xref: /linux/security/commoncap.c (revision 811f35ff59b6f99ae272d6f5b96bc9e974f88196)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Common capabilities, needed by capability.o.
3  */
4 
5 #include <linux/capability.h>
6 #include <linux/audit.h>
7 #include <linux/init.h>
8 #include <linux/kernel.h>
9 #include <linux/lsm_hooks.h>
10 #include <linux/file.h>
11 #include <linux/mm.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/skbuff.h>
16 #include <linux/netlink.h>
17 #include <linux/ptrace.h>
18 #include <linux/xattr.h>
19 #include <linux/hugetlb.h>
20 #include <linux/mount.h>
21 #include <linux/sched.h>
22 #include <linux/prctl.h>
23 #include <linux/securebits.h>
24 #include <linux/user_namespace.h>
25 #include <linux/binfmts.h>
26 #include <linux/personality.h>
27 #include <linux/mnt_idmapping.h>
28 
29 /*
30  * If a non-root user executes a setuid-root binary in
31  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
32  * However if fE is also set, then the intent is for only
33  * the file capabilities to be applied, and the setuid-root
34  * bit is left on either to change the uid (plausible) or
35  * to get full privilege on a kernel without file capabilities
36  * support.  So in that case we do not raise capabilities.
37  *
38  * Warn if that happens, once per boot.
39  */
40 static void warn_setuid_and_fcaps_mixed(const char *fname)
41 {
42 	static int warned;
43 	if (!warned) {
44 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
45 			" effective capabilities. Therefore not raising all"
46 			" capabilities.\n", fname);
47 		warned = 1;
48 	}
49 }
50 
51 /**
52  * cap_capable - Determine whether a task has a particular effective capability
53  * @cred: The credentials to use
54  * @targ_ns:  The user namespace in which we need the capability
55  * @cap: The capability to check for
56  * @opts: Bitmask of options defined in include/linux/security.h
57  *
58  * Determine whether the nominated task has the specified capability amongst
59  * its effective set, returning 0 if it does, -ve if it does not.
60  *
61  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
62  * and has_capability() functions.  That is, it has the reverse semantics:
63  * cap_has_capability() returns 0 when a task has a capability, but the
64  * kernel's capable() and has_capability() returns 1 for this case.
65  */
66 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
67 		int cap, unsigned int opts)
68 {
69 	struct user_namespace *ns = targ_ns;
70 
71 	/* See if cred has the capability in the target user namespace
72 	 * by examining the target user namespace and all of the target
73 	 * user namespace's parents.
74 	 */
75 	for (;;) {
76 		/* Do we have the necessary capabilities? */
77 		if (ns == cred->user_ns)
78 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
79 
80 		/*
81 		 * If we're already at a lower level than we're looking for,
82 		 * we're done searching.
83 		 */
84 		if (ns->level <= cred->user_ns->level)
85 			return -EPERM;
86 
87 		/*
88 		 * The owner of the user namespace in the parent of the
89 		 * user namespace has all caps.
90 		 */
91 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
92 			return 0;
93 
94 		/*
95 		 * If you have a capability in a parent user ns, then you have
96 		 * it over all children user namespaces as well.
97 		 */
98 		ns = ns->parent;
99 	}
100 
101 	/* We never get here */
102 }
103 
104 /**
105  * cap_settime - Determine whether the current process may set the system clock
106  * @ts: The time to set
107  * @tz: The timezone to set
108  *
109  * Determine whether the current process may set the system clock and timezone
110  * information, returning 0 if permission granted, -ve if denied.
111  */
112 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
113 {
114 	if (!capable(CAP_SYS_TIME))
115 		return -EPERM;
116 	return 0;
117 }
118 
119 /**
120  * cap_ptrace_access_check - Determine whether the current process may access
121  *			   another
122  * @child: The process to be accessed
123  * @mode: The mode of attachment.
124  *
125  * If we are in the same or an ancestor user_ns and have all the target
126  * task's capabilities, then ptrace access is allowed.
127  * If we have the ptrace capability to the target user_ns, then ptrace
128  * access is allowed.
129  * Else denied.
130  *
131  * Determine whether a process may access another, returning 0 if permission
132  * granted, -ve if denied.
133  */
134 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
135 {
136 	int ret = 0;
137 	const struct cred *cred, *child_cred;
138 	const kernel_cap_t *caller_caps;
139 
140 	rcu_read_lock();
141 	cred = current_cred();
142 	child_cred = __task_cred(child);
143 	if (mode & PTRACE_MODE_FSCREDS)
144 		caller_caps = &cred->cap_effective;
145 	else
146 		caller_caps = &cred->cap_permitted;
147 	if (cred->user_ns == child_cred->user_ns &&
148 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
149 		goto out;
150 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
151 		goto out;
152 	ret = -EPERM;
153 out:
154 	rcu_read_unlock();
155 	return ret;
156 }
157 
158 /**
159  * cap_ptrace_traceme - Determine whether another process may trace the current
160  * @parent: The task proposed to be the tracer
161  *
162  * If parent is in the same or an ancestor user_ns and has all current's
163  * capabilities, then ptrace access is allowed.
