1 /* Common capabilities, needed by capability.o.
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License as published by
5 * the Free Software Foundation; either version 2 of the License, or
6 * (at your option) any later version.
10 #include <linux/capability.h>
11 #include <linux/audit.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/lsm_hooks.h>
15 #include <linux/file.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/skbuff.h>
21 #include <linux/netlink.h>
22 #include <linux/ptrace.h>
23 #include <linux/xattr.h>
24 #include <linux/hugetlb.h>
25 #include <linux/mount.h>
26 #include <linux/sched.h>
27 #include <linux/prctl.h>
28 #include <linux/securebits.h>
29 #include <linux/user_namespace.h>
30 #include <linux/binfmts.h>
31 #include <linux/personality.h>
34 * If a non-root user executes a setuid-root binary in
35 * !secure(SECURE_NOROOT) mode, then we raise capabilities.
36 * However if fE is also set, then the intent is for only
37 * the file capabilities to be applied, and the setuid-root
38 * bit is left on either to change the uid (plausible) or
39 * to get full privilege on a kernel without file capabilities
40 * support. So in that case we do not raise capabilities.
42 * Warn if that happens, once per boot.
44 static void warn_setuid_and_fcaps_mixed(const char *fname
)
48 printk(KERN_INFO
"warning: `%s' has both setuid-root and"
49 " effective capabilities. Therefore not raising all"
50 " capabilities.\n", fname
);
56 * cap_capable - Determine whether a task has a particular effective capability
57 * @cred: The credentials to use
58 * @ns: The user namespace in which we need the capability
59 * @cap: The capability to check for
60 * @audit: Whether to write an audit message or not
62 * Determine whether the nominated task has the specified capability amongst
63 * its effective set, returning 0 if it does, -ve if it does not.
65 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
66 * and has_capability() functions. That is, it has the reverse semantics:
67 * cap_has_capability() returns 0 when a task has a capability, but the
68 * kernel's capable() and has_capability() returns 1 for this case.
70 int cap_capable(const struct cred
*cred
, struct user_namespace
*targ_ns
,
71 int cap
, unsigned int opts
)
73 struct user_namespace
*ns
= targ_ns
;
75 /* See if cred has the capability in the target user namespace
76 * by examining the target user namespace and all of the target
77 * user namespace's parents.
80 /* Do we have the necessary capabilities? */
81 if (ns
== cred
->user_ns
)
82 return cap_raised(cred
->cap_effective
, cap
) ? 0 : -EPERM
;
85 * If we're already at a lower level than we're looking for,
86 * we're done searching.
88 if (ns
->level
<= cred
->user_ns
->level
)
92 * The owner of the user namespace in the parent of the
93 * user namespace has all caps.
95 if ((ns
->parent
== cred
->user_ns
) && uid_eq(ns
->owner
, cred
->euid
))
99 * If you have a capability in a parent user ns, then you have
100 * it over all children user namespaces as well.
105 /* We never get here */
109 * cap_settime - Determine whether the current process may set the system clock
110 * @ts: The time to set
111 * @tz: The timezone to set
113 * Determine whether the current process may set the system clock and timezone
114 * information, returning 0 if permission granted, -ve if denied.
116 int cap_settime(const struct timespec64
*ts
, const struct timezone
*tz
)
118 if (!capable(CAP_SYS_TIME
))
124 * cap_ptrace_access_check - Determine whether the current process may access
126 * @child: The process to be accessed
127 * @mode: The mode of attachment.
129 * If we are in the same or an ancestor user_ns and have all the target
130 * task's capabilities, then ptrace access is allowed.
131 * If we have the ptrace capability to the target user_ns, then ptrace
135 * Determine whether a process may access another, returning 0 if permission
136 * granted, -ve if denied.
138 int cap_ptrace_access_check(struct task_struct
*child
, unsigned int mode
)
141 const struct cred
*cred
, *child_cred
;
142 const kernel_cap_t
*caller_caps
;
145 cred
= current_cred();
146 child_cred
= __task_cred(child
);
147 if (mode
& PTRACE_MODE_FSCREDS
)
148 caller_caps
= &cred
->cap_effective
;
150 caller_caps
= &cred
->cap_permitted
;
151 if (cred
->user_ns
== child_cred
->user_ns
&&
152 cap_issubset(child_cred
->cap_permitted
, *caller_caps
))
154 if (ns_capable(child_cred
->user_ns
, CAP_SYS_PTRACE
))
163 * cap_ptrace_traceme - Determine whether another process may trace the current
164 * @parent: The task proposed to be the tracer
166 * If parent is in the same or an ancestor user_ns and has all current's
167 * capabilities, then ptrace access is allowed.
168 * If parent has the ptrace capability to current's user_ns, then ptrace
172 * Determine whether the nominated task is permitted to trace the current
173 * process, returning 0 if permission is granted, -ve if denied.
175 int cap_ptrace_traceme(struct task_struct
*parent
)
178 const struct cred
*cred
, *child_cred
;
181 cred
= __task_cred(parent
);
182 child_cred
= current_cred();
183 if (cred
->user_ns
== child_cred
->user_ns
&&
184 cap_issubset(child_cred
->cap_permitted
, cred
->cap_permitted
))
186 if (has_ns_capability(parent
, child_cred
->user_ns
, CAP_SYS_PTRACE
))
195 * cap_capget - Retrieve a task's capability sets
196 * @target: The task from which to retrieve the capability sets
197 * @effective: The place to record the effective set
198 * @inheritable: The place to record the inheritable set
199 * @permitted: The place to record the permitted set
201 * This function retrieves the capabilities of the nominated task and returns
202 * them to the caller.
