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,
72 int cap, unsigned int opts)
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)
226 /* they are so limited unless the current task has the CAP_SETPCAP
229 if (cap_capable(current_cred(), current_cred()->user_ns,
230 CAP_SETPCAP, CAP_OPT_NONE) == 0)
236 * cap_capset - Validate and apply proposed changes to current's capabilities
237 * @new: The proposed new credentials; alterations should be made here
238 * @old: The current task's current credentials
239 * @effective: A pointer to the proposed new effective capabilities set
240 * @inheritable: A pointer to the proposed new inheritable capabilities set
241 * @permitted: A pointer to the proposed new permitted capabilities set
243 * This function validates and applies a proposed mass change to the current
244 * process's capability sets. The changes are made to the proposed new
245 * credentials, and assuming no error, will be committed by the caller of LSM.
247 int cap_capset(struct cred *new,
248 const struct cred *old,
249 const kernel_cap_t *effective,
250 const kernel_cap_t *inheritable,
251 const kernel_cap_t *permitted)
253 if (cap_inh_is_capped() &&
254 !cap_issubset(*inheritable,
255 cap_combine(old->cap_inheritable,
256 old->cap_permitted)))
257 /* incapable of using this inheritable set */
260 if (!cap_issubset(*inheritable,
261 cap_combine(old->cap_inheritable,
263 /* no new pI capabilities outside bounding set */
266 /* verify restrictions on target's new Permitted set */
267 if (!cap_issubset(*permitted, old->cap_permitted))
270 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
271 if (!cap_issubset(*effective, *permitted))
274 new->cap_effective = *effective;
275 new->cap_inheritable = *inheritable;
276 new->cap_permitted = *permitted;
279 * Mask off ambient bits that are no longer both permitted and
282 new->cap_ambient = cap_intersect(new->cap_ambient,
283 cap_intersect(*permitted,
285 if (WARN_ON(!cap_ambient_invariant_ok(new)))
291 * cap_inode_need_killpriv - Determine if inode change affects privileges
292 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
294 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
295 * affects the security markings on that inode, and if it is, should
296 * inode_killpriv() be invoked or the change rejected.
298 * Returns 1 if security.capability has a value, meaning inode_killpriv()
299 * is required, 0 otherwise, meaning inode_killpriv() is not required.
301 int cap_inode_need_killpriv(struct dentry *dentry)
303 struct inode *inode = d_backing_inode(dentry);
306 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
311 * cap_inode_killpriv - Erase the security markings on an inode
312 * @dentry: The inode/dentry to alter
314 * Erase the privilege-enhancing security markings on an inode.
316 * Returns 0 if successful, -ve on error.
318 int cap_inode_killpriv(struct dentry *dentry)
322 error = __vfs_removexattr(dentry, XATTR_NAME_CAPS);
323 if (error == -EOPNOTSUPP)
328 static bool rootid_owns_currentns(kuid_t kroot)
330 struct user_namespace *ns;
332 if (!uid_valid(kroot))
335 for (ns = current_user_ns(); ; ns = ns->parent) {
336 if (from_kuid(ns, kroot) == 0)
338 if (ns == &init_user_ns)
345 static __u32 sansflags(__u32 m)
347 return m & ~VFS_CAP_FLAGS_EFFECTIVE;
350 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
352 if (size != XATTR_CAPS_SZ_2)
354 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
357 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
359 if (size != XATTR_CAPS_SZ_3)
361 return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
365 * getsecurity: We are called for security.* before any attempt to read the
366 * xattr from the inode itself.
