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>
29 * If a non-root user executes a setuid-root binary in
30 * !secure(SECURE_NOROOT) mode, then we raise capabilities.
31 * However if fE is also set, then the intent is for only
32 * the file capabilities to be applied, and the setuid-root
33 * bit is left on either to change the uid (plausible) or
34 * to get full privilege on a kernel without file capabilities
35 * support. So in that case we do not raise capabilities.
37 * Warn if that happens, once per boot.
39 static void warn_setuid_and_fcaps_mixed(const char *fname)
43 printk(KERN_INFO "warning: `%s' has both setuid-root and"
44 " effective capabilities. Therefore not raising all"
45 " capabilities.\n", fname);
51 * cap_capable - Determine whether a task has a particular effective capability
52 * @cred: The credentials to use
53 * @ns: The user namespace in which we need the capability
54 * @cap: The capability to check for
55 * @opts: Bitmask of options defined in include/linux/security.h
57 * Determine whether the nominated task has the specified capability amongst
58 * its effective set, returning 0 if it does, -ve if it does not.
60 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
61 * and has_capability() functions. That is, it has the reverse semantics:
62 * cap_has_capability() returns 0 when a task has a capability, but the
63 * kernel's capable() and has_capability() returns 1 for this case.
65 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
66 int cap, unsigned int opts)
68 struct user_namespace *ns = targ_ns;
70 /* See if cred has the capability in the target user namespace
71 * by examining the target user namespace and all of the target
72 * user namespace's parents.
75 /* Do we have the necessary capabilities? */
76 if (ns == cred->user_ns)
77 return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
80 * If we're already at a lower level than we're looking for,
81 * we're done searching.
83 if (ns->level <= cred->user_ns->level)
87 * The owner of the user namespace in the parent of the
88 * user namespace has all caps.
90 if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
94 * If you have a capability in a parent user ns, then you have
95 * it over all children user namespaces as well.
100 /* We never get here */
104 * cap_settime - Determine whether the current process may set the system clock
105 * @ts: The time to set
106 * @tz: The timezone to set
108 * Determine whether the current process may set the system clock and timezone
109 * information, returning 0 if permission granted, -ve if denied.
111 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
113 if (!capable(CAP_SYS_TIME))
119 * cap_ptrace_access_check - Determine whether the current process may access
121 * @child: The process to be accessed
122 * @mode: The mode of attachment.
124 * If we are in the same or an ancestor user_ns and have all the target
125 * task's capabilities, then ptrace access is allowed.
126 * If we have the ptrace capability to the target user_ns, then ptrace
130 * Determine whether a process may access another, returning 0 if permission
131 * granted, -ve if denied.
133 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
136 const struct cred *cred, *child_cred;
137 const kernel_cap_t *caller_caps;
140 cred = current_cred();
141 child_cred = __task_cred(child);
142 if (mode & PTRACE_MODE_FSCREDS)
143 caller_caps = &cred->cap_effective;
145 caller_caps = &cred->cap_permitted;
146 if (cred->user_ns == child_cred->user_ns &&
147 cap_issubset(child_cred->cap_permitted, *caller_caps))
149 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
158 * cap_ptrace_traceme - Determine whether another process may trace the current
159 * @parent: The task proposed to be the tracer
161 * If parent is in the same or an ancestor user_ns and has all current's
162 * capabilities, then ptrace access is allowed.
163 * If parent has the ptrace capability to current's user_ns, then ptrace
167 * Determine whether the nominated task is permitted to trace the current
168 * process, returning 0 if permission is granted, -ve if denied.
170 int cap_ptrace_traceme(struct task_struct *parent)
173 const struct cred *cred, *child_cred;
176 cred = __task_cred(parent);
177 child_cred = current_cred();
178 if (cred->user_ns == child_cred->user_ns &&
179 cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
181 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
190 * cap_capget - Retrieve a task's capability sets
191 * @target: The task from which to retrieve the capability sets
192 * @effective: The place to record the effective set
193 * @inheritable: The place to record the inheritable set
194 * @permitted: The place to record the permitted set
196 * This function retrieves the capabilities of the nominated task and returns
197 * them to the caller.
199 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
200 kernel_cap_t *inheritable, kernel_cap_t *permitted)
202 const struct cred *cred;
204 /* Derived from kernel/capability.c:sys_capget. */
206 cred = __task_cred(target);
207 *effective = cred->cap_effective;
208 *inheritable = cred->cap_inheritable;
209 *permitted = cred->cap_permitted;
215 * Determine whether the inheritable capabilities are limited to the old
216 * permitted set. Returns 1 if they are limited, 0 if they are not.
