xref: /openbmc/linux/security/commoncap.c (revision 0545810f)
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  * @idmap:	idmap 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 idmap of
314  * the vfsmount must be passed through @idmap. This function will then
315  * take care to map the inode according to @idmap before checking
316  * permissions. On non-idmapped mounts or if permission checking is to be
317  * performed on the raw inode simply passs @nop_mnt_idmap.
318  *
319  * Return: 0 if successful, -ve on error.
320  */
321 int cap_inode_killpriv(struct mnt_idmap *idmap, struct dentry *dentry)
322 {
323 	int error;
324 
325 	error = __vfs_removexattr(idmap, dentry, XATTR_NAME_CAPS);
326 	if (error == -EOPNOTSUPP)
327 		error = 0;
328 	return error;
329 }
330 
331 static bool rootid_owns_currentns(vfsuid_t rootvfsuid)
332 {
333 	struct user_namespace *ns;
334 	kuid_t kroot;
335 
336 	if (!vfsuid_valid(rootvfsuid))
337 		return false;
338 
339 	kroot = vfsuid_into_kuid(rootvfsuid);
340 	for (ns = current_user_ns();; ns = ns->parent) {
341 		if (from_kuid(ns, kroot) == 0)
342 			return true;
343 		if (ns == &init_user_ns)
344 			break;
345 	}
346 
347 	return false;
348 }
349 
350 static __u32 sansflags(__u32 m)
351 {
352 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
353 }
354 
355 static bool is_v2header(int size, const struct vfs_cap_data *cap)
356 {
357 	if (size != XATTR_CAPS_SZ_2)
358 		return false;
359 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
360 }
361 
362 static bool is_v3header(int size, const struct vfs_cap_data *cap)
363 {
364 	if (size != XATTR_CAPS_SZ_3)
365 		return false;
366 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
367 }
368 
369 /*
370  * getsecurity: We are called for security.* before any attempt to read the
371  * xattr from the inode itself.
372  *
373  * This gives us a chance to read the on-disk value and convert it.  If we
374  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
375  *
376  * Note we are not called by vfs_getxattr_alloc(), but that is only called
377  * by the integrity subsystem, which really wants the unconverted values -
378  * so that's good.
379  */
380 int cap_inode_getsecurity(struct mnt_idmap *idmap,
381 			  struct inode *inode, const char *name, void **buffer,
382 			  bool alloc)
383 {
384 	int size;
385 	kuid_t kroot;
386 	vfsuid_t vfsroot;
387 	u32 nsmagic, magic;
388 	uid_t root, mappedroot;
389 	char *tmpbuf = NULL;
390 	struct vfs_cap_data *cap;
391 	struct vfs_ns_cap_data *nscap = NULL;
392 	struct dentry *dentry;
393 	struct user_namespace *fs_ns;
394 
395 	if (strcmp(name, "capability") != 0)
396 		return -EOPNOTSUPP;
397 
398 	dentry = d_find_any_alias(inode);
399 	if (!dentry)
400 		return -EINVAL;
401 	size = vfs_getxattr_alloc(idmap, dentry, XATTR_NAME_CAPS, &tmpbuf,
402 				  sizeof(struct vfs_ns_cap_data), GFP_NOFS);
403 	dput(dentry);
404 	/* gcc11 complains if we don't check for !tmpbuf */
405 	if (size < 0 || !tmpbuf)
406 		goto out_free;
407 
408 	fs_ns = inode->i_sb->s_user_ns;
409 	cap = (struct vfs_cap_data *) tmpbuf;
410 	if (is_v2header(size, cap)) {
411 		root = 0;
412 	} else if (is_v3header(size, cap)) {
413 		nscap = (struct vfs_ns_cap_data *) tmpbuf;
414 		root = le32_to_cpu(nscap->rootid);
415 	} else {
416 		size = -EINVAL;
417 		goto out_free;
418 	}
419 
420 	kroot = make_kuid(fs_ns, root);
421 
422 	/* If this is an idmapped mount shift the kuid. */
423 	vfsroot = make_vfsuid(idmap, fs_ns, kroot);
424 
425 	/* If the root kuid maps to a valid uid in current ns, then return
426 	 * this as a nscap. */
427 	mappedroot = from_kuid(current_user_ns(), vfsuid_into_kuid(vfsroot));
428 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
429 		size = sizeof(struct vfs_ns_cap_data);
430 		if (alloc) {
431 			if (!nscap) {
432 				/* v2 -> v3 conversion */
433 				nscap = kzalloc(size, GFP_ATOMIC);
434 				if (!nscap) {
435 					size = -ENOMEM;
436 					goto out_free;
437 				}
438 				nsmagic = VFS_CAP_REVISION_3;
439 				magic = le32_to_cpu(cap->magic_etc);
440 				if (magic & VFS_CAP_FLAGS_EFFECTIVE)
441 					nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
442 				memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
443 				nscap->magic_etc = cpu_to_le32(nsmagic);
444 			} else {
445 				/* use allocated v3 buffer */
446 				tmpbuf = NULL;
447 			}
