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(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,
375 uid_t root, mappedroot;
377 struct vfs_cap_data *cap;
378 struct vfs_ns_cap_data *nscap = NULL;
379 struct dentry *dentry;
380 struct user_namespace *fs_ns;
382 if (strcmp(name, "capability") != 0)
385 dentry = d_find_any_alias(inode);
389 size = sizeof(struct vfs_ns_cap_data);
390 ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS,
391 &tmpbuf, size, GFP_NOFS);
394 if (ret < 0 || !tmpbuf) {
399 fs_ns = inode->i_sb->s_user_ns;
400 cap = (struct vfs_cap_data *) tmpbuf;
401 if (is_v2header((size_t) ret, cap)) {
403 } else if (is_v3header((size_t) ret, cap)) {
404 nscap = (struct vfs_ns_cap_data *) tmpbuf;
405 root = le32_to_cpu(nscap->rootid);
411 kroot = make_kuid(fs_ns, root);
413 /* If the root kuid maps to a valid uid in current ns, then return
414 * this as a nscap. */
415 mappedroot = from_kuid(current_user_ns(), kroot);
416 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
417 size = sizeof(struct vfs_ns_cap_data);
420 /* v2 -> v3 conversion */
421 nscap = kzalloc(size, GFP_ATOMIC);
426 nsmagic = VFS_CAP_REVISION_3;
427 magic = le32_to_cpu(cap->magic_etc);
428 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
429 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
430 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
431 nscap->magic_etc = cpu_to_le32(nsmagic);
433 /* use allocated v3 buffer */
436 nscap->rootid = cpu_to_le32(mappedroot);
442 if (!rootid_owns_currentns(kroot)) {
447 /* This comes from a parent namespace. Return as a v2 capability */
448 size = sizeof(struct vfs_cap_data);
451 /* v3 -> v2 conversion */
452 cap = kzalloc(size, GFP_ATOMIC);
457 magic = VFS_CAP_REVISION_2;
458 nsmagic = le32_to_cpu(nscap->magic_etc);
459 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
460 magic |= VFS_CAP_FLAGS_EFFECTIVE;
461 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
462 cap->magic_etc = cpu_to_le32(magic);
464 /* use unconverted v2 */
474 static kuid_t rootid_from_xattr(const void *value, size_t size,
475 struct user_namespace *task_ns)
477 const struct vfs_ns_cap_data *nscap = value;
480 if (size == XATTR_CAPS_SZ_3)
481 rootid = le32_to_cpu(nscap->rootid);
483 return make_kuid(task_ns, rootid);
486 static bool validheader(size_t size, const struct vfs_cap_data *cap)
488 return is_v2header(size, cap) || is_v3header(size, cap);
492 * User requested a write of security.capability. If needed, update the
493 * xattr to change from v2 to v3, or to fixup the v3 rootid.
495 * If all is ok, we return the new size, on error return < 0.
497 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size)
499 struct vfs_ns_cap_data *nscap;
501 const struct vfs_cap_data *cap = *ivalue;
502 __u32 magic, nsmagic;
503 struct inode *inode = d_backing_inode(dentry);
504 struct user_namespace *task_ns = current_user_ns(),
505 *fs_ns = inode->i_sb->s_user_ns;
511 if (!validheader(size, cap))
513 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
515 if (size == XATTR_CAPS_SZ_2)
516 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
517 /* user is privileged, just write the v2 */
520 rootid = rootid_from_xattr(*ivalue, size, task_ns);
521 if (!uid_valid(rootid))
524 nsrootid = from_kuid(fs_ns, rootid);
528 newsize = sizeof(struct vfs_ns_cap_data);
529 nscap = kmalloc(newsize, GFP_ATOMIC);
532 nscap->rootid = cpu_to_le32(nsrootid);
533 nsmagic = VFS_CAP_REVISION_3;
534 magic = le32_to_cpu(cap->magic_etc);
535 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
536 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
537 nscap->magic_etc = cpu_to_le32(nsmagic);
538 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
546 * Calculate the new process capability sets from the capability sets attached
549 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
550 struct linux_binprm *bprm,
554 struct cred *new = bprm->cred;
558 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
561 if (caps->magic_etc & VFS_CAP_REVISION_MASK)
564 CAP_FOR_EACH_U32(i) {
565 __u32 permitted = caps->permitted.cap[i];
566 __u32 inheritable = caps->inheritable.cap[i];
569 * pP' = (X & fP) | (pI & fI)
570 * The addition of pA' is handled later.
