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(vfsuid_t rootvfsuid)
333 struct user_namespace *ns;
336 if (!vfsuid_valid(rootvfsuid))
339 kroot = vfsuid_into_kuid(rootvfsuid);
340 for (ns = current_user_ns();; ns = ns->parent) {
341 if (from_kuid(ns, kroot) == 0)
343 if (ns == &init_user_ns)
350 static __u32 sansflags(__u32 m)
352 return m & ~VFS_CAP_FLAGS_EFFECTIVE;
355 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
357 if (size != XATTR_CAPS_SZ_2)
359 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
362 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
364 if (size != XATTR_CAPS_SZ_3)
366 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
370 * getsecurity: We are called for security.* before any attempt to read the
371 * xattr from the inode itself.
373 * This gives us a chance to read the on-disk value and convert it. If we
374 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
376 * Note we are not called by vfs_getxattr_alloc(), but that is only called
377 * by the integrity subsystem, which really wants the unconverted values -
380 int cap_inode_getsecurity(struct user_namespace *mnt_userns,
381 struct inode *inode, const char *name, void **buffer,
388 uid_t root, mappedroot;
390 struct vfs_cap_data *cap;
391 struct vfs_ns_cap_data *nscap = NULL;
392 struct dentry *dentry;
393 struct user_namespace *fs_ns;
395 if (strcmp(name, "capability") != 0)
398 dentry = d_find_any_alias(inode);
402 size = sizeof(struct vfs_ns_cap_data);
403 ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
404 &tmpbuf, size, GFP_NOFS);
407 if (ret < 0 || !tmpbuf) {
412 fs_ns = inode->i_sb->s_user_ns;
413 cap = (struct vfs_cap_data *) tmpbuf;
414 if (is_v2header((size_t) ret, cap)) {
416 } else if (is_v3header((size_t) ret, cap)) {
417 nscap = (struct vfs_ns_cap_data *) tmpbuf;
418 root = le32_to_cpu(nscap->rootid);
424 kroot = make_kuid(fs_ns, root);
426 /* If this is an idmapped mount shift the kuid. */
427 vfsroot = make_vfsuid(mnt_userns, fs_ns, kroot);
429 /* If the root kuid maps to a valid uid in current ns, then return
430 * this as a nscap. */
431 mappedroot = from_kuid(current_user_ns(), vfsuid_into_kuid(vfsroot));
432 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
433 size = sizeof(struct vfs_ns_cap_data);
436 /* v2 -> v3 conversion */
437 nscap = kzalloc(size, GFP_ATOMIC);
442 nsmagic = VFS_CAP_REVISION_3;
443 magic = le32_to_cpu(cap->magic_etc);
444 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
445 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
446 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
447 nscap->magic_etc = cpu_to_le32(nsmagic);
449 /* use allocated v3 buffer */
452 nscap->rootid = cpu_to_le32(mappedroot);
458 if (!rootid_owns_currentns(vfsroot)) {
463 /* This comes from a parent namespace. Return as a v2 capability */
464 size = sizeof(struct vfs_cap_data);
467 /* v3 -> v2 conversion */
468 cap = kzalloc(size, GFP_ATOMIC);
473 magic = VFS_CAP_REVISION_2;
474 nsmagic = le32_to_cpu(nscap->magic_etc);
475 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
476 magic |= VFS_CAP_FLAGS_EFFECTIVE;
477 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
478 cap->magic_etc = cpu_to_le32(magic);
480 /* use unconverted v2 */
491 * rootid_from_xattr - translate root uid of vfs caps
493 * @value: vfs caps value which may be modified by this function
494 * @size: size of @ivalue
495 * @task_ns: user namespace of the caller
497 static vfsuid_t rootid_from_xattr(const void *value, size_t size,
498 struct user_namespace *task_ns)
500 const struct vfs_ns_cap_data *nscap = value;
503 if (size == XATTR_CAPS_SZ_3)
504 rootid = le32_to_cpu(nscap->rootid);
506 return VFSUIDT_INIT(make_kuid(task_ns, rootid));
509 static bool validheader(size_t size, const struct vfs_cap_data *cap)
511 return is_v2header(size, cap) || is_v3header(size, cap);
515 * cap_convert_nscap - check vfs caps
517 * @mnt_userns: user namespace of the mount the inode was found from
518 * @dentry: used to retrieve inode to check permissions on
519 * @ivalue: vfs caps value which may be modified by this function
520 * @size: size of @ivalue
522 * User requested a write of security.capability. If needed, update the
523 * xattr to change from v2 to v3, or to fixup the v3 rootid.
