xref: /linux/security/commoncap.c (revision 4f77c3462308c62ffe7129cc18b9ac937f44b5a5)
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 #include <uapi/linux/lsm.h>
29 
30 /*
31  * If a non-root user executes a setuid-root binary in
32  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
33  * However if fE is also set, then the intent is for only
34  * the file capabilities to be applied, and the setuid-root
35  * bit is left on either to change the uid (plausible) or
36  * to get full privilege on a kernel without file capabilities
37  * support.  So in that case we do not raise capabilities.
38  *
39  * Warn if that happens, once per boot.
40  */
41 static void warn_setuid_and_fcaps_mixed(const char *fname)
42 {
43 	static int warned;
44 	if (!warned) {
45 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
46 			" effective capabilities. Therefore not raising all"
47 			" capabilities.\n", fname);
48 		warned = 1;
49 	}
50 }
51 
52 /**
53  * cap_capable - Determine whether a task has a particular effective capability
54  * @cred: The credentials to use
55  * @targ_ns:  The user namespace in which we need the capability
56  * @cap: The capability to check for
57  * @opts: Bitmask of options defined in include/linux/security.h
58  *
59  * Determine whether the nominated task has the specified capability amongst
60  * its effective set, returning 0 if it does, -ve if it does not.
61  *
62  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
63  * and has_capability() functions.  That is, it has the reverse semantics:
64  * cap_has_capability() returns 0 when a task has a capability, but the
65  * kernel's capable() and has_capability() returns 1 for this case.
66  */
67 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
68 		int cap, unsigned int opts)
69 {
70 	struct user_namespace *ns = targ_ns;
71 
72 	/* See if cred has the capability in the target user namespace
73 	 * by examining the target user namespace and all of the target
74 	 * user namespace's parents.
75 	 */
76 	for (;;) {
77 		/* Do we have the necessary capabilities? */
78 		if (ns == cred->user_ns)
79 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
80 
81 		/*
82 		 * If we're already at a lower level than we're looking for,
83 		 * we're done searching.
84 		 */
85 		if (ns->level <= cred->user_ns->level)
86 			return -EPERM;
87 
88 		/*
89 		 * The owner of the user namespace in the parent of the
90 		 * user namespace has all caps.
91 		 */
92 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
93 			return 0;
94 
95 		/*
96 		 * If you have a capability in a parent user ns, then you have
97 		 * it over all children user namespaces as well.
98 		 */
99 		ns = ns->parent;
100 	}
101 
102 	/* We never get here */
103 }
104 
105 /**
106  * cap_settime - Determine whether the current process may set the system clock
107  * @ts: The time to set
108  * @tz: The timezone to set
109  *
110  * Determine whether the current process may set the system clock and timezone
111  * information, returning 0 if permission granted, -ve if denied.
112  */
113 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
114 {
115 	if (!capable(CAP_SYS_TIME))
116 		return -EPERM;
117 	return 0;
118 }
119 
120 /**
121  * cap_ptrace_access_check - Determine whether the current process may access
122  *			   another
123  * @child: The process to be accessed
124  * @mode: The mode of attachment.
125  *
126  * If we are in the same or an ancestor user_ns and have all the target
127  * task's capabilities, then ptrace access is allowed.
128  * If we have the ptrace capability to the target user_ns, then ptrace
129  * access is allowed.
130  * Else denied.
131  *
132  * Determine whether a process may access another, returning 0 if permission
133  * granted, -ve if denied.
134  */
135 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
136 {
137 	int ret = 0;
138 	const struct cred *cred, *child_cred;
139 	const kernel_cap_t *caller_caps;
140 
141 	rcu_read_lock();
142 	cred = current_cred();
143 	child_cred = __task_cred(child);
144 	if (mode & PTRACE_MODE_FSCREDS)
145 		caller_caps = &cred->cap_effective;
146 	else
147 		caller_caps = &cred->cap_permitted;
148 	if (cred->user_ns == child_cred->user_ns &&
149 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
150 		goto out;
151 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
152 		goto out;
153 	ret = -EPERM;
154 out:
155 	rcu_read_unlock();
156 	return ret;
157 }
158 
159 /**
160  * cap_ptrace_traceme - Determine whether another process may trace the current
161  * @parent: The task proposed to be the tracer
162  *
163  * If parent is in the same or an ancestor user_ns and has all current's
164  * capabilities, then ptrace access is allowed.
165  * If parent has the ptrace capability to current's user_ns, then ptrace
166  * access is allowed.
167  * Else denied.
168  *
169  * Determine whether the nominated task is permitted to trace the current
170  * process, returning 0 if permission is granted, -ve if denied.
171  */
172 int cap_ptrace_traceme(struct task_struct *parent)
173 {
174 	int ret = 0;
175 	const struct cred *cred, *child_cred;
176 
177 	rcu_read_lock();
178 	cred = __task_cred(parent);
179 	child_cred = current_cred();
180 	if (cred->user_ns == child_cred->user_ns &&
181 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
182 		goto out;
183 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
184 		goto out;
185 	ret = -EPERM;
186 out:
187 	rcu_read_unlock();
188 	return ret;
189 }
190 
191 /**
192  * cap_capget - Retrieve a task's capability sets
193  * @target: The task from which to retrieve the capability sets
194  * @effective: The place to record the effective set
195  * @inheritable: The place to record the inheritable set
196  * @permitted: The place to record the permitted set
197  *
198  * This function retrieves the capabilities of the nominated task and returns
199  * them to the caller.
200  */
201 int cap_capget(const struct task_struct *target, kernel_cap_t *effective,
202 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
203 {
204 	const struct cred *cred;
205 
206 	/* Derived from kernel/capability.c:sys_capget. */
207 	rcu_read_lock();
208 	cred = __task_cred(target);
209 	*effective   = cred->cap_effective;
210 	*inheritable = cred->cap_inheritable;
211 	*permitted   = cred->cap_permitted;
212 	rcu_read_unlock();
213 	return 0;
214 }
215 
216 /*
217  * Determine whether the inheritable capabilities are limited to the old
218  * permitted set.  Returns 1 if they are limited, 0 if they are not.
