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/sched/autogroup.h>
53 #include <linux/sched/loadavg.h>
54 #include <linux/sched/stat.h>
55 #include <linux/sched/mm.h>
56 #include <linux/sched/coredump.h>
57 #include <linux/sched/task.h>
58 #include <linux/sched/cputime.h>
59 #include <linux/rcupdate.h>
60 #include <linux/uidgid.h>
61 #include <linux/cred.h>
63 #include <linux/kmsg_dump.h>
64 /* Move somewhere else to avoid recompiling? */
65 #include <generated/utsrelease.h>
67 #include <linux/uaccess.h>
69 #include <asm/unistd.h>
71 #ifndef SET_UNALIGN_CTL
72 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
74 #ifndef GET_UNALIGN_CTL
75 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
78 # define SET_FPEMU_CTL(a, b) (-EINVAL)
81 # define GET_FPEMU_CTL(a, b) (-EINVAL)
84 # define SET_FPEXC_CTL(a, b) (-EINVAL)
87 # define GET_FPEXC_CTL(a, b) (-EINVAL)
90 # define GET_ENDIAN(a, b) (-EINVAL)
93 # define SET_ENDIAN(a, b) (-EINVAL)
96 # define GET_TSC_CTL(a) (-EINVAL)
99 # define SET_TSC_CTL(a) (-EINVAL)
101 #ifndef MPX_ENABLE_MANAGEMENT
102 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
104 #ifndef MPX_DISABLE_MANAGEMENT
105 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
108 # define GET_FP_MODE(a) (-EINVAL)
111 # define SET_FP_MODE(a,b) (-EINVAL)
115 * this is where the system-wide overflow UID and GID are defined, for
116 * architectures that now have 32-bit UID/GID but didn't in the past
119 int overflowuid = DEFAULT_OVERFLOWUID;
120 int overflowgid = DEFAULT_OVERFLOWGID;
122 EXPORT_SYMBOL(overflowuid);
123 EXPORT_SYMBOL(overflowgid);
126 * the same as above, but for filesystems which can only store a 16-bit
127 * UID and GID. as such, this is needed on all architectures
130 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
131 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
133 EXPORT_SYMBOL(fs_overflowuid);
134 EXPORT_SYMBOL(fs_overflowgid);
137 * Returns true if current's euid is same as p's uid or euid,
138 * or has CAP_SYS_NICE to p's user_ns.
140 * Called with rcu_read_lock, creds are safe
142 static bool set_one_prio_perm(struct task_struct *p)
144 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
146 if (uid_eq(pcred->uid, cred->euid) ||
147 uid_eq(pcred->euid, cred->euid))
149 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
155 * set the priority of a task
156 * - the caller must hold the RCU read lock
158 static int set_one_prio(struct task_struct *p, int niceval, int error)
162 if (!set_one_prio_perm(p)) {
166 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
170 no_nice = security_task_setnice(p, niceval);
177 set_user_nice(p, niceval);
182 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
184 struct task_struct *g, *p;
185 struct user_struct *user;
186 const struct cred *cred = current_cred();
191 if (which > PRIO_USER || which < PRIO_PROCESS)
194 /* normalize: avoid signed division (rounding problems) */
196 if (niceval < MIN_NICE)
198 if (niceval > MAX_NICE)
202 read_lock(&tasklist_lock);
206 p = find_task_by_vpid(who);
210 error = set_one_prio(p, niceval, error);
214 pgrp = find_vpid(who);
216 pgrp = task_pgrp(current);
217 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
218 error = set_one_prio(p, niceval, error);
219 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
222 uid = make_kuid(cred->user_ns, who);
226 else if (!uid_eq(uid, cred->uid)) {
227 user = find_user(uid);
229 goto out_unlock; /* No processes for this user */
231 do_each_thread(g, p) {
232 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
233 error = set_one_prio(p, niceval, error);
234 } while_each_thread(g, p);
235 if (!uid_eq(uid, cred->uid))
236 free_uid(user); /* For find_user() */
240 read_unlock(&tasklist_lock);
247 * Ugh. To avoid negative return values, "getpriority()" will
248 * not return the normal nice-value, but a negated value that
249 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
250 * to stay compatible.
252 SYSCALL_DEFINE2(getpriority, int, which, int, who)
254 struct task_struct *g, *p;
255 struct user_struct *user;
256 const struct cred *cred = current_cred();
257 long niceval, retval = -ESRCH;
261 if (which > PRIO_USER || which < PRIO_PROCESS)
265 read_lock(&tasklist_lock);
269 p = find_task_by_vpid(who);
273 niceval = nice_to_rlimit(task_nice(p));
274 if (niceval > retval)
280 pgrp = find_vpid(who);
282 pgrp = task_pgrp(current);
283 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
284 niceval = nice_to_rlimit(task_nice(p));
285 if (niceval > retval)
287 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
290 uid = make_kuid(cred->user_ns, who);
294 else if (!uid_eq(uid, cred->uid)) {
295 user = find_user(uid);
297 goto out_unlock; /* No processes for this user */
299 do_each_thread(g, p) {
300 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
301 niceval = nice_to_rlimit(task_nice(p));
302 if (niceval > retval)
305 } while_each_thread(g, p);
306 if (!uid_eq(uid, cred->uid))
307 free_uid(user); /* for find_user() */
311 read_unlock(&tasklist_lock);
318 * Unprivileged users may change the real gid to the effective gid
319 * or vice versa. (BSD-style)
321 * If you set the real gid at all, or set the effective gid to a value not
322 * equal to the real gid, then the saved gid is set to the new effective gid.
324 * This makes it possible for a setgid program to completely drop its
325 * privileges, which is often a useful assertion to make when you are doing
326 * a security audit over a program.
328 * The general idea is that a program which uses just setregid() will be
329 * 100% compatible with BSD. A program which uses just setgid() will be
330 * 100% compatible with POSIX with saved IDs.
332 * SMP: There are not races, the GIDs are checked only by filesystem
333 * operations (as far as semantic preservation is concerned).