448 			nscap->rootid = cpu_to_le32(mappedroot);
449 			*buffer = nscap;
450 		}
451 		goto out_free;
452 	}
453 
454 	if (!rootid_owns_currentns(vfsroot)) {
455 		size = -EOVERFLOW;
456 		goto out_free;
457 	}
458 
459 	/* This comes from a parent namespace.  Return as a v2 capability */
460 	size = sizeof(struct vfs_cap_data);
461 	if (alloc) {
462 		if (nscap) {
463 			/* v3 -> v2 conversion */
464 			cap = kzalloc(size, GFP_ATOMIC);
465 			if (!cap) {
466 				size = -ENOMEM;
467 				goto out_free;
468 			}
469 			magic = VFS_CAP_REVISION_2;
470 			nsmagic = le32_to_cpu(nscap->magic_etc);
471 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
472 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
473 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
474 			cap->magic_etc = cpu_to_le32(magic);
475 		} else {
476 			/* use unconverted v2 */
477 			tmpbuf = NULL;
478 		}
479 		*buffer = cap;
480 	}
481 out_free:
482 	kfree(tmpbuf);
483 	return size;
484 }
485 
486 /**
487  * rootid_from_xattr - translate root uid of vfs caps
488  *
489  * @value:	vfs caps value which may be modified by this function
490  * @size:	size of @ivalue
491  * @task_ns:	user namespace of the caller
492  */
493 static vfsuid_t rootid_from_xattr(const void *value, size_t size,
494 				  struct user_namespace *task_ns)
495 {
496 	const struct vfs_ns_cap_data *nscap = value;
497 	uid_t rootid = 0;
498 
499 	if (size == XATTR_CAPS_SZ_3)
500 		rootid = le32_to_cpu(nscap->rootid);
501 
502 	return VFSUIDT_INIT(make_kuid(task_ns, rootid));
503 }
504 
505 static bool validheader(size_t size, const struct vfs_cap_data *cap)
506 {
507 	return is_v2header(size, cap) || is_v3header(size, cap);
508 }
509 
510 /**
511  * cap_convert_nscap - check vfs caps
512  *
513  * @idmap:	idmap of the mount the inode was found from
514  * @dentry:	used to retrieve inode to check permissions on
515  * @ivalue:	vfs caps value which may be modified by this function
516  * @size:	size of @ivalue
517  *
518  * User requested a write of security.capability.  If needed, update the
519  * xattr to change from v2 to v3, or to fixup the v3 rootid.
520  *
521  * If the inode has been found through an idmapped mount the idmap of
522  * the vfsmount must be passed through @idmap. This function will then
523  * take care to map the inode according to @idmap before checking
524  * permissions. On non-idmapped mounts or if permission checking is to be
525  * performed on the raw inode simply passs @nop_mnt_idmap.
526  *
527  * Return: On success, return the new size; on error, return < 0.
528  */
529 int cap_convert_nscap(struct mnt_idmap *idmap, struct dentry *dentry,
530 		      const void **ivalue, size_t size)
531 {
532 	struct vfs_ns_cap_data *nscap;
533 	uid_t nsrootid;
534 	const struct vfs_cap_data *cap = *ivalue;
535 	__u32 magic, nsmagic;
536 	struct inode *inode = d_backing_inode(dentry);
537 	struct user_namespace *task_ns = current_user_ns(),
538 		*fs_ns = inode->i_sb->s_user_ns;
539 	kuid_t rootid;
540 	vfsuid_t vfsrootid;
541 	size_t newsize;
542 
543 	if (!*ivalue)
544 		return -EINVAL;
545 	if (!validheader(size, cap))
546 		return -EINVAL;
547 	if (!capable_wrt_inode_uidgid(idmap, inode, CAP_SETFCAP))
548 		return -EPERM;
549 	if (size == XATTR_CAPS_SZ_2 && (idmap == &nop_mnt_idmap))
550 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
551 			/* user is privileged, just write the v2 */
552 			return size;
553 
554 	vfsrootid = rootid_from_xattr(*ivalue, size, task_ns);
555 	if (!vfsuid_valid(vfsrootid))
556 		return -EINVAL;
557 
558 	rootid = from_vfsuid(idmap, fs_ns, vfsrootid);
559 	if (!uid_valid(rootid))
560 		return -EINVAL;
561 
562 	nsrootid = from_kuid(fs_ns, rootid);
563 	if (nsrootid == -1)
564 		return -EINVAL;
565 
566 	newsize = sizeof(struct vfs_ns_cap_data);
567 	nscap = kmalloc(newsize, GFP_ATOMIC);
568 	if (!nscap)
569 		return -ENOMEM;
570 	nscap->rootid = cpu_to_le32(nsrootid);
571 	nsmagic = VFS_CAP_REVISION_3;
572 	magic = le32_to_cpu(cap->magic_etc);
573 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
574 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
575 	nscap->magic_etc = cpu_to_le32(nsmagic);
576 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
577 
578 	*ivalue = nscap;
579 	return newsize;
580 }
581 
582 /*
583  * Calculate the new process capability sets from the capability sets attached
584  * to a file.
