1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Common capabilities, needed by capability.o.
5 #include <linux/capability.h>
6 #include <linux/audit.h>
7 #include <linux/init.h>
8 #include <linux/kernel.h>
9 #include <linux/lsm_hooks.h>
10 #include <linux/file.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/skbuff.h>
16 #include <linux/netlink.h>
17 #include <linux/ptrace.h>
18 #include <linux/xattr.h>
19 #include <linux/hugetlb.h>
20 #include <linux/mount.h>
21 #include <linux/sched.h>
22 #include <linux/prctl.h>
23 #include <linux/securebits.h>
24 #include <linux/user_namespace.h>
25 #include <linux/binfmts.h>
26 #include <linux/personality.h>
27 #include <linux/mnt_idmapping.h>
30 * If a non-root user executes a setuid-root binary in
31 * !secure(SECURE_NOROOT) mode, then we raise capabilities.
32 * However if fE is also set, then the intent is for only
33 * the file capabilities to be applied, and the setuid-root
34 * bit is left on either to change the uid (plausible) or
35 * to get full privilege on a kernel without file capabilities
36 * support. So in that case we do not raise capabilities.
38 * Warn if that happens, once per boot.
40 static void warn_setuid_and_fcaps_mixed(const char *fname)
44 printk(KERN_INFO "warning: `%s' has both setuid-root and"
45 " effective capabilities. Therefore not raising all"
46 " capabilities.\n", fname);
52 * cap_capable - Determine whether a task has a particular effective capability
53 * @cred: The credentials to use
54 * @targ_ns: The user namespace in which we need the capability
55 * @cap: The capability to check for
56 * @opts: Bitmask of options defined in include/linux/security.h
58 * Determine whether the nominated task has the specified capability amongst
59 * its effective set, returning 0 if it does, -ve if it does not.
61 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
62 * and has_capability() functions. That is, it has the reverse semantics:
63 * cap_has_capability() returns 0 when a task has a capability, but the
64 * kernel's capable() and has_capability() returns 1 for this case.
66 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
67 int cap, unsigned int opts)
69 struct user_namespace *ns = targ_ns;
71 /* See if cred has the capability in the target user namespace
72 * by examining the target user namespace and all of the target
73 * user namespace's parents.
76 /* Do we have the necessary capabilities? */
77 if (ns == cred->user_ns)
78 return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
81 * If we're already at a lower level than we're looking for,
82 * we're done searching.
84 if (ns->level <= cred->user_ns->level)
88 * The owner of the user namespace in the parent of the
89 * user namespace has all caps.
91 if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
95 * If you have a capability in a parent user ns, then you have
96 * it over all children user namespaces as well.
101 /* We never get here */
105 * cap_settime - Determine whether the current process may set the system clock
106 * @ts: The time to set
107 * @tz: The timezone to set
109 * Determine whether the current process may set the system clock and timezone
110 * information, returning 0 if permission granted, -ve if denied.
112 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
114 if (!capable(CAP_SYS_TIME))
120 * cap_ptrace_access_check - Determine whether the current process may access
122 * @child: The process to be accessed
123 * @mode: The mode of attachment.
125 * If we are in the same or an ancestor user_ns and have all the target
126 * task's capabilities, then ptrace access is allowed.
127 * If we have the ptrace capability to the target user_ns, then ptrace
131 * Determine whether a process may access another, returning 0 if permission
132 * granted, -ve if denied.
134 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
137 const struct cred *cred, *child_cred;
138 const kernel_cap_t *caller_caps;
141 cred = current_cred();
142 child_cred = __task_cred(child);
143 if (mode & PTRACE_MODE_FSCREDS)
144 caller_caps = &cred->cap_effective;
146 caller_caps = &cred->cap_permitted;
147 if (cred->user_ns == child_cred->user_ns &&
148 cap_issubset(child_cred->cap_permitted, *caller_caps))
150 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
159 * cap_ptrace_traceme - Determine whether another process may trace the current
160 * @parent: The task proposed to be the tracer
162 * If parent is in the same or an ancestor user_ns and has all current's
163 * capabilities, then ptrace access is allowed.
164 * If parent has the ptrace capability to current's user_ns, then ptrace
168 * Determine whether the nominated task is permitted to trace the current
169 * process, returning 0 if permission is granted, -ve if denied.