218 static inline int cap_inh_is_capped(void)
220 /* they are so limited unless the current task has the CAP_SETPCAP
223 if (cap_capable(current_cred(), current_cred()->user_ns,
224 CAP_SETPCAP, CAP_OPT_NONE) == 0)
230 * cap_capset - Validate and apply proposed changes to current's capabilities
231 * @new: The proposed new credentials; alterations should be made here
232 * @old: The current task's current credentials
233 * @effective: A pointer to the proposed new effective capabilities set
234 * @inheritable: A pointer to the proposed new inheritable capabilities set
235 * @permitted: A pointer to the proposed new permitted capabilities set
237 * This function validates and applies a proposed mass change to the current
238 * process's capability sets. The changes are made to the proposed new
239 * credentials, and assuming no error, will be committed by the caller of LSM.
241 int cap_capset(struct cred *new,
242 const struct cred *old,
243 const kernel_cap_t *effective,
244 const kernel_cap_t *inheritable,
245 const kernel_cap_t *permitted)
247 if (cap_inh_is_capped() &&
248 !cap_issubset(*inheritable,
249 cap_combine(old->cap_inheritable,
250 old->cap_permitted)))
251 /* incapable of using this inheritable set */
254 if (!cap_issubset(*inheritable,
255 cap_combine(old->cap_inheritable,
257 /* no new pI capabilities outside bounding set */
260 /* verify restrictions on target's new Permitted set */
261 if (!cap_issubset(*permitted, old->cap_permitted))
264 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
265 if (!cap_issubset(*effective, *permitted))
268 new->cap_effective = *effective;
269 new->cap_inheritable = *inheritable;
270 new->cap_permitted = *permitted;
273 * Mask off ambient bits that are no longer both permitted and
276 new->cap_ambient = cap_intersect(new->cap_ambient,
277 cap_intersect(*permitted,
279 if (WARN_ON(!cap_ambient_invariant_ok(new)))
285 * cap_inode_need_killpriv - Determine if inode change affects privileges
286 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
288 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
289 * affects the security markings on that inode, and if it is, should
290 * inode_killpriv() be invoked or the change rejected.
292 * Returns 1 if security.capability has a value, meaning inode_killpriv()
293 * is required, 0 otherwise, meaning inode_killpriv() is not required.
295 int cap_inode_need_killpriv(struct dentry *dentry)
297 struct inode *inode = d_backing_inode(dentry);
300 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
305 * cap_inode_killpriv - Erase the security markings on an inode
307 * @mnt_userns: user namespace of the mount the inode was found from
308 * @dentry: The inode/dentry to alter
310 * Erase the privilege-enhancing security markings on an inode.
312 * If the inode has been found through an idmapped mount the user namespace of
313 * the vfsmount must be passed through @mnt_userns. This function will then
314 * take care to map the inode according to @mnt_userns before checking
315 * permissions. On non-idmapped mounts or if permission checking is to be
316 * performed on the raw inode simply passs init_user_ns.
318 * Returns 0 if successful, -ve on error.
320 int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry)
324 error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS);
325 if (error == -EOPNOTSUPP)
330 static bool rootid_owns_currentns(kuid_t kroot)
332 struct user_namespace *ns;
334 if (!uid_valid(kroot))
337 for (ns = current_user_ns(); ; ns = ns->parent) {
338 if (from_kuid(ns, kroot) == 0)
340 if (ns == &init_user_ns)
347 static __u32 sansflags(__u32 m)
349 return m & ~VFS_CAP_FLAGS_EFFECTIVE;
352 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
354 if (size != XATTR_CAPS_SZ_2)
356 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
359 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
361 if (size != XATTR_CAPS_SZ_3)
363 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
367 * getsecurity: We are called for security.* before any attempt to read the
368 * xattr from the inode itself.
