xref: /openbmc/linux/security/commoncap.c (revision 1f0214a8)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Common capabilities, needed by capability.o.
3  */
4 
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>
11 #include <linux/mm.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>
27 #include <linux/mnt_idmapping.h>
28 
29 /*
30  * If a non-root user executes a setuid-root binary in
31  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
32  * However if fE is also set, then the intent is for only
33  * the file capabilities to be applied, and the setuid-root
34  * bit is left on either to change the uid (plausible) or
35  * to get full privilege on a kernel without file capabilities
36  * support.  So in that case we do not raise capabilities.
37  *
38  * Warn if that happens, once per boot.
39  */
40 static void warn_setuid_and_fcaps_mixed(const char *fname)
41 {
42 	static int warned;
43 	if (!warned) {
44 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
45 			" effective capabilities. Therefore not raising all"
46 			" capabilities.\n", fname);
47 		warned = 1;
48 	}
49 }
50 
51 /**
52  * cap_capable - Determine whether a task has a particular effective capability
53  * @cred: The credentials to use
54  * @targ_ns:  The user namespace in which we need the capability
55  * @cap: The capability to check for
56  * @opts: Bitmask of options defined in include/linux/security.h
57  *
58  * Determine whether the nominated task has the specified capability amongst
59  * its effective set, returning 0 if it does, -ve if it does not.
60  *
61  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
62  * and has_capability() functions.  That is, it has the reverse semantics:
63  * cap_has_capability() returns 0 when a task has a capability, but the
64  * kernel's capable() and has_capability() returns 1 for this case.
65  */
66 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
67 		int cap, unsigned int opts)
68 {
69 	struct user_namespace *ns = targ_ns;
70 
71 	/* See if cred has the capability in the target user namespace
72 	 * by examining the target user namespace and all of the target
73 	 * user namespace's parents.
74 	 */
75 	for (;;) {
76 		/* Do we have the necessary capabilities? */
77 		if (ns == cred->user_ns)
78 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
79 
80 		/*
81 		 * If we're already at a lower level than we're looking for,
82 		 * we're done searching.
83 		 */
84 		if (ns->level <= cred->user_ns->level)
85 			return -EPERM;
86 
87 		/*
88 		 * The owner of the user namespace in the parent of the
89 		 * user namespace has all caps.
90 		 */
91 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
92 			return 0;
93 
94 		/*
95 		 * If you have a capability in a parent user ns, then you have
96 		 * it over all children user namespaces as well.
97 		 */
98 		ns = ns->parent;
99 	}
100 
101 	/* We never get here */
102 }
103 
104 /**
105  * cap_settime - Determine whether the current process may set the system clock
106  * @ts: The time to set
107  * @tz: The timezone to set
108  *
109  * Determine whether the current process may set the system clock and timezone
110  * information, returning 0 if permission granted, -ve if denied.
111  */
112 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
113 {
114 	if (!capable(CAP_SYS_TIME))
115 		return -EPERM;
116 	return 0;
117 }
118 
119 /**
120  * cap_ptrace_access_check - Determine whether the current process may access
121  *			   another
122  * @child: The process to be accessed
123  * @mode: The mode of attachment.
124  *
125  * If we are in the same or an ancestor user_ns and have all the target
126  * task's capabilities, then ptrace access is allowed.
127  * If we have the ptrace capability to the target user_ns, then ptrace
128  * access is allowed.
129  * Else denied.
130  *
131  * Determine whether a process may access another, returning 0 if permission
132  * granted, -ve if denied.
133  */
134 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
135 {
136 	int ret = 0;
137 	const struct cred *cred, *child_cred;
138 	const kernel_cap_t *caller_caps;
139 
140 	rcu_read_lock();
141 	cred = current_cred();
142 	child_cred = __task_cred(child);
143 	if (mode & PTRACE_MODE_FSCREDS)
144 		caller_caps = &cred->cap_effective;
145 	else
146 		caller_caps = &cred->cap_permitted;
147 	if (cred->user_ns == child_cred->user_ns &&
148 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
149 		goto out;
150 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
151 		goto out;
152 	ret = -EPERM;
153 out:
154 	rcu_read_unlock();
155 	return ret;
156 }
157 
158 /**
159  * cap_ptrace_traceme - Determine whether another process may trace the current
160  * @parent: The task proposed to be the tracer
161  *
162  * If parent is in the same or an ancestor user_ns and has all current's
163  * capabilities, then ptrace access is allowed.
164  * If parent has the ptrace capability to current's user_ns, then ptrace
165  * access is allowed.
166  * Else denied.
167  *
168  * Determine whether the nominated task is permitted to trace the current
169  * process, returning 0 if permission is granted, -ve if denied.
170  */
171 int cap_ptrace_traceme(struct task_struct *parent)
172 {
173 	int ret = 0;
174 	const struct cred *cred, *child_cred;
175 
176 	rcu_read_lock();
177 	cred = __task_cred(parent);
178 	child_cred = current_cred();
179 	if (cred->user_ns == child_cred->user_ns &&
180 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
181 		goto out;
182 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
183 		goto out;
184 	ret = -EPERM;
185 out:
186 	rcu_read_unlock();
187 	return ret;
188 }
189 
190 /**
191  * cap_capget - Retrieve a task's capability sets
192  * @target: The task from which to retrieve the capability sets
193  * @effective: The place to record the effective set
194  * @inheritable: The place to record the inheritable set
195  * @permitted: The place to record the permitted set
196  *
197  * This function retrieves the capabilities of the nominated task and returns
198  * them to the caller.
199  */
200 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
201 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
202 {
203 	const struct cred *cred;
204 
205 	/* Derived from kernel/capability.c:sys_capget. */
206 	rcu_read_lock();
207 	cred = __task_cred(target);
208 	*effective   = cred->cap_effective;
209 	*inheritable = cred->cap_inheritable;
210 	*permitted   = cred->cap_permitted;
211 	rcu_read_unlock();
212 	return 0;
213 }
214 
215 /*
216  * Determine whether the inheritable capabilities are limited to the old
217  * permitted set.  Returns 1 if they are limited, 0 if they are not.