585  */
586 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
587 					  struct linux_binprm *bprm,
588 					  bool *effective,
589 					  bool *has_fcap)
590 {
591 	struct cred *new = bprm->cred;
592 	int ret = 0;
593 
594 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
595 		*effective = true;
596 
597 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
598 		*has_fcap = true;
599 
600 	/*
601 	 * pP' = (X & fP) | (pI & fI)
602 	 * The addition of pA' is handled later.
603 	 */
604 	new->cap_permitted.val =
605 		(new->cap_bset.val & caps->permitted.val) |
606 		(new->cap_inheritable.val & caps->inheritable.val);
607 
608 	if (caps->permitted.val & ~new->cap_permitted.val)
609 		/* insufficient to execute correctly */
610 		ret = -EPERM;
611 
612 	/*
613 	 * For legacy apps, with no internal support for recognizing they
614 	 * do not have enough capabilities, we return an error if they are
615 	 * missing some "forced" (aka file-permitted) capabilities.
616 	 */
617 	return *effective ? ret : 0;
618 }
619 
620 /**
621  * get_vfs_caps_from_disk - retrieve vfs caps from disk
622  *
623  * @idmap:	idmap of the mount the inode was found from
624  * @dentry:	dentry from which @inode is retrieved
625  * @cpu_caps:	vfs capabilities
626  *
627  * Extract the on-exec-apply capability sets for an executable file.
628  *
629  * If the inode has been found through an idmapped mount the idmap of
630  * the vfsmount must be passed through @idmap. This function will then
631  * take care to map the inode according to @idmap before checking
632  * permissions. On non-idmapped mounts or if permission checking is to be
633  * performed on the raw inode simply passs @nop_mnt_idmap.
634  */
635 int get_vfs_caps_from_disk(struct mnt_idmap *idmap,
636 			   const struct dentry *dentry,
637 			   struct cpu_vfs_cap_data *cpu_caps)
638 {
639 	struct inode *inode = d_backing_inode(dentry);
640 	__u32 magic_etc;
641 	int size;
642 	struct vfs_ns_cap_data data, *nscaps = &data;
643 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
644 	kuid_t rootkuid;
645 	vfsuid_t rootvfsuid;
646 	struct user_namespace *fs_ns;
647 
648 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
649 
650 	if (!inode)
651 		return -ENODATA;
652 
653 	fs_ns = inode->i_sb->s_user_ns;
654 	size = __vfs_getxattr((struct dentry *)dentry, inode,
655 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
656 	if (size == -ENODATA || size == -EOPNOTSUPP)
657 		/* no data, that's ok */
658 		return -ENODATA;
659 
660 	if (size < 0)
661 		return size;
662 
663 	if (size < sizeof(magic_etc))
664 		return -EINVAL;
665 
666 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
667 
668 	rootkuid = make_kuid(fs_ns, 0);
669 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
670 	case VFS_CAP_REVISION_1:
671 		if (size != XATTR_CAPS_SZ_1)
672 			return -EINVAL;
673 		break;
674 	case VFS_CAP_REVISION_2:
675 		if (size != XATTR_CAPS_SZ_2)
676 			return -EINVAL;
677 		break;
678 	case VFS_CAP_REVISION_3:
679 		if (size != XATTR_CAPS_SZ_3)
680 			return -EINVAL;
681 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
682 		break;
683 
684 	default:
685 		return -EINVAL;
686 	}
687 
688 	rootvfsuid = make_vfsuid(idmap, fs_ns, rootkuid);
689 	if (!vfsuid_valid(rootvfsuid))
690 		return -ENODATA;
691 
692 	/* Limit the caps to the mounter of the filesystem
693 	 * or the more limited uid specified in the xattr.
694 	 */
695 	if (!rootid_owns_currentns(rootvfsuid))
696 		return -ENODATA;
697 
698 	cpu_caps->permitted.val = le32_to_cpu(caps->data[0].permitted);
699 	cpu_caps->inheritable.val = le32_to_cpu(caps->data[0].inheritable);
700 
701 	/*
702 	 * Rev1 had just a single 32-bit word, later expanded
703 	 * to a second one for the high bits
704 	 */
705 	if ((magic_etc & VFS_CAP_REVISION_MASK) != VFS_CAP_REVISION_1) {
706 		cpu_caps->permitted.val += (u64)le32_to_cpu(caps->data[1].permitted) << 32;
707 		cpu_caps->inheritable.val += (u64)le32_to_cpu(caps->data[1].inheritable) << 32;
708 	}
709 
710 	cpu_caps->permitted.val &= CAP_VALID_MASK;
711 	cpu_caps->inheritable.val &= CAP_VALID_MASK;
712 
713 	cpu_caps->rootid = vfsuid_into_kuid(rootvfsuid);
714 
715 	return 0;
716 }
717 
718 /*
719  * Attempt to get the on-exec apply capability sets for an executable file from
720  * its xattrs and, if present, apply them to the proposed credentials being
721  * constructed by execve().