368 * This gives us a chance to read the on-disk value and convert it. If we
369 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
371 * Note we are not called by vfs_getxattr_alloc(), but that is only called
372 * by the integrity subsystem, which really wants the unconverted values -
375 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
381 uid_t root, mappedroot;
383 struct vfs_cap_data *cap;
384 struct vfs_ns_cap_data *nscap = NULL;
385 struct dentry *dentry;
386 struct user_namespace *fs_ns;
388 if (strcmp(name, "capability") != 0)
391 dentry = d_find_any_alias(inode);
395 size = sizeof(struct vfs_ns_cap_data);
396 ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS,
397 &tmpbuf, size, GFP_NOFS);
400 if (ret < 0 || !tmpbuf)
403 fs_ns = inode->i_sb->s_user_ns;
404 cap = (struct vfs_cap_data *) tmpbuf;
405 if (is_v2header((size_t) ret, cap)) {
407 } else if (is_v3header((size_t) ret, cap)) {
408 nscap = (struct vfs_ns_cap_data *) tmpbuf;
409 root = le32_to_cpu(nscap->rootid);
415 kroot = make_kuid(fs_ns, root);
417 /* If the root kuid maps to a valid uid in current ns, then return
418 * this as a nscap. */
419 mappedroot = from_kuid(current_user_ns(), kroot);
420 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
421 size = sizeof(struct vfs_ns_cap_data);
424 /* v2 -> v3 conversion */
425 nscap = kzalloc(size, GFP_ATOMIC);
430 nsmagic = VFS_CAP_REVISION_3;
431 magic = le32_to_cpu(cap->magic_etc);
432 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
433 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
434 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
435 nscap->magic_etc = cpu_to_le32(nsmagic);
437 /* use allocated v3 buffer */
440 nscap->rootid = cpu_to_le32(mappedroot);
446 if (!rootid_owns_currentns(kroot)) {
451 /* This comes from a parent namespace. Return as a v2 capability */
452 size = sizeof(struct vfs_cap_data);
455 /* v3 -> v2 conversion */
456 cap = kzalloc(size, GFP_ATOMIC);
461 magic = VFS_CAP_REVISION_2;
462 nsmagic = le32_to_cpu(nscap->magic_etc);
463 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
464 magic |= VFS_CAP_FLAGS_EFFECTIVE;
465 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
466 cap->magic_etc = cpu_to_le32(magic);
468 /* use unconverted v2 */
478 static kuid_t rootid_from_xattr(const void *value, size_t size,
479 struct user_namespace *task_ns)
481 const struct vfs_ns_cap_data *nscap = value;
484 if (size == XATTR_CAPS_SZ_3)
485 rootid = le32_to_cpu(nscap->rootid);
487 return make_kuid(task_ns, rootid);
490 static bool validheader(size_t size, const struct vfs_cap_data *cap)
492 return is_v2header(size, cap) || is_v3header(size, cap);
496 * User requested a write of security.capability. If needed, update the
497 * xattr to change from v2 to v3, or to fixup the v3 rootid.
499 * If all is ok, we return the new size, on error return < 0.
501 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size)
503 struct vfs_ns_cap_data *nscap;
505 const struct vfs_cap_data *cap = *ivalue;
506 __u32 magic, nsmagic;
507 struct inode *inode = d_backing_inode(dentry);
508 struct user_namespace *task_ns = current_user_ns(),
509 *fs_ns = inode->i_sb->s_user_ns;
515 if (!validheader(size, cap))
517 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
519 if (size == XATTR_CAPS_SZ_2)
520 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
521 /* user is privileged, just write the v2 */
524 rootid = rootid_from_xattr(*ivalue, size, task_ns);
525 if (!uid_valid(rootid))
528 nsrootid = from_kuid(fs_ns, rootid);
532 newsize = sizeof(struct vfs_ns_cap_data);
533 nscap = kmalloc(newsize, GFP_ATOMIC);
536 nscap->rootid = cpu_to_le32(nsrootid);
537 nsmagic = VFS_CAP_REVISION_3;
538 magic = le32_to_cpu(cap->magic_etc);
539 if (magic & VFS_CAP_FLAGS_EFFECTIVE)
540 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
541 nscap->magic_etc = cpu_to_le32(nsmagic);
542 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
550 * Calculate the new process capability sets from the capability sets attached
553 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
554 struct linux_binprm *bprm,
558 struct cred *new = bprm->cred;
562 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
565 if (caps->magic_etc & VFS_CAP_REVISION_MASK)
568 CAP_FOR_EACH_U32(i) {
569 __u32 permitted = caps->permitted.cap[i];
570 __u32 inheritable = caps->inheritable.cap[i];
573 * pP' = (X & fP) | (pI & fI)
574 * The addition of pA' is handled later.