335 #ifdef CONFIG_MULTIUSER
336 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
338 struct user_namespace *ns = current_user_ns();
339 const struct cred *old;
344 krgid = make_kgid(ns, rgid);
345 kegid = make_kgid(ns, egid);
347 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
349 if ((egid != (gid_t) -1) && !gid_valid(kegid))
352 new = prepare_creds();
355 old = current_cred();
358 if (rgid != (gid_t) -1) {
359 if (gid_eq(old->gid, krgid) ||
360 gid_eq(old->egid, krgid) ||
361 ns_capable(old->user_ns, CAP_SETGID))
366 if (egid != (gid_t) -1) {
367 if (gid_eq(old->gid, kegid) ||
368 gid_eq(old->egid, kegid) ||
369 gid_eq(old->sgid, kegid) ||
370 ns_capable(old->user_ns, CAP_SETGID))
376 if (rgid != (gid_t) -1 ||
377 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
378 new->sgid = new->egid;
379 new->fsgid = new->egid;
381 return commit_creds(new);
389 * setgid() is implemented like SysV w/ SAVED_IDS
391 * SMP: Same implicit races as above.
393 SYSCALL_DEFINE1(setgid, gid_t, gid)
395 struct user_namespace *ns = current_user_ns();
396 const struct cred *old;
401 kgid = make_kgid(ns, gid);
402 if (!gid_valid(kgid))
405 new = prepare_creds();
408 old = current_cred();
411 if (ns_capable(old->user_ns, CAP_SETGID))
412 new->gid = new->egid = new->sgid = new->fsgid = kgid;
413 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
414 new->egid = new->fsgid = kgid;
418 return commit_creds(new);
426 * change the user struct in a credentials set to match the new UID
428 static int set_user(struct cred *new)
430 struct user_struct *new_user;
432 new_user = alloc_uid(new->uid);
437 * We don't fail in case of NPROC limit excess here because too many
438 * poorly written programs don't check set*uid() return code, assuming
439 * it never fails if called by root. We may still enforce NPROC limit
440 * for programs doing set*uid()+execve() by harmlessly deferring the
441 * failure to the execve() stage.
443 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
444 new_user != INIT_USER)
445 current->flags |= PF_NPROC_EXCEEDED;
447 current->flags &= ~PF_NPROC_EXCEEDED;
450 new->user = new_user;
455 * Unprivileged users may change the real uid to the effective uid
456 * or vice versa. (BSD-style)
458 * If you set the real uid at all, or set the effective uid to a value not
459 * equal to the real uid, then the saved uid is set to the new effective uid.
461 * This makes it possible for a setuid program to completely drop its
462 * privileges, which is often a useful assertion to make when you are doing
463 * a security audit over a program.
465 * The general idea is that a program which uses just setreuid() will be
466 * 100% compatible with BSD. A program which uses just setuid() will be
467 * 100% compatible with POSIX with saved IDs.
469 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
471 struct user_namespace *ns = current_user_ns();
472 const struct cred *old;
477 kruid = make_kuid(ns, ruid);
478 keuid = make_kuid(ns, euid);
480 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
482 if ((euid != (uid_t) -1) && !uid_valid(keuid))
485 new = prepare_creds();
488 old = current_cred();
491 if (ruid != (uid_t) -1) {
493 if (!uid_eq(old->uid, kruid) &&
494 !uid_eq(old->euid, kruid) &&
495 !ns_capable(old->user_ns, CAP_SETUID))
499 if (euid != (uid_t) -1) {
501 if (!uid_eq(old->uid, keuid) &&
502 !uid_eq(old->euid, keuid) &&
503 !uid_eq(old->suid, keuid) &&
504 !ns_capable(old->user_ns, CAP_SETUID))
508 if (!uid_eq(new->uid, old->uid)) {
509 retval = set_user(new);
513 if (ruid != (uid_t) -1 ||
514 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
515 new->suid = new->euid;
516 new->fsuid = new->euid;
518 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
522 return commit_creds(new);
530 * setuid() is implemented like SysV with SAVED_IDS
532 * Note that SAVED_ID's is deficient in that a setuid root program
533 * like sendmail, for example, cannot set its uid to be a normal
534 * user and then switch back, because if you're root, setuid() sets
535 * the saved uid too. If you don't like this, blame the bright people
536 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
537 * will allow a root program to temporarily drop privileges and be able to
538 * regain them by swapping the real and effective uid.
540 SYSCALL_DEFINE1(setuid, uid_t, uid)
542 struct user_namespace *ns = current_user_ns();
543 const struct cred *old;
548 kuid = make_kuid(ns, uid);
549 if (!uid_valid(kuid))
552 new = prepare_creds();
555 old = current_cred();
558 if (ns_capable(old->user_ns, CAP_SETUID)) {
559 new->suid = new->uid = kuid;
560 if (!uid_eq(kuid, old->uid)) {
561 retval = set_user(new);
565 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
569 new->fsuid = new->euid = kuid;
571 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
575 return commit_creds(new);
584 * This function implements a generic ability to update ruid, euid,
585 * and suid. This allows you to implement the 4.4 compatible seteuid().
587 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
589 struct user_namespace *ns = current_user_ns();
590 const struct cred *old;
593 kuid_t kruid, keuid, ksuid;
595 kruid = make_kuid(ns, ruid);
596 keuid = make_kuid(ns, euid);
597 ksuid = make_kuid(ns, suid);
599 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
602 if ((euid != (uid_t) -1) && !uid_valid(keuid))
605 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
608 new = prepare_creds();
612 old = current_cred();
615 if (!ns_capable(old->user_ns, CAP_SETUID)) {
616 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
617 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
619 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
620 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
622 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
623 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
627 if (ruid != (uid_t) -1) {
629 if (!uid_eq(kruid, old->uid)) {
630 retval = set_user(new);
635 if (euid != (uid_t) -1)
637 if (suid != (uid_t) -1)
639 new->fsuid = new->euid;
641 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
645 return commit_creds(new);
652 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
654 const struct cred *cred = current_cred();
656 uid_t ruid, euid, suid;
658 ruid = from_kuid_munged(cred->user_ns, cred->uid);
659 euid = from_kuid_munged(cred->user_ns, cred->euid);
660 suid = from_kuid_munged(cred->user_ns, cred->suid);
662 retval = put_user(ruid, ruidp);
664 retval = put_user(euid, euidp);
666 return put_user(suid, suidp);
672 * Same as above, but for rgid, egid, sgid.