722  */
723 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
724 			 bool *effective, bool *has_fcap)
725 {
726 	int rc = 0;
727 	struct cpu_vfs_cap_data vcaps;
728 
729 	cap_clear(bprm->cred->cap_permitted);
730 
731 	if (!file_caps_enabled)
732 		return 0;
733 
734 	if (!mnt_may_suid(file->f_path.mnt))
735 		return 0;
736 
737 	/*
738 	 * This check is redundant with mnt_may_suid() but is kept to make
739 	 * explicit that capability bits are limited to s_user_ns and its
740 	 * descendants.
741 	 */
742 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
743 		return 0;
744 
745 	rc = get_vfs_caps_from_disk(file_mnt_idmap(file),
746 				    file->f_path.dentry, &vcaps);
747 	if (rc < 0) {
748 		if (rc == -EINVAL)
749 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
750 					bprm->filename);
751 		else if (rc == -ENODATA)
752 			rc = 0;
753 		goto out;
754 	}
755 
756 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
757 
758 out:
759 	if (rc)
760 		cap_clear(bprm->cred->cap_permitted);
761 
762 	return rc;
763 }
764 
765 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
766 
767 static inline bool __is_real(kuid_t uid, struct cred *cred)
768 { return uid_eq(cred->uid, uid); }
769 
770 static inline bool __is_eff(kuid_t uid, struct cred *cred)
771 { return uid_eq(cred->euid, uid); }
772 
773 static inline bool __is_suid(kuid_t uid, struct cred *cred)
774 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
775 
776 /*
777  * handle_privileged_root - Handle case of privileged root
778  * @bprm: The execution parameters, including the proposed creds
779  * @has_fcap: Are any file capabilities set?
780  * @effective: Do we have effective root privilege?
781  * @root_uid: This namespace' root UID WRT initial USER namespace
782  *
783  * Handle the case where root is privileged and hasn't been neutered by
784  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
785  * set UID root and nothing is changed.  If we are root, cap_permitted is
786  * updated.  If we have become set UID root, the effective bit is set.
787  */
788 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
789 				   bool *effective, kuid_t root_uid)
790 {
791 	const struct cred *old = current_cred();
792 	struct cred *new = bprm->cred;
793 
794 	if (!root_privileged())
795 		return;
796 	/*
797 	 * If the legacy file capability is set, then don't set privs
798 	 * for a setuid root binary run by a non-root user.  Do set it
799 	 * for a root user just to cause least surprise to an admin.
800 	 */
801 	if (has_fcap && __is_suid(root_uid, new)) {
802 		warn_setuid_and_fcaps_mixed(bprm->filename);
803 		return;
804 	}
805 	/*
806 	 * To support inheritance of root-permissions and suid-root
807 	 * executables under compatibility mode, we override the
808 	 * capability sets for the file.
809 	 */
810 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
811 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
812 		new->cap_permitted = cap_combine(old->cap_bset,
813 						 old->cap_inheritable);
814 	}
815 	/*
816 	 * If only the real uid is 0, we do not set the effective bit.
817 	 */
818 	if (__is_eff(root_uid, new))
819 		*effective = true;
820 }
821 
822 #define __cap_gained(field, target, source) \
823 	!cap_issubset(target->cap_##field, source->cap_##field)
824 #define __cap_grew(target, source, cred) \
825 	!cap_issubset(cred->cap_##target, cred->cap_##source)
826 #define __cap_full(field, cred) \
827 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
828 
829 static inline bool __is_setuid(struct cred *new, const struct cred *old)
830 { return !uid_eq(new->euid, old->uid); }
831 
832 static inline bool __is_setgid(struct cred *new, const struct cred *old)
833 { return !gid_eq(new->egid, old->gid); }
834 
835 /*
836  * 1) Audit candidate if current->cap_effective is set
837  *
838  * We do not bother to audit if 3 things are true:
839  *   1) cap_effective has all caps
840  *   2) we became root *OR* are were already root
841  *   3) root is supposed to have all caps (SECURE_NOROOT)
842  * Since this is just a normal root execing a process.
843  *
844  * Number 1 above might fail if you don't have a full bset, but I think
845  * that is interesting information to audit.