572 new->cap_permitted.cap[i] =
573 (new->cap_bset.cap[i] & permitted) |
574 (new->cap_inheritable.cap[i] & inheritable);
576 if (permitted & ~new->cap_permitted.cap[i])
577 /* insufficient to execute correctly */
582 * For legacy apps, with no internal support for recognizing they
583 * do not have enough capabilities, we return an error if they are
584 * missing some "forced" (aka file-permitted) capabilities.
586 return *effective ? ret : 0;
590 * Extract the on-exec-apply capability sets for an executable file.
592 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
594 struct inode *inode = d_backing_inode(dentry);
598 struct vfs_ns_cap_data data, *nscaps = &data;
599 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
601 struct user_namespace *fs_ns;
603 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
608 fs_ns = inode->i_sb->s_user_ns;
609 size = __vfs_getxattr((struct dentry *)dentry, inode,
610 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
611 if (size == -ENODATA || size == -EOPNOTSUPP)
612 /* no data, that's ok */
618 if (size < sizeof(magic_etc))
621 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
623 rootkuid = make_kuid(fs_ns, 0);
624 switch (magic_etc & VFS_CAP_REVISION_MASK) {
625 case VFS_CAP_REVISION_1:
626 if (size != XATTR_CAPS_SZ_1)
628 tocopy = VFS_CAP_U32_1;
630 case VFS_CAP_REVISION_2:
631 if (size != XATTR_CAPS_SZ_2)
633 tocopy = VFS_CAP_U32_2;
635 case VFS_CAP_REVISION_3:
636 if (size != XATTR_CAPS_SZ_3)
638 tocopy = VFS_CAP_U32_3;
639 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
645 /* Limit the caps to the mounter of the filesystem
646 * or the more limited uid specified in the xattr.
648 if (!rootid_owns_currentns(rootkuid))
651 CAP_FOR_EACH_U32(i) {
654 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
655 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
658 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
659 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
661 cpu_caps->rootid = rootkuid;
667 * Attempt to get the on-exec apply capability sets for an executable file from
668 * its xattrs and, if present, apply them to the proposed credentials being
669 * constructed by execve().
671 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_fcap)
674 struct cpu_vfs_cap_data vcaps;
676 cap_clear(bprm->cred->cap_permitted);
678 if (!file_caps_enabled)
681 if (!mnt_may_suid(bprm->file->f_path.mnt))
685 * This check is redundant with mnt_may_suid() but is kept to make
686 * explicit that capability bits are limited to s_user_ns and its
689 if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
692 rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
695 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
697 else if (rc == -ENODATA)
702 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
706 cap_clear(bprm->cred->cap_permitted);
711 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
713 static inline bool __is_real(kuid_t uid, struct cred *cred)
714 { return uid_eq(cred->uid, uid); }
716 static inline bool __is_eff(kuid_t uid, struct cred *cred)
717 { return uid_eq(cred->euid, uid); }
719 static inline bool __is_suid(kuid_t uid, struct cred *cred)
720 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
723 * handle_privileged_root - Handle case of privileged root
724 * @bprm: The execution parameters, including the proposed creds
725 * @has_fcap: Are any file capabilities set?
726 * @effective: Do we have effective root privilege?
727 * @root_uid: This namespace' root UID WRT initial USER namespace
729 * Handle the case where root is privileged and hasn't been neutered by
730 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
731 * set UID root and nothing is changed. If we are root, cap_permitted is
732 * updated. If we have become set UID root, the effective bit is set.
734 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
735 bool *effective, kuid_t root_uid)
737 const struct cred *old = current_cred();
738 struct cred *new = bprm->cred;
740 if (!root_privileged())
743 * If the legacy file capability is set, then don't set privs
744 * for a setuid root binary run by a non-root user. Do set it
745 * for a root user just to cause least surprise to an admin.
