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
306 * @dentry: The inode/dentry to alter
308 * Erase the privilege-enhancing security markings on an inode.
310 * Returns 0 if successful, -ve on error.
312 int cap_inode_killpriv(struct dentry *dentry)
316 error = __vfs_removexattr(&init_user_ns, dentry, XATTR_NAME_CAPS);
317 if (error == -EOPNOTSUPP)
322 static bool rootid_owns_currentns(kuid_t kroot)
324 struct user_namespace *ns;
326 if (!uid_valid(kroot))
329 for (ns = current_user_ns(); ; ns = ns->parent) {
330 if (from_kuid(ns, kroot) == 0)
332 if (ns == &init_user_ns)
339 static __u32 sansflags(__u32 m)
341 return m & ~VFS_CAP_FLAGS_EFFECTIVE;
344 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
346 if (size != XATTR_CAPS_SZ_2)
348 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
351 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
353 if (size != XATTR_CAPS_SZ_3)
355 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
359 * getsecurity: We are called for security.* before any attempt to read the
360 * xattr from the inode itself.
362 * This gives us a chance to read the on-disk value and convert it. If we
363 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
365 * Note we are not called by vfs_getxattr_alloc(), but that is only called
366 * by the integrity subsystem, which really wants the unconverted values -
369 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
374 uid_t root, mappedroot;
376 struct vfs_cap_data *cap;
377 struct vfs_ns_cap_data *nscap;
378 struct dentry *dentry;
379 struct user_namespace *fs_ns;
381 if (strcmp(name, "capability") != 0)
384 dentry = d_find_any_alias(inode);
388 size = sizeof(struct vfs_ns_cap_data);
389 ret = (int)vfs_getxattr_alloc(&init_user_ns, dentry, XATTR_NAME_CAPS,
390 &tmpbuf, size, GFP_NOFS);
396 fs_ns = inode->i_sb->s_user_ns;
397 cap = (struct vfs_cap_data *) tmpbuf;
398 if (is_v2header((size_t) ret, cap)) {
399 /* If this is sizeof(vfs_cap_data) then we're ok with the
400 * on-disk value, so return that. */
406 } else if (!is_v3header((size_t) ret, cap)) {
411 nscap = (struct vfs_ns_cap_data *) tmpbuf;
412 root = le32_to_cpu(nscap->rootid);
413 kroot = make_kuid(fs_ns, root);
415 /* If the root kuid maps to a valid uid in current ns, then return
416 * this as a nscap. */
417 mappedroot = from_kuid(current_user_ns(), kroot);
418 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
421 nscap->rootid = cpu_to_le32(mappedroot);
427 if (!rootid_owns_currentns(kroot)) {
432 /* This comes from a parent namespace. Return as a v2 capability */
433 size = sizeof(struct vfs_cap_data);
435 *buffer = kmalloc(size, GFP_ATOMIC);
437 struct vfs_cap_data *cap = *buffer;
438 __le32 nsmagic, magic;
439 magic = VFS_CAP_REVISION_2;
440 nsmagic = le32_to_cpu(nscap->magic_etc);
441 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
442 magic |= VFS_CAP_FLAGS_EFFECTIVE;
443 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
444 cap->magic_etc = cpu_to_le32(magic);
454 * rootid_from_xattr - translate root uid of vfs caps
456 * @value: vfs caps value which may be modified by this function
457 * @size: size of @ivalue
458 * @task_ns: user namespace of the caller
459 * @mnt_userns: user namespace of the mount the inode was found from
461 * If the inode has been found through an idmapped mount the user namespace of
462 * the vfsmount must be passed through @mnt_userns. This function will then
463 * take care to map the inode according to @mnt_userns before checking
464 * permissions. On non-idmapped mounts or if permission checking is to be
465 * performed on the raw inode simply passs init_user_ns.
467 static kuid_t rootid_from_xattr(const void *value, size_t size,
468 struct user_namespace *task_ns,
469 struct user_namespace *mnt_userns)
471 const struct vfs_ns_cap_data *nscap = value;
475 if (size == XATTR_CAPS_SZ_3)
476 rootid = le32_to_cpu(nscap->rootid);
478 rootkid = make_kuid(task_ns, rootid);
479 return kuid_from_mnt(mnt_userns, rootkid);
482 static bool validheader(size_t size, const struct vfs_cap_data *cap)
484 return is_v2header(size, cap) || is_v3header(size, cap);
488 * cap_convert_nscap - check vfs caps
490 * @mnt_userns: user namespace of the mount the inode was found from
491 * @dentry: used to retrieve inode to check permissions on
492 * @ivalue: vfs caps value which may be modified by this function
493 * @size: size of @ivalue
495 * User requested a write of security.capability. If needed, update the
496 * xattr to change from v2 to v3, or to fixup the v3 rootid.