846  *
847  * A number of other conditions require logging:
848  * 2) something prevented setuid root getting all caps
849  * 3) non-setuid root gets fcaps
850  * 4) non-setuid root gets ambient
851  */
852 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
853 				     kuid_t root, bool has_fcap)
854 {
855 	bool ret = false;
856 
857 	if ((__cap_grew(effective, ambient, new) &&
858 	     !(__cap_full(effective, new) &&
859 	       (__is_eff(root, new) || __is_real(root, new)) &&
860 	       root_privileged())) ||
861 	    (root_privileged() &&
862 	     __is_suid(root, new) &&
863 	     !__cap_full(effective, new)) ||
864 	    (!__is_setuid(new, old) &&
865 	     ((has_fcap &&
866 	       __cap_gained(permitted, new, old)) ||
867 	      __cap_gained(ambient, new, old))))
868 
869 		ret = true;
870 
871 	return ret;
872 }
873 
874 /**
875  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
876  * @bprm: The execution parameters, including the proposed creds
877  * @file: The file to pull the credentials from
878  *
879  * Set up the proposed credentials for a new execution context being
880  * constructed by execve().  The proposed creds in @bprm->cred is altered,
881  * which won't take effect immediately.
882  *
883  * Return: 0 if successful, -ve on error.
884  */
885 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
886 {
887 	/* Process setpcap binaries and capabilities for uid 0 */
888 	const struct cred *old = current_cred();
889 	struct cred *new = bprm->cred;
890 	bool effective = false, has_fcap = false, is_setid;
891 	int ret;
892 	kuid_t root_uid;
893 
894 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
895 		return -EPERM;
896 
897 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
898 	if (ret < 0)
899 		return ret;
900 
901 	root_uid = make_kuid(new->user_ns, 0);
902 
903 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
904 
905 	/* if we have fs caps, clear dangerous personality flags */
906 	if (__cap_gained(permitted, new, old))
907 		bprm->per_clear |= PER_CLEAR_ON_SETID;
908 
909 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
910 	 * credentials unless they have the appropriate permit.
911 	 *
912 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
913 	 */
914 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
915 
916 	if ((is_setid || __cap_gained(permitted, new, old)) &&
917 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
918 	     !ptracer_capable(current, new->user_ns))) {
919 		/* downgrade; they get no more than they had, and maybe less */
920 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
921 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
922 			new->euid = new->uid;
923 			new->egid = new->gid;
924 		}
925 		new->cap_permitted = cap_intersect(new->cap_permitted,
926 						   old->cap_permitted);
927 	}
928 
929 	new->suid = new->fsuid = new->euid;
930 	new->sgid = new->fsgid = new->egid;
931 
932 	/* File caps or setid cancels ambient. */
933 	if (has_fcap || is_setid)
934 		cap_clear(new->cap_ambient);
935 
936 	/*
937 	 * Now that we've computed pA', update pP' to give:
938 	 *   pP' = (X & fP) | (pI & fI) | pA'
939 	 */
940 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
941 
942 	/*
943 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
944 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
945 	 */
946 	if (effective)
947 		new->cap_effective = new->cap_permitted;
948 	else
949 		new->cap_effective = new->cap_ambient;
950 
951 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
952 		return -EPERM;
953 
954 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
955 		ret = audit_log_bprm_fcaps(bprm, new, old);
956 		if (ret < 0)
957 			return ret;
958 	}
959 
960 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
961 
962 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
963 		return -EPERM;
964 
965 	/* Check for privilege-elevated exec. */
966 	if (is_setid ||
967 	    (!__is_real(root_uid, new) &&
968 	     (effective ||
969 	      __cap_grew(permitted, ambient, new))))
970 		bprm->secureexec = 1;
971 
972 	return 0;
973 }
974 
975 /**
976  * cap_inode_setxattr - Determine whether an xattr may be altered
977  * @dentry: The inode/dentry being altered
978  * @name: The name of the xattr to be changed
979  * @value: The value that the xattr will be changed to
980  * @size: The size of value
981  * @flags: The replacement flag
982  *
983  * Determine whether an xattr may be altered or set on an inode, returning 0 if
984  * permission is granted, -ve if denied.
985  *
986  * This is used to make sure security xattrs don't get updated or set by those
987  * who aren't privileged to do so.
988  */
989 int cap_inode_setxattr(struct dentry *dentry, const char *name,
990 		       const void *value, size_t size, int flags)
991 {
992 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
993 
994 	/* Ignore non-security xattrs */
995 	if (strncmp(name, XATTR_SECURITY_PREFIX,
996 			XATTR_SECURITY_PREFIX_LEN) != 0)
997 		return 0;
998 
999 	/*
1000 	 * For XATTR_NAME_CAPS the check will be done in
1001 	 * cap_convert_nscap(), called by setxattr()
1002 	 */
1003 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
1004 		return 0;
1005 
1006 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1007 		return -EPERM;
1008 	return 0;
1009 }
1010 
1011 /**
1012  * cap_inode_removexattr - Determine whether an xattr may be removed
1013  *
1014  * @idmap:	idmap of the mount the inode was found from
1015  * @dentry:	The inode/dentry being altered
1016  * @name:	The name of the xattr to be changed
1017  *
1018  * Determine whether an xattr may be removed from an inode, returning 0 if
1019  * permission is granted, -ve if denied.
