someone reads this member in procfs while we're
1893 * updating -- it may get partly updated results. It's
1894 * known and acceptable trade off: we leave it as is to
1895 * not introduce additional locks here making the kernel
1898 if (prctl_map.auxv_size)
1899 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1901 up_write(&mm->mmap_sem);
1904 #endif /* CONFIG_CHECKPOINT_RESTORE */
1906 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1910 * This doesn't move the auxiliary vector itself since it's pinned to
1911 * mm_struct, but it permits filling the vector with new values. It's
1912 * up to the caller to provide sane values here, otherwise userspace
1913 * tools which use this vector might be unhappy.
1915 unsigned long user_auxv[AT_VECTOR_SIZE];
1917 if (len > sizeof(user_auxv))
1920 if (copy_from_user(user_auxv, (const void __user *)addr, len))
1923 /* Make sure the last entry is always AT_NULL */
1924 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1925 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1927 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1930 memcpy(mm->saved_auxv, user_auxv, len);
1931 task_unlock(current);
1936 static int prctl_set_mm(int opt, unsigned long addr,
1937 unsigned long arg4, unsigned long arg5)
1939 struct mm_struct *mm = current->mm;
1940 struct prctl_mm_map prctl_map;
1941 struct vm_area_struct *vma;
1944 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1945 opt != PR_SET_MM_MAP &&
1946 opt != PR_SET_MM_MAP_SIZE)))
1949 #ifdef CONFIG_CHECKPOINT_RESTORE
1950 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1951 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1954 if (!capable(CAP_SYS_RESOURCE))
1957 if (opt == PR_SET_MM_EXE_FILE)
1958 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1960 if (opt == PR_SET_MM_AUXV)
1961 return prctl_set_auxv(mm, addr, arg4);
1963 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1968 down_write(&mm->mmap_sem);
1969 vma = find_vma(mm, addr);
1971 prctl_map.start_code = mm->start_code;
1972 prctl_map.end_code = mm->end_code;
1973 prctl_map.start_data = mm->start_data;
1974 prctl_map.end_data = mm->end_data;
1975 prctl_map.start_brk = mm->start_brk;
1976 prctl_map.brk = mm->brk;
1977 prctl_map.start_stack = mm->start_stack;
1978 prctl_map.arg_start = mm->arg_start;
1979 prctl_map.arg_end = mm->arg_end;
1980 prctl_map.env_start = mm->env_start;
1981 prctl_map.env_end = mm->env_end;
1982 prctl_map.auxv = NULL;
1983 prctl_map.auxv_size = 0;
1984 prctl_map.exe_fd = -1;
1987 case PR_SET_MM_START_CODE:
1988 prctl_map.start_code = addr;
1990 case PR_SET_MM_END_CODE:
1991 prctl_map.end_code = addr;
1993 case PR_SET_MM_START_DATA:
1994 prctl_map.start_data = addr;
1996 case PR_SET_MM_END_DATA:
1997 prctl_map.end_data = addr;
1999 case PR_SET_MM_START_STACK:
2000 prctl_map.start_stack = addr;
2002 case PR_SET_MM_START_BRK:
2003 prctl_map.start_brk = addr;
2006 prctl_map.brk = addr;
2008 case PR_SET_MM_ARG_START:
2009 prctl_map.arg_start = addr;
2011 case PR_SET_MM_ARG_END:
2012 prctl_map.arg_end = addr;
2014 case PR_SET_MM_ENV_START:
2015 prctl_map.env_start = addr;
2017 case PR_SET_MM_ENV_END:
2018 prctl_map.env_end = addr;
2024 error = validate_prctl_map(&prctl_map);
2030 * If command line arguments and environment
2031 * are placed somewhere else on stack, we can
2032 * set them up here, ARG_START/END to setup
2033 * command line argumets and ENV_START/END
2036 case PR_SET_MM_START_STACK:
2037 case PR_SET_MM_ARG_START:
2038 case PR_SET_MM_ARG_END:
2039 case PR_SET_MM_ENV_START:
2040 case PR_SET_MM_ENV_END:
2047 mm->start_code = prctl_map.start_code;
2048 mm->end_code = prctl_map.end_code;
2049 mm->start_data = prctl_map.start_data;
2050 mm->end_data = prctl_map.end_data;
2051 mm->start_brk = prctl_map.start_brk;
2052 mm->brk = prctl_map.brk;
2053 mm->start_stack = prctl_map.start_stack;
2054 mm->arg_start = prctl_map.arg_start;
2055 mm->arg_end = prctl_map.arg_end;
2056 mm->env_start = prctl_map.env_start;
2057 mm->env_end = prctl_map.env_end;
2061 up_write(&mm->mmap_sem);
2065 #ifdef CONFIG_CHECKPOINT_RESTORE
2066 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2068 return put_user(me->clear_child_tid, tid_addr);
2071 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2078 static int prctl_update_vma_anon_name(struct vm_area_struct *vma,
2079 struct vm_area_struct **prev,
2080 unsigned long start, unsigned long end,
2081 const char __user *name_addr)
2083 struct mm_struct *mm = vma->vm_mm;
2087 if (name_addr == vma_get_anon_name(vma)) {
2092 pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
2093 *prev = vma_merge(mm, *prev, start, end, vma->vm_flags, vma->anon_vma,
2094 vma->vm_file, pgoff, vma_policy(vma),
2095 vma->vm_userfaultfd_ctx, name_addr);
2103 if (start != vma->vm_start) {
2104 error = split_vma(mm, vma, start, 1);
2109 if (end != vma->vm_end) {
2110 error = split_vma(mm, vma, end, 0);
2117 vma->anon_name = name_addr;
2120 if (error == -ENOMEM)
2125 static int prctl_set_vma_anon_name(unsigned long start, unsigned long end,
2129 struct vm_area_struct *vma, *prev;
2130 int unmapped_error = 0;
2131 int error = -EINVAL;
2134 * If the interval [start,end) covers some unmapped address
2135 * ranges, just ignore them, but return -ENOMEM at the end.
2136 * - this matches the handling in madvise.
2138 vma = find_vma_prev(current->mm, start, &prev);
2139 if (vma && start > vma->vm_start)
2143 /* Still start < end. */
2148 /* Here start < (end|vma->vm_end). */
2149 if (start < vma->vm_start) {
2150 unmapped_error = -ENOMEM;
2151 start = vma->vm_start;
2156 /* Here vma->vm_start <= start < (end|vma->vm_end) */
2161 /* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */
2162 error = prctl_update_vma_anon_name(vma, &prev, start, tmp,
2163 (const char __user *)arg);
2167 if (prev && start < prev->vm_end)
2168 start = prev->vm_end;
2169 error = unmapped_error;
2173 vma = prev->vm_next;
2174 else /* madvise_remove dropped mmap_sem */
2175 vma = find_vma(current->mm, start);
2179 static int prctl_set_vma(unsigned long opt, unsigned long start,
2180 unsigned long len_in, unsigned long arg)
2182 struct mm_struct *mm = current->mm;
2187 if (start & ~PAGE_MASK)
2189 len = (len_in + ~PAGE_MASK) & PAGE_MASK;
2191 /* Check to see whether len was rounded up from small -ve to zero */
2202 down_write(&mm->mmap_sem);
2205 case PR_SET_VMA_ANON_NAME:
2206 error = prctl_set_vma_anon_name(start, end, arg);
2212 up_write(&mm->mmap_sem);
2216 #else /* CONFIG_MMU */
2217 static int prctl_set_vma(unsigned long opt, unsigned long start,
2218 unsigned long len_in, unsigned long arg)
2224 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2225 unsigned long, arg4, unsigned long, arg5)
2227 struct task_struct *me = current;
2228 struct task_struct *tsk;
2229 unsigned char comm[sizeof(me->comm)];
2232 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2233 if (error != -ENOSYS)
2238 case PR_SET_PDEATHSIG:
2239 if (!valid_signal(arg2)) {
2243 me->pdeath_signal = arg2;