525 * If the inode has been found through an idmapped mount the user namespace of
526 * the vfsmount must be passed through @mnt_userns. This function will then
527 * take care to map the inode according to @mnt_userns before checking
528 * permissions. On non-idmapped mounts or if permission checking is to be
529 * performed on the raw inode simply passs init_user_ns.
531 * Return: On success, return the new size; on error, return < 0.
533 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
534 const void **ivalue, size_t size)
536 struct vfs_ns_cap_data *nscap;
538 const struct vfs_cap_data *cap = *ivalue;
539 __u32 magic, nsmagic;
540 struct inode *inode = d_backing_inode(dentry);
541 struct user_namespace *task_ns = current_user_ns(),
542 *fs_ns = inode->i_sb->s_user_ns;
549 if (!validheader(size, cap))
551 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
553 if (size == XATTR_CAPS_SZ_2 && (mnt_userns == fs_ns))
554 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
555 /* user is privileged, just write the v2 */
558 vfsrootid = rootid_from_xattr(*ivalue, size, task_ns);
559 if (!vfsuid_valid(vfsrootid))
562 rootid = from_vfsuid(mnt_userns, fs_ns, vfsrootid);
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;
657 struct user_namespace *fs_ns;
659 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
664 fs_ns = inode->i_sb->s_user_ns;
665 size = __vfs_getxattr((struct dentry *)dentry, inode,
666 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
667 if (size == -ENODATA || size == -EOPNOTSUPP)
668 /* no data, that's ok */
674 if (size < sizeof(magic_etc))
677 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
679 rootkuid = make_kuid(fs_ns, 0);
680 switch (magic_etc & VFS_CAP_REVISION_MASK) {
681 case VFS_CAP_REVISION_1:
682 if (size != XATTR_CAPS_SZ_1)
684 tocopy = VFS_CAP_U32_1;
686 case VFS_CAP_REVISION_2:
687 if (size != XATTR_CAPS_SZ_2)
689 tocopy = VFS_CAP_U32_2;
691 case VFS_CAP_REVISION_3:
692 if (size != XATTR_CAPS_SZ_3)
694 tocopy = VFS_CAP_U32_3;
695 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
702 rootvfsuid = make_vfsuid(mnt_userns, fs_ns, rootkuid);
703 if (!vfsuid_valid(rootvfsuid))
706 /* Limit the caps to the mounter of the filesystem
707 * or the more limited uid specified in the xattr.
709 if (!rootid_owns_currentns(rootvfsuid))
712 CAP_FOR_EACH_U32(i) {
715 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
716 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
719 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
720 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
722 cpu_caps->rootid = vfsuid_into_kuid(rootvfsuid);
728 * Attempt to get the on-exec apply capability sets for an executable file from
729 * its xattrs and, if present, apply them to the proposed credentials being
730 * constructed by execve().
732 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
733 bool *effective, bool *has_fcap)
736 struct cpu_vfs_cap_data vcaps;
738 cap_clear(bprm->cred->cap_permitted);
740 if (!file_caps_enabled)
743 if (!mnt_may_suid(file->f_path.mnt))
747 * This check is redundant with mnt_may_suid() but is kept to make
748 * explicit that capability bits are limited to s_user_ns and its
751 if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
754 rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
755 file->f_path.dentry, &vcaps);
758 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
760 else if (rc == -ENODATA)
765 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
769 cap_clear(bprm->cred->cap_permitted);
774 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
776 static inline bool __is_real(kuid_t uid, struct cred *cred)
777 { return uid_eq(cred->uid, uid); }
779 static inline bool __is_eff(kuid_t uid, struct cred *cred)
780 { return uid_eq(cred->euid, uid); }
782 static inline bool __is_suid(kuid_t uid, struct cred *cred)
783 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
786 * handle_privileged_root - Handle case of privileged root
787 * @bprm: The execution parameters, including the proposed creds
788 * @has_fcap: Are any file capabilities set?
789 * @effective: Do we have effective root privilege?