576 new->cap_permitted.cap[i] =
577 (new->cap_bset.cap[i] & permitted) |
578 (new->cap_inheritable.cap[i] & inheritable);
580 if (permitted & ~new->cap_permitted.cap[i])
581 /* insufficient to execute correctly */
586 * For legacy apps, with no internal support for recognizing they
587 * do not have enough capabilities, we return an error if they are
588 * missing some "forced" (aka file-permitted) capabilities.
590 return *effective ? ret : 0;
594 * Extract the on-exec-apply capability sets for an executable file.
596 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
598 struct inode *inode = d_backing_inode(dentry);
602 struct vfs_ns_cap_data data, *nscaps = &data;
603 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
605 struct user_namespace *fs_ns;
607 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
612 fs_ns = inode->i_sb->s_user_ns;
613 size = __vfs_getxattr((struct dentry *)dentry, inode,
614 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
615 if (size == -ENODATA || size == -EOPNOTSUPP)
616 /* no data, that's ok */
622 if (size < sizeof(magic_etc))
625 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
627 rootkuid = make_kuid(fs_ns, 0);
628 switch (magic_etc & VFS_CAP_REVISION_MASK) {
629 case VFS_CAP_REVISION_1:
630 if (size != XATTR_CAPS_SZ_1)
632 tocopy = VFS_CAP_U32_1;
634 case VFS_CAP_REVISION_2:
635 if (size != XATTR_CAPS_SZ_2)
637 tocopy = VFS_CAP_U32_2;
639 case VFS_CAP_REVISION_3:
640 if (size != XATTR_CAPS_SZ_3)
642 tocopy = VFS_CAP_U32_3;
643 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
649 /* Limit the caps to the mounter of the filesystem
650 * or the more limited uid specified in the xattr.
652 if (!rootid_owns_currentns(rootkuid))
655 CAP_FOR_EACH_U32(i) {
658 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
659 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
662 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
663 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
669 * Attempt to get the on-exec apply capability sets for an executable file from
670 * its xattrs and, if present, apply them to the proposed credentials being
671 * constructed by execve().
673 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_fcap)
676 struct cpu_vfs_cap_data vcaps;
678 cap_clear(bprm->cred->cap_permitted);
680 if (!file_caps_enabled)
683 if (!mnt_may_suid(bprm->file->f_path.mnt))
687 * This check is redundant with mnt_may_suid() but is kept to make
688 * explicit that capability bits are limited to s_user_ns and its
691 if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
694 rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
697 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
699 else if (rc == -ENODATA)
704 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
706 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
707 __func__, rc, bprm->filename);
711 cap_clear(bprm->cred->cap_permitted);
716 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
718 static inline bool __is_real(kuid_t uid, struct cred *cred)
719 { return uid_eq(cred->uid, uid); }
721 static inline bool __is_eff(kuid_t uid, struct cred *cred)
722 { return uid_eq(cred->euid, uid); }
724 static inline bool __is_suid(kuid_t uid, struct cred *cred)
725 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
728 * handle_privileged_root - Handle case of privileged root
729 * @bprm: The execution parameters, including the proposed creds
730 * @has_fcap: Are any file capabilities set?
731 * @effective: Do we have effective root privilege?
732 * @root_uid: This namespace' root UID WRT initial USER namespace
734 * Handle the case where root is privileged and hasn't been neutered by
735 * SECURE_NOROOT. If file capabilities are set, they won't be combined with
736 * set UID root and nothing is changed. If we are root, cap_permitted is
737 * updated. If we have become set UID root, the effective bit is set.
739 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
740 bool *effective, kuid_t root_uid)
742 const struct cred *old = current_cred();
743 struct cred *new = bprm->cred;
745 if (!root_privileged())
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_fcap && __is_suid(root_uid, new)) {
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 (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
762 /* pP' = (cap_bset & ~0) | (pI & ~0) */
763 new->cap_permitted = cap_combine(old->cap_bset,
764 old->cap_inheritable);
767 * If only the real uid is 0, we do not set the effective bit.