164  * If parent has the ptrace capability to current's user_ns, then ptrace
165  * access is allowed.
166  * Else denied.
167  *
168  * Determine whether the nominated task is permitted to trace the current
169  * process, returning 0 if permission is granted, -ve if denied.
170  */
171 int cap_ptrace_traceme(struct task_struct *parent)
172 {
173 	int ret = 0;
174 	const struct cred *cred, *child_cred;
175 
176 	rcu_read_lock();
177 	cred = __task_cred(parent);
178 	child_cred = current_cred();
179 	if (cred->user_ns == child_cred->user_ns &&
180 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
181 		goto out;
182 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
183 		goto out;
184 	ret = -EPERM;
185 out:
186 	rcu_read_unlock();
187 	return ret;
188 }
189 
190 /**
191  * cap_capget - Retrieve a task's capability sets
192  * @target: The task from which to retrieve the capability sets
193  * @effective: The place to record the effective set
194  * @inheritable: The place to record the inheritable set
195  * @permitted: The place to record the permitted set
196  *
197  * This function retrieves the capabilities of the nominated task and returns
198  * them to the caller.
199  */
200 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
201 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
202 {
203 	const struct cred *cred;
204 
205 	/* Derived from kernel/capability.c:sys_capget. */
206 	rcu_read_lock();
207 	cred = __task_cred(target);
208 	*effective   = cred->cap_effective;
209 	*inheritable = cred->cap_inheritable;
210 	*permitted   = cred->cap_permitted;
211 	rcu_read_unlock();
212 	return 0;
213 }
214 
215 /*
216  * Determine whether the inheritable capabilities are limited to the old
217  * permitted set.  Returns 1 if they are limited, 0 if they are not.
218  */
219 static inline int cap_inh_is_capped(void)
220 {
221 	/* they are so limited unless the current task has the CAP_SETPCAP
222 	 * capability
223 	 */
224 	if (cap_capable(current_cred(), current_cred()->user_ns,
225 			CAP_SETPCAP, CAP_OPT_NONE) == 0)
226 		return 0;
227 	return 1;
228 }
229 
230 /**
231  * cap_capset - Validate and apply proposed changes to current's capabilities
232  * @new: The proposed new credentials; alterations should be made here
233  * @old: The current task's current credentials
234  * @effective: A pointer to the proposed new effective capabilities set
235  * @inheritable: A pointer to the proposed new inheritable capabilities set
236  * @permitted: A pointer to the proposed new permitted capabilities set
237  *
238  * This function validates and applies a proposed mass change to the current
239  * process's capability sets.  The changes are made to the proposed new
240  * credentials, and assuming no error, will be committed by the caller of LSM.
241  */
242 int cap_capset(struct cred *new,
243 	       const struct cred *old,
244 	       const kernel_cap_t *effective,
245 	       const kernel_cap_t *inheritable,
246 	       const kernel_cap_t *permitted)
247 {
248 	if (cap_inh_is_capped() &&
249 	    !cap_issubset(*inheritable,
250 			  cap_combine(old->cap_inheritable,
251 				      old->cap_permitted)))
252 		/* incapable of using this inheritable set */
253 		return -EPERM;
254 
255 	if (!cap_issubset(*inheritable,
256 			  cap_combine(old->cap_inheritable,
257 				      old->cap_bset)))
258 		/* no new pI capabilities outside bounding set */
259 		return -EPERM;
260 
261 	/* verify restrictions on target's new Permitted set */
262 	if (!cap_issubset(*permitted, old->cap_permitted))
263 		return -EPERM;
264 
265 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
266 	if (!cap_issubset(*effective, *permitted))
267 		return -EPERM;
268 
269 	new->cap_effective   = *effective;
270 	new->cap_inheritable = *inheritable;
271 	new->cap_permitted   = *permitted;
272 
273 	/*
274 	 * Mask off ambient bits that are no longer both permitted and
275 	 * inheritable.
276 	 */
277 	new->cap_ambient = cap_intersect(new->cap_ambient,
278 					 cap_intersect(*permitted,
279 						       *inheritable));
280 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
281 		return -EINVAL;
282 	return 0;
283 }
284 
285 /**
286  * cap_inode_need_killpriv - Determine if inode change affects privileges
287  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
288  *
289  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
290  * affects the security markings on that inode, and if it is, should
291  * inode_killpriv() be invoked or the change rejected.
292  *
293  * Return: 1 if security.capability has a value, meaning inode_killpriv()
294  * is required, 0 otherwise, meaning inode_killpriv() is not required.