498 * If the inode has been found through an idmapped mount the user namespace of
499 * the vfsmount must be passed through @mnt_userns. This function will then
500 * take care to map the inode according to @mnt_userns before checking
501 * permissions. On non-idmapped mounts or if permission checking is to be
502 * performed on the raw inode simply passs init_user_ns.
504 * If all is ok, we return the new size, on error return < 0.
506 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
507 const void **ivalue, size_t size)
509 struct vfs_ns_cap_data *nscap;
511 const struct vfs_cap_data *cap = *ivalue;
512 __u32 magic, nsmagic;
513 struct inode *inode = d_backing_inode(dentry);
514 struct user_namespace *task_ns = current_user_ns(),
515 *fs_ns = inode->i_sb->s_user_ns;
521 if (!validheader(size, cap))
523 if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
525 if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
526 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
527 /* user is privileged, just write the v2 */
530 rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
531 if (!uid_valid(rootid))
534 nsrootid = from_kuid(fs_ns, rootid);
538 newsize = sizeof(struct vfs_ns_cap_data);
539 nscap = kmalloc(newsize, GFP_ATOMIC);
542 nscap->rootid = cpu_to_le32(nsrootid);
543 nsmagic = VFS_CAP_REVISION_3;
544 magic = le32_to_cpu(cap->magic_etc);
545 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
546 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
547 nscap->magic_etc = cpu_to_le32(nsmagic);
548 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
555 * Calculate the new process capability sets from the capability sets attached
558 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
559 struct linux_binprm *bprm,
563 struct cred *new = bprm->cred;
567 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
570 if (caps->magic_etc & VFS_CAP_REVISION_MASK)
573 CAP_FOR_EACH_U32(i) {
574 __u32 permitted = caps->permitted.cap[i];
575 __u32 inheritable = caps->inheritable.cap[i];
578 * pP' = (X & fP) | (pI & fI)
579 * The addition of pA' is handled later.
581 new->cap_permitted.cap[i] =
582 (new->cap_bset.cap[i] & permitted) |
583 (new->cap_inheritable.cap[i] & inheritable);
585 if (permitted & ~new->cap_permitted.cap[i])
586 /* insufficient to execute correctly */
591 * For legacy apps, with no internal support for recognizing they
592 * do not have enough capabilities, we return an error if they are
593 * missing some "forced" (aka file-permitted) capabilities.
595 return *effective ? ret : 0;
599 * Extract the on-exec-apply capability sets for an executable file.
601 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
603 struct inode *inode = d_backing_inode(dentry);
607 struct vfs_ns_cap_data data, *nscaps = &data;
608 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
610 struct user_namespace *fs_ns;
612 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
617 fs_ns = inode->i_sb->s_user_ns;
618 size = __vfs_getxattr((struct dentry *)dentry, inode,
619 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
620 if (size == -ENODATA || size == -EOPNOTSUPP)
621 /* no data, that's ok */
627 if (size < sizeof(magic_etc))
630 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
632 rootkuid = make_kuid(fs_ns, 0);
633 switch (magic_etc & VFS_CAP_REVISION_MASK) {
634 case VFS_CAP_REVISION_1:
635 if (size != XATTR_CAPS_SZ_1)
637 tocopy = VFS_CAP_U32_1;
639 case VFS_CAP_REVISION_2:
640 if (size != XATTR_CAPS_SZ_2)
642 tocopy = VFS_CAP_U32_2;
644 case VFS_CAP_REVISION_3:
645 if (size != XATTR_CAPS_SZ_3)
647 tocopy = VFS_CAP_U32_3;
648 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
654 /* Limit the caps to the mounter of the filesystem
655 * or the more limited uid specified in the xattr.
657 if (!rootid_owns_currentns(rootkuid))
660 CAP_FOR_EACH_U32(i) {
663 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
664 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
667 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
668 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
670 cpu_caps->rootid = rootkuid;
676 * Attempt to get the on-exec apply capability sets for an executable file from
677 * its xattrs and, if present, apply them to the proposed credentials being
678 * constructed by execve().