769 if (__is_eff(root_uid, new))
773 #define __cap_gained(field, target, source) \
774 !cap_issubset(target->cap_##field, source->cap_##field)
775 #define __cap_grew(target, source, cred) \
776 !cap_issubset(cred->cap_##target, cred->cap_##source)
777 #define __cap_full(field, cred) \
778 cap_issubset(CAP_FULL_SET, cred->cap_##field)
780 static inline bool __is_setuid(struct cred *new, const struct cred *old)
781 { return !uid_eq(new->euid, old->uid); }
783 static inline bool __is_setgid(struct cred *new, const struct cred *old)
784 { return !gid_eq(new->egid, old->gid); }
787 * 1) Audit candidate if current->cap_effective is set
789 * We do not bother to audit if 3 things are true:
790 * 1) cap_effective has all caps
791 * 2) we became root *OR* are were already root
792 * 3) root is supposed to have all caps (SECURE_NOROOT)
793 * Since this is just a normal root execing a process.
795 * Number 1 above might fail if you don't have a full bset, but I think
796 * that is interesting information to audit.
798 * A number of other conditions require logging:
799 * 2) something prevented setuid root getting all caps
800 * 3) non-setuid root gets fcaps
801 * 4) non-setuid root gets ambient
803 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
804 kuid_t root, bool has_fcap)
808 if ((__cap_grew(effective, ambient, new) &&
809 !(__cap_full(effective, new) &&
810 (__is_eff(root, new) || __is_real(root, new)) &&
811 root_privileged())) ||
812 (root_privileged() &&
813 __is_suid(root, new) &&
814 !__cap_full(effective, new)) ||
815 (!__is_setuid(new, old) &&
817 __cap_gained(permitted, new, old)) ||
818 __cap_gained(ambient, new, old))))
826 * cap_bprm_set_creds - Set up the proposed credentials for execve().
827 * @bprm: The execution parameters, including the proposed creds
829 * Set up the proposed credentials for a new execution context being
830 * constructed by execve(). The proposed creds in @bprm->cred is altered,
831 * which won't take effect immediately. Returns 0 if successful, -ve on error.
833 int cap_bprm_set_creds(struct linux_binprm *bprm)
835 const struct cred *old = current_cred();
836 struct cred *new = bprm->cred;
837 bool effective = false, has_fcap = false, is_setid;
841 new->cap_ambient = old->cap_ambient;
842 if (WARN_ON(!cap_ambient_invariant_ok(old)))
845 ret = get_file_caps(bprm, &effective, &has_fcap);
849 root_uid = make_kuid(new->user_ns, 0);
851 handle_privileged_root(bprm, has_fcap, &effective, root_uid);
853 /* if we have fs caps, clear dangerous personality flags */
854 if (__cap_gained(permitted, new, old))
855 bprm->per_clear |= PER_CLEAR_ON_SETID;
857 /* Don't let someone trace a set[ug]id/setpcap binary with the revised
858 * credentials unless they have the appropriate permit.
860 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
862 is_setid = __is_setuid(new, old) || __is_setgid(new, old);
864 if ((is_setid || __cap_gained(permitted, new, old)) &&
865 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
866 !ptracer_capable(current, new->user_ns))) {
867 /* downgrade; they get no more than they had, and maybe less */
868 if (!ns_capable(new->user_ns, CAP_SETUID) ||
869 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
870 new->euid = new->uid;
871 new->egid = new->gid;
873 new->cap_permitted = cap_intersect(new->cap_permitted,
877 new->suid = new->fsuid = new->euid;
878 new->sgid = new->fsgid = new->egid;
880 /* File caps or setid cancels ambient. */
881 if (has_fcap || is_setid)
882 cap_clear(new->cap_ambient);
885 * Now that we've computed pA', update pP' to give:
886 * pP' = (X & fP) | (pI & fI) | pA'
888 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
891 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set,
892 * this is the same as pE' = (fE ? pP' : 0) | pA'.
