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/rcupdate.h>
59 #include <linux/uidgid.h>
60 #include <linux/cred.h>
62 #include <linux/kmsg_dump.h>
63 /* Move somewhere else to avoid recompiling? */
64 #include <generated/utsrelease.h>
66 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
70 #ifndef SET_UNALIGN_CTL
71 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
73 #ifndef GET_UNALIGN_CTL
74 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
77 # define SET_FPEMU_CTL(a, b) (-EINVAL)
80 # define GET_FPEMU_CTL(a, b) (-EINVAL)
83 # define SET_FPEXC_CTL(a, b) (-EINVAL)
86 # define GET_FPEXC_CTL(a, b) (-EINVAL)
89 # define GET_ENDIAN(a, b) (-EINVAL)
92 # define SET_ENDIAN(a, b) (-EINVAL)
95 # define GET_TSC_CTL(a) (-EINVAL)
98 # define SET_TSC_CTL(a) (-EINVAL)
100 #ifndef MPX_ENABLE_MANAGEMENT
101 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
103 #ifndef MPX_DISABLE_MANAGEMENT
104 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
107 # define GET_FP_MODE(a) (-EINVAL)
110 # define SET_FP_MODE(a,b) (-EINVAL)
114 * this is where the system-wide overflow UID and GID are defined, for
115 * architectures that now have 32-bit UID/GID but didn't in the past
118 int overflowuid = DEFAULT_OVERFLOWUID;
119 int overflowgid = DEFAULT_OVERFLOWGID;
121 EXPORT_SYMBOL(overflowuid);
122 EXPORT_SYMBOL(overflowgid);
125 * the same as above, but for filesystems which can only store a 16-bit
126 * UID and GID. as such, this is needed on all architectures
129 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
130 int fs_overflowgid = DEFAULT_FS_OVERFLOWUID;
132 EXPORT_SYMBOL(fs_overflowuid);
133 EXPORT_SYMBOL(fs_overflowgid);
136 * Returns true if current's euid is same as p's uid or euid,
137 * or has CAP_SYS_NICE to p's user_ns.
139 * Called with rcu_read_lock, creds are safe
141 static bool set_one_prio_perm(struct task_struct *p)
143 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
145 if (uid_eq(pcred->uid, cred->euid) ||
146 uid_eq(pcred->euid, cred->euid))
148 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
154 * set the priority of a task
155 * - the caller must hold the RCU read lock
157 static int set_one_prio(struct task_struct *p, int niceval, int error)
161 if (!set_one_prio_perm(p)) {
165 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
169 no_nice = security_task_setnice(p, niceval);
176 set_user_nice(p, niceval);
181 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
183 struct task_struct *g, *p;
184 struct user_struct *user;
185 const struct cred *cred = current_cred();
190 if (which > PRIO_USER || which < PRIO_PROCESS)
193 /* normalize: avoid signed division (rounding problems) */
195 if (niceval < MIN_NICE)
197 if (niceval > MAX_NICE)
201 read_lock(&tasklist_lock);
205 p = find_task_by_vpid(who);
209 error = set_one_prio(p, niceval, error);
213 pgrp = find_vpid(who);
215 pgrp = task_pgrp(current);
216 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
217 error = set_one_prio(p, niceval, error);
218 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
221 uid = make_kuid(cred->user_ns, who);
225 else if (!uid_eq(uid, cred->uid)) {
226 user = find_user(uid);
228 goto out_unlock; /* No processes for this user */
230 do_each_thread(g, p) {
231 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
232 error = set_one_prio(p, niceval, error);
233 } while_each_thread(g, p);
234 if (!uid_eq(uid, cred->uid))
235 free_uid(user); /* For find_user() */
239 read_unlock(&tasklist_lock);
246 * Ugh. To avoid negative return values, "getpriority()" will
247 * not return the normal nice-value, but a negated value that
248 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
249 * to stay compatible.
251 SYSCALL_DEFINE2(getpriority, int, which, int, who)
253 struct task_struct *g, *p;
254 struct user_struct *user;
255 const struct cred *cred = current_cred();
256 long niceval, retval = -ESRCH;
260 if (which > PRIO_USER || which < PRIO_PROCESS)
264 read_lock(&tasklist_lock);
268 p = find_task_by_vpid(who);
272 niceval = nice_to_rlimit(task_nice(p));
273 if (niceval > retval)
279 pgrp = find_vpid(who);
281 pgrp = task_pgrp(current);
282 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
283 niceval = nice_to_rlimit(task_nice(p));
284 if (niceval > retval)
286 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
289 uid = make_kuid(cred->user_ns, who);
293 else if (!uid_eq(uid, cred->uid)) {
294 user = find_user(uid);
296 goto out_unlock; /* No processes for this user */
298 do_each_thread(g, p) {
299 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
300 niceval = nice_to_rlimit(task_nice(p));
301 if (niceval > retval)
304 } while_each_thread(g, p);
305 if (!uid_eq(uid, cred->uid))
306 free_uid(user); /* for find_user() */
310 read_unlock(&tasklist_lock);
317 * Unprivileged users may change the real gid to the effective gid
318 * or vice versa. (BSD-style)
320 * If you set the real gid at all, or set the effective gid to a value not
321 * equal to the real gid, then the saved gid is set to the new effective gid.
323 * This makes it possible for a setgid program to completely drop its
324 * privileges, which is often a useful assertion to make when you are doing
325 * a security audit over a program.
327 * The general idea is that a program which uses just setregid() will be
328 * 100% compatible with BSD. A program which uses just setgid() will be
329 * 100% compatible with POSIX with saved IDs.
331 * SMP: There are not races, the GIDs are checked only by filesystem
332 * operations (as far as semantic preservation is concerned).
334 #ifdef CONFIG_MULTIUSER
335 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
337 struct user_namespace *ns = current_user_ns();
338 const struct cred *old;
343 krgid = make_kgid(ns, rgid);
344 kegid = make_kgid(ns, egid);
346 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
348 if ((egid != (gid_t) -1) && !gid_valid(kegid))
351 new = prepare_creds();
354 old = current_cred();
357 if (rgid != (gid_t) -1) {
358 if (gid_eq(old->gid, krgid) ||
359 gid_eq(old->egid, krgid) ||
360 ns_capable(old->user_ns, CAP_SETGID))
365 if (egid != (gid_t) -1) {
366 if (gid_eq(old->gid, kegid) ||
367 gid_eq(old->egid, kegid) ||
368 gid_eq(old->sgid, kegid) ||
369 ns_capable(old->user_ns, CAP_SETGID))
375 if (rgid != (gid_t) -1 ||
376 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
377 new->sgid = new->egid;
378 new->fsgid = new->egid;
380 return commit_creds(new);
388 * setgid() is implemented like SysV w/ SAVED_IDS
390 * SMP: Same implicit races as above.