790 * @root_uid: This namespace' root UID WRT initial USER namespace
792 * Handle the case where root is privileged and hasn't been neutered by
793 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
794 * set UID root and nothing is changed. If we are root, cap_permitted is
795 * updated. If we have become set UID root, the effective bit is set.
797 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
798 bool *effective, kuid_t root_uid)
800 const struct cred *old = current_cred();
801 struct cred *new = bprm->cred;
803 if (!root_privileged())
806 * If the legacy file capability is set, then don't set privs
807 * for a setuid root binary run by a non-root user. Do set it
808 * for a root user just to cause least surprise to an admin.
810 if (has_fcap && __is_suid(root_uid, new)) {
811 warn_setuid_and_fcaps_mixed(bprm->filename);
815 * To support inheritance of root-permissions and suid-root
816 * executables under compatibility mode, we override the
817 * capability sets for the file.
819 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
820 /* pP' = (cap_bset & ~0) | (pI & ~0) */
821 new->cap_permitted = cap_combine(old->cap_bset,
822 old->cap_inheritable);
825 * If only the real uid is 0, we do not set the effective bit.
827 if (__is_eff(root_uid, new))
831 #define __cap_gained(field, target, source) \
832 !cap_issubset(target->cap_##field, source->cap_##field)
833 #define __cap_grew(target, source, cred) \
834 !cap_issubset(cred->cap_##target, cred->cap_##source)
835 #define __cap_full(field, cred) \
836 cap_issubset(CAP_FULL_SET, cred->cap_##field)
838 static inline bool __is_setuid(struct cred *new, const struct cred *old)
839 { return !uid_eq(new->euid, old->uid); }
841 static inline bool __is_setgid(struct cred *new, const struct cred *old)
842 { return !gid_eq(new->egid, old->gid); }
845 * 1) Audit candidate if current->cap_effective is set
847 * We do not bother to audit if 3 things are true:
848 * 1) cap_effective has all caps
849 * 2) we became root *OR* are were already root
850 * 3) root is supposed to have all caps (SECURE_NOROOT)
851 * Since this is just a normal root execing a process.
853 * Number 1 above might fail if you don't have a full bset, but I think
854 * that is interesting information to audit.
856 * A number of other conditions require logging:
857 * 2) something prevented setuid root getting all caps
858 * 3) non-setuid root gets fcaps
859 * 4) non-setuid root gets ambient
861 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
862 kuid_t root, bool has_fcap)
866 if ((__cap_grew(effective, ambient, new) &&
867 !(__cap_full(effective, new) &&
868 (__is_eff(root, new) || __is_real(root, new)) &&
869 root_privileged())) ||
870 (root_privileged() &&
871 __is_suid(root, new) &&
872 !__cap_full(effective, new)) ||
873 (!__is_setuid(new, old) &&
875 __cap_gained(permitted, new, old)) ||
876 __cap_gained(ambient, new, old))))
884 * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
885 * @bprm: The execution parameters, including the proposed creds
886 * @file: The file to pull the credentials from
888 * Set up the proposed credentials for a new execution context being
889 * constructed by execve(). The proposed creds in @bprm->cred is altered,
890 * which won't take effect immediately.
892 * Return: 0 if successful, -ve on error.
894 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
896 /* Process setpcap binaries and capabilities for uid 0 */
897 const struct cred *old = current_cred();
898 struct cred *new = bprm->cred;
899 bool effective = false, has_fcap = false, is_setid;
903 if (WARN_ON(!cap_ambient_invariant_ok(old)))
906 ret = get_file_caps(bprm, file, &effective, &has_fcap);
910 root_uid = make_kuid(new->user_ns, 0);
912 handle_privileged_root(bprm, has_fcap, &effective, root_uid);
914 /* if we have fs caps, clear dangerous personality flags */
915 if (__cap_gained(permitted, new, old))
916 bprm->per_clear |= PER_CLEAR_ON_SETID;
918 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
919 * credentials unless they have the appropriate permit.