295  */
296 int cap_inode_need_killpriv(struct dentry *dentry)
297 {
298 	struct inode *inode = d_backing_inode(dentry);
299 	int error;
300 
301 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
302 	return error > 0;
303 }
304 
305 /**
306  * cap_inode_killpriv - Erase the security markings on an inode
307  *
308  * @mnt_userns:	user namespace of the mount the inode was found from
309  * @dentry:	The inode/dentry to alter
310  *
311  * Erase the privilege-enhancing security markings on an inode.
312  *
313  * If the inode has been found through an idmapped mount the user namespace of
314  * the vfsmount must be passed through @mnt_userns. This function will then
315  * take care to map the inode according to @mnt_userns before checking
316  * permissions. On non-idmapped mounts or if permission checking is to be
317  * performed on the raw inode simply passs init_user_ns.
318  *
319  * Return: 0 if successful, -ve on error.
320  */
321 int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry)
322 {
323 	int error;
324 
325 	error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS);
326 	if (error == -EOPNOTSUPP)
327 		error = 0;
328 	return error;
329 }
330 
331 static bool rootid_owns_currentns(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 user_namespace *mnt_userns,
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(mnt_userns, 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(mnt_userns, 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  * @mnt_userns:	user namespace 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 user namespace of
522  * the vfsmount must be passed through @mnt_userns. This function will then
523  * take care to map the inode according to @mnt_userns before checking
524  * permissions. On non-idmapped mounts or if permission checking is to be
525  * performed on the raw inode simply passs init_user_ns.
526  *
527  * Return: On success, return the new size; on error, return < 0.
528  */
529 int cap_convert_nscap(struct user_namespace *mnt_userns, 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(mnt_userns, inode, CAP_SETFCAP))
548 		return -EPERM;
549 	if (size == XATTR_CAPS_SZ_2 && (mnt_userns == fs_ns))
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(mnt_userns, 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 	unsigned i;
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 	CAP_FOR_EACH_U32(i) {
602 		__u32 permitted = caps->permitted.cap[i];
603 		__u32 inheritable = caps->inheritable.cap[i];
604 
605 		/*
606 		 * pP' = (X & fP) | (pI & fI)
607 		 * The addition of pA' is handled later.
608 		 */
609 		new->cap_permitted.cap[i] =
610 			(new->cap_bset.cap[i] & permitted) |
611 			(new->cap_inheritable.cap[i] & inheritable);
612 
613 		if (permitted & ~new->cap_permitted.cap[i])
614 			/* insufficient to execute correctly */
615 			ret = -EPERM;
616 	}
617 
618 	/*
619 	 * For legacy apps, with no internal support for recognizing they
620 	 * do not have enough capabilities, we return an error if they are
621 	 * missing some "forced" (aka file-permitted) capabilities.
622 	 */
623 	return *effective ? ret : 0;
624 }
625 
626 /**
627  * get_vfs_caps_from_disk - retrieve vfs caps from disk
628  *
629  * @mnt_userns:	user namespace of the mount the inode was found from
630  * @dentry:	dentry from which @inode is retrieved
631  * @cpu_caps:	vfs capabilities
632  *
633  * Extract the on-exec-apply capability sets for an executable file.
634  *
635  * If the inode has been found through an idmapped mount the user namespace of
636  * the vfsmount must be passed through @mnt_userns. This function will then
637  * take care to map the inode according to @mnt_userns before checking
638  * permissions. On non-idmapped mounts or if permission checking is to be
639  * performed on the raw inode simply passs init_user_ns.
640  */
641 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
642 			   const struct dentry *dentry,
643 			   struct cpu_vfs_cap_data *cpu_caps)
644 {
645 	struct inode *inode = d_backing_inode(dentry);
646 	__u32 magic_etc;
647 	unsigned tocopy, i;
648 	int size;
649 	struct vfs_ns_cap_data data, *nscaps = &data;
650 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
651 	kuid_t rootkuid;
652 	vfsuid_t rootvfsuid;
653 	struct user_namespace *fs_ns;
654 
655 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
656 
657 	if (!inode)
658 		return -ENODATA;
659 
660 	fs_ns = inode->i_sb->s_user_ns;
661 	size = __vfs_getxattr((struct dentry *)dentry, inode,
662 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
663 	if (size == -ENODATA || size == -EOPNOTSUPP)
664 		/* no data, that's ok */
665 		return -ENODATA;
666 
667 	if (size < 0)
668 		return size;
669 
670 	if (size < sizeof(magic_etc))
671 		return -EINVAL;
672 
673 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
674 
675 	rootkuid = make_kuid(fs_ns, 0);
676 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
677 	case VFS_CAP_REVISION_1:
678 		if (size != XATTR_CAPS_SZ_1)
679 			return -EINVAL;
680 		tocopy = VFS_CAP_U32_1;
681 		break;
682 	case VFS_CAP_REVISION_2:
683 		if (size != XATTR_CAPS_SZ_2)
684 			return -EINVAL;
685 		tocopy = VFS_CAP_U32_2;
686 		break;
687 	case VFS_CAP_REVISION_3:
688 		if (size != XATTR_CAPS_SZ_3)
689 			return -EINVAL;
690 		tocopy = VFS_CAP_U32_3;
691 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
692 		break;
693 
694 	default:
695 		return -EINVAL;
696 	}
697 
698 	rootvfsuid = make_vfsuid(mnt_userns, fs_ns, rootkuid);
699 	if (!vfsuid_valid(rootvfsuid))
700 		return -ENODATA;
701 
702 	/* Limit the caps to the mounter of the filesystem
703 	 * or the more limited uid specified in the xattr.