392 SYSCALL_DEFINE1(setgid, gid_t, gid)
394 struct user_namespace *ns = current_user_ns();
395 const struct cred *old;
400 kgid = make_kgid(ns, gid);
401 if (!gid_valid(kgid))
404 new = prepare_creds();
407 old = current_cred();
410 if (ns_capable(old->user_ns, CAP_SETGID))
411 new->gid = new->egid = new->sgid = new->fsgid = kgid;
412 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
413 new->egid = new->fsgid = kgid;
417 return commit_creds(new);
425 * change the user struct in a credentials set to match the new UID
427 static int set_user(struct cred *new)
429 struct user_struct *new_user;
431 new_user = alloc_uid(new->uid);
436 * We don't fail in case of NPROC limit excess here because too many
437 * poorly written programs don't check set*uid() return code, assuming
438 * it never fails if called by root. We may still enforce NPROC limit
439 * for programs doing set*uid()+execve() by harmlessly deferring the
440 * failure to the execve() stage.
442 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
443 new_user != INIT_USER)
444 current->flags |= PF_NPROC_EXCEEDED;
446 current->flags &= ~PF_NPROC_EXCEEDED;
449 new->user = new_user;
454 * Unprivileged users may change the real uid to the effective uid
455 * or vice versa. (BSD-style)
457 * If you set the real uid at all, or set the effective uid to a value not
458 * equal to the real uid, then the saved uid is set to the new effective uid.
460 * This makes it possible for a setuid program to completely drop its
461 * privileges, which is often a useful assertion to make when you are doing
462 * a security audit over a program.
464 * The general idea is that a program which uses just setreuid() will be
465 * 100% compatible with BSD. A program which uses just setuid() will be
466 * 100% compatible with POSIX with saved IDs.
468 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
470 struct user_namespace *ns = current_user_ns();
471 const struct cred *old;
476 kruid = make_kuid(ns, ruid);
477 keuid = make_kuid(ns, euid);
479 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
481 if ((euid != (uid_t) -1) && !uid_valid(keuid))
484 new = prepare_creds();
487 old = current_cred();
490 if (ruid != (uid_t) -1) {
492 if (!uid_eq(old->uid, kruid) &&
493 !uid_eq(old->euid, kruid) &&
494 !ns_capable(old->user_ns, CAP_SETUID))
498 if (euid != (uid_t) -1) {
500 if (!uid_eq(old->uid, keuid) &&
501 !uid_eq(old->euid, keuid) &&
502 !uid_eq(old->suid, keuid) &&
503 !ns_capable(old->user_ns, CAP_SETUID))
507 if (!uid_eq(new->uid, old->uid)) {
508 retval = set_user(new);
512 if (ruid != (uid_t) -1 ||
513 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
514 new->suid = new->euid;
515 new->fsuid = new->euid;
517 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
521 return commit_creds(new);
529 * setuid() is implemented like SysV with SAVED_IDS
531 * Note that SAVED_ID's is deficient in that a setuid root program
532 * like sendmail, for example, cannot set its uid to be a normal
533 * user and then switch back, because if you're root, setuid() sets
534 * the saved uid too. If you don't like this, blame the bright people
535 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
536 * will allow a root program to temporarily drop privileges and be able to
537 * regain them by swapping the real and effective uid.
539 SYSCALL_DEFINE1(setuid, uid_t, uid)
541 struct user_namespace *ns = current_user_ns();
542 const struct cred *old;
547 kuid = make_kuid(ns, uid);
548 if (!uid_valid(kuid))
551 new = prepare_creds();
554 old = current_cred();
557 if (ns_capable(old->user_ns, CAP_SETUID)) {
558 new->suid = new->uid = kuid;
559 if (!uid_eq(kuid, old->uid)) {
560 retval = set_user(new);
564 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
568 new->fsuid = new->euid = kuid;
570 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
574 return commit_creds(new);
583 * This function implements a generic ability to update ruid, euid,
584 * and suid. This allows you to implement the 4.4 compatible seteuid().
586 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
588 struct user_namespace *ns = current_user_ns();
589 const struct cred *old;
592 kuid_t kruid, keuid, ksuid;
594 kruid = make_kuid(ns, ruid);
595 keuid = make_kuid(ns, euid);
596 ksuid = make_kuid(ns, suid);
598 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
601 if ((euid != (uid_t) -1) && !uid_valid(keuid))
604 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
607 new = prepare_creds();
611 old = current_cred();
614 if (!ns_capable(old->user_ns, CAP_SETUID)) {
615 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
616 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
618 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
619 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
621 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
622 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
626 if (ruid != (uid_t) -1) {
628 if (!uid_eq(kruid, old->uid)) {
629 retval = set_user(new);
634 if (euid != (uid_t) -1)
636 if (suid != (uid_t) -1)
638 new->fsuid = new->euid;
640 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
644 return commit_creds(new);
651 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
653 const struct cred *cred = current_cred();
655 uid_t ruid, euid, suid;
657 ruid = from_kuid_munged(cred->user_ns, cred->uid);
658 euid = from_kuid_munged(cred->user_ns, cred->euid);
659 suid = from_kuid_munged(cred->user_ns, cred->suid);
661 retval = put_user(ruid, ruidp);
663 retval = put_user(euid, euidp);
665 return put_user(suid, suidp);
671 * Same as above, but for rgid, egid, sgid.
