1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/export.h>
10 #include <linux/utsname.h>
11 #include <linux/mman.h>
12 #include <linux/reboot.h>
13 #include <linux/prctl.h>
14 #include <linux/highuid.h>
16 #include <linux/kmod.h>
17 #include <linux/perf_event.h>
18 #include <linux/resource.h>
19 #include <linux/kernel.h>
20 #include <linux/workqueue.h>
21 #include <linux/capability.h>
22 #include <linux/device.h>
23 #include <linux/key.h>
24 #include <linux/times.h>
25 #include <linux/posix-timers.h>
26 #include <linux/security.h>
27 #include <linux/dcookies.h>
28 #include <linux/suspend.h>
29 #include <linux/tty.h>
30 #include <linux/signal.h>
31 #include <linux/cn_proc.h>
32 #include <linux/getcpu.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/seccomp.h>
35 #include <linux/cpu.h>
36 #include <linux/personality.h>
37 #include <linux/ptrace.h>
38 #include <linux/fs_struct.h>
39 #include <linux/file.h>
40 #include <linux/mount.h>
41 #include <linux/gfp.h>
42 #include <linux/syscore_ops.h>
43 #include <linux/version.h>
44 #include <linux/ctype.h>
46 #include <linux/compat.h>
47 #include <linux/syscalls.h>
48 #include <linux/kprobes.h>
49 #include <linux/user_namespace.h>
50 #include <linux/binfmts.h>
52 #include <linux/sched.h>
53 #include <linux/sched/autogroup.h>
54 #include <linux/sched/loadavg.h>
55 #include <linux/sched/stat.h>
56 #include <linux/sched/mm.h>
57 #include <linux/sched/coredump.h>
58 #include <linux/sched/task.h>
59 #include <linux/sched/cputime.h>
60 #include <linux/rcupdate.h>
61 #include <linux/uidgid.h>
62 #include <linux/cred.h>
64 #include <linux/nospec.h>
66 #include <linux/kmsg_dump.h>
67 /* Move somewhere else to avoid recompiling? */
68 #include <generated/utsrelease.h>
70 #include <linux/uaccess.h>
72 #include <asm/unistd.h>
76 #ifndef SET_UNALIGN_CTL
77 # define SET_UNALIGN_CTL(a, b) (-EINVAL)
79 #ifndef GET_UNALIGN_CTL
80 # define GET_UNALIGN_CTL(a, b) (-EINVAL)
83 # define SET_FPEMU_CTL(a, b) (-EINVAL)
86 # define GET_FPEMU_CTL(a, b) (-EINVAL)
89 # define SET_FPEXC_CTL(a, b) (-EINVAL)
92 # define GET_FPEXC_CTL(a, b) (-EINVAL)
95 # define GET_ENDIAN(a, b) (-EINVAL)
98 # define SET_ENDIAN(a, b) (-EINVAL)
101 # define GET_TSC_CTL(a) (-EINVAL)
104 # define SET_TSC_CTL(a) (-EINVAL)
106 #ifndef MPX_ENABLE_MANAGEMENT
107 # define MPX_ENABLE_MANAGEMENT() (-EINVAL)
109 #ifndef MPX_DISABLE_MANAGEMENT
110 # define MPX_DISABLE_MANAGEMENT() (-EINVAL)
113 # define GET_FP_MODE(a) (-EINVAL)
116 # define SET_FP_MODE(a,b) (-EINVAL)
119 # define SVE_SET_VL(a) (-EINVAL)
122 # define SVE_GET_VL() (-EINVAL)
126 * this is where the system-wide overflow UID and GID are defined, for
127 * architectures that now have 32-bit UID/GID but didn't in the past
130 int overflowuid = DEFAULT_OVERFLOWUID;
131 int overflowgid = DEFAULT_OVERFLOWGID;
133 EXPORT_SYMBOL(overflowuid);
134 EXPORT_SYMBOL(overflowgid);
137 * the same as above, but for filesystems which can only store a 16-bit
138 * UID and GID. as such, this is needed on all architectures
141 int fs_overflowuid = DEFAULT_FS_OVERFLOWUID;
142 int fs_overflowgid = DEFAULT_FS_OVERFLOWGID;
144 EXPORT_SYMBOL(fs_overflowuid);
145 EXPORT_SYMBOL(fs_overflowgid);
148 * Returns true if current's euid is same as p's uid or euid,
149 * or has CAP_SYS_NICE to p's user_ns.
151 * Called with rcu_read_lock, creds are safe
153 static bool set_one_prio_perm(struct task_struct *p)
155 const struct cred *cred = current_cred(), *pcred = __task_cred(p);
157 if (uid_eq(pcred->uid, cred->euid) ||
158 uid_eq(pcred->euid, cred->euid))
160 if (ns_capable(pcred->user_ns, CAP_SYS_NICE))
166 * set the priority of a task
167 * - the caller must hold the RCU read lock
169 static int set_one_prio(struct task_struct *p, int niceval, int error)
173 if (!set_one_prio_perm(p)) {
177 if (niceval < task_nice(p) && !can_nice(p, niceval)) {
181 no_nice = security_task_setnice(p, niceval);
188 set_user_nice(p, niceval);
193 SYSCALL_DEFINE3(setpriority, int, which, int, who, int, niceval)
195 struct task_struct *g, *p;
196 struct user_struct *user;
197 const struct cred *cred = current_cred();
202 if (which > PRIO_USER || which < PRIO_PROCESS)
205 /* normalize: avoid signed division (rounding problems) */
207 if (niceval < MIN_NICE)
209 if (niceval > MAX_NICE)
213 read_lock(&tasklist_lock);
217 p = find_task_by_vpid(who);
221 error = set_one_prio(p, niceval, error);
225 pgrp = find_vpid(who);
227 pgrp = task_pgrp(current);
228 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
229 error = set_one_prio(p, niceval, error);
230 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
233 uid = make_kuid(cred->user_ns, who);
237 else if (!uid_eq(uid, cred->uid)) {
238 user = find_user(uid);
240 goto out_unlock; /* No processes for this user */
242 do_each_thread(g, p) {
243 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p))
244 error = set_one_prio(p, niceval, error);
245 } while_each_thread(g, p);
246 if (!uid_eq(uid, cred->uid))
247 free_uid(user); /* For find_user() */
251 read_unlock(&tasklist_lock);
258 * Ugh. To avoid negative return values, "getpriority()" will
259 * not return the normal nice-value, but a negated value that
260 * has been offset by 20 (ie it returns 40..1 instead of -20..19)
261 * to stay compatible.
263 SYSCALL_DEFINE2(getpriority, int, which, int, who)
265 struct task_struct *g, *p;
266 struct user_struct *user;
267 const struct cred *cred = current_cred();
268 long niceval, retval = -ESRCH;
272 if (which > PRIO_USER || which < PRIO_PROCESS)
276 read_lock(&tasklist_lock);
280 p = find_task_by_vpid(who);
284 niceval = nice_to_rlimit(task_nice(p));
285 if (niceval > retval)
291 pgrp = find_vpid(who);
293 pgrp = task_pgrp(current);
294 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
295 niceval = nice_to_rlimit(task_nice(p));
296 if (niceval > retval)
298 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
301 uid = make_kuid(cred->user_ns, who);
305 else if (!uid_eq(uid, cred->uid)) {
306 user = find_user(uid);
308 goto out_unlock; /* No processes for this user */
310 do_each_thread(g, p) {
311 if (uid_eq(task_uid(p), uid) && task_pid_vnr(p)) {
312 niceval = nice_to_rlimit(task_nice(p));
313 if (niceval > retval)
316 } while_each_thread(g, p);
317 if (!uid_eq(uid, cred->uid))
318 free_uid(user); /* for find_user() */
322 read_unlock(&tasklist_lock);
329 * Unprivileged users may change the real gid to the effective gid
330 * or vice versa. (BSD-style)
332 * If you set the real gid at all, or set the effective gid to a value not
333 * equal to the real gid, then the saved gid is set to the new effective gid.
335 * This makes it possible for a setgid program to completely drop its
336 * privileges, which is often a useful assertion to make when you are doing
337 * a security audit over a program.
339 * The general idea is that a program which uses just setregid() will be
340 * 100% compatible with BSD. A program which uses just setgid() will be
341 * 100% compatible with POSIX with saved IDs.
343 * SMP: There are not races, the GIDs are checked only by filesystem
344 * operations (as far as semantic preservation is concerned).