1020  *
1021  * If the inode has been found through an idmapped mount the idmap of
1022  * the vfsmount must be passed through @idmap. This function will then
1023  * take care to map the inode according to @idmap before checking
1024  * permissions. On non-idmapped mounts or if permission checking is to be
1025  * performed on the raw inode simply pass @nop_mnt_idmap.
1026  *
1027  * This is used to make sure security xattrs don't get removed by those who
1028  * aren't privileged to remove them.
1029  */
1030 int cap_inode_removexattr(struct mnt_idmap *idmap,
1031 			  struct dentry *dentry, const char *name)
1032 {
1033 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1034 
1035 	/* Ignore non-security xattrs */
1036 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1037 			XATTR_SECURITY_PREFIX_LEN) != 0)
1038 		return 0;
1039 
1040 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1041 		/* security.capability gets namespaced */
1042 		struct inode *inode = d_backing_inode(dentry);
1043 		if (!inode)
1044 			return -EINVAL;
1045 		if (!capable_wrt_inode_uidgid(idmap, inode, CAP_SETFCAP))
1046 			return -EPERM;
1047 		return 0;
1048 	}
1049 
1050 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1051 		return -EPERM;
1052 	return 0;
1053 }
1054 
1055 /*
1056  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1057  * a process after a call to setuid, setreuid, or setresuid.
1058  *
1059  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1060  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1061  *  cleared.
1062  *
1063  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1064  *  capabilities of the process are cleared.
1065  *
1066  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1067  *  capabilities are set to the permitted capabilities.
1068  *
1069  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1070  *  never happen.
1071  *
1072  *  -astor
1073  *
1074  * cevans - New behaviour, Oct '99
1075  * A process may, via prctl(), elect to keep its capabilities when it
1076  * calls setuid() and switches away from uid==0. Both permitted and
1077  * effective sets will be retained.
1078  * Without this change, it was impossible for a daemon to drop only some
1079  * of its privilege. The call to setuid(!=0) would drop all privileges!
1080  * Keeping uid 0 is not an option because uid 0 owns too many vital
1081  * files..
1082  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1083  */
1084 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1085 {
1086 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1087 
1088 	if ((uid_eq(old->uid, root_uid) ||
1089 	     uid_eq(old->euid, root_uid) ||
1090 	     uid_eq(old->suid, root_uid)) &&
1091 	    (!uid_eq(new->uid, root_uid) &&
1092 	     !uid_eq(new->euid, root_uid) &&
1093 	     !uid_eq(new->suid, root_uid))) {
1094 		if (!issecure(SECURE_KEEP_CAPS)) {
1095 			cap_clear(new->cap_permitted);
1096 			cap_clear(new->cap_effective);
1097 		}
1098 
1099 		/*
1100 		 * Pre-ambient programs expect setresuid to nonroot followed
1101 		 * by exec to drop capabilities.  We should make sure that
1102 		 * this remains the case.
1103 		 */
1104 		cap_clear(new->cap_ambient);
1105 	}
1106 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1107 		cap_clear(new->cap_effective);
1108 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1109 		new->cap_effective = new->cap_permitted;
1110 }
1111 
1112 /**
1113  * cap_task_fix_setuid - Fix up the results of setuid() call
1114  * @new: The proposed credentials
1115  * @old: The current task's current credentials
1116  * @flags: Indications of what has changed
1117  *
1118  * Fix up the results of setuid() call before the credential changes are
1119  * actually applied.
1120  *
1121  * Return: 0 to grant the changes, -ve to deny them.
1122  */
1123 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1124 {
1125 	switch (flags) {
1126 	case LSM_SETID_RE:
1127 	case LSM_SETID_ID:
1128 	case LSM_SETID_RES:
1129 		/* juggle the capabilities to follow [RES]UID changes unless
1130 		 * otherwise suppressed */
1131 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1132 			cap_emulate_setxuid(new, old);
1133 		break;
1134 
1135 	case LSM_SETID_FS:
1136 		/* juggle the capabilties to follow FSUID changes, unless
1137 		 * otherwise suppressed
1138 		 *
1139 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1140 		 *          if not, we might be a bit too harsh here.