673 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
675 struct user_namespace *ns = current_user_ns();
676 const struct cred *old;
679 kgid_t krgid, kegid, ksgid;
681 krgid = make_kgid(ns, rgid);
682 kegid = make_kgid(ns, egid);
683 ksgid = make_kgid(ns, sgid);
685 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
687 if ((egid != (gid_t) -1) && !gid_valid(kegid))
689 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
692 new = prepare_creds();
695 old = current_cred();
698 if (!ns_capable(old->user_ns, CAP_SETGID)) {
699 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
700 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
702 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
703 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
705 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
706 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
710 if (rgid != (gid_t) -1)
712 if (egid != (gid_t) -1)
714 if (sgid != (gid_t) -1)
716 new->fsgid = new->egid;
718 return commit_creds(new);
725 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
727 const struct cred *cred = current_cred();
729 gid_t rgid, egid, sgid;
731 rgid = from_kgid_munged(cred->user_ns, cred->gid);
732 egid = from_kgid_munged(cred->user_ns, cred->egid);
733 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
735 retval = put_user(rgid, rgidp);
737 retval = put_user(egid, egidp);
739 retval = put_user(sgid, sgidp);
747 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
748 * is used for "access()" and for the NFS daemon (letting nfsd stay at
749 * whatever uid it wants to). It normally shadows "euid", except when
750 * explicitly set by setfsuid() or for access..
752 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
754 const struct cred *old;
759 old = current_cred();
760 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
762 kuid = make_kuid(old->user_ns, uid);
763 if (!uid_valid(kuid))
766 new = prepare_creds();
770 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
771 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
772 ns_capable(old->user_ns, CAP_SETUID)) {
773 if (!uid_eq(kuid, old->fsuid)) {
775 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
789 * Samma på svenska..
791 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
793 const struct cred *old;
798 old = current_cred();
799 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
801 kgid = make_kgid(old->user_ns, gid);
802 if (!gid_valid(kgid))
805 new = prepare_creds();
809 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
810 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
811 ns_capable(old->user_ns, CAP_SETGID)) {
812 if (!gid_eq(kgid, old->fsgid)) {
825 #endif /* CONFIG_MULTIUSER */
828 * sys_getpid - return the thread group id of the current process
830 * Note, despite the name, this returns the tgid not the pid. The tgid and
831 * the pid are identical unless CLONE_THREAD was specified on clone() in
832 * which case the tgid is the same in all threads of the same group.
834 * This is SMP safe as current->tgid does not change.
836 SYSCALL_DEFINE0(getpid)
838 return task_tgid_vnr(current);
841 /* Thread ID - the internal kernel "pid" */
842 SYSCALL_DEFINE0(gettid)
844 return task_pid_vnr(current);
848 * Accessing ->real_parent is not SMP-safe, it could
849 * change from under us. However, we can use a stale
850 * value of ->real_parent under rcu_read_lock(), see
851 * release_task()->call_rcu(delayed_put_task_struct).
853 SYSCALL_DEFINE0(getppid)
858 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
864 SYSCALL_DEFINE0(getuid)
866 /* Only we change this so SMP safe */
867 return from_kuid_munged(current_user_ns(), current_uid());
870 SYSCALL_DEFINE0(geteuid)
872 /* Only we change this so SMP safe */
873 return from_kuid_munged(current_user_ns(), current_euid());
876 SYSCALL_DEFINE0(getgid)
878 /* Only we change this so SMP safe */
879 return from_kgid_munged(current_user_ns(), current_gid());
882 SYSCALL_DEFINE0(getegid)
884 /* Only we change this so SMP safe */
885 return from_kgid_munged(current_user_ns(), current_egid());
888 void do_sys_times(struct tms *tms)
890 u64 tgutime, tgstime, cutime, cstime;
892 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
893 cutime = current->signal->cutime;
894 cstime = current->signal->cstime;
895 tms->tms_utime = nsec_to_clock_t(tgutime);
896 tms->tms_stime = nsec_to_clock_t(tgstime);
897 tms->tms_cutime = nsec_to_clock_t(cutime);
898 tms->tms_cstime = nsec_to_clock_t(cstime);
901 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
907 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
910 force_successful_syscall_return();
911 return (long) jiffies_64_to_clock_t(get_jiffies_64());
915 * This needs some heavy checking ...
916 * I just haven't the stomach for it. I also don't fully
917 * understand sessions/pgrp etc. Let somebody who does explain it.
919 * OK, I think I have the protection semantics right.... this is really
920 * only important on a multi-user system anyway, to make sure one user
921 * can't send a signal to a process owned by another. -TYT, 12/12/91
923 * !PF_FORKNOEXEC check to conform completely to POSIX.
925 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
927 struct task_struct *p;
928 struct task_struct *group_leader = current->group_leader;
933 pid = task_pid_vnr(group_leader);
940 /* From this point forward we keep holding onto the tasklist lock
941 * so that our parent does not change from under us. -DaveM
943 write_lock_irq(&tasklist_lock);
946 p = find_task_by_vpid(pid);
951 if (!thread_group_leader(p))
954 if (same_thread_group(p->real_parent, group_leader)) {
956 if (task_session(p) != task_session(group_leader))
959 if (!(p->flags & PF_FORKNOEXEC))
963 if (p != group_leader)
968 if (p->signal->leader)
973 struct task_struct *g;
975 pgrp = find_vpid(pgid);
976 g = pid_task(pgrp, PIDTYPE_PGID);
977 if (!g || task_session(g) != task_session(group_leader))
981 err = security_task_setpgid(p, pgid);
985 if (task_pgrp(p) != pgrp)
986 change_pid(p, PIDTYPE_PGID, pgrp);
990 /* All paths lead to here, thus we are safe. -DaveM */
991 write_unlock_irq(&tasklist_lock);
996 SYSCALL_DEFINE1(getpgid, pid_t, pid)
998 struct task_struct *p;
1004 grp = task_pgrp(current);
1007 p = find_task_by_vpid(pid);
1014 retval = security_task_getpgid(p);
1018 retval = pid_vnr(grp);
1024 #ifdef __ARCH_WANT_SYS_GETPGRP
1026 SYSCALL_DEFINE0(getpgrp)
1028 return sys_getpgid(0);
1033 SYSCALL_DEFINE1(getsid, pid_t, pid)
1035 struct task_struct *p;
1041 sid = task_session(current);
1044 p = find_task_by_vpid(pid);
1047 sid = task_session(p);
1051 retval = security_task_getsid(p);
1055 retval = pid_vnr(sid);
1061 static void set_special_pids(struct pid *pid)
1063 struct task_struct *curr = current->group_leader;
1065 if (task_session(curr) != pid)
1066 change_pid(curr, PIDTYPE_SID, pid);
1068 if (task_pgrp(curr) != pid)
1069 change_pid(curr, PIDTYPE_PGID, pid);
1072 SYSCALL_DEFINE0(setsid)
1074 struct task_struct *group_leader = current->group_leader;
1075 struct pid *sid = task_pid(group_leader);
1076 pid_t session = pid_vnr(sid);
1079 write_lock_irq(&tasklist_lock);
1080 /* Fail if I am already a session leader */
1081 if (group_leader->signal->leader)
1084 /* Fail if a process group id already exists that equals the
1085 * proposed session id.