204 int cap_capget(struct task_struct
*target
, kernel_cap_t
*effective
,
205 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
207 const struct cred
*cred
;
209 /* Derived from kernel/capability.c:sys_capget. */
211 cred
= __task_cred(target
);
212 *effective
= cred
->cap_effective
;
213 *inheritable
= cred
->cap_inheritable
;
214 *permitted
= cred
->cap_permitted
;
220 * Determine whether the inheritable capabilities are limited to the old
221 * permitted set. Returns 1 if they are limited, 0 if they are not.
223 static inline int cap_inh_is_capped(void)
225 /* they are so limited unless the current task has the CAP_SETPCAP
228 if (cap_capable(current_cred(), current_cred()->user_ns
,
229 CAP_SETPCAP
, CAP_OPT_NONE
) == 0)
235 * cap_capset - Validate and apply proposed changes to current's capabilities
236 * @new: The proposed new credentials; alterations should be made here
237 * @old: The current task's current credentials
238 * @effective: A pointer to the proposed new effective capabilities set
239 * @inheritable: A pointer to the proposed new inheritable capabilities set
240 * @permitted: A pointer to the proposed new permitted capabilities set
242 * This function validates and applies a proposed mass change to the current
243 * process's capability sets. The changes are made to the proposed new
244 * credentials, and assuming no error, will be committed by the caller of LSM.
246 int cap_capset(struct cred
*new,
247 const struct cred
*old
,
248 const kernel_cap_t
*effective
,
249 const kernel_cap_t
*inheritable
,
250 const kernel_cap_t
*permitted
)
252 if (cap_inh_is_capped() &&
253 !cap_issubset(*inheritable
,
254 cap_combine(old
->cap_inheritable
,
255 old
->cap_permitted
)))
256 /* incapable of using this inheritable set */
259 if (!cap_issubset(*inheritable
,
260 cap_combine(old
->cap_inheritable
,
262 /* no new pI capabilities outside bounding set */
265 /* verify restrictions on target's new Permitted set */
266 if (!cap_issubset(*permitted
, old
->cap_permitted
))
269 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
270 if (!cap_issubset(*effective
, *permitted
))
273 new->cap_effective
= *effective
;
274 new->cap_inheritable
= *inheritable
;
275 new->cap_permitted
= *permitted
;
278 * Mask off ambient bits that are no longer both permitted and
281 new->cap_ambient
= cap_intersect(new->cap_ambient
,
282 cap_intersect(*permitted
,
284 if (WARN_ON(!cap_ambient_invariant_ok(new)))
290 * cap_inode_need_killpriv - Determine if inode change affects privileges
291 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
293 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
294 * affects the security markings on that inode, and if it is, should
295 * inode_killpriv() be invoked or the change rejected.
297 * Returns 1 if security.capability has a value, meaning inode_killpriv()
298 * is required, 0 otherwise, meaning inode_killpriv() is not required.
300 int cap_inode_need_killpriv(struct dentry
*dentry
)
302 struct inode
*inode
= d_backing_inode(dentry
);
305 error
= __vfs_getxattr(dentry
, inode
, XATTR_NAME_CAPS
, NULL
, 0);
310 * cap_inode_killpriv - Erase the security markings on an inode
311 * @dentry: The inode/dentry to alter
313 * Erase the privilege-enhancing security markings on an inode.
315 * Returns 0 if successful, -ve on error.
317 int cap_inode_killpriv(struct dentry
*dentry
)
321 error
= __vfs_removexattr(dentry
, XATTR_NAME_CAPS
);
322 if (error
== -EOPNOTSUPP
)
327 static bool rootid_owns_currentns(kuid_t kroot
)
329 struct user_namespace
*ns
;
331 if (!uid_valid(kroot
))
334 for (ns
= current_user_ns(); ; ns
= ns
->parent
) {
335 if (from_kuid(ns
, kroot
) == 0)
337 if (ns
== &init_user_ns
)
344 static __u32
sansflags(__u32 m
)
346 return m
& ~VFS_CAP_FLAGS_EFFECTIVE
;
349 static bool is_v2header(size_t size
, const struct vfs_cap_data
*cap
)
351 if (size
!= XATTR_CAPS_SZ_2
)
353 return sansflags(le32_to_cpu(cap
->magic_etc
)) == VFS_CAP_REVISION_2
;
356 static bool is_v3header(size_t size
, const struct vfs_cap_data
*cap
)
358 if (size
!= XATTR_CAPS_SZ_3
)
360 return sansflags(le32_to_cpu(cap
->magic_etc
)) == VFS_CAP_REVISION_3
;
364 * getsecurity: We are called for security.* before any attempt to read the
365 * xattr from the inode itself.