171 int cap_ptrace_traceme(struct task_struct *parent)
174 const struct cred *cred, *child_cred;
177 cred = __task_cred(parent);
178 child_cred = current_cred();
179 if (cred->user_ns == child_cred->user_ns &&
180 cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
182 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
191 * cap_capget - Retrieve a task's capability sets
192 * @target: The task from which to retrieve the capability sets
193 * @effective: The place to record the effective set
194 * @inheritable: The place to record the inheritable set
195 * @permitted: The place to record the permitted set
197 * This function retrieves the capabilities of the nominated task and returns
198 * them to the caller.
200 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
201 kernel_cap_t *inheritable, kernel_cap_t *permitted)
203 const struct cred *cred;
205 /* Derived from kernel/capability.c:sys_capget. */
207 cred = __task_cred(target);
208 *effective = cred->cap_effective;
209 *inheritable = cred->cap_inheritable;
210 *permitted = cred->cap_permitted;
216 * Determine whether the inheritable capabilities are limited to the old
217 * permitted set. Returns 1 if they are limited, 0 if they are not.
219 static inline int cap_inh_is_capped(void)
221 /* they are so limited unless the current task has the CAP_SETPCAP
224 if (cap_capable(current_cred(), current_cred()->user_ns,
225 CAP_SETPCAP, CAP_OPT_NONE) == 0)
231 * cap_capset - Validate and apply proposed changes to current's capabilities
232 * @new: The proposed new credentials; alterations should be made here
233 * @old: The current task's current credentials
234 * @effective: A pointer to the proposed new effective capabilities set
235 * @inheritable: A pointer to the proposed new inheritable capabilities set
236 * @permitted: A pointer to the proposed new permitted capabilities set
238 * This function validates and applies a proposed mass change to the current
239 * process's capability sets. The changes are made to the proposed new
240 * credentials, and assuming no error, will be committed by the caller of LSM.
242 int cap_capset(struct cred *new,
243 const struct cred *old,
244 const kernel_cap_t *effective,
245 const kernel_cap_t *inheritable,
246 const kernel_cap_t *permitted)
248 if (cap_inh_is_capped() &&
249 !cap_issubset(*inheritable,
250 cap_combine(old->cap_inheritable,
251 old->cap_permitted)))
252 /* incapable of using this inheritable set */
255 if (!cap_issubset(*inheritable,
256 cap_combine(old->cap_inheritable,
258 /* no new pI capabilities outside bounding set */
261 /* verify restrictions on target's new Permitted set */
262 if (!cap_issubset(*permitted, old->cap_permitted))
265 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
266 if (!cap_issubset(*effective, *permitted))
269 new->cap_effective = *effective;
270 new->cap_inheritable = *inheritable;
271 new->cap_permitted = *permitted;
274 * Mask off ambient bits that are no longer both permitted and
277 new->cap_ambient = cap_intersect(new->cap_ambient,
278 cap_intersect(*permitted,
280 if (WARN_ON(!cap_ambient_invariant_ok(new)))
286 * cap_inode_need_killpriv - Determine if inode change affects privileges
287 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
289 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
290 * affects the security markings on that inode, and if it is, should
291 * inode_killpriv() be invoked or the change rejected.
293 * Return: 1 if security.capability has a value, meaning inode_killpriv()
294 * is required, 0 otherwise, meaning inode_killpriv() is not required.
296 int cap_inode_need_killpriv(struct dentry *dentry)
298 struct inode *inode = d_backing_inode(dentry);
301 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
306 * cap_inode_killpriv - Erase the security markings on an inode
308 * @mnt_userns: user namespace of the mount the inode was found from
309 * @dentry: The inode/dentry to alter
311 * Erase the privilege-enhancing security markings on an inode.
313 * If the inode has been found through an idmapped mount the user namespace of
314 * the vfsmount must be passed through @mnt_userns. This function will then
315 * take care to map the inode according to @mnt_userns before checking
316 * permissions. On non-idmapped mounts or if permission checking is to be
317 * performed on the raw inode simply passs init_user_ns.
319 * Return: 0 if successful, -ve on error.
321 int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry)
325 error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS);
326 if (error == -EOPNOTSUPP)
331 static bool rootid_owns_currentns(kuid_t kroot)
333 struct user_namespace *ns;
335 if (!uid_valid(kroot))
338 for (ns = current_user_ns(); ; ns = ns->parent) {
339 if (from_kuid(ns, kroot) == 0)
341 if (ns == &init_user_ns)
348 static __u32 sansflags(__u32 m)
350 return m & ~VFS_CAP_FLAGS_EFFECTIVE;
353 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
355 if (size != XATTR_CAPS_SZ_2)
357 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
360 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
362 if (size != XATTR_CAPS_SZ_3)
364 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
368 * getsecurity: We are called for security.* before any attempt to read the
369 * xattr from the inode itself.