674 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
676 struct user_namespace *ns = current_user_ns();
677 const struct cred *old;
680 kgid_t krgid, kegid, ksgid;
682 krgid = make_kgid(ns, rgid);
683 kegid = make_kgid(ns, egid);
684 ksgid = make_kgid(ns, sgid);
686 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
688 if ((egid != (gid_t) -1) && !gid_valid(kegid))
690 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
693 new = prepare_creds();
696 old = current_cred();
699 if (!ns_capable(old->user_ns, CAP_SETGID)) {
700 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
701 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
703 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
704 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
706 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
707 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
711 if (rgid != (gid_t) -1)
713 if (egid != (gid_t) -1)
715 if (sgid != (gid_t) -1)
717 new->fsgid = new->egid;
719 return commit_creds(new);
726 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
728 const struct cred *cred = current_cred();
730 gid_t rgid, egid, sgid;
732 rgid = from_kgid_munged(cred->user_ns, cred->gid);
733 egid = from_kgid_munged(cred->user_ns, cred->egid);
734 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
736 retval = put_user(rgid, rgidp);
738 retval = put_user(egid, egidp);
740 retval = put_user(sgid, sgidp);
748 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
749 * is used for "access()" and for the NFS daemon (letting nfsd stay at
750 * whatever uid it wants to). It normally shadows "euid", except when
751 * explicitly set by setfsuid() or for access..
753 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
755 const struct cred *old;
760 old = current_cred();
761 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
763 kuid = make_kuid(old->user_ns, uid);
764 if (!uid_valid(kuid))
767 new = prepare_creds();
771 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
772 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
773 ns_capable(old->user_ns, CAP_SETUID)) {
774 if (!uid_eq(kuid, old->fsuid)) {
776 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
790 * Samma på svenska..
792 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
794 const struct cred *old;
799 old = current_cred();
800 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
802 kgid = make_kgid(old->user_ns, gid);
803 if (!gid_valid(kgid))
806 new = prepare_creds();
810 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
811 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
812 ns_capable(old->user_ns, CAP_SETGID)) {
813 if (!gid_eq(kgid, old->fsgid)) {
826 #endif /* CONFIG_MULTIUSER */
829 * sys_getpid - return the thread group id of the current process
831 * Note, despite the name, this returns the tgid not the pid. The tgid and
832 * the pid are identical unless CLONE_THREAD was specified on clone() in
833 * which case the tgid is the same in all threads of the same group.
835 * This is SMP safe as current->tgid does not change.
837 SYSCALL_DEFINE0(getpid)
839 return task_tgid_vnr(current);
842 /* Thread ID - the internal kernel "pid" */
843 SYSCALL_DEFINE0(gettid)
845 return task_pid_vnr(current);
849 * Accessing ->real_parent is not SMP-safe, it could
850 * change from under us. However, we can use a stale
851 * value of ->real_parent under rcu_read_lock(), see
852 * release_task()->call_rcu(delayed_put_task_struct).
854 SYSCALL_DEFINE0(getppid)
859 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
865 SYSCALL_DEFINE0(getuid)
867 /* Only we change this so SMP safe */
868 return from_kuid_munged(current_user_ns(), current_uid());
871 SYSCALL_DEFINE0(geteuid)
873 /* Only we change this so SMP safe */
874 return from_kuid_munged(current_user_ns(), current_euid());
877 SYSCALL_DEFINE0(getgid)
879 /* Only we change this so SMP safe */
880 return from_kgid_munged(current_user_ns(), current_gid());
883 SYSCALL_DEFINE0(getegid)
885 /* Only we change this so SMP safe */
886 return from_kgid_munged(current_user_ns(), current_egid());
889 void do_sys_times(struct tms *tms)
891 u64 tgutime, tgstime, cutime, cstime;
893 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
894 cutime = current->signal->cutime;
895 cstime = current->signal->cstime;
896 tms->tms_utime = nsec_to_clock_t(tgutime);
897 tms->tms_stime = nsec_to_clock_t(tgstime);
898 tms->tms_cutime = nsec_to_clock_t(cutime);
899 tms->tms_cstime = nsec_to_clock_t(cstime);
902 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
908 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
911 force_successful_syscall_return();
912 return (long) jiffies_64_to_clock_t(get_jiffies_64());
916 * This needs some heavy checking ...
917 * I just haven't the stomach for it. I also don't fully
918 * understand sessions/pgrp etc. Let somebody who does explain it.
920 * OK, I think I have the protection semantics right.... this is really
921 * only important on a multi-user system anyway, to make sure one user
922 * can't send a signal to a process owned by another. -TYT, 12/12/91
924 * !PF_FORKNOEXEC check to conform completely to POSIX.
926 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
928 struct task_struct *p;
929 struct task_struct *group_leader = current->group_leader;
934 pid = task_pid_vnr(group_leader);
941 /* From this point forward we keep holding onto the tasklist lock
942 * so that our parent does not change from under us. -DaveM
944 write_lock_irq(&tasklist_lock);
947 p = find_task_by_vpid(pid);
952 if (!thread_group_leader(p))
955 if (same_thread_group(p->real_parent, group_leader)) {
957 if (task_session(p) != task_session(group_leader))
960 if (!(p->flags & PF_FORKNOEXEC))
964 if (p != group_leader)
969 if (p->signal->leader)
974 struct task_struct *g;
976 pgrp = find_vpid(pgid);
977 g = pid_task(pgrp, PIDTYPE_PGID);
978 if (!g || task_session(g) != task_session(group_leader))
982 err = security_task_setpgid(p, pgid);
986 if (task_pgrp(p) != pgrp)
987 change_pid(p, PIDTYPE_PGID, pgrp);
991 /* All paths lead to here, thus we are safe. -DaveM */
992 write_unlock_irq(&tasklist_lock);
997 SYSCALL_DEFINE1(getpgid, pid_t, pid)
999 struct task_struct *p;
1005 grp = task_pgrp(current);
1008 p = find_task_by_vpid(pid);
1015 retval = security_task_getpgid(p);
1019 retval = pid_vnr(grp);
1025 #ifdef __ARCH_WANT_SYS_GETPGRP
1027 SYSCALL_DEFINE0(getpgrp)
1029 return sys_getpgid(0);
1034 SYSCALL_DEFINE1(getsid, pid_t, pid)
1036 struct task_struct *p;
1042 sid = task_session(current);
1045 p = find_task_by_vpid(pid);
1048 sid = task_session(p);
1052 retval = security_task_getsid(p);
1056 retval = pid_vnr(sid);
1062 static void set_special_pids(struct pid *pid)
1064 struct task_struct *curr = current->group_leader;
1066 if (task_session(curr) != pid)
1067 change_pid(curr, PIDTYPE_SID, pid);
1069 if (task_pgrp(curr) != pid)
1070 change_pid(curr, PIDTYPE_PGID, pid);
1073 SYSCALL_DEFINE0(setsid)
1075 struct task_struct *group_leader = current->group_leader;
1076 struct pid *sid = task_pid(group_leader);
1077 pid_t session = pid_vnr(sid);
1080 write_lock_irq(&tasklist_lock);
1081 /* Fail if I am already a session leader */
1082 if (group_leader->signal->leader)
1085 /* Fail if a process group id already exists that equals the
1086 * proposed session id.