218  */
219 static inline int cap_inh_is_capped(void)
220 {
221 	/* they are so limited unless the current task has the CAP_SETPCAP
222 	 * capability
223 	 */
224 	if (cap_capable(current_cred(), current_cred()->user_ns,
225 			CAP_SETPCAP, CAP_OPT_NONE) == 0)
226 		return 0;
227 	return 1;
228 }
229 
230 /**
231  * cap_capset - Validate and apply proposed changes to current's capabilities
232  * @new: The proposed new credentials; alterations should be made here
233  * @old: The current task's current credentials
234  * @effective: A pointer to the proposed new effective capabilities set
235  * @inheritable: A pointer to the proposed new inheritable capabilities set
236  * @permitted: A pointer to the proposed new permitted capabilities set
237  *
238  * This function validates and applies a proposed mass change to the current
239  * process's capability sets.  The changes are made to the proposed new
240  * credentials, and assuming no error, will be committed by the caller of LSM.
241  */
242 int cap_capset(struct cred *new,
243 	       const struct cred *old,
244 	       const kernel_cap_t *effective,
245 	       const kernel_cap_t *inheritable,
246 	       const kernel_cap_t *permitted)
247 {
248 	if (cap_inh_is_capped() &&
249 	    !cap_issubset(*inheritable,
250 			  cap_combine(old->cap_inheritable,
251 				      old->cap_permitted)))
252 		/* incapable of using this inheritable set */
253 		return -EPERM;
254 
255 	if (!cap_issubset(*inheritable,
256 			  cap_combine(old->cap_inheritable,
257 				      old->cap_bset)))
258 		/* no new pI capabilities outside bounding set */
259 		return -EPERM;
260 
261 	/* verify restrictions on target's new Permitted set */
262 	if (!cap_issubset(*permitted, old->cap_permitted))
263 		return -EPERM;
264 
265 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
266 	if (!cap_issubset(*effective, *permitted))
267 		return -EPERM;
268 
269 	new->cap_effective   = *effective;
270 	new->cap_inheritable = *inheritable;
271 	new->cap_permitted   = *permitted;
272 
273 	/*
274 	 * Mask off ambient bits that are no longer both permitted and
275 	 * inheritable.
276 	 */
277 	new->cap_ambient = cap_intersect(new->cap_ambient,
278 					 cap_intersect(*permitted,
279 						       *inheritable));
280 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
281 		return -EINVAL;
282 	return 0;
283 }
284 
285 /**
286  * cap_inode_need_killpriv - Determine if inode change affects privileges
287  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
288  *
289  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
290  * affects the security markings on that inode, and if it is, should
291  * inode_killpriv() be invoked or the change rejected.
292  *
293  * Return: 1 if security.capability has a value, meaning inode_killpriv()
294  * is required, 0 otherwise, meaning inode_killpriv() is not required.
295  */
296 int cap_inode_need_killpriv(struct dentry *dentry)
297 {
298 	struct inode *inode = d_backing_inode(dentry);
299 	int error;
300 
301 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
302 	return error > 0;
303 }
304 
305 /**
306  * cap_inode_killpriv - Erase the security markings on an inode
307  *
308  * @mnt_userns:	user namespace of the mount the inode was found from
309  * @dentry:	The inode/dentry to alter
310  *
311  * Erase the privilege-enhancing security markings on an inode.
312  *
313  * If the inode has been found through an idmapped mount the user namespace of
314  * the vfsmount must be passed through @mnt_userns. This function will then
315  * take care to map the inode according to @mnt_userns before checking
316  * permissions. On non-idmapped mounts or if permission checking is to be
317  * performed on the raw inode simply passs init_user_ns.
318  *
319  * Return: 0 if successful, -ve on error.
320  */
321 int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry)
322 {
323 	int error;
324 
325 	error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS);
326 	if (error == -EOPNOTSUPP)
327 		error = 0;
328 	return error;
329 }
330 
331 static bool rootid_owns_currentns(kuid_t kroot)
332 {
333 	struct user_namespace *ns;
334 
335 	if (!uid_valid(kroot))
336 		return false;
337 
338 	for (ns = current_user_ns(); ; ns = ns->parent) {
339 		if (from_kuid(ns, kroot) == 0)
340 			return true;
341 		if (ns == &init_user_ns)
342 			break;
343 	}
344 
345 	return false;
346 }
347 
348 static __u32 sansflags(__u32 m)
349 {
350 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
351 }
352 
353 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
354 {
355 	if (size != XATTR_CAPS_SZ_2)
356 		return false;
357 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
358 }
359 
360 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
361 {
362 	if (size != XATTR_CAPS_SZ_3)
363 		return false;
364 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
365 }
366 
367 /*
368  * getsecurity: We are called for security.* before any attempt to read the
369  * xattr from the inode itself.
370  *
371  * This gives us a chance to read the on-disk value and convert it.  If we
372  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
373  *
374  * Note we are not called by vfs_getxattr_alloc(), but that is only called
375  * by the integrity subsystem, which really wants the unconverted values -
376  * so that's good.
