1 /* Common capabilities, needed by capability.o.
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License as published by
5 * the Free Software Foundation; either version 2 of the License, or
6 * (at your option) any later version.
10 #include <linux/capability.h>
11 #include <linux/audit.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/lsm_hooks.h>
16 #include <linux/file.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/skbuff.h>
22 #include <linux/netlink.h>
23 #include <linux/ptrace.h>
24 #include <linux/xattr.h>
25 #include <linux/hugetlb.h>
26 #include <linux/mount.h>
27 #include <linux/sched.h>
28 #include <linux/prctl.h>
29 #include <linux/securebits.h>
30 #include <linux/user_namespace.h>
31 #include <linux/binfmts.h>
32 #include <linux/personality.h>
35 * If a non-root user executes a setuid-root binary in
36 * !secure(SECURE_NOROOT) mode, then we raise capabilities.
37 * However if fE is also set, then the intent is for only
38 * the file capabilities to be applied, and the setuid-root
39 * bit is left on either to change the uid (plausible) or
40 * to get full privilege on a kernel without file capabilities
41 * support. So in that case we do not raise capabilities.
43 * Warn if that happens, once per boot.
45 static void warn_setuid_and_fcaps_mixed(const char *fname)
49 printk(KERN_INFO "warning: `%s' has both setuid-root and"
50 " effective capabilities. Therefore not raising all"
51 " capabilities.\n", fname);
57 * cap_capable - Determine whether a task has a particular effective capability
58 * @cred: The credentials to use
59 * @ns: The user namespace in which we need the capability
60 * @cap: The capability to check for
61 * @audit: Whether to write an audit message or not
63 * Determine whether the nominated task has the specified capability amongst
64 * its effective set, returning 0 if it does, -ve if it does not.
66 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
67 * and has_capability() functions. That is, it has the reverse semantics:
68 * cap_has_capability() returns 0 when a task has a capability, but the
69 * kernel's capable() and has_capability() returns 1 for this case.
71 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
74 struct user_namespace *ns = targ_ns;
76 /* See if cred has the capability in the target user namespace
77 * by examining the target user namespace and all of the target
78 * user namespace's parents.
81 /* Do we have the necessary capabilities? */
82 if (ns == cred->user_ns)
83 return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
86 * If we're already at a lower level than we're looking for,
87 * we're done searching.
89 if (ns->level <= cred->user_ns->level)
93 * The owner of the user namespace in the parent of the
94 * user namespace has all caps.
96 if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
100 * If you have a capability in a parent user ns, then you have
101 * it over all children user namespaces as well.
106 /* We never get here */
110 * cap_settime - Determine whether the current process may set the system clock
111 * @ts: The time to set
112 * @tz: The timezone to set
114 * Determine whether the current process may set the system clock and timezone
115 * information, returning 0 if permission granted, -ve if denied.
117 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
119 if (!capable(CAP_SYS_TIME))
125 * cap_ptrace_access_check - Determine whether the current process may access
127 * @child: The process to be accessed
128 * @mode: The mode of attachment.
130 * If we are in the same or an ancestor user_ns and have all the target
131 * task's capabilities, then ptrace access is allowed.
132 * If we have the ptrace capability to the target user_ns, then ptrace
136 * Determine whether a process may access another, returning 0 if permission
137 * granted, -ve if denied.
139 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
142 const struct cred *cred, *child_cred;
143 const kernel_cap_t *caller_caps;
146 cred = current_cred();
147 child_cred = __task_cred(child);
148 if (mode & PTRACE_MODE_FSCREDS)
149 caller_caps = &cred->cap_effective;
151 caller_caps = &cred->cap_permitted;
152 if (cred->user_ns == child_cred->user_ns &&
153 cap_issubset(child_cred->cap_permitted, *caller_caps))
155 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
164 * cap_ptrace_traceme - Determine whether another process may trace the current
165 * @parent: The task proposed to be the tracer
167 * If parent is in the same or an ancestor user_ns and has all current's
168 * capabilities, then ptrace access is allowed.
169 * If parent has the ptrace capability to current's user_ns, then ptrace
173 * Determine whether the nominated task is permitted to trace the current
174 * process, returning 0 if permission is granted, -ve if denied.