367 * This gives us a chance to read the on-disk value and convert it. If we
368 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
370 * Note we are not called by vfs_getxattr_alloc(), but that is only called
371 * by the integrity subsystem, which really wants the unconverted values -
374 int cap_inode_getsecurity(struct inode
*inode
, const char *name
, void **buffer
,
379 uid_t root
, mappedroot
;
381 struct vfs_cap_data
*cap
;
382 struct vfs_ns_cap_data
*nscap
;
383 struct dentry
*dentry
;
384 struct user_namespace
*fs_ns
;
386 if (strcmp(name
, "capability") != 0)
389 dentry
= d_find_any_alias(inode
);
393 size
= sizeof(struct vfs_ns_cap_data
);
394 ret
= (int) vfs_getxattr_alloc(dentry
, XATTR_NAME_CAPS
,
395 &tmpbuf
, size
, GFP_NOFS
);
401 fs_ns
= inode
->i_sb
->s_user_ns
;
402 cap
= (struct vfs_cap_data
*) tmpbuf
;
403 if (is_v2header((size_t) ret
, cap
)) {
404 /* If this is sizeof(vfs_cap_data) then we're ok with the
405 * on-disk value, so return that. */
411 } else if (!is_v3header((size_t) ret
, cap
)) {
416 nscap
= (struct vfs_ns_cap_data
*) tmpbuf
;
417 root
= le32_to_cpu(nscap
->rootid
);
418 kroot
= make_kuid(fs_ns
, root
);
420 /* If the root kuid maps to a valid uid in current ns, then return
421 * this as a nscap. */
422 mappedroot
= from_kuid(current_user_ns(), kroot
);
423 if (mappedroot
!= (uid_t
)-1 && mappedroot
!= (uid_t
)0) {
426 nscap
->rootid
= cpu_to_le32(mappedroot
);
432 if (!rootid_owns_currentns(kroot
)) {
437 /* This comes from a parent namespace. Return as a v2 capability */
438 size
= sizeof(struct vfs_cap_data
);
440 *buffer
= kmalloc(size
, GFP_ATOMIC
);
442 struct vfs_cap_data
*cap
= *buffer
;
443 __le32 nsmagic
, magic
;
444 magic
= VFS_CAP_REVISION_2
;
445 nsmagic
= le32_to_cpu(nscap
->magic_etc
);
446 if (nsmagic
& VFS_CAP_FLAGS_EFFECTIVE
)
447 magic
|= VFS_CAP_FLAGS_EFFECTIVE
;
448 memcpy(&cap
->data
, &nscap
->data
, sizeof(__le32
) * 2 * VFS_CAP_U32
);
449 cap
->magic_etc
= cpu_to_le32(magic
);
458 static kuid_t
rootid_from_xattr(const void *value
, size_t size
,
459 struct user_namespace
*task_ns
)
461 const struct vfs_ns_cap_data
*nscap
= value
;
464 if (size
== XATTR_CAPS_SZ_3
)
465 rootid
= le32_to_cpu(nscap
->rootid
);
467 return make_kuid(task_ns
, rootid
);
470 static bool validheader(size_t size
, const struct vfs_cap_data
*cap
)
472 return is_v2header(size
, cap
) || is_v3header(size
, cap
);
476 * User requested a write of security.capability. If needed, update the
477 * xattr to change from v2 to v3, or to fixup the v3 rootid.
479 * If all is ok, we return the new size, on error return < 0.
481 int cap_convert_nscap(struct dentry
*dentry
, void **ivalue
, size_t size
)
483 struct vfs_ns_cap_data
*nscap
;
485 const struct vfs_cap_data
*cap
= *ivalue
;
486 __u32 magic
, nsmagic
;
487 struct inode
*inode
= d_backing_inode(dentry
);
488 struct user_namespace
*task_ns
= current_user_ns(),
489 *fs_ns
= inode
->i_sb
->s_user_ns
;
495 if (!validheader(size
, cap
))
497 if (!capable_wrt_inode_uidgid(inode
, CAP_SETFCAP
))
499 if (size
== XATTR_CAPS_SZ_2
)
500 if (ns_capable(inode
->i_sb
->s_user_ns
, CAP_SETFCAP
))
501 /* user is privileged, just write the v2 */
504 rootid
= rootid_from_xattr(*ivalue
, size
, task_ns
);
505 if (!uid_valid(rootid
))
508 nsrootid
= from_kuid(fs_ns
, rootid
);
512 newsize
= sizeof(struct vfs_ns_cap_data
);
513 nscap
= kmalloc(newsize
, GFP_ATOMIC
);
516 nscap
->rootid
= cpu_to_le32(nsrootid
);
517 nsmagic
= VFS_CAP_REVISION_3
;
518 magic
= le32_to_cpu(cap
->magic_etc
);
519 if (magic
& VFS_CAP_FLAGS_EFFECTIVE
)
520 nsmagic
|= VFS_CAP_FLAGS_EFFECTIVE
;
521 nscap
->magic_etc
= cpu_to_le32(nsmagic
);
522 memcpy(&nscap
->data
, &cap
->data
, sizeof(__le32
) * 2 * VFS_CAP_U32
);
530 * Calculate the new process capability sets from the capability sets attached
533 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data
*caps
,
534 struct linux_binprm
*bprm
,
538 struct cred
*new = bprm
->cred
;
542 if (caps
->magic_etc
& VFS_CAP_FLAGS_EFFECTIVE
)
545 if (caps
->magic_etc
& VFS_CAP_REVISION_MASK
)
548 CAP_FOR_EACH_U32(i
) {
549 __u32 permitted
= caps
->permitted
.cap
[i
];
550 __u32 inheritable
= caps
->inheritable
.cap
[i
];
553 * pP' = (X & fP) | (pI & fI)
554 * The addition of pA' is handled later.