377  */
378 int cap_inode_getsecurity(struct user_namespace *mnt_userns,
379 			  struct inode *inode, const char *name, void **buffer,
380 			  bool alloc)
381 {
382 	int size, ret;
383 	kuid_t kroot;
384 	u32 nsmagic, magic;
385 	uid_t root, mappedroot;
386 	char *tmpbuf = NULL;
387 	struct vfs_cap_data *cap;
388 	struct vfs_ns_cap_data *nscap = NULL;
389 	struct dentry *dentry;
390 	struct user_namespace *fs_ns;
391 
392 	if (strcmp(name, "capability") != 0)
393 		return -EOPNOTSUPP;
394 
395 	dentry = d_find_any_alias(inode);
396 	if (!dentry)
397 		return -EINVAL;
398 
399 	size = sizeof(struct vfs_ns_cap_data);
400 	ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
401 				      &tmpbuf, size, GFP_NOFS);
402 	dput(dentry);
403 
404 	if (ret < 0 || !tmpbuf)
405 		return ret;
406 
407 	fs_ns = inode->i_sb->s_user_ns;
408 	cap = (struct vfs_cap_data *) tmpbuf;
409 	if (is_v2header((size_t) ret, cap)) {
410 		root = 0;
411 	} else if (is_v3header((size_t) ret, cap)) {
412 		nscap = (struct vfs_ns_cap_data *) tmpbuf;
413 		root = le32_to_cpu(nscap->rootid);
414 	} else {
415 		size = -EINVAL;
416 		goto out_free;
417 	}
418 
419 	kroot = make_kuid(fs_ns, root);
420 
421 	/* If this is an idmapped mount shift the kuid. */
422 	kroot = mapped_kuid_fs(mnt_userns, fs_ns, kroot);
423 
424 	/* If the root kuid maps to a valid uid in current ns, then return
425 	 * this as a nscap. */
426 	mappedroot = from_kuid(current_user_ns(), kroot);
427 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
428 		size = sizeof(struct vfs_ns_cap_data);
429 		if (alloc) {
430 			if (!nscap) {
431 				/* v2 -> v3 conversion */
432 				nscap = kzalloc(size, GFP_ATOMIC);
433 				if (!nscap) {
434 					size = -ENOMEM;
435 					goto out_free;
436 				}
437 				nsmagic = VFS_CAP_REVISION_3;
438 				magic = le32_to_cpu(cap->magic_etc);
439 				if (magic & VFS_CAP_FLAGS_EFFECTIVE)
440 					nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
441 				memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
442 				nscap->magic_etc = cpu_to_le32(nsmagic);
443 			} else {
444 				/* use allocated v3 buffer */
445 				tmpbuf = NULL;
446 			}
447 			nscap->rootid = cpu_to_le32(mappedroot);
448 			*buffer = nscap;
449 		}
450 		goto out_free;
451 	}
452 
453 	if (!rootid_owns_currentns(kroot)) {
454 		size = -EOVERFLOW;
455 		goto out_free;
456 	}
457 
458 	/* This comes from a parent namespace.  Return as a v2 capability */
459 	size = sizeof(struct vfs_cap_data);
460 	if (alloc) {
461 		if (nscap) {
462 			/* v3 -> v2 conversion */
463 			cap = kzalloc(size, GFP_ATOMIC);
464 			if (!cap) {
465 				size = -ENOMEM;
466 				goto out_free;
467 			}
468 			magic = VFS_CAP_REVISION_2;
469 			nsmagic = le32_to_cpu(nscap->magic_etc);
470 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
471 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
472 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
473 			cap->magic_etc = cpu_to_le32(magic);
474 		} else {
475 			/* use unconverted v2 */
476 			tmpbuf = NULL;
477 		}
478 		*buffer = cap;
479 	}
480 out_free:
481 	kfree(tmpbuf);
482 	return size;
483 }
484 
485 /**
486  * rootid_from_xattr - translate root uid of vfs caps
487  *
488  * @value:	vfs caps value which may be modified by this function
489  * @size:	size of @ivalue
490  * @task_ns:	user namespace of the caller
491  * @mnt_userns:	user namespace of the mount the inode was found from
492  * @fs_userns:	user namespace of the filesystem
493  *
494  * If the inode has been found through an idmapped mount the user namespace of
495  * the vfsmount must be passed through @mnt_userns. This function will then
496  * take care to map the inode according to @mnt_userns before checking
497  * permissions. On non-idmapped mounts or if permission checking is to be
498  * performed on the raw inode simply passs init_user_ns.
499  */
500 static kuid_t rootid_from_xattr(const void *value, size_t size,
501 				struct user_namespace *task_ns,
502 				struct user_namespace *mnt_userns,
503 				struct user_namespace *fs_userns)
504 {
505 	const struct vfs_ns_cap_data *nscap = value;
506 	kuid_t rootkid;
507 	uid_t rootid = 0;
508 
509 	if (size == XATTR_CAPS_SZ_3)
510 		rootid = le32_to_cpu(nscap->rootid);
511 
512 	rootkid = make_kuid(task_ns, rootid);
513 	return mapped_kuid_user(mnt_userns, fs_userns, rootkid);
514 }
515 
516 static bool validheader(size_t size, const struct vfs_cap_data *cap)
517 {
518 	return is_v2header(size, cap) || is_v3header(size, cap);
519 }
520 
521 /**
522  * cap_convert_nscap - check vfs caps
523  *
524  * @mnt_userns:	user namespace of the mount the inode was found from
525  * @dentry:	used to retrieve inode to check permissions on
526  * @ivalue:	vfs caps value which may be modified by this function
527  * @size:	size of @ivalue
528  *
529  * User requested a write of security.capability.  If needed, update the
530  * xattr to change from v2 to v3, or to fixup the v3 rootid.
531  *
532  * If the inode has been found through an idmapped mount the user namespace of
533  * the vfsmount must be passed through @mnt_userns. This function will then
534  * take care to map the inode according to @mnt_userns before checking
535  * permissions. On non-idmapped mounts or if permission checking is to be
536  * performed on the raw inode simply passs init_user_ns.
537  *
538  * Return: On success, return the new size; on error, return < 0.