747 if (has_fcap && __is_suid(root_uid, new)) {
748 warn_setuid_and_fcaps_mixed(bprm->filename);
752 * To support inheritance of root-permissions and suid-root
753 * executables under compatibility mode, we override the
754 * capability sets for the file.
756 if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
757 /* pP' = (cap_bset & ~0) | (pI & ~0) */
758 new->cap_permitted = cap_combine(old->cap_bset,
759 old->cap_inheritable);
762 * If only the real uid is 0, we do not set the effective bit.
764 if (__is_eff(root_uid, new))
768 #define __cap_gained(field, target, source) \
769 !cap_issubset(target->cap_##field, source->cap_##field)
770 #define __cap_grew(target, source, cred) \
771 !cap_issubset(cred->cap_##target, cred->cap_##source)
772 #define __cap_full(field, cred) \
773 cap_issubset(CAP_FULL_SET, cred->cap_##field)
775 static inline bool __is_setuid(struct cred *new, const struct cred *old)
776 { return !uid_eq(new->euid, old->uid); }
778 static inline bool __is_setgid(struct cred *new, const struct cred *old)
779 { return !gid_eq(new->egid, old->gid); }
782 * 1) Audit candidate if current->cap_effective is set
784 * We do not bother to audit if 3 things are true:
785 * 1) cap_effective has all caps
786 * 2) we became root *OR* are were already root
787 * 3) root is supposed to have all caps (SECURE_NOROOT)
788 * Since this is just a normal root execing a process.
790 * Number 1 above might fail if you don't have a full bset, but I think
791 * that is interesting information to audit.
793 * A number of other conditions require logging:
794 * 2) something prevented setuid root getting all caps
795 * 3) non-setuid root gets fcaps
796 * 4) non-setuid root gets ambient
798 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
799 kuid_t root, bool has_fcap)
803 if ((__cap_grew(effective, ambient, new) &&
804 !(__cap_full(effective, new) &&
805 (__is_eff(root, new) || __is_real(root, new)) &&
806 root_privileged())) ||
807 (root_privileged() &&
808 __is_suid(root, new) &&
809 !__cap_full(effective, new)) ||
810 (!__is_setuid(new, old) &&
812 __cap_gained(permitted, new, old)) ||
813 __cap_gained(ambient, new, old))))
821 * cap_bprm_set_creds - Set up the proposed credentials for execve().
822 * @bprm: The execution parameters, including the proposed creds
824 * Set up the proposed credentials for a new execution context being
825 * constructed by execve(). The proposed creds in @bprm->cred is altered,
826 * which won't take effect immediately. Returns 0 if successful, -ve on error.
828 int cap_bprm_set_creds(struct linux_binprm *bprm)
830 const struct cred *old = current_cred();
831 struct cred *new = bprm->cred;
832 bool effective = false, has_fcap = false, is_setid;
836 new->cap_ambient = old->cap_ambient;
837 if (WARN_ON(!cap_ambient_invariant_ok(old)))
840 ret = get_file_caps(bprm, &effective, &has_fcap);
844 root_uid = make_kuid(new->user_ns, 0);
846 handle_privileged_root(bprm, has_fcap, &effective, root_uid);
848 /* if we have fs caps, clear dangerous personality flags */
849 if (__cap_gained(permitted, new, old))
850 bprm->per_clear |= PER_CLEAR_ON_SETID;
852 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
853 * credentials unless they have the appropriate permit.
855 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
857 is_setid = __is_setuid(new, old) || __is_setgid(new, old);
859 if ((is_setid || __cap_gained(permitted, new, old)) &&
860 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
861 !ptracer_capable(current, new->user_ns))) {
862 /* downgrade; they get no more than they had, and maybe less */
863 if (!ns_capable(new->user_ns, CAP_SETUID) ||
864 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
865 new->euid = new->uid;
866 new->egid = new->gid;
868 new->cap_permitted = cap_intersect(new->cap_permitted,
872 new->suid = new->fsuid = new->euid;
873 new->sgid = new->fsgid = new->egid;
875 /* File caps or setid cancels ambient. */
876 if (has_fcap || is_setid)
877 cap_clear(new->cap_ambient);
880 * Now that we've computed pA', update pP' to give:
881 * pP' = (X & fP) | (pI & fI) | pA'
883 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
886 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
887 * this is the same as pE' = (fE ? pP' : 0) | pA'.