176 int cap_ptrace_traceme(struct task_struct *parent)
179 const struct cred *cred, *child_cred;
182 cred = __task_cred(parent);
183 child_cred = current_cred();
184 if (cred->user_ns == child_cred->user_ns &&
185 cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
187 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
196 * cap_capget - Retrieve a task's capability sets
197 * @target: The task from which to retrieve the capability sets
198 * @effective: The place to record the effective set
199 * @inheritable: The place to record the inheritable set
200 * @permitted: The place to record the permitted set
202 * This function retrieves the capabilities of the nominated task and returns
203 * them to the caller.
205 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
206 kernel_cap_t *inheritable, kernel_cap_t *permitted)
208 const struct cred *cred;
210 /* Derived from kernel/capability.c:sys_capget. */
212 cred = __task_cred(target);
213 *effective = cred->cap_effective;
214 *inheritable = cred->cap_inheritable;
215 *permitted = cred->cap_permitted;
221 * Determine whether the inheritable capabilities are limited to the old
222 * permitted set. Returns 1 if they are limited, 0 if they are not.
224 static inline int cap_inh_is_capped(void)
227 /* they are so limited unless the current task has the CAP_SETPCAP
230 if (cap_capable(current_cred(), current_cred()->user_ns,
231 CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
237 * cap_capset - Validate and apply proposed changes to current's capabilities
238 * @new: The proposed new credentials; alterations should be made here
239 * @old: The current task's current credentials
240 * @effective: A pointer to the proposed new effective capabilities set
241 * @inheritable: A pointer to the proposed new inheritable capabilities set
242 * @permitted: A pointer to the proposed new permitted capabilities set
244 * This function validates and applies a proposed mass change to the current
245 * process's capability sets. The changes are made to the proposed new
246 * credentials, and assuming no error, will be committed by the caller of LSM.
248 int cap_capset(struct cred *new,
249 const struct cred *old,
250 const kernel_cap_t *effective,
251 const kernel_cap_t *inheritable,
252 const kernel_cap_t *permitted)
254 if (cap_inh_is_capped() &&
255 !cap_issubset(*inheritable,
256 cap_combine(old->cap_inheritable,
257 old->cap_permitted)))
258 /* incapable of using this inheritable set */
261 if (!cap_issubset(*inheritable,
262 cap_combine(old->cap_inheritable,
264 /* no new pI capabilities outside bounding set */
267 /* verify restrictions on target's new Permitted set */
268 if (!cap_issubset(*permitted, old->cap_permitted))
271 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
272 if (!cap_issubset(*effective, *permitted))
275 new->cap_effective = *effective;
276 new->cap_inheritable = *inheritable;
277 new->cap_permitted = *permitted;
280 * Mask off ambient bits that are no longer both permitted and
283 new->cap_ambient = cap_intersect(new->cap_ambient,
284 cap_intersect(*permitted,
286 if (WARN_ON(!cap_ambient_invariant_ok(new)))
292 * cap_inode_need_killpriv - Determine if inode change affects privileges
293 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
295 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
296 * affects the security markings on that inode, and if it is, should
297 * inode_killpriv() be invoked or the change rejected.
299 * Returns 1 if security.capability has a value, meaning inode_killpriv()
300 * is required, 0 otherwise, meaning inode_killpriv() is not required.
302 int cap_inode_need_killpriv(struct dentry *dentry)
304 struct inode *inode = d_backing_inode(dentry);
307 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
312 * cap_inode_killpriv - Erase the security markings on an inode
313 * @dentry: The inode/dentry to alter
315 * Erase the privilege-enhancing security markings on an inode.
317 * Returns 0 if successful, -ve on error.
319 int cap_inode_killpriv(struct dentry *dentry)
323 error = __vfs_removexattr(dentry, XATTR_NAME_CAPS);
324 if (error == -EOPNOTSUPP)
329 static bool rootid_owns_currentns(kuid_t kroot)
331 struct user_namespace *ns;
333 if (!uid_valid(kroot))
336 for (ns = current_user_ns(); ; ns = ns->parent) {
337 if (from_kuid(ns, kroot) == 0)
339 if (ns == &init_user_ns)
346 static __u32 sansflags(__u32 m)
348 return m & ~VFS_CAP_FLAGS_EFFECTIVE;
351 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
353 if (size != XATTR_CAPS_SZ_2)
355 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
358 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
360 if (size != XATTR_CAPS_SZ_3)
362 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
366 * getsecurity: We are called for security.* before any attempt to read the
367 * xattr from the inode itself.