346 #ifdef CONFIG_MULTIUSER
347 long __sys_setregid(gid_t rgid, gid_t egid)
349 struct user_namespace *ns = current_user_ns();
350 const struct cred *old;
355 krgid = make_kgid(ns, rgid);
356 kegid = make_kgid(ns, egid);
358 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
360 if ((egid != (gid_t) -1) && !gid_valid(kegid))
363 new = prepare_creds();
366 old = current_cred();
369 if (rgid != (gid_t) -1) {
370 if (gid_eq(old->gid, krgid) ||
371 gid_eq(old->egid, krgid) ||
372 ns_capable(old->user_ns, CAP_SETGID))
377 if (egid != (gid_t) -1) {
378 if (gid_eq(old->gid, kegid) ||
379 gid_eq(old->egid, kegid) ||
380 gid_eq(old->sgid, kegid) ||
381 ns_capable(old->user_ns, CAP_SETGID))
387 if (rgid != (gid_t) -1 ||
388 (egid != (gid_t) -1 && !gid_eq(kegid, old->gid)))
389 new->sgid = new->egid;
390 new->fsgid = new->egid;
392 return commit_creds(new);
399 SYSCALL_DEFINE2(setregid, gid_t, rgid, gid_t, egid)
401 return __sys_setregid(rgid, egid);
405 * setgid() is implemented like SysV w/ SAVED_IDS
407 * SMP: Same implicit races as above.
409 long __sys_setgid(gid_t gid)
411 struct user_namespace *ns = current_user_ns();
412 const struct cred *old;
417 kgid = make_kgid(ns, gid);
418 if (!gid_valid(kgid))
421 new = prepare_creds();
424 old = current_cred();
427 if (ns_capable(old->user_ns, CAP_SETGID))
428 new->gid = new->egid = new->sgid = new->fsgid = kgid;
429 else if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->sgid))
430 new->egid = new->fsgid = kgid;
434 return commit_creds(new);
441 SYSCALL_DEFINE1(setgid, gid_t, gid)
443 return __sys_setgid(gid);
447 * change the user struct in a credentials set to match the new UID
449 static int set_user(struct cred *new)
451 struct user_struct *new_user;
453 new_user = alloc_uid(new->uid);
458 * We don't fail in case of NPROC limit excess here because too many
459 * poorly written programs don't check set*uid() return code, assuming
460 * it never fails if called by root. We may still enforce NPROC limit
461 * for programs doing set*uid()+execve() by harmlessly deferring the
462 * failure to the execve() stage.
464 if (atomic_read(&new_user->processes) >= rlimit(RLIMIT_NPROC) &&
465 new_user != INIT_USER)
466 current->flags |= PF_NPROC_EXCEEDED;
468 current->flags &= ~PF_NPROC_EXCEEDED;
471 new->user = new_user;
476 * Unprivileged users may change the real uid to the effective uid
477 * or vice versa. (BSD-style)
479 * If you set the real uid at all, or set the effective uid to a value not
480 * equal to the real uid, then the saved uid is set to the new effective uid.
482 * This makes it possible for a setuid program to completely drop its
483 * privileges, which is often a useful assertion to make when you are doing
484 * a security audit over a program.
486 * The general idea is that a program which uses just setreuid() will be
487 * 100% compatible with BSD. A program which uses just setuid() will be
488 * 100% compatible with POSIX with saved IDs.
490 long __sys_setreuid(uid_t ruid, uid_t euid)
492 struct user_namespace *ns = current_user_ns();
493 const struct cred *old;
498 kruid = make_kuid(ns, ruid);
499 keuid = make_kuid(ns, euid);
501 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
503 if ((euid != (uid_t) -1) && !uid_valid(keuid))
506 new = prepare_creds();
509 old = current_cred();
512 if (ruid != (uid_t) -1) {
514 if (!uid_eq(old->uid, kruid) &&
515 !uid_eq(old->euid, kruid) &&
516 !ns_capable(old->user_ns, CAP_SETUID))
520 if (euid != (uid_t) -1) {
522 if (!uid_eq(old->uid, keuid) &&
523 !uid_eq(old->euid, keuid) &&
524 !uid_eq(old->suid, keuid) &&
525 !ns_capable(old->user_ns, CAP_SETUID))
529 if (!uid_eq(new->uid, old->uid)) {
530 retval = set_user(new);
534 if (ruid != (uid_t) -1 ||
535 (euid != (uid_t) -1 && !uid_eq(keuid, old->uid)))
536 new->suid = new->euid;
537 new->fsuid = new->euid;
539 retval = security_task_fix_setuid(new, old, LSM_SETID_RE);
543 return commit_creds(new);
550 SYSCALL_DEFINE2(setreuid, uid_t, ruid, uid_t, euid)
552 return __sys_setreuid(ruid, euid);
556 * setuid() is implemented like SysV with SAVED_IDS
558 * Note that SAVED_ID's is deficient in that a setuid root program
559 * like sendmail, for example, cannot set its uid to be a normal
560 * user and then switch back, because if you're root, setuid() sets
561 * the saved uid too. If you don't like this, blame the bright people
562 * in the POSIX committee and/or USG. Note that the BSD-style setreuid()
563 * will allow a root program to temporarily drop privileges and be able to
564 * regain them by swapping the real and effective uid.
566 long __sys_setuid(uid_t uid)
568 struct user_namespace *ns = current_user_ns();
569 const struct cred *old;
574 kuid = make_kuid(ns, uid);
575 if (!uid_valid(kuid))
578 new = prepare_creds();
581 old = current_cred();
584 if (ns_capable(old->user_ns, CAP_SETUID)) {
585 new->suid = new->uid = kuid;
586 if (!uid_eq(kuid, old->uid)) {
587 retval = set_user(new);
591 } else if (!uid_eq(kuid, old->uid) && !uid_eq(kuid, new->suid)) {
595 new->fsuid = new->euid = kuid;
597 retval = security_task_fix_setuid(new, old, LSM_SETID_ID);
601 return commit_creds(new);
608 SYSCALL_DEFINE1(setuid, uid_t, uid)
610 return __sys_setuid(uid);
615 * This function implements a generic ability to update ruid, euid,
616 * and suid. This allows you to implement the 4.4 compatible seteuid().
618 long __sys_setresuid(uid_t ruid, uid_t euid, uid_t suid)
620 struct user_namespace *ns = current_user_ns();
621 const struct cred *old;
624 kuid_t kruid, keuid, ksuid;
626 kruid = make_kuid(ns, ruid);
627 keuid = make_kuid(ns, euid);
628 ksuid = make_kuid(ns, suid);
630 if ((ruid != (uid_t) -1) && !uid_valid(kruid))
633 if ((euid != (uid_t) -1) && !uid_valid(keuid))
636 if ((suid != (uid_t) -1) && !uid_valid(ksuid))
639 new = prepare_creds();
643 old = current_cred();
646 if (!ns_capable(old->user_ns, CAP_SETUID)) {
647 if (ruid != (uid_t) -1 && !uid_eq(kruid, old->uid) &&
648 !uid_eq(kruid, old->euid) && !uid_eq(kruid, old->suid))
650 if (euid != (uid_t) -1 && !uid_eq(keuid, old->uid) &&
651 !uid_eq(keuid, old->euid) && !uid_eq(keuid, old->suid))
653 if (suid != (uid_t) -1 && !uid_eq(ksuid, old->uid) &&
654 !uid_eq(ksuid, old->euid) && !uid_eq(ksuid, old->suid))
658 if (ruid != (uid_t) -1) {
660 if (!uid_eq(kruid, old->uid)) {
661 retval = set_user(new);
666 if (euid != (uid_t) -1)
668 if (suid != (uid_t) -1)
670 new->fsuid = new->euid;
672 retval = security_task_fix_setuid(new, old, LSM_SETID_RES);
676 return commit_creds(new);
683 SYSCALL_DEFINE3(setresuid, uid_t, ruid, uid_t, euid, uid_t, suid)
685 return __sys_setresuid(ruid, euid, suid);
688 SYSCALL_DEFINE3(getresuid, uid_t __user *, ruidp, uid_t __user *, euidp, uid_t __user *, suidp)
690 const struct cred *cred = current_cred();
692 uid_t ruid, euid, suid;
694 ruid = from_kuid_munged(cred->user_ns, cred->uid);
695 euid = from_kuid_munged(cred->user_ns, cred->euid);
696 suid = from_kuid_munged(cred->user_ns, cred->suid);
698 retval = put_user(ruid, ruidp);
700 retval = put_user(euid, euidp);
702 return put_user(suid, suidp);
708 * Same as above, but for rgid, egid, sgid.