371 * This gives us a chance to read the on-disk value and convert it. If we
372 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
374 * Note we are not called by vfs_getxattr_alloc(), but that is only called
375 * by the integrity subsystem, which really wants the unconverted values -
378 int cap_inode_getsecurity(struct user_namespace *mnt_userns,
379 struct inode *inode, const char *name, void **buffer,
385 uid_t root, mappedroot;
387 struct vfs_cap_data *cap;
388 struct vfs_ns_cap_data *nscap = NULL;
389 struct dentry *dentry;
390 struct user_namespace *fs_ns;
392 if (strcmp(name, "capability") != 0)
395 dentry = d_find_any_alias(inode);
399 size = sizeof(struct vfs_ns_cap_data);
400 ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
401 &tmpbuf, size, GFP_NOFS);
404 if (ret < 0 || !tmpbuf)
407 fs_ns = inode->i_sb->s_user_ns;
408 cap = (struct vfs_cap_data *) tmpbuf;
409 if (is_v2header((size_t) ret, cap)) {
411 } else if (is_v3header((size_t) ret, cap)) {
412 nscap = (struct vfs_ns_cap_data *) tmpbuf;
413 root = le32_to_cpu(nscap->rootid);
419 kroot = make_kuid(fs_ns, root);
421 /* If this is an idmapped mount shift the kuid. */
422 kroot = kuid_into_mnt(mnt_userns, kroot);
424 /* If the root kuid maps to a valid uid in current ns, then return
425 * this as a nscap. */
426 mappedroot = from_kuid(current_user_ns(), kroot);
427 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
428 size = sizeof(struct vfs_ns_cap_data);
431 /* v2 -> v3 conversion */
432 nscap = kzalloc(size, GFP_ATOMIC);
437 nsmagic = VFS_CAP_REVISION_3;
438 magic = le32_to_cpu(cap->magic_etc);
439 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
440 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
441 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
442 nscap->magic_etc = cpu_to_le32(nsmagic);
444 /* use allocated v3 buffer */
447 nscap->rootid = cpu_to_le32(mappedroot);
453 if (!rootid_owns_currentns(kroot)) {
458 /* This comes from a parent namespace. Return as a v2 capability */
459 size = sizeof(struct vfs_cap_data);
462 /* v3 -> v2 conversion */
463 cap = kzalloc(size, GFP_ATOMIC);
468 magic = VFS_CAP_REVISION_2;
469 nsmagic = le32_to_cpu(nscap->magic_etc);
470 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
471 magic |= VFS_CAP_FLAGS_EFFECTIVE;
472 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
473 cap->magic_etc = cpu_to_le32(magic);
475 /* use unconverted v2 */
486 * rootid_from_xattr - translate root uid of vfs caps
488 * @value: vfs caps value which may be modified by this function
489 * @size: size of @ivalue
490 * @task_ns: user namespace of the caller
491 * @mnt_userns: user namespace of the mount the inode was found from
493 * If the inode has been found through an idmapped mount the user namespace of
494 * the vfsmount must be passed through @mnt_userns. This function will then
495 * take care to map the inode according to @mnt_userns before checking
496 * permissions. On non-idmapped mounts or if permission checking is to be
497 * performed on the raw inode simply passs init_user_ns.
499 static kuid_t rootid_from_xattr(const void *value, size_t size,
500 struct user_namespace *task_ns,
501 struct user_namespace *mnt_userns)
503 const struct vfs_ns_cap_data *nscap = value;
507 if (size == XATTR_CAPS_SZ_3)
508 rootid = le32_to_cpu(nscap->rootid);
510 rootkid = make_kuid(task_ns, rootid);
511 return kuid_from_mnt(mnt_userns, rootkid);
514 static bool validheader(size_t size, const struct vfs_cap_data *cap)
516 return is_v2header(size, cap) || is_v3header(size, cap);
520 * cap_convert_nscap - check vfs caps
522 * @mnt_userns: user namespace of the mount the inode was found from
523 * @dentry: used to retrieve inode to check permissions on
524 * @ivalue: vfs caps value which may be modified by this function
525 * @size: size of @ivalue
527 * User requested a write of security.capability. If needed, update the
528 * xattr to change from v2 to v3, or to fixup the v3 rootid.
530 * If the inode has been found through an idmapped mount the user namespace of
531 * the vfsmount must be passed through @mnt_userns. This function will then
532 * take care to map the inode according to @mnt_userns before checking
533 * permissions. On non-idmapped mounts or if permission checking is to be
534 * performed on the raw inode simply passs init_user_ns.