1088 if (pid_task(sid, PIDTYPE_PGID))
1091 group_leader->signal->leader = 1;
1092 set_special_pids(sid);
1094 proc_clear_tty(group_leader);
1098 write_unlock_irq(&tasklist_lock);
1100 proc_sid_connector(group_leader);
1101 sched_autogroup_create_attach(group_leader);
1106 DECLARE_RWSEM(uts_sem);
1108 #ifdef COMPAT_UTS_MACHINE
1109 #define override_architecture(name) \
1110 (personality(current->personality) == PER_LINUX32 && \
1111 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1112 sizeof(COMPAT_UTS_MACHINE)))
1114 #define override_architecture(name) 0
1118 * Work around broken programs that cannot handle "Linux 3.0".
1119 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1120 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1122 static int override_release(char __user *release, size_t len)
1126 if (current->personality & UNAME26) {
1127 const char *rest = UTS_RELEASE;
1128 char buf[65] = { 0 };
1134 if (*rest == '.' && ++ndots >= 3)
1136 if (!isdigit(*rest) && *rest != '.')
1140 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1141 copy = clamp_t(size_t, len, 1, sizeof(buf));
1142 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1143 ret = copy_to_user(release, buf, copy + 1);
1148 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1152 down_read(&uts_sem);
1153 if (copy_to_user(name, utsname(), sizeof *name))
1157 if (!errno && override_release(name->release, sizeof(name->release)))
1159 if (!errno && override_architecture(name))
1164 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1168 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1175 down_read(&uts_sem);
1176 if (copy_to_user(name, utsname(), sizeof(*name)))
1180 if (!error && override_release(name->release, sizeof(name->release)))
1182 if (!error && override_architecture(name))
1187 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1193 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1196 down_read(&uts_sem);
1197 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1199 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1200 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1202 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1203 error |= __copy_to_user(&name->release, &utsname()->release,
1205 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1206 error |= __copy_to_user(&name->version, &utsname()->version,
1208 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1209 error |= __copy_to_user(&name->machine, &utsname()->machine,
1211 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1214 if (!error && override_architecture(name))
1216 if (!error && override_release(name->release, sizeof(name->release)))
1218 return error ? -EFAULT : 0;
1222 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1225 char tmp[__NEW_UTS_LEN];
1227 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1230 if (len < 0 || len > __NEW_UTS_LEN)
1232 down_write(&uts_sem);
1234 if (!copy_from_user(tmp, name, len)) {
1235 struct new_utsname *u = utsname();
1237 memcpy(u->nodename, tmp, len);
1238 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1240 uts_proc_notify(UTS_PROC_HOSTNAME);
1246 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1248 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1251 struct new_utsname *u;
1255 down_read(&uts_sem);
1257 i = 1 + strlen(u->nodename);
1261 if (copy_to_user(name, u->nodename, i))
1270 * Only setdomainname; getdomainname can be implemented by calling
1273 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1276 char tmp[__NEW_UTS_LEN];
1278 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1280 if (len < 0 || len > __NEW_UTS_LEN)
1283 down_write(&uts_sem);
1285 if (!copy_from_user(tmp, name, len)) {
1286 struct new_utsname *u = utsname();
1288 memcpy(u->domainname, tmp, len);
1289 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1291 uts_proc_notify(UTS_PROC_DOMAINNAME);
1297 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1299 struct rlimit value;
1302 ret = do_prlimit(current, resource, NULL, &value);
1304 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1309 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1312 * Back compatibility for getrlimit. Needed for some apps.
1314 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1315 struct rlimit __user *, rlim)
1318 if (resource >= RLIM_NLIMITS)
1321 task_lock(current->group_leader);
1322 x = current->signal->rlim[resource];
1323 task_unlock(current->group_leader);
1324 if (x.rlim_cur > 0x7FFFFFFF)
1325 x.rlim_cur = 0x7FFFFFFF;
1326 if (x.rlim_max > 0x7FFFFFFF)
1327 x.rlim_max = 0x7FFFFFFF;
1328 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1333 static inline bool rlim64_is_infinity(__u64 rlim64)
1335 #if BITS_PER_LONG < 64
1336 return rlim64 >= ULONG_MAX;
1338 return rlim64 == RLIM64_INFINITY;
1342 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1344 if (rlim->rlim_cur == RLIM_INFINITY)
1345 rlim64->rlim_cur = RLIM64_INFINITY;
1347 rlim64->rlim_cur = rlim->rlim_cur;
1348 if (rlim->rlim_max == RLIM_INFINITY)
1349 rlim64->rlim_max = RLIM64_INFINITY;
1351 rlim64->rlim_max = rlim->rlim_max;
1354 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1356 if (rlim64_is_infinity(rlim64->rlim_cur))
1357 rlim->rlim_cur = RLIM_INFINITY;
1359 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1360 if (rlim64_is_infinity(rlim64->rlim_max))
1361 rlim->rlim_max = RLIM_INFINITY;
1363 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1366 /* make sure you are allowed to change @tsk limits before calling this */
1367 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1368 struct rlimit *new_rlim, struct rlimit *old_rlim)
1370 struct rlimit *rlim;
1373 if (resource >= RLIM_NLIMITS)
1376 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1378 if (resource == RLIMIT_NOFILE &&
1379 new_rlim->rlim_max > sysctl_nr_open)
1383 /* protect tsk->signal and tsk->sighand from disappearing */
1384 read_lock(&tasklist_lock);
1385 if (!tsk->sighand) {
1390 rlim = tsk->signal->rlim + resource;
1391 task_lock(tsk->group_leader);
1393 /* Keep the capable check against init_user_ns until
1394 cgroups can contain all limits */
1395 if (new_rlim->rlim_max > rlim->rlim_max &&
1396 !capable(CAP_SYS_RESOURCE))
1399 retval = security_task_setrlimit(tsk, resource, new_rlim);
1400 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1402 * The caller is asking for an immediate RLIMIT_CPU
1403 * expiry. But we use the zero value to mean "it was
1404 * never set". So let's cheat and make it one second
1407 new_rlim->rlim_cur = 1;
1416 task_unlock(tsk->group_leader);