219  */
220 static inline int cap_inh_is_capped(void)
221 {
222 	/* they are so limited unless the current task has the CAP_SETPCAP
223 	 * capability
224 	 */
225 	if (cap_capable(current_cred(), current_cred()->user_ns,
226 			CAP_SETPCAP, CAP_OPT_NONE) == 0)
227 		return 0;
228 	return 1;
229 }
230 
231 /**
232  * cap_capset - Validate and apply proposed changes to current's capabilities
233  * @new: The proposed new credentials; alterations should be made here
234  * @old: The current task's current credentials
235  * @effective: A pointer to the proposed new effective capabilities set
236  * @inheritable: A pointer to the proposed new inheritable capabilities set
237  * @permitted: A pointer to the proposed new permitted capabilities set
238  *
239  * This function validates and applies a proposed mass change to the current
240  * process's capability sets.  The changes are made to the proposed new
241  * credentials, and assuming no error, will be committed by the caller of LSM.
242  */
243 int cap_capset(struct cred *new,
244 	       const struct cred *old,
245 	       const kernel_cap_t *effective,
246 	       const kernel_cap_t *inheritable,
247 	       const kernel_cap_t *permitted)
248 {
249 	if (cap_inh_is_capped() &&
250 	    !cap_issubset(*inheritable,
251 			  cap_combine(old->cap_inheritable,
252 				      old->cap_permitted)))
253 		/* incapable of using this inheritable set */
254 		return -EPERM;
255 
256 	if (!cap_issubset(*inheritable,
257 			  cap_combine(old->cap_inheritable,
258 				      old->cap_bset)))
259 		/* no new pI capabilities outside bounding set */
260 		return -EPERM;
261 
262 	/* verify restrictions on target's new Permitted set */
263 	if (!cap_issubset(*permitted, old->cap_permitted))
264 		return -EPERM;
265 
266 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
267 	if (!cap_issubset(*effective, *permitted))
268 		return -EPERM;
269 
270 	new->cap_effective   = *effective;
271 	new->cap_inheritable = *inheritable;
272 	new->cap_permitted   = *permitted;
273 
274 	/*
275 	 * Mask off ambient bits that are no longer both permitted and
276 	 * inheritable.
277 	 */
278 	new->cap_ambient = cap_intersect(new->cap_ambient,
279 					 cap_intersect(*permitted,
280 						       *inheritable));
281 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
282 		return -EINVAL;
283 	return 0;
284 }
285 
286 /**
287  * cap_inode_need_killpriv - Determine if inode change affects privileges
288  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
289  *
290  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
291  * affects the security markings on that inode, and if it is, should
292  * inode_killpriv() be invoked or the change rejected.
293  *
294  * Return: 1 if security.capability has a value, meaning inode_killpriv()
295  * is required, 0 otherwise, meaning inode_killpriv() is not required.
296  */
297 int cap_inode_need_killpriv(struct dentry *dentry)
298 {
299 	struct inode *inode = d_backing_inode(dentry);
300 	int error;
301 
302 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
303 	return error > 0;
304 }
305 
306 /**
307  * cap_inode_killpriv - Erase the security markings on an inode
308  *
309  * @idmap:	idmap of the mount the inode was found from
310  * @dentry:	The inode/dentry to alter
311  *
312  * Erase the privilege-enhancing security markings on an inode.
313  *
314  * If the inode has been found through an idmapped mount the idmap of
315  * the vfsmount must be passed through @idmap. This function will then
316  * take care to map the inode according to @idmap before checking
317  * permissions. On non-idmapped mounts or if permission checking is to be
318  * performed on the raw inode simply pass @nop_mnt_idmap.
319  *
320  * Return: 0 if successful, -ve on error.
321  */
322 int cap_inode_killpriv(struct mnt_idmap *idmap, struct dentry *dentry)
323 {
324 	int error;
325 
326 	error = __vfs_removexattr(idmap, dentry, XATTR_NAME_CAPS);
327 	if (error == -EOPNOTSUPP)
328 		error = 0;
329 	return error;
330 }
331 
332 static bool rootid_owns_currentns(vfsuid_t rootvfsuid)
333 {
334 	struct user_namespace *ns;
335 	kuid_t kroot;
336 
337 	if (!vfsuid_valid(rootvfsuid))
338 		return false;
339 
340 	kroot = vfsuid_into_kuid(rootvfsuid);
341 	for (ns = current_user_ns();; ns = ns->parent) {
342 		if (from_kuid(ns, kroot) == 0)
343 			return true;
344 		if (ns == &init_user_ns)
345 			break;
346 	}
347 
348 	return false;
349 }
350 
351 static __u32 sansflags(__u32 m)
352 {
353 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
354 }
355 
356 static bool is_v2header(int size, const struct vfs_cap_data *cap)
357 {
358 	if (size != XATTR_CAPS_SZ_2)
359 		return false;
360 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
361 }
362 
363 static bool is_v3header(int size, const struct vfs_cap_data *cap)
364 {
365 	if (size != XATTR_CAPS_SZ_3)
366 		return false;
367 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
368 }
369 
370 /*
371  * getsecurity: We are called for security.* before any attempt to read the
372  * xattr from the inode itself.
373  *
374  * This gives us a chance to read the on-disk value and convert it.  If we
375  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
376  *
377  * Note we are not called by vfs_getxattr_alloc(), but that is only called
378  * by the integrity subsystem, which really wants the unconverted values -
379  * so that's good.