710 long __sys_setresgid(gid_t rgid, gid_t egid, gid_t sgid)
712 struct user_namespace *ns = current_user_ns();
713 const struct cred *old;
716 kgid_t krgid, kegid, ksgid;
718 krgid = make_kgid(ns, rgid);
719 kegid = make_kgid(ns, egid);
720 ksgid = make_kgid(ns, sgid);
722 if ((rgid != (gid_t) -1) && !gid_valid(krgid))
724 if ((egid != (gid_t) -1) && !gid_valid(kegid))
726 if ((sgid != (gid_t) -1) && !gid_valid(ksgid))
729 new = prepare_creds();
732 old = current_cred();
735 if (!ns_capable(old->user_ns, CAP_SETGID)) {
736 if (rgid != (gid_t) -1 && !gid_eq(krgid, old->gid) &&
737 !gid_eq(krgid, old->egid) && !gid_eq(krgid, old->sgid))
739 if (egid != (gid_t) -1 && !gid_eq(kegid, old->gid) &&
740 !gid_eq(kegid, old->egid) && !gid_eq(kegid, old->sgid))
742 if (sgid != (gid_t) -1 && !gid_eq(ksgid, old->gid) &&
743 !gid_eq(ksgid, old->egid) && !gid_eq(ksgid, old->sgid))
747 if (rgid != (gid_t) -1)
749 if (egid != (gid_t) -1)
751 if (sgid != (gid_t) -1)
753 new->fsgid = new->egid;
755 return commit_creds(new);
762 SYSCALL_DEFINE3(setresgid, gid_t, rgid, gid_t, egid, gid_t, sgid)
764 return __sys_setresgid(rgid, egid, sgid);
767 SYSCALL_DEFINE3(getresgid, gid_t __user *, rgidp, gid_t __user *, egidp, gid_t __user *, sgidp)
769 const struct cred *cred = current_cred();
771 gid_t rgid, egid, sgid;
773 rgid = from_kgid_munged(cred->user_ns, cred->gid);
774 egid = from_kgid_munged(cred->user_ns, cred->egid);
775 sgid = from_kgid_munged(cred->user_ns, cred->sgid);
777 retval = put_user(rgid, rgidp);
779 retval = put_user(egid, egidp);
781 retval = put_user(sgid, sgidp);
789 * "setfsuid()" sets the fsuid - the uid used for filesystem checks. This
790 * is used for "access()" and for the NFS daemon (letting nfsd stay at
791 * whatever uid it wants to). It normally shadows "euid", except when
792 * explicitly set by setfsuid() or for access..
794 long __sys_setfsuid(uid_t uid)
796 const struct cred *old;
801 old = current_cred();
802 old_fsuid = from_kuid_munged(old->user_ns, old->fsuid);
804 kuid = make_kuid(old->user_ns, uid);
805 if (!uid_valid(kuid))
808 new = prepare_creds();
812 if (uid_eq(kuid, old->uid) || uid_eq(kuid, old->euid) ||
813 uid_eq(kuid, old->suid) || uid_eq(kuid, old->fsuid) ||
814 ns_capable(old->user_ns, CAP_SETUID)) {
815 if (!uid_eq(kuid, old->fsuid)) {
817 if (security_task_fix_setuid(new, old, LSM_SETID_FS) == 0)
830 SYSCALL_DEFINE1(setfsuid, uid_t, uid)
832 return __sys_setfsuid(uid);
836 * Samma på svenska..
838 long __sys_setfsgid(gid_t gid)
840 const struct cred *old;
845 old = current_cred();
846 old_fsgid = from_kgid_munged(old->user_ns, old->fsgid);
848 kgid = make_kgid(old->user_ns, gid);
849 if (!gid_valid(kgid))
852 new = prepare_creds();
856 if (gid_eq(kgid, old->gid) || gid_eq(kgid, old->egid) ||
857 gid_eq(kgid, old->sgid) || gid_eq(kgid, old->fsgid) ||
858 ns_capable(old->user_ns, CAP_SETGID)) {
859 if (!gid_eq(kgid, old->fsgid)) {
873 SYSCALL_DEFINE1(setfsgid, gid_t, gid)
875 return __sys_setfsgid(gid);
877 #endif /* CONFIG_MULTIUSER */
880 * sys_getpid - return the thread group id of the current process
882 * Note, despite the name, this returns the tgid not the pid. The tgid and
883 * the pid are identical unless CLONE_THREAD was specified on clone() in
884 * which case the tgid is the same in all threads of the same group.
886 * This is SMP safe as current->tgid does not change.
888 SYSCALL_DEFINE0(getpid)
890 return task_tgid_vnr(current);
893 /* Thread ID - the internal kernel "pid" */
894 SYSCALL_DEFINE0(gettid)
896 return task_pid_vnr(current);
900 * Accessing ->real_parent is not SMP-safe, it could
901 * change from under us. However, we can use a stale
902 * value of ->real_parent under rcu_read_lock(), see
903 * release_task()->call_rcu(delayed_put_task_struct).
905 SYSCALL_DEFINE0(getppid)
910 pid = task_tgid_vnr(rcu_dereference(current->real_parent));
916 SYSCALL_DEFINE0(getuid)
918 /* Only we change this so SMP safe */
919 return from_kuid_munged(current_user_ns(), current_uid());
922 SYSCALL_DEFINE0(geteuid)
924 /* Only we change this so SMP safe */
925 return from_kuid_munged(current_user_ns(), current_euid());
928 SYSCALL_DEFINE0(getgid)
930 /* Only we change this so SMP safe */
931 return from_kgid_munged(current_user_ns(), current_gid());
934 SYSCALL_DEFINE0(getegid)
936 /* Only we change this so SMP safe */
937 return from_kgid_munged(current_user_ns(), current_egid());
940 static void do_sys_times(struct tms *tms)
942 u64 tgutime, tgstime, cutime, cstime;
944 thread_group_cputime_adjusted(current, &tgutime, &tgstime);
945 cutime = current->signal->cutime;
946 cstime = current->signal->cstime;
947 tms->tms_utime = nsec_to_clock_t(tgutime);
948 tms->tms_stime = nsec_to_clock_t(tgstime);
949 tms->tms_cutime = nsec_to_clock_t(cutime);
950 tms->tms_cstime = nsec_to_clock_t(cstime);
953 SYSCALL_DEFINE1(times, struct tms __user *, tbuf)
959 if (copy_to_user(tbuf, &tmp, sizeof(struct tms)))
962 force_successful_syscall_return();
963 return (long) jiffies_64_to_clock_t(get_jiffies_64());
967 static compat_clock_t clock_t_to_compat_clock_t(clock_t x)
969 return compat_jiffies_to_clock_t(clock_t_to_jiffies(x));
972 COMPAT_SYSCALL_DEFINE1(times, struct compat_tms __user *, tbuf)
976 struct compat_tms tmp;
979 /* Convert our struct tms to the compat version. */
980 tmp.tms_utime = clock_t_to_compat_clock_t(tms.tms_utime);
981 tmp.tms_stime = clock_t_to_compat_clock_t(tms.tms_stime);
982 tmp.tms_cutime = clock_t_to_compat_clock_t(tms.tms_cutime);
983 tmp.tms_cstime = clock_t_to_compat_clock_t(tms.tms_cstime);
984 if (copy_to_user(tbuf, &tmp, sizeof(tmp)))
987 force_successful_syscall_return();
988 return compat_jiffies_to_clock_t(jiffies);
993 * This needs some heavy checking ...
994 * I just haven't the stomach for it. I also don't fully
995 * understand sessions/pgrp etc. Let somebody who does explain it.
997 * OK, I think I have the protection semantics right.... this is really
998 * only important on a multi-user system anyway, to make sure one user
999 * can't send a signal to a process owned by another. -TYT, 12/12/91
1001 * !PF_FORKNOEXEC check to conform completely to POSIX.