704 	 */
705 	if (!rootid_owns_currentns(rootvfsuid))
706 		return -ENODATA;
707 
708 	CAP_FOR_EACH_U32(i) {
709 		if (i >= tocopy)
710 			break;
711 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
712 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
713 	}
714 
715 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
716 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
717 
718 	cpu_caps->rootid = vfsuid_into_kuid(rootvfsuid);
719 
720 	return 0;
721 }
722 
723 /*
724  * Attempt to get the on-exec apply capability sets for an executable file from
725  * its xattrs and, if present, apply them to the proposed credentials being
726  * constructed by execve().
727  */
728 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
729 			 bool *effective, bool *has_fcap)
730 {
731 	int rc = 0;
732 	struct cpu_vfs_cap_data vcaps;
733 
734 	cap_clear(bprm->cred->cap_permitted);
735 
736 	if (!file_caps_enabled)
737 		return 0;
738 
739 	if (!mnt_may_suid(file->f_path.mnt))
740 		return 0;
741 
742 	/*
743 	 * This check is redundant with mnt_may_suid() but is kept to make
744 	 * explicit that capability bits are limited to s_user_ns and its
745 	 * descendants.
746 	 */
747 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
748 		return 0;
749 
750 	rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
751 				    file->f_path.dentry, &vcaps);
752 	if (rc < 0) {
753 		if (rc == -EINVAL)
754 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
755 					bprm->filename);
756 		else if (rc == -ENODATA)
757 			rc = 0;
758 		goto out;
759 	}
760 
761 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
762 
763 out:
764 	if (rc)
765 		cap_clear(bprm->cred->cap_permitted);
766 
767 	return rc;
768 }
769 
770 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
771 
772 static inline bool __is_real(kuid_t uid, struct cred *cred)
773 { return uid_eq(cred->uid, uid); }
774 
775 static inline bool __is_eff(kuid_t uid, struct cred *cred)
776 { return uid_eq(cred->euid, uid); }
777 
778 static inline bool __is_suid(kuid_t uid, struct cred *cred)
779 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
780 
781 /*
782  * handle_privileged_root - Handle case of privileged root
783  * @bprm: The execution parameters, including the proposed creds
784  * @has_fcap: Are any file capabilities set?
785  * @effective: Do we have effective root privilege?
786  * @root_uid: This namespace' root UID WRT initial USER namespace
787  *
788  * Handle the case where root is privileged and hasn't been neutered by
789  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
790  * set UID root and nothing is changed.  If we are root, cap_permitted is
791  * updated.  If we have become set UID root, the effective bit is set.
792  */
793 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
794 				   bool *effective, kuid_t root_uid)
795 {
796 	const struct cred *old = current_cred();
797 	struct cred *new = bprm->cred;
798 
799 	if (!root_privileged())
800 		return;
801 	/*
802 	 * If the legacy file capability is set, then don't set privs
803 	 * for a setuid root binary run by a non-root user.  Do set it
804 	 * for a root user just to cause least surprise to an admin.
805 	 */
806 	if (has_fcap && __is_suid(root_uid, new)) {
807 		warn_setuid_and_fcaps_mixed(bprm->filename);
808 		return;
809 	}
810 	/*
811 	 * To support inheritance of root-permissions and suid-root
812 	 * executables under compatibility mode, we override the
813 	 * capability sets for the file.
814 	 */
815 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
816 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
817 		new->cap_permitted = cap_combine(old->cap_bset,
818 						 old->cap_inheritable);
819 	}
820 	/*
821 	 * If only the real uid is 0, we do not set the effective bit.
822 	 */
823 	if (__is_eff(root_uid, new))
824 		*effective = true;
825 }
826 
827 #define __cap_gained(field, target, source) \
828 	!cap_issubset(target->cap_##field, source->cap_##field)
829 #define __cap_grew(target, source, cred) \
830 	!cap_issubset(cred->cap_##target, cred->cap_##source)
831 #define __cap_full(field, cred) \
832 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
833 
834 static inline bool __is_setuid(struct cred *new, const struct cred *old)
835 { return !uid_eq(new->euid, old->uid); }
836 
837 static inline bool __is_setgid(struct cred *new, const struct cred *old)
838 { return !gid_eq(new->egid, old->gid); }
839 
840 /*
841  * 1) Audit candidate if current->cap_effective is set
842  *
843  * We do not bother to audit if 3 things are true:
844  *   1) cap_effective has all caps
845  *   2) we became root *OR* are were already root
846  *   3) root is supposed to have all caps (SECURE_NOROOT)
847  * Since this is just a normal root execing a process.