539  */
540 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
541 		      const void **ivalue, size_t size)
542 {
543 	struct vfs_ns_cap_data *nscap;
544 	uid_t nsrootid;
545 	const struct vfs_cap_data *cap = *ivalue;
546 	__u32 magic, nsmagic;
547 	struct inode *inode = d_backing_inode(dentry);
548 	struct user_namespace *task_ns = current_user_ns(),
549 		*fs_ns = inode->i_sb->s_user_ns;
550 	kuid_t rootid;
551 	size_t newsize;
552 
553 	if (!*ivalue)
554 		return -EINVAL;
555 	if (!validheader(size, cap))
556 		return -EINVAL;
557 	if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
558 		return -EPERM;
559 	if (size == XATTR_CAPS_SZ_2 && (mnt_userns == fs_ns))
560 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
561 			/* user is privileged, just write the v2 */
562 			return size;
563 
564 	rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns, fs_ns);
565 	if (!uid_valid(rootid))
566 		return -EINVAL;
567 
568 	nsrootid = from_kuid(fs_ns, rootid);
569 	if (nsrootid == -1)
570 		return -EINVAL;
571 
572 	newsize = sizeof(struct vfs_ns_cap_data);
573 	nscap = kmalloc(newsize, GFP_ATOMIC);
574 	if (!nscap)
575 		return -ENOMEM;
576 	nscap->rootid = cpu_to_le32(nsrootid);
577 	nsmagic = VFS_CAP_REVISION_3;
578 	magic = le32_to_cpu(cap->magic_etc);
579 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
580 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
581 	nscap->magic_etc = cpu_to_le32(nsmagic);
582 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
583 
584 	*ivalue = nscap;
585 	return newsize;
586 }
587 
588 /*
589  * Calculate the new process capability sets from the capability sets attached
590  * to a file.
591  */
592 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
593 					  struct linux_binprm *bprm,
594 					  bool *effective,
595 					  bool *has_fcap)
596 {
597 	struct cred *new = bprm->cred;
598 	unsigned i;
599 	int ret = 0;
600 
601 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
602 		*effective = true;
603 
604 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
605 		*has_fcap = true;
606 
607 	CAP_FOR_EACH_U32(i) {
608 		__u32 permitted = caps->permitted.cap[i];
609 		__u32 inheritable = caps->inheritable.cap[i];
610 
611 		/*
612 		 * pP' = (X & fP) | (pI & fI)
613 		 * The addition of pA' is handled later.
614 		 */
615 		new->cap_permitted.cap[i] =
616 			(new->cap_bset.cap[i] & permitted) |
617 			(new->cap_inheritable.cap[i] & inheritable);
618 
619 		if (permitted & ~new->cap_permitted.cap[i])
620 			/* insufficient to execute correctly */
621 			ret = -EPERM;
622 	}
623 
624 	/*
625 	 * For legacy apps, with no internal support for recognizing they
626 	 * do not have enough capabilities, we return an error if they are
627 	 * missing some "forced" (aka file-permitted) capabilities.
628 	 */
629 	return *effective ? ret : 0;
630 }
631 
632 /**
633  * get_vfs_caps_from_disk - retrieve vfs caps from disk
634  *
635  * @mnt_userns:	user namespace of the mount the inode was found from
636  * @dentry:	dentry from which @inode is retrieved
637  * @cpu_caps:	vfs capabilities
638  *
639  * Extract the on-exec-apply capability sets for an executable file.
640  *
641  * If the inode has been found through an idmapped mount the user namespace of
642  * the vfsmount must be passed through @mnt_userns. This function will then
643  * take care to map the inode according to @mnt_userns before checking
644  * permissions. On non-idmapped mounts or if permission checking is to be
645  * performed on the raw inode simply passs init_user_ns.
646  */
647 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
648 			   const struct dentry *dentry,
649 			   struct cpu_vfs_cap_data *cpu_caps)
650 {
651 	struct inode *inode = d_backing_inode(dentry);
652 	__u32 magic_etc;
653 	unsigned tocopy, i;
654 	int size;
655 	struct vfs_ns_cap_data data, *nscaps = &data;
656 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
657 	kuid_t rootkuid;
658 	struct user_namespace *fs_ns;
659 
660 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
661 
662 	if (!inode)
663 		return -ENODATA;
664 
665 	fs_ns = inode->i_sb->s_user_ns;
666 	size = __vfs_getxattr((struct dentry *)dentry, inode,
667 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
668 	if (size == -ENODATA || size == -EOPNOTSUPP)
669 		/* no data, that's ok */
670 		return -ENODATA;
671 
672 	if (size < 0)
673 		return size;
674 
675 	if (size < sizeof(magic_etc))
676 		return -EINVAL;
677 
678 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
679 
680 	rootkuid = make_kuid(fs_ns, 0);
681 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
682 	case VFS_CAP_REVISION_1:
683 		if (size != XATTR_CAPS_SZ_1)
684 			return -EINVAL;
685 		tocopy = VFS_CAP_U32_1;
686 		break;
687 	case VFS_CAP_REVISION_2:
688 		if (size != XATTR_CAPS_SZ_2)
689 			return -EINVAL;
690 		tocopy = VFS_CAP_U32_2;
691 		break;
692 	case VFS_CAP_REVISION_3:
693 		if (size != XATTR_CAPS_SZ_3)
694 			return -EINVAL;
695 		tocopy = VFS_CAP_U32_3;
696 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
697 		break;
698 
699 	default:
700 		return -EINVAL;
701 	}
702 	/* Limit the caps to the mounter of the filesystem
703 	 * or the more limited uid specified in the xattr.
704 	 */
705 	rootkuid = mapped_kuid_fs(mnt_userns, fs_ns, rootkuid);
706 	if (!rootid_owns_currentns(rootkuid))
707 		return -ENODATA;
708 
709 	CAP_FOR_EACH_U32(i) {
710 		if (i >= tocopy)
711 			break;
712 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
713 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
714 	}
715 
716 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
717 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
718 
719 	cpu_caps->rootid = rootkuid;
720 
721 	return 0;
722 }
723 
724 /*
725  * Attempt to get the on-exec apply capability sets for an executable file from
726  * its xattrs and, if present, apply them to the proposed credentials being
727  * constructed by execve().