370 * This gives us a chance to read the on-disk value and convert it. If we
371 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
373 * Note we are not called by vfs_getxattr_alloc(), but that is only called
374 * by the integrity subsystem, which really wants the unconverted values -
377 int cap_inode_getsecurity(struct user_namespace *mnt_userns,
378 struct inode *inode, const char *name, void **buffer,
384 uid_t root, mappedroot;
386 struct vfs_cap_data *cap;
387 struct vfs_ns_cap_data *nscap = NULL;
388 struct dentry *dentry;
389 struct user_namespace *fs_ns;
391 if (strcmp(name, "capability") != 0)
394 dentry = d_find_any_alias(inode);
398 size = sizeof(struct vfs_ns_cap_data);
399 ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
400 &tmpbuf, size, GFP_NOFS);
406 fs_ns = inode->i_sb->s_user_ns;
407 cap = (struct vfs_cap_data *) tmpbuf;
408 if (is_v2header((size_t) ret, cap)) {
410 } else if (is_v3header((size_t) ret, cap)) {
411 nscap = (struct vfs_ns_cap_data *) tmpbuf;
412 root = le32_to_cpu(nscap->rootid);
418 kroot = make_kuid(fs_ns, root);
420 /* If this is an idmapped mount shift the kuid. */
421 kroot = kuid_into_mnt(mnt_userns, kroot);
423 /* If the root kuid maps to a valid uid in current ns, then return
424 * this as a nscap. */
425 mappedroot = from_kuid(current_user_ns(), kroot);
426 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
427 size = sizeof(struct vfs_ns_cap_data);
430 /* v2 -> v3 conversion */
431 nscap = kzalloc(size, GFP_ATOMIC);
436 nsmagic = VFS_CAP_REVISION_3;
437 magic = le32_to_cpu(cap->magic_etc);
438 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
439 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
440 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
441 nscap->magic_etc = cpu_to_le32(nsmagic);
443 /* use allocated v3 buffer */
446 nscap->rootid = cpu_to_le32(mappedroot);
452 if (!rootid_owns_currentns(kroot)) {
457 /* This comes from a parent namespace. Return as a v2 capability */
458 size = sizeof(struct vfs_cap_data);
461 /* v3 -> v2 conversion */
462 cap = kzalloc(size, GFP_ATOMIC);
467 magic = VFS_CAP_REVISION_2;
468 nsmagic = le32_to_cpu(nscap->magic_etc);
469 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
470 magic |= VFS_CAP_FLAGS_EFFECTIVE;
471 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
472 cap->magic_etc = cpu_to_le32(magic);
474 /* use unconverted v2 */
485 * rootid_from_xattr - translate root uid of vfs caps
487 * @value: vfs caps value which may be modified by this function
488 * @size: size of @ivalue
489 * @task_ns: user namespace of the caller
490 * @mnt_userns: user namespace of the mount the inode was found from
492 * If the inode has been found through an idmapped mount the user namespace of
493 * the vfsmount must be passed through @mnt_userns. This function will then
494 * take care to map the inode according to @mnt_userns before checking
495 * permissions. On non-idmapped mounts or if permission checking is to be
496 * performed on the raw inode simply passs init_user_ns.
498 static kuid_t rootid_from_xattr(const void *value, size_t size,
499 struct user_namespace *task_ns,
500 struct user_namespace *mnt_userns)
502 const struct vfs_ns_cap_data *nscap = value;
506 if (size == XATTR_CAPS_SZ_3)
507 rootid = le32_to_cpu(nscap->rootid);
509 rootkid = make_kuid(task_ns, rootid);
510 return kuid_from_mnt(mnt_userns, rootkid);
513 static bool validheader(size_t size, const struct vfs_cap_data *cap)
515 return is_v2header(size, cap) || is_v3header(size, cap);
519 * cap_convert_nscap - check vfs caps
521 * @mnt_userns: user namespace of the mount the inode was found from
522 * @dentry: used to retrieve inode to check permissions on
523 * @ivalue: vfs caps value which may be modified by this function
524 * @size: size of @ivalue
526 * User requested a write of security.capability. If needed, update the
527 * xattr to change from v2 to v3, or to fixup the v3 rootid.
529 * If the inode has been found through an idmapped mount the user namespace of
530 * the vfsmount must be passed through @mnt_userns. This function will then
531 * take care to map the inode according to @mnt_userns before checking
532 * permissions. On non-idmapped mounts or if permission checking is to be
533 * performed on the raw inode simply passs init_user_ns.
535 * If all is ok, we return the new size, on error return < 0.