680 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
681 bool *effective, bool *has_fcap)
684 struct cpu_vfs_cap_data vcaps;
686 cap_clear(bprm->cred->cap_permitted);
688 if (!file_caps_enabled)
691 if (!mnt_may_suid(file->f_path.mnt))
695 * This check is redundant with mnt_may_suid() but is kept to make
696 * explicit that capability bits are limited to s_user_ns and its
699 if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
702 rc = get_vfs_caps_from_disk(file->f_path.dentry, &vcaps);
705 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
707 else if (rc == -ENODATA)
712 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
716 cap_clear(bprm->cred->cap_permitted);
721 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
723 static inline bool __is_real(kuid_t uid, struct cred *cred)
724 { return uid_eq(cred->uid, uid); }
726 static inline bool __is_eff(kuid_t uid, struct cred *cred)
727 { return uid_eq(cred->euid, uid); }
729 static inline bool __is_suid(kuid_t uid, struct cred *cred)
730 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
733 * handle_privileged_root - Handle case of privileged root
734 * @bprm: The execution parameters, including the proposed creds
735 * @has_fcap: Are any file capabilities set?
736 * @effective: Do we have effective root privilege?
737 * @root_uid: This namespace' root UID WRT initial USER namespace
739 * Handle the case where root is privileged and hasn't been neutered by
740 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
741 * set UID root and nothing is changed. If we are root, cap_permitted is
742 * updated. If we have become set UID root, the effective bit is set.
744 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
745 bool *effective, kuid_t root_uid)
747 const struct cred *old = current_cred();
748 struct cred *new = bprm->cred;
750 if (!root_privileged())
753 * If the legacy file capability is set, then don't set privs
754 * for a setuid root binary run by a non-root user. Do set it
755 * for a root user just to cause least surprise to an admin.
757 if (has_fcap && __is_suid(root_uid, new)) {
758 warn_setuid_and_fcaps_mixed(bprm->filename);
762 * To support inheritance of root-permissions and suid-root
763 * executables under compatibility mode, we override the
764 * capability sets for the file.
766 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
767 /* pP' = (cap_bset & ~0) | (pI & ~0) */
768 new->cap_permitted = cap_combine(old->cap_bset,
769 old->cap_inheritable);
772 * If only the real uid is 0, we do not set the effective bit.
774 if (__is_eff(root_uid, new))
778 #define __cap_gained(field, target, source) \
779 !cap_issubset(target->cap_##field, source->cap_##field)
780 #define __cap_grew(target, source, cred) \
781 !cap_issubset(cred->cap_##target, cred->cap_##source)
782 #define __cap_full(field, cred) \
783 cap_issubset(CAP_FULL_SET, cred->cap_##field)
785 static inline bool __is_setuid(struct cred *new, const struct cred *old)
786 { return !uid_eq(new->euid, old->uid); }
788 static inline bool __is_setgid(struct cred *new, const struct cred *old)
789 { return !gid_eq(new->egid, old->gid); }
792 * 1) Audit candidate if current->cap_effective is set
794 * We do not bother to audit if 3 things are true:
795 * 1) cap_effective has all caps
796 * 2) we became root *OR* are were already root
797 * 3) root is supposed to have all caps (SECURE_NOROOT)
798 * Since this is just a normal root execing a process.
800 * Number 1 above might fail if you don't have a full bset, but I think
801 * that is interesting information to audit.
803 * A number of other conditions require logging:
804 * 2) something prevented setuid root getting all caps
805 * 3) non-setuid root gets fcaps
806 * 4) non-setuid root gets ambient
808 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
809 kuid_t root, bool has_fcap)
813 if ((__cap_grew(effective, ambient, new) &&
814 !(__cap_full(effective, new) &&
815 (__is_eff(root, new) || __is_real(root, new)) &&
816 root_privileged())) ||
817 (root_privileged() &&
818 __is_suid(root, new) &&
819 !__cap_full(effective, new)) ||
820 (!__is_setuid(new, old) &&
822 __cap_gained(permitted, new, old)) ||
823 __cap_gained(ambient, new, old))))
831 * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
832 * @bprm: The execution parameters, including the proposed creds
833 * @file: The file to pull the credentials from
835 * Set up the proposed credentials for a new execution context being
836 * constructed by execve(). The proposed creds in @bprm->cred is altered,
837 * which won't take effect immediately. Returns 0 if successful, -ve on error.