848  *
849  * Number 1 above might fail if you don't have a full bset, but I think
850  * that is interesting information to audit.
851  *
852  * A number of other conditions require logging:
853  * 2) something prevented setuid root getting all caps
854  * 3) non-setuid root gets fcaps
855  * 4) non-setuid root gets ambient
856  */
857 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
858 				     kuid_t root, bool has_fcap)
859 {
860 	bool ret = false;
861 
862 	if ((__cap_grew(effective, ambient, new) &&
863 	     !(__cap_full(effective, new) &&
864 	       (__is_eff(root, new) || __is_real(root, new)) &&
865 	       root_privileged())) ||
866 	    (root_privileged() &&
867 	     __is_suid(root, new) &&
868 	     !__cap_full(effective, new)) ||
869 	    (!__is_setuid(new, old) &&
870 	     ((has_fcap &&
871 	       __cap_gained(permitted, new, old)) ||
872 	      __cap_gained(ambient, new, old))))
873 
874 		ret = true;
875 
876 	return ret;
877 }
878 
879 /**
880  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
881  * @bprm: The execution parameters, including the proposed creds
882  * @file: The file to pull the credentials from
883  *
884  * Set up the proposed credentials for a new execution context being
885  * constructed by execve().  The proposed creds in @bprm->cred is altered,
886  * which won't take effect immediately.
887  *
888  * Return: 0 if successful, -ve on error.
889  */
890 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
891 {
892 	/* Process setpcap binaries and capabilities for uid 0 */
893 	const struct cred *old = current_cred();
894 	struct cred *new = bprm->cred;
895 	bool effective = false, has_fcap = false, is_setid;
896 	int ret;
897 	kuid_t root_uid;
898 
899 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
900 		return -EPERM;
901 
902 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
903 	if (ret < 0)
904 		return ret;
905 
906 	root_uid = make_kuid(new->user_ns, 0);
907 
908 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
909 
910 	/* if we have fs caps, clear dangerous personality flags */
911 	if (__cap_gained(permitted, new, old))
912 		bprm->per_clear |= PER_CLEAR_ON_SETID;
913 
914 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
915 	 * credentials unless they have the appropriate permit.
916 	 *
917 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
918 	 */
919 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
920 
921 	if ((is_setid || __cap_gained(permitted, new, old)) &&
922 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
923 	     !ptracer_capable(current, new->user_ns))) {
924 		/* downgrade; they get no more than they had, and maybe less */
925 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
926 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
927 			new->euid = new->uid;
928 			new->egid = new->gid;
929 		}
930 		new->cap_permitted = cap_intersect(new->cap_permitted,
931 						   old->cap_permitted);
932 	}
933 
934 	new->suid = new->fsuid = new->euid;
935 	new->sgid = new->fsgid = new->egid;
936 
937 	/* File caps or setid cancels ambient. */
938 	if (has_fcap || is_setid)
939 		cap_clear(new->cap_ambient);
940 
941 	/*
942 	 * Now that we've computed pA', update pP' to give:
943 	 *   pP' = (X & fP) | (pI & fI) | pA'
944 	 */
945 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
946 
947 	/*
948 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
949 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
950 	 */
951 	if (effective)
952 		new->cap_effective = new->cap_permitted;
953 	else
954 		new->cap_effective = new->cap_ambient;
955 
956 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
957 		return -EPERM;
958 
959 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
960 		ret = audit_log_bprm_fcaps(bprm, new, old);
961 		if (ret < 0)
962 			return ret;
963 	}
964 
965 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
966 
967 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
968 		return -EPERM;
969 
970 	/* Check for privilege-elevated exec. */
971 	if (is_setid ||
972 	    (!__is_real(root_uid, new) &&
973 	     (effective ||
974 	      __cap_grew(permitted, ambient, new))))
975 		bprm->secureexec = 1;
976 
977 	return 0;
978 }
979 
980 /**
981  * cap_inode_setxattr - Determine whether an xattr may be altered
982  * @dentry: The inode/dentry being altered
983  * @name: The name of the xattr to be changed
984  * @value: The value that the xattr will be changed to
985  * @size: The size of value
986  * @flags: The replacement flag
987  *
988  * Determine whether an xattr may be altered or set on an inode, returning 0 if
989  * permission is granted, -ve if denied.
990  *
991  * This is used to make sure security xattrs don't get updated or set by those
992  * who aren't privileged to do so.