556 new->cap_permitted
.cap
[i
] =
557 (new->cap_bset
.cap
[i
] & permitted
) |
558 (new->cap_inheritable
.cap
[i
] & inheritable
);
560 if (permitted
& ~new->cap_permitted
.cap
[i
])
561 /* insufficient to execute correctly */
566 * For legacy apps, with no internal support for recognizing they
567 * do not have enough capabilities, we return an error if they are
568 * missing some "forced" (aka file-permitted) capabilities.
570 return *effective
? ret
: 0;
574 * Extract the on-exec-apply capability sets for an executable file.
576 int get_vfs_caps_from_disk(const struct dentry
*dentry
, struct cpu_vfs_cap_data
*cpu_caps
)
578 struct inode
*inode
= d_backing_inode(dentry
);
582 struct vfs_ns_cap_data data
, *nscaps
= &data
;
583 struct vfs_cap_data
*caps
= (struct vfs_cap_data
*) &data
;
585 struct user_namespace
*fs_ns
;
587 memset(cpu_caps
, 0, sizeof(struct cpu_vfs_cap_data
));
592 fs_ns
= inode
->i_sb
->s_user_ns
;
593 size
= __vfs_getxattr((struct dentry
*)dentry
, inode
,
594 XATTR_NAME_CAPS
, &data
, XATTR_CAPS_SZ
);
595 if (size
== -ENODATA
|| size
== -EOPNOTSUPP
)
596 /* no data, that's ok */
602 if (size
< sizeof(magic_etc
))
605 cpu_caps
->magic_etc
= magic_etc
= le32_to_cpu(caps
->magic_etc
);
607 rootkuid
= make_kuid(fs_ns
, 0);
608 switch (magic_etc
& VFS_CAP_REVISION_MASK
) {
609 case VFS_CAP_REVISION_1
:
610 if (size
!= XATTR_CAPS_SZ_1
)
612 tocopy
= VFS_CAP_U32_1
;
614 case VFS_CAP_REVISION_2
:
615 if (size
!= XATTR_CAPS_SZ_2
)
617 tocopy
= VFS_CAP_U32_2
;
619 case VFS_CAP_REVISION_3
:
620 if (size
!= XATTR_CAPS_SZ_3
)
622 tocopy
= VFS_CAP_U32_3
;
623 rootkuid
= make_kuid(fs_ns
, le32_to_cpu(nscaps
->rootid
));
629 /* Limit the caps to the mounter of the filesystem
630 * or the more limited uid specified in the xattr.
632 if (!rootid_owns_currentns(rootkuid
))
635 CAP_FOR_EACH_U32(i
) {
638 cpu_caps
->permitted
.cap
[i
] = le32_to_cpu(caps
->data
[i
].permitted
);
639 cpu_caps
->inheritable
.cap
[i
] = le32_to_cpu(caps
->data
[i
].inheritable
);
642 cpu_caps
->permitted
.cap
[CAP_LAST_U32
] &= CAP_LAST_U32_VALID_MASK
;
643 cpu_caps
->inheritable
.cap
[CAP_LAST_U32
] &= CAP_LAST_U32_VALID_MASK
;
649 * Attempt to get the on-exec apply capability sets for an executable file from
650 * its xattrs and, if present, apply them to the proposed credentials being
651 * constructed by execve().
653 static int get_file_caps(struct linux_binprm
*bprm
, bool *effective
, bool *has_fcap
)
656 struct cpu_vfs_cap_data vcaps
;
658 cap_clear(bprm
->cred
->cap_permitted
);
660 if (!file_caps_enabled
)
663 if (path_nosuid(&bprm
->file
->f_path
))
667 * This check is redundant with mnt_may_suid() but is kept to make
668 * explicit that capability bits are limited to s_user_ns and its
671 if (!current_in_userns(bprm
->file
->f_path
.mnt
->mnt_sb
->s_user_ns
))
674 rc
= get_vfs_caps_from_disk(bprm
->file
->f_path
.dentry
, &vcaps
);
677 printk(KERN_NOTICE
"Invalid argument reading file caps for %s\n",
679 else if (rc
== -ENODATA
)
684 rc
= bprm_caps_from_vfs_caps(&vcaps
, bprm
, effective
, has_fcap
);
688 cap_clear(bprm
->cred
->cap_permitted
);
693 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT
); }
695 static inline bool __is_real(kuid_t uid
, struct cred
*cred
)
696 { return uid_eq(cred
->uid
, uid
); }
698 static inline bool __is_eff(kuid_t uid
, struct cred
*cred
)
699 { return uid_eq(cred
->euid
, uid
); }
701 static inline bool __is_suid(kuid_t uid
, struct cred
*cred
)
702 { return !__is_real(uid
, cred
) && __is_eff(uid
, cred
); }
705 * handle_privileged_root - Handle case of privileged root
706 * @bprm: The execution parameters, including the proposed creds
707 * @has_fcap: Are any file capabilities set?
708 * @effective: Do we have effective root privilege?
709 * @root_uid: This namespace' root UID WRT initial USER namespace
711 * Handle the case where root is privileged and hasn't been neutered by
712 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
713 * set UID root and nothing is changed. If we are root, cap_permitted is
714 * updated. If we have become set UID root, the effective bit is set.