380  */
381 int cap_inode_getsecurity(struct mnt_idmap *idmap,
382 			  struct inode *inode, const char *name, void **buffer,
383 			  bool alloc)
384 {
385 	int size;
386 	kuid_t kroot;
387 	vfsuid_t vfsroot;
388 	u32 nsmagic, magic;
389 	uid_t root, mappedroot;
390 	char *tmpbuf = NULL;
391 	struct vfs_cap_data *cap;
392 	struct vfs_ns_cap_data *nscap = NULL;
393 	struct dentry *dentry;
394 	struct user_namespace *fs_ns;
395 
396 	if (strcmp(name, "capability") != 0)
397 		return -EOPNOTSUPP;
398 
399 	dentry = d_find_any_alias(inode);
400 	if (!dentry)
401 		return -EINVAL;
402 	size = vfs_getxattr_alloc(idmap, dentry, XATTR_NAME_CAPS, &tmpbuf,
403 				  sizeof(struct vfs_ns_cap_data), GFP_NOFS);
404 	dput(dentry);
405 	/* gcc11 complains if we don't check for !tmpbuf */
406 	if (size < 0 || !tmpbuf)
407 		goto out_free;
408 
409 	fs_ns = inode->i_sb->s_user_ns;
410 	cap = (struct vfs_cap_data *) tmpbuf;
411 	if (is_v2header(size, cap)) {
412 		root = 0;
413 	} else if (is_v3header(size, cap)) {
414 		nscap = (struct vfs_ns_cap_data *) tmpbuf;
415 		root = le32_to_cpu(nscap->rootid);
416 	} else {
417 		size = -EINVAL;
418 		goto out_free;
419 	}
420 
421 	kroot = make_kuid(fs_ns, root);
422 
423 	/* If this is an idmapped mount shift the kuid. */
424 	vfsroot = make_vfsuid(idmap, fs_ns, kroot);
425 
426 	/* If the root kuid maps to a valid uid in current ns, then return
427 	 * this as a nscap. */
428 	mappedroot = from_kuid(current_user_ns(), vfsuid_into_kuid(vfsroot));
429 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
430 		size = sizeof(struct vfs_ns_cap_data);
431 		if (alloc) {
432 			if (!nscap) {
433 				/* v2 -> v3 conversion */
434 				nscap = kzalloc(size, GFP_ATOMIC);
435 				if (!nscap) {
436 					size = -ENOMEM;
437 					goto out_free;
438 				}
439 				nsmagic = VFS_CAP_REVISION_3;
440 				magic = le32_to_cpu(cap->magic_etc);
441 				if (magic & VFS_CAP_FLAGS_EFFECTIVE)
442 					nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
443 				memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
444 				nscap->magic_etc = cpu_to_le32(nsmagic);
445 			} else {
446 				/* use allocated v3 buffer */
447 				tmpbuf = NULL;
448 			}
449 			nscap->rootid = cpu_to_le32(mappedroot);
450 			*buffer = nscap;
451 		}
452 		goto out_free;
453 	}
454 
455 	if (!rootid_owns_currentns(vfsroot)) {
456 		size = -EOVERFLOW;
457 		goto out_free;
458 	}
459 
460 	/* This comes from a parent namespace.  Return as a v2 capability */
461 	size = sizeof(struct vfs_cap_data);
462 	if (alloc) {
463 		if (nscap) {
464 			/* v3 -> v2 conversion */
465 			cap = kzalloc(size, GFP_ATOMIC);
466 			if (!cap) {
467 				size = -ENOMEM;
468 				goto out_free;
469 			}
470 			magic = VFS_CAP_REVISION_2;
471 			nsmagic = le32_to_cpu(nscap->magic_etc);
472 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
473 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
474 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
475 			cap->magic_etc = cpu_to_le32(magic);
476 		} else {
477 			/* use unconverted v2 */
478 			tmpbuf = NULL;
479 		}
480 		*buffer = cap;
481 	}
482 out_free:
483 	kfree(tmpbuf);
484 	return size;
485 }
486 
487 /**
488  * rootid_from_xattr - translate root uid of vfs caps
489  *
490  * @value:	vfs caps value which may be modified by this function
491  * @size:	size of @ivalue
492  * @task_ns:	user namespace of the caller
493  */
494 static vfsuid_t rootid_from_xattr(const void *value, size_t size,
495 				  struct user_namespace *task_ns)
496 {
497 	const struct vfs_ns_cap_data *nscap = value;
498 	uid_t rootid = 0;
499 
500 	if (size == XATTR_CAPS_SZ_3)
501 		rootid = le32_to_cpu(nscap->rootid);
502 
503 	return VFSUIDT_INIT(make_kuid(task_ns, rootid));
504 }
505 
506 static bool validheader(size_t size, const struct vfs_cap_data *cap)
507 {
508 	return is_v2header(size, cap) || is_v3header(size, cap);
509 }
510 
511 /**
512  * cap_convert_nscap - check vfs caps
513  *
514  * @idmap:	idmap of the mount the inode was found from
515  * @dentry:	used to retrieve inode to check permissions on
516  * @ivalue:	vfs caps value which may be modified by this function
517  * @size:	size of @ivalue
518  *
519  * User requested a write of security.capability.  If needed, update the
520  * xattr to change from v2 to v3, or to fixup the v3 rootid.
521  *
522  * If the inode has been found through an idmapped mount the idmap of
523  * the vfsmount must be passed through @idmap. This function will then
524  * take care to map the inode according to @idmap before checking
525  * permissions. On non-idmapped mounts or if permission checking is to be
526  * performed on the raw inode simply pass @nop_mnt_idmap.
527  *
528  * Return: On success, return the new size; on error, return < 0.