1419 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1420 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1421 * very long-standing error, and fixing it now risks breakage of
1422 * applications, so we live with it
1424 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1425 new_rlim->rlim_cur != RLIM_INFINITY &&
1426 IS_ENABLED(CONFIG_POSIX_TIMERS))
1427 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1429 read_unlock(&tasklist_lock);
1433 /* rcu lock must be held */
1434 static int check_prlimit_permission(struct task_struct *task,
1437 const struct cred *cred = current_cred(), *tcred;
1440 if (current == task)
1443 tcred = __task_cred(task);
1444 id_match = (uid_eq(cred->uid, tcred->euid) &&
1445 uid_eq(cred->uid, tcred->suid) &&
1446 uid_eq(cred->uid, tcred->uid) &&
1447 gid_eq(cred->gid, tcred->egid) &&
1448 gid_eq(cred->gid, tcred->sgid) &&
1449 gid_eq(cred->gid, tcred->gid));
1450 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1453 return security_task_prlimit(cred, tcred, flags);
1456 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1457 const struct rlimit64 __user *, new_rlim,
1458 struct rlimit64 __user *, old_rlim)
1460 struct rlimit64 old64, new64;
1461 struct rlimit old, new;
1462 struct task_struct *tsk;
1463 unsigned int checkflags = 0;
1467 checkflags |= LSM_PRLIMIT_READ;
1470 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1472 rlim64_to_rlim(&new64, &new);
1473 checkflags |= LSM_PRLIMIT_WRITE;
1477 tsk = pid ? find_task_by_vpid(pid) : current;
1482 ret = check_prlimit_permission(tsk, checkflags);
1487 get_task_struct(tsk);
1490 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1491 old_rlim ? &old : NULL);
1493 if (!ret && old_rlim) {
1494 rlim_to_rlim64(&old, &old64);
1495 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1499 put_task_struct(tsk);
1503 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1505 struct rlimit new_rlim;
1507 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1509 return do_prlimit(current, resource, &new_rlim, NULL);
1513 * It would make sense to put struct rusage in the task_struct,
1514 * except that would make the task_struct be *really big*. After
1515 * task_struct gets moved into malloc'ed memory, it would
1516 * make sense to do this. It will make moving the rest of the information
1517 * a lot simpler! (Which we're not doing right now because we're not
1518 * measuring them yet).
1520 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1521 * races with threads incrementing their own counters. But since word
1522 * reads are atomic, we either get new values or old values and we don't
1523 * care which for the sums. We always take the siglock to protect reading
1524 * the c* fields from p->signal from races with exit.c updating those
1525 * fields when reaping, so a sample either gets all the additions of a
1526 * given child after it's reaped, or none so this sample is before reaping.
1529 * We need to take the siglock for CHILDEREN, SELF and BOTH
1530 * for the cases current multithreaded, non-current single threaded
1531 * non-current multithreaded. Thread traversal is now safe with
1533 * Strictly speaking, we donot need to take the siglock if we are current and
1534 * single threaded, as no one else can take our signal_struct away, no one
1535 * else can reap the children to update signal->c* counters, and no one else
1536 * can race with the signal-> fields. If we do not take any lock, the
1537 * signal-> fields could be read out of order while another thread was just
1538 * exiting. So we should place a read memory barrier when we avoid the lock.
1539 * On the writer side, write memory barrier is implied in __exit_signal
1540 * as __exit_signal releases the siglock spinlock after updating the signal->
1541 * fields. But we don't do this yet to keep things simple.
1545 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1547 r->ru_nvcsw += t->nvcsw;
1548 r->ru_nivcsw += t->nivcsw;
1549 r->ru_minflt += t->min_flt;
1550 r->ru_majflt += t->maj_flt;
1551 r->ru_inblock += task_io_get_inblock(t);
1552 r->ru_oublock += task_io_get_oublock(t);
1555 void getrusage(struct task_struct *p, int who, struct rusage *r)
1557 struct task_struct *t;
1558 unsigned long flags;
1559 u64 tgutime, tgstime, utime, stime;
1560 unsigned long maxrss = 0;
1562 memset((char *)r, 0, sizeof (*r));
1565 if (who == RUSAGE_THREAD) {
1566 task_cputime_adjusted(current, &utime, &stime);
1567 accumulate_thread_rusage(p, r);
1568 maxrss = p->signal->maxrss;
1572 if (!lock_task_sighand(p, &flags))
1577 case RUSAGE_CHILDREN:
1578 utime = p->signal->cutime;
1579 stime = p->signal->cstime;
1580 r->ru_nvcsw = p->signal->cnvcsw;
1581 r->ru_nivcsw = p->signal->cnivcsw;
1582 r->ru_minflt = p->signal->cmin_flt;
1583 r->ru_majflt = p->signal->cmaj_flt;
1584 r->ru_inblock = p->signal->cinblock;
1585 r->ru_oublock = p->signal->coublock;
1586 maxrss = p->signal->cmaxrss;
1588 if (who == RUSAGE_CHILDREN)
1592 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1595 r->ru_nvcsw += p->signal->nvcsw;
1596 r->ru_nivcsw += p->signal->nivcsw;
1597 r->ru_minflt += p->signal->min_flt;
1598 r->ru_majflt += p->signal->maj_flt;
1599 r->ru_inblock += p->signal->inblock;
1600 r->ru_oublock += p->signal->oublock;
1601 if (maxrss < p->signal->maxrss)
1602 maxrss = p->signal->maxrss;
1605 accumulate_thread_rusage(t, r);
1606 } while_each_thread(p, t);
1612 unlock_task_sighand(p, &flags);
1615 r->ru_utime = ns_to_timeval(utime);
1616 r->ru_stime = ns_to_timeval(stime);
1618 if (who != RUSAGE_CHILDREN) {
1619 struct mm_struct *mm = get_task_mm(p);
1622 setmax_mm_hiwater_rss(&maxrss, mm);
1626 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1629 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1633 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1634 who != RUSAGE_THREAD)
1637 getrusage(current, who, &r);
1638 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1641 #ifdef CONFIG_COMPAT
1642 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1646 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1647 who != RUSAGE_THREAD)
1650 getrusage(current, who, &r);
1651 return put_compat_rusage(&r, ru);
1655 SYSCALL_DEFINE1(umask, int, mask)
1657 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1661 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1664 struct file *old_exe, *exe_file;
1665 struct inode *inode;
1672 inode = file_inode(exe.file);
1675 * Because the original mm->exe_file points to executable file, make
1676 * sure that this one is executable as well, to avoid breaking an
1680 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1683 err = inode_permission(inode, MAY_EXEC);
1688 * Forbid mm->exe_file change if old file still mapped.
