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 static 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 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
918 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
921 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
925 struct compat_tms tmp;
928 /* Convert our struct tms to the compat version. */
929 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
930 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
931 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
932 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
933 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
936 force_successful_syscall_return();
937 return compat_jiffies_to_clock_t(jiffies);
942 * This needs some heavy checking ...
943 * I just haven't the stomach for it. I also don't fully
944 * understand sessions/pgrp etc. Let somebody who does explain it.
946 * OK, I think I have the protection semantics right.... this is really
947 * only important on a multi-user system anyway, to make sure one user
948 * can't send a signal to a process owned by another. -TYT, 12/12/91
950 * !PF_FORKNOEXEC check to conform completely to POSIX.
952 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
954 struct task_struct *p;
955 struct task_struct *group_leader = current->group_leader;
960 pid = task_pid_vnr(group_leader);
967 /* From this point forward we keep holding onto the tasklist lock
968 * so that our parent does not change from under us. -DaveM
970 write_lock_irq(&tasklist_lock);
973 p = find_task_by_vpid(pid);
978 if (!thread_group_leader(p))
981 if (same_thread_group(p->real_parent, group_leader)) {
983 if (task_session(p) != task_session(group_leader))
986 if (!(p->flags & PF_FORKNOEXEC))
990 if (p != group_leader)
995 if (p->signal->leader)
1000 struct task_struct *g;
1002 pgrp = find_vpid(pgid);
1003 g = pid_task(pgrp, PIDTYPE_PGID);
1004 if (!g || task_session(g) != task_session(group_leader))
1008 err = security_task_setpgid(p, pgid);
1012 if (task_pgrp(p) != pgrp)
1013 change_pid(p, PIDTYPE_PGID, pgrp);
1017 /* All paths lead to here, thus we are safe. -DaveM */
1018 write_unlock_irq(&tasklist_lock);
1023 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1025 struct task_struct *p;
1031 grp = task_pgrp(current);
1034 p = find_task_by_vpid(pid);
1041 retval = security_task_getpgid(p);
1045 retval = pid_vnr(grp);
1051 #ifdef __ARCH_WANT_SYS_GETPGRP
1053 SYSCALL_DEFINE0(getpgrp)
1055 return sys_getpgid(0);
1060 SYSCALL_DEFINE1(getsid, pid_t, pid)
1062 struct task_struct *p;
1068 sid = task_session(current);
1071 p = find_task_by_vpid(pid);
1074 sid = task_session(p);
1078 retval = security_task_getsid(p);
1082 retval = pid_vnr(sid);
1088 static void set_special_pids(struct pid *pid)
1090 struct task_struct *curr = current->group_leader;
1092 if (task_session(curr) != pid)
1093 change_pid(curr, PIDTYPE_SID, pid);
1095 if (task_pgrp(curr) != pid)
1096 change_pid(curr, PIDTYPE_PGID, pid);
1099 SYSCALL_DEFINE0(setsid)
1101 struct task_struct *group_leader = current->group_leader;
1102 struct pid *sid = task_pid(group_leader);
1103 pid_t session = pid_vnr(sid);
1106 write_lock_irq(&tasklist_lock);
1107 /* Fail if I am already a session leader */
1108 if (group_leader->signal->leader)
1111 /* Fail if a process group id already exists that equals the
1112 * proposed session id.
1114 if (pid_task(sid, PIDTYPE_PGID))
1117 group_leader->signal->leader = 1;
1118 set_special_pids(sid);
1120 proc_clear_tty(group_leader);
1124 write_unlock_irq(&tasklist_lock);
1126 proc_sid_connector(group_leader);
1127 sched_autogroup_create_attach(group_leader);
1132 DECLARE_RWSEM(uts_sem);
1134 #ifdef COMPAT_UTS_MACHINE
1135 #define override_architecture(name) \
1136 (personality(current->personality) == PER_LINUX32 && \
1137 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1138 sizeof(COMPAT_UTS_MACHINE)))
1140 #define override_architecture(name) 0
1144 * Work around broken programs that cannot handle "Linux 3.0".
1145 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1146 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1148 static int override_release(char __user *release, size_t len)
1152 if (current->personality & UNAME26) {
1153 const char *rest = UTS_RELEASE;
1154 char buf[65] = { 0 };
1160 if (*rest == '.' && ++ndots >= 3)
1162 if (!isdigit(*rest) && *rest != '.')