537 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
538 const void **ivalue, size_t size)
540 struct vfs_ns_cap_data *nscap;
542 const struct vfs_cap_data *cap = *ivalue;
543 __u32 magic, nsmagic;
544 struct inode *inode = d_backing_inode(dentry);
545 struct user_namespace *task_ns = current_user_ns(),
546 *fs_ns = inode->i_sb->s_user_ns;
552 if (!validheader(size, cap))
554 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
556 if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
557 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
558 /* user is privileged, just write the v2 */
561 rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
562 if (!uid_valid(rootid))
565 nsrootid = from_kuid(fs_ns, rootid);
569 newsize = sizeof(struct vfs_ns_cap_data);
570 nscap = kmalloc(newsize, GFP_ATOMIC);
573 nscap->rootid = cpu_to_le32(nsrootid);
574 nsmagic = VFS_CAP_REVISION_3;
575 magic = le32_to_cpu(cap->magic_etc);
576 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
577 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
578 nscap->magic_etc = cpu_to_le32(nsmagic);
579 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
586 * Calculate the new process capability sets from the capability sets attached
589 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
590 struct linux_binprm *bprm,
594 struct cred *new = bprm->cred;
598 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
601 if (caps->magic_etc & VFS_CAP_REVISION_MASK)
604 CAP_FOR_EACH_U32(i) {
605 __u32 permitted = caps->permitted.cap[i];
606 __u32 inheritable = caps->inheritable.cap[i];
609 * pP' = (X & fP) | (pI & fI)
610 * The addition of pA' is handled later.
612 new->cap_permitted.cap[i] =
613 (new->cap_bset.cap[i] & permitted) |
614 (new->cap_inheritable.cap[i] & inheritable);
616 if (permitted & ~new->cap_permitted.cap[i])
617 /* insufficient to execute correctly */
622 * For legacy apps, with no internal support for recognizing they
623 * do not have enough capabilities, we return an error if they are
624 * missing some "forced" (aka file-permitted) capabilities.
626 return *effective ? ret : 0;
630 * get_vfs_caps_from_disk - retrieve vfs caps from disk
632 * @mnt_userns: user namespace of the mount the inode was found from
633 * @dentry: dentry from which @inode is retrieved
634 * @cpu_caps: vfs capabilities
636 * Extract the on-exec-apply capability sets for an executable file.
638 * If the inode has been found through an idmapped mount the user namespace of
639 * the vfsmount must be passed through @mnt_userns. This function will then
640 * take care to map the inode according to @mnt_userns before checking
641 * permissions. On non-idmapped mounts or if permission checking is to be
642 * performed on the raw inode simply passs init_user_ns.
644 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
645 const struct dentry *dentry,
646 struct cpu_vfs_cap_data *cpu_caps)
648 struct inode *inode = d_backing_inode(dentry);
652 struct vfs_ns_cap_data data, *nscaps = &data;
653 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
655 struct user_namespace *fs_ns;
657 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
662 fs_ns = inode->i_sb->s_user_ns;
663 size = __vfs_getxattr((struct dentry *)dentry, inode,
664 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
665 if (size == -ENODATA || size == -EOPNOTSUPP)
666 /* no data, that's ok */
672 if (size < sizeof(magic_etc))
675 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
677 rootkuid = make_kuid(fs_ns, 0);
678 switch (magic_etc & VFS_CAP_REVISION_MASK) {
679 case VFS_CAP_REVISION_1:
680 if (size != XATTR_CAPS_SZ_1)
682 tocopy = VFS_CAP_U32_1;
684 case VFS_CAP_REVISION_2:
685 if (size != XATTR_CAPS_SZ_2)
687 tocopy = VFS_CAP_U32_2;
689 case VFS_CAP_REVISION_3:
690 if (size != XATTR_CAPS_SZ_3)
692 tocopy = VFS_CAP_U32_3;
693 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
699 /* Limit the caps to the mounter of the filesystem
700 * or the more limited uid specified in the xattr.
702 rootkuid = kuid_into_mnt(mnt_userns, rootkuid);
703 if (!rootid_owns_currentns(rootkuid))
706 CAP_FOR_EACH_U32(i) {
709 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
710 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
713 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
714 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
716 cpu_caps->rootid = rootkuid;
722 * Attempt to get the on-exec apply capability sets for an executable file from
723 * its xattrs and, if present, apply them to the proposed credentials being
724 * constructed by execve().