890 new->cap_effective = new->cap_permitted;
892 new->cap_effective = new->cap_ambient;
894 if (WARN_ON(!cap_ambient_invariant_ok(new)))
897 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
898 ret = audit_log_bprm_fcaps(bprm, new, old);
903 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
905 if (WARN_ON(!cap_ambient_invariant_ok(new)))
908 /* Check for privilege-elevated exec. */
909 bprm->cap_elevated = 0;
911 (!__is_real(root_uid, new) &&
913 __cap_grew(permitted, ambient, new))))
914 bprm->cap_elevated = 1;
920 * cap_inode_setxattr - Determine whether an xattr may be altered
921 * @dentry: The inode/dentry being altered
922 * @name: The name of the xattr to be changed
923 * @value: The value that the xattr will be changed to
924 * @size: The size of value
925 * @flags: The replacement flag
927 * Determine whether an xattr may be altered or set on an inode, returning 0 if
928 * permission is granted, -ve if denied.
930 * This is used to make sure security xattrs don't get updated or set by those
931 * who aren't privileged to do so.
933 int cap_inode_setxattr(struct dentry *dentry, const char *name,
934 const void *value, size_t size, int flags)
936 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
938 /* Ignore non-security xattrs */
939 if (strncmp(name, XATTR_SECURITY_PREFIX,
940 XATTR_SECURITY_PREFIX_LEN) != 0)
944 * For XATTR_NAME_CAPS the check will be done in
945 * cap_convert_nscap(), called by setxattr()
947 if (strcmp(name, XATTR_NAME_CAPS) == 0)
950 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
956 * cap_inode_removexattr - Determine whether an xattr may be removed
957 * @dentry: The inode/dentry being altered
958 * @name: The name of the xattr to be changed
960 * Determine whether an xattr may be removed from an inode, returning 0 if
961 * permission is granted, -ve if denied.
963 * This is used to make sure security xattrs don't get removed by those who
964 * aren't privileged to remove them.
966 int cap_inode_removexattr(struct dentry *dentry, const char *name)
968 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
970 /* Ignore non-security xattrs */
971 if (strncmp(name, XATTR_SECURITY_PREFIX,
972 XATTR_SECURITY_PREFIX_LEN) != 0)
975 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
976 /* security.capability gets namespaced */
977 struct inode *inode = d_backing_inode(dentry);
980 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
985 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
991 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
992 * a process after a call to setuid, setreuid, or setresuid.
994 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
995 * {r,e,s}uid != 0, the permitted and effective capabilities are
998 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
999 * capabilities of the process are cleared.
1001 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1002 * capabilities are set to the permitted capabilities.
1004 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1009 * cevans - New behaviour, Oct '99
1010 * A process may, via prctl(), elect to keep its capabilities when it
1011 * calls setuid() and switches away from uid==0. Both permitted and
1012 * effective sets will be retained.
1013 * Without this change, it was impossible for a daemon to drop only some
1014 * of its privilege. The call to setuid(!=0) would drop all privileges!
1015 * Keeping uid 0 is not an option because uid 0 owns too many vital
1017 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1019 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1021 kuid_t root_uid = make_kuid(old->user_ns, 0);
1023 if ((uid_eq(old->uid, root_uid) ||
1024 uid_eq(old->euid, root_uid) ||
1025 uid_eq(old->suid, root_uid)) &&
1026 (!uid_eq(new->uid, root_uid) &&
1027 !uid_eq(new->euid, root_uid) &&
1028 !uid_eq(new->suid, root_uid))) {
1029 if (!issecure(SECURE_KEEP_CAPS)) {
1030 cap_clear(new->cap_permitted);
1031 cap_clear(new->cap_effective);
1035 * Pre-ambient programs expect setresuid to nonroot followed
1036 * by exec to drop capabilities. We should make sure that
1037 * this remains the case.