1166 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1167 copy = clamp_t(size_t, len, 1, sizeof(buf));
1168 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1169 ret = copy_to_user(release, buf, copy + 1);
1174 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1178 down_read(&uts_sem);
1179 if (copy_to_user(name, utsname(), sizeof *name))
1183 if (!errno && override_release(name->release, sizeof(name->release)))
1185 if (!errno && override_architecture(name))
1190 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1194 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1201 down_read(&uts_sem);
1202 if (copy_to_user(name, utsname(), sizeof(*name)))
1206 if (!error && override_release(name->release, sizeof(name->release)))
1208 if (!error && override_architecture(name))
1213 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1219 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1222 down_read(&uts_sem);
1223 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1225 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1226 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1228 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1229 error |= __copy_to_user(&name->release, &utsname()->release,
1231 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1232 error |= __copy_to_user(&name->version, &utsname()->version,
1234 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1235 error |= __copy_to_user(&name->machine, &utsname()->machine,
1237 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1240 if (!error && override_architecture(name))
1242 if (!error && override_release(name->release, sizeof(name->release)))
1244 return error ? -EFAULT : 0;
1248 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1251 char tmp[__NEW_UTS_LEN];
1253 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1256 if (len < 0 || len > __NEW_UTS_LEN)
1258 down_write(&uts_sem);
1260 if (!copy_from_user(tmp, name, len)) {
1261 struct new_utsname *u = utsname();
1263 memcpy(u->nodename, tmp, len);
1264 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1266 uts_proc_notify(UTS_PROC_HOSTNAME);
1272 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1274 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1277 struct new_utsname *u;
1281 down_read(&uts_sem);
1283 i = 1 + strlen(u->nodename);
1287 if (copy_to_user(name, u->nodename, i))
1296 * Only setdomainname; getdomainname can be implemented by calling
1299 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1302 char tmp[__NEW_UTS_LEN];
1304 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1306 if (len < 0 || len > __NEW_UTS_LEN)
1309 down_write(&uts_sem);
1311 if (!copy_from_user(tmp, name, len)) {
1312 struct new_utsname *u = utsname();
1314 memcpy(u->domainname, tmp, len);
1315 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1317 uts_proc_notify(UTS_PROC_DOMAINNAME);
1323 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1325 struct rlimit value;
1328 ret = do_prlimit(current, resource, NULL, &value);
1330 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1335 #ifdef CONFIG_COMPAT
1337 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1338 struct compat_rlimit __user *, rlim)
1341 struct compat_rlimit r32;
1343 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1346 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1347 r.rlim_cur = RLIM_INFINITY;
1349 r.rlim_cur = r32.rlim_cur;
1350 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1351 r.rlim_max = RLIM_INFINITY;
1353 r.rlim_max = r32.rlim_max;
1354 return do_prlimit(current, resource, &r, NULL);
1357 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1358 struct compat_rlimit __user *, rlim)
1363 ret = do_prlimit(current, resource, NULL, &r);
1365 struct compat_rlimit r32;
1366 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1367 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1369 r32.rlim_cur = r.rlim_cur;
1370 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1371 r32.rlim_max = COMPAT_RLIM_INFINITY;
1373 r32.rlim_max = r.rlim_max;
1375 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1383 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1386 * Back compatibility for getrlimit. Needed for some apps.
1388 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1389 struct rlimit __user *, rlim)
1392 if (resource >= RLIM_NLIMITS)
1395 task_lock(current->group_leader);
1396 x = current->signal->rlim[resource];
1397 task_unlock(current->group_leader);
1398 if (x.rlim_cur > 0x7FFFFFFF)
1399 x.rlim_cur = 0x7FFFFFFF;
1400 if (x.rlim_max > 0x7FFFFFFF)
1401 x.rlim_max = 0x7FFFFFFF;
1402 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1405 #ifdef CONFIG_COMPAT
1406 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1407 struct compat_rlimit __user *, rlim)
1411 if (resource >= RLIM_NLIMITS)
1414 task_lock(current->group_leader);
1415 r = current->signal->rlim[resource];
1416 task_unlock(current->group_leader);
1417 if (r.rlim_cur > 0x7FFFFFFF)
1418 r.rlim_cur = 0x7FFFFFFF;