1141 		 */
1142 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1143 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1144 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1145 				new->cap_effective =
1146 					cap_drop_fs_set(new->cap_effective);
1147 
1148 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1149 				new->cap_effective =
1150 					cap_raise_fs_set(new->cap_effective,
1151 							 new->cap_permitted);
1152 		}
1153 		break;
1154 
1155 	default:
1156 		return -EINVAL;
1157 	}
1158 
1159 	return 0;
1160 }
1161 
1162 /*
1163  * Rationale: code calling task_setscheduler, task_setioprio, and
1164  * task_setnice, assumes that
1165  *   . if capable(cap_sys_nice), then those actions should be allowed
1166  *   . if not capable(cap_sys_nice), but acting on your own processes,
1167  *   	then those actions should be allowed
1168  * This is insufficient now since you can call code without suid, but
1169  * yet with increased caps.
1170  * So we check for increased caps on the target process.
1171  */
1172 static int cap_safe_nice(struct task_struct *p)
1173 {
1174 	int is_subset, ret = 0;
1175 
1176 	rcu_read_lock();
1177 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1178 				 current_cred()->cap_permitted);
1179 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1180 		ret = -EPERM;
1181 	rcu_read_unlock();
1182 
1183 	return ret;
1184 }
1185 
1186 /**
1187  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1188  * @p: The task to affect
1189  *
1190  * Detemine if the requested scheduler policy change is permitted for the
1191  * specified task.
1192  *
1193  * Return: 0 if permission is granted, -ve if denied.
1194  */
1195 int cap_task_setscheduler(struct task_struct *p)
1196 {
1197 	return cap_safe_nice(p);
1198 }
1199 
1200 /**
1201  * cap_task_setioprio - Detemine if I/O priority change is permitted
1202  * @p: The task to affect
1203  * @ioprio: The I/O priority to set
1204  *
1205  * Detemine if the requested I/O priority change is permitted for the specified
1206  * task.
1207  *
1208  * Return: 0 if permission is granted, -ve if denied.
1209  */
1210 int cap_task_setioprio(struct task_struct *p, int ioprio)
1211 {
1212 	return cap_safe_nice(p);
1213 }
1214 
1215 /**
1216  * cap_task_setnice - Detemine if task priority change is permitted
1217  * @p: The task to affect
1218  * @nice: The nice value to set
1219  *
1220  * Detemine if the requested task priority change is permitted for the
1221  * specified task.
1222  *
1223  * Return: 0 if permission is granted, -ve if denied.
1224  */
1225 int cap_task_setnice(struct task_struct *p, int nice)
1226 {
1227 	return cap_safe_nice(p);
1228 }
1229 
1230 /*
1231  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1232  * the current task's bounding set.  Returns 0 on success, -ve on error.
1233  */
1234 static int cap_prctl_drop(unsigned long cap)
1235 {
1236 	struct cred *new;
1237 
1238 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1239 		return -EPERM;
1240 	if (!cap_valid(cap))
1241 		return -EINVAL;
1242 
1243 	new = prepare_creds();
1244 	if (!new)
1245 		return -ENOMEM;
1246 	cap_lower(new->cap_bset, cap);
1247 	return commit_creds(new);
1248 }
1249 
1250 /**
1251  * cap_task_prctl - Implement process control functions for this security module
1252  * @option: The process control function requested
1253  * @arg2: The argument data for this function
1254  * @arg3: The argument data for this function
1255  * @arg4: The argument data for this function
1256  * @arg5: The argument data for this function
1257  *
1258  * Allow process control functions (sys_prctl()) to alter capabilities; may
1259  * also deny access to other functions not otherwise implemented here.
1260  *
1261  * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1262  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1263  * modules will consider performing the function.
1264  */
1265 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1266 		   unsigned long arg4, unsigned long arg5)
1267 {
1268 	const struct cred *old = current_cred();
1269 	struct cred *new;
1270 
1271 	switch (option) {
1272 	case PR_CAPBSET_READ:
1273 		if (!cap_valid(arg2))
1274 			return -EINVAL;
1275 		return !!cap_raised(old->cap_bset, arg2);
1276 
1277 	case PR_CAPBSET_DROP:
1278 		return cap_prctl_drop(arg2);
1279 
1280 	/*
1281 	 * The next four prctl's remain to assist with transitioning a
1282 	 * system from legacy UID=0 based privilege (when filesystem
1283 	 * capabilities are not in use) to a system using filesystem
1284 	 * capabilities only - as the POSIX.1e draft intended.
1285 	 *
1286 	 * Note:
1287 	 *
1288 	 *  PR_SET_SECUREBITS =
1289 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1290 	 *    | issecure_mask(SECURE_NOROOT)
1291 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1292 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1293 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1294 	 *
1295 	 * will ensure that the current process and all of its
1296 	 * children will be locked into a pure
1297 	 * capability-based-privilege environment.