369 * This gives us a chance to read the on-disk value and convert it. If we
370 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
372 * Note we are not called by vfs_getxattr_alloc(), but that is only called
373 * by the integrity subsystem, which really wants the unconverted values -
376 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
382 uid_t root, mappedroot;
384 struct vfs_cap_data *cap;
385 struct vfs_ns_cap_data *nscap = NULL;
386 struct dentry *dentry;
387 struct user_namespace *fs_ns;
389 if (strcmp(name, "capability") != 0)
392 dentry = d_find_any_alias(inode);
396 size = sizeof(struct vfs_ns_cap_data);
397 ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS,
398 &tmpbuf, size, GFP_NOFS);
401 if (ret < 0 || !tmpbuf) {
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 the root kuid maps to a valid uid in current ns, then return
421 * this as a nscap. */
422 mappedroot = from_kuid(current_user_ns(), kroot);
423 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
424 size = sizeof(struct vfs_ns_cap_data);
427 /* v2 -> v3 conversion */
428 nscap = kzalloc(size, GFP_ATOMIC);
433 nsmagic = VFS_CAP_REVISION_3;
434 magic = le32_to_cpu(cap->magic_etc);
435 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
436 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
437 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
438 nscap->magic_etc = cpu_to_le32(nsmagic);
440 /* use allocated v3 buffer */
443 nscap->rootid = cpu_to_le32(mappedroot);
449 if (!rootid_owns_currentns(kroot)) {
454 /* This comes from a parent namespace. Return as a v2 capability */
455 size = sizeof(struct vfs_cap_data);
458 /* v3 -> v2 conversion */
459 cap = kzalloc(size, GFP_ATOMIC);
464 magic = VFS_CAP_REVISION_2;
465 nsmagic = le32_to_cpu(nscap->magic_etc);
466 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
467 magic |= VFS_CAP_FLAGS_EFFECTIVE;
468 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
469 cap->magic_etc = cpu_to_le32(magic);
471 /* use unconverted v2 */
481 static kuid_t rootid_from_xattr(const void *value, size_t size,
482 struct user_namespace *task_ns)
484 const struct vfs_ns_cap_data *nscap = value;
487 if (size == XATTR_CAPS_SZ_3)
488 rootid = le32_to_cpu(nscap->rootid);
490 return make_kuid(task_ns, rootid);
493 static bool validheader(size_t size, const struct vfs_cap_data *cap)
495 return is_v2header(size, cap) || is_v3header(size, cap);
499 * User requested a write of security.capability. If needed, update the
500 * xattr to change from v2 to v3, or to fixup the v3 rootid.
502 * If all is ok, we return the new size, on error return < 0.
504 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size)
506 struct vfs_ns_cap_data *nscap;
508 const struct vfs_cap_data *cap = *ivalue;
509 __u32 magic, nsmagic;
510 struct inode *inode = d_backing_inode(dentry);
511 struct user_namespace *task_ns = current_user_ns(),
512 *fs_ns = inode->i_sb->s_user_ns;
518 if (!validheader(size, cap))
520 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
522 if (size == XATTR_CAPS_SZ_2)
523 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
524 /* user is privileged, just write the v2 */
527 rootid = rootid_from_xattr(*ivalue, size, task_ns);
528 if (!uid_valid(rootid))
531 nsrootid = from_kuid(fs_ns, rootid);
535 newsize = sizeof(struct vfs_ns_cap_data);
536 nscap = kmalloc(newsize, GFP_ATOMIC);
539 nscap->rootid = cpu_to_le32(nsrootid);
540 nsmagic = VFS_CAP_REVISION_3;
541 magic = le32_to_cpu(cap->magic_etc);
542 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
543 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
544 nscap->magic_etc = cpu_to_le32(nsmagic);
545 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
553 * Calculate the new process capability sets from the capability sets attached
556 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
557 struct linux_binprm *bprm,
561 struct cred *new = bprm->cred;
565 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
568 if (caps->magic_etc & VFS_CAP_REVISION_MASK)
571 CAP_FOR_EACH_U32(i) {
572 __u32 permitted = caps->permitted.cap[i];
573 __u32 inheritable = caps->inheritable.cap[i];
576 * pP' = (X & fP) | (pI & fI)
577 * The addition of pA' is handled later.