529  */
530 int cap_convert_nscap(struct mnt_idmap *idmap, struct dentry *dentry,
531 		      const void **ivalue, size_t size)
532 {
533 	struct vfs_ns_cap_data *nscap;
534 	uid_t nsrootid;
535 	const struct vfs_cap_data *cap = *ivalue;
536 	__u32 magic, nsmagic;
537 	struct inode *inode = d_backing_inode(dentry);
538 	struct user_namespace *task_ns = current_user_ns(),
539 		*fs_ns = inode->i_sb->s_user_ns;
540 	kuid_t rootid;
541 	vfsuid_t vfsrootid;
542 	size_t newsize;
543 
544 	if (!*ivalue)
545 		return -EINVAL;
546 	if (!validheader(size, cap))
547 		return -EINVAL;
548 	if (!capable_wrt_inode_uidgid(idmap, inode, CAP_SETFCAP))
549 		return -EPERM;
550 	if (size == XATTR_CAPS_SZ_2 && (idmap == &nop_mnt_idmap))
551 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
552 			/* user is privileged, just write the v2 */
553 			return size;
554 
555 	vfsrootid = rootid_from_xattr(*ivalue, size, task_ns);
556 	if (!vfsuid_valid(vfsrootid))
557 		return -EINVAL;
558 
559 	rootid = from_vfsuid(idmap, fs_ns, vfsrootid);
560 	if (!uid_valid(rootid))
561 		return -EINVAL;
562 
563 	nsrootid = from_kuid(fs_ns, rootid);
564 	if (nsrootid == -1)
565 		return -EINVAL;
566 
567 	newsize = sizeof(struct vfs_ns_cap_data);
568 	nscap = kmalloc(newsize, GFP_ATOMIC);
569 	if (!nscap)
570 		return -ENOMEM;
571 	nscap->rootid = cpu_to_le32(nsrootid);
572 	nsmagic = VFS_CAP_REVISION_3;
573 	magic = le32_to_cpu(cap->magic_etc);
574 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
575 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
576 	nscap->magic_etc = cpu_to_le32(nsmagic);
577 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
578 
579 	*ivalue = nscap;
580 	return newsize;
581 }
582 
583 /*
584  * Calculate the new process capability sets from the capability sets attached
585  * to a file.
586  */
587 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
588 					  struct linux_binprm *bprm,
589 					  bool *effective,
590 					  bool *has_fcap)
591 {
592 	struct cred *new = bprm->cred;
593 	int ret = 0;
594 
595 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
596 		*effective = true;
597 
598 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
599 		*has_fcap = true;
600 
601 	/*
602 	 * pP' = (X & fP) | (pI & fI)
603 	 * The addition of pA' is handled later.
604 	 */
605 	new->cap_permitted.val =
606 		(new->cap_bset.val & caps->permitted.val) |
607 		(new->cap_inheritable.val & caps->inheritable.val);
608 
609 	if (caps->permitted.val & ~new->cap_permitted.val)
610 		/* insufficient to execute correctly */
611 		ret = -EPERM;
612 
613 	/*
614 	 * For legacy apps, with no internal support for recognizing they
615 	 * do not have enough capabilities, we return an error if they are
616 	 * missing some "forced" (aka file-permitted) capabilities.
617 	 */
618 	return *effective ? ret : 0;
619 }
620 
621 /**
622  * get_vfs_caps_from_disk - retrieve vfs caps from disk
623  *
624  * @idmap:	idmap of the mount the inode was found from
625  * @dentry:	dentry from which @inode is retrieved
626  * @cpu_caps:	vfs capabilities
627  *
628  * Extract the on-exec-apply capability sets for an executable file.
629  *
630  * If the inode has been found through an idmapped mount the idmap of
631  * the vfsmount must be passed through @idmap. This function will then
632  * take care to map the inode according to @idmap before checking
633  * permissions. On non-idmapped mounts or if permission checking is to be
634  * performed on the raw inode simply pass @nop_mnt_idmap.
635  */
636 int get_vfs_caps_from_disk(struct mnt_idmap *idmap,
637 			   const struct dentry *dentry,
638 			   struct cpu_vfs_cap_data *cpu_caps)
639 {
640 	struct inode *inode = d_backing_inode(dentry);
641 	__u32 magic_etc;
642 	int size;
643 	struct vfs_ns_cap_data data, *nscaps = &data;
644 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
645 	kuid_t rootkuid;
646 	vfsuid_t rootvfsuid;
647 	struct user_namespace *fs_ns;
648 
649 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
650 
651 	if (!inode)
652 		return -ENODATA;
653 
654 	fs_ns = inode->i_sb->s_user_ns;
655 	size = __vfs_getxattr((struct dentry *)dentry, inode,
656 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
657 	if (size == -ENODATA || size == -EOPNOTSUPP)
658 		/* no data, that's ok */
659 		return -ENODATA;
660 
661 	if (size < 0)
662 		return size;
663 
664 	if (size < sizeof(magic_etc))
665 		return -EINVAL;
666 
667 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
668 
669 	rootkuid = make_kuid(fs_ns, 0);
670 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
671 	case VFS_CAP_REVISION_1:
672 		if (size != XATTR_CAPS_SZ_1)
673 			return -EINVAL;
674 		break;
675 	case VFS_CAP_REVISION_2:
676 		if (size != XATTR_CAPS_SZ_2)
677 			return -EINVAL;
678 		break;
679 	case VFS_CAP_REVISION_3:
680 		if (size != XATTR_CAPS_SZ_3)
681 			return -EINVAL;
682 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
683 		break;
684 
685 	default:
686 		return -EINVAL;
687 	}
688 
689 	rootvfsuid = make_vfsuid(idmap, fs_ns, rootkuid);
690 	if (!vfsuid_valid(rootvfsuid))
691 		return -ENODATA;
692 
693 	/* Limit the caps to the mounter of the filesystem
694 	 * or the more limited uid specified in the xattr.