1298 	 */
1299 	case PR_SET_SECUREBITS:
1300 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1301 		     & (old->securebits ^ arg2))			/*[1]*/
1302 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1303 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1304 		    || (cap_capable(current_cred(),
1305 				    current_cred()->user_ns,
1306 				    CAP_SETPCAP,
1307 				    CAP_OPT_NONE) != 0)			/*[4]*/
1308 			/*
1309 			 * [1] no changing of bits that are locked
1310 			 * [2] no unlocking of locks
1311 			 * [3] no setting of unsupported bits
1312 			 * [4] doing anything requires privilege (go read about
1313 			 *     the "sendmail capabilities bug")
1314 			 */
1315 		    )
1316 			/* cannot change a locked bit */
1317 			return -EPERM;
1318 
1319 		new = prepare_creds();
1320 		if (!new)
1321 			return -ENOMEM;
1322 		new->securebits = arg2;
1323 		return commit_creds(new);
1324 
1325 	case PR_GET_SECUREBITS:
1326 		return old->securebits;
1327 
1328 	case PR_GET_KEEPCAPS:
1329 		return !!issecure(SECURE_KEEP_CAPS);
1330 
1331 	case PR_SET_KEEPCAPS:
1332 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1333 			return -EINVAL;
1334 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1335 			return -EPERM;
1336 
1337 		new = prepare_creds();
1338 		if (!new)
1339 			return -ENOMEM;
1340 		if (arg2)
1341 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1342 		else
1343 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1344 		return commit_creds(new);
1345 
1346 	case PR_CAP_AMBIENT:
1347 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1348 			if (arg3 | arg4 | arg5)
1349 				return -EINVAL;
1350 
1351 			new = prepare_creds();
1352 			if (!new)
1353 				return -ENOMEM;
1354 			cap_clear(new->cap_ambient);
1355 			return commit_creds(new);
1356 		}
1357 
1358 		if (((!cap_valid(arg3)) | arg4 | arg5))
1359 			return -EINVAL;
1360 
1361 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1362 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1363 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1364 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1365 			return -EINVAL;
1366 		} else {
1367 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1368 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1369 			     !cap_raised(current_cred()->cap_inheritable,
1370 					 arg3) ||
1371 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1372 				return -EPERM;
1373 
1374 			new = prepare_creds();
1375 			if (!new)
1376 				return -ENOMEM;
1377 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1378 				cap_raise(new->cap_ambient, arg3);
1379 			else
1380 				cap_lower(new->cap_ambient, arg3);
1381 			return commit_creds(new);
1382 		}
1383 
1384 	default:
1385 		/* No functionality available - continue with default */
1386 		return -ENOSYS;
1387 	}
1388 }
1389 
1390 /**
1391  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1392  * @mm: The VM space in which the new mapping is to be made
1393  * @pages: The size of the mapping
1394  *
1395  * Determine whether the allocation of a new virtual mapping by the current
1396  * task is permitted.
1397  *
1398  * Return: 1 if permission is granted, 0 if not.
1399  */
1400 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1401 {
1402 	int cap_sys_admin = 0;
1403 
1404 	if (cap_capable(current_cred(), &init_user_ns,
1405 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1406 		cap_sys_admin = 1;
1407 
1408 	return cap_sys_admin;
1409 }
1410 
1411 /**
1412  * cap_mmap_addr - check if able to map given addr
1413  * @addr: address attempting to be mapped
1414  *
1415  * If the process is attempting to map memory below dac_mmap_min_addr they need
1416  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1417  * capability security module.
1418  *
1419  * Return: 0 if this mapping should be allowed or -EPERM if not.
1420  */
1421 int cap_mmap_addr(unsigned long addr)
1422 {
1423 	int ret = 0;
1424 
1425 	if (addr < dac_mmap_min_addr) {
1426 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1427 				  CAP_OPT_NONE);
1428 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1429 		if (ret == 0)
1430 			current->flags |= PF_SUPERPRIV;
1431 	}
1432 	return ret;
1433 }
1434 
1435 int cap_mmap_file(struct file *file, unsigned long reqprot,
1436 		  unsigned long prot, unsigned long flags)
1437 {
1438 	return 0;
1439 }
1440 
1441 #ifdef CONFIG_SECURITY
1442 
1443 static struct security_hook_list capability_hooks[] __ro_after_init = {
1444 	LSM_HOOK_INIT(capable, cap_capable),
1445 	LSM_HOOK_INIT(settime, cap_settime),
1446 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1447 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1448 	LSM_HOOK_INIT(capget, cap_capget),
1449 	LSM_HOOK_INIT(capset, cap_capset),
1450 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1451 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1452 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1453 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1454 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1455 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1456 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1457 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1458 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1459 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1460 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1461 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1462 };
1463 
1464 static int __init capability_init(void)
1465 {
1466 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1467 				"capability");
1468 	return 0;
1469 }
1470 
1471 DEFINE_LSM(capability) = {
1472 	.name = "capability",
1473 	.order = LSM_ORDER_FIRST,
1474 	.init = capability_init,
1475 };
1476 
1477 #endif /* CONFIG_SECURITY */
1478