895 new->cap_effective = new->cap_permitted;
897 new->cap_effective = new->cap_ambient;
899 if (WARN_ON(!cap_ambient_invariant_ok(new)))
902 if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
903 ret = audit_log_bprm_fcaps(bprm, new, old);
908 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
910 if (WARN_ON(!cap_ambient_invariant_ok(new)))
913 /* Check for privilege-elevated exec. */
914 bprm->cap_elevated = 0;
916 (!__is_real(root_uid, new) &&
918 __cap_grew(permitted, ambient, new))))
919 bprm->cap_elevated = 1;
925 * cap_inode_setxattr - Determine whether an xattr may be altered
926 * @dentry: The inode/dentry being altered
927 * @name: The name of the xattr to be changed
928 * @value: The value that the xattr will be changed to
929 * @size: The size of value
930 * @flags: The replacement flag
932 * Determine whether an xattr may be altered or set on an inode, returning 0 if
933 * permission is granted, -ve if denied.
935 * This is used to make sure security xattrs don't get updated or set by those
936 * who aren't privileged to do so.
938 int cap_inode_setxattr(struct dentry *dentry, const char *name,
939 const void *value, size_t size, int flags)
941 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
943 /* Ignore non-security xattrs */
944 if (strncmp(name, XATTR_SECURITY_PREFIX,
945 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
949 * For XATTR_NAME_CAPS the check will be done in
950 * cap_convert_nscap(), called by setxattr()
952 if (strcmp(name, XATTR_NAME_CAPS) == 0)
955 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
961 * cap_inode_removexattr - Determine whether an xattr may be removed
962 * @dentry: The inode/dentry being altered
963 * @name: The name of the xattr to be changed
965 * Determine whether an xattr may be removed from an inode, returning 0 if
966 * permission is granted, -ve if denied.
968 * This is used to make sure security xattrs don't get removed by those who
969 * aren't privileged to remove them.
971 int cap_inode_removexattr(struct dentry *dentry, const char *name)
973 struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
975 /* Ignore non-security xattrs */
976 if (strncmp(name, XATTR_SECURITY_PREFIX,
977 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
980 if (strcmp(name, XATTR_NAME_CAPS) == 0) {
981 /* security.capability gets namespaced */
982 struct inode *inode = d_backing_inode(dentry);
985 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
990 if (!ns_capable(user_ns, CAP_SYS_ADMIN))
996 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
997 * a process after a call to setuid, setreuid, or setresuid.
999 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1000 * {r,e,s}uid != 0, the permitted and effective capabilities are
1003 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1004 * capabilities of the process are cleared.
1006 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1007 * capabilities are set to the permitted capabilities.
1009 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1014 * cevans - New behaviour, Oct '99
1015 * A process may, via prctl(), elect to keep its capabilities when it
1016 * calls setuid() and switches away from uid==0. Both permitted and
1017 * effective sets will be retained.
1018 * Without this change, it was impossible for a daemon to drop only some
1019 * of its privilege. The call to setuid(!=0) would drop all privileges!
1020 * Keeping uid 0 is not an option because uid 0 owns too many vital
1022 * Thanks to Olaf Kirch and Peter Benie for spotting this.
1024 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1026 kuid_t root_uid = make_kuid(old->user_ns, 0);
1028 if ((uid_eq(old->uid, root_uid) ||
1029 uid_eq(old->euid, root_uid) ||
1030 uid_eq(old->suid, root_uid)) &&
1031 (!uid_eq(new->uid, root_uid) &&
1032 !uid_eq(new->euid, root_uid) &&
1033 !uid_eq(new->suid, root_uid))) {
1034 if (!issecure(SECURE_KEEP_CAPS)) {
1035 cap_clear(new->cap_permitted);
1036 cap_clear(new->cap_effective);
1040 * Pre-ambient programs expect setresuid to nonroot followed
1041 * by exec to drop capabilities. We should make sure that
1042 * this remains the case.