1003 SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1005 struct task_struct *p;
1006 struct task_struct *group_leader = current->group_leader;
1011 pid = task_pid_vnr(group_leader);
1018 /* From this point forward we keep holding onto the tasklist lock
1019 * so that our parent does not change from under us. -DaveM
1021 write_lock_irq(&tasklist_lock);
1024 p = find_task_by_vpid(pid);
1029 if (!thread_group_leader(p))
1032 if (same_thread_group(p->real_parent, group_leader)) {
1034 if (task_session(p) != task_session(group_leader))
1037 if (!(p->flags & PF_FORKNOEXEC))
1041 if (p != group_leader)
1046 if (p->signal->leader)
1051 struct task_struct *g;
1053 pgrp = find_vpid(pgid);
1054 g = pid_task(pgrp, PIDTYPE_PGID);
1055 if (!g || task_session(g) != task_session(group_leader))
1059 err = security_task_setpgid(p, pgid);
1063 if (task_pgrp(p) != pgrp)
1064 change_pid(p, PIDTYPE_PGID, pgrp);
1068 /* All paths lead to here, thus we are safe. -DaveM */
1069 write_unlock_irq(&tasklist_lock);
1074 static int do_getpgid(pid_t pid)
1076 struct task_struct *p;
1082 grp = task_pgrp(current);
1085 p = find_task_by_vpid(pid);
1092 retval = security_task_getpgid(p);
1096 retval = pid_vnr(grp);
1102 SYSCALL_DEFINE1(getpgid, pid_t, pid)
1104 return do_getpgid(pid);
1107 #ifdef __ARCH_WANT_SYS_GETPGRP
1109 SYSCALL_DEFINE0(getpgrp)
1111 return do_getpgid(0);
1116 SYSCALL_DEFINE1(getsid, pid_t, pid)
1118 struct task_struct *p;
1124 sid = task_session(current);
1127 p = find_task_by_vpid(pid);
1130 sid = task_session(p);
1134 retval = security_task_getsid(p);
1138 retval = pid_vnr(sid);
1144 static void set_special_pids(struct pid *pid)
1146 struct task_struct *curr = current->group_leader;
1148 if (task_session(curr) != pid)
1149 change_pid(curr, PIDTYPE_SID, pid);
1151 if (task_pgrp(curr) != pid)
1152 change_pid(curr, PIDTYPE_PGID, pid);
1155 int ksys_setsid(void)
1157 struct task_struct *group_leader = current->group_leader;
1158 struct pid *sid = task_pid(group_leader);
1159 pid_t session = pid_vnr(sid);
1162 write_lock_irq(&tasklist_lock);
1163 /* Fail if I am already a session leader */
1164 if (group_leader->signal->leader)
1167 /* Fail if a process group id already exists that equals the
1168 * proposed session id.
1170 if (pid_task(sid, PIDTYPE_PGID))
1173 group_leader->signal->leader = 1;
1174 set_special_pids(sid);
1176 proc_clear_tty(group_leader);
1180 write_unlock_irq(&tasklist_lock);
1182 proc_sid_connector(group_leader);
1183 sched_autogroup_create_attach(group_leader);
1188 SYSCALL_DEFINE0(setsid)
1190 return ksys_setsid();
1193 DECLARE_RWSEM(uts_sem);
1195 #ifdef COMPAT_UTS_MACHINE
1196 #define override_architecture(name) \
1197 (personality(current->personality) == PER_LINUX32 && \
1198 copy_to_user(name->machine, COMPAT_UTS_MACHINE, \
1199 sizeof(COMPAT_UTS_MACHINE)))
1201 #define override_architecture(name) 0
1205 * Work around broken programs that cannot handle "Linux 3.0".
1206 * Instead we map 3.x to 2.6.40+x, so e.g. 3.0 would be 2.6.40
1207 * And we map 4.x to 2.6.60+x, so 4.0 would be 2.6.60.
1209 static int override_release(char __user *release, size_t len)
1213 if (current->personality & UNAME26) {
1214 const char *rest = UTS_RELEASE;
1215 char buf[65] = { 0 };
1221 if (*rest == '.' && ++ndots >= 3)
1223 if (!isdigit(*rest) && *rest != '.')
1227 v = ((LINUX_VERSION_CODE >> 8) & 0xff) + 60;
1228 copy = clamp_t(size_t, len, 1, sizeof(buf));
1229 copy = scnprintf(buf, copy, "2.6.%u%s", v, rest);
1230 ret = copy_to_user(release, buf, copy + 1);
1235 SYSCALL_DEFINE1(newuname, struct new_utsname __user *, name)
1237 struct new_utsname tmp;
1239 down_read(&uts_sem);
1240 memcpy(&tmp, utsname(), sizeof(tmp));
1242 if (copy_to_user(name, &tmp, sizeof(tmp)))
1245 if (override_release(name->release, sizeof(name->release)))
1247 if (override_architecture(name))
1252 #ifdef __ARCH_WANT_SYS_OLD_UNAME
1256 SYSCALL_DEFINE1(uname, struct old_utsname __user *, name)
1258 struct old_utsname tmp;
1263 down_read(&uts_sem);
1264 memcpy(&tmp, utsname(), sizeof(tmp));
1266 if (copy_to_user(name, &tmp, sizeof(tmp)))
1269 if (override_release(name->release, sizeof(name->release)))
1271 if (override_architecture(name))
1276 SYSCALL_DEFINE1(olduname, struct oldold_utsname __user *, name)
1278 struct oldold_utsname tmp;
1283 memset(&tmp, 0, sizeof(tmp));
1285 down_read(&uts_sem);
1286 memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
1287 memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
1288 memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
1289 memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
1290 memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
1292 if (copy_to_user(name, &tmp, sizeof(tmp)))
1295 if (override_architecture(name))
1297 if (override_release(name->release, sizeof(name->release)))
1303 SYSCALL_DEFINE2(sethostname, char __user *, name, int, len)
1306 char tmp[__NEW_UTS_LEN];
1308 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1311 if (len < 0 || len > __NEW_UTS_LEN)
1314 if (!copy_from_user(tmp, name, len)) {
1315 struct new_utsname *u;
1317 down_write(&uts_sem);
1319 memcpy(u->nodename, tmp, len);
1320 memset(u->nodename + len, 0, sizeof(u->nodename) - len);
1322 uts_proc_notify(UTS_PROC_HOSTNAME);
1328 #ifdef __ARCH_WANT_SYS_GETHOSTNAME
1330 SYSCALL_DEFINE2(gethostname, char __user *, name, int, len)
1333 struct new_utsname *u;
1334 char tmp[__NEW_UTS_LEN + 1];
1338 down_read(&uts_sem);
1340 i = 1 + strlen(u->nodename);
1343 memcpy(tmp, u->nodename, i);
1345 if (copy_to_user(name, tmp, i))
1353 * Only setdomainname; getdomainname can be implemented by calling
1356 SYSCALL_DEFINE2(setdomainname, char __user *, name, int, len)
1359 char tmp[__NEW_UTS_LEN];
1361 if (!ns_capable(current->nsproxy->uts_ns->user_ns, CAP_SYS_ADMIN))
1363 if (len < 0 || len > __NEW_UTS_LEN)
1367 if (!copy_from_user(tmp, name, len)) {
1368 struct new_utsname *u;
1370 down_write(&uts_sem);
1372 memcpy(u->domainname, tmp, len);
1373 memset(u->domainname + len, 0, sizeof(u->domainname) - len);
1375 uts_proc_notify(UTS_PROC_DOMAINNAME);
1381 SYSCALL_DEFINE2(getrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1383 struct rlimit value;
1386 ret = do_prlimit(current, resource, NULL, &value);
1388 ret = copy_to_user(rlim, &value, sizeof(*rlim)) ? -EFAULT : 0;
1393 #ifdef CONFIG_COMPAT
1395 COMPAT_SYSCALL_DEFINE2(setrlimit, unsigned int, resource,
1396 struct compat_rlimit __user *, rlim)
1399 struct compat_rlimit r32;
1401 if (copy_from_user(&r32, rlim, sizeof(struct compat_rlimit)))
1404 if (r32.rlim_cur == COMPAT_RLIM_INFINITY)
1405 r.rlim_cur = RLIM_INFINITY;
1407 r.rlim_cur = r32.rlim_cur;
1408 if (r32.rlim_max == COMPAT_RLIM_INFINITY)
1409 r.rlim_max = RLIM_INFINITY;
1411 r.rlim_max = r32.rlim_max;
1412 return do_prlimit(current, resource, &r, NULL);
1415 COMPAT_SYSCALL_DEFINE2(getrlimit, unsigned int, resource,
1416 struct compat_rlimit __user *, rlim)
1421 ret = do_prlimit(current, resource, NULL, &r);
1423 struct compat_rlimit r32;
1424 if (r.rlim_cur > COMPAT_RLIM_INFINITY)
1425 r32.rlim_cur = COMPAT_RLIM_INFINITY;
1427 r32.rlim_cur = r.rlim_cur;
1428 if (r.rlim_max > COMPAT_RLIM_INFINITY)
1429 r32.rlim_max = COMPAT_RLIM_INFINITY;
1431 r32.rlim_max = r.rlim_max;
1433 if (copy_to_user(rlim, &r32, sizeof(struct compat_rlimit)))
1441 #ifdef __ARCH_WANT_SYS_OLD_GETRLIMIT
1444 * Back compatibility for getrlimit. Needed for some apps.