1039 cap_clear(new->cap_ambient);
1041 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1042 cap_clear(new->cap_effective);
1043 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1044 new->cap_effective = new->cap_permitted;
1048 * cap_task_fix_setuid - Fix up the results of setuid() call
1049 * @new: The proposed credentials
1050 * @old: The current task's current credentials
1051 * @flags: Indications of what has changed
1053 * Fix up the results of setuid() call before the credential changes are
1054 * actually applied, returning 0 to grant the changes, -ve to deny them.
1056 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1062 /* juggle the capabilities to follow [RES]UID changes unless
1063 * otherwise suppressed */
1064 if (!issecure(SECURE_NO_SETUID_FIXUP))
1065 cap_emulate_setxuid(new, old);
1069 /* juggle the capabilties to follow FSUID changes, unless
1070 * otherwise suppressed
1072 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1073 * if not, we might be a bit too harsh here.
1075 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1076 kuid_t root_uid = make_kuid(old->user_ns, 0);
1077 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1078 new->cap_effective =
1079 cap_drop_fs_set(new->cap_effective);
1081 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1082 new->cap_effective =
1083 cap_raise_fs_set(new->cap_effective,
1084 new->cap_permitted);
1096 * Rationale: code calling task_setscheduler, task_setioprio, and
1097 * task_setnice, assumes that
1098 * . if capable(cap_sys_nice), then those actions should be allowed
1099 * . if not capable(cap_sys_nice), but acting on your own processes,
1100 * then those actions should be allowed
1101 * This is insufficient now since you can call code without suid, but
1102 * yet with increased caps.
1103 * So we check for increased caps on the target process.
1105 static int cap_safe_nice(struct task_struct *p)
1107 int is_subset, ret = 0;
1110 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1111 current_cred()->cap_permitted);
1112 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1120 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1121 * @p: The task to affect
1123 * Detemine if the requested scheduler policy change is permitted for the
1124 * specified task, returning 0 if permission is granted, -ve if denied.
1126 int cap_task_setscheduler(struct task_struct *p)
1128 return cap_safe_nice(p);
1132 * cap_task_ioprio - Detemine if I/O priority change is permitted
1133 * @p: The task to affect
1134 * @ioprio: The I/O priority to set
1136 * Detemine if the requested I/O priority change is permitted for the specified
1137 * task, returning 0 if permission is granted, -ve if denied.
1139 int cap_task_setioprio(struct task_struct *p, int ioprio)
1141 return cap_safe_nice(p);
1145 * cap_task_ioprio - Detemine if task priority change is permitted
1146 * @p: The task to affect
1147 * @nice: The nice value to set
1149 * Detemine if the requested task priority change is permitted for the
1150 * specified task, returning 0 if permission is granted, -ve if denied.
1152 int cap_task_setnice(struct task_struct *p, int nice)
1154 return cap_safe_nice(p);
1158 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1159 * the current task's bounding set. Returns 0 on success, -ve on error.
1161 static int cap_prctl_drop(unsigned long cap)
1165 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1167 if (!cap_valid(cap))
1170 new = prepare_creds();
1173 cap_lower(new->cap_bset, cap);
1174 return commit_creds(new);
1178 * cap_task_prctl - Implement process control functions for this security module
1179 * @option: The process control function requested
1180 * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1182 * Allow process control functions (sys_prctl()) to alter capabilities; may
1183 * also deny access to other functions not otherwise implemented here.
1185 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1186 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1187 * modules will consider performing the function.
1189 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1190 unsigned long arg4, unsigned long arg5)
1192 const struct cred *old = current_cred();
1196 case PR_CAPBSET_READ:
1197 if (!cap_valid(arg2))
1199 return !!cap_raised(old->cap_bset, arg2);
1201 case PR_CAPBSET_DROP:
1202 return cap_prctl_drop(arg2);
1205 * The next four prctl's remain to assist with transitioning a
1206 * system from legacy UID=0 based privilege (when filesystem
1207 * capabilities are not in use) to a system using filesystem
1208 * capabilities only - as the POSIX.1e draft intended.