839 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
841 /* Process setpcap binaries and capabilities for uid 0 */
842 const struct cred *old = current_cred();
843 struct cred *new = bprm->cred;
844 bool effective = false, has_fcap = false, is_setid;
848 if (WARN_ON(!cap_ambient_invariant_ok(old)))
851 ret = get_file_caps(bprm, file, &effective, &has_fcap);
855 root_uid = make_kuid(new->user_ns, 0);
857 handle_privileged_root(bprm, has_fcap, &effective, root_uid);
859 /* if we have fs caps, clear dangerous personality flags */
860 if (__cap_gained(permitted, new, old))
861 bprm->per_clear |= PER_CLEAR_ON_SETID;
863 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
864 * credentials unless they have the appropriate permit.
866 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
868 is_setid = __is_setuid(new, old) || __is_setgid(new, old);
870 if ((is_setid || __cap_gained(permitted, new, old)) &&
871 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
872 !ptracer_capable(current, new->user_ns))) {
873 /* downgrade; they get no more than they had, and maybe less */
874 if (!ns_capable(new->user_ns, CAP_SETUID) ||
875 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
876 new->euid = new->uid;
877 new->egid = new->gid;
879 new->cap_permitted = cap_intersect(new->cap_permitted,
883 new->suid = new->fsuid = new->euid;
884 new->sgid = new->fsgid = new->egid;
886 /* File caps or setid cancels ambient. */
887 if (has_fcap || is_setid)
888 cap_clear(new->cap_ambient);
891 * Now that we've computed pA', update pP' to give:
892 * pP' = (X & fP) | (pI & fI) | pA'
894 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
897 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
898 * this is the same as pE' = (fE ? pP' : 0) | pA'.
901 new->cap_effective = new->cap_permitted;
903 new->cap_effective = new->cap_ambient;
905 if (WARN_ON(!cap_ambient_invariant_ok(new)))
908 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
909 ret = audit_log_bprm_fcaps(bprm, new, old);
914 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
916 if (WARN_ON(!cap_ambient_invariant_ok(new)))
919 /* Check for privilege-elevated exec. */
921 (!__is_real(root_uid, new) &&
923 __cap_grew(permitted, ambient, new))))
924 bprm->secureexec = 1;
930 * cap_inode_setxattr - Determine whether an xattr may be altered
931 * @dentry: The inode/dentry being altered
932 * @name: The name of the xattr to be changed
933 * @value: The value that the xattr will be changed to
934 * @size: The size of value
935 * @flags: The replacement flag
937 * Determine whether an xattr may be altered or set on an inode, returning 0 if
938 * permission is granted, -ve if denied.
940 * This is used to make sure security xattrs don't get updated or set by those
941 * who aren't privileged to do so.
943 int cap_inode_setxattr(struct dentry *dentry, const char *name,
944 const void *value, size_t size, int flags)
946 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
948 /* Ignore non-security xattrs */
949 if (strncmp(name, XATTR_SECURITY_PREFIX,
950 XATTR_SECURITY_PREFIX_LEN) != 0)
954 * For XATTR_NAME_CAPS the check will be done in
955 * cap_convert_nscap(), called by setxattr()
957 if (strcmp(name, XATTR_NAME_CAPS) == 0)
960 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
966 * cap_inode_removexattr - Determine whether an xattr may be removed
967 * @dentry: The inode/dentry being altered
968 * @name: The name of the xattr to be changed
970 * Determine whether an xattr may be removed from an inode, returning 0 if
971 * permission is granted, -ve if denied.
973 * This is used to make sure security xattrs don't get removed by those who
974 * aren't privileged to remove them.
976 int cap_inode_removexattr(struct dentry *dentry, const char *name)
978 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
980 /* Ignore non-security xattrs */
981 if (strncmp(name, XATTR_SECURITY_PREFIX,
982 XATTR_SECURITY_PREFIX_LEN) != 0)
985 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
986 /* security.capability gets namespaced */
987 struct inode *inode = d_backing_inode(dentry);
990 if (!capable_wrt_inode_uidgid(&init_user_ns, inode,
996 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1002 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1003 * a process after a call to setuid, setreuid, or setresuid.
1005 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1006 * {r,e,s}uid != 0, the permitted and effective capabilities are
1009 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1010 * capabilities of the process are cleared.
1012 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1013 * capabilities are set to the permitted capabilities.
1015 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1020 * cevans - New behaviour, Oct '99
1021 * A process may, via prctl(), elect to keep its capabilities when it
1022 * calls setuid() and switches away from uid==0. Both permitted and
1023 * effective sets will be retained.