1419 if (r.rlim_max > 0x7FFFFFFF)
1420 r.rlim_max = 0x7FFFFFFF;
1422 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1423 put_user(r.rlim_max, &rlim->rlim_max))
1431 static inline bool rlim64_is_infinity(__u64 rlim64)
1433 #if BITS_PER_LONG < 64
1434 return rlim64 >= ULONG_MAX;
1436 return rlim64 == RLIM64_INFINITY;
1440 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1442 if (rlim->rlim_cur == RLIM_INFINITY)
1443 rlim64->rlim_cur = RLIM64_INFINITY;
1445 rlim64->rlim_cur = rlim->rlim_cur;
1446 if (rlim->rlim_max == RLIM_INFINITY)
1447 rlim64->rlim_max = RLIM64_INFINITY;
1449 rlim64->rlim_max = rlim->rlim_max;
1452 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1454 if (rlim64_is_infinity(rlim64->rlim_cur))
1455 rlim->rlim_cur = RLIM_INFINITY;
1457 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1458 if (rlim64_is_infinity(rlim64->rlim_max))
1459 rlim->rlim_max = RLIM_INFINITY;
1461 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1464 /* make sure you are allowed to change @tsk limits before calling this */
1465 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1466 struct rlimit *new_rlim, struct rlimit *old_rlim)
1468 struct rlimit *rlim;
1471 if (resource >= RLIM_NLIMITS)
1474 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1476 if (resource == RLIMIT_NOFILE &&
1477 new_rlim->rlim_max > sysctl_nr_open)
1481 /* protect tsk->signal and tsk->sighand from disappearing */
1482 read_lock(&tasklist_lock);
1483 if (!tsk->sighand) {
1488 rlim = tsk->signal->rlim + resource;
1489 task_lock(tsk->group_leader);
1491 /* Keep the capable check against init_user_ns until
1492 cgroups can contain all limits */
1493 if (new_rlim->rlim_max > rlim->rlim_max &&
1494 !capable(CAP_SYS_RESOURCE))
1497 retval = security_task_setrlimit(tsk, resource, new_rlim);
1498 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1500 * The caller is asking for an immediate RLIMIT_CPU
1501 * expiry. But we use the zero value to mean "it was
1502 * never set". So let's cheat and make it one second
1505 new_rlim->rlim_cur = 1;
1514 task_unlock(tsk->group_leader);
1517 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1518 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1519 * very long-standing error, and fixing it now risks breakage of
1520 * applications, so we live with it
1522 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1523 new_rlim->rlim_cur != RLIM_INFINITY &&
1524 IS_ENABLED(CONFIG_POSIX_TIMERS))
1525 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1527 read_unlock(&tasklist_lock);
1531 /* rcu lock must be held */
1532 static int check_prlimit_permission(struct task_struct *task,
1535 const struct cred *cred = current_cred(), *tcred;
1538 if (current == task)
1541 tcred = __task_cred(task);
1542 id_match = (uid_eq(cred->uid, tcred->euid) &&
1543 uid_eq(cred->uid, tcred->suid) &&
1544 uid_eq(cred->uid, tcred->uid) &&
1545 gid_eq(cred->gid, tcred->egid) &&
1546 gid_eq(cred->gid, tcred->sgid) &&
1547 gid_eq(cred->gid, tcred->gid));
1548 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1551 return security_task_prlimit(cred, tcred, flags);
1554 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1555 const struct rlimit64 __user *, new_rlim,
1556 struct rlimit64 __user *, old_rlim)
1558 struct rlimit64 old64, new64;
1559 struct rlimit old, new;
1560 struct task_struct *tsk;
1561 unsigned int checkflags = 0;
1565 checkflags |= LSM_PRLIMIT_READ;
1568 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1570 rlim64_to_rlim(&new64, &new);
1571 checkflags |= LSM_PRLIMIT_WRITE;
1575 tsk = pid ? find_task_by_vpid(pid) : current;
1580 ret = check_prlimit_permission(tsk, checkflags);
1585 get_task_struct(tsk);
1588 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1589 old_rlim ? &old : NULL);
1591 if (!ret && old_rlim) {
1592 rlim_to_rlim64(&old, &old64);
1593 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1597 put_task_struct(tsk);
1601 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1603 struct rlimit new_rlim;
1605 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1607 return do_prlimit(current, resource, &new_rlim, NULL);
1611 * It would make sense to put struct rusage in the task_struct,
1612 * except that would make the task_struct be *really big*. After
1613 * task_struct gets moved into malloc'ed memory, it would
1614 * make sense to do this. It will make moving the rest of the information
1615 * a lot simpler! (Which we're not doing right now because we're not
1616 * measuring them yet).
1618 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1619 * races with threads incrementing their own counters. But since word
1620 * reads are atomic, we either get new values or old values and we don't
1621 * care which for the sums. We always take the siglock to protect reading
1622 * the c* fields from p->signal from races with exit.c updating those
1623 * fields when reaping, so a sample either gets all the additions of a
1624 * given child after it's reaped, or none so this sample is before reaping.