716 static void handle_privileged_root(struct linux_binprm
*bprm
, bool has_fcap
,
717 bool *effective
, kuid_t root_uid
)
719 const struct cred
*old
= current_cred();
720 struct cred
*new = bprm
->cred
;
722 if (!root_privileged())
725 * If the legacy file capability is set, then don't set privs
726 * for a setuid root binary run by a non-root user. Do set it
727 * for a root user just to cause least surprise to an admin.
729 if (has_fcap
&& __is_suid(root_uid
, new)) {
730 warn_setuid_and_fcaps_mixed(bprm
->filename
);
734 * To support inheritance of root-permissions and suid-root
735 * executables under compatibility mode, we override the
736 * capability sets for the file.
738 if (__is_eff(root_uid
, new) || __is_real(root_uid
, new)) {
739 /* pP' = (cap_bset & ~0) | (pI & ~0) */
740 new->cap_permitted
= cap_combine(old
->cap_bset
,
741 old
->cap_inheritable
);
744 * If only the real uid is 0, we do not set the effective bit.
746 if (__is_eff(root_uid
, new))
750 #define __cap_gained(field, target, source) \
751 !cap_issubset(target->cap_##field, source->cap_##field)
752 #define __cap_grew(target, source, cred) \
753 !cap_issubset(cred->cap_##target, cred->cap_##source)
754 #define __cap_full(field, cred) \
755 cap_issubset(CAP_FULL_SET, cred->cap_##field)
757 static inline bool __is_setuid(struct cred
*new, const struct cred
*old
)
758 { return !uid_eq(new->euid
, old
->uid
); }
760 static inline bool __is_setgid(struct cred
*new, const struct cred
*old
)
761 { return !gid_eq(new->egid
, old
->gid
); }
764 * 1) Audit candidate if current->cap_effective is set
766 * We do not bother to audit if 3 things are true:
767 * 1) cap_effective has all caps
768 * 2) we became root *OR* are were already root
769 * 3) root is supposed to have all caps (SECURE_NOROOT)
770 * Since this is just a normal root execing a process.
772 * Number 1 above might fail if you don't have a full bset, but I think
773 * that is interesting information to audit.
775 * A number of other conditions require logging:
776 * 2) something prevented setuid root getting all caps
777 * 3) non-setuid root gets fcaps
778 * 4) non-setuid root gets ambient
780 static inline bool nonroot_raised_pE(struct cred
*new, const struct cred
*old
,
781 kuid_t root
, bool has_fcap
)
785 if ((__cap_grew(effective
, ambient
, new) &&
786 !(__cap_full(effective
, new) &&
787 (__is_eff(root
, new) || __is_real(root
, new)) &&
788 root_privileged())) ||
789 (root_privileged() &&
790 __is_suid(root
, new) &&
791 !__cap_full(effective
, new)) ||
792 (!__is_setuid(new, old
) &&
794 __cap_gained(permitted
, new, old
)) ||
795 __cap_gained(ambient
, new, old
))))
803 * cap_bprm_set_creds - Set up the proposed credentials for execve().
804 * @bprm: The execution parameters, including the proposed creds
806 * Set up the proposed credentials for a new execution context being
807 * constructed by execve(). The proposed creds in @bprm->cred is altered,
808 * which won't take effect immediately. Returns 0 if successful, -ve on error.
810 int cap_bprm_set_creds(struct linux_binprm
*bprm
)
812 const struct cred
*old
= current_cred();
813 struct cred
*new = bprm
->cred
;
814 bool effective
= false, has_fcap
= false, is_setid
;
818 if (WARN_ON(!cap_ambient_invariant_ok(old
)))
821 ret
= get_file_caps(bprm
, &effective
, &has_fcap
);
825 root_uid
= make_kuid(new->user_ns
, 0);
827 handle_privileged_root(bprm
, has_fcap
, &effective
, root_uid
);
829 /* if we have fs caps, clear dangerous personality flags */
830 if (__cap_gained(permitted
, new, old
))
831 bprm
->per_clear
|= PER_CLEAR_ON_SETID
;
833 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
834 * credentials unless they have the appropriate permit.
836 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
838 is_setid
= __is_setuid(new, old
) || __is_setgid(new, old
);
840 if ((is_setid
|| __cap_gained(permitted
, new, old
)) &&
841 ((bprm
->unsafe
& ~LSM_UNSAFE_PTRACE
) ||
842 !ptracer_capable(current
, new->user_ns
))) {
843 /* downgrade; they get no more than they had, and maybe less */
844 if (!ns_capable(new->user_ns
, CAP_SETUID
) ||
845 (bprm
->unsafe
& LSM_UNSAFE_NO_NEW_PRIVS
)) {
846 new->euid
= new->uid
;
847 new->egid
= new->gid
;
849 new->cap_permitted
= cap_intersect(new->cap_permitted
,
853 new->suid
= new->fsuid
= new->euid
;
854 new->sgid
= new->fsgid
= new->egid
;
856 /* File caps or setid cancels ambient. */
857 if (has_fcap
|| is_setid
)
858 cap_clear(new->cap_ambient
);
861 * Now that we've computed pA', update pP' to give:
862 * pP' = (X & fP) | (pI & fI) | pA'
864 new->cap_permitted
= cap_combine(new->cap_permitted
, new->cap_ambient
);
867 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
868 * this is the same as pE' = (fE ? pP' : 0) | pA'.