536 * Return: On success, return the new size; on error, return < 0.
538 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
539 const void **ivalue, size_t size)
541 struct vfs_ns_cap_data *nscap;
543 const struct vfs_cap_data *cap = *ivalue;
544 __u32 magic, nsmagic;
545 struct inode *inode = d_backing_inode(dentry);
546 struct user_namespace *task_ns = current_user_ns(),
547 *fs_ns = inode->i_sb->s_user_ns;
553 if (!validheader(size, cap))
555 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
557 if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
558 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
559 /* user is privileged, just write the v2 */
562 rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
563 if (!uid_valid(rootid))
566 nsrootid = from_kuid(fs_ns, rootid);
570 newsize = sizeof(struct vfs_ns_cap_data);
571 nscap = kmalloc(newsize, GFP_ATOMIC);
574 nscap->rootid = cpu_to_le32(nsrootid);
575 nsmagic = VFS_CAP_REVISION_3;
576 magic = le32_to_cpu(cap->magic_etc);
577 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
578 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
579 nscap->magic_etc = cpu_to_le32(nsmagic);
580 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
587 * Calculate the new process capability sets from the capability sets attached
590 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
591 struct linux_binprm *bprm,
595 struct cred *new = bprm->cred;
599 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
602 if (caps->magic_etc & VFS_CAP_REVISION_MASK)
605 CAP_FOR_EACH_U32(i) {
606 __u32 permitted = caps->permitted.cap[i];
607 __u32 inheritable = caps->inheritable.cap[i];
610 * pP' = (X & fP) | (pI & fI)
611 * The addition of pA' is handled later.
613 new->cap_permitted.cap[i] =
614 (new->cap_bset.cap[i] & permitted) |
615 (new->cap_inheritable.cap[i] & inheritable);
617 if (permitted & ~new->cap_permitted.cap[i])
618 /* insufficient to execute correctly */
623 * For legacy apps, with no internal support for recognizing they
624 * do not have enough capabilities, we return an error if they are
625 * missing some "forced" (aka file-permitted) capabilities.
627 return *effective ? ret : 0;
631 * get_vfs_caps_from_disk - retrieve vfs caps from disk
633 * @mnt_userns: user namespace of the mount the inode was found from
634 * @dentry: dentry from which @inode is retrieved
635 * @cpu_caps: vfs capabilities
637 * Extract the on-exec-apply capability sets for an executable file.
639 * If the inode has been found through an idmapped mount the user namespace of
640 * the vfsmount must be passed through @mnt_userns. This function will then
641 * take care to map the inode according to @mnt_userns before checking
642 * permissions. On non-idmapped mounts or if permission checking is to be
643 * performed on the raw inode simply passs init_user_ns.
645 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
646 const struct dentry *dentry,
647 struct cpu_vfs_cap_data *cpu_caps)
649 struct inode *inode = d_backing_inode(dentry);
653 struct vfs_ns_cap_data data, *nscaps = &data;
654 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
656 struct user_namespace *fs_ns;
658 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
663 fs_ns = inode->i_sb->s_user_ns;
664 size = __vfs_getxattr((struct dentry *)dentry, inode,
665 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
666 if (size == -ENODATA || size == -EOPNOTSUPP)
667 /* no data, that's ok */
673 if (size < sizeof(magic_etc))
676 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
678 rootkuid = make_kuid(fs_ns, 0);
679 switch (magic_etc & VFS_CAP_REVISION_MASK) {
680 case VFS_CAP_REVISION_1:
681 if (size != XATTR_CAPS_SZ_1)
683 tocopy = VFS_CAP_U32_1;
685 case VFS_CAP_REVISION_2:
686 if (size != XATTR_CAPS_SZ_2)
688 tocopy = VFS_CAP_U32_2;
690 case VFS_CAP_REVISION_3:
691 if (size != XATTR_CAPS_SZ_3)
693 tocopy = VFS_CAP_U32_3;
694 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
700 /* Limit the caps to the mounter of the filesystem
701 * or the more limited uid specified in the xattr.
703 rootkuid = kuid_into_mnt(mnt_userns, rootkuid);
704 if (!rootid_owns_currentns(rootkuid))
707 CAP_FOR_EACH_U32(i) {
710 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
711 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
714 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
715 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
717 cpu_caps->rootid = rootkuid;
723 * Attempt to get the on-exec apply capability sets for an executable file from
724 * its xattrs and, if present, apply them to the proposed credentials being
725 * constructed by execve().