1627 * We need to take the siglock for CHILDEREN, SELF and BOTH
1628 * for the cases current multithreaded, non-current single threaded
1629 * non-current multithreaded. Thread traversal is now safe with
1631 * Strictly speaking, we donot need to take the siglock if we are current and
1632 * single threaded, as no one else can take our signal_struct away, no one
1633 * else can reap the children to update signal->c* counters, and no one else
1634 * can race with the signal-> fields. If we do not take any lock, the
1635 * signal-> fields could be read out of order while another thread was just
1636 * exiting. So we should place a read memory barrier when we avoid the lock.
1637 * On the writer side, write memory barrier is implied in __exit_signal
1638 * as __exit_signal releases the siglock spinlock after updating the signal->
1639 * fields. But we don't do this yet to keep things simple.
1643 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1645 r->ru_nvcsw += t->nvcsw;
1646 r->ru_nivcsw += t->nivcsw;
1647 r->ru_minflt += t->min_flt;
1648 r->ru_majflt += t->maj_flt;
1649 r->ru_inblock += task_io_get_inblock(t);
1650 r->ru_oublock += task_io_get_oublock(t);
1653 void getrusage(struct task_struct *p, int who, struct rusage *r)
1655 struct task_struct *t;
1656 unsigned long flags;
1657 u64 tgutime, tgstime, utime, stime;
1658 unsigned long maxrss = 0;
1660 memset((char *)r, 0, sizeof (*r));
1663 if (who == RUSAGE_THREAD) {
1664 task_cputime_adjusted(current, &utime, &stime);
1665 accumulate_thread_rusage(p, r);
1666 maxrss = p->signal->maxrss;
1670 if (!lock_task_sighand(p, &flags))
1675 case RUSAGE_CHILDREN:
1676 utime = p->signal->cutime;
1677 stime = p->signal->cstime;
1678 r->ru_nvcsw = p->signal->cnvcsw;
1679 r->ru_nivcsw = p->signal->cnivcsw;
1680 r->ru_minflt = p->signal->cmin_flt;
1681 r->ru_majflt = p->signal->cmaj_flt;
1682 r->ru_inblock = p->signal->cinblock;
1683 r->ru_oublock = p->signal->coublock;
1684 maxrss = p->signal->cmaxrss;
1686 if (who == RUSAGE_CHILDREN)
1690 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1693 r->ru_nvcsw += p->signal->nvcsw;
1694 r->ru_nivcsw += p->signal->nivcsw;
1695 r->ru_minflt += p->signal->min_flt;
1696 r->ru_majflt += p->signal->maj_flt;
1697 r->ru_inblock += p->signal->inblock;
1698 r->ru_oublock += p->signal->oublock;
1699 if (maxrss < p->signal->maxrss)
1700 maxrss = p->signal->maxrss;
1703 accumulate_thread_rusage(t, r);
1704 } while_each_thread(p, t);
1710 unlock_task_sighand(p, &flags);
1713 r->ru_utime = ns_to_timeval(utime);
1714 r->ru_stime = ns_to_timeval(stime);
1716 if (who != RUSAGE_CHILDREN) {
1717 struct mm_struct *mm = get_task_mm(p);
1720 setmax_mm_hiwater_rss(&maxrss, mm);
1724 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1727 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1731 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1732 who != RUSAGE_THREAD)
1735 getrusage(current, who, &r);
1736 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1739 #ifdef CONFIG_COMPAT
1740 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1744 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1745 who != RUSAGE_THREAD)
1748 getrusage(current, who, &r);
1749 return put_compat_rusage(&r, ru);
1753 SYSCALL_DEFINE1(umask, int, mask)
1755 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1759 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1762 struct file *old_exe, *exe_file;
1763 struct inode *inode;
1770 inode = file_inode(exe.file);
1773 * Because the original mm->exe_file points to executable file, make
1774 * sure that this one is executable as well, to avoid breaking an
1778 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1781 err = inode_permission(inode, MAY_EXEC);
1786 * Forbid mm->exe_file change if old file still mapped.
1788 exe_file = get_mm_exe_file(mm);
1791 struct vm_area_struct *vma;
1793 down_read(&mm->mmap_sem);
1794 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1797 if (path_equal(&vma->vm_file->f_path,
1802 up_read(&mm->mmap_sem);
1807 /* set the new file, lockless */
1809 old_exe = xchg(&mm->exe_file, exe.file);
1816 up_read(&mm->mmap_sem);
1822 * WARNING: we don't require any capability here so be very careful
1823 * in what is allowed for modification from userspace.