695 	 */
696 	if (!rootid_owns_currentns(rootvfsuid))
697 		return -ENODATA;
698 
699 	cpu_caps->permitted.val = le32_to_cpu(caps->data[0].permitted);
700 	cpu_caps->inheritable.val = le32_to_cpu(caps->data[0].inheritable);
701 
702 	/*
703 	 * Rev1 had just a single 32-bit word, later expanded
704 	 * to a second one for the high bits
705 	 */
706 	if ((magic_etc & VFS_CAP_REVISION_MASK) != VFS_CAP_REVISION_1) {
707 		cpu_caps->permitted.val += (u64)le32_to_cpu(caps->data[1].permitted) << 32;
708 		cpu_caps->inheritable.val += (u64)le32_to_cpu(caps->data[1].inheritable) << 32;
709 	}
710 
711 	cpu_caps->permitted.val &= CAP_VALID_MASK;
712 	cpu_caps->inheritable.val &= CAP_VALID_MASK;
713 
714 	cpu_caps->rootid = vfsuid_into_kuid(rootvfsuid);
715 
716 	return 0;
717 }
718 
719 /*
720  * Attempt to get the on-exec apply capability sets for an executable file from
721  * its xattrs and, if present, apply them to the proposed credentials being
722  * constructed by execve().
723  */
724 static int get_file_caps(struct linux_binprm *bprm, const struct file *file,
725 			 bool *effective, bool *has_fcap)
726 {
727 	int rc = 0;
728 	struct cpu_vfs_cap_data vcaps;
729 
730 	cap_clear(bprm->cred->cap_permitted);
731 
732 	if (!file_caps_enabled)
733 		return 0;
734 
735 	if (!mnt_may_suid(file->f_path.mnt))
736 		return 0;
737 
738 	/*
739 	 * This check is redundant with mnt_may_suid() but is kept to make
740 	 * explicit that capability bits are limited to s_user_ns and its
741 	 * descendants.
742 	 */
743 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
744 		return 0;
745 
746 	rc = get_vfs_caps_from_disk(file_mnt_idmap(file),
747 				    file->f_path.dentry, &vcaps);
748 	if (rc < 0) {
749 		if (rc == -EINVAL)
750 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
751 					bprm->filename);
752 		else if (rc == -ENODATA)
753 			rc = 0;
754 		goto out;
755 	}
756 
757 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
758 
759 out:
760 	if (rc)
761 		cap_clear(bprm->cred->cap_permitted);
762 
763 	return rc;
764 }
765 
766 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
767 
768 static inline bool __is_real(kuid_t uid, struct cred *cred)
769 { return uid_eq(cred->uid, uid); }
770 
771 static inline bool __is_eff(kuid_t uid, struct cred *cred)
772 { return uid_eq(cred->euid, uid); }
773 
774 static inline bool __is_suid(kuid_t uid, struct cred *cred)
775 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
776 
777 /*
778  * handle_privileged_root - Handle case of privileged root
779  * @bprm: The execution parameters, including the proposed creds
780  * @has_fcap: Are any file capabilities set?
781  * @effective: Do we have effective root privilege?
782  * @root_uid: This namespace' root UID WRT initial USER namespace
783  *
784  * Handle the case where root is privileged and hasn't been neutered by
785  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
786  * set UID root and nothing is changed.  If we are root, cap_permitted is
787  * updated.  If we have become set UID root, the effective bit is set.
788  */
789 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
790 				   bool *effective, kuid_t root_uid)
791 {
792 	const struct cred *old = current_cred();
793 	struct cred *new = bprm->cred;
794 
795 	if (!root_privileged())
796 		return;
797 	/*
798 	 * If the legacy file capability is set, then don't set privs
799 	 * for a setuid root binary run by a non-root user.  Do set it
800 	 * for a root user just to cause least surprise to an admin.
801 	 */
802 	if (has_fcap && __is_suid(root_uid, new)) {
803 		warn_setuid_and_fcaps_mixed(bprm->filename);
804 		return;
805 	}
806 	/*
807 	 * To support inheritance of root-permissions and suid-root
808 	 * executables under compatibility mode, we override the
809 	 * capability sets for the file.
810 	 */
811 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
812 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
813 		new->cap_permitted = cap_combine(old->cap_bset,
814 						 old->cap_inheritable);
815 	}
816 	/*
817 	 * If only the real uid is 0, we do not set the effective bit.
818 	 */
819 	if (__is_eff(root_uid, new))
820 		*effective = true;
821 }
822 
823 #define __cap_gained(field, target, source) \
824 	!cap_issubset(target->cap_##field, source->cap_##field)
825 #define __cap_grew(target, source, cred) \
826 	!cap_issubset(cred->cap_##target, cred->cap_##source)
827 #define __cap_full(field, cred) \
828 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
829 
830 static inline bool __is_setuid(struct cred *new, const struct cred *old)
831 { return !uid_eq(new->euid, old->uid); }
832 
833 static inline bool __is_setgid(struct cred *new, const struct cred *old)
834 { return !gid_eq(new->egid, old->gid); }
835 
836 /*
837  * 1) Audit candidate if current->cap_effective is set
838  *
839  * We do not bother to audit if 3 things are true:
840  *   1) cap_effective has all caps
841  *   2) we became root *OR* are were already root
842  *   3) root is supposed to have all caps (SECURE_NOROOT)
843  * Since this is just a normal root execing a process.
844  *
845  * Number 1 above might fail if you don't have a full bset, but I think
846  * that is interesting information to audit.
847  *
848  * A number of other conditions require logging:
849  * 2) something prevented setuid root getting all caps
850  * 3) non-setuid root gets fcaps
851  * 4) non-setuid root gets ambient
852  */
853 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
854 				     kuid_t root, bool has_fcap)
855 {
856 	bool ret = false;
857 
858 	if ((__cap_grew(effective, ambient, new) &&
859 	     !(__cap_full(effective, new) &&
860 	       (__is_eff(root, new) || __is_real(root, new)) &&
861 	       root_privileged())) ||
862 	    (root_privileged() &&
863 	     __is_suid(root, new) &&
864 	     !__cap_full(effective, new)) ||
865 	    (!__is_setuid(new, old) &&
866 	     ((has_fcap &&
867 	       __cap_gained(permitted, new, old)) ||
868 	      __cap_gained(ambient, new, old))))
869 
870 		ret = true;
871 
872 	return ret;
873 }
874 
875 /**
876  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
877  * @bprm: The execution parameters, including the proposed creds
878  * @file: The file to pull the credentials from
879  *
880  * Set up the proposed credentials for a new execution context being
881  * constructed by execve().  The proposed creds in @bprm->cred is altered,
882  * which won't take effect immediately.