993  */
994 int cap_inode_setxattr(struct dentry *dentry, const char *name,
995 		       const void *value, size_t size, int flags)
996 {
997 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
998 
999 	/* Ignore non-security xattrs */
1000 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1001 			XATTR_SECURITY_PREFIX_LEN) != 0)
1002 		return 0;
1003 
1004 	/*
1005 	 * For XATTR_NAME_CAPS the check will be done in
1006 	 * cap_convert_nscap(), called by setxattr()
1007 	 */
1008 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
1009 		return 0;
1010 
1011 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1012 		return -EPERM;
1013 	return 0;
1014 }
1015 
1016 /**
1017  * cap_inode_removexattr - Determine whether an xattr may be removed
1018  *
1019  * @mnt_userns:	User namespace of the mount the inode was found from
1020  * @dentry:	The inode/dentry being altered
1021  * @name:	The name of the xattr to be changed
1022  *
1023  * Determine whether an xattr may be removed from an inode, returning 0 if
1024  * permission is granted, -ve if denied.
1025  *
1026  * If the inode has been found through an idmapped mount the user namespace of
1027  * the vfsmount must be passed through @mnt_userns. This function will then
1028  * take care to map the inode according to @mnt_userns before checking
1029  * permissions. On non-idmapped mounts or if permission checking is to be
1030  * performed on the raw inode simply passs init_user_ns.
1031  *
1032  * This is used to make sure security xattrs don't get removed by those who
1033  * aren't privileged to remove them.
1034  */
1035 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1036 			  struct dentry *dentry, const char *name)
1037 {
1038 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1039 
1040 	/* Ignore non-security xattrs */
1041 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1042 			XATTR_SECURITY_PREFIX_LEN) != 0)
1043 		return 0;
1044 
1045 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1046 		/* security.capability gets namespaced */
1047 		struct inode *inode = d_backing_inode(dentry);
1048 		if (!inode)
1049 			return -EINVAL;
1050 		if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1051 			return -EPERM;
1052 		return 0;
1053 	}
1054 
1055 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1056 		return -EPERM;
1057 	return 0;
1058 }
1059 
1060 /*
1061  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1062  * a process after a call to setuid, setreuid, or setresuid.
1063  *
1064  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1065  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1066  *  cleared.
1067  *
1068  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1069  *  capabilities of the process are cleared.
1070  *
1071  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1072  *  capabilities are set to the permitted capabilities.
1073  *
1074  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1075  *  never happen.
1076  *
1077  *  -astor
1078  *
1079  * cevans - New behaviour, Oct '99
1080  * A process may, via prctl(), elect to keep its capabilities when it
1081  * calls setuid() and switches away from uid==0. Both permitted and
1082  * effective sets will be retained.
1083  * Without this change, it was impossible for a daemon to drop only some
1084  * of its privilege. The call to setuid(!=0) would drop all privileges!
1085  * Keeping uid 0 is not an option because uid 0 owns too many vital
1086  * files..
1087  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1088  */
1089 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1090 {
1091 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1092 
1093 	if ((uid_eq(old->uid, root_uid) ||
1094 	     uid_eq(old->euid, root_uid) ||
1095 	     uid_eq(old->suid, root_uid)) &&
1096 	    (!uid_eq(new->uid, root_uid) &&
1097 	     !uid_eq(new->euid, root_uid) &&
1098 	     !uid_eq(new->suid, root_uid))) {
1099 		if (!issecure(SECURE_KEEP_CAPS)) {
1100 			cap_clear(new->cap_permitted);
1101 			cap_clear(new->cap_effective);
1102 		}
1103 
1104 		/*
1105 		 * Pre-ambient programs expect setresuid to nonroot followed
1106 		 * by exec to drop capabilities.  We should make sure that
1107 		 * this remains the case.
1108 		 */
1109 		cap_clear(new->cap_ambient);
1110 	}
1111 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1112 		cap_clear(new->cap_effective);
1113 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1114 		new->cap_effective = new->cap_permitted;
1115 }
1116 
1117 /**
1118  * cap_task_fix_setuid - Fix up the results of setuid() call
1119  * @new: The proposed credentials
1120  * @old: The current task's current credentials
1121  * @flags: Indications of what has changed
1122  *
1123  * Fix up the results of setuid() call before the credential changes are
1124  * actually applied.
1125  *
1126  * Return: 0 to grant the changes, -ve to deny them.
1127  */
1128 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1129 {
1130 	switch (flags) {
1131 	case LSM_SETID_RE:
1132 	case LSM_SETID_ID:
1133 	case LSM_SETID_RES:
1134 		/* juggle the capabilities to follow [RES]UID changes unless
1135 		 * otherwise suppressed */
1136 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1137 			cap_emulate_setxuid(new, old);
1138 		break;
1139 
1140 	case LSM_SETID_FS:
1141 		/* juggle the capabilties to follow FSUID changes, unless
1142 		 * otherwise suppressed
1143 		 *
1144 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1145 		 *          if not, we might be a bit too harsh here.