1825 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1827 unsigned long mmap_max_addr = TASK_SIZE;
1828 struct mm_struct *mm = current->mm;
1829 int error = -EINVAL, i;
1831 static const unsigned char offsets[] = {
1832 offsetof(struct prctl_mm_map, start_code),
1833 offsetof(struct prctl_mm_map, end_code),
1834 offsetof(struct prctl_mm_map, start_data),
1835 offsetof(struct prctl_mm_map, end_data),
1836 offsetof(struct prctl_mm_map, start_brk),
1837 offsetof(struct prctl_mm_map, brk),
1838 offsetof(struct prctl_mm_map, start_stack),
1839 offsetof(struct prctl_mm_map, arg_start),
1840 offsetof(struct prctl_mm_map, arg_end),
1841 offsetof(struct prctl_mm_map, env_start),
1842 offsetof(struct prctl_mm_map, env_end),
1846 * Make sure the members are not somewhere outside
1847 * of allowed address space.
1849 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1850 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1852 if ((unsigned long)val >= mmap_max_addr ||
1853 (unsigned long)val < mmap_min_addr)
1858 * Make sure the pairs are ordered.
1860 #define __prctl_check_order(__m1, __op, __m2) \
1861 ((unsigned long)prctl_map->__m1 __op \
1862 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1863 error = __prctl_check_order(start_code, <, end_code);
1864 error |= __prctl_check_order(start_data, <, end_data);
1865 error |= __prctl_check_order(start_brk, <=, brk);
1866 error |= __prctl_check_order(arg_start, <=, arg_end);
1867 error |= __prctl_check_order(env_start, <=, env_end);
1870 #undef __prctl_check_order
1875 * @brk should be after @end_data in traditional maps.
1877 if (prctl_map->start_brk <= prctl_map->end_data ||
1878 prctl_map->brk <= prctl_map->end_data)
1882 * Neither we should allow to override limits if they set.
1884 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1885 prctl_map->start_brk, prctl_map->end_data,
1886 prctl_map->start_data))
1890 * Someone is trying to cheat the auxv vector.
1892 if (prctl_map->auxv_size) {
1893 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1898 * Finally, make sure the caller has the rights to
1899 * change /proc/pid/exe link: only local root should
1902 if (prctl_map->exe_fd != (u32)-1) {
1903 struct user_namespace *ns = current_user_ns();
1904 const struct cred *cred = current_cred();
1906 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1907 !gid_eq(cred->gid, make_kgid(ns, 0)))
1916 #ifdef CONFIG_CHECKPOINT_RESTORE
1917 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1919 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1920 unsigned long user_auxv[AT_VECTOR_SIZE];
1921 struct mm_struct *mm = current->mm;
1924 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1925 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1927 if (opt == PR_SET_MM_MAP_SIZE)
1928 return put_user((unsigned int)sizeof(prctl_map),
1929 (unsigned int __user *)addr);
1931 if (data_size != sizeof(prctl_map))
1934 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1937 error = validate_prctl_map(&prctl_map);
1941 if (prctl_map.auxv_size) {
1942 memset(user_auxv, 0, sizeof(user_auxv));
1943 if (copy_from_user(user_auxv,
1944 (const void __user *)prctl_map.auxv,
1945 prctl_map.auxv_size))
1948 /* Last entry must be AT_NULL as specification requires */
1949 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1950 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1953 if (prctl_map.exe_fd != (u32)-1) {
1954 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1959 down_write(&mm->mmap_sem);
1962 * We don't validate if these members are pointing to
1963 * real present VMAs because application may have correspond
1964 * VMAs already unmapped and kernel uses these members for statistics
1965 * output in procfs mostly, except
1967 * - @start_brk/@brk which are used in do_brk but kernel lookups
1968 * for VMAs when updating these memvers so anything wrong written
1969 * here cause kernel to swear at userspace program but won't lead
1970 * to any problem in kernel itself
1973 mm->start_code = prctl_map.start_code;
1974 mm->end_code = prctl_map.end_code;
1975 mm->start_data = prctl_map.start_data;
1976 mm->end_data = prctl_map.end_data;
1977 mm->start_brk = prctl_map.start_brk;
1978 mm->brk = prctl_map.brk;
1979 mm->start_stack = prctl_map.start_stack;
1980 mm->arg_start = prctl_map.arg_start;
1981 mm->arg_end = prctl_map.arg_end;
1982 mm->env_start = prctl_map.env_start;
1983 mm->env_end = prctl_map.env_end;
1986 * Note this update of @saved_auxv is lockless thus
1987 * if someone reads this member in procfs while we're
1988 * updating -- it may get partly updated results. It's
1989 * known and acceptable trade off: we leave it as is to
1990 * not introduce additional locks here making the kernel
1993 if (prctl_map.auxv_size)
1994 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1996 up_write(&mm->mmap_sem);
1999 #endif /* CONFIG_CHECKPOINT_RESTORE */
2001 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2005 * This doesn't move the auxiliary vector itself since it's pinned to
2006 * mm_struct, but it permits filling the vector with new values. It's
2007 * up to the caller to provide sane values here, otherwise userspace
2008 * tools which use this vector might be unhappy.