883  *
884  * Return: 0 if successful, -ve on error.
885  */
886 int cap_bprm_creds_from_file(struct linux_binprm *bprm, const struct file *file)
887 {
888 	/* Process setpcap binaries and capabilities for uid 0 */
889 	const struct cred *old = current_cred();
890 	struct cred *new = bprm->cred;
891 	bool effective = false, has_fcap = false, is_setid;
892 	int ret;
893 	kuid_t root_uid;
894 
895 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
896 		return -EPERM;
897 
898 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
899 	if (ret < 0)
900 		return ret;
901 
902 	root_uid = make_kuid(new->user_ns, 0);
903 
904 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
905 
906 	/* if we have fs caps, clear dangerous personality flags */
907 	if (__cap_gained(permitted, new, old))
908 		bprm->per_clear |= PER_CLEAR_ON_SETID;
909 
910 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
911 	 * credentials unless they have the appropriate permit.
912 	 *
913 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
914 	 */
915 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
916 
917 	if ((is_setid || __cap_gained(permitted, new, old)) &&
918 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
919 	     !ptracer_capable(current, new->user_ns))) {
920 		/* downgrade; they get no more than they had, and maybe less */
921 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
922 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
923 			new->euid = new->uid;
924 			new->egid = new->gid;
925 		}
926 		new->cap_permitted = cap_intersect(new->cap_permitted,
927 						   old->cap_permitted);
928 	}
929 
930 	new->suid = new->fsuid = new->euid;
931 	new->sgid = new->fsgid = new->egid;
932 
933 	/* File caps or setid cancels ambient. */
934 	if (has_fcap || is_setid)
935 		cap_clear(new->cap_ambient);
936 
937 	/*
938 	 * Now that we've computed pA', update pP' to give:
939 	 *   pP' = (X & fP) | (pI & fI) | pA'
940 	 */
941 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
942 
943 	/*
944 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
945 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
946 	 */
947 	if (effective)
948 		new->cap_effective = new->cap_permitted;
949 	else
950 		new->cap_effective = new->cap_ambient;
951 
952 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
953 		return -EPERM;
954 
955 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
956 		ret = audit_log_bprm_fcaps(bprm, new, old);
957 		if (ret < 0)
958 			return ret;
959 	}
960 
961 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
962 
963 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
964 		return -EPERM;
965 
966 	/* Check for privilege-elevated exec. */
967 	if (is_setid ||
968 	    (!__is_real(root_uid, new) &&
969 	     (effective ||
970 	      __cap_grew(permitted, ambient, new))))
971 		bprm->secureexec = 1;
972 
973 	return 0;
974 }
975 
976 /**
977  * cap_inode_setxattr - Determine whether an xattr may be altered
978  * @dentry: The inode/dentry being altered
979  * @name: The name of the xattr to be changed
980  * @value: The value that the xattr will be changed to
981  * @size: The size of value
982  * @flags: The replacement flag
983  *
984  * Determine whether an xattr may be altered or set on an inode, returning 0 if
985  * permission is granted, -ve if denied.
986  *
987  * This is used to make sure security xattrs don't get updated or set by those
988  * who aren't privileged to do so.
989  */
990 int cap_inode_setxattr(struct dentry *dentry, const char *name,
991 		       const void *value, size_t size, int flags)
992 {
993 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
994 
995 	/* Ignore non-security xattrs */
996 	if (strncmp(name, XATTR_SECURITY_PREFIX,
997 			XATTR_SECURITY_PREFIX_LEN) != 0)
998 		return 0;
999 
1000 	/*
1001 	 * For XATTR_NAME_CAPS the check will be done in
1002 	 * cap_convert_nscap(), called by setxattr()
1003 	 */
1004 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
1005 		return 0;
1006 
1007 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1008 		return -EPERM;
1009 	return 0;
1010 }
1011 
1012 /**
1013  * cap_inode_removexattr - Determine whether an xattr may be removed
1014  *
1015  * @idmap:	idmap of the mount the inode was found from
1016  * @dentry:	The inode/dentry being altered
1017  * @name:	The name of the xattr to be changed
1018  *
1019  * Determine whether an xattr may be removed from an inode, returning 0 if
1020  * permission is granted, -ve if denied.
1021  *
1022  * If the inode has been found through an idmapped mount the idmap of
1023  * the vfsmount must be passed through @idmap. This function will then
1024  * take care to map the inode according to @idmap before checking
1025  * permissions. On non-idmapped mounts or if permission checking is to be
1026  * performed on the raw inode simply pass @nop_mnt_idmap.
1027  *
1028  * This is used to make sure security xattrs don't get removed by those who
1029  * aren't privileged to remove them.
1030  */
1031 int cap_inode_removexattr(struct mnt_idmap *idmap,
1032 			  struct dentry *dentry, const char *name)
1033 {
1034 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1035 
1036 	/* Ignore non-security xattrs */
1037 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1038 			XATTR_SECURITY_PREFIX_LEN) != 0)
1039 		return 0;
1040 
1041 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1042 		/* security.capability gets namespaced */
1043 		struct inode *inode = d_backing_inode(dentry);
1044 		if (!inode)
1045 			return -EINVAL;
1046 		if (!capable_wrt_inode_uidgid(idmap, inode, CAP_SETFCAP))
1047 			return -EPERM;
1048 		return 0;
1049 	}
1050 
1051 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1052 		return -EPERM;
1053 	return 0;
1054 }
1055 
1056 /*
1057  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1058  * a process after a call to setuid, setreuid, or setresuid.
1059  *
1060  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1061  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1062  *  cleared.
1063  *
1064  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1065  *  capabilities of the process are cleared.
1066  *
1067  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1068  *  capabilities are set to the permitted capabilities.
1069  *
1070  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1071  *  never happen.