2245 case PR_GET_PDEATHSIG:
2246 error = put_user(me->pdeath_signal, (int __user *)arg2);
2248 case PR_GET_DUMPABLE:
2249 error = get_dumpable(me->mm);
2251 case PR_SET_DUMPABLE:
2252 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2256 set_dumpable(me->mm, arg2);
2259 case PR_SET_UNALIGN:
2260 error = SET_UNALIGN_CTL(me, arg2);
2262 case PR_GET_UNALIGN:
2263 error = GET_UNALIGN_CTL(me, arg2);
2266 error = SET_FPEMU_CTL(me, arg2);
2269 error = GET_FPEMU_CTL(me, arg2);
2272 error = SET_FPEXC_CTL(me, arg2);
2275 error = GET_FPEXC_CTL(me, arg2);
2278 error = PR_TIMING_STATISTICAL;
2281 if (arg2 != PR_TIMING_STATISTICAL)
2285 comm[sizeof(me->comm) - 1] = 0;
2286 if (strncpy_from_user(comm, (char __user *)arg2,
2287 sizeof(me->comm) - 1) < 0)
2289 set_task_comm(me, comm);
2290 proc_comm_connector(me);
2293 get_task_comm(comm, me);
2294 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2298 error = GET_ENDIAN(me, arg2);
2301 error = SET_ENDIAN(me, arg2);
2303 case PR_GET_SECCOMP:
2304 error = prctl_get_seccomp();
2306 case PR_SET_SECCOMP:
2307 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2310 error = GET_TSC_CTL(arg2);
2313 error = SET_TSC_CTL(arg2);
2315 case PR_TASK_PERF_EVENTS_DISABLE:
2316 error = perf_event_task_disable();
2318 case PR_TASK_PERF_EVENTS_ENABLE:
2319 error = perf_event_task_enable();
2321 case PR_GET_TIMERSLACK:
2322 if (current->timer_slack_ns > ULONG_MAX)
2325 error = current->timer_slack_ns;
2327 case PR_SET_TIMERSLACK:
2329 current->timer_slack_ns =
2330 current->default_timer_slack_ns;
2332 current->timer_slack_ns = arg2;
2338 case PR_MCE_KILL_CLEAR:
2341 current->flags &= ~PF_MCE_PROCESS;
2343 case PR_MCE_KILL_SET:
2344 current->flags |= PF_MCE_PROCESS;
2345 if (arg3 == PR_MCE_KILL_EARLY)
2346 current->flags |= PF_MCE_EARLY;
2347 else if (arg3 == PR_MCE_KILL_LATE)
2348 current->flags &= ~PF_MCE_EARLY;
2349 else if (arg3 == PR_MCE_KILL_DEFAULT)
2351 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2359 case PR_MCE_KILL_GET:
2360 if (arg2 | arg3 | arg4 | arg5)
2362 if (current->flags & PF_MCE_PROCESS)
2363 error = (current->flags & PF_MCE_EARLY) ?
2364 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2366 error = PR_MCE_KILL_DEFAULT;
2369 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2371 case PR_GET_TID_ADDRESS:
2372 error = prctl_get_tid_address(me, (int __user **)arg2);
2374 case PR_SET_TIMERSLACK_PID:
2375 if (task_pid_vnr(current) != (pid_t)arg3 &&
2376 !capable(CAP_SYS_NICE))
2379 tsk = find_task_by_vpid((pid_t)arg3);
2384 get_task_struct(tsk);
2387 tsk->timer_slack_ns =
2388 tsk->default_timer_slack_ns;
2390 tsk->timer_slack_ns = arg2;
2391 put_task_struct(tsk);
2394 case PR_SET_CHILD_SUBREAPER:
2395 me->signal->is_child_subreaper = !!arg2;
2397 case PR_GET_CHILD_SUBREAPER:
2398 error = put_user(me->signal->is_child_subreaper,
2399 (int __user *)arg2);
2401 case PR_SET_NO_NEW_PRIVS:
2402 if (arg2 != 1 || arg3 || arg4 || arg5)
2405 task_set_no_new_privs(current);
2407 case PR_GET_NO_NEW_PRIVS:
2408 if (arg2 || arg3 || arg4 || arg5)
2410 return task_no_new_privs(current) ? 1 : 0;
2411 case PR_GET_THP_DISABLE:
2412 if (arg2 || arg3 || arg4 || arg5)
2414 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2416 case PR_SET_THP_DISABLE:
2417 if (arg3 || arg4 || arg5)
2419 down_write(&me->mm->mmap_sem);
2421 me->mm->def_flags |= VM_NOHUGEPAGE;
2423 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2424 up_write(&me->mm->mmap_sem);
2426 case PR_MPX_ENABLE_MANAGEMENT:
2427 if (arg2 || arg3 || arg4 || arg5)
2429 error = MPX_ENABLE_MANAGEMENT();
2431 case PR_MPX_DISABLE_MANAGEMENT:
2432 if (arg2 || arg3 || arg4 || arg5)
2434 error = MPX_DISABLE_MANAGEMENT();
2436 case PR_SET_FP_MODE:
2437 error = SET_FP_MODE(me, arg2);
2439 case PR_GET_FP_MODE:
2440 error = GET_FP_MODE(me);
2443 error = prctl_set_vma(arg2, arg3, arg4, arg5);
2452 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2453 struct getcpu_cache __user *, unused)
2456 int cpu = raw_smp_processor_id();
2459 err |= put_user(cpu, cpup);
2461 err |= put_user(cpu_to_node(cpu), nodep);
2462 return err ? -EFAULT : 0;
2466 * do_sysinfo - fill in sysinfo struct
2467 * @info: pointer to buffer to fill
2469 static int do_sysinfo(struct sysinfo *info)
2471 unsigned long mem_total, sav_total;
2472 unsigned int mem_unit, bitcount;
2475 memset(info, 0, sizeof(struct sysinfo));
2477 get_monotonic_boottime(&tp);
2478 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2480 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2482 info->procs = nr_threads;
2488 * If the sum of all the available memory (i.e. ram + swap)
2489 * is less than can be stored in a 32 bit unsigned long then
2490 * we can be binary compatible with 2.2.x kernels. If not,
2491 * well, in that case 2.2.x was broken anyways...
2493 * -Erik Andersen <andersee@debian.org>
2496 mem_total = info->totalram + info->totalswap;
2497 if (mem_total < info->totalram || mem_total < info->totalswap)
2500 mem_unit = info->mem_unit;
2501 while (mem_unit > 1) {
2504 sav_total = mem_total;
2506 if (mem_total < sav_total)
2511 * If mem_total did not overflow, multiply all memory values by
2512 * info->mem_unit and set it to 1. This leaves things compatible
2513 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2518 info->totalram <<= bitcount;
2519 info->freeram <<= bitcount;
2520 info->sharedram <<= bitcount;
2521 info->bufferram <<= bitcount;
2522 info->totalswap <<= bitcount;
2523 info->freeswap <<= bitcount;
2524 info->totalhigh <<= bitcount;
2525 info->freehigh <<= bitcount;
2531 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2537 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2543 #ifdef CONFIG_COMPAT
2544 struct compat_sysinfo {
2558 char _f[20-2*sizeof(u32)-sizeof(int)];
2561 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2567 /* Check to see if any memory value is too large for 32-bit and scale
2570 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2573 while (s.mem_unit < PAGE_SIZE) {
2578 s.totalram >>= bitcount;
2579 s.freeram >>= bitcount;
2580 s.sharedram >>= bitcount;
2581 s.bufferram >>= bitcount;
2582 s.totalswap >>= bitcount;
2583 s.freeswap >>= bitcount;
2584 s.totalhigh >>= bitcount;
2585 s.freehigh >>= bitcount;
2588 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2589 __put_user(s.uptime, &info->uptime) ||
2590 __put_user(s.loads[0], &info->loads[0]) ||
2591 __put_user(s.loads[1], &info->loads[1]) ||
2592 __put_user(s.loads[2], &info->loads[2]) ||
2593 __put_user(s.totalram, &info->totalram) ||
2594 __put_user(s.freeram, &info->freeram) ||
2595 __put_user(s.sharedram, &info->sharedram) ||
2596 __put_user(s.bufferram, &info->bufferram) ||
2597 __put_user(s.totalswap, &info->totalswap) ||
2598 __put_user(s.freeswap, &info->freeswap) ||
2599 __put_user(s.procs, &info->procs) ||
2600 __put_user(s.totalhigh, &info->totalhigh) ||
2601 __put_user(s.freehigh, &info->freehigh) ||
2602 __put_user(s.mem_unit, &info->mem_unit))
2607 #endif /* CONFIG_COMPAT */