1087 if (pid_task(sid, PIDTYPE_PGID))
1090 group_leader->signal->leader = 1;
1091 set_special_pids(sid);
1093 proc_clear_tty(group_leader);
1097 write_unlock_irq(&tasklist_lock);
1099 proc_sid_connector(group_leader);
1100 sched_autogroup_create_attach(group_leader);
1105 DECLARE_RWSEM(uts_sem);
1107 #ifdef COMPAT_UTS_MACHINE
1108 #define override_architecture(name) \
1109 (personality(current->personality) == PER_LINUX32 && \
1110 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1111 sizeof(COMPAT_UTS_MACHINE)))
1113 #define override_architecture(name) 0
1117 * Work around broken programs that cannot handle "Linux 3.0".
1118 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1119 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1121 static int override_release(char __user *release, size_t len)
1125 if (current->personality & UNAME26) {
1126 const char *rest = UTS_RELEASE;
1127 char buf[65] = { 0 };
1133 if (*rest == '.' && ++ndots >= 3)
1135 if (!isdigit(*rest) && *rest != '.')
1139 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1140 copy = clamp_t(size_t, len, 1, sizeof(buf));
1141 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1142 ret = copy_to_user(release, buf, copy + 1);
1147 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1151 down_read(&uts_sem);
1152 if (copy_to_user(name, utsname(), sizeof *name))
1156 if (!errno && override_release(name->release, sizeof(name->release)))
1158 if (!errno && override_architecture(name))
1163 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1167 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1174 down_read(&uts_sem);
1175 if (copy_to_user(name, utsname(), sizeof(*name)))
1179 if (!error && override_release(name->release, sizeof(name->release)))
1181 if (!error && override_architecture(name))
1186 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1192 if (!access_ok(VERIFY_WRITE, name, sizeof(struct oldold_utsname)))
1195 down_read(&uts_sem);
1196 error = __copy_to_user(&name->sysname, &utsname()->sysname,
1198 error |= __put_user(0, name->sysname + __OLD_UTS_LEN);
1199 error |= __copy_to_user(&name->nodename, &utsname()->nodename,
1201 error |= __put_user(0, name->nodename + __OLD_UTS_LEN);
1202 error |= __copy_to_user(&name->release, &utsname()->release,
1204 error |= __put_user(0, name->release + __OLD_UTS_LEN);
1205 error |= __copy_to_user(&name->version, &utsname()->version,
1207 error |= __put_user(0, name->version + __OLD_UTS_LEN);
1208 error |= __copy_to_user(&name->machine, &utsname()->machine,
1210 error |= __put_user(0, name->machine + __OLD_UTS_LEN);
1213 if (!error && override_architecture(name))
1215 if (!error && override_release(name->release, sizeof(name->release)))
1217 return error ? -EFAULT : 0;
1221 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1224 char tmp[__NEW_UTS_LEN];
1226 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1229 if (len < 0 || len > __NEW_UTS_LEN)
1231 down_write(&uts_sem);
1233 if (!copy_from_user(tmp, name, len)) {
1234 struct new_utsname *u = utsname();
1236 memcpy(u->nodename, tmp, len);
1237 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1239 uts_proc_notify(UTS_PROC_HOSTNAME);
1245 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1247 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1250 struct new_utsname *u;
1254 down_read(&uts_sem);
1256 i = 1 + strlen(u->nodename);
1260 if (copy_to_user(name, u->nodename, i))
1269 * Only setdomainname; getdomainname can be implemented by calling
1272 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1275 char tmp[__NEW_UTS_LEN];
1277 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1279 if (len < 0 || len > __NEW_UTS_LEN)
1282 down_write(&uts_sem);
1284 if (!copy_from_user(tmp, name, len)) {
1285 struct new_utsname *u = utsname();
1287 memcpy(u->domainname, tmp, len);
1288 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1290 uts_proc_notify(UTS_PROC_DOMAINNAME);
1296 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1298 struct rlimit value;
1301 ret = do_prlimit(current, resource, NULL, &value);
1303 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1308 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1311 * Back compatibility for getrlimit. Needed for some apps.
1313 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1314 struct rlimit __user *, rlim)
1317 if (resource >= RLIM_NLIMITS)
1320 task_lock(current->group_leader);
1321 x = current->signal->rlim[resource];
1322 task_unlock(current->group_leader);
1323 if (x.rlim_cur > 0x7FFFFFFF)
1324 x.rlim_cur = 0x7FFFFFFF;
1325 if (x.rlim_max > 0x7FFFFFFF)
1326 x.rlim_max = 0x7FFFFFFF;
1327 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1332 static inline bool rlim64_is_infinity(__u64 rlim64)
1334 #if BITS_PER_LONG < 64
1335 return rlim64 >= ULONG_MAX;
1337 return rlim64 == RLIM64_INFINITY;
1341 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1343 if (rlim->rlim_cur == RLIM_INFINITY)
1344 rlim64->rlim_cur = RLIM64_INFINITY;
1346 rlim64->rlim_cur = rlim->rlim_cur;
1347 if (rlim->rlim_max == RLIM_INFINITY)
1348 rlim64->rlim_max = RLIM64_INFINITY;
1350 rlim64->rlim_max = rlim->rlim_max;
1353 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1355 if (rlim64_is_infinity(rlim64->rlim_cur))
1356 rlim->rlim_cur = RLIM_INFINITY;
1358 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1359 if (rlim64_is_infinity(rlim64->rlim_max))
1360 rlim->rlim_max = RLIM_INFINITY;
1362 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1365 /* make sure you are allowed to change @tsk limits before calling this */
1366 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1367 struct rlimit *new_rlim, struct rlimit *old_rlim)
1369 struct rlimit *rlim;
1372 if (resource >= RLIM_NLIMITS)
1375 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1377 if (resource == RLIMIT_NOFILE &&
1378 new_rlim->rlim_max > sysctl_nr_open)
1382 /* protect tsk->signal and tsk->sighand from disappearing */
1383 read_lock(&tasklist_lock);
1384 if (!tsk->sighand) {
1389 rlim = tsk->signal->rlim + resource;
1390 task_lock(tsk->group_leader);
1392 /* Keep the capable check against init_user_ns until
1393 cgroups can contain all limits */
1394 if (new_rlim->rlim_max > rlim->rlim_max &&
1395 !capable(CAP_SYS_RESOURCE))
1398 retval = security_task_setrlimit(tsk->group_leader,
1399 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)
1436 const struct cred *cred = current_cred(), *tcred;
1438 if (current == task)
1441 tcred = __task_cred(task);
1442 if (uid_eq(cred->uid, tcred->euid) &&
1443 uid_eq(cred->uid, tcred->suid) &&
1444 uid_eq(cred->uid, tcred->uid) &&
1445 gid_eq(cred->gid, tcred->egid) &&
1446 gid_eq(cred->gid, tcred->sgid) &&
1447 gid_eq(cred->gid, tcred->gid))
1449 if (ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1455 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1456 const struct rlimit64 __user *, new_rlim,
1457 struct rlimit64 __user *, old_rlim)
1459 struct rlimit64 old64, new64;
1460 struct rlimit old, new;
1461 struct task_struct *tsk;
1465 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1467 rlim64_to_rlim(&new64, &new);
1471 tsk = pid ? find_task_by_vpid(pid) : current;
1476 ret = check_prlimit_permission(tsk);
1481 get_task_struct(tsk);
1484 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1485 old_rlim ? &old : NULL);
1487 if (!ret && old_rlim) {
1488 rlim_to_rlim64(&old, &old64);
1489 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1493 put_task_struct(tsk);
1497 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1499 struct rlimit new_rlim;
1501 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1503 return do_prlimit(current, resource, &new_rlim, NULL);
1507 * It would make sense to put struct rusage in the task_struct,
1508 * except that would make the task_struct be *really big*. After
1509 * task_struct gets moved into malloc'ed memory, it would
1510 * make sense to do this. It will make moving the rest of the information
1511 * a lot simpler! (Which we're not doing right now because we're not
1512 * measuring them yet).