579 new->cap_permitted.cap[i] =
580 (new->cap_bset.cap[i] & permitted) |
581 (new->cap_inheritable.cap[i] & inheritable);
583 if (permitted & ~new->cap_permitted.cap[i])
584 /* insufficient to execute correctly */
589 * For legacy apps, with no internal support for recognizing they
590 * do not have enough capabilities, we return an error if they are
591 * missing some "forced" (aka file-permitted) capabilities.
593 return *effective ? ret : 0;
597 * Extract the on-exec-apply capability sets for an executable file.
599 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
601 struct inode *inode = d_backing_inode(dentry);
605 struct vfs_ns_cap_data data, *nscaps = &data;
606 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
608 struct user_namespace *fs_ns;
610 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
615 fs_ns = inode->i_sb->s_user_ns;
616 size = __vfs_getxattr((struct dentry *)dentry, inode,
617 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
618 if (size == -ENODATA || size == -EOPNOTSUPP)
619 /* no data, that's ok */
625 if (size < sizeof(magic_etc))
628 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
630 rootkuid = make_kuid(fs_ns, 0);
631 switch (magic_etc & VFS_CAP_REVISION_MASK) {
632 case VFS_CAP_REVISION_1:
633 if (size != XATTR_CAPS_SZ_1)
635 tocopy = VFS_CAP_U32_1;
637 case VFS_CAP_REVISION_2:
638 if (size != XATTR_CAPS_SZ_2)
640 tocopy = VFS_CAP_U32_2;
642 case VFS_CAP_REVISION_3:
643 if (size != XATTR_CAPS_SZ_3)
645 tocopy = VFS_CAP_U32_3;
646 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
652 /* Limit the caps to the mounter of the filesystem
653 * or the more limited uid specified in the xattr.
655 if (!rootid_owns_currentns(rootkuid))
658 CAP_FOR_EACH_U32(i) {
661 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
662 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
665 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
666 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
672 * Attempt to get the on-exec apply capability sets for an executable file from
673 * its xattrs and, if present, apply them to the proposed credentials being
674 * constructed by execve().
676 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap)
679 struct cpu_vfs_cap_data vcaps;
681 cap_clear(bprm->cred->cap_permitted);
683 if (!file_caps_enabled)
686 if (!mnt_may_suid(bprm->file->f_path.mnt))
690 * This check is redundant with mnt_may_suid() but is kept to make
691 * explicit that capability bits are limited to s_user_ns and its
694 if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
697 rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
700 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
702 else if (rc == -ENODATA)
707 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap);
709 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
710 __func__, rc, bprm->filename);
714 cap_clear(bprm->cred->cap_permitted);
720 * cap_bprm_set_creds - Set up the proposed credentials for execve().
721 * @bprm: The execution parameters, including the proposed creds
723 * Set up the proposed credentials for a new execution context being
724 * constructed by execve(). The proposed creds in @bprm->cred is altered,
725 * which won't take effect immediately. Returns 0 if successful, -ve on error.
727 int cap_bprm_set_creds(struct linux_binprm *bprm)
729 const struct cred *old = current_cred();
730 struct cred *new = bprm->cred;
731 bool effective, has_cap = false, is_setid;
735 new->cap_ambient = old->cap_ambient;
736 if (WARN_ON(!cap_ambient_invariant_ok(old)))
740 ret = get_file_caps(bprm, &effective, &has_cap);
744 root_uid = make_kuid(new->user_ns, 0);
746 if (!issecure(SECURE_NOROOT)) {
748 * If the legacy file capability is set, then don't set privs
749 * for a setuid root binary run by a non-root user. Do set it
750 * for a root user just to cause least surprise to an admin.
752 if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) {
753 warn_setuid_and_fcaps_mixed(bprm->filename);
757 * To support inheritance of root-permissions and suid-root
758 * executables under compatibility mode, we override the
759 * capability sets for the file.
761 * If only the real uid is 0, we do not set the effective bit.
763 if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) {
764 /* pP' = (cap_bset & ~0) | (pI & ~0) */
765 new->cap_permitted = cap_combine(old->cap_bset,
766 old->cap_inheritable);
768 if (uid_eq(new->euid, root_uid))
773 /* if we have fs caps, clear dangerous personality flags */
774 if (!cap_issubset(new->cap_permitted, old->cap_permitted))
775 bprm->per_clear |= PER_CLEAR_ON_SETID;
778 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
779 * credentials unless they have the appropriate permit.