1044 cap_clear(new->cap_ambient);
1046 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1047 cap_clear(new->cap_effective);
1048 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1049 new->cap_effective = new->cap_permitted;
1053 * cap_task_fix_setuid - Fix up the results of setuid() call
1054 * @new: The proposed credentials
1055 * @old: The current task's current credentials
1056 * @flags: Indications of what has changed
1058 * Fix up the results of setuid() call before the credential changes are
1059 * actually applied, returning 0 to grant the changes, -ve to deny them.
1061 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1067 /* juggle the capabilities to follow [RES]UID changes unless
1068 * otherwise suppressed */
1069 if (!issecure(SECURE_NO_SETUID_FIXUP))
1070 cap_emulate_setxuid(new, old);
1074 /* juggle the capabilties to follow FSUID changes, unless
1075 * otherwise suppressed
1077 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1078 * if not, we might be a bit too harsh here.
1080 if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1081 kuid_t root_uid = make_kuid(old->user_ns, 0);
1082 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1083 new->cap_effective =
1084 cap_drop_fs_set(new->cap_effective);
1086 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1087 new->cap_effective =
1088 cap_raise_fs_set(new->cap_effective,
1089 new->cap_permitted);
1101 * Rationale: code calling task_setscheduler, task_setioprio, and
1102 * task_setnice, assumes that
1103 * . if capable(cap_sys_nice), then those actions should be allowed
1104 * . if not capable(cap_sys_nice), but acting on your own processes,
1105 * then those actions should be allowed
1106 * This is insufficient now since you can call code without suid, but
1107 * yet with increased caps.
1108 * So we check for increased caps on the target process.
1110 static int cap_safe_nice(struct task_struct *p)
1112 int is_subset, ret = 0;
1115 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1116 current_cred()->cap_permitted);
1117 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1125 * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1126 * @p: The task to affect
1128 * Detemine if the requested scheduler policy change is permitted for the
1129 * specified task, returning 0 if permission is granted, -ve if denied.
1131 int cap_task_setscheduler(struct task_struct *p)
1133 return cap_safe_nice(p);
1137 * cap_task_ioprio - Detemine if I/O priority change is permitted
1138 * @p: The task to affect
1139 * @ioprio: The I/O priority to set
1141 * Detemine if the requested I/O priority change is permitted for the specified
1142 * task, returning 0 if permission is granted, -ve if denied.
1144 int cap_task_setioprio(struct task_struct *p, int ioprio)
1146 return cap_safe_nice(p);
1150 * cap_task_ioprio - Detemine if task priority change is permitted
1151 * @p: The task to affect
1152 * @nice: The nice value to set
1154 * Detemine if the requested task priority change is permitted for the
1155 * specified task, returning 0 if permission is granted, -ve if denied.
1157 int cap_task_setnice(struct task_struct *p, int nice)
1159 return cap_safe_nice(p);
1163 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from
1164 * the current task's bounding set. Returns 0 on success, -ve on error.
1166 static int cap_prctl_drop(unsigned long cap)
1170 if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1172 if (!cap_valid(cap))
1175 new = prepare_creds();
1178 cap_lower(new->cap_bset, cap);
1179 return commit_creds(new);
1183 * cap_task_prctl - Implement process control functions for this security module
1184 * @option: The process control function requested
1185 * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1187 * Allow process control functions (sys_prctl()) to alter capabilities; may
1188 * also deny access to other functions not otherwise implemented here.
1190 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1191 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM
1192 * modules will consider performing the function.
1194 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1195 unsigned long arg4, unsigned long arg5)
1197 const struct cred *old = current_cred();
1201 case PR_CAPBSET_READ:
1202 if (!cap_valid(arg2))
1204 return !!cap_raised(old->cap_bset, arg2);
1206 case PR_CAPBSET_DROP:
1207 return cap_prctl_drop(arg2);
1210 * The next four prctl's remain to assist with transitioning a
1211 * system from legacy UID=0 based privilege (when filesystem
1212 * capabilities are not in use) to a system using filesystem
1213 * capabilities only - as the POSIX.1e draft intended.