1514 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1515 * races with threads incrementing their own counters. But since word
1516 * reads are atomic, we either get new values or old values and we don't
1517 * care which for the sums. We always take the siglock to protect reading
1518 * the c* fields from p->signal from races with exit.c updating those
1519 * fields when reaping, so a sample either gets all the additions of a
1520 * given child after it's reaped, or none so this sample is before reaping.
1523 * We need to take the siglock for CHILDEREN, SELF and BOTH
1524 * for the cases current multithreaded, non-current single threaded
1525 * non-current multithreaded. Thread traversal is now safe with
1527 * Strictly speaking, we donot need to take the siglock if we are current and
1528 * single threaded, as no one else can take our signal_struct away, no one
1529 * else can reap the children to update signal->c* counters, and no one else
1530 * can race with the signal-> fields. If we do not take any lock, the
1531 * signal-> fields could be read out of order while another thread was just
1532 * exiting. So we should place a read memory barrier when we avoid the lock.
1533 * On the writer side, write memory barrier is implied in __exit_signal
1534 * as __exit_signal releases the siglock spinlock after updating the signal->
1535 * fields. But we don't do this yet to keep things simple.
1539 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1541 r->ru_nvcsw += t->nvcsw;
1542 r->ru_nivcsw += t->nivcsw;
1543 r->ru_minflt += t->min_flt;
1544 r->ru_majflt += t->maj_flt;
1545 r->ru_inblock += task_io_get_inblock(t);
1546 r->ru_oublock += task_io_get_oublock(t);
1549 static void k_getrusage(struct task_struct *p, int who, struct rusage *r)
1551 struct task_struct *t;
1552 unsigned long flags;
1553 u64 tgutime, tgstime, utime, stime;
1554 unsigned long maxrss = 0;
1556 memset((char *)r, 0, sizeof (*r));
1559 if (who == RUSAGE_THREAD) {
1560 task_cputime_adjusted(current, &utime, &stime);
1561 accumulate_thread_rusage(p, r);
1562 maxrss = p->signal->maxrss;
1566 if (!lock_task_sighand(p, &flags))
1571 case RUSAGE_CHILDREN:
1572 utime = p->signal->cutime;
1573 stime = p->signal->cstime;
1574 r->ru_nvcsw = p->signal->cnvcsw;
1575 r->ru_nivcsw = p->signal->cnivcsw;
1576 r->ru_minflt = p->signal->cmin_flt;
1577 r->ru_majflt = p->signal->cmaj_flt;
1578 r->ru_inblock = p->signal->cinblock;
1579 r->ru_oublock = p->signal->coublock;
1580 maxrss = p->signal->cmaxrss;
1582 if (who == RUSAGE_CHILDREN)
1586 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1589 r->ru_nvcsw += p->signal->nvcsw;
1590 r->ru_nivcsw += p->signal->nivcsw;
1591 r->ru_minflt += p->signal->min_flt;
1592 r->ru_majflt += p->signal->maj_flt;
1593 r->ru_inblock += p->signal->inblock;
1594 r->ru_oublock += p->signal->oublock;
1595 if (maxrss < p->signal->maxrss)
1596 maxrss = p->signal->maxrss;
1599 accumulate_thread_rusage(t, r);
1600 } while_each_thread(p, t);
1606 unlock_task_sighand(p, &flags);
1609 r->ru_utime = ns_to_timeval(utime);
1610 r->ru_stime = ns_to_timeval(stime);
1612 if (who != RUSAGE_CHILDREN) {
1613 struct mm_struct *mm = get_task_mm(p);
1616 setmax_mm_hiwater_rss(&maxrss, mm);
1620 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1623 int getrusage(struct task_struct *p, int who, struct rusage __user *ru)
1627 k_getrusage(p, who, &r);
1628 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1631 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1633 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1634 who != RUSAGE_THREAD)
1636 return getrusage(current, who, ru);
1639 #ifdef CONFIG_COMPAT
1640 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1644 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1645 who != RUSAGE_THREAD)
1648 k_getrusage(current, who, &r);
1649 return put_compat_rusage(&r, ru);
1653 SYSCALL_DEFINE1(umask, int, mask)
1655 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1659 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1662 struct file *old_exe, *exe_file;
1663 struct inode *inode;
1670 inode = file_inode(exe.file);
1673 * Because the original mm->exe_file points to executable file, make
1674 * sure that this one is executable as well, to avoid breaking an
1678 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1681 err = inode_permission(inode, MAY_EXEC);
1686 * Forbid mm->exe_file change if old file still mapped.