1024 * Without this change, it was impossible for a daemon to drop only some
1025 * of its privilege. The call to setuid(!=0) would drop all privileges!
1026 * Keeping uid 0 is not an option because uid 0 owns too many vital
1028 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1030 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1032 kuid_t root_uid = make_kuid(old->user_ns, 0);
1034 if ((uid_eq(old->uid, root_uid) ||
1035 uid_eq(old->euid, root_uid) ||
1036 uid_eq(old->suid, root_uid)) &&
1037 (!uid_eq(new->uid, root_uid) &&
1038 !uid_eq(new->euid, root_uid) &&
1039 !uid_eq(new->suid, root_uid))) {
1040 if (!issecure(SECURE_KEEP_CAPS)) {
1041 cap_clear(new->cap_permitted);
1042 cap_clear(new->cap_effective);
1046 * Pre-ambient programs expect setresuid to nonroot followed
1047 * by exec to drop capabilities. We should make sure that
1048 * this remains the case.
1050 cap_clear(new->cap_ambient);
1052 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1053 cap_clear(new->cap_effective);
1054 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1055 new->cap_effective = new->cap_permitted;
1059 * cap_task_fix_setuid - Fix up the results of setuid() call
1060 * @new: The proposed credentials
1061 * @old: The current task's current credentials
1062 * @flags: Indications of what has changed
1064 * Fix up the results of setuid() call before the credential changes are
1065 * actually applied, returning 0 to grant the changes, -ve to deny them.
1067 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1073 /* juggle the capabilities to follow [RES]UID changes unless
1074 * otherwise suppressed */
1075 if (!issecure(SECURE_NO_SETUID_FIXUP))
1076 cap_emulate_setxuid(new, old);
1080 /* juggle the capabilties to follow FSUID changes, unless
1081 * otherwise suppressed
1083 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1084 * if not, we might be a bit too harsh here.
1086 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1087 kuid_t root_uid = make_kuid(old->user_ns, 0);
1088 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1089 new->cap_effective =
1090 cap_drop_fs_set(new->cap_effective);
1092 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1093 new->cap_effective =
1094 cap_raise_fs_set(new->cap_effective,
1095 new->cap_permitted);
1107 * Rationale: code calling task_setscheduler, task_setioprio, and
1108 * task_setnice, assumes that
1109 * . if capable(cap_sys_nice), then those actions should be allowed
1110 * . if not capable(cap_sys_nice), but acting on your own processes,
1111 * then those actions should be allowed
1112 * This is insufficient now since you can call code without suid, but
1113 * yet with increased caps.
1114 * So we check for increased caps on the target process.
1116 static int cap_safe_nice(struct task_struct *p)
1118 int is_subset, ret = 0;
1121 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1122 current_cred()->cap_permitted);
1123 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1131 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1132 * @p: The task to affect
1134 * Detemine if the requested scheduler policy change is permitted for the
1135 * specified task, returning 0 if permission is granted, -ve if denied.
1137 int cap_task_setscheduler(struct task_struct *p)
1139 return cap_safe_nice(p);
1143 * cap_task_ioprio - Detemine if I/O priority change is permitted
1144 * @p: The task to affect
1145 * @ioprio: The I/O priority to set
1147 * Detemine if the requested I/O priority change is permitted for the specified
1148 * task, returning 0 if permission is granted, -ve if denied.
1150 int cap_task_setioprio(struct task_struct *p, int ioprio)
1152 return cap_safe_nice(p);
1156 * cap_task_ioprio - Detemine if task priority change is permitted
1157 * @p: The task to affect
1158 * @nice: The nice value to set
1160 * Detemine if the requested task priority change is permitted for the
1161 * specified task, returning 0 if permission is granted, -ve if denied.
1163 int cap_task_setnice(struct task_struct *p, int nice)
1165 return cap_safe_nice(p);
1169 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1170 * the current task's bounding set. Returns 0 on success, -ve on error.
1172 static int cap_prctl_drop(unsigned long cap)
1176 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1178 if (!cap_valid(cap))
1181 new = prepare_creds();
1184 cap_lower(new->cap_bset, cap);
1185 return commit_creds(new);
1189 * cap_task_prctl - Implement process control functions for this security module
1190 * @option: The process control function requested
1191 * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1193 * Allow process control functions (sys_prctl()) to alter capabilities; may
1194 * also deny access to other functions not otherwise implemented here.
1196 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1197 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1198 * modules will consider performing the function.