1146 		 */
1147 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1148 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1149 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1150 				new->cap_effective =
1151 					cap_drop_fs_set(new->cap_effective);
1152 
1153 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1154 				new->cap_effective =
1155 					cap_raise_fs_set(new->cap_effective,
1156 							 new->cap_permitted);
1157 		}
1158 		break;
1159 
1160 	default:
1161 		return -EINVAL;
1162 	}
1163 
1164 	return 0;
1165 }
1166 
1167 /*
1168  * Rationale: code calling task_setscheduler, task_setioprio, and
1169  * task_setnice, assumes that
1170  *   . if capable(cap_sys_nice), then those actions should be allowed
1171  *   . if not capable(cap_sys_nice), but acting on your own processes,
1172  *   	then those actions should be allowed
1173  * This is insufficient now since you can call code without suid, but
1174  * yet with increased caps.
1175  * So we check for increased caps on the target process.
1176  */
1177 static int cap_safe_nice(struct task_struct *p)
1178 {
1179 	int is_subset, ret = 0;
1180 
1181 	rcu_read_lock();
1182 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1183 				 current_cred()->cap_permitted);
1184 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1185 		ret = -EPERM;
1186 	rcu_read_unlock();
1187 
1188 	return ret;
1189 }
1190 
1191 /**
1192  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1193  * @p: The task to affect
1194  *
1195  * Detemine if the requested scheduler policy change is permitted for the
1196  * specified task.
1197  *
1198  * Return: 0 if permission is granted, -ve if denied.
1199  */
1200 int cap_task_setscheduler(struct task_struct *p)
1201 {
1202 	return cap_safe_nice(p);
1203 }
1204 
1205 /**
1206  * cap_task_setioprio - Detemine if I/O priority change is permitted
1207  * @p: The task to affect
1208  * @ioprio: The I/O priority to set
1209  *
1210  * Detemine if the requested I/O priority change is permitted for the specified
1211  * task.
1212  *
1213  * Return: 0 if permission is granted, -ve if denied.
1214  */
1215 int cap_task_setioprio(struct task_struct *p, int ioprio)
1216 {
1217 	return cap_safe_nice(p);
1218 }
1219 
1220 /**
1221  * cap_task_setnice - Detemine if task priority change is permitted
1222  * @p: The task to affect
1223  * @nice: The nice value to set
1224  *
1225  * Detemine if the requested task priority change is permitted for the
1226  * specified task.
1227  *
1228  * Return: 0 if permission is granted, -ve if denied.
1229  */
1230 int cap_task_setnice(struct task_struct *p, int nice)
1231 {
1232 	return cap_safe_nice(p);
1233 }
1234 
1235 /*
1236  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1237  * the current task's bounding set.  Returns 0 on success, -ve on error.
1238  */
1239 static int cap_prctl_drop(unsigned long cap)
1240 {
1241 	struct cred *new;
1242 
1243 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1244 		return -EPERM;
1245 	if (!cap_valid(cap))
1246 		return -EINVAL;
1247 
1248 	new = prepare_creds();
1249 	if (!new)
1250 		return -ENOMEM;
1251 	cap_lower(new->cap_bset, cap);
1252 	return commit_creds(new);
1253 }
1254 
1255 /**
1256  * cap_task_prctl - Implement process control functions for this security module
1257  * @option: The process control function requested
1258  * @arg2: The argument data for this function
1259  * @arg3: The argument data for this function
1260  * @arg4: The argument data for this function
1261  * @arg5: The argument data for this function
1262  *
1263  * Allow process control functions (sys_prctl()) to alter capabilities; may
1264  * also deny access to other functions not otherwise implemented here.
1265  *
1266  * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1267  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1268  * modules will consider performing the function.
1269  */
1270 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1271 		   unsigned long arg4, unsigned long arg5)
1272 {
1273 	const struct cred *old = current_cred();
1274 	struct cred *new;
1275 
1276 	switch (option) {
1277 	case PR_CAPBSET_READ:
1278 		if (!cap_valid(arg2))
1279 			return -EINVAL;
1280 		return !!cap_raised(old->cap_bset, arg2);
1281 
1282 	case PR_CAPBSET_DROP:
1283 		return cap_prctl_drop(arg2);
1284 
1285 	/*
1286 	 * The next four prctl's remain to assist with transitioning a
1287 	 * system from legacy UID=0 based privilege (when filesystem
1288 	 * capabilities are not in use) to a system using filesystem
1289 	 * capabilities only - as the POSIX.1e draft intended.
1290 	 *
1291 	 * Note:
1292 	 *
1293 	 *  PR_SET_SECUREBITS =
1294 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1295 	 *    | issecure_mask(SECURE_NOROOT)
1296 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1297 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1298 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1299 	 *
1300 	 * will ensure that the current process and all of its
1301 	 * children will be locked into a pure
1302 	 * capability-based-privilege environment.