726 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
727 bool *effective, bool *has_fcap)
730 struct cpu_vfs_cap_data vcaps;
732 cap_clear(bprm->cred->cap_permitted);
734 if (!file_caps_enabled)
737 if (!mnt_may_suid(file->f_path.mnt))
741 * This check is redundant with mnt_may_suid() but is kept to make
742 * explicit that capability bits are limited to s_user_ns and its
745 if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
748 rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
749 file->f_path.dentry, &vcaps);
752 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
754 else if (rc == -ENODATA)
759 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
763 cap_clear(bprm->cred->cap_permitted);
768 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
770 static inline bool __is_real(kuid_t uid, struct cred *cred)
771 { return uid_eq(cred->uid, uid); }
773 static inline bool __is_eff(kuid_t uid, struct cred *cred)
774 { return uid_eq(cred->euid, uid); }
776 static inline bool __is_suid(kuid_t uid, struct cred *cred)
777 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
780 * handle_privileged_root - Handle case of privileged root
781 * @bprm: The execution parameters, including the proposed creds
782 * @has_fcap: Are any file capabilities set?
783 * @effective: Do we have effective root privilege?
784 * @root_uid: This namespace' root UID WRT initial USER namespace
786 * Handle the case where root is privileged and hasn't been neutered by
787 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
788 * set UID root and nothing is changed. If we are root, cap_permitted is
789 * updated. If we have become set UID root, the effective bit is set.
791 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
792 bool *effective, kuid_t root_uid)
794 const struct cred *old = current_cred();
795 struct cred *new = bprm->cred;
797 if (!root_privileged())
800 * If the legacy file capability is set, then don't set privs
801 * for a setuid root binary run by a non-root user. Do set it
802 * for a root user just to cause least surprise to an admin.
804 if (has_fcap && __is_suid(root_uid, new)) {
805 warn_setuid_and_fcaps_mixed(bprm->filename);
809 * To support inheritance of root-permissions and suid-root
810 * executables under compatibility mode, we override the
811 * capability sets for the file.
813 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
814 /* pP' = (cap_bset & ~0) | (pI & ~0) */
815 new->cap_permitted = cap_combine(old->cap_bset,
816 old->cap_inheritable);
819 * If only the real uid is 0, we do not set the effective bit.
821 if (__is_eff(root_uid, new))
825 #define __cap_gained(field, target, source) \
826 !cap_issubset(target->cap_##field, source->cap_##field)
827 #define __cap_grew(target, source, cred) \
828 !cap_issubset(cred->cap_##target, cred->cap_##source)
829 #define __cap_full(field, cred) \
830 cap_issubset(CAP_FULL_SET, cred->cap_##field)
832 static inline bool __is_setuid(struct cred *new, const struct cred *old)
833 { return !uid_eq(new->euid, old->uid); }
835 static inline bool __is_setgid(struct cred *new, const struct cred *old)
836 { return !gid_eq(new->egid, old->gid); }
839 * 1) Audit candidate if current->cap_effective is set
841 * We do not bother to audit if 3 things are true:
842 * 1) cap_effective has all caps
843 * 2) we became root *OR* are were already root
844 * 3) root is supposed to have all caps (SECURE_NOROOT)
845 * Since this is just a normal root execing a process.
847 * Number 1 above might fail if you don't have a full bset, but I think
848 * that is interesting information to audit.
850 * A number of other conditions require logging:
851 * 2) something prevented setuid root getting all caps
852 * 3) non-setuid root gets fcaps
853 * 4) non-setuid root gets ambient
855 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
856 kuid_t root, bool has_fcap)
860 if ((__cap_grew(effective, ambient, new) &&
861 !(__cap_full(effective, new) &&
862 (__is_eff(root, new) || __is_real(root, new)) &&
863 root_privileged())) ||
864 (root_privileged() &&
865 __is_suid(root, new) &&
866 !__cap_full(effective, new)) ||
867 (!__is_setuid(new, old) &&
869 __cap_gained(permitted, new, old)) ||
870 __cap_gained(ambient, new, old))))
878 * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
879 * @bprm: The execution parameters, including the proposed creds
880 * @file: The file to pull the credentials from
882 * Set up the proposed credentials for a new execution context being
883 * constructed by execve(). The proposed creds in @bprm->cred is altered,
884 * which won't take effect immediately. Returns 0 if successful, -ve on error.
886 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
888 /* Process setpcap binaries and capabilities for uid 0 */
889 const struct cred *old = current_cred();
890 struct cred *new = bprm->cred;
891 bool effective = false, has_fcap = false, is_setid;
895 if (WARN_ON(!cap_ambient_invariant_ok(old)))
898 ret = get_file_caps(bprm, file, &effective, &has_fcap);
902 root_uid = make_kuid(new->user_ns, 0);
904 handle_privileged_root(bprm, has_fcap, &effective, root_uid);
906 /* if we have fs caps, clear dangerous personality flags */
907 if (__cap_gained(permitted, new, old))
908 bprm->per_clear |= PER_CLEAR_ON_SETID;
910 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
911 * credentials unless they have the appropriate permit.