1200 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1201 unsigned long arg4, unsigned long arg5)
1203 const struct cred *old = current_cred();
1207 case PR_CAPBSET_READ:
1208 if (!cap_valid(arg2))
1210 return !!cap_raised(old->cap_bset, arg2);
1212 case PR_CAPBSET_DROP:
1213 return cap_prctl_drop(arg2);
1216 * The next four prctl's remain to assist with transitioning a
1217 * system from legacy UID=0 based privilege (when filesystem
1218 * capabilities are not in use) to a system using filesystem
1219 * capabilities only - as the POSIX.1e draft intended.
1223 * PR_SET_SECUREBITS =
1224 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1225 * | issecure_mask(SECURE_NOROOT)
1226 * | issecure_mask(SECURE_NOROOT_LOCKED)
1227 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1228 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1230 * will ensure that the current process and all of its
1231 * children will be locked into a pure
1232 * capability-based-privilege environment.
1234 case PR_SET_SECUREBITS:
1235 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1236 & (old->securebits ^ arg2)) /*[1]*/
1237 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1238 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1239 || (cap_capable(current_cred(),
1240 current_cred()->user_ns,
1242 CAP_OPT_NONE) != 0) /*[4]*/
1244 * [1] no changing of bits that are locked
1245 * [2] no unlocking of locks
1246 * [3] no setting of unsupported bits
1247 * [4] doing anything requires privilege (go read about
1248 * the "sendmail capabilities bug")
1251 /* cannot change a locked bit */
1254 new = prepare_creds();
1257 new->securebits = arg2;
1258 return commit_creds(new);
1260 case PR_GET_SECUREBITS:
1261 return old->securebits;
1263 case PR_GET_KEEPCAPS:
1264 return !!issecure(SECURE_KEEP_CAPS);
1266 case PR_SET_KEEPCAPS:
1267 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1269 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1272 new = prepare_creds();
1276 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1278 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1279 return commit_creds(new);
1281 case PR_CAP_AMBIENT:
1282 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1283 if (arg3 | arg4 | arg5)
1286 new = prepare_creds();
1289 cap_clear(new->cap_ambient);
1290 return commit_creds(new);
1293 if (((!cap_valid(arg3)) | arg4 | arg5))
1296 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1297 return !!cap_raised(current_cred()->cap_ambient, arg3);
1298 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1299 arg2 != PR_CAP_AMBIENT_LOWER) {
1302 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1303 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1304 !cap_raised(current_cred()->cap_inheritable,
1306 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1309 new = prepare_creds();
1312 if (arg2 == PR_CAP_AMBIENT_RAISE)
1313 cap_raise(new->cap_ambient, arg3);
1315 cap_lower(new->cap_ambient, arg3);
1316 return commit_creds(new);
1320 /* No functionality available - continue with default */
1326 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1327 * @mm: The VM space in which the new mapping is to be made
1328 * @pages: The size of the mapping
1330 * Determine whether the allocation of a new virtual mapping by the current
1331 * task is permitted, returning 1 if permission is granted, 0 if not.
1333 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1335 int cap_sys_admin = 0;
1337 if (cap_capable(current_cred(), &init_user_ns,
1338 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1341 return cap_sys_admin;
1345 * cap_mmap_addr - check if able to map given addr
1346 * @addr: address attempting to be mapped
1348 * If the process is attempting to map memory below dac_mmap_min_addr they need
1349 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1350 * capability security module. Returns 0 if this mapping should be allowed
1353 int cap_mmap_addr(unsigned long addr)
1357 if (addr < dac_mmap_min_addr) {
1358 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1360 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1362 current->flags |= PF_SUPERPRIV;
1367 int cap_mmap_file(struct file *file, unsigned long reqprot,
1368 unsigned long prot, unsigned long flags)
1373 #ifdef CONFIG_SECURITY
1375 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1376 LSM_HOOK_INIT(capable, cap_capable),
1377 LSM_HOOK_INIT(settime, cap_settime),
1378 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1379 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1380 LSM_HOOK_INIT(capget, cap_capget),
1381 LSM_HOOK_INIT(capset, cap_capset),
1382 LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1383 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1384 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1385 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1386 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1387 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1388 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1389 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1390 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1391 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1392 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1393 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1396 static int __init capability_init(void)
1398 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1403 DEFINE_LSM(capability) = {
1404 .name = "capability",
1405 .order = LSM_ORDER_FIRST,
1406 .init = capability_init,
1409 #endif /* CONFIG_SECURITY */