781 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
783 is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid);
786 !cap_issubset(new->cap_permitted, old->cap_permitted)) &&
787 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
788 !ptracer_capable(current, new->user_ns))) {
789 /* downgrade; they get no more than they had, and maybe less */
790 if (!ns_capable(new->user_ns, CAP_SETUID) ||
791 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
792 new->euid = new->uid;
793 new->egid = new->gid;
795 new->cap_permitted = cap_intersect(new->cap_permitted,
799 new->suid = new->fsuid = new->euid;
800 new->sgid = new->fsgid = new->egid;
802 /* File caps or setid cancels ambient. */
803 if (has_cap || is_setid)
804 cap_clear(new->cap_ambient);
807 * Now that we've computed pA', update pP' to give:
808 * pP' = (X & fP) | (pI & fI) | pA'
810 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
813 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
814 * this is the same as pE' = (fE ? pP' : 0) | pA'.
817 new->cap_effective = new->cap_permitted;
819 new->cap_effective = new->cap_ambient;
821 if (WARN_ON(!cap_ambient_invariant_ok(new)))
825 * Audit candidate if current->cap_effective is set
827 * We do not bother to audit if 3 things are true:
828 * 1) cap_effective has all caps
830 * 3) root is supposed to have all caps (SECURE_NOROOT)
831 * Since this is just a normal root execing a process.
833 * Number 1 above might fail if you don't have a full bset, but I think
834 * that is interesting information to audit.
836 if (!cap_issubset(new->cap_effective, new->cap_ambient)) {
837 if (!cap_issubset(CAP_FULL_SET, new->cap_effective) ||
838 !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) ||
839 issecure(SECURE_NOROOT)) {
840 ret = audit_log_bprm_fcaps(bprm, new, old);
846 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
848 if (WARN_ON(!cap_ambient_invariant_ok(new)))
851 /* Check for privilege-elevated exec. */
852 bprm->cap_elevated = 0;
854 bprm->cap_elevated = 1;
855 } else if (!uid_eq(new->uid, root_uid)) {
857 !cap_issubset(new->cap_permitted, new->cap_ambient))
858 bprm->cap_elevated = 1;
865 * cap_inode_setxattr - Determine whether an xattr may be altered
866 * @dentry: The inode/dentry being altered
867 * @name: The name of the xattr to be changed
868 * @value: The value that the xattr will be changed to
869 * @size: The size of value
870 * @flags: The replacement flag
872 * Determine whether an xattr may be altered or set on an inode, returning 0 if
873 * permission is granted, -ve if denied.
875 * This is used to make sure security xattrs don't get updated or set by those
876 * who aren't privileged to do so.
878 int cap_inode_setxattr(struct dentry *dentry, const char *name,
879 const void *value, size_t size, int flags)
881 /* Ignore non-security xattrs */
882 if (strncmp(name, XATTR_SECURITY_PREFIX,
883 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
887 * For XATTR_NAME_CAPS the check will be done in
888 * cap_convert_nscap(), called by setxattr()
890 if (strcmp(name, XATTR_NAME_CAPS) == 0)
893 if (!capable(CAP_SYS_ADMIN))
899 * cap_inode_removexattr - Determine whether an xattr may be removed
900 * @dentry: The inode/dentry being altered
901 * @name: The name of the xattr to be changed
903 * Determine whether an xattr may be removed from an inode, returning 0 if
904 * permission is granted, -ve if denied.
906 * This is used to make sure security xattrs don't get removed by those who
907 * aren't privileged to remove them.
909 int cap_inode_removexattr(struct dentry *dentry, const char *name)
911 /* Ignore non-security xattrs */
912 if (strncmp(name, XATTR_SECURITY_PREFIX,
913 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
916 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
917 /* security.capability gets namespaced */
918 struct inode *inode = d_backing_inode(dentry);
921 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
926 if (!capable(CAP_SYS_ADMIN))
932 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
933 * a process after a call to setuid, setreuid, or setresuid.
935 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
936 * {r,e,s}uid != 0, the permitted and effective capabilities are
939 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
940 * capabilities of the process are cleared.
942 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
943 * capabilities are set to the permitted capabilities.
945 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
950 * cevans - New behaviour, Oct '99
951 * A process may, via prctl(), elect to keep its capabilities when it
952 * calls setuid() and switches away from uid==0. Both permitted and
953 * effective sets will be retained.
954 * Without this change, it was impossible for a daemon to drop only some
955 * of its privilege. The call to setuid(!=0) would drop all privileges!