1688 exe_file = get_mm_exe_file(mm);
1691 struct vm_area_struct *vma;
1693 down_read(&mm->mmap_sem);
1694 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1697 if (path_equal(&vma->vm_file->f_path,
1702 up_read(&mm->mmap_sem);
1707 /* set the new file, lockless */
1709 old_exe = xchg(&mm->exe_file, exe.file);
1716 up_read(&mm->mmap_sem);
1722 * WARNING: we don't require any capability here so be very careful
1723 * in what is allowed for modification from userspace.
1725 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1727 unsigned long mmap_max_addr = TASK_SIZE;
1728 struct mm_struct *mm = current->mm;
1729 int error = -EINVAL, i;
1731 static const unsigned char offsets[] = {
1732 offsetof(struct prctl_mm_map, start_code),
1733 offsetof(struct prctl_mm_map, end_code),
1734 offsetof(struct prctl_mm_map, start_data),
1735 offsetof(struct prctl_mm_map, end_data),
1736 offsetof(struct prctl_mm_map, start_brk),
1737 offsetof(struct prctl_mm_map, brk),
1738 offsetof(struct prctl_mm_map, start_stack),
1739 offsetof(struct prctl_mm_map, arg_start),
1740 offsetof(struct prctl_mm_map, arg_end),
1741 offsetof(struct prctl_mm_map, env_start),
1742 offsetof(struct prctl_mm_map, env_end),
1746 * Make sure the members are not somewhere outside
1747 * of allowed address space.
1749 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1750 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1752 if ((unsigned long)val >= mmap_max_addr ||
1753 (unsigned long)val < mmap_min_addr)
1758 * Make sure the pairs are ordered.
1760 #define __prctl_check_order(__m1, __op, __m2) \
1761 ((unsigned long)prctl_map->__m1 __op \
1762 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1763 error = __prctl_check_order(start_code, <, end_code);
1764 error |= __prctl_check_order(start_data, <, end_data);
1765 error |= __prctl_check_order(start_brk, <=, brk);
1766 error |= __prctl_check_order(arg_start, <=, arg_end);
1767 error |= __prctl_check_order(env_start, <=, env_end);
1770 #undef __prctl_check_order
1775 * @brk should be after @end_data in traditional maps.
1777 if (prctl_map->start_brk <= prctl_map->end_data ||
1778 prctl_map->brk <= prctl_map->end_data)
1782 * Neither we should allow to override limits if they set.
1784 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1785 prctl_map->start_brk, prctl_map->end_data,
1786 prctl_map->start_data))
1790 * Someone is trying to cheat the auxv vector.
1792 if (prctl_map->auxv_size) {
1793 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1798 * Finally, make sure the caller has the rights to
1799 * change /proc/pid/exe link: only local root should
1802 if (prctl_map->exe_fd != (u32)-1) {
1803 struct user_namespace *ns = current_user_ns();
1804 const struct cred *cred = current_cred();
1806 if (!uid_eq(cred->uid, make_kuid(ns, 0)) ||
1807 !gid_eq(cred->gid, make_kgid(ns, 0)))
1816 #ifdef CONFIG_CHECKPOINT_RESTORE
1817 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1819 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1820 unsigned long user_auxv[AT_VECTOR_SIZE];
1821 struct mm_struct *mm = current->mm;
1824 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1825 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1827 if (opt == PR_SET_MM_MAP_SIZE)
1828 return put_user((unsigned int)sizeof(prctl_map),
1829 (unsigned int __user *)addr);
1831 if (data_size != sizeof(prctl_map))
1834 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1837 error = validate_prctl_map(&prctl_map);
1841 if (prctl_map.auxv_size) {
1842 memset(user_auxv, 0, sizeof(user_auxv));
1843 if (copy_from_user(user_auxv,
1844 (const void __user *)prctl_map.auxv,
1845 prctl_map.auxv_size))
1848 /* Last entry must be AT_NULL as specification requires */
1849 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1850 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
1853 if (prctl_map.exe_fd != (u32)-1) {
1854 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
1859 down_write(&mm->mmap_sem);
1862 * We don't validate if these members are pointing to
1863 * real present VMAs because application may have correspond
1864 * VMAs already unmapped and kernel uses these members for statistics
1865 * output in procfs mostly, except
1867 * - @start_brk/@brk which are used in do_brk but kernel lookups
1868 * for VMAs when updating these memvers so anything wrong written
1869 * here cause kernel to swear at userspace program but won't lead
1870 * to any problem in kernel itself
1873 mm->start_code = prctl_map.start_code;
1874 mm->end_code = prctl_map.end_code;
1875 mm->start_data = prctl_map.start_data;
1876 mm->end_data = prctl_map.end_data;
1877 mm->start_brk = prctl_map.start_brk;
1878 mm->brk = prctl_map.brk;
1879 mm->start_stack = prctl_map.start_stack;
1880 mm->arg_start = prctl_map.arg_start;
1881 mm->arg_end = prctl_map.arg_end;
1882 mm->env_start = prctl_map.env_start;
1883 mm->env_end = prctl_map.env_end;
1886 * Note this update of @saved_auxv is lockless thus
1887 * if someone reads this member in procfs while we're
1888 * updating -- it may get partly updated results. It's
1889 * known and acceptable trade off: we leave it as is to
1890 * not introduce additional locks here making the kernel
1893 if (prctl_map.auxv_size)
1894 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
1896 up_write(&mm->mmap_sem);
1899 #endif /* CONFIG_CHECKPOINT_RESTORE */
1901 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
1905 * This doesn't move the auxiliary vector itself since it's pinned to
1906 * mm_struct, but it permits filling the vector with new values. It's
1907 * up to the caller to provide sane values here, otherwise userspace
1908 * tools which use this vector might be unhappy.