913 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
915 is_setid = __is_setuid(new, old) || __is_setgid(new, old);
917 if ((is_setid || __cap_gained(permitted, new, old)) &&
918 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
919 !ptracer_capable(current, new->user_ns))) {
920 /* downgrade; they get no more than they had, and maybe less */
921 if (!ns_capable(new->user_ns, CAP_SETUID) ||
922 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
923 new->euid = new->uid;
924 new->egid = new->gid;
926 new->cap_permitted = cap_intersect(new->cap_permitted,
930 new->suid = new->fsuid = new->euid;
931 new->sgid = new->fsgid = new->egid;
933 /* File caps or setid cancels ambient. */
934 if (has_fcap || is_setid)
935 cap_clear(new->cap_ambient);
938 * Now that we've computed pA', update pP' to give:
939 * pP' = (X & fP) | (pI & fI) | pA'
941 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
944 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
945 * this is the same as pE' = (fE ? pP' : 0) | pA'.
948 new->cap_effective = new->cap_permitted;
950 new->cap_effective = new->cap_ambient;
952 if (WARN_ON(!cap_ambient_invariant_ok(new)))
955 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
956 ret = audit_log_bprm_fcaps(bprm, new, old);
961 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
963 if (WARN_ON(!cap_ambient_invariant_ok(new)))
966 /* Check for privilege-elevated exec. */
968 (!__is_real(root_uid, new) &&
970 __cap_grew(permitted, ambient, new))))
971 bprm->secureexec = 1;
977 * cap_inode_setxattr - Determine whether an xattr may be altered
978 * @dentry: The inode/dentry being altered
979 * @name: The name of the xattr to be changed
980 * @value: The value that the xattr will be changed to
981 * @size: The size of value
982 * @flags: The replacement flag
984 * Determine whether an xattr may be altered or set on an inode, returning 0 if
985 * permission is granted, -ve if denied.
987 * This is used to make sure security xattrs don't get updated or set by those
988 * who aren't privileged to do so.
990 int cap_inode_setxattr(struct dentry *dentry, const char *name,
991 const void *value, size_t size, int flags)
993 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
995 /* Ignore non-security xattrs */
996 if (strncmp(name, XATTR_SECURITY_PREFIX,
997 XATTR_SECURITY_PREFIX_LEN) != 0)
1001 * For XATTR_NAME_CAPS the check will be done in
1002 * cap_convert_nscap(), called by setxattr()
1004 if (strcmp(name, XATTR_NAME_CAPS) == 0)
1007 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1013 * cap_inode_removexattr - Determine whether an xattr may be removed
1015 * @mnt_userns: User namespace of the mount the inode was found from
1016 * @dentry: The inode/dentry being altered
1017 * @name: The name of the xattr to be changed
1019 * Determine whether an xattr may be removed from an inode, returning 0 if
1020 * permission is granted, -ve if denied.
1022 * If the inode has been found through an idmapped mount the user namespace of
1023 * the vfsmount must be passed through @mnt_userns. This function will then
1024 * take care to map the inode according to @mnt_userns before checking
1025 * permissions. On non-idmapped mounts or if permission checking is to be
1026 * performed on the raw inode simply passs init_user_ns.
1028 * This is used to make sure security xattrs don't get removed by those who
1029 * aren't privileged to remove them.
1031 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1032 struct dentry *dentry, const char *name)
1034 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1036 /* Ignore non-security xattrs */
1037 if (strncmp(name, XATTR_SECURITY_PREFIX,
1038 XATTR_SECURITY_PREFIX_LEN) != 0)
1041 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1042 /* security.capability gets namespaced */
1043 struct inode *inode = d_backing_inode(dentry);
1046 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1051 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1057 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1058 * a process after a call to setuid, setreuid, or setresuid.
1060 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1061 * {r,e,s}uid != 0, the permitted and effective capabilities are
1064 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1065 * capabilities of the process are cleared.
1067 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1068 * capabilities are set to the permitted capabilities.
1070 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1075 * cevans - New behaviour, Oct '99
1076 * A process may, via prctl(), elect to keep its capabilities when it
1077 * calls setuid() and switches away from uid==0. Both permitted and
1078 * effective sets will be retained.