1690 exe_file = get_mm_exe_file(mm);
1693 struct vm_area_struct *vma;
1695 down_read(&mm->mmap_sem);
1696 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1699 if (path_equal(&vma->vm_file->f_path,
1704 up_read(&mm->mmap_sem);
1709 /* set the new file, lockless */
1711 old_exe = xchg(&mm->exe_file, exe.file);
1718 up_read(&mm->mmap_sem);
1724 * WARNING: we don't require any capability here so be very careful
1725 * in what is allowed for modification from userspace.
1727 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1729 unsigned long mmap_max_addr = TASK_SIZE;
1730 struct mm_struct *mm = current->mm;
1731 int error = -EINVAL, i;
1733 static const unsigned char offsets[] = {
1734 offsetof(struct prctl_mm_map, start_code),
1735 offsetof(struct prctl_mm_map, end_code),
1736 offsetof(struct prctl_mm_map, start_data),
1737 offsetof(struct prctl_mm_map, end_data),
1738 offsetof(struct prctl_mm_map, start_brk),
1739 offsetof(struct prctl_mm_map, brk),
1740 offsetof(struct prctl_mm_map, start_stack),
1741 offsetof(struct prctl_mm_map, arg_start),
1742 offsetof(struct prctl_mm_map, arg_end),
1743 offsetof(struct prctl_mm_map, env_start),
1744 offsetof(struct prctl_mm_map, env_end),
1748 * Make sure the members are not somewhere outside
1749 * of allowed address space.
1751 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1752 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1754 if ((unsigned long)val >= mmap_max_addr ||
1755 (unsigned long)val < mmap_min_addr)
1760 * Make sure the pairs are ordered.
1762 #define __prctl_check_order(__m1, __op, __m2) \
1763 ((unsigned long)prctl_map->__m1 __op \
1764 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1765 error = __prctl_check_order(start_code, <, end_code);
1766 error |= __prctl_check_order(start_data, <, end_data);
1767 error |= __prctl_check_order(start_brk, <=, brk);
1768 error |= __prctl_check_order(arg_start, <=, arg_end);
1769 error |= __prctl_check_order(env_start, <=, env_end);
1772 #undef __prctl_check_order
1777 * @brk should be after @end_data in traditional maps.
1779 if (prctl_map->start_brk <= prctl_map->end_data ||
1780 prctl_map->brk <= prctl_map->end_data)
1784 * Neither we should allow to override limits if they set.
1786 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1787 prctl_map->start_brk, prctl_map->end_data,
1788 prctl_map->start_data))
1792 * Someone is trying to cheat the auxv vector.
1794 if (prctl_map->auxv_size) {
1795 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1800 * Finally, make sure the caller has the rights to
1801 * change /proc/pid/exe link: only local root should
1804 if (prctl_map->exe_fd != (u32)-1) {
1805 struct user_namespace *ns = current_user_ns();
1806 const struct cred *cred = current_cred();
1808 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1809 !gid_eq(cred->gid, make_kgid(ns, 0)))
1818 #ifdef CONFIG_CHECKPOINT_RESTORE
1819 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1821 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1822 unsigned long user_auxv[AT_VECTOR_SIZE];
1823 struct mm_struct *mm = current->mm;
1826 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1827 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1829 if (opt == PR_SET_MM_MAP_SIZE)
1830 return put_user((unsigned int)sizeof(prctl_map),
1831 (unsigned int __user *)addr);
1833 if (data_size != sizeof(prctl_map))
1836 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1839 error = validate_prctl_map(&prctl_map);
1843 if (prctl_map.auxv_size) {
1844 memset(user_auxv, 0, sizeof(user_auxv));
1845 if (copy_from_user(user_auxv,
1846 (const void __user *)prctl_map.auxv,
1847 prctl_map.auxv_size))
1850 /* Last entry must be AT_NULL as specification requires */
1851 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1852 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1855 if (prctl_map.exe_fd != (u32)-1) {
1856 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1861 down_write(&mm->mmap_sem);
1864 * We don't validate if these members are pointing to
1865 * real present VMAs because application may have correspond
1866 * VMAs already unmapped and kernel uses these members for statistics
1867 * output in procfs mostly, except
1869 * - @start_brk/@brk which are used in do_brk but kernel lookups
1870 * for VMAs when updating these memvers so anything wrong written
1871 * here cause kernel to swear at userspace program but won't lead
1872 * to any problem in kernel itself
1875 mm->start_code = prctl_map.start_code;
1876 mm->end_code = prctl_map.end_code;
1877 mm->start_data = prctl_map.start_data;
1878 mm->end_data = prctl_map.end_data;
1879 mm->start_brk = prctl_map.start_brk;
1880 mm->brk = prctl_map.brk;
1881 mm->start_stack = prctl_map.start_stack;
1882 mm->arg_start = prctl_map.arg_start;
1883 mm->arg_end = prctl_map.arg_end;
1884 mm->env_start = prctl_map.env_start;
1885 mm->env_end = prctl_map.env_end;
1888 * Note this update of @saved_auxv is lockless thus
1889 * if someone reads this member in procfs while we're
1890 * updating -- it may get partly updated results. It's
1891 * known and acceptable trade off: we leave it as is to
1892 * not introduce additional locks here making the kernel
1895 if (prctl_map.auxv_size)
1896 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1898 up_write(&mm->mmap_sem);
1901 #endif /* CONFIG_CHECKPOINT_RESTORE */
1903 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1907 * This doesn't move the auxiliary vector itself since it's pinned to
1908 * mm_struct, but it permits filling the vector with new values. It's
1909 * up to the caller to provide sane values here, otherwise userspace
1910 * tools which use this vector might be unhappy.