1446 SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1447 struct rlimit __user *, rlim)
1450 if (resource >= RLIM_NLIMITS)
1453 resource = array_index_nospec(resource, RLIM_NLIMITS);
1454 task_lock(current->group_leader);
1455 x = current->signal->rlim[resource];
1456 task_unlock(current->group_leader);
1457 if (x.rlim_cur > 0x7FFFFFFF)
1458 x.rlim_cur = 0x7FFFFFFF;
1459 if (x.rlim_max > 0x7FFFFFFF)
1460 x.rlim_max = 0x7FFFFFFF;
1461 return copy_to_user(rlim, &x, sizeof(x)) ? -EFAULT : 0;
1464 #ifdef CONFIG_COMPAT
1465 COMPAT_SYSCALL_DEFINE2(old_getrlimit, unsigned int, resource,
1466 struct compat_rlimit __user *, rlim)
1470 if (resource >= RLIM_NLIMITS)
1473 resource = array_index_nospec(resource, RLIM_NLIMITS);
1474 task_lock(current->group_leader);
1475 r = current->signal->rlim[resource];
1476 task_unlock(current->group_leader);
1477 if (r.rlim_cur > 0x7FFFFFFF)
1478 r.rlim_cur = 0x7FFFFFFF;
1479 if (r.rlim_max > 0x7FFFFFFF)
1480 r.rlim_max = 0x7FFFFFFF;
1482 if (put_user(r.rlim_cur, &rlim->rlim_cur) ||
1483 put_user(r.rlim_max, &rlim->rlim_max))
1491 static inline bool rlim64_is_infinity(__u64 rlim64)
1493 #if BITS_PER_LONG < 64
1494 return rlim64 >= ULONG_MAX;
1496 return rlim64 == RLIM64_INFINITY;
1500 static void rlim_to_rlim64(const struct rlimit *rlim, struct rlimit64 *rlim64)
1502 if (rlim->rlim_cur == RLIM_INFINITY)
1503 rlim64->rlim_cur = RLIM64_INFINITY;
1505 rlim64->rlim_cur = rlim->rlim_cur;
1506 if (rlim->rlim_max == RLIM_INFINITY)
1507 rlim64->rlim_max = RLIM64_INFINITY;
1509 rlim64->rlim_max = rlim->rlim_max;
1512 static void rlim64_to_rlim(const struct rlimit64 *rlim64, struct rlimit *rlim)
1514 if (rlim64_is_infinity(rlim64->rlim_cur))
1515 rlim->rlim_cur = RLIM_INFINITY;
1517 rlim->rlim_cur = (unsigned long)rlim64->rlim_cur;
1518 if (rlim64_is_infinity(rlim64->rlim_max))
1519 rlim->rlim_max = RLIM_INFINITY;
1521 rlim->rlim_max = (unsigned long)rlim64->rlim_max;
1524 /* make sure you are allowed to change @tsk limits before calling this */
1525 int do_prlimit(struct task_struct *tsk, unsigned int resource,
1526 struct rlimit *new_rlim, struct rlimit *old_rlim)
1528 struct rlimit *rlim;
1531 if (resource >= RLIM_NLIMITS)
1534 if (new_rlim->rlim_cur > new_rlim->rlim_max)
1536 if (resource == RLIMIT_NOFILE &&
1537 new_rlim->rlim_max > sysctl_nr_open)
1541 /* protect tsk->signal and tsk->sighand from disappearing */
1542 read_lock(&tasklist_lock);
1543 if (!tsk->sighand) {
1548 rlim = tsk->signal->rlim + resource;
1549 task_lock(tsk->group_leader);
1551 /* Keep the capable check against init_user_ns until
1552 cgroups can contain all limits */
1553 if (new_rlim->rlim_max > rlim->rlim_max &&
1554 !capable(CAP_SYS_RESOURCE))
1557 retval = security_task_setrlimit(tsk, resource, new_rlim);
1558 if (resource == RLIMIT_CPU && new_rlim->rlim_cur == 0) {
1560 * The caller is asking for an immediate RLIMIT_CPU
1561 * expiry. But we use the zero value to mean "it was
1562 * never set". So let's cheat and make it one second
1565 new_rlim->rlim_cur = 1;
1574 task_unlock(tsk->group_leader);
1577 * RLIMIT_CPU handling. Note that the kernel fails to return an error
1578 * code if it rejected the user's attempt to set RLIMIT_CPU. This is a
1579 * very long-standing error, and fixing it now risks breakage of
1580 * applications, so we live with it
1582 if (!retval && new_rlim && resource == RLIMIT_CPU &&
1583 new_rlim->rlim_cur != RLIM_INFINITY &&
1584 IS_ENABLED(CONFIG_POSIX_TIMERS))
1585 update_rlimit_cpu(tsk, new_rlim->rlim_cur);
1587 read_unlock(&tasklist_lock);
1591 /* rcu lock must be held */
1592 static int check_prlimit_permission(struct task_struct *task,
1595 const struct cred *cred = current_cred(), *tcred;
1598 if (current == task)
1601 tcred = __task_cred(task);
1602 id_match = (uid_eq(cred->uid, tcred->euid) &&
1603 uid_eq(cred->uid, tcred->suid) &&
1604 uid_eq(cred->uid, tcred->uid) &&
1605 gid_eq(cred->gid, tcred->egid) &&
1606 gid_eq(cred->gid, tcred->sgid) &&
1607 gid_eq(cred->gid, tcred->gid));
1608 if (!id_match && !ns_capable(tcred->user_ns, CAP_SYS_RESOURCE))
1611 return security_task_prlimit(cred, tcred, flags);
1614 SYSCALL_DEFINE4(prlimit64, pid_t, pid, unsigned int, resource,
1615 const struct rlimit64 __user *, new_rlim,
1616 struct rlimit64 __user *, old_rlim)
1618 struct rlimit64 old64, new64;
1619 struct rlimit old, new;
1620 struct task_struct *tsk;
1621 unsigned int checkflags = 0;
1625 checkflags |= LSM_PRLIMIT_READ;
1628 if (copy_from_user(&new64, new_rlim, sizeof(new64)))
1630 rlim64_to_rlim(&new64, &new);
1631 checkflags |= LSM_PRLIMIT_WRITE;
1635 tsk = pid ? find_task_by_vpid(pid) : current;
1640 ret = check_prlimit_permission(tsk, checkflags);
1645 get_task_struct(tsk);
1648 ret = do_prlimit(tsk, resource, new_rlim ? &new : NULL,
1649 old_rlim ? &old : NULL);
1651 if (!ret && old_rlim) {
1652 rlim_to_rlim64(&old, &old64);
1653 if (copy_to_user(old_rlim, &old64, sizeof(old64)))
1657 put_task_struct(tsk);
1661 SYSCALL_DEFINE2(setrlimit, unsigned int, resource, struct rlimit __user *, rlim)
1663 struct rlimit new_rlim;
1665 if (copy_from_user(&new_rlim, rlim, sizeof(*rlim)))
1667 return do_prlimit(current, resource, &new_rlim, NULL);
1671 * It would make sense to put struct rusage in the task_struct,
1672 * except that would make the task_struct be *really big*. After
1673 * task_struct gets moved into malloc'ed memory, it would
1674 * make sense to do this. It will make moving the rest of the information
1675 * a lot simpler! (Which we're not doing right now because we're not
1676 * measuring them yet).
1678 * When sampling multiple threads for RUSAGE_SELF, under SMP we might have
1679 * races with threads incrementing their own counters. But since word
1680 * reads are atomic, we either get new values or old values and we don't
1681 * care which for the sums. We always take the siglock to protect reading
1682 * the c* fields from p->signal from races with exit.c updating those
1683 * fields when reaping, so a sample either gets all the additions of a
1684 * given child after it's reaped, or none so this sample is before reaping.
1687 * We need to take the siglock for CHILDEREN, SELF and BOTH
1688 * for the cases current multithreaded, non-current single threaded
1689 * non-current multithreaded. Thread traversal is now safe with
1691 * Strictly speaking, we donot need to take the siglock if we are current and
1692 * single threaded, as no one else can take our signal_struct away, no one
1693 * else can reap the children to update signal->c* counters, and no one else
1694 * can race with the signal-> fields. If we do not take any lock, the
1695 * signal-> fields could be read out of order while another thread was just
1696 * exiting. So we should place a read memory barrier when we avoid the lock.
1697 * On the writer side, write memory barrier is implied in __exit_signal
1698 * as __exit_signal releases the siglock spinlock after updating the signal->
1699 * fields. But we don't do this yet to keep things simple.