921 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
923 is_setid = __is_setuid(new, old) || __is_setgid(new, old);
925 if ((is_setid || __cap_gained(permitted, new, old)) &&
926 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
927 !ptracer_capable(current, new->user_ns))) {
928 /* downgrade; they get no more than they had, and maybe less */
929 if (!ns_capable(new->user_ns, CAP_SETUID) ||
930 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
931 new->euid = new->uid;
932 new->egid = new->gid;
934 new->cap_permitted = cap_intersect(new->cap_permitted,
938 new->suid = new->fsuid = new->euid;
939 new->sgid = new->fsgid = new->egid;
941 /* File caps or setid cancels ambient. */
942 if (has_fcap || is_setid)
943 cap_clear(new->cap_ambient);
946 * Now that we've computed pA', update pP' to give:
947 * pP' = (X & fP) | (pI & fI) | pA'
949 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
952 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
953 * this is the same as pE' = (fE ? pP' : 0) | pA'.
956 new->cap_effective = new->cap_permitted;
958 new->cap_effective = new->cap_ambient;
960 if (WARN_ON(!cap_ambient_invariant_ok(new)))
963 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
964 ret = audit_log_bprm_fcaps(bprm, new, old);
969 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
971 if (WARN_ON(!cap_ambient_invariant_ok(new)))
974 /* Check for privilege-elevated exec. */
976 (!__is_real(root_uid, new) &&
978 __cap_grew(permitted, ambient, new))))
979 bprm->secureexec = 1;
985 * cap_inode_setxattr - Determine whether an xattr may be altered
986 * @dentry: The inode/dentry being altered
987 * @name: The name of the xattr to be changed
988 * @value: The value that the xattr will be changed to
989 * @size: The size of value
990 * @flags: The replacement flag
992 * Determine whether an xattr may be altered or set on an inode, returning 0 if
993 * permission is granted, -ve if denied.
995 * This is used to make sure security xattrs don't get updated or set by those
996 * who aren't privileged to do so.
998 int cap_inode_setxattr(struct dentry *dentry, const char *name,
999 const void *value, size_t size, int flags)
1001 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1003 /* Ignore non-security xattrs */
1004 if (strncmp(name, XATTR_SECURITY_PREFIX,
1005 XATTR_SECURITY_PREFIX_LEN) != 0)
1009 * For XATTR_NAME_CAPS the check will be done in
1010 * cap_convert_nscap(), called by setxattr()
1012 if (strcmp(name, XATTR_NAME_CAPS) == 0)
1015 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1021 * cap_inode_removexattr - Determine whether an xattr may be removed
1023 * @mnt_userns: User namespace of the mount the inode was found from
1024 * @dentry: The inode/dentry being altered
1025 * @name: The name of the xattr to be changed
1027 * Determine whether an xattr may be removed from an inode, returning 0 if
1028 * permission is granted, -ve if denied.
1030 * If the inode has been found through an idmapped mount the user namespace of
1031 * the vfsmount must be passed through @mnt_userns. This function will then
1032 * take care to map the inode according to @mnt_userns before checking
1033 * permissions. On non-idmapped mounts or if permission checking is to be
1034 * performed on the raw inode simply passs init_user_ns.
1036 * This is used to make sure security xattrs don't get removed by those who
1037 * aren't privileged to remove them.
1039 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1040 struct dentry *dentry, const char *name)
1042 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1044 /* Ignore non-security xattrs */
1045 if (strncmp(name, XATTR_SECURITY_PREFIX,
1046 XATTR_SECURITY_PREFIX_LEN) != 0)
1049 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1050 /* security.capability gets namespaced */
1051 struct inode *inode = d_backing_inode(dentry);
1054 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1059 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1065 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1066 * a process after a call to setuid, setreuid, or setresuid.
1068 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1069 * {r,e,s}uid != 0, the permitted and effective capabilities are
1072 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1073 * capabilities of the process are cleared.
1075 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1076 * capabilities are set to the permitted capabilities.
1078 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1083 * cevans - New behaviour, Oct '99
1084 * A process may, via prctl(), elect to keep its capabilities when it
1085 * calls setuid() and switches away from uid==0. Both permitted and
1086 * effective sets will be retained.
1087 * Without this change, it was impossible for a daemon to drop only some
1088 * of its privilege. The call to setuid(!=0) would drop all privileges!