871 new->cap_effective
= new->cap_permitted
;
873 new->cap_effective
= new->cap_ambient
;
875 if (WARN_ON(!cap_ambient_invariant_ok(new)))
878 if (nonroot_raised_pE(new, old
, root_uid
, has_fcap
)) {
879 ret
= audit_log_bprm_fcaps(bprm
, new, old
);
884 new->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
886 if (WARN_ON(!cap_ambient_invariant_ok(new)))
889 /* Check for privilege-elevated exec. */
890 bprm
->cap_elevated
= 0;
892 (!__is_real(root_uid
, new) &&
894 __cap_grew(permitted
, ambient
, new))))
895 bprm
->cap_elevated
= 1;
901 * cap_inode_setxattr - Determine whether an xattr may be altered
902 * @dentry: The inode/dentry being altered
903 * @name: The name of the xattr to be changed
904 * @value: The value that the xattr will be changed to
905 * @size: The size of value
906 * @flags: The replacement flag
908 * Determine whether an xattr may be altered or set on an inode, returning 0 if
909 * permission is granted, -ve if denied.
911 * This is used to make sure security xattrs don't get updated or set by those
912 * who aren't privileged to do so.
914 int cap_inode_setxattr(struct dentry
*dentry
, const char *name
,
915 const void *value
, size_t size
, int flags
)
917 struct user_namespace
*user_ns
= dentry
->d_sb
->s_user_ns
;
919 /* Ignore non-security xattrs */
920 if (strncmp(name
, XATTR_SECURITY_PREFIX
,
921 sizeof(XATTR_SECURITY_PREFIX
) - 1) != 0)
925 * For XATTR_NAME_CAPS the check will be done in
926 * cap_convert_nscap(), called by setxattr()
928 if (strcmp(name
, XATTR_NAME_CAPS
) == 0)
931 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
937 * cap_inode_removexattr - Determine whether an xattr may be removed
938 * @dentry: The inode/dentry being altered
939 * @name: The name of the xattr to be changed
941 * Determine whether an xattr may be removed from an inode, returning 0 if
942 * permission is granted, -ve if denied.
944 * This is used to make sure security xattrs don't get removed by those who
945 * aren't privileged to remove them.
947 int cap_inode_removexattr(struct dentry
*dentry
, const char *name
)
949 struct user_namespace
*user_ns
= dentry
->d_sb
->s_user_ns
;
951 /* Ignore non-security xattrs */
952 if (strncmp(name
, XATTR_SECURITY_PREFIX
,
953 sizeof(XATTR_SECURITY_PREFIX
) - 1) != 0)
956 if (strcmp(name
, XATTR_NAME_CAPS
) == 0) {
957 /* security.capability gets namespaced */
958 struct inode
*inode
= d_backing_inode(dentry
);
961 if (!capable_wrt_inode_uidgid(inode
, CAP_SETFCAP
))
966 if (!ns_capable(user_ns
, CAP_SYS_ADMIN
))
972 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
973 * a process after a call to setuid, setreuid, or setresuid.
975 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
976 * {r,e,s}uid != 0, the permitted and effective capabilities are
979 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
980 * capabilities of the process are cleared.
982 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
983 * capabilities are set to the permitted capabilities.
985 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
990 * cevans - New behaviour, Oct '99
991 * A process may, via prctl(), elect to keep its capabilities when it
992 * calls setuid() and switches away from uid==0. Both permitted and
993 * effective sets will be retained.
994 * Without this change, it was impossible for a daemon to drop only some
995 * of its privilege. The call to setuid(!=0) would drop all privileges!
996 * Keeping uid 0 is not an option because uid 0 owns too many vital
998 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1000 static inline void cap_emulate_setxuid(struct cred
*new, const struct cred
*old
)
1002 kuid_t root_uid
= make_kuid(old
->user_ns
, 0);
1004 if ((uid_eq(old
->uid
, root_uid
) ||
1005 uid_eq(old
->euid
, root_uid
) ||
1006 uid_eq(old
->suid
, root_uid
)) &&
1007 (!uid_eq(new->uid
, root_uid
) &&
1008 !uid_eq(new->euid
, root_uid
) &&
1009 !uid_eq(new->suid
, root_uid
))) {
1010 if (!issecure(SECURE_KEEP_CAPS
)) {
1011 cap_clear(new->cap_permitted
);
1012 cap_clear(new->cap_effective
);
1016 * Pre-ambient programs expect setresuid to nonroot followed
1017 * by exec to drop capabilities. We should make sure that
1018 * this remains the case.
1020 cap_clear(new->cap_ambient
);
1022 if (uid_eq(old
->euid
, root_uid
) && !uid_eq(new->euid
, root_uid
))
1023 cap_clear(new->cap_effective
);
1024 if (!uid_eq(old
->euid
, root_uid
) && uid_eq(new->euid
, root_uid
))
1025 new->cap_effective
= new->cap_permitted
;
1029 * cap_task_fix_setuid - Fix up the results of setuid() call
1030 * @new: The proposed credentials
1031 * @old: The current task's current credentials
1032 * @flags: Indications of what has changed
1034 * Fix up the results of setuid() call before the credential changes are
1035 * actually applied, returning 0 to grant the changes, -ve to deny them.
1037 int cap_task_fix_setuid(struct cred
*new, const struct cred
*old
, int flags
)
1043 /* juggle the capabilities to follow [RES]UID changes unless
1044 * otherwise suppressed */
1045 if (!issecure(SECURE_NO_SETUID_FIXUP
))
1046 cap_emulate_setxuid(new, old
);
1050 /* juggle the capabilties to follow FSUID changes, unless
1051 * otherwise suppressed
1053 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1054 * if not, we might be a bit too harsh here.