1703 static void accumulate_thread_rusage(struct task_struct *t, struct rusage *r)
1705 r->ru_nvcsw += t->nvcsw;
1706 r->ru_nivcsw += t->nivcsw;
1707 r->ru_minflt += t->min_flt;
1708 r->ru_majflt += t->maj_flt;
1709 r->ru_inblock += task_io_get_inblock(t);
1710 r->ru_oublock += task_io_get_oublock(t);
1713 void getrusage(struct task_struct *p, int who, struct rusage *r)
1715 struct task_struct *t;
1716 unsigned long flags;
1717 u64 tgutime, tgstime, utime, stime;
1718 unsigned long maxrss = 0;
1720 memset((char *)r, 0, sizeof (*r));
1723 if (who == RUSAGE_THREAD) {
1724 task_cputime_adjusted(current, &utime, &stime);
1725 accumulate_thread_rusage(p, r);
1726 maxrss = p->signal->maxrss;
1730 if (!lock_task_sighand(p, &flags))
1735 case RUSAGE_CHILDREN:
1736 utime = p->signal->cutime;
1737 stime = p->signal->cstime;
1738 r->ru_nvcsw = p->signal->cnvcsw;
1739 r->ru_nivcsw = p->signal->cnivcsw;
1740 r->ru_minflt = p->signal->cmin_flt;
1741 r->ru_majflt = p->signal->cmaj_flt;
1742 r->ru_inblock = p->signal->cinblock;
1743 r->ru_oublock = p->signal->coublock;
1744 maxrss = p->signal->cmaxrss;
1746 if (who == RUSAGE_CHILDREN)
1750 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1753 r->ru_nvcsw += p->signal->nvcsw;
1754 r->ru_nivcsw += p->signal->nivcsw;
1755 r->ru_minflt += p->signal->min_flt;
1756 r->ru_majflt += p->signal->maj_flt;
1757 r->ru_inblock += p->signal->inblock;
1758 r->ru_oublock += p->signal->oublock;
1759 if (maxrss < p->signal->maxrss)
1760 maxrss = p->signal->maxrss;
1763 accumulate_thread_rusage(t, r);
1764 } while_each_thread(p, t);
1770 unlock_task_sighand(p, &flags);
1773 r->ru_utime = ns_to_timeval(utime);
1774 r->ru_stime = ns_to_timeval(stime);
1776 if (who != RUSAGE_CHILDREN) {
1777 struct mm_struct *mm = get_task_mm(p);
1780 setmax_mm_hiwater_rss(&maxrss, mm);
1784 r->ru_maxrss = maxrss * (PAGE_SIZE / 1024); /* convert pages to KBs */
1787 SYSCALL_DEFINE2(getrusage, int, who, struct rusage __user *, ru)
1791 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1792 who != RUSAGE_THREAD)
1795 getrusage(current, who, &r);
1796 return copy_to_user(ru, &r, sizeof(r)) ? -EFAULT : 0;
1799 #ifdef CONFIG_COMPAT
1800 COMPAT_SYSCALL_DEFINE2(getrusage, int, who, struct compat_rusage __user *, ru)
1804 if (who != RUSAGE_SELF && who != RUSAGE_CHILDREN &&
1805 who != RUSAGE_THREAD)
1808 getrusage(current, who, &r);
1809 return put_compat_rusage(&r, ru);
1813 SYSCALL_DEFINE1(umask, int, mask)
1815 mask = xchg(¤t->fs->umask, mask & S_IRWXUGO);
1819 static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1822 struct file *old_exe, *exe_file;
1823 struct inode *inode;
1830 inode = file_inode(exe.file);
1833 * Because the original mm->exe_file points to executable file, make
1834 * sure that this one is executable as well, to avoid breaking an
1838 if (!S_ISREG(inode->i_mode) || path_noexec(&exe.file->f_path))
1841 err = inode_permission(inode, MAY_EXEC);
1846 * Forbid mm->exe_file change if old file still mapped.
1848 exe_file = get_mm_exe_file(mm);
1851 struct vm_area_struct *vma;
1853 down_read(&mm->mmap_sem);
1854 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1857 if (path_equal(&vma->vm_file->f_path,
1862 up_read(&mm->mmap_sem);
1867 /* set the new file, lockless */
1869 old_exe = xchg(&mm->exe_file, exe.file);
1876 up_read(&mm->mmap_sem);
1882 * WARNING: we don't require any capability here so be very careful
1883 * in what is allowed for modification from userspace.
1885 static int validate_prctl_map(struct prctl_mm_map *prctl_map)
1887 unsigned long mmap_max_addr = TASK_SIZE;
1888 struct mm_struct *mm = current->mm;
1889 int error = -EINVAL, i;
1891 static const unsigned char offsets[] = {
1892 offsetof(struct prctl_mm_map, start_code),
1893 offsetof(struct prctl_mm_map, end_code),
1894 offsetof(struct prctl_mm_map, start_data),
1895 offsetof(struct prctl_mm_map, end_data),
1896 offsetof(struct prctl_mm_map, start_brk),
1897 offsetof(struct prctl_mm_map, brk),
1898 offsetof(struct prctl_mm_map, start_stack),
1899 offsetof(struct prctl_mm_map, arg_start),
1900 offsetof(struct prctl_mm_map, arg_end),
1901 offsetof(struct prctl_mm_map, env_start),
1902 offsetof(struct prctl_mm_map, env_end),
1906 * Make sure the members are not somewhere outside
1907 * of allowed address space.
1909 for (i = 0; i < ARRAY_SIZE(offsets); i++) {
1910 u64 val = *(u64 *)((char *)prctl_map + offsets[i]);
1912 if ((unsigned long)val >= mmap_max_addr ||
1913 (unsigned long)val < mmap_min_addr)
1918 * Make sure the pairs are ordered.
1920 #define __prctl_check_order(__m1, __op, __m2) \
1921 ((unsigned long)prctl_map->__m1 __op \
1922 (unsigned long)prctl_map->__m2) ? 0 : -EINVAL
1923 error = __prctl_check_order(start_code, <, end_code);
1924 error |= __prctl_check_order(start_data,<=, end_data);
1925 error |= __prctl_check_order(start_brk, <=, brk);
1926 error |= __prctl_check_order(arg_start, <=, arg_end);
1927 error |= __prctl_check_order(env_start, <=, env_end);
1930 #undef __prctl_check_order
1935 * Neither we should allow to override limits if they set.
1937 if (check_data_rlimit(rlimit(RLIMIT_DATA), prctl_map->brk,
1938 prctl_map->start_brk, prctl_map->end_data,
1939 prctl_map->start_data))
1943 * Someone is trying to cheat the auxv vector.
1945 if (prctl_map->auxv_size) {
1946 if (!prctl_map->auxv || prctl_map->auxv_size > sizeof(mm->saved_auxv))
1951 * Finally, make sure the caller has the rights to
1952 * change /proc/pid/exe link: only local sys admin should
1955 if (prctl_map->exe_fd != (u32)-1) {
1956 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN))
1965 #ifdef CONFIG_CHECKPOINT_RESTORE
1966 static int prctl_set_mm_map(int opt, const void __user *addr, unsigned long data_size)
1968 struct prctl_mm_map prctl_map = { .exe_fd = (u32)-1, };
1969 unsigned long user_auxv[AT_VECTOR_SIZE];
1970 struct mm_struct *mm = current->mm;
1973 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
1974 BUILD_BUG_ON(sizeof(struct prctl_mm_map) > 256);
1976 if (opt == PR_SET_MM_MAP_SIZE)
1977 return put_user((unsigned int)sizeof(prctl_map),
1978 (unsigned int __user *)addr);
1980 if (data_size != sizeof(prctl_map))
1983 if (copy_from_user(&prctl_map, addr, sizeof(prctl_map)))
1986 error = validate_prctl_map(&prctl_map);
1990 if (prctl_map.auxv_size) {
1991 memset(user_auxv, 0, sizeof(user_auxv));
1992 if (copy_from_user(user_auxv,
1993 (const void __user *)prctl_map.auxv,
1994 prctl_map.auxv_size))
1997 /* Last entry must be AT_NULL as specification requires */
1998 user_auxv[AT_VECTOR_SIZE - 2] = AT_NULL;
1999 user_auxv[AT_VECTOR_SIZE - 1] = AT_NULL;
2002 if (prctl_map.exe_fd != (u32)-1) {
2003 error = prctl_set_mm_exe_file(mm, prctl_map.exe_fd);
2009 * arg_lock protects concurent updates but we still need mmap_sem for
2010 * read to exclude races with sys_brk.