1079 * Without this change, it was impossible for a daemon to drop only some
1080 * of its privilege. The call to setuid(!=0) would drop all privileges!
1081 * Keeping uid 0 is not an option because uid 0 owns too many vital
1083 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1085 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1087 kuid_t root_uid = make_kuid(old->user_ns, 0);
1089 if ((uid_eq(old->uid, root_uid) ||
1090 uid_eq(old->euid, root_uid) ||
1091 uid_eq(old->suid, root_uid)) &&
1092 (!uid_eq(new->uid, root_uid) &&
1093 !uid_eq(new->euid, root_uid) &&
1094 !uid_eq(new->suid, root_uid))) {
1095 if (!issecure(SECURE_KEEP_CAPS)) {
1096 cap_clear(new->cap_permitted);
1097 cap_clear(new->cap_effective);
1101 * Pre-ambient programs expect setresuid to nonroot followed
1102 * by exec to drop capabilities. We should make sure that
1103 * this remains the case.
1105 cap_clear(new->cap_ambient);
1107 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1108 cap_clear(new->cap_effective);
1109 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1110 new->cap_effective = new->cap_permitted;
1114 * cap_task_fix_setuid - Fix up the results of setuid() call
1115 * @new: The proposed credentials
1116 * @old: The current task's current credentials
1117 * @flags: Indications of what has changed
1119 * Fix up the results of setuid() call before the credential changes are
1120 * actually applied, returning 0 to grant the changes, -ve to deny them.
1122 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1128 /* juggle the capabilities to follow [RES]UID changes unless
1129 * otherwise suppressed */
1130 if (!issecure(SECURE_NO_SETUID_FIXUP))
1131 cap_emulate_setxuid(new, old);
1135 /* juggle the capabilties to follow FSUID changes, unless
1136 * otherwise suppressed
1138 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1139 * if not, we might be a bit too harsh here.
1141 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1142 kuid_t root_uid = make_kuid(old->user_ns, 0);
1143 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1144 new->cap_effective =
1145 cap_drop_fs_set(new->cap_effective);
1147 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1148 new->cap_effective =
1149 cap_raise_fs_set(new->cap_effective,
1150 new->cap_permitted);
1162 * Rationale: code calling task_setscheduler, task_setioprio, and
1163 * task_setnice, assumes that
1164 * . if capable(cap_sys_nice), then those actions should be allowed
1165 * . if not capable(cap_sys_nice), but acting on your own processes,
1166 * then those actions should be allowed
1167 * This is insufficient now since you can call code without suid, but
1168 * yet with increased caps.
1169 * So we check for increased caps on the target process.
1171 static int cap_safe_nice(struct task_struct *p)
1173 int is_subset, ret = 0;
1176 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1177 current_cred()->cap_permitted);
1178 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1186 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1187 * @p: The task to affect
1189 * Detemine if the requested scheduler policy change is permitted for the
1190 * specified task, returning 0 if permission is granted, -ve if denied.
1192 int cap_task_setscheduler(struct task_struct *p)
1194 return cap_safe_nice(p);
1198 * cap_task_ioprio - Detemine if I/O priority change is permitted
1199 * @p: The task to affect
1200 * @ioprio: The I/O priority to set
1202 * Detemine if the requested I/O priority change is permitted for the specified
1203 * task, returning 0 if permission is granted, -ve if denied.
1205 int cap_task_setioprio(struct task_struct *p, int ioprio)
1207 return cap_safe_nice(p);
1211 * cap_task_ioprio - Detemine if task priority change is permitted
1212 * @p: The task to affect
1213 * @nice: The nice value to set
1215 * Detemine if the requested task priority change is permitted for the
1216 * specified task, returning 0 if permission is granted, -ve if denied.
1218 int cap_task_setnice(struct task_struct *p, int nice)
1220 return cap_safe_nice(p);
1224 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1225 * the current task's bounding set. Returns 0 on success, -ve on error.
1227 static int cap_prctl_drop(unsigned long cap)
1231 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1233 if (!cap_valid(cap))
1236 new = prepare_creds();
1239 cap_lower(new->cap_bset, cap);
1240 return commit_creds(new);
1244 * cap_task_prctl - Implement process control functions for this security module
1245 * @option: The process control function requested
1246 * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1248 * Allow process control functions (sys_prctl()) to alter capabilities; may
1249 * also deny access to other functions not otherwise implemented here.
1251 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1252 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1253 * modules will consider performing the function.