2010 unsigned long user_auxv[AT_VECTOR_SIZE];
2012 if (len > sizeof(user_auxv))
2015 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2018 /* Make sure the last entry is always AT_NULL */
2019 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2020 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2022 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2025 memcpy(mm->saved_auxv, user_auxv, len);
2026 task_unlock(current);
2031 static int prctl_set_mm(int opt, unsigned long addr,
2032 unsigned long arg4, unsigned long arg5)
2034 struct mm_struct *mm = current->mm;
2035 struct prctl_mm_map prctl_map;
2036 struct vm_area_struct *vma;
2039 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2040 opt != PR_SET_MM_MAP &&
2041 opt != PR_SET_MM_MAP_SIZE)))
2044 #ifdef CONFIG_CHECKPOINT_RESTORE
2045 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2046 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2049 if (!capable(CAP_SYS_RESOURCE))
2052 if (opt == PR_SET_MM_EXE_FILE)
2053 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2055 if (opt == PR_SET_MM_AUXV)
2056 return prctl_set_auxv(mm, addr, arg4);
2058 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2063 down_write(&mm->mmap_sem);
2064 vma = find_vma(mm, addr);
2066 prctl_map.start_code = mm->start_code;
2067 prctl_map.end_code = mm->end_code;
2068 prctl_map.start_data = mm->start_data;
2069 prctl_map.end_data = mm->end_data;
2070 prctl_map.start_brk = mm->start_brk;
2071 prctl_map.brk = mm->brk;
2072 prctl_map.start_stack = mm->start_stack;
2073 prctl_map.arg_start = mm->arg_start;
2074 prctl_map.arg_end = mm->arg_end;
2075 prctl_map.env_start = mm->env_start;
2076 prctl_map.env_end = mm->env_end;
2077 prctl_map.auxv = NULL;
2078 prctl_map.auxv_size = 0;
2079 prctl_map.exe_fd = -1;
2082 case PR_SET_MM_START_CODE:
2083 prctl_map.start_code = addr;
2085 case PR_SET_MM_END_CODE:
2086 prctl_map.end_code = addr;
2088 case PR_SET_MM_START_DATA:
2089 prctl_map.start_data = addr;
2091 case PR_SET_MM_END_DATA:
2092 prctl_map.end_data = addr;
2094 case PR_SET_MM_START_STACK:
2095 prctl_map.start_stack = addr;
2097 case PR_SET_MM_START_BRK:
2098 prctl_map.start_brk = addr;
2101 prctl_map.brk = addr;
2103 case PR_SET_MM_ARG_START:
2104 prctl_map.arg_start = addr;
2106 case PR_SET_MM_ARG_END:
2107 prctl_map.arg_end = addr;
2109 case PR_SET_MM_ENV_START:
2110 prctl_map.env_start = addr;
2112 case PR_SET_MM_ENV_END:
2113 prctl_map.env_end = addr;
2119 error = validate_prctl_map(&prctl_map);
2125 * If command line arguments and environment
2126 * are placed somewhere else on stack, we can
2127 * set them up here, ARG_START/END to setup
2128 * command line argumets and ENV_START/END
2131 case PR_SET_MM_START_STACK:
2132 case PR_SET_MM_ARG_START:
2133 case PR_SET_MM_ARG_END:
2134 case PR_SET_MM_ENV_START:
2135 case PR_SET_MM_ENV_END:
2142 mm->start_code = prctl_map.start_code;
2143 mm->end_code = prctl_map.end_code;
2144 mm->start_data = prctl_map.start_data;
2145 mm->end_data = prctl_map.end_data;
2146 mm->start_brk = prctl_map.start_brk;
2147 mm->brk = prctl_map.brk;
2148 mm->start_stack = prctl_map.start_stack;
2149 mm->arg_start = prctl_map.arg_start;
2150 mm->arg_end = prctl_map.arg_end;
2151 mm->env_start = prctl_map.env_start;
2152 mm->env_end = prctl_map.env_end;
2156 up_write(&mm->mmap_sem);
2160 #ifdef CONFIG_CHECKPOINT_RESTORE
2161 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2163 return put_user(me->clear_child_tid, tid_addr);
2166 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2172 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2175 * If task has has_child_subreaper - all its decendants
2176 * already have these flag too and new decendants will
2177 * inherit it on fork, skip them.