2012 down_read(&mm->mmap_sem);
2015 * We don't validate if these members are pointing to
2016 * real present VMAs because application may have correspond
2017 * VMAs already unmapped and kernel uses these members for statistics
2018 * output in procfs mostly, except
2020 * - @start_brk/@brk which are used in do_brk but kernel lookups
2021 * for VMAs when updating these memvers so anything wrong written
2022 * here cause kernel to swear at userspace program but won't lead
2023 * to any problem in kernel itself
2026 spin_lock(&mm->arg_lock);
2027 mm->start_code = prctl_map.start_code;
2028 mm->end_code = prctl_map.end_code;
2029 mm->start_data = prctl_map.start_data;
2030 mm->end_data = prctl_map.end_data;
2031 mm->start_brk = prctl_map.start_brk;
2032 mm->brk = prctl_map.brk;
2033 mm->start_stack = prctl_map.start_stack;
2034 mm->arg_start = prctl_map.arg_start;
2035 mm->arg_end = prctl_map.arg_end;
2036 mm->env_start = prctl_map.env_start;
2037 mm->env_end = prctl_map.env_end;
2038 spin_unlock(&mm->arg_lock);
2041 * Note this update of @saved_auxv is lockless thus
2042 * if someone reads this member in procfs while we're
2043 * updating -- it may get partly updated results. It's
2044 * known and acceptable trade off: we leave it as is to
2045 * not introduce additional locks here making the kernel
2048 if (prctl_map.auxv_size)
2049 memcpy(mm->saved_auxv, user_auxv, sizeof(user_auxv));
2051 up_read(&mm->mmap_sem);
2054 #endif /* CONFIG_CHECKPOINT_RESTORE */
2056 static int prctl_set_auxv(struct mm_struct *mm, unsigned long addr,
2060 * This doesn't move the auxiliary vector itself since it's pinned to
2061 * mm_struct, but it permits filling the vector with new values. It's
2062 * up to the caller to provide sane values here, otherwise userspace
2063 * tools which use this vector might be unhappy.
2065 unsigned long user_auxv[AT_VECTOR_SIZE];
2067 if (len > sizeof(user_auxv))
2070 if (copy_from_user(user_auxv, (const void __user *)addr, len))
2073 /* Make sure the last entry is always AT_NULL */
2074 user_auxv[AT_VECTOR_SIZE - 2] = 0;
2075 user_auxv[AT_VECTOR_SIZE - 1] = 0;
2077 BUILD_BUG_ON(sizeof(user_auxv) != sizeof(mm->saved_auxv));
2080 memcpy(mm->saved_auxv, user_auxv, len);
2081 task_unlock(current);
2086 static int prctl_set_mm(int opt, unsigned long addr,
2087 unsigned long arg4, unsigned long arg5)
2089 struct mm_struct *mm = current->mm;
2090 struct prctl_mm_map prctl_map;
2091 struct vm_area_struct *vma;
2094 if (arg5 || (arg4 && (opt != PR_SET_MM_AUXV &&
2095 opt != PR_SET_MM_MAP &&
2096 opt != PR_SET_MM_MAP_SIZE)))
2099 #ifdef CONFIG_CHECKPOINT_RESTORE
2100 if (opt == PR_SET_MM_MAP || opt == PR_SET_MM_MAP_SIZE)
2101 return prctl_set_mm_map(opt, (const void __user *)addr, arg4);
2104 if (!capable(CAP_SYS_RESOURCE))
2107 if (opt == PR_SET_MM_EXE_FILE)
2108 return prctl_set_mm_exe_file(mm, (unsigned int)addr);
2110 if (opt == PR_SET_MM_AUXV)
2111 return prctl_set_auxv(mm, addr, arg4);
2113 if (addr >= TASK_SIZE || addr < mmap_min_addr)
2118 down_write(&mm->mmap_sem);
2119 vma = find_vma(mm, addr);
2121 prctl_map.start_code = mm->start_code;
2122 prctl_map.end_code = mm->end_code;
2123 prctl_map.start_data = mm->start_data;
2124 prctl_map.end_data = mm->end_data;
2125 prctl_map.start_brk = mm->start_brk;
2126 prctl_map.brk = mm->brk;
2127 prctl_map.start_stack = mm->start_stack;
2128 prctl_map.arg_start = mm->arg_start;
2129 prctl_map.arg_end = mm->arg_end;
2130 prctl_map.env_start = mm->env_start;
2131 prctl_map.env_end = mm->env_end;
2132 prctl_map.auxv = NULL;
2133 prctl_map.auxv_size = 0;
2134 prctl_map.exe_fd = -1;
2137 case PR_SET_MM_START_CODE:
2138 prctl_map.start_code = addr;
2140 case PR_SET_MM_END_CODE:
2141 prctl_map.end_code = addr;
2143 case PR_SET_MM_START_DATA:
2144 prctl_map.start_data = addr;
2146 case PR_SET_MM_END_DATA:
2147 prctl_map.end_data = addr;
2149 case PR_SET_MM_START_STACK:
2150 prctl_map.start_stack = addr;
2152 case PR_SET_MM_START_BRK:
2153 prctl_map.start_brk = addr;
2156 prctl_map.brk = addr;
2158 case PR_SET_MM_ARG_START:
2159 prctl_map.arg_start = addr;
2161 case PR_SET_MM_ARG_END:
2162 prctl_map.arg_end = addr;
2164 case PR_SET_MM_ENV_START:
2165 prctl_map.env_start = addr;
2167 case PR_SET_MM_ENV_END:
2168 prctl_map.env_end = addr;
2174 error = validate_prctl_map(&prctl_map);
2180 * If command line arguments and environment
2181 * are placed somewhere else on stack, we can
2182 * set them up here, ARG_START/END to setup
2183 * command line argumets and ENV_START/END
2186 case PR_SET_MM_START_STACK:
2187 case PR_SET_MM_ARG_START:
2188 case PR_SET_MM_ARG_END:
2189 case PR_SET_MM_ENV_START:
2190 case PR_SET_MM_ENV_END:
2197 mm->start_code = prctl_map.start_code;
2198 mm->end_code = prctl_map.end_code;
2199 mm->start_data = prctl_map.start_data;
2200 mm->end_data = prctl_map.end_data;
2201 mm->start_brk = prctl_map.start_brk;
2202 mm->brk = prctl_map.brk;
2203 mm->start_stack = prctl_map.start_stack;
2204 mm->arg_start = prctl_map.arg_start;
2205 mm->arg_end = prctl_map.arg_end;
2206 mm->env_start = prctl_map.env_start;
2207 mm->env_end = prctl_map.env_end;
2211 up_write(&mm->mmap_sem);
2215 #ifdef CONFIG_CHECKPOINT_RESTORE
2216 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2218 return put_user(me->clear_child_tid, tid_addr);
2221 static int prctl_get_tid_address(struct task_struct *me, int __user **tid_addr)
2227 static int propagate_has_child_subreaper(struct task_struct *p, void *data)
2230 * If task has has_child_subreaper - all its decendants
2231 * already have these flag too and new decendants will
2232 * inherit it on fork, skip them.
2234 * If we've found child_reaper - skip descendants in
2235 * it's subtree as they will never get out pidns.