1912 unsigned long user_auxv[AT_VECTOR_SIZE];
1914 if (len > sizeof(user_auxv))
1917 if (copy_from_user(user_auxv, (const void __user *)addr, len))
1920 /* Make sure the last entry is always AT_NULL */
1921 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1922 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1924 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1927 memcpy(mm->saved_auxv, user_auxv, len);
1928 task_unlock(current);
1933 static int prctl_set_mm(int opt, unsigned long addr,
1934 unsigned long arg4, unsigned long arg5)
1936 struct mm_struct *mm = current->mm;
1937 struct prctl_mm_map prctl_map;
1938 struct vm_area_struct *vma;
1941 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1942 opt != PR_SET_MM_MAP &&
1943 opt != PR_SET_MM_MAP_SIZE)))
1946 #ifdef CONFIG_CHECKPOINT_RESTORE
1947 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1948 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1951 if (!capable(CAP_SYS_RESOURCE))
1954 if (opt == PR_SET_MM_EXE_FILE)
1955 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1957 if (opt == PR_SET_MM_AUXV)
1958 return prctl_set_auxv(mm, addr, arg4);
1960 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1965 down_write(&mm->mmap_sem);
1966 vma = find_vma(mm, addr);
1968 prctl_map.start_code = mm->start_code;
1969 prctl_map.end_code = mm->end_code;
1970 prctl_map.start_data = mm->start_data;
1971 prctl_map.end_data = mm->end_data;
1972 prctl_map.start_brk = mm->start_brk;
1973 prctl_map.brk = mm->brk;
1974 prctl_map.start_stack = mm->start_stack;
1975 prctl_map.arg_start = mm->arg_start;
1976 prctl_map.arg_end = mm->arg_end;
1977 prctl_map.env_start = mm->env_start;
1978 prctl_map.env_end = mm->env_end;
1979 prctl_map.auxv = NULL;
1980 prctl_map.auxv_size = 0;
1981 prctl_map.exe_fd = -1;
1984 case PR_SET_MM_START_CODE:
1985 prctl_map.start_code = addr;
1987 case PR_SET_MM_END_CODE:
1988 prctl_map.end_code = addr;
1990 case PR_SET_MM_START_DATA:
1991 prctl_map.start_data = addr;
1993 case PR_SET_MM_END_DATA:
1994 prctl_map.end_data = addr;
1996 case PR_SET_MM_START_STACK:
1997 prctl_map.start_stack = addr;
1999 case PR_SET_MM_START_BRK:
2000 prctl_map.start_brk = addr;
2003 prctl_map.brk = addr;
2005 case PR_SET_MM_ARG_START:
2006 prctl_map.arg_start = addr;
2008 case PR_SET_MM_ARG_END:
2009 prctl_map.arg_end = addr;
2011 case PR_SET_MM_ENV_START:
2012 prctl_map.env_start = addr;
2014 case PR_SET_MM_ENV_END:
2015 prctl_map.env_end = addr;
2021 error = validate_prctl_map(&prctl_map);
2027 * If command line arguments and environment
2028 * are placed somewhere else on stack, we can
2029 * set them up here, ARG_START/END to setup
2030 * command line argumets and ENV_START/END
2033 case PR_SET_MM_START_STACK:
2034 case PR_SET_MM_ARG_START:
2035 case PR_SET_MM_ARG_END:
2036 case PR_SET_MM_ENV_START:
2037 case PR_SET_MM_ENV_END:
2044 mm->start_code = prctl_map.start_code;
2045 mm->end_code = prctl_map.end_code;
2046 mm->start_data = prctl_map.start_data;
2047 mm->end_data = prctl_map.end_data;
2048 mm->start_brk = prctl_map.start_brk;
2049 mm->brk = prctl_map.brk;
2050 mm->start_stack = prctl_map.start_stack;
2051 mm->arg_start = prctl_map.arg_start;
2052 mm->arg_end = prctl_map.arg_end;
2053 mm->env_start = prctl_map.env_start;
2054 mm->env_end = prctl_map.env_end;
2058 up_write(&mm->mmap_sem);
2062 #ifdef CONFIG_CHECKPOINT_RESTORE
2063 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2065 return put_user(me->clear_child_tid, tid_addr);
2068 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2074 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2077 * If task has has_child_subreaper - all its decendants
2078 * already have these flag too and new decendants will
2079 * inherit it on fork, skip them.
2081 * If we've found child_reaper - skip descendants in
2082 * it's subtree as they will never get out pidns.
2084 if (p->signal->has_child_subreaper ||
2085 is_child_reaper(task_pid(p)))
2088 p->signal->has_child_subreaper = 1;
2092 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2093 unsigned long, arg4, unsigned long, arg5)
2095 struct task_struct *me = current;
2096 unsigned char comm[sizeof(me->comm)];
2099 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2100 if (error != -ENOSYS)
2105 case PR_SET_PDEATHSIG:
2106 if (!valid_signal(arg2)) {
2110 me->pdeath_signal = arg2;
2112 case PR_GET_PDEATHSIG:
2113 error = put_user(me->pdeath_signal, (int __user *)arg2);
2115 case PR_GET_DUMPABLE:
2116 error = get_dumpable(me->mm);
2118 case PR_SET_DUMPABLE:
2119 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2123 set_dumpable(me->mm, arg2);
2126 case PR_SET_UNALIGN:
2127 error = SET_UNALIGN_CTL(me, arg2);
2129 case PR_GET_UNALIGN:
2130 error = GET_UNALIGN_CTL(me, arg2);
2133 error = SET_FPEMU_CTL(me, arg2);
2136 error = GET_FPEMU_CTL(me, arg2);
2139 error = SET_FPEXC_CTL(me, arg2);
2142 error = GET_FPEXC_CTL(me, arg2);
2145 error = PR_TIMING_STATISTICAL;
2148 if (arg2 != PR_TIMING_STATISTICAL)
2152 comm[sizeof(me->comm) - 1] = 0;
2153 if (strncpy_from_user(comm, (char __user *)arg2,
2154 sizeof(me->comm) - 1) < 0)
2156 set_task_comm(me, comm);
2157 proc_comm_connector(me);
2160 get_task_comm(comm, me);
2161 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2165 error = GET_ENDIAN(me, arg2);
2168 error = SET_ENDIAN(me, arg2);
2170 case PR_GET_SECCOMP:
2171 error = prctl_get_seccomp();
2173 case PR_SET_SECCOMP:
2174 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2177 error = GET_TSC_CTL(arg2);
2180 error = SET_TSC_CTL(arg2);
2182 case PR_TASK_PERF_EVENTS_DISABLE:
2183 error = perf_event_task_disable();
2185 case PR_TASK_PERF_EVENTS_ENABLE:
2186 error = perf_event_task_enable();
2188 case PR_GET_TIMERSLACK:
2189 if (current->timer_slack_ns > ULONG_MAX)
2192 error = current->timer_slack_ns;
2194 case PR_SET_TIMERSLACK:
2196 current->timer_slack_ns =
2197 current->default_timer_slack_ns;
2199 current->timer_slack_ns = arg2;
2205 case PR_MCE_KILL_CLEAR:
2208 current->flags &= ~PF_MCE_PROCESS;
2210 case PR_MCE_KILL_SET:
2211 current->flags |= PF_MCE_PROCESS;
2212 if (arg3 == PR_MCE_KILL_EARLY)
2213 current->flags |= PF_MCE_EARLY;
2214 else if (arg3 == PR_MCE_KILL_LATE)
2215 current->flags &= ~PF_MCE_EARLY;
2216 else if (arg3 == PR_MCE_KILL_DEFAULT)
2218 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2226 case PR_MCE_KILL_GET:
2227 if (arg2 | arg3 | arg4 | arg5)
2229 if (current->flags & PF_MCE_PROCESS)
2230 error = (current->flags & PF_MCE_EARLY) ?