2179 * If we've found child_reaper - skip descendants in
2180 * it's subtree as they will never get out pidns.
2182 if (p->signal->has_child_subreaper ||
2183 is_child_reaper(task_pid(p)))
2186 p->signal->has_child_subreaper = 1;
2190 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2191 unsigned long, arg4, unsigned long, arg5)
2193 struct task_struct *me = current;
2194 unsigned char comm[sizeof(me->comm)];
2197 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2198 if (error != -ENOSYS)
2203 case PR_SET_PDEATHSIG:
2204 if (!valid_signal(arg2)) {
2208 me->pdeath_signal = arg2;
2210 case PR_GET_PDEATHSIG:
2211 error = put_user(me->pdeath_signal, (int __user *)arg2);
2213 case PR_GET_DUMPABLE:
2214 error = get_dumpable(me->mm);
2216 case PR_SET_DUMPABLE:
2217 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2221 set_dumpable(me->mm, arg2);
2224 case PR_SET_UNALIGN:
2225 error = SET_UNALIGN_CTL(me, arg2);
2227 case PR_GET_UNALIGN:
2228 error = GET_UNALIGN_CTL(me, arg2);
2231 error = SET_FPEMU_CTL(me, arg2);
2234 error = GET_FPEMU_CTL(me, arg2);
2237 error = SET_FPEXC_CTL(me, arg2);
2240 error = GET_FPEXC_CTL(me, arg2);
2243 error = PR_TIMING_STATISTICAL;
2246 if (arg2 != PR_TIMING_STATISTICAL)
2250 comm[sizeof(me->comm) - 1] = 0;
2251 if (strncpy_from_user(comm, (char __user *)arg2,
2252 sizeof(me->comm) - 1) < 0)
2254 set_task_comm(me, comm);
2255 proc_comm_connector(me);
2258 get_task_comm(comm, me);
2259 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2263 error = GET_ENDIAN(me, arg2);
2266 error = SET_ENDIAN(me, arg2);
2268 case PR_GET_SECCOMP:
2269 error = prctl_get_seccomp();
2271 case PR_SET_SECCOMP:
2272 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2275 error = GET_TSC_CTL(arg2);
2278 error = SET_TSC_CTL(arg2);
2280 case PR_TASK_PERF_EVENTS_DISABLE:
2281 error = perf_event_task_disable();
2283 case PR_TASK_PERF_EVENTS_ENABLE:
2284 error = perf_event_task_enable();
2286 case PR_GET_TIMERSLACK:
2287 if (current->timer_slack_ns > ULONG_MAX)
2290 error = current->timer_slack_ns;
2292 case PR_SET_TIMERSLACK:
2294 current->timer_slack_ns =
2295 current->default_timer_slack_ns;
2297 current->timer_slack_ns = arg2;
2303 case PR_MCE_KILL_CLEAR:
2306 current->flags &= ~PF_MCE_PROCESS;
2308 case PR_MCE_KILL_SET:
2309 current->flags |= PF_MCE_PROCESS;
2310 if (arg3 == PR_MCE_KILL_EARLY)
2311 current->flags |= PF_MCE_EARLY;
2312 else if (arg3 == PR_MCE_KILL_LATE)
2313 current->flags &= ~PF_MCE_EARLY;
2314 else if (arg3 == PR_MCE_KILL_DEFAULT)
2316 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2324 case PR_MCE_KILL_GET:
2325 if (arg2 | arg3 | arg4 | arg5)
2327 if (current->flags & PF_MCE_PROCESS)
2328 error = (current->flags & PF_MCE_EARLY) ?