1072  *
1073  *  -astor
1074  *
1075  * cevans - New behaviour, Oct '99
1076  * A process may, via prctl(), elect to keep its capabilities when it
1077  * calls setuid() and switches away from uid==0. Both permitted and
1078  * effective sets will be retained.
1079  * Without this change, it was impossible for a daemon to drop only some
1080  * of its privilege. The call to setuid(!=0) would drop all privileges!
1081  * Keeping uid 0 is not an option because uid 0 owns too many vital
1082  * files..
1083  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1084  */
1085 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1086 {
1087 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1088 
1089 	if ((uid_eq(old->uid, root_uid) ||
1090 	     uid_eq(old->euid, root_uid) ||
1091 	     uid_eq(old->suid, root_uid)) &&
1092 	    (!uid_eq(new->uid, root_uid) &&
1093 	     !uid_eq(new->euid, root_uid) &&
1094 	     !uid_eq(new->suid, root_uid))) {
1095 		if (!issecure(SECURE_KEEP_CAPS)) {
1096 			cap_clear(new->cap_permitted);
1097 			cap_clear(new->cap_effective);
1098 		}
1099 
1100 		/*
1101 		 * Pre-ambient programs expect setresuid to nonroot followed
1102 		 * by exec to drop capabilities.  We should make sure that
1103 		 * this remains the case.
1104 		 */
1105 		cap_clear(new->cap_ambient);
1106 	}
1107 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1108 		cap_clear(new->cap_effective);
1109 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1110 		new->cap_effective = new->cap_permitted;
1111 }
1112 
1113 /**
1114  * cap_task_fix_setuid - Fix up the results of setuid() call
1115  * @new: The proposed credentials
1116  * @old: The current task's current credentials
1117  * @flags: Indications of what has changed
1118  *
1119  * Fix up the results of setuid() call before the credential changes are
1120  * actually applied.
1121  *
1122  * Return: 0 to grant the changes, -ve to deny them.
1123  */
1124 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1125 {
1126 	switch (flags) {
1127 	case LSM_SETID_RE:
1128 	case LSM_SETID_ID:
1129 	case LSM_SETID_RES:
1130 		/* juggle the capabilities to follow [RES]UID changes unless
1131 		 * otherwise suppressed */
1132 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1133 			cap_emulate_setxuid(new, old);
1134 		break;
1135 
1136 	case LSM_SETID_FS:
1137 		/* juggle the capabilities to follow FSUID changes, unless
1138 		 * otherwise suppressed
1139 		 *
1140 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1141 		 *          if not, we might be a bit too harsh here.
1142 		 */
1143 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1144 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1145 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1146 				new->cap_effective =
1147 					cap_drop_fs_set(new->cap_effective);
1148 
1149 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1150 				new->cap_effective =
1151 					cap_raise_fs_set(new->cap_effective,
1152 							 new->cap_permitted);
1153 		}
1154 		break;
1155 
1156 	default:
1157 		return -EINVAL;
1158 	}
1159 
1160 	return 0;
1161 }
1162 
1163 /*
1164  * Rationale: code calling task_setscheduler, task_setioprio, and
1165  * task_setnice, assumes that
1166  *   . if capable(cap_sys_nice), then those actions should be allowed
1167  *   . if not capable(cap_sys_nice), but acting on your own processes,
1168  *   	then those actions should be allowed
1169  * This is insufficient now since you can call code without suid, but
1170  * yet with increased caps.
1171  * So we check for increased caps on the target process.
1172  */
1173 static int cap_safe_nice(struct task_struct *p)
1174 {
1175 	int is_subset, ret = 0;
1176 
1177 	rcu_read_lock();
1178 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1179 				 current_cred()->cap_permitted);
1180 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1181 		ret = -EPERM;
1182 	rcu_read_unlock();
1183 
1184 	return ret;
1185 }
1186 
1187 /**
1188  * cap_task_setscheduler - Determine if scheduler policy change is permitted
1189  * @p: The task to affect
1190  *
1191  * Determine if the requested scheduler policy change is permitted for the
1192  * specified task.
1193  *
1194  * Return: 0 if permission is granted, -ve if denied.
1195  */
1196 int cap_task_setscheduler(struct task_struct *p)
1197 {
1198 	return cap_safe_nice(p);
1199 }
1200 
1201 /**
1202  * cap_task_setioprio - Determine if I/O priority change is permitted
1203  * @p: The task to affect
1204  * @ioprio: The I/O priority to set
1205  *
1206  * Determine if the requested I/O priority change is permitted for the specified
1207  * task.
1208  *
1209  * Return: 0 if permission is granted, -ve if denied.
1210  */
1211 int cap_task_setioprio(struct task_struct *p, int ioprio)
1212 {
1213 	return cap_safe_nice(p);
1214 }
1215 
1216 /**
1217  * cap_task_setnice - Determine if task priority change is permitted
1218  * @p: The task to affect
1219  * @nice: The nice value to set
1220  *
1221  * Determine if the requested task priority change is permitted for the
1222  * specified task.
1223  *
1224  * Return: 0 if permission is granted, -ve if denied.
1225  */
1226 int cap_task_setnice(struct task_struct *p, int nice)
1227 {
1228 	return cap_safe_nice(p);
1229 }
1230 
1231 /*
1232  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1233  * the current task's bounding set.  Returns 0 on success, -ve on error.
1234  */
1235 static int cap_prctl_drop(unsigned long cap)
1236 {
1237 	struct cred *new;
1238 
1239 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1240 		return -EPERM;
1241 	if (!cap_valid(cap))
1242 		return -EINVAL;
1243 
1244 	new = prepare_creds();
1245 	if (!new)
1246 		return -ENOMEM;
1247 	cap_lower(new->cap_bset, cap);
1248 	return commit_creds(new);
1249 }
1250 
1251 /**
1252  * cap_task_prctl - Implement process control functions for this security module
1253  * @option: The process control function requested
1254  * @arg2: The argument data for this function
1255  * @arg3: The argument data for this function
1256  * @arg4: The argument data for this function
1257  * @arg5: The argument data for this function
1258  *
1259  * Allow process control functions (sys_prctl()) to alter capabilities; may
1260  * also deny access to other functions not otherwise implemented here.