2231 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2233 error = PR_MCE_KILL_DEFAULT;
2236 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2238 case PR_GET_TID_ADDRESS:
2239 error = prctl_get_tid_address(me, (int __user **)arg2);
2241 case PR_SET_CHILD_SUBREAPER:
2242 me->signal->is_child_subreaper = !!arg2;
2246 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2248 case PR_GET_CHILD_SUBREAPER:
2249 error = put_user(me->signal->is_child_subreaper,
2250 (int __user *)arg2);
2252 case PR_SET_NO_NEW_PRIVS:
2253 if (arg2 != 1 || arg3 || arg4 || arg5)
2256 task_set_no_new_privs(current);
2258 case PR_GET_NO_NEW_PRIVS:
2259 if (arg2 || arg3 || arg4 || arg5)
2261 return task_no_new_privs(current) ? 1 : 0;
2262 case PR_GET_THP_DISABLE:
2263 if (arg2 || arg3 || arg4 || arg5)
2265 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2267 case PR_SET_THP_DISABLE:
2268 if (arg3 || arg4 || arg5)
2270 if (down_write_killable(&me->mm->mmap_sem))
2273 me->mm->def_flags |= VM_NOHUGEPAGE;
2275 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2276 up_write(&me->mm->mmap_sem);
2278 case PR_MPX_ENABLE_MANAGEMENT:
2279 if (arg2 || arg3 || arg4 || arg5)
2281 error = MPX_ENABLE_MANAGEMENT();
2283 case PR_MPX_DISABLE_MANAGEMENT:
2284 if (arg2 || arg3 || arg4 || arg5)
2286 error = MPX_DISABLE_MANAGEMENT();
2288 case PR_SET_FP_MODE:
2289 error = SET_FP_MODE(me, arg2);
2291 case PR_GET_FP_MODE:
2292 error = GET_FP_MODE(me);
2301 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2302 struct getcpu_cache __user *, unused)
2305 int cpu = raw_smp_processor_id();
2308 err |= put_user(cpu, cpup);
2310 err |= put_user(cpu_to_node(cpu), nodep);
2311 return err ? -EFAULT : 0;
2315 * do_sysinfo - fill in sysinfo struct
2316 * @info: pointer to buffer to fill
2318 static int do_sysinfo(struct sysinfo *info)
2320 unsigned long mem_total, sav_total;
2321 unsigned int mem_unit, bitcount;
2324 memset(info, 0, sizeof(struct sysinfo));
2326 get_monotonic_boottime(&tp);
2327 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2329 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2331 info->procs = nr_threads;
2337 * If the sum of all the available memory (i.e. ram + swap)
2338 * is less than can be stored in a 32 bit unsigned long then
2339 * we can be binary compatible with 2.2.x kernels. If not,
2340 * well, in that case 2.2.x was broken anyways...
2342 * -Erik Andersen <andersee@debian.org>
2345 mem_total = info->totalram + info->totalswap;
2346 if (mem_total < info->totalram || mem_total < info->totalswap)
2349 mem_unit = info->mem_unit;
2350 while (mem_unit > 1) {
2353 sav_total = mem_total;
2355 if (mem_total < sav_total)
2360 * If mem_total did not overflow, multiply all memory values by
2361 * info->mem_unit and set it to 1. This leaves things compatible
2362 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2367 info->totalram <<= bitcount;
2368 info->freeram <<= bitcount;
2369 info->sharedram <<= bitcount;
2370 info->bufferram <<= bitcount;
2371 info->totalswap <<= bitcount;
2372 info->freeswap <<= bitcount;
2373 info->totalhigh <<= bitcount;
2374 info->freehigh <<= bitcount;
2380 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2386 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2392 #ifdef CONFIG_COMPAT
2393 struct compat_sysinfo {
2407 char _f[20-2*sizeof(u32)-sizeof(int)];
2410 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2416 /* Check to see if any memory value is too large for 32-bit and scale
2419 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2422 while (s.mem_unit < PAGE_SIZE) {
2427 s.totalram >>= bitcount;
2428 s.freeram >>= bitcount;
2429 s.sharedram >>= bitcount;
2430 s.bufferram >>= bitcount;
2431 s.totalswap >>= bitcount;
2432 s.freeswap >>= bitcount;
2433 s.totalhigh >>= bitcount;
2434 s.freehigh >>= bitcount;
2437 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2438 __put_user(s.uptime, &info->uptime) ||
2439 __put_user(s.loads[0], &info->loads[0]) ||
2440 __put_user(s.loads[1], &info->loads[1]) ||
2441 __put_user(s.loads[2], &info->loads[2]) ||
2442 __put_user(s.totalram, &info->totalram) ||
2443 __put_user(s.freeram, &info->freeram) ||
2444 __put_user(s.sharedram, &info->sharedram) ||
2445 __put_user(s.bufferram, &info->bufferram) ||
2446 __put_user(s.totalswap, &info->totalswap) ||
2447 __put_user(s.freeswap, &info->freeswap) ||
2448 __put_user(s.procs, &info->procs) ||
2449 __put_user(s.totalhigh, &info->totalhigh) ||
2450 __put_user(s.freehigh, &info->freehigh) ||
2451 __put_user(s.mem_unit, &info->mem_unit))
2456 #endif /* CONFIG_COMPAT */