1303 	 */
1304 	case PR_SET_SECUREBITS:
1305 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1306 		     & (old->securebits ^ arg2))			/*[1]*/
1307 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1308 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1309 		    || (cap_capable(current_cred(),
1310 				    current_cred()->user_ns,
1311 				    CAP_SETPCAP,
1312 				    CAP_OPT_NONE) != 0)			/*[4]*/
1313 			/*
1314 			 * [1] no changing of bits that are locked
1315 			 * [2] no unlocking of locks
1316 			 * [3] no setting of unsupported bits
1317 			 * [4] doing anything requires privilege (go read about
1318 			 *     the "sendmail capabilities bug")
1319 			 */
1320 		    )
1321 			/* cannot change a locked bit */
1322 			return -EPERM;
1323 
1324 		new = prepare_creds();
1325 		if (!new)
1326 			return -ENOMEM;
1327 		new->securebits = arg2;
1328 		return commit_creds(new);
1329 
1330 	case PR_GET_SECUREBITS:
1331 		return old->securebits;
1332 
1333 	case PR_GET_KEEPCAPS:
1334 		return !!issecure(SECURE_KEEP_CAPS);
1335 
1336 	case PR_SET_KEEPCAPS:
1337 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1338 			return -EINVAL;
1339 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1340 			return -EPERM;
1341 
1342 		new = prepare_creds();
1343 		if (!new)
1344 			return -ENOMEM;
1345 		if (arg2)
1346 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1347 		else
1348 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1349 		return commit_creds(new);
1350 
1351 	case PR_CAP_AMBIENT:
1352 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1353 			if (arg3 | arg4 | arg5)
1354 				return -EINVAL;
1355 
1356 			new = prepare_creds();
1357 			if (!new)
1358 				return -ENOMEM;
1359 			cap_clear(new->cap_ambient);
1360 			return commit_creds(new);
1361 		}
1362 
1363 		if (((!cap_valid(arg3)) | arg4 | arg5))
1364 			return -EINVAL;
1365 
1366 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1367 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1368 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1369 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1370 			return -EINVAL;
1371 		} else {
1372 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1373 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1374 			     !cap_raised(current_cred()->cap_inheritable,
1375 					 arg3) ||
1376 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1377 				return -EPERM;
1378 
1379 			new = prepare_creds();
1380 			if (!new)
1381 				return -ENOMEM;
1382 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1383 				cap_raise(new->cap_ambient, arg3);
1384 			else
1385 				cap_lower(new->cap_ambient, arg3);
1386 			return commit_creds(new);
1387 		}
1388 
1389 	default:
1390 		/* No functionality available - continue with default */
1391 		return -ENOSYS;
1392 	}
1393 }
1394 
1395 /**
1396  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1397  * @mm: The VM space in which the new mapping is to be made
1398  * @pages: The size of the mapping
1399  *
1400  * Determine whether the allocation of a new virtual mapping by the current
1401  * task is permitted.
1402  *
1403  * Return: 1 if permission is granted, 0 if not.
1404  */
1405 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1406 {
1407 	int cap_sys_admin = 0;
1408 
1409 	if (cap_capable(current_cred(), &init_user_ns,
1410 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1411 		cap_sys_admin = 1;
1412 
1413 	return cap_sys_admin;
1414 }
1415 
1416 /**
1417  * cap_mmap_addr - check if able to map given addr
1418  * @addr: address attempting to be mapped
1419  *
1420  * If the process is attempting to map memory below dac_mmap_min_addr they need
1421  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1422  * capability security module.
1423  *
1424  * Return: 0 if this mapping should be allowed or -EPERM if not.
1425  */
1426 int cap_mmap_addr(unsigned long addr)
1427 {
1428 	int ret = 0;
1429 
1430 	if (addr < dac_mmap_min_addr) {
1431 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1432 				  CAP_OPT_NONE);
1433 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1434 		if (ret == 0)
1435 			current->flags |= PF_SUPERPRIV;
1436 	}
1437 	return ret;
1438 }
1439 
1440 int cap_mmap_file(struct file *file, unsigned long reqprot,
1441 		  unsigned long prot, unsigned long flags)
1442 {
1443 	return 0;
1444 }
1445 
1446 #ifdef CONFIG_SECURITY
1447 
1448 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1449 	LSM_HOOK_INIT(capable, cap_capable),
1450 	LSM_HOOK_INIT(settime, cap_settime),
1451 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1452 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1453 	LSM_HOOK_INIT(capget, cap_capget),
1454 	LSM_HOOK_INIT(capset, cap_capset),
1455 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1456 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1457 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1458 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1459 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1460 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1461 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1462 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1463 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1464 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1465 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1466 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1467 };
1468 
1469 static int __init capability_init(void)
1470 {
1471 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1472 				"capability");
1473 	return 0;
1474 }
1475 
1476 DEFINE_LSM(capability) = {
1477 	.name = "capability",
1478 	.order = LSM_ORDER_FIRST,
1479 	.init = capability_init,
1480 };
1481 
1482 #endif /* CONFIG_SECURITY */
1483