727 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
728 bool *effective, bool *has_fcap)
731 struct cpu_vfs_cap_data vcaps;
733 cap_clear(bprm->cred->cap_permitted);
735 if (!file_caps_enabled)
738 if (!mnt_may_suid(file->f_path.mnt))
742 * This check is redundant with mnt_may_suid() but is kept to make
743 * explicit that capability bits are limited to s_user_ns and its
746 if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
749 rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
750 file->f_path.dentry, &vcaps);
753 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
755 else if (rc == -ENODATA)
760 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
764 cap_clear(bprm->cred->cap_permitted);
769 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
771 static inline bool __is_real(kuid_t uid, struct cred *cred)
772 { return uid_eq(cred->uid, uid); }
774 static inline bool __is_eff(kuid_t uid, struct cred *cred)
775 { return uid_eq(cred->euid, uid); }
777 static inline bool __is_suid(kuid_t uid, struct cred *cred)
778 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
781 * handle_privileged_root - Handle case of privileged root
782 * @bprm: The execution parameters, including the proposed creds
783 * @has_fcap: Are any file capabilities set?
784 * @effective: Do we have effective root privilege?
785 * @root_uid: This namespace' root UID WRT initial USER namespace
787 * Handle the case where root is privileged and hasn't been neutered by
788 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
789 * set UID root and nothing is changed. If we are root, cap_permitted is
790 * updated. If we have become set UID root, the effective bit is set.
792 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
793 bool *effective, kuid_t root_uid)
795 const struct cred *old = current_cred();
796 struct cred *new = bprm->cred;
798 if (!root_privileged())
801 * If the legacy file capability is set, then don't set privs
802 * for a setuid root binary run by a non-root user. Do set it
803 * for a root user just to cause least surprise to an admin.
805 if (has_fcap && __is_suid(root_uid, new)) {
806 warn_setuid_and_fcaps_mixed(bprm->filename);
810 * To support inheritance of root-permissions and suid-root
811 * executables under compatibility mode, we override the
812 * capability sets for the file.
814 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
815 /* pP' = (cap_bset & ~0) | (pI & ~0) */
816 new->cap_permitted = cap_combine(old->cap_bset,
817 old->cap_inheritable);
820 * If only the real uid is 0, we do not set the effective bit.
822 if (__is_eff(root_uid, new))
826 #define __cap_gained(field, target, source) \
827 !cap_issubset(target->cap_##field, source->cap_##field)
828 #define __cap_grew(target, source, cred) \
829 !cap_issubset(cred->cap_##target, cred->cap_##source)
830 #define __cap_full(field, cred) \
831 cap_issubset(CAP_FULL_SET, cred->cap_##field)
833 static inline bool __is_setuid(struct cred *new, const struct cred *old)
834 { return !uid_eq(new->euid, old->uid); }
836 static inline bool __is_setgid(struct cred *new, const struct cred *old)
837 { return !gid_eq(new->egid, old->gid); }
840 * 1) Audit candidate if current->cap_effective is set
842 * We do not bother to audit if 3 things are true:
843 * 1) cap_effective has all caps
844 * 2) we became root *OR* are were already root
845 * 3) root is supposed to have all caps (SECURE_NOROOT)
846 * Since this is just a normal root execing a process.
848 * Number 1 above might fail if you don't have a full bset, but I think
849 * that is interesting information to audit.
851 * A number of other conditions require logging:
852 * 2) something prevented setuid root getting all caps
853 * 3) non-setuid root gets fcaps
854 * 4) non-setuid root gets ambient
856 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
857 kuid_t root, bool has_fcap)
861 if ((__cap_grew(effective, ambient, new) &&
862 !(__cap_full(effective, new) &&
863 (__is_eff(root, new) || __is_real(root, new)) &&
864 root_privileged())) ||
865 (root_privileged() &&
866 __is_suid(root, new) &&
867 !__cap_full(effective, new)) ||
868 (!__is_setuid(new, old) &&
870 __cap_gained(permitted, new, old)) ||
871 __cap_gained(ambient, new, old))))
879 * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
880 * @bprm: The execution parameters, including the proposed creds
881 * @file: The file to pull the credentials from
883 * Set up the proposed credentials for a new execution context being
884 * constructed by execve(). The proposed creds in @bprm->cred is altered,
885 * which won't take effect immediately.
887 * Return: 0 if successful, -ve on error.