1910 unsigned long user_auxv[AT_VECTOR_SIZE];
1912 if (len > sizeof(user_auxv))
1915 if (copy_from_user(user_auxv, (const void __user *)addr, len))
1918 /* Make sure the last entry is always AT_NULL */
1919 user_auxv[AT_VECTOR_SIZE - 2] = 0;
1920 user_auxv[AT_VECTOR_SIZE - 1] = 0;
1922 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1925 memcpy(mm->saved_auxv, user_auxv, len);
1926 task_unlock(current);
1931 static int prctl_set_mm(int opt, unsigned long addr,
1932 unsigned long arg4, unsigned long arg5)
1934 struct mm_struct *mm = current->mm;
1935 struct prctl_mm_map prctl_map;
1936 struct vm_area_struct *vma;
1939 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
1940 opt != PR_SET_MM_MAP &&
1941 opt != PR_SET_MM_MAP_SIZE)))
1944 #ifdef CONFIG_CHECKPOINT_RESTORE
1945 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
1946 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
1949 if (!capable(CAP_SYS_RESOURCE))
1952 if (opt == PR_SET_MM_EXE_FILE)
1953 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
1955 if (opt == PR_SET_MM_AUXV)
1956 return prctl_set_auxv(mm, addr, arg4);
1958 if (addr >= TASK_SIZE || addr < mmap_min_addr)
1963 down_write(&mm->mmap_sem);
1964 vma = find_vma(mm, addr);
1966 prctl_map.start_code = mm->start_code;
1967 prctl_map.end_code = mm->end_code;
1968 prctl_map.start_data = mm->start_data;
1969 prctl_map.end_data = mm->end_data;
1970 prctl_map.start_brk = mm->start_brk;
1971 prctl_map.brk = mm->brk;
1972 prctl_map.start_stack = mm->start_stack;
1973 prctl_map.arg_start = mm->arg_start;
1974 prctl_map.arg_end = mm->arg_end;
1975 prctl_map.env_start = mm->env_start;
1976 prctl_map.env_end = mm->env_end;
1977 prctl_map.auxv = NULL;
1978 prctl_map.auxv_size = 0;
1979 prctl_map.exe_fd = -1;
1982 case PR_SET_MM_START_CODE:
1983 prctl_map.start_code = addr;
1985 case PR_SET_MM_END_CODE:
1986 prctl_map.end_code = addr;
1988 case PR_SET_MM_START_DATA:
1989 prctl_map.start_data = addr;
1991 case PR_SET_MM_END_DATA:
1992 prctl_map.end_data = addr;
1994 case PR_SET_MM_START_STACK:
1995 prctl_map.start_stack = addr;
1997 case PR_SET_MM_START_BRK:
1998 prctl_map.start_brk = addr;
2001 prctl_map.brk = addr;
2003 case PR_SET_MM_ARG_START:
2004 prctl_map.arg_start = addr;
2006 case PR_SET_MM_ARG_END:
2007 prctl_map.arg_end = addr;
2009 case PR_SET_MM_ENV_START:
2010 prctl_map.env_start = addr;
2012 case PR_SET_MM_ENV_END:
2013 prctl_map.env_end = addr;
2019 error = validate_prctl_map(&prctl_map);
2025 * If command line arguments and environment
2026 * are placed somewhere else on stack, we can
2027 * set them up here, ARG_START/END to setup
2028 * command line argumets and ENV_START/END
2031 case PR_SET_MM_START_STACK:
2032 case PR_SET_MM_ARG_START:
2033 case PR_SET_MM_ARG_END:
2034 case PR_SET_MM_ENV_START:
2035 case PR_SET_MM_ENV_END:
2042 mm->start_code = prctl_map.start_code;
2043 mm->end_code = prctl_map.end_code;
2044 mm->start_data = prctl_map.start_data;
2045 mm->end_data = prctl_map.end_data;
2046 mm->start_brk = prctl_map.start_brk;
2047 mm->brk = prctl_map.brk;
2048 mm->start_stack = prctl_map.start_stack;
2049 mm->arg_start = prctl_map.arg_start;
2050 mm->arg_end = prctl_map.arg_end;
2051 mm->env_start = prctl_map.env_start;
2052 mm->env_end = prctl_map.env_end;
2056 up_write(&mm->mmap_sem);
2060 #ifdef CONFIG_CHECKPOINT_RESTORE
2061 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2063 return put_user(me->clear_child_tid, tid_addr);
2066 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2072 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2075 * If task has has_child_subreaper - all its decendants
2076 * already have these flag too and new decendants will
2077 * inherit it on fork, skip them.
2079 * If we've found child_reaper - skip descendants in
2080 * it's subtree as they will never get out pidns.
2082 if (p->signal->has_child_subreaper ||
2083 is_child_reaper(task_pid(p)))
2086 p->signal->has_child_subreaper = 1;
2090 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2091 unsigned long, arg4, unsigned long, arg5)
2093 struct task_struct *me = current;
2094 unsigned char comm[sizeof(me->comm)];
2097 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2098 if (error != -ENOSYS)
2103 case PR_SET_PDEATHSIG:
2104 if (!valid_signal(arg2)) {
2108 me->pdeath_signal = arg2;
2110 case PR_GET_PDEATHSIG:
2111 error = put_user(me->pdeath_signal, (int __user *)arg2);
2113 case PR_GET_DUMPABLE:
2114 error = get_dumpable(me->mm);
2116 case PR_SET_DUMPABLE:
2117 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2121 set_dumpable(me->mm, arg2);
2124 case PR_SET_UNALIGN:
2125 error = SET_UNALIGN_CTL(me, arg2);
2127 case PR_GET_UNALIGN:
2128 error = GET_UNALIGN_CTL(me, arg2);
2131 error = SET_FPEMU_CTL(me, arg2);
2134 error = GET_FPEMU_CTL(me, arg2);
2137 error = SET_FPEXC_CTL(me, arg2);
2140 error = GET_FPEXC_CTL(me, arg2);
2143 error = PR_TIMING_STATISTICAL;
2146 if (arg2 != PR_TIMING_STATISTICAL)
2150 comm[sizeof(me->comm) - 1] = 0;
2151 if (strncpy_from_user(comm, (char __user *)arg2,
2152 sizeof(me->comm) - 1) < 0)
2154 set_task_comm(me, comm);
2155 proc_comm_connector(me);
2158 get_task_comm(comm, me);
2159 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2163 error = GET_ENDIAN(me, arg2);
2166 error = SET_ENDIAN(me, arg2);
2168 case PR_GET_SECCOMP:
2169 error = prctl_get_seccomp();
2171 case PR_SET_SECCOMP:
2172 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2175 error = GET_TSC_CTL(arg2);
2178 error = SET_TSC_CTL(arg2);
2180 case PR_TASK_PERF_EVENTS_DISABLE:
2181 error = perf_event_task_disable();
2183 case PR_TASK_PERF_EVENTS_ENABLE:
2184 error = perf_event_task_enable();
2186 case PR_GET_TIMERSLACK:
2187 if (current->timer_slack_ns > ULONG_MAX)
2190 error = current->timer_slack_ns;
2192 case PR_SET_TIMERSLACK:
2194 current->timer_slack_ns =
2195 current->default_timer_slack_ns;
2197 current->timer_slack_ns = arg2;
2203 case PR_MCE_KILL_CLEAR:
2206 current->flags &= ~PF_MCE_PROCESS;
2208 case PR_MCE_KILL_SET:
2209 current->flags |= PF_MCE_PROCESS;
2210 if (arg3 == PR_MCE_KILL_EARLY)
2211 current->flags |= PF_MCE_EARLY;
2212 else if (arg3 == PR_MCE_KILL_LATE)
2213 current->flags &= ~PF_MCE_EARLY;
2214 else if (arg3 == PR_MCE_KILL_DEFAULT)
2216 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2224 case PR_MCE_KILL_GET:
2225 if (arg2 | arg3 | arg4 | arg5)
2227 if (current->flags & PF_MCE_PROCESS)
2228 error = (current->flags & PF_MCE_EARLY) ?