1089 * Keeping uid 0 is not an option because uid 0 owns too many vital
1091 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1093 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1095 kuid_t root_uid = make_kuid(old->user_ns, 0);
1097 if ((uid_eq(old->uid, root_uid) ||
1098 uid_eq(old->euid, root_uid) ||
1099 uid_eq(old->suid, root_uid)) &&
1100 (!uid_eq(new->uid, root_uid) &&
1101 !uid_eq(new->euid, root_uid) &&
1102 !uid_eq(new->suid, root_uid))) {
1103 if (!issecure(SECURE_KEEP_CAPS)) {
1104 cap_clear(new->cap_permitted);
1105 cap_clear(new->cap_effective);
1109 * Pre-ambient programs expect setresuid to nonroot followed
1110 * by exec to drop capabilities. We should make sure that
1111 * this remains the case.
1113 cap_clear(new->cap_ambient);
1115 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1116 cap_clear(new->cap_effective);
1117 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1118 new->cap_effective = new->cap_permitted;
1122 * cap_task_fix_setuid - Fix up the results of setuid() call
1123 * @new: The proposed credentials
1124 * @old: The current task's current credentials
1125 * @flags: Indications of what has changed
1127 * Fix up the results of setuid() call before the credential changes are
1130 * Return: 0 to grant the changes, -ve to deny them.
1132 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1138 /* juggle the capabilities to follow [RES]UID changes unless
1139 * otherwise suppressed */
1140 if (!issecure(SECURE_NO_SETUID_FIXUP))
1141 cap_emulate_setxuid(new, old);
1145 /* juggle the capabilties to follow FSUID changes, unless
1146 * otherwise suppressed
1148 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1149 * if not, we might be a bit too harsh here.
1151 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1152 kuid_t root_uid = make_kuid(old->user_ns, 0);
1153 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1154 new->cap_effective =
1155 cap_drop_fs_set(new->cap_effective);
1157 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1158 new->cap_effective =
1159 cap_raise_fs_set(new->cap_effective,
1160 new->cap_permitted);
1172 * Rationale: code calling task_setscheduler, task_setioprio, and
1173 * task_setnice, assumes that
1174 * . if capable(cap_sys_nice), then those actions should be allowed
1175 * . if not capable(cap_sys_nice), but acting on your own processes,
1176 * then those actions should be allowed
1177 * This is insufficient now since you can call code without suid, but
1178 * yet with increased caps.
1179 * So we check for increased caps on the target process.
1181 static int cap_safe_nice(struct task_struct *p)
1183 int is_subset, ret = 0;
1186 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1187 current_cred()->cap_permitted);
1188 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1196 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1197 * @p: The task to affect
1199 * Detemine if the requested scheduler policy change is permitted for the
1202 * Return: 0 if permission is granted, -ve if denied.
1204 int cap_task_setscheduler(struct task_struct *p)
1206 return cap_safe_nice(p);
1210 * cap_task_setioprio - Detemine if I/O priority change is permitted
1211 * @p: The task to affect
1212 * @ioprio: The I/O priority to set
1214 * Detemine if the requested I/O priority change is permitted for the specified
1217 * Return: 0 if permission is granted, -ve if denied.
1219 int cap_task_setioprio(struct task_struct *p, int ioprio)
1221 return cap_safe_nice(p);
1225 * cap_task_setnice - Detemine if task priority change is permitted
1226 * @p: The task to affect
1227 * @nice: The nice value to set
1229 * Detemine if the requested task priority change is permitted for the
1232 * Return: 0 if permission is granted, -ve if denied.
1234 int cap_task_setnice(struct task_struct *p, int nice)
1236 return cap_safe_nice(p);
1240 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1241 * the current task's bounding set. Returns 0 on success, -ve on error.
1243 static int cap_prctl_drop(unsigned long cap)
1247 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1249 if (!cap_valid(cap))
1252 new = prepare_creds();
1255 cap_lower(new->cap_bset, cap);
1256 return commit_creds(new);
1260 * cap_task_prctl - Implement process control functions for this security module
1261 * @option: The process control function requested
1262 * @arg2: The argument data for this function
1263 * @arg3: The argument data for this function
1264 * @arg4: The argument data for this function
1265 * @arg5: The argument data for this function
1267 * Allow process control functions (sys_prctl()) to alter capabilities; may
1268 * also deny access to other functions not otherwise implemented here.
1270 * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1271 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1272 * modules will consider performing the function.