889 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
891 /* Process setpcap binaries and capabilities for uid 0 */
892 const struct cred *old = current_cred();
893 struct cred *new = bprm->cred;
894 bool effective = false, has_fcap = false, is_setid;
898 if (WARN_ON(!cap_ambient_invariant_ok(old)))
901 ret = get_file_caps(bprm, file, &effective, &has_fcap);
905 root_uid = make_kuid(new->user_ns, 0);
907 handle_privileged_root(bprm, has_fcap, &effective, root_uid);
909 /* if we have fs caps, clear dangerous personality flags */
910 if (__cap_gained(permitted, new, old))
911 bprm->per_clear |= PER_CLEAR_ON_SETID;
913 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
914 * credentials unless they have the appropriate permit.
916 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
918 is_setid = __is_setuid(new, old) || __is_setgid(new, old);
920 if ((is_setid || __cap_gained(permitted, new, old)) &&
921 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
922 !ptracer_capable(current, new->user_ns))) {
923 /* downgrade; they get no more than they had, and maybe less */
924 if (!ns_capable(new->user_ns, CAP_SETUID) ||
925 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
926 new->euid = new->uid;
927 new->egid = new->gid;
929 new->cap_permitted = cap_intersect(new->cap_permitted,
933 new->suid = new->fsuid = new->euid;
934 new->sgid = new->fsgid = new->egid;
936 /* File caps or setid cancels ambient. */
937 if (has_fcap || is_setid)
938 cap_clear(new->cap_ambient);
941 * Now that we've computed pA', update pP' to give:
942 * pP' = (X & fP) | (pI & fI) | pA'
944 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
947 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
948 * this is the same as pE' = (fE ? pP' : 0) | pA'.
951 new->cap_effective = new->cap_permitted;
953 new->cap_effective = new->cap_ambient;
955 if (WARN_ON(!cap_ambient_invariant_ok(new)))
958 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
959 ret = audit_log_bprm_fcaps(bprm, new, old);
964 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
966 if (WARN_ON(!cap_ambient_invariant_ok(new)))
969 /* Check for privilege-elevated exec. */
971 (!__is_real(root_uid, new) &&
973 __cap_grew(permitted, ambient, new))))
974 bprm->secureexec = 1;
980 * cap_inode_setxattr - Determine whether an xattr may be altered
981 * @dentry: The inode/dentry being altered
982 * @name: The name of the xattr to be changed
983 * @value: The value that the xattr will be changed to
984 * @size: The size of value
985 * @flags: The replacement flag
987 * Determine whether an xattr may be altered or set on an inode, returning 0 if
988 * permission is granted, -ve if denied.
990 * This is used to make sure security xattrs don't get updated or set by those
991 * who aren't privileged to do so.
993 int cap_inode_setxattr(struct dentry *dentry, const char *name,
994 const void *value, size_t size, int flags)
996 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
998 /* Ignore non-security xattrs */
999 if (strncmp(name, XATTR_SECURITY_PREFIX,
1000 XATTR_SECURITY_PREFIX_LEN) != 0)
1004 * For XATTR_NAME_CAPS the check will be done in
1005 * cap_convert_nscap(), called by setxattr()
1007 if (strcmp(name, XATTR_NAME_CAPS) == 0)
1010 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1016 * cap_inode_removexattr - Determine whether an xattr may be removed
1018 * @mnt_userns: User namespace of the mount the inode was found from
1019 * @dentry: The inode/dentry being altered
1020 * @name: The name of the xattr to be changed
1022 * Determine whether an xattr may be removed from an inode, returning 0 if
1023 * permission is granted, -ve if denied.
1025 * If the inode has been found through an idmapped mount the user namespace of
1026 * the vfsmount must be passed through @mnt_userns. This function will then
1027 * take care to map the inode according to @mnt_userns before checking
1028 * permissions. On non-idmapped mounts or if permission checking is to be
1029 * performed on the raw inode simply passs init_user_ns.
1031 * This is used to make sure security xattrs don't get removed by those who
1032 * aren't privileged to remove them.
1034 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1035 struct dentry *dentry, const char *name)
1037 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1039 /* Ignore non-security xattrs */
1040 if (strncmp(name, XATTR_SECURITY_PREFIX,
1041 XATTR_SECURITY_PREFIX_LEN) != 0)
1044 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1045 /* security.capability gets namespaced */
1046 struct inode *inode = d_backing_inode(dentry);
1049 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1054 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1060 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1061 * a process after a call to setuid, setreuid, or setresuid.
1063 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1064 * {r,e,s}uid != 0, the permitted and effective capabilities are
1067 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1068 * capabilities of the process are cleared.
1070 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1071 * capabilities are set to the permitted capabilities.
1073 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1078 * cevans - New behaviour, Oct '99
1079 * A process may, via prctl(), elect to keep its capabilities when it
1080 * calls setuid() and switches away from uid==0. Both permitted and
1081 * effective sets will be retained.