1212 * PR_SET_SECUREBITS =
1213 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1214 * | issecure_mask(SECURE_NOROOT)
1215 * | issecure_mask(SECURE_NOROOT_LOCKED)
1216 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1217 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1219 * will ensure that the current process and all of its
1220 * children will be locked into a pure
1221 * capability-based-privilege environment.
1223 case PR_SET_SECUREBITS:
1224 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1225 & (old->securebits ^ arg2)) /*[1]*/
1226 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1227 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1228 || (cap_capable(current_cred(),
1229 current_cred()->user_ns,
1231 CAP_OPT_NONE) != 0) /*[4]*/
1233 * [1] no changing of bits that are locked
1234 * [2] no unlocking of locks
1235 * [3] no setting of unsupported bits
1236 * [4] doing anything requires privilege (go read about
1237 * the "sendmail capabilities bug")
1240 /* cannot change a locked bit */
1243 new = prepare_creds();
1246 new->securebits = arg2;
1247 return commit_creds(new);
1249 case PR_GET_SECUREBITS:
1250 return old->securebits;
1252 case PR_GET_KEEPCAPS:
1253 return !!issecure(SECURE_KEEP_CAPS);
1255 case PR_SET_KEEPCAPS:
1256 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1258 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1261 new = prepare_creds();
1265 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1267 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1268 return commit_creds(new);
1270 case PR_CAP_AMBIENT:
1271 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1272 if (arg3 | arg4 | arg5)
1275 new = prepare_creds();
1278 cap_clear(new->cap_ambient);
1279 return commit_creds(new);
1282 if (((!cap_valid(arg3)) | arg4 | arg5))
1285 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1286 return !!cap_raised(current_cred()->cap_ambient, arg3);
1287 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1288 arg2 != PR_CAP_AMBIENT_LOWER) {
1291 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1292 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1293 !cap_raised(current_cred()->cap_inheritable,
1295 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1298 new = prepare_creds();
1301 if (arg2 == PR_CAP_AMBIENT_RAISE)
1302 cap_raise(new->cap_ambient, arg3);
1304 cap_lower(new->cap_ambient, arg3);
1305 return commit_creds(new);
1309 /* No functionality available - continue with default */
1315 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1316 * @mm: The VM space in which the new mapping is to be made
1317 * @pages: The size of the mapping
1319 * Determine whether the allocation of a new virtual mapping by the current
1320 * task is permitted, returning 1 if permission is granted, 0 if not.
1322 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1324 int cap_sys_admin = 0;
1326 if (cap_capable(current_cred(), &init_user_ns,
1327 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1330 return cap_sys_admin;
1334 * cap_mmap_addr - check if able to map given addr
1335 * @addr: address attempting to be mapped
1337 * If the process is attempting to map memory below dac_mmap_min_addr they need
1338 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1339 * capability security module. Returns 0 if this mapping should be allowed
1342 int cap_mmap_addr(unsigned long addr)
1346 if (addr < dac_mmap_min_addr) {
1347 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1349 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1351 current->flags |= PF_SUPERPRIV;
1356 int cap_mmap_file(struct file *file, unsigned long reqprot,
1357 unsigned long prot, unsigned long flags)
1362 #ifdef CONFIG_SECURITY
1364 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1365 LSM_HOOK_INIT(capable, cap_capable),
1366 LSM_HOOK_INIT(settime, cap_settime),
1367 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1368 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1369 LSM_HOOK_INIT(capget, cap_capget),
1370 LSM_HOOK_INIT(capset, cap_capset),
1371 LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1372 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1373 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1374 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1375 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1376 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1377 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1378 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1379 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1380 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1381 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1382 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1385 static int __init capability_init(void)
1387 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1392 DEFINE_LSM(capability) = {
1393 .name = "capability",
1394 .order = LSM_ORDER_FIRST,
1395 .init = capability_init,
1398 #endif /* CONFIG_SECURITY */