728  */
729 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
730 			 bool *effective, bool *has_fcap)
731 {
732 	int rc = 0;
733 	struct cpu_vfs_cap_data vcaps;
734 
735 	cap_clear(bprm->cred->cap_permitted);
736 
737 	if (!file_caps_enabled)
738 		return 0;
739 
740 	if (!mnt_may_suid(file->f_path.mnt))
741 		return 0;
742 
743 	/*
744 	 * This check is redundant with mnt_may_suid() but is kept to make
745 	 * explicit that capability bits are limited to s_user_ns and its
746 	 * descendants.
747 	 */
748 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
749 		return 0;
750 
751 	rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
752 				    file->f_path.dentry, &vcaps);
753 	if (rc < 0) {
754 		if (rc == -EINVAL)
755 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
756 					bprm->filename);
757 		else if (rc == -ENODATA)
758 			rc = 0;
759 		goto out;
760 	}
761 
762 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
763 
764 out:
765 	if (rc)
766 		cap_clear(bprm->cred->cap_permitted);
767 
768 	return rc;
769 }
770 
771 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
772 
773 static inline bool __is_real(kuid_t uid, struct cred *cred)
774 { return uid_eq(cred->uid, uid); }
775 
776 static inline bool __is_eff(kuid_t uid, struct cred *cred)
777 { return uid_eq(cred->euid, uid); }
778 
779 static inline bool __is_suid(kuid_t uid, struct cred *cred)
780 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
781 
782 /*
783  * handle_privileged_root - Handle case of privileged root
784  * @bprm: The execution parameters, including the proposed creds
785  * @has_fcap: Are any file capabilities set?
786  * @effective: Do we have effective root privilege?
787  * @root_uid: This namespace' root UID WRT initial USER namespace
788  *
789  * Handle the case where root is privileged and hasn't been neutered by
790  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
791  * set UID root and nothing is changed.  If we are root, cap_permitted is
792  * updated.  If we have become set UID root, the effective bit is set.
793  */
794 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
795 				   bool *effective, kuid_t root_uid)
796 {
797 	const struct cred *old = current_cred();
798 	struct cred *new = bprm->cred;
799 
800 	if (!root_privileged())
801 		return;
802 	/*
803 	 * If the legacy file capability is set, then don't set privs
804 	 * for a setuid root binary run by a non-root user.  Do set it
805 	 * for a root user just to cause least surprise to an admin.
806 	 */
807 	if (has_fcap && __is_suid(root_uid, new)) {
808 		warn_setuid_and_fcaps_mixed(bprm->filename);
809 		return;
810 	}
811 	/*
812 	 * To support inheritance of root-permissions and suid-root
813 	 * executables under compatibility mode, we override the
814 	 * capability sets for the file.
815 	 */
816 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
817 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
818 		new->cap_permitted = cap_combine(old->cap_bset,
819 						 old->cap_inheritable);
820 	}
821 	/*
822 	 * If only the real uid is 0, we do not set the effective bit.
823 	 */
824 	if (__is_eff(root_uid, new))
825 		*effective = true;
826 }
827 
828 #define __cap_gained(field, target, source) \
829 	!cap_issubset(target->cap_##field, source->cap_##field)
830 #define __cap_grew(target, source, cred) \
831 	!cap_issubset(cred->cap_##target, cred->cap_##source)
832 #define __cap_full(field, cred) \
833 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
834 
835 static inline bool __is_setuid(struct cred *new, const struct cred *old)
836 { return !uid_eq(new->euid, old->uid); }
837 
838 static inline bool __is_setgid(struct cred *new, const struct cred *old)
839 { return !gid_eq(new->egid, old->gid); }
840 
841 /*
842  * 1) Audit candidate if current->cap_effective is set
843  *
844  * We do not bother to audit if 3 things are true:
845  *   1) cap_effective has all caps
846  *   2) we became root *OR* are were already root
847  *   3) root is supposed to have all caps (SECURE_NOROOT)
848  * Since this is just a normal root execing a process.
849  *
850  * Number 1 above might fail if you don't have a full bset, but I think
851  * that is interesting information to audit.
852  *
853  * A number of other conditions require logging:
854  * 2) something prevented setuid root getting all caps
855  * 3) non-setuid root gets fcaps
856  * 4) non-setuid root gets ambient
857  */
858 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
859 				     kuid_t root, bool has_fcap)
860 {
861 	bool ret = false;
862 
863 	if ((__cap_grew(effective, ambient, new) &&
864 	     !(__cap_full(effective, new) &&
865 	       (__is_eff(root, new) || __is_real(root, new)) &&
866 	       root_privileged())) ||
867 	    (root_privileged() &&
868 	     __is_suid(root, new) &&
869 	     !__cap_full(effective, new)) ||
870 	    (!__is_setuid(new, old) &&
871 	     ((has_fcap &&
872 	       __cap_gained(permitted, new, old)) ||
873 	      __cap_gained(ambient, new, old))))
874 
875 		ret = true;
876 
877 	return ret;
878 }
879 
880 /**
881  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
882  * @bprm: The execution parameters, including the proposed creds
883  * @file: The file to pull the credentials from
884  *
885  * Set up the proposed credentials for a new execution context being
886  * constructed by execve().  The proposed creds in @bprm->cred is altered,
887  * which won't take effect immediately.
888  *
889  * Return: 0 if successful, -ve on error.