1082 * Without this change, it was impossible for a daemon to drop only some
1083 * of its privilege. The call to setuid(!=0) would drop all privileges!
1084 * Keeping uid 0 is not an option because uid 0 owns too many vital
1086 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1088 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1090 kuid_t root_uid = make_kuid(old->user_ns, 0);
1092 if ((uid_eq(old->uid, root_uid) ||
1093 uid_eq(old->euid, root_uid) ||
1094 uid_eq(old->suid, root_uid)) &&
1095 (!uid_eq(new->uid, root_uid) &&
1096 !uid_eq(new->euid, root_uid) &&
1097 !uid_eq(new->suid, root_uid))) {
1098 if (!issecure(SECURE_KEEP_CAPS)) {
1099 cap_clear(new->cap_permitted);
1100 cap_clear(new->cap_effective);
1104 * Pre-ambient programs expect setresuid to nonroot followed
1105 * by exec to drop capabilities. We should make sure that
1106 * this remains the case.
1108 cap_clear(new->cap_ambient);
1110 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1111 cap_clear(new->cap_effective);
1112 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1113 new->cap_effective = new->cap_permitted;
1117 * cap_task_fix_setuid - Fix up the results of setuid() call
1118 * @new: The proposed credentials
1119 * @old: The current task's current credentials
1120 * @flags: Indications of what has changed
1122 * Fix up the results of setuid() call before the credential changes are
1125 * Return: 0 to grant the changes, -ve to deny them.
1127 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1133 /* juggle the capabilities to follow [RES]UID changes unless
1134 * otherwise suppressed */
1135 if (!issecure(SECURE_NO_SETUID_FIXUP))
1136 cap_emulate_setxuid(new, old);
1140 /* juggle the capabilties to follow FSUID changes, unless
1141 * otherwise suppressed
1143 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1144 * if not, we might be a bit too harsh here.
1146 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1147 kuid_t root_uid = make_kuid(old->user_ns, 0);
1148 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1149 new->cap_effective =
1150 cap_drop_fs_set(new->cap_effective);
1152 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1153 new->cap_effective =
1154 cap_raise_fs_set(new->cap_effective,
1155 new->cap_permitted);
1167 * Rationale: code calling task_setscheduler, task_setioprio, and
1168 * task_setnice, assumes that
1169 * . if capable(cap_sys_nice), then those actions should be allowed
1170 * . if not capable(cap_sys_nice), but acting on your own processes,
1171 * then those actions should be allowed
1172 * This is insufficient now since you can call code without suid, but
1173 * yet with increased caps.
1174 * So we check for increased caps on the target process.
1176 static int cap_safe_nice(struct task_struct *p)
1178 int is_subset, ret = 0;
1181 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1182 current_cred()->cap_permitted);
1183 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1191 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1192 * @p: The task to affect
1194 * Detemine if the requested scheduler policy change is permitted for the
1197 * Return: 0 if permission is granted, -ve if denied.
1199 int cap_task_setscheduler(struct task_struct *p)
1201 return cap_safe_nice(p);
1205 * cap_task_setioprio - Detemine if I/O priority change is permitted
1206 * @p: The task to affect
1207 * @ioprio: The I/O priority to set
1209 * Detemine if the requested I/O priority change is permitted for the specified
1212 * Return: 0 if permission is granted, -ve if denied.
1214 int cap_task_setioprio(struct task_struct *p, int ioprio)
1216 return cap_safe_nice(p);
1220 * cap_task_setnice - Detemine if task priority change is permitted
1221 * @p: The task to affect
1222 * @nice: The nice value to set
1224 * Detemine if the requested task priority change is permitted for the
1227 * Return: 0 if permission is granted, -ve if denied.
1229 int cap_task_setnice(struct task_struct *p, int nice)
1231 return cap_safe_nice(p);
1235 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1236 * the current task's bounding set. Returns 0 on success, -ve on error.
1238 static int cap_prctl_drop(unsigned long cap)
1242 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1244 if (!cap_valid(cap))
1247 new = prepare_creds();
1250 cap_lower(new->cap_bset, cap);
1251 return commit_creds(new);
1255 * cap_task_prctl - Implement process control functions for this security module
1256 * @option: The process control function requested
1257 * @arg2: The argument data for this function
1258 * @arg3: The argument data for this function
1259 * @arg4: The argument data for this function
1260 * @arg5: The argument data for this function
1262 * Allow process control functions (sys_prctl()) to alter capabilities; may
1263 * also deny access to other functions not otherwise implemented here.