956 * Keeping uid 0 is not an option because uid 0 owns too many vital
958 * Thanks to Olaf Kirch and Peter Benie for spotting this.
960 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
962 kuid_t root_uid = make_kuid(old->user_ns, 0);
964 if ((uid_eq(old->uid, root_uid) ||
965 uid_eq(old->euid, root_uid) ||
966 uid_eq(old->suid, root_uid)) &&
967 (!uid_eq(new->uid, root_uid) &&
968 !uid_eq(new->euid, root_uid) &&
969 !uid_eq(new->suid, root_uid))) {
970 if (!issecure(SECURE_KEEP_CAPS)) {
971 cap_clear(new->cap_permitted);
972 cap_clear(new->cap_effective);
976 * Pre-ambient programs expect setresuid to nonroot followed
977 * by exec to drop capabilities. We should make sure that
978 * this remains the case.
980 cap_clear(new->cap_ambient);
982 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
983 cap_clear(new->cap_effective);
984 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
985 new->cap_effective = new->cap_permitted;
989 * cap_task_fix_setuid - Fix up the results of setuid() call
990 * @new: The proposed credentials
991 * @old: The current task's current credentials
992 * @flags: Indications of what has changed
994 * Fix up the results of setuid() call before the credential changes are
995 * actually applied, returning 0 to grant the changes, -ve to deny them.
997 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1003 /* juggle the capabilities to follow [RES]UID changes unless
1004 * otherwise suppressed */
1005 if (!issecure(SECURE_NO_SETUID_FIXUP))
1006 cap_emulate_setxuid(new, old);
1010 /* juggle the capabilties to follow FSUID changes, unless
1011 * otherwise suppressed
1013 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1014 * if not, we might be a bit too harsh here.
1016 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1017 kuid_t root_uid = make_kuid(old->user_ns, 0);
1018 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1019 new->cap_effective =
1020 cap_drop_fs_set(new->cap_effective);
1022 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1023 new->cap_effective =
1024 cap_raise_fs_set(new->cap_effective,
1025 new->cap_permitted);
1037 * Rationale: code calling task_setscheduler, task_setioprio, and
1038 * task_setnice, assumes that
1039 * . if capable(cap_sys_nice), then those actions should be allowed
1040 * . if not capable(cap_sys_nice), but acting on your own processes,
1041 * then those actions should be allowed
1042 * This is insufficient now since you can call code without suid, but
1043 * yet with increased caps.
1044 * So we check for increased caps on the target process.
1046 static int cap_safe_nice(struct task_struct *p)
1048 int is_subset, ret = 0;
1051 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1052 current_cred()->cap_permitted);
1053 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1061 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1062 * @p: The task to affect
1064 * Detemine if the requested scheduler policy change is permitted for the
1065 * specified task, returning 0 if permission is granted, -ve if denied.
1067 int cap_task_setscheduler(struct task_struct *p)
1069 return cap_safe_nice(p);
1073 * cap_task_ioprio - Detemine if I/O priority change is permitted
1074 * @p: The task to affect
1075 * @ioprio: The I/O priority to set
1077 * Detemine if the requested I/O priority change is permitted for the specified
1078 * task, returning 0 if permission is granted, -ve if denied.
1080 int cap_task_setioprio(struct task_struct *p, int ioprio)
1082 return cap_safe_nice(p);
1086 * cap_task_ioprio - Detemine if task priority change is permitted
1087 * @p: The task to affect
1088 * @nice: The nice value to set
1090 * Detemine if the requested task priority change is permitted for the
1091 * specified task, returning 0 if permission is granted, -ve if denied.
1093 int cap_task_setnice(struct task_struct *p, int nice)
1095 return cap_safe_nice(p);
1099 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1100 * the current task's bounding set. Returns 0 on success, -ve on error.
1102 static int cap_prctl_drop(unsigned long cap)
1106 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1108 if (!cap_valid(cap))
1111 new = prepare_creds();
1114 cap_lower(new->cap_bset, cap);
1115 return commit_creds(new);
1119 * cap_task_prctl - Implement process control functions for this security module
1120 * @option: The process control function requested
1121 * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1123 * Allow process control functions (sys_prctl()) to alter capabilities; may
1124 * also deny access to other functions not otherwise implemented here.
1126 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1127 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1128 * modules will consider performing the function.