890  */
891 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
892 {
893 	/* Process setpcap binaries and capabilities for uid 0 */
894 	const struct cred *old = current_cred();
895 	struct cred *new = bprm->cred;
896 	bool effective = false, has_fcap = false, is_setid;
897 	int ret;
898 	kuid_t root_uid;
899 
900 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
901 		return -EPERM;
902 
903 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
904 	if (ret < 0)
905 		return ret;
906 
907 	root_uid = make_kuid(new->user_ns, 0);
908 
909 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
910 
911 	/* if we have fs caps, clear dangerous personality flags */
912 	if (__cap_gained(permitted, new, old))
913 		bprm->per_clear |= PER_CLEAR_ON_SETID;
914 
915 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
916 	 * credentials unless they have the appropriate permit.
917 	 *
918 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
919 	 */
920 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
921 
922 	if ((is_setid || __cap_gained(permitted, new, old)) &&
923 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
924 	     !ptracer_capable(current, new->user_ns))) {
925 		/* downgrade; they get no more than they had, and maybe less */
926 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
927 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
928 			new->euid = new->uid;
929 			new->egid = new->gid;
930 		}
931 		new->cap_permitted = cap_intersect(new->cap_permitted,
932 						   old->cap_permitted);
933 	}
934 
935 	new->suid = new->fsuid = new->euid;
936 	new->sgid = new->fsgid = new->egid;
937 
938 	/* File caps or setid cancels ambient. */
939 	if (has_fcap || is_setid)
940 		cap_clear(new->cap_ambient);
941 
942 	/*
943 	 * Now that we've computed pA', update pP' to give:
944 	 *   pP' = (X & fP) | (pI & fI) | pA'
945 	 */
946 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
947 
948 	/*
949 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
950 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
951 	 */
952 	if (effective)
953 		new->cap_effective = new->cap_permitted;
954 	else
955 		new->cap_effective = new->cap_ambient;
956 
957 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
958 		return -EPERM;
959 
960 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
961 		ret = audit_log_bprm_fcaps(bprm, new, old);
962 		if (ret < 0)
963 			return ret;
964 	}
965 
966 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
967 
968 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
969 		return -EPERM;
970 
971 	/* Check for privilege-elevated exec. */
972 	if (is_setid ||
973 	    (!__is_real(root_uid, new) &&
974 	     (effective ||
975 	      __cap_grew(permitted, ambient, new))))
976 		bprm->secureexec = 1;
977 
978 	return 0;
979 }
980 
981 /**
982  * cap_inode_setxattr - Determine whether an xattr may be altered
983  * @dentry: The inode/dentry being altered
984  * @name: The name of the xattr to be changed
985  * @value: The value that the xattr will be changed to
986  * @size: The size of value
987  * @flags: The replacement flag
988  *
989  * Determine whether an xattr may be altered or set on an inode, returning 0 if
990  * permission is granted, -ve if denied.
991  *
992  * This is used to make sure security xattrs don't get updated or set by those
993  * who aren't privileged to do so.
994  */
995 int cap_inode_setxattr(struct dentry *dentry, const char *name,
996 		       const void *value, size_t size, int flags)
997 {
998 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
999 
1000 	/* Ignore non-security xattrs */
1001 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1002 			XATTR_SECURITY_PREFIX_LEN) != 0)
1003 		return 0;
1004 
1005 	/*
1006 	 * For XATTR_NAME_CAPS the check will be done in
1007 	 * cap_convert_nscap(), called by setxattr()
1008 	 */
1009 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
1010 		return 0;
1011 
1012 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1013 		return -EPERM;
1014 	return 0;
1015 }
1016 
1017 /**
1018  * cap_inode_removexattr - Determine whether an xattr may be removed
1019  *
1020  * @mnt_userns:	User namespace of the mount the inode was found from
1021  * @dentry:	The inode/dentry being altered
1022  * @name:	The name of the xattr to be changed
1023  *
1024  * Determine whether an xattr may be removed from an inode, returning 0 if
1025  * permission is granted, -ve if denied.
1026  *
1027  * If the inode has been found through an idmapped mount the user namespace of
1028  * the vfsmount must be passed through @mnt_userns. This function will then
1029  * take care to map the inode according to @mnt_userns before checking
1030  * permissions. On non-idmapped mounts or if permission checking is to be
1031  * performed on the raw inode simply passs init_user_ns.
1032  *
1033  * This is used to make sure security xattrs don't get removed by those who
1034  * aren't privileged to remove them.
1035  */
1036 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1037 			  struct dentry *dentry, const char *name)
1038 {
1039 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1040 
1041 	/* Ignore non-security xattrs */
1042 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1043 			XATTR_SECURITY_PREFIX_LEN) != 0)
1044 		return 0;
1045 
1046 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1047 		/* security.capability gets namespaced */
1048 		struct inode *inode = d_backing_inode(dentry);
1049 		if (!inode)
1050 			return -EINVAL;
1051 		if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1052 			return -EPERM;
1053 		return 0;
1054 	}
1055 
1056 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1057 		return -EPERM;
1058 	return 0;
1059 }
1060 
1061 /*
1062  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1063  * a process after a call to setuid, setreuid, or setresuid.
1064  *
1065  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1066  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1067  *  cleared.
1068  *
1069  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1070  *  capabilities of the process are cleared.
1071  *
1072  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1073  *  capabilities are set to the permitted capabilities.
1074  *
1075  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1076  *  never happen.
1077  *
1078  *  -astor
1079  *
1080  * cevans - New behaviour, Oct '99
1081  * A process may, via prctl(), elect to keep its capabilities when it
1082  * calls setuid() and switches away from uid==0. Both permitted and
1083  * effective sets will be retained.
1084  * Without this change, it was impossible for a daemon to drop only some
1085  * of its privilege. The call to setuid(!=0) would drop all privileges!
1086  * Keeping uid 0 is not an option because uid 0 owns too many vital
1087  * files..