1255 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1256 unsigned long arg4, unsigned long arg5)
1258 const struct cred *old = current_cred();
1262 case PR_CAPBSET_READ:
1263 if (!cap_valid(arg2))
1265 return !!cap_raised(old->cap_bset, arg2);
1267 case PR_CAPBSET_DROP:
1268 return cap_prctl_drop(arg2);
1271 * The next four prctl's remain to assist with transitioning a
1272 * system from legacy UID=0 based privilege (when filesystem
1273 * capabilities are not in use) to a system using filesystem
1274 * capabilities only - as the POSIX.1e draft intended.
1278 * PR_SET_SECUREBITS =
1279 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1280 * | issecure_mask(SECURE_NOROOT)
1281 * | issecure_mask(SECURE_NOROOT_LOCKED)
1282 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1283 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1285 * will ensure that the current process and all of its
1286 * children will be locked into a pure
1287 * capability-based-privilege environment.
1289 case PR_SET_SECUREBITS:
1290 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1291 & (old->securebits ^ arg2)) /*[1]*/
1292 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1293 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1294 || (cap_capable(current_cred(),
1295 current_cred()->user_ns,
1297 CAP_OPT_NONE) != 0) /*[4]*/
1299 * [1] no changing of bits that are locked
1300 * [2] no unlocking of locks
1301 * [3] no setting of unsupported bits
1302 * [4] doing anything requires privilege (go read about
1303 * the "sendmail capabilities bug")
1306 /* cannot change a locked bit */
1309 new = prepare_creds();
1312 new->securebits = arg2;
1313 return commit_creds(new);
1315 case PR_GET_SECUREBITS:
1316 return old->securebits;
1318 case PR_GET_KEEPCAPS:
1319 return !!issecure(SECURE_KEEP_CAPS);
1321 case PR_SET_KEEPCAPS:
1322 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1324 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1327 new = prepare_creds();
1331 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1333 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1334 return commit_creds(new);
1336 case PR_CAP_AMBIENT:
1337 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1338 if (arg3 | arg4 | arg5)
1341 new = prepare_creds();
1344 cap_clear(new->cap_ambient);
1345 return commit_creds(new);
1348 if (((!cap_valid(arg3)) | arg4 | arg5))
1351 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1352 return !!cap_raised(current_cred()->cap_ambient, arg3);
1353 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1354 arg2 != PR_CAP_AMBIENT_LOWER) {
1357 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1358 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1359 !cap_raised(current_cred()->cap_inheritable,
1361 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1364 new = prepare_creds();
1367 if (arg2 == PR_CAP_AMBIENT_RAISE)
1368 cap_raise(new->cap_ambient, arg3);
1370 cap_lower(new->cap_ambient, arg3);
1371 return commit_creds(new);
1375 /* No functionality available - continue with default */
1381 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1382 * @mm: The VM space in which the new mapping is to be made
1383 * @pages: The size of the mapping
1385 * Determine whether the allocation of a new virtual mapping by the current
1386 * task is permitted, returning 1 if permission is granted, 0 if not.
1388 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1390 int cap_sys_admin = 0;
1392 if (cap_capable(current_cred(), &init_user_ns,
1393 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1396 return cap_sys_admin;
1400 * cap_mmap_addr - check if able to map given addr
1401 * @addr: address attempting to be mapped
1403 * If the process is attempting to map memory below dac_mmap_min_addr they need
1404 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1405 * capability security module. Returns 0 if this mapping should be allowed
1408 int cap_mmap_addr(unsigned long addr)
1412 if (addr < dac_mmap_min_addr) {
1413 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1415 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1417 current->flags |= PF_SUPERPRIV;
1422 int cap_mmap_file(struct file *file, unsigned long reqprot,
1423 unsigned long prot, unsigned long flags)
1428 #ifdef CONFIG_SECURITY
1430 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1431 LSM_HOOK_INIT(capable, cap_capable),
1432 LSM_HOOK_INIT(settime, cap_settime),
1433 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1434 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1435 LSM_HOOK_INIT(capget, cap_capget),
1436 LSM_HOOK_INIT(capset, cap_capset),
1437 LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1438 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1439 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1440 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1441 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1442 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1443 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1444 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1445 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1446 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1447 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1448 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1451 static int __init capability_init(void)
1453 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1458 DEFINE_LSM(capability) = {
1459 .name = "capability",
1460 .order = LSM_ORDER_FIRST,
1461 .init = capability_init,
1464 #endif /* CONFIG_SECURITY */