1261  *
1262  * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1263  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1264  * modules will consider performing the function.
1265  */
1266 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1267 		   unsigned long arg4, unsigned long arg5)
1268 {
1269 	const struct cred *old = current_cred();
1270 	struct cred *new;
1271 
1272 	switch (option) {
1273 	case PR_CAPBSET_READ:
1274 		if (!cap_valid(arg2))
1275 			return -EINVAL;
1276 		return !!cap_raised(old->cap_bset, arg2);
1277 
1278 	case PR_CAPBSET_DROP:
1279 		return cap_prctl_drop(arg2);
1280 
1281 	/*
1282 	 * The next four prctl's remain to assist with transitioning a
1283 	 * system from legacy UID=0 based privilege (when filesystem
1284 	 * capabilities are not in use) to a system using filesystem
1285 	 * capabilities only - as the POSIX.1e draft intended.
1286 	 *
1287 	 * Note:
1288 	 *
1289 	 *  PR_SET_SECUREBITS =
1290 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1291 	 *    | issecure_mask(SECURE_NOROOT)
1292 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1293 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1294 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1295 	 *
1296 	 * will ensure that the current process and all of its
1297 	 * children will be locked into a pure
1298 	 * capability-based-privilege environment.
1299 	 */
1300 	case PR_SET_SECUREBITS:
1301 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1302 		     & (old->securebits ^ arg2))			/*[1]*/
1303 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1304 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1305 		    || (cap_capable(current_cred(),
1306 				    current_cred()->user_ns,
1307 				    CAP_SETPCAP,
1308 				    CAP_OPT_NONE) != 0)			/*[4]*/
1309 			/*
1310 			 * [1] no changing of bits that are locked
1311 			 * [2] no unlocking of locks
1312 			 * [3] no setting of unsupported bits
1313 			 * [4] doing anything requires privilege (go read about
1314 			 *     the "sendmail capabilities bug")
1315 			 */
1316 		    )
1317 			/* cannot change a locked bit */
1318 			return -EPERM;
1319 
1320 		new = prepare_creds();
1321 		if (!new)
1322 			return -ENOMEM;
1323 		new->securebits = arg2;
1324 		return commit_creds(new);
1325 
1326 	case PR_GET_SECUREBITS:
1327 		return old->securebits;
1328 
1329 	case PR_GET_KEEPCAPS:
1330 		return !!issecure(SECURE_KEEP_CAPS);
1331 
1332 	case PR_SET_KEEPCAPS:
1333 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1334 			return -EINVAL;
1335 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1336 			return -EPERM;
1337 
1338 		new = prepare_creds();
1339 		if (!new)
1340 			return -ENOMEM;
1341 		if (arg2)
1342 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1343 		else
1344 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1345 		return commit_creds(new);
1346 
1347 	case PR_CAP_AMBIENT:
1348 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1349 			if (arg3 | arg4 | arg5)
1350 				return -EINVAL;
1351 
1352 			new = prepare_creds();
1353 			if (!new)
1354 				return -ENOMEM;
1355 			cap_clear(new->cap_ambient);
1356 			return commit_creds(new);
1357 		}
1358 
1359 		if (((!cap_valid(arg3)) | arg4 | arg5))
1360 			return -EINVAL;
1361 
1362 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1363 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1364 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1365 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1366 			return -EINVAL;
1367 		} else {
1368 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1369 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1370 			     !cap_raised(current_cred()->cap_inheritable,
1371 					 arg3) ||
1372 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1373 				return -EPERM;
1374 
1375 			new = prepare_creds();
1376 			if (!new)
1377 				return -ENOMEM;
1378 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1379 				cap_raise(new->cap_ambient, arg3);
1380 			else
1381 				cap_lower(new->cap_ambient, arg3);
1382 			return commit_creds(new);
1383 		}
1384 
1385 	default:
1386 		/* No functionality available - continue with default */
1387 		return -ENOSYS;
1388 	}
1389 }
1390 
1391 /**
1392  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1393  * @mm: The VM space in which the new mapping is to be made
1394  * @pages: The size of the mapping
1395  *
1396  * Determine whether the allocation of a new virtual mapping by the current
1397  * task is permitted.
1398  *
1399  * Return: 1 if permission is granted, 0 if not.
1400  */
1401 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1402 {
1403 	int cap_sys_admin = 0;
1404 
1405 	if (cap_capable(current_cred(), &init_user_ns,
1406 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1407 		cap_sys_admin = 1;
1408 
1409 	return cap_sys_admin;
1410 }
1411 
1412 /**
1413  * cap_mmap_addr - check if able to map given addr
1414  * @addr: address attempting to be mapped
1415  *
1416  * If the process is attempting to map memory below dac_mmap_min_addr they need
1417  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1418  * capability security module.
1419  *
1420  * Return: 0 if this mapping should be allowed or -EPERM if not.
1421  */
1422 int cap_mmap_addr(unsigned long addr)
1423 {
1424 	int ret = 0;
1425 
1426 	if (addr < dac_mmap_min_addr) {
1427 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1428 				  CAP_OPT_NONE);
1429 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1430 		if (ret == 0)
1431 			current->flags |= PF_SUPERPRIV;
1432 	}
1433 	return ret;
1434 }
1435 
1436 int cap_mmap_file(struct file *file, unsigned long reqprot,
1437 		  unsigned long prot, unsigned long flags)
1438 {
1439 	return 0;
1440 }
1441 
1442 #ifdef CONFIG_SECURITY
1443 
1444 static const struct lsm_id capability_lsmid = {
1445 	.name = "capability",
1446 	.id = LSM_ID_CAPABILITY,
1447 };
1448 
1449 static struct security_hook_list capability_hooks[] __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_lsmid);
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