1056 if (!issecure(SECURE_NO_SETUID_FIXUP
)) {
1057 kuid_t root_uid
= make_kuid(old
->user_ns
, 0);
1058 if (uid_eq(old
->fsuid
, root_uid
) && !uid_eq(new->fsuid
, root_uid
))
1059 new->cap_effective
=
1060 cap_drop_fs_set(new->cap_effective
);
1062 if (!uid_eq(old
->fsuid
, root_uid
) && uid_eq(new->fsuid
, root_uid
))
1063 new->cap_effective
=
1064 cap_raise_fs_set(new->cap_effective
,
1065 new->cap_permitted
);
1077 * Rationale: code calling task_setscheduler, task_setioprio, and
1078 * task_setnice, assumes that
1079 * . if capable(cap_sys_nice), then those actions should be allowed
1080 * . if not capable(cap_sys_nice), but acting on your own processes,
1081 * then those actions should be allowed
1082 * This is insufficient now since you can call code without suid, but
1083 * yet with increased caps.
1084 * So we check for increased caps on the target process.
1086 static int cap_safe_nice(struct task_struct
*p
)
1088 int is_subset
, ret
= 0;
1091 is_subset
= cap_issubset(__task_cred(p
)->cap_permitted
,
1092 current_cred()->cap_permitted
);
1093 if (!is_subset
&& !ns_capable(__task_cred(p
)->user_ns
, CAP_SYS_NICE
))
1101 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1102 * @p: The task to affect
1104 * Detemine if the requested scheduler policy change is permitted for the
1105 * specified task, returning 0 if permission is granted, -ve if denied.
1107 int cap_task_setscheduler(struct task_struct
*p
)
1109 return cap_safe_nice(p
);
1113 * cap_task_ioprio - Detemine if I/O priority change is permitted
1114 * @p: The task to affect
1115 * @ioprio: The I/O priority to set
1117 * Detemine if the requested I/O priority change is permitted for the specified
1118 * task, returning 0 if permission is granted, -ve if denied.
1120 int cap_task_setioprio(struct task_struct
*p
, int ioprio
)
1122 return cap_safe_nice(p
);
1126 * cap_task_ioprio - Detemine if task priority change is permitted
1127 * @p: The task to affect
1128 * @nice: The nice value to set
1130 * Detemine if the requested task priority change is permitted for the
1131 * specified task, returning 0 if permission is granted, -ve if denied.
1133 int cap_task_setnice(struct task_struct
*p
, int nice
)
1135 return cap_safe_nice(p
);
1139 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1140 * the current task's bounding set. Returns 0 on success, -ve on error.
1142 static int cap_prctl_drop(unsigned long cap
)
1146 if (!ns_capable(current_user_ns(), CAP_SETPCAP
))
1148 if (!cap_valid(cap
))
1151 new = prepare_creds();
1154 cap_lower(new->cap_bset
, cap
);
1155 return commit_creds(new);
1159 * cap_task_prctl - Implement process control functions for this security module
1160 * @option: The process control function requested
1161 * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1163 * Allow process control functions (sys_prctl()) to alter capabilities; may
1164 * also deny access to other functions not otherwise implemented here.
1166 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1167 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1168 * modules will consider performing the function.
1170 int cap_task_prctl(int option
, unsigned long arg2
, unsigned long arg3
,
1171 unsigned long arg4
, unsigned long arg5
)
1173 const struct cred
*old
= current_cred();
1177 case PR_CAPBSET_READ
:
1178 if (!cap_valid(arg2
))
1180 return !!cap_raised(old
->cap_bset
, arg2
);
1182 case PR_CAPBSET_DROP
:
1183 return cap_prctl_drop(arg2
);
1186 * The next four prctl's remain to assist with transitioning a
1187 * system from legacy UID=0 based privilege (when filesystem
1188 * capabilities are not in use) to a system using filesystem
1189 * capabilities only - as the POSIX.1e draft intended.
1193 * PR_SET_SECUREBITS =
1194 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1195 * | issecure_mask(SECURE_NOROOT)
1196 * | issecure_mask(SECURE_NOROOT_LOCKED)
1197 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1198 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1200 * will ensure that the current process and all of its
1201 * children will be locked into a pure
1202 * capability-based-privilege environment.