1130 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1131 unsigned long arg4, unsigned long arg5)
1133 const struct cred *old = current_cred();
1137 case PR_CAPBSET_READ:
1138 if (!cap_valid(arg2))
1140 return !!cap_raised(old->cap_bset, arg2);
1142 case PR_CAPBSET_DROP:
1143 return cap_prctl_drop(arg2);
1146 * The next four prctl's remain to assist with transitioning a
1147 * system from legacy UID=0 based privilege (when filesystem
1148 * capabilities are not in use) to a system using filesystem
1149 * capabilities only - as the POSIX.1e draft intended.
1153 * PR_SET_SECUREBITS =
1154 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1155 * | issecure_mask(SECURE_NOROOT)
1156 * | issecure_mask(SECURE_NOROOT_LOCKED)
1157 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1158 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1160 * will ensure that the current process and all of its
1161 * children will be locked into a pure
1162 * capability-based-privilege environment.
1164 case PR_SET_SECUREBITS:
1165 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1166 & (old->securebits ^ arg2)) /*[1]*/
1167 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1168 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1169 || (cap_capable(current_cred(),
1170 current_cred()->user_ns, CAP_SETPCAP,
1171 SECURITY_CAP_AUDIT) != 0) /*[4]*/
1173 * [1] no changing of bits that are locked
1174 * [2] no unlocking of locks
1175 * [3] no setting of unsupported bits
1176 * [4] doing anything requires privilege (go read about
1177 * the "sendmail capabilities bug")
1180 /* cannot change a locked bit */
1183 new = prepare_creds();
1186 new->securebits = arg2;
1187 return commit_creds(new);
1189 case PR_GET_SECUREBITS:
1190 return old->securebits;
1192 case PR_GET_KEEPCAPS:
1193 return !!issecure(SECURE_KEEP_CAPS);
1195 case PR_SET_KEEPCAPS:
1196 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1198 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1201 new = prepare_creds();
1205 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1207 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1208 return commit_creds(new);
1210 case PR_CAP_AMBIENT:
1211 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1212 if (arg3 | arg4 | arg5)
1215 new = prepare_creds();
1218 cap_clear(new->cap_ambient);
1219 return commit_creds(new);
1222 if (((!cap_valid(arg3)) | arg4 | arg5))
1225 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1226 return !!cap_raised(current_cred()->cap_ambient, arg3);
1227 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1228 arg2 != PR_CAP_AMBIENT_LOWER) {
1231 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1232 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1233 !cap_raised(current_cred()->cap_inheritable,
1235 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1238 new = prepare_creds();
1241 if (arg2 == PR_CAP_AMBIENT_RAISE)
1242 cap_raise(new->cap_ambient, arg3);
1244 cap_lower(new->cap_ambient, arg3);
1245 return commit_creds(new);
1249 /* No functionality available - continue with default */
1255 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1256 * @mm: The VM space in which the new mapping is to be made
1257 * @pages: The size of the mapping
1259 * Determine whether the allocation of a new virtual mapping by the current
1260 * task is permitted, returning 1 if permission is granted, 0 if not.
1262 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1264 int cap_sys_admin = 0;
1266 if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1267 SECURITY_CAP_NOAUDIT) == 0)
1269 return cap_sys_admin;
1273 * cap_mmap_addr - check if able to map given addr
1274 * @addr: address attempting to be mapped
1276 * If the process is attempting to map memory below dac_mmap_min_addr they need
1277 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1278 * capability security module. Returns 0 if this mapping should be allowed
1281 int cap_mmap_addr(unsigned long addr)
1285 if (addr < dac_mmap_min_addr) {
1286 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1287 SECURITY_CAP_AUDIT);
1288 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1290 current->flags |= PF_SUPERPRIV;
1295 int cap_mmap_file(struct file *file, unsigned long reqprot,
1296 unsigned long prot, unsigned long flags)
1301 #ifdef CONFIG_SECURITY
1303 struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1304 LSM_HOOK_INIT(capable, cap_capable),
1305 LSM_HOOK_INIT(settime, cap_settime),
1306 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1307 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1308 LSM_HOOK_INIT(capget, cap_capget),
1309 LSM_HOOK_INIT(capset, cap_capset),
1310 LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1311 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1312 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1313 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1314 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1315 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1316 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1317 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1318 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1319 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1320 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1321 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1324 void __init capability_add_hooks(void)
1326 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1330 #endif /* CONFIG_SECURITY */