1274 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1275 unsigned long arg4, unsigned long arg5)
1277 const struct cred *old = current_cred();
1281 case PR_CAPBSET_READ:
1282 if (!cap_valid(arg2))
1284 return !!cap_raised(old->cap_bset, arg2);
1286 case PR_CAPBSET_DROP:
1287 return cap_prctl_drop(arg2);
1290 * The next four prctl's remain to assist with transitioning a
1291 * system from legacy UID=0 based privilege (when filesystem
1292 * capabilities are not in use) to a system using filesystem
1293 * capabilities only - as the POSIX.1e draft intended.
1297 * PR_SET_SECUREBITS =
1298 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1299 * | issecure_mask(SECURE_NOROOT)
1300 * | issecure_mask(SECURE_NOROOT_LOCKED)
1301 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1302 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1304 * will ensure that the current process and all of its
1305 * children will be locked into a pure
1306 * capability-based-privilege environment.
1308 case PR_SET_SECUREBITS:
1309 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1310 & (old->securebits ^ arg2)) /*[1]*/
1311 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1312 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1313 || (cap_capable(current_cred(),
1314 current_cred()->user_ns,
1316 CAP_OPT_NONE) != 0) /*[4]*/
1318 * [1] no changing of bits that are locked
1319 * [2] no unlocking of locks
1320 * [3] no setting of unsupported bits
1321 * [4] doing anything requires privilege (go read about
1322 * the "sendmail capabilities bug")
1325 /* cannot change a locked bit */
1328 new = prepare_creds();
1331 new->securebits = arg2;
1332 return commit_creds(new);
1334 case PR_GET_SECUREBITS:
1335 return old->securebits;
1337 case PR_GET_KEEPCAPS:
1338 return !!issecure(SECURE_KEEP_CAPS);
1340 case PR_SET_KEEPCAPS:
1341 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1343 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1346 new = prepare_creds();
1350 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1352 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1353 return commit_creds(new);
1355 case PR_CAP_AMBIENT:
1356 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1357 if (arg3 | arg4 | arg5)
1360 new = prepare_creds();
1363 cap_clear(new->cap_ambient);
1364 return commit_creds(new);
1367 if (((!cap_valid(arg3)) | arg4 | arg5))
1370 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1371 return !!cap_raised(current_cred()->cap_ambient, arg3);
1372 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1373 arg2 != PR_CAP_AMBIENT_LOWER) {
1376 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1377 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1378 !cap_raised(current_cred()->cap_inheritable,
1380 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1383 new = prepare_creds();
1386 if (arg2 == PR_CAP_AMBIENT_RAISE)
1387 cap_raise(new->cap_ambient, arg3);
1389 cap_lower(new->cap_ambient, arg3);
1390 return commit_creds(new);
1394 /* No functionality available - continue with default */
1400 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1401 * @mm: The VM space in which the new mapping is to be made
1402 * @pages: The size of the mapping
1404 * Determine whether the allocation of a new virtual mapping by the current
1405 * task is permitted.
1407 * Return: 1 if permission is granted, 0 if not.
1409 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1411 int cap_sys_admin = 0;
1413 if (cap_capable(current_cred(), &init_user_ns,
1414 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1417 return cap_sys_admin;
1421 * cap_mmap_addr - check if able to map given addr
1422 * @addr: address attempting to be mapped
1424 * If the process is attempting to map memory below dac_mmap_min_addr they need
1425 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1426 * capability security module.
1428 * Return: 0 if this mapping should be allowed or -EPERM if not.
1430 int cap_mmap_addr(unsigned long addr)
1434 if (addr < dac_mmap_min_addr) {
1435 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1437 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1439 current->flags |= PF_SUPERPRIV;
1444 int cap_mmap_file(struct file *file, unsigned long reqprot,
1445 unsigned long prot, unsigned long flags)
1450 #ifdef CONFIG_SECURITY
1452 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1453 LSM_HOOK_INIT(capable, cap_capable),
1454 LSM_HOOK_INIT(settime, cap_settime),
1455 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1456 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1457 LSM_HOOK_INIT(capget, cap_capget),
1458 LSM_HOOK_INIT(capset, cap_capset),
1459 LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1460 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1461 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1462 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1463 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1464 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1465 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1466 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1467 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1468 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1469 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1470 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1473 static int __init capability_init(void)
1475 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1480 DEFINE_LSM(capability) = {
1481 .name = "capability",
1482 .order = LSM_ORDER_FIRST,
1483 .init = capability_init,
1486 #endif /* CONFIG_SECURITY */