1217 * PR_SET_SECUREBITS =
1218 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1219 * | issecure_mask(SECURE_NOROOT)
1220 * | issecure_mask(SECURE_NOROOT_LOCKED)
1221 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
1222 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1224 * will ensure that the current process and all of its
1225 * children will be locked into a pure
1226 * capability-based-privilege environment.
1228 case PR_SET_SECUREBITS:
1229 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1230 & (old->securebits ^ arg2)) /*[1]*/
1231 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/
1232 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
1233 || (cap_capable(current_cred(),
1234 current_cred()->user_ns,
1236 CAP_OPT_NONE) != 0) /*[4]*/
1238 * [1] no changing of bits that are locked
1239 * [2] no unlocking of locks
1240 * [3] no setting of unsupported bits
1241 * [4] doing anything requires privilege (go read about
1242 * the "sendmail capabilities bug")
1245 /* cannot change a locked bit */
1248 new = prepare_creds();
1251 new->securebits = arg2;
1252 return commit_creds(new);
1254 case PR_GET_SECUREBITS:
1255 return old->securebits;
1257 case PR_GET_KEEPCAPS:
1258 return !!issecure(SECURE_KEEP_CAPS);
1260 case PR_SET_KEEPCAPS:
1261 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1263 if (issecure(SECURE_KEEP_CAPS_LOCKED))
1266 new = prepare_creds();
1270 new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1272 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1273 return commit_creds(new);
1275 case PR_CAP_AMBIENT:
1276 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1277 if (arg3 | arg4 | arg5)
1280 new = prepare_creds();
1283 cap_clear(new->cap_ambient);
1284 return commit_creds(new);
1287 if (((!cap_valid(arg3)) | arg4 | arg5))
1290 if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1291 return !!cap_raised(current_cred()->cap_ambient, arg3);
1292 } else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1293 arg2 != PR_CAP_AMBIENT_LOWER) {
1296 if (arg2 == PR_CAP_AMBIENT_RAISE &&
1297 (!cap_raised(current_cred()->cap_permitted, arg3) ||
1298 !cap_raised(current_cred()->cap_inheritable,
1300 issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1303 new = prepare_creds();
1306 if (arg2 == PR_CAP_AMBIENT_RAISE)
1307 cap_raise(new->cap_ambient, arg3);
1309 cap_lower(new->cap_ambient, arg3);
1310 return commit_creds(new);
1314 /* No functionality available - continue with default */
1320 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1321 * @mm: The VM space in which the new mapping is to be made
1322 * @pages: The size of the mapping
1324 * Determine whether the allocation of a new virtual mapping by the current
1325 * task is permitted, returning 1 if permission is granted, 0 if not.
1327 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1329 int cap_sys_admin = 0;
1331 if (cap_capable(current_cred(), &init_user_ns,
1332 CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1335 return cap_sys_admin;
1339 * cap_mmap_addr - check if able to map given addr
1340 * @addr: address attempting to be mapped
1342 * If the process is attempting to map memory below dac_mmap_min_addr they need
1343 * CAP_SYS_RAWIO. The other parameters to this function are unused by the
1344 * capability security module. Returns 0 if this mapping should be allowed
1347 int cap_mmap_addr(unsigned long addr)
1351 if (addr < dac_mmap_min_addr) {
1352 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1354 /* set PF_SUPERPRIV if it turns out we allow the low mmap */
1356 current->flags |= PF_SUPERPRIV;
1361 int cap_mmap_file(struct file *file, unsigned long reqprot,
1362 unsigned long prot, unsigned long flags)
1367 #ifdef CONFIG_SECURITY
1369 struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1370 LSM_HOOK_INIT(capable, cap_capable),
1371 LSM_HOOK_INIT(settime, cap_settime),
1372 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1373 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1374 LSM_HOOK_INIT(capget, cap_capget),
1375 LSM_HOOK_INIT(capset, cap_capset),
1376 LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1377 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1378 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1379 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1380 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1381 LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1382 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1383 LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1384 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1385 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1386 LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1387 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1390 void __init capability_add_hooks(void)
1392 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1396 #endif /* CONFIG_SECURITY */