1265 * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1266 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1267 * modules will consider performing the function.
1269 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1270 unsigned long arg4, unsigned long arg5)
1272 const struct cred *old = current_cred();
1276 case PR_CAPBSET_READ:
1277 if (!cap_valid(arg2))
1279 return !!cap_raised(old->cap_bset, arg2);
1281 case PR_CAPBSET_DROP:
1282 return cap_prctl_drop(arg2);
1285 * The next four prctl's remain to assist with transitioning a
1286 * system from legacy UID=0 based privilege (when filesystem
1287 * capabilities are not in use) to a system using filesystem
1288 * capabilities only - as the POSIX.1e draft intended.
1292 * PR_SET_SECUREBITS =
1293 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1294 * | issecure_mask(SECURE_NOROOT)
1295 * | issecure_mask(SECURE_NOROOT_LOCKED)
1296 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1297 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1299 * will ensure that the current process and all of its
1300 * children will be locked into a pure
1301 * capability-based-privilege environment.
1303 case PR_SET_SECUREBITS:
1304 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1305 & (old->securebits ^ arg2)) /*[1]*/
1306 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1307 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1308 || (cap_capable(current_cred(),
1309 current_cred()->user_ns,
1311 CAP_OPT_NONE) != 0) /*[4]*/
1313 * [1] no changing of bits that are locked
1314 * [2] no unlocking of locks
1315 * [3] no setting of unsupported bits
1316 * [4] doing anything requires privilege (go read about
1317 * the "sendmail capabilities bug")
1320 /* cannot change a locked bit */
1323 new = prepare_creds();
1326 new->securebits = arg2;
1327 return commit_creds(new);
1329 case PR_GET_SECUREBITS:
1330 return old->securebits;
1332 case PR_GET_KEEPCAPS:
1333 return !!issecure(SECURE_KEEP_CAPS);
1335 case PR_SET_KEEPCAPS:
1336 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1338 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1341 new = prepare_creds();
1345 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1347 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1348 return commit_creds(new);
1350 case PR_CAP_AMBIENT:
1351 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1352 if (arg3 | arg4 | arg5)
1355 new = prepare_creds();
1358 cap_clear(new->cap_ambient);
1359 return commit_creds(new);
1362 if (((!cap_valid(arg3)) | arg4 | arg5))
1365 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1366 return !!cap_raised(current_cred()->cap_ambient, arg3);
1367 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1368 arg2 != PR_CAP_AMBIENT_LOWER) {
1371 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1372 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1373 !cap_raised(current_cred()->cap_inheritable,
1375 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1378 new = prepare_creds();
1381 if (arg2 == PR_CAP_AMBIENT_RAISE)
1382 cap_raise(new->cap_ambient, arg3);
1384 cap_lower(new->cap_ambient, arg3);
1385 return commit_creds(new);
1389 /* No functionality available - continue with default */
1395 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1396 * @mm: The VM space in which the new mapping is to be made
1397 * @pages: The size of the mapping
1399 * Determine whether the allocation of a new virtual mapping by the current
1400 * task is permitted.
1402 * Return: 1 if permission is granted, 0 if not.
1404 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1406 int cap_sys_admin = 0;
1408 if (cap_capable(current_cred(), &init_user_ns,
1409 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1412 return cap_sys_admin;
1416 * cap_mmap_addr - check if able to map given addr
1417 * @addr: address attempting to be mapped
1419 * If the process is attempting to map memory below dac_mmap_min_addr they need
1420 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1421 * capability security module.
1423 * Return: 0 if this mapping should be allowed or -EPERM if not.
1425 int cap_mmap_addr(unsigned long addr)
1429 if (addr < dac_mmap_min_addr) {
1430 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1432 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1434 current->flags |= PF_SUPERPRIV;
1439 int cap_mmap_file(struct file *file, unsigned long reqprot,
1440 unsigned long prot, unsigned long flags)
1445 #ifdef CONFIG_SECURITY
1447 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1448 LSM_HOOK_INIT(capable, cap_capable),
1449 LSM_HOOK_INIT(settime, cap_settime),
1450 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1451 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1452 LSM_HOOK_INIT(capget, cap_capget),
1453 LSM_HOOK_INIT(capset, cap_capset),
1454 LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1455 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1456 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1457 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1458 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1459 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1460 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1461 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1462 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1463 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1464 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1465 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1468 static int __init capability_init(void)
1470 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1475 DEFINE_LSM(capability) = {
1476 .name = "capability",
1477 .order = LSM_ORDER_FIRST,
1478 .init = capability_init,
1481 #endif /* CONFIG_SECURITY */