2237 if (p->signal->has_child_subreaper ||
2238 is_child_reaper(task_pid(p)))
2241 p->signal->has_child_subreaper = 1;
2245 int __weak arch_prctl_spec_ctrl_get(struct task_struct *t, unsigned long which)
2250 int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,
2256 SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3,
2257 unsigned long, arg4, unsigned long, arg5)
2259 struct task_struct *me = current;
2260 unsigned char comm[sizeof(me->comm)];
2263 error = security_task_prctl(option, arg2, arg3, arg4, arg5);
2264 if (error != -ENOSYS)
2269 case PR_SET_PDEATHSIG:
2270 if (!valid_signal(arg2)) {
2274 me->pdeath_signal = arg2;
2276 case PR_GET_PDEATHSIG:
2277 error = put_user(me->pdeath_signal, (int __user *)arg2);
2279 case PR_GET_DUMPABLE:
2280 error = get_dumpable(me->mm);
2282 case PR_SET_DUMPABLE:
2283 if (arg2 != SUID_DUMP_DISABLE && arg2 != SUID_DUMP_USER) {
2287 set_dumpable(me->mm, arg2);
2290 case PR_SET_UNALIGN:
2291 error = SET_UNALIGN_CTL(me, arg2);
2293 case PR_GET_UNALIGN:
2294 error = GET_UNALIGN_CTL(me, arg2);
2297 error = SET_FPEMU_CTL(me, arg2);
2300 error = GET_FPEMU_CTL(me, arg2);
2303 error = SET_FPEXC_CTL(me, arg2);
2306 error = GET_FPEXC_CTL(me, arg2);
2309 error = PR_TIMING_STATISTICAL;
2312 if (arg2 != PR_TIMING_STATISTICAL)
2316 comm[sizeof(me->comm) - 1] = 0;
2317 if (strncpy_from_user(comm, (char __user *)arg2,
2318 sizeof(me->comm) - 1) < 0)
2320 set_task_comm(me, comm);
2321 proc_comm_connector(me);
2324 get_task_comm(comm, me);
2325 if (copy_to_user((char __user *)arg2, comm, sizeof(comm)))
2329 error = GET_ENDIAN(me, arg2);
2332 error = SET_ENDIAN(me, arg2);
2334 case PR_GET_SECCOMP:
2335 error = prctl_get_seccomp();
2337 case PR_SET_SECCOMP:
2338 error = prctl_set_seccomp(arg2, (char __user *)arg3);
2341 error = GET_TSC_CTL(arg2);
2344 error = SET_TSC_CTL(arg2);
2346 case PR_TASK_PERF_EVENTS_DISABLE:
2347 error = perf_event_task_disable();
2349 case PR_TASK_PERF_EVENTS_ENABLE:
2350 error = perf_event_task_enable();
2352 case PR_GET_TIMERSLACK:
2353 if (current->timer_slack_ns > ULONG_MAX)
2356 error = current->timer_slack_ns;
2358 case PR_SET_TIMERSLACK:
2360 current->timer_slack_ns =
2361 current->default_timer_slack_ns;
2363 current->timer_slack_ns = arg2;
2369 case PR_MCE_KILL_CLEAR:
2372 current->flags &= ~PF_MCE_PROCESS;
2374 case PR_MCE_KILL_SET:
2375 current->flags |= PF_MCE_PROCESS;
2376 if (arg3 == PR_MCE_KILL_EARLY)
2377 current->flags |= PF_MCE_EARLY;
2378 else if (arg3 == PR_MCE_KILL_LATE)
2379 current->flags &= ~PF_MCE_EARLY;
2380 else if (arg3 == PR_MCE_KILL_DEFAULT)
2382 ~(PF_MCE_EARLY|PF_MCE_PROCESS);
2390 case PR_MCE_KILL_GET:
2391 if (arg2 | arg3 | arg4 | arg5)
2393 if (current->flags & PF_MCE_PROCESS)
2394 error = (current->flags & PF_MCE_EARLY) ?
2395 PR_MCE_KILL_EARLY : PR_MCE_KILL_LATE;
2397 error = PR_MCE_KILL_DEFAULT;
2400 error = prctl_set_mm(arg2, arg3, arg4, arg5);
2402 case PR_GET_TID_ADDRESS:
2403 error = prctl_get_tid_address(me, (int __user **)arg2);
2405 case PR_SET_CHILD_SUBREAPER:
2406 me->signal->is_child_subreaper = !!arg2;
2410 walk_process_tree(me, propagate_has_child_subreaper, NULL);
2412 case PR_GET_CHILD_SUBREAPER:
2413 error = put_user(me->signal->is_child_subreaper,
2414 (int __user *)arg2);
2416 case PR_SET_NO_NEW_PRIVS:
2417 if (arg2 != 1 || arg3 || arg4 || arg5)
2420 task_set_no_new_privs(current);
2422 case PR_GET_NO_NEW_PRIVS:
2423 if (arg2 || arg3 || arg4 || arg5)
2425 return task_no_new_privs(current) ? 1 : 0;
2426 case PR_GET_THP_DISABLE:
2427 if (arg2 || arg3 || arg4 || arg5)
2429 error = !!test_bit(MMF_DISABLE_THP, &me->mm->flags);
2431 case PR_SET_THP_DISABLE:
2432 if (arg3 || arg4 || arg5)
2434 if (down_write_killable(&me->mm->mmap_sem))
2437 set_bit(MMF_DISABLE_THP, &me->mm->flags);
2439 clear_bit(MMF_DISABLE_THP, &me->mm->flags);
2440 up_write(&me->mm->mmap_sem);
2442 case PR_MPX_ENABLE_MANAGEMENT:
2443 if (arg2 || arg3 || arg4 || arg5)
2445 error = MPX_ENABLE_MANAGEMENT();
2447 case PR_MPX_DISABLE_MANAGEMENT:
2448 if (arg2 || arg3 || arg4 || arg5)
2450 error = MPX_DISABLE_MANAGEMENT();
2452 case PR_SET_FP_MODE:
2453 error = SET_FP_MODE(me, arg2);
2455 case PR_GET_FP_MODE:
2456 error = GET_FP_MODE(me);
2459 error = SVE_SET_VL(arg2);
2462 error = SVE_GET_VL();
2464 case PR_GET_SPECULATION_CTRL:
2465 if (arg3 || arg4 || arg5)
2467 error = arch_prctl_spec_ctrl_get(me, arg2);
2469 case PR_SET_SPECULATION_CTRL:
2472 error = arch_prctl_spec_ctrl_set(me, arg2, arg3);
2481 SYSCALL_DEFINE3(getcpu, unsigned __user *, cpup, unsigned __user *, nodep,
2482 struct getcpu_cache __user *, unused)
2485 int cpu = raw_smp_processor_id();
2488 err |= put_user(cpu, cpup);
2490 err |= put_user(cpu_to_node(cpu), nodep);
2491 return err ? -EFAULT : 0;
2495 * do_sysinfo - fill in sysinfo struct
2496 * @info: pointer to buffer to fill
2498 static int do_sysinfo(struct sysinfo *info)
2500 unsigned long mem_total, sav_total;
2501 unsigned int mem_unit, bitcount;
2502 struct timespec64 tp;
2504 memset(info, 0, sizeof(struct sysinfo));
2506 ktime_get_boottime_ts64(&tp);
2507 info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
2509 get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
2511 info->procs = nr_threads;
2517 * If the sum of all the available memory (i.e. ram + swap)
2518 * is less than can be stored in a 32 bit unsigned long then
2519 * we can be binary compatible with 2.2.x kernels. If not,
2520 * well, in that case 2.2.x was broken anyways...
2522 * -Erik Andersen <andersee@debian.org>
2525 mem_total = info->totalram + info->totalswap;
2526 if (mem_total < info->totalram || mem_total < info->totalswap)
2529 mem_unit = info->mem_unit;
2530 while (mem_unit > 1) {
2533 sav_total = mem_total;
2535 if (mem_total < sav_total)
2540 * If mem_total did not overflow, multiply all memory values by
2541 * info->mem_unit and set it to 1. This leaves things compatible
2542 * with 2.2.x, and also retains compatibility with earlier 2.4.x
2547 info->totalram <<= bitcount;
2548 info->freeram <<= bitcount;
2549 info->sharedram <<= bitcount;
2550 info->bufferram <<= bitcount;
2551 info->totalswap <<= bitcount;
2552 info->freeswap <<= bitcount;
2553 info->totalhigh <<= bitcount;
2554 info->freehigh <<= bitcount;
2560 SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
2566 if (copy_to_user(info, &val, sizeof(struct sysinfo)))
2572 #ifdef CONFIG_COMPAT
2573 struct compat_sysinfo {
2587 char _f[20-2*sizeof(u32)-sizeof(int)];
2590 COMPAT_SYSCALL_DEFINE1(sysinfo, struct compat_sysinfo __user *, info)
2596 /* Check to see if any memory value is too large for 32-bit and scale
2599 if (upper_32_bits(s.totalram) || upper_32_bits(s.totalswap)) {
2602 while (s.mem_unit < PAGE_SIZE) {
2607 s.totalram >>= bitcount;
2608 s.freeram >>= bitcount;
2609 s.sharedram >>= bitcount;
2610 s.bufferram >>= bitcount;
2611 s.totalswap >>= bitcount;
2612 s.freeswap >>= bitcount;
2613 s.totalhigh >>= bitcount;
2614 s.freehigh >>= bitcount;
2617 if (!access_ok(VERIFY_WRITE, info, sizeof(struct compat_sysinfo)) ||
2618 __put_user(s.uptime, &info->uptime) ||
2619 __put_user(s.loads[0], &info->loads[0]) ||
2620 __put_user(s.loads[1], &info->loads[1]) ||
2621 __put_user(s.loads[2], &info->loads[2]) ||
2622 __put_user(s.totalram, &info->totalram) ||
2623 __put_user(s.freeram, &info->freeram) ||
2624 __put_user(s.sharedram, &info->sharedram) ||
2625 __put_user(s.bufferram, &info->bufferram) ||
2626 __put_user(s.totalswap, &info->totalswap) ||
2627 __put_user(s.freeswap, &info->freeswap) ||
2628 __put_user(s.procs, &info->procs) ||
2629 __put_user(s.totalhigh, &info->totalhigh) ||
2630 __put_user(s.freehigh, &info->freehigh) ||
2631 __put_user(s.mem_unit, &info->mem_unit))
2636 #endif /* CONFIG_COMPAT */