2329 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2331 error = PR_MCE_KILL_DEFAULT;
2334 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2336 case PR_GET_TID_ADDRESS:
2337 error = prctl_get_tid_address(me, (int __user **)arg2);
2339 case PR_SET_CHILD_SUBREAPER:
2340 me->signal->is_child_subreaper = !!arg2;
2344 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2346 case PR_GET_CHILD_SUBREAPER:
2347 error = put_user(me->signal->is_child_subreaper,
2348 (int __user *)arg2);
2350 case PR_SET_NO_NEW_PRIVS:
2351 if (arg2 != 1 || arg3 || arg4 || arg5)
2354 task_set_no_new_privs(current);
2356 case PR_GET_NO_NEW_PRIVS:
2357 if (arg2 || arg3 || arg4 || arg5)
2359 return task_no_new_privs(current) ? 1 : 0;
2360 case PR_GET_THP_DISABLE:
2361 if (arg2 || arg3 || arg4 || arg5)
2363 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2365 case PR_SET_THP_DISABLE:
2366 if (arg3 || arg4 || arg5)
2368 if (down_write_killable(&me->mm->mmap_sem))
2371 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2373 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2374 up_write(&me->mm->mmap_sem);
2376 case PR_MPX_ENABLE_MANAGEMENT:
2377 if (arg2 || arg3 || arg4 || arg5)
2379 error = MPX_ENABLE_MANAGEMENT();
2381 case PR_MPX_DISABLE_MANAGEMENT:
2382 if (arg2 || arg3 || arg4 || arg5)
2384 error = MPX_DISABLE_MANAGEMENT();
2386 case PR_SET_FP_MODE:
2387 error = SET_FP_MODE(me, arg2);
2389 case PR_GET_FP_MODE:
2390 error = GET_FP_MODE(me);
2399 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2400 struct getcpu_cache __user *, unused)
2403 int cpu = raw_smp_processor_id();
2406 err |= put_user(cpu, cpup);
2408 err |= put_user(cpu_to_node(cpu), nodep);
2409 return err ? -EFAULT : 0;
2413 * do_sysinfo - fill in sysinfo struct
2414 * @info: pointer to buffer to fill
2416 static int do_sysinfo(struct sysinfo *info)
2418 unsigned long mem_total, sav_total;
2419 unsigned int mem_unit, bitcount;
2422 memset(info, 0, sizeof(struct sysinfo));
2424 get_monotonic_boottime(&tp);
2425 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2427 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2429 info->procs = nr_threads;
2435 * If the sum of all the available memory (i.e. ram + swap)
2436 * is less than can be stored in a 32 bit unsigned long then
2437 * we can be binary compatible with 2.2.x kernels. If not,
2438 * well, in that case 2.2.x was broken anyways...
2440 * -Erik Andersen <andersee@debian.org>
2443 mem_total = info->totalram + info->totalswap;
2444 if (mem_total < info->totalram || mem_total < info->totalswap)
2447 mem_unit = info->mem_unit;
2448 while (mem_unit > 1) {
2451 sav_total = mem_total;
2453 if (mem_total < sav_total)
2458 * If mem_total did not overflow, multiply all memory values by
2459 * info->mem_unit and set it to 1. This leaves things compatible
2460 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2465 info->totalram <<= bitcount;
2466 info->freeram <<= bitcount;
2467 info->sharedram <<= bitcount;
2468 info->bufferram <<= bitcount;
2469 info->totalswap <<= bitcount;
2470 info->freeswap <<= bitcount;
2471 info->totalhigh <<= bitcount;
2472 info->freehigh <<= bitcount;
2478 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2484 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2490 #ifdef CONFIG_COMPAT
2491 struct compat_sysinfo {
2505 char _f[20-2*sizeof(u32)-sizeof(int)];
2508 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2514 /* Check to see if any memory value is too large for 32-bit and scale
2517 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2520 while (s.mem_unit < PAGE_SIZE) {
2525 s.totalram >>= bitcount;
2526 s.freeram >>= bitcount;
2527 s.sharedram >>= bitcount;
2528 s.bufferram >>= bitcount;
2529 s.totalswap >>= bitcount;
2530 s.freeswap >>= bitcount;
2531 s.totalhigh >>= bitcount;
2532 s.freehigh >>= bitcount;
2535 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2536 __put_user(s.uptime, &info->uptime) ||
2537 __put_user(s.loads[0], &info->loads[0]) ||
2538 __put_user(s.loads[1], &info->loads[1]) ||
2539 __put_user(s.loads[2], &info->loads[2]) ||
2540 __put_user(s.totalram, &info->totalram) ||
2541 __put_user(s.freeram, &info->freeram) ||
2542 __put_user(s.sharedram, &info->sharedram) ||
2543 __put_user(s.bufferram, &info->bufferram) ||
2544 __put_user(s.totalswap, &info->totalswap) ||
2545 __put_user(s.freeswap, &info->freeswap) ||
2546 __put_user(s.procs, &info->procs) ||
2547 __put_user(s.totalhigh, &info->totalhigh) ||
2548 __put_user(s.freehigh, &info->freehigh) ||
2549 __put_user(s.mem_unit, &info->mem_unit))
2554 #endif /* CONFIG_COMPAT */