1204 case PR_SET_SECUREBITS
:
1205 if ((((old
->securebits
& SECURE_ALL_LOCKS
) >> 1)
1206 & (old
->securebits
^ arg2
)) /*[1]*/
1207 || ((old
->securebits
& SECURE_ALL_LOCKS
& ~arg2
)) /*[2]*/
1208 || (arg2
& ~(SECURE_ALL_LOCKS
| SECURE_ALL_BITS
)) /*[3]*/
1209 || (cap_capable(current_cred(),
1210 current_cred()->user_ns
,
1212 CAP_OPT_NONE
) != 0) /*[4]*/
1214 * [1] no changing of bits that are locked
1215 * [2] no unlocking of locks
1216 * [3] no setting of unsupported bits
1217 * [4] doing anything requires privilege (go read about
1218 * the "sendmail capabilities bug")
1221 /* cannot change a locked bit */
1224 new = prepare_creds();
1227 new->securebits
= arg2
;
1228 return commit_creds(new);
1230 case PR_GET_SECUREBITS
:
1231 return old
->securebits
;
1233 case PR_GET_KEEPCAPS
:
1234 return !!issecure(SECURE_KEEP_CAPS
);
1236 case PR_SET_KEEPCAPS
:
1237 if (arg2
> 1) /* Note, we rely on arg2 being unsigned here */
1239 if (issecure(SECURE_KEEP_CAPS_LOCKED
))
1242 new = prepare_creds();
1246 new->securebits
|= issecure_mask(SECURE_KEEP_CAPS
);
1248 new->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
1249 return commit_creds(new);
1251 case PR_CAP_AMBIENT
:
1252 if (arg2
== PR_CAP_AMBIENT_CLEAR_ALL
) {
1253 if (arg3
| arg4
| arg5
)
1256 new = prepare_creds();
1259 cap_clear(new->cap_ambient
);
1260 return commit_creds(new);
1263 if (((!cap_valid(arg3
)) | arg4
| arg5
))
1266 if (arg2
== PR_CAP_AMBIENT_IS_SET
) {
1267 return !!cap_raised(current_cred()->cap_ambient
, arg3
);
1268 } else if (arg2
!= PR_CAP_AMBIENT_RAISE
&&
1269 arg2
!= PR_CAP_AMBIENT_LOWER
) {
1272 if (arg2
== PR_CAP_AMBIENT_RAISE
&&
1273 (!cap_raised(current_cred()->cap_permitted
, arg3
) ||
1274 !cap_raised(current_cred()->cap_inheritable
,
1276 issecure(SECURE_NO_CAP_AMBIENT_RAISE
)))
1279 new = prepare_creds();
1282 if (arg2
== PR_CAP_AMBIENT_RAISE
)
1283 cap_raise(new->cap_ambient
, arg3
);
1285 cap_lower(new->cap_ambient
, arg3
);
1286 return commit_creds(new);
1290 /* No functionality available - continue with default */
1296 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1297 * @mm: The VM space in which the new mapping is to be made
1298 * @pages: The size of the mapping
1300 * Determine whether the allocation of a new virtual mapping by the current
1301 * task is permitted, returning 1 if permission is granted, 0 if not.
1303 int cap_vm_enough_memory(struct mm_struct
*mm
, long pages
)
1305 int cap_sys_admin
= 0;
1307 if (cap_capable(current_cred(), &init_user_ns
,
1308 CAP_SYS_ADMIN
, CAP_OPT_NOAUDIT
) == 0)
1311 return cap_sys_admin
;
1315 * cap_mmap_addr - check if able to map given addr
1316 * @addr: address attempting to be mapped
1318 * If the process is attempting to map memory below dac_mmap_min_addr they need
1319 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1320 * capability security module. Returns 0 if this mapping should be allowed
1323 int cap_mmap_addr(unsigned long addr
)
1327 if (addr
< dac_mmap_min_addr
) {
1328 ret
= cap_capable(current_cred(), &init_user_ns
, CAP_SYS_RAWIO
,
1330 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1332 current
->flags
|= PF_SUPERPRIV
;
1336 EXPORT_SYMBOL_GPL(cap_mmap_addr
);
1338 int cap_mmap_file(struct file
*file
, unsigned long reqprot
,
1339 unsigned long prot
, unsigned long flags
)
1343 EXPORT_SYMBOL_GPL(cap_mmap_file
);
1345 #ifdef CONFIG_SECURITY
1347 struct security_hook_list capability_hooks
[] __lsm_ro_after_init
= {
1348 LSM_HOOK_INIT(capable
, cap_capable
),
1349 LSM_HOOK_INIT(settime
, cap_settime
),
1350 LSM_HOOK_INIT(ptrace_access_check
, cap_ptrace_access_check
),
1351 LSM_HOOK_INIT(ptrace_traceme
, cap_ptrace_traceme
),
1352 LSM_HOOK_INIT(capget
, cap_capget
),
1353 LSM_HOOK_INIT(capset
, cap_capset
),
1354 LSM_HOOK_INIT(bprm_set_creds
, cap_bprm_set_creds
),
1355 LSM_HOOK_INIT(inode_need_killpriv
, cap_inode_need_killpriv
),
1356 LSM_HOOK_INIT(inode_killpriv
, cap_inode_killpriv
),
1357 LSM_HOOK_INIT(inode_getsecurity
, cap_inode_getsecurity
),
1358 LSM_HOOK_INIT(mmap_addr
, cap_mmap_addr
),
1359 LSM_HOOK_INIT(mmap_file
, cap_mmap_file
),
1360 LSM_HOOK_INIT(task_fix_setuid
, cap_task_fix_setuid
),
1361 LSM_HOOK_INIT(task_prctl
, cap_task_prctl
),
1362 LSM_HOOK_INIT(task_setscheduler
, cap_task_setscheduler
),
1363 LSM_HOOK_INIT(task_setioprio
, cap_task_setioprio
),
1364 LSM_HOOK_INIT(task_setnice
, cap_task_setnice
),
1365 LSM_HOOK_INIT(vm_enough_memory
, cap_vm_enough_memory
),
1368 static int __init
capability_init(void)
1370 security_add_hooks(capability_hooks
, ARRAY_SIZE(capability_hooks
),
1375 DEFINE_LSM(capability
) = {
1376 .name
= "capability",
1377 .order
= LSM_ORDER_FIRST
,
1378 .init
= capability_init
,
1381 #endif /* CONFIG_SECURITY */