2229 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2231 error = PR_MCE_KILL_DEFAULT;
2234 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2236 case PR_GET_TID_ADDRESS:
2237 error = prctl_get_tid_address(me, (int __user **)arg2);
2239 case PR_SET_CHILD_SUBREAPER:
2240 me->signal->is_child_subreaper = !!arg2;
2244 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2246 case PR_GET_CHILD_SUBREAPER:
2247 error = put_user(me->signal->is_child_subreaper,
2248 (int __user *)arg2);
2250 case PR_SET_NO_NEW_PRIVS:
2251 if (arg2 != 1 || arg3 || arg4 || arg5)
2254 task_set_no_new_privs(current);
2256 case PR_GET_NO_NEW_PRIVS:
2257 if (arg2 || arg3 || arg4 || arg5)
2259 return task_no_new_privs(current) ? 1 : 0;
2260 case PR_GET_THP_DISABLE:
2261 if (arg2 || arg3 || arg4 || arg5)
2263 error = !!(me->mm->def_flags & VM_NOHUGEPAGE);
2265 case PR_SET_THP_DISABLE:
2266 if (arg3 || arg4 || arg5)
2268 if (down_write_killable(&me->mm->mmap_sem))
2271 me->mm->def_flags |= VM_NOHUGEPAGE;
2273 me->mm->def_flags &= ~VM_NOHUGEPAGE;
2274 up_write(&me->mm->mmap_sem);
2276 case PR_MPX_ENABLE_MANAGEMENT:
2277 if (arg2 || arg3 || arg4 || arg5)
2279 error = MPX_ENABLE_MANAGEMENT();
2281 case PR_MPX_DISABLE_MANAGEMENT:
2282 if (arg2 || arg3 || arg4 || arg5)
2284 error = MPX_DISABLE_MANAGEMENT();
2286 case PR_SET_FP_MODE:
2287 error = SET_FP_MODE(me, arg2);
2289 case PR_GET_FP_MODE:
2290 error = GET_FP_MODE(me);
2299 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2300 struct getcpu_cache __user *, unused)
2303 int cpu = raw_smp_processor_id();
2306 err |= put_user(cpu, cpup);
2308 err |= put_user(cpu_to_node(cpu), nodep);
2309 return err ? -EFAULT : 0;
2313 * do_sysinfo - fill in sysinfo struct
2314 * @info: pointer to buffer to fill
2316 static int do_sysinfo(struct sysinfo *info)
2318 unsigned long mem_total, sav_total;
2319 unsigned int mem_unit, bitcount;
2322 memset(info, 0, sizeof(struct sysinfo));
2324 get_monotonic_boottime(&tp);
2325 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2327 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2329 info->procs = nr_threads;
2335 * If the sum of all the available memory (i.e. ram + swap)
2336 * is less than can be stored in a 32 bit unsigned long then
2337 * we can be binary compatible with 2.2.x kernels. If not,
2338 * well, in that case 2.2.x was broken anyways...
2340 * -Erik Andersen <andersee@debian.org>
2343 mem_total = info->totalram + info->totalswap;
2344 if (mem_total < info->totalram || mem_total < info->totalswap)
2347 mem_unit = info->mem_unit;
2348 while (mem_unit > 1) {
2351 sav_total = mem_total;
2353 if (mem_total < sav_total)
2358 * If mem_total did not overflow, multiply all memory values by
2359 * info->mem_unit and set it to 1. This leaves things compatible
2360 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2365 info->totalram <<= bitcount;
2366 info->freeram <<= bitcount;
2367 info->sharedram <<= bitcount;
2368 info->bufferram <<= bitcount;
2369 info->totalswap <<= bitcount;
2370 info->freeswap <<= bitcount;
2371 info->totalhigh <<= bitcount;
2372 info->freehigh <<= bitcount;
2378 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2384 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2390 #ifdef CONFIG_COMPAT
2391 struct compat_sysinfo {
2405 char _f[20-2*sizeof(u32)-sizeof(int)];
2408 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2414 /* Check to see if any memory value is too large for 32-bit and scale
2417 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2420 while (s.mem_unit < PAGE_SIZE) {
2425 s.totalram >>= bitcount;
2426 s.freeram >>= bitcount;
2427 s.sharedram >>= bitcount;
2428 s.bufferram >>= bitcount;
2429 s.totalswap >>= bitcount;
2430 s.freeswap >>= bitcount;
2431 s.totalhigh >>= bitcount;
2432 s.freehigh >>= bitcount;
2435 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2436 __put_user(s.uptime, &info->uptime) ||
2437 __put_user(s.loads[0], &info->loads[0]) ||
2438 __put_user(s.loads[1], &info->loads[1]) ||
2439 __put_user(s.loads[2], &info->loads[2]) ||
2440 __put_user(s.totalram, &info->totalram) ||
2441 __put_user(s.freeram, &info->freeram) ||
2442 __put_user(s.sharedram, &info->sharedram) ||
2443 __put_user(s.bufferram, &info->bufferram) ||
2444 __put_user(s.totalswap, &info->totalswap) ||
2445 __put_user(s.freeswap, &info->freeswap) ||
2446 __put_user(s.procs, &info->procs) ||
2447 __put_user(s.totalhigh, &info->totalhigh) ||
2448 __put_user(s.freehigh, &info->freehigh) ||
2449 __put_user(s.mem_unit, &info->mem_unit))
2454 #endif /* CONFIG_COMPAT */