1088  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1089  */
1090 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1091 {
1092 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1093 
1094 	if ((uid_eq(old->uid, root_uid) ||
1095 	     uid_eq(old->euid, root_uid) ||
1096 	     uid_eq(old->suid, root_uid)) &&
1097 	    (!uid_eq(new->uid, root_uid) &&
1098 	     !uid_eq(new->euid, root_uid) &&
1099 	     !uid_eq(new->suid, root_uid))) {
1100 		if (!issecure(SECURE_KEEP_CAPS)) {
1101 			cap_clear(new->cap_permitted);
1102 			cap_clear(new->cap_effective);
1103 		}
1104 
1105 		/*
1106 		 * Pre-ambient programs expect setresuid to nonroot followed
1107 		 * by exec to drop capabilities.  We should make sure that
1108 		 * this remains the case.
1109 		 */
1110 		cap_clear(new->cap_ambient);
1111 	}
1112 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1113 		cap_clear(new->cap_effective);
1114 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1115 		new->cap_effective = new->cap_permitted;
1116 }
1117 
1118 /**
1119  * cap_task_fix_setuid - Fix up the results of setuid() call
1120  * @new: The proposed credentials
1121  * @old: The current task's current credentials
1122  * @flags: Indications of what has changed
1123  *
1124  * Fix up the results of setuid() call before the credential changes are
1125  * actually applied.
1126  *
1127  * Return: 0 to grant the changes, -ve to deny them.
1128  */
1129 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1130 {
1131 	switch (flags) {
1132 	case LSM_SETID_RE:
1133 	case LSM_SETID_ID:
1134 	case LSM_SETID_RES:
1135 		/* juggle the capabilities to follow [RES]UID changes unless
1136 		 * otherwise suppressed */
1137 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1138 			cap_emulate_setxuid(new, old);
1139 		break;
1140 
1141 	case LSM_SETID_FS:
1142 		/* juggle the capabilties to follow FSUID changes, unless
1143 		 * otherwise suppressed
1144 		 *
1145 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1146 		 *          if not, we might be a bit too harsh here.
1147 		 */
1148 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1149 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1150 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1151 				new->cap_effective =
1152 					cap_drop_fs_set(new->cap_effective);
1153 
1154 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1155 				new->cap_effective =
1156 					cap_raise_fs_set(new->cap_effective,
1157 							 new->cap_permitted);
1158 		}
1159 		break;
1160 
1161 	default:
1162 		return -EINVAL;
1163 	}
1164 
1165 	return 0;
1166 }
1167 
1168 /*
1169  * Rationale: code calling task_setscheduler, task_setioprio, and
1170  * task_setnice, assumes that
1171  *   . if capable(cap_sys_nice), then those actions should be allowed
1172  *   . if not capable(cap_sys_nice), but acting on your own processes,
1173  *   	then those actions should be allowed
1174  * This is insufficient now since you can call code without suid, but
1175  * yet with increased caps.
1176  * So we check for increased caps on the target process.
1177  */
1178 static int cap_safe_nice(struct task_struct *p)
1179 {
1180 	int is_subset, ret = 0;
1181 
1182 	rcu_read_lock();
1183 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1184 				 current_cred()->cap_permitted);
1185 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1186 		ret = -EPERM;
1187 	rcu_read_unlock();
1188 
1189 	return ret;
1190 }
1191 
1192 /**
1193  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1194  * @p: The task to affect
1195  *
1196  * Detemine if the requested scheduler policy change is permitted for the
1197  * specified task.
1198  *
1199  * Return: 0 if permission is granted, -ve if denied.
1200  */
1201 int cap_task_setscheduler(struct task_struct *p)
1202 {
1203 	return cap_safe_nice(p);
1204 }
1205 
1206 /**
1207  * cap_task_setioprio - Detemine if I/O priority change is permitted
1208  * @p: The task to affect
1209  * @ioprio: The I/O priority to set
1210  *
1211  * Detemine if the requested I/O priority change is permitted for the specified
1212  * task.
1213  *
1214  * Return: 0 if permission is granted, -ve if denied.
1215  */
1216 int cap_task_setioprio(struct task_struct *p, int ioprio)
1217 {
1218 	return cap_safe_nice(p);
1219 }
1220 
1221 /**
1222  * cap_task_setnice - Detemine if task priority change is permitted
1223  * @p: The task to affect
1224  * @nice: The nice value to set
1225  *
1226  * Detemine if the requested task priority change is permitted for the
1227  * specified task.
1228  *
1229  * Return: 0 if permission is granted, -ve if denied.
1230  */
1231 int cap_task_setnice(struct task_struct *p, int nice)
1232 {
1233 	return cap_safe_nice(p);
1234 }
1235 
1236 /*
1237  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1238  * the current task's bounding set.  Returns 0 on success, -ve on error.
1239  */
1240 static int cap_prctl_drop(unsigned long cap)
1241 {
1242 	struct cred *new;
1243 
1244 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1245 		return -EPERM;
1246 	if (!cap_valid(cap))
1247 		return -EINVAL;
1248 
1249 	new = prepare_creds();
1250 	if (!new)
1251 		return -ENOMEM;
1252 	cap_lower(new->cap_bset, cap);
1253 	return commit_creds(new);
1254 }
1255 
1256 /**
1257  * cap_task_prctl - Implement process control functions for this security module
1258  * @option: The process control function requested
1259  * @arg2: The argument data for this function
1260  * @arg3: The argument data for this function
1261  * @arg4: The argument data for this function
1262  * @arg5: The argument data for this function
1263  *
1264  * Allow process control functions (sys_prctl()) to alter capabilities; may
1265  * also deny access to other functions not otherwise implemented here.
1266  *
1267  * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1268  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1269  * modules will consider performing the function.
1270  */
1271 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1272 		   unsigned long arg4, unsigned long arg5)
1273 {
1274 	const struct cred *old = current_cred();
1275 	struct cred *new;
1276 
1277 	switch (option) {
1278 	case PR_CAPBSET_READ:
1279 		if (!cap_valid(arg2))
1280 			return -EINVAL;
1281 		return !!cap_raised(old->cap_bset, arg2);
1282 
1283 	case PR_CAPBSET_DROP:
1284 		return cap_prctl_drop(arg2);
1285 
1286 	/*
1287 	 * The next four prctl's remain to assist with transitioning a
1288 	 * system from legacy UID=0 based privilege (when filesystem
1289 	 * capabilities are not in use) to a system using filesystem
1290 	 * capabilities only - as the POSIX.1e draft intended.
1291 	 *
1292 	 * Note:
1293 	 *
1294 	 *  PR_SET_SECUREBITS =
1295 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1296 	 *    | issecure_mask(SECURE_NOROOT)
1297 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1298 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1299 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1300 	 *
1301 	 * will ensure that the current process and all of its
1302 	 * children will be locked into a pure
1303 	 * capability-based-privilege environment.
1304 	 */
1305 	case PR_SET_SECUREBITS:
1306 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1307 		     & (old->securebits ^ arg2))			/*[1]*/
1308 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1309 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1310 		    || (cap_capable(current_cred(),
1311 				    current_cred()->user_ns,
1312 				    CAP_SETPCAP,
1313 				    CAP_OPT_NONE) != 0)			/*[4]*/
1314 			/*
1315 			 * [1] no changing of bits that are locked
1316 			 * [2] no unlocking of locks
1317 			 * [3] no setting of unsupported bits
1318 			 * [4] doing anything requires privilege (go read about
1319 			 *     the "sendmail capabilities bug")
1320 			 */
1321 		    )
1322 			/* cannot change a locked bit */
1323 			return -EPERM;
1324 
1325 		new = prepare_creds();
1326 		if (!new)
1327 			return -ENOMEM;
1328 		new->securebits = arg2;
1329 		return commit_creds(new);
1330 
1331 	case PR_GET_SECUREBITS:
1332 		return old->securebits;
1333 
1334 	case PR_GET_KEEPCAPS:
1335 		return !!issecure(SECURE_KEEP_CAPS);
1336 
1337 	case PR_SET_KEEPCAPS:
1338 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1339 			return -EINVAL;
1340 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1341 			return -EPERM;
1342 
1343 		new = prepare_creds();
1344 		if (!new)
1345 			return -ENOMEM;
1346 		if (arg2)
1347 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1348 		else
1349 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1350 		return commit_creds(new);
1351 
1352 	case PR_CAP_AMBIENT:
1353 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1354 			if (arg3 | arg4 | arg5)
1355 				return -EINVAL;
1356 
1357 			new = prepare_creds();
1358 			if (!new)
1359 				return -ENOMEM;
1360 			cap_clear(new->cap_ambient);
1361 			return commit_creds(new);
1362 		}
1363 
1364 		if (((!cap_valid(arg3)) | arg4 | arg5))
1365 			return -EINVAL;
1366 
1367 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1368 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1369 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1370 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1371 			return -EINVAL;
1372 		} else {
1373 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1374 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1375 			     !cap_raised(current_cred()->cap_inheritable,
1376 					 arg3) ||
1377 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1378 				return -EPERM;
1379 
1380 			new = prepare_creds();
1381 			if (!new)
1382 				return -ENOMEM;
1383 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1384 				cap_raise(new->cap_ambient, arg3);
1385 			else
1386 				cap_lower(new->cap_ambient, arg3);
1387 			return commit_creds(new);
1388 		}
1389 
1390 	default:
1391 		/* No functionality available - continue with default */
1392 		return -ENOSYS;
1393 	}
1394 }
1395 
1396 /**
1397  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1398  * @mm: The VM space in which the new mapping is to be made
1399  * @pages: The size of the mapping
1400  *
1401  * Determine whether the allocation of a new virtual mapping by the current
1402  * task is permitted.
1403  *
1404  * Return: 1 if permission is granted, 0 if not.
1405  */
1406 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1407 {
1408 	int cap_sys_admin = 0;
1409 
1410 	if (cap_capable(current_cred(), &init_user_ns,
1411 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1412 		cap_sys_admin = 1;
1413 
1414 	return cap_sys_admin;
1415 }
1416 
1417 /**
1418  * cap_mmap_addr - check if able to map given addr
1419  * @addr: address attempting to be mapped
1420  *
1421  * If the process is attempting to map memory below dac_mmap_min_addr they need
1422  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1423  * capability security module.
1424  *
1425  * Return: 0 if this mapping should be allowed or -EPERM if not.
1426  */
1427 int cap_mmap_addr(unsigned long addr)
1428 {
1429 	int ret = 0;
1430 
1431 	if (addr < dac_mmap_min_addr) {
1432 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1433 				  CAP_OPT_NONE);
1434 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1435 		if (ret == 0)
1436 			current->flags |= PF_SUPERPRIV;
1437 	}
1438 	return ret;
1439 }
1440 
1441 int cap_mmap_file(struct file *file, unsigned long reqprot,
1442 		  unsigned long prot, unsigned long flags)
1443 {
1444 	return 0;
1445 }
1446 
1447 #ifdef CONFIG_SECURITY
1448 
1449 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1450 	LSM_HOOK_INIT(capable, cap_capable),
1451 	LSM_HOOK_INIT(settime, cap_settime),
1452 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1453 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1454 	LSM_HOOK_INIT(capget, cap_capget),
1455 	LSM_HOOK_INIT(capset, cap_capset),
1456 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1457 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1458 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1459 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1460 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1461 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1462 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1463 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1464 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1465 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1466 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1467 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1468 };
1469 
1470 static int __init capability_init(void)
1471 {
1472 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1473 				"capability");
1474 	return 0;
1475 }
1476 
1477 DEFINE_LSM(capability) = {
1478 	.name = "capability",
1479 	.order = LSM_ORDER_FIRST,
1480 	.init = capability_init,
1481 };
1482 
1483 #endif /* CONFIG_SECURITY */
1484