xref: /openbmc/linux/security/commoncap.c (revision 52cdded0)
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 
28 /*
29  * If a non-root user executes a setuid-root binary in
30  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
31  * However if fE is also set, then the intent is for only
32  * the file capabilities to be applied, and the setuid-root
33  * bit is left on either to change the uid (plausible) or
34  * to get full privilege on a kernel without file capabilities
35  * support.  So in that case we do not raise capabilities.
36  *
37  * Warn if that happens, once per boot.
38  */
39 static void warn_setuid_and_fcaps_mixed(const char *fname)
40 {
41 	static int warned;
42 	if (!warned) {
43 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
44 			" effective capabilities. Therefore not raising all"
45 			" capabilities.\n", fname);
46 		warned = 1;
47 	}
48 }
49 
50 /**
51  * cap_capable - Determine whether a task has a particular effective capability
52  * @cred: The credentials to use
53  * @ns:  The user namespace in which we need the capability
54  * @cap: The capability to check for
55  * @opts: Bitmask of options defined in include/linux/security.h
56  *
57  * Determine whether the nominated task has the specified capability amongst
58  * its effective set, returning 0 if it does, -ve if it does not.
59  *
60  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
61  * and has_capability() functions.  That is, it has the reverse semantics:
62  * cap_has_capability() returns 0 when a task has a capability, but the
63  * kernel's capable() and has_capability() returns 1 for this case.
64  */
65 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
66 		int cap, unsigned int opts)
67 {
68 	struct user_namespace *ns = targ_ns;
69 
70 	/* See if cred has the capability in the target user namespace
71 	 * by examining the target user namespace and all of the target
72 	 * user namespace's parents.
73 	 */
74 	for (;;) {
75 		/* Do we have the necessary capabilities? */
76 		if (ns == cred->user_ns)
77 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
78 
79 		/*
80 		 * If we're already at a lower level than we're looking for,
81 		 * we're done searching.
82 		 */
83 		if (ns->level <= cred->user_ns->level)
84 			return -EPERM;
85 
86 		/*
87 		 * The owner of the user namespace in the parent of the
88 		 * user namespace has all caps.
89 		 */
90 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
91 			return 0;
92 
93 		/*
94 		 * If you have a capability in a parent user ns, then you have
95 		 * it over all children user namespaces as well.
96 		 */
97 		ns = ns->parent;
98 	}
99 
100 	/* We never get here */
101 }
102 
103 /**
104  * cap_settime - Determine whether the current process may set the system clock
105  * @ts: The time to set
106  * @tz: The timezone to set
107  *
108  * Determine whether the current process may set the system clock and timezone
109  * information, returning 0 if permission granted, -ve if denied.
110  */
111 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
112 {
113 	if (!capable(CAP_SYS_TIME))
114 		return -EPERM;
115 	return 0;
116 }
117 
118 /**
119  * cap_ptrace_access_check - Determine whether the current process may access
120  *			   another
121  * @child: The process to be accessed
122  * @mode: The mode of attachment.
123  *
124  * If we are in the same or an ancestor user_ns and have all the target
125  * task's capabilities, then ptrace access is allowed.
126  * If we have the ptrace capability to the target user_ns, then ptrace
127  * access is allowed.
128  * Else denied.
129  *
130  * Determine whether a process may access another, returning 0 if permission
131  * granted, -ve if denied.
132  */
133 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
134 {
135 	int ret = 0;
136 	const struct cred *cred, *child_cred;
137 	const kernel_cap_t *caller_caps;
138 
139 	rcu_read_lock();
140 	cred = current_cred();
141 	child_cred = __task_cred(child);
142 	if (mode & PTRACE_MODE_FSCREDS)
143 		caller_caps = &cred->cap_effective;
144 	else
145 		caller_caps = &cred->cap_permitted;
146 	if (cred->user_ns == child_cred->user_ns &&
147 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
148 		goto out;
149 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
150 		goto out;
151 	ret = -EPERM;
152 out:
153 	rcu_read_unlock();
154 	return ret;
155 }
156 
157 /**
158  * cap_ptrace_traceme - Determine whether another process may trace the current
159  * @parent: The task proposed to be the tracer
160  *
161  * If parent is in the same or an ancestor user_ns and has all current's
162  * capabilities, then ptrace access is allowed.
163  * If parent has the ptrace capability to current's user_ns, then ptrace
164  * access is allowed.
165  * Else denied.
166  *
167  * Determine whether the nominated task is permitted to trace the current
168  * process, returning 0 if permission is granted, -ve if denied.
169  */
170 int cap_ptrace_traceme(struct task_struct *parent)
171 {
172 	int ret = 0;
173 	const struct cred *cred, *child_cred;
174 
175 	rcu_read_lock();
176 	cred = __task_cred(parent);
177 	child_cred = current_cred();
178 	if (cred->user_ns == child_cred->user_ns &&
179 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
180 		goto out;
181 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
182 		goto out;
183 	ret = -EPERM;
184 out:
185 	rcu_read_unlock();
186 	return ret;
187 }
188 
189 /**
190  * cap_capget - Retrieve a task's capability sets
191  * @target: The task from which to retrieve the capability sets
192  * @effective: The place to record the effective set
193  * @inheritable: The place to record the inheritable set
194  * @permitted: The place to record the permitted set
195  *
196  * This function retrieves the capabilities of the nominated task and returns
197  * them to the caller.
198  */
199 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
200 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
201 {
202 	const struct cred *cred;
203 
204 	/* Derived from kernel/capability.c:sys_capget. */
205 	rcu_read_lock();
206 	cred = __task_cred(target);
207 	*effective   = cred->cap_effective;
208 	*inheritable = cred->cap_inheritable;
209 	*permitted   = cred->cap_permitted;
210 	rcu_read_unlock();
211 	return 0;
212 }
213 
214 /*
215  * Determine whether the inheritable capabilities are limited to the old
216  * permitted set.  Returns 1 if they are limited, 0 if they are not.
217  */
218 static inline int cap_inh_is_capped(void)
219 {
220 	/* they are so limited unless the current task has the CAP_SETPCAP
221 	 * capability
222 	 */
223 	if (cap_capable(current_cred(), current_cred()->user_ns,
224 			CAP_SETPCAP, CAP_OPT_NONE) == 0)
225 		return 0;
226 	return 1;
227 }
228 
229 /**
230  * cap_capset - Validate and apply proposed changes to current's capabilities
231  * @new: The proposed new credentials; alterations should be made here
232  * @old: The current task's current credentials
233  * @effective: A pointer to the proposed new effective capabilities set
234  * @inheritable: A pointer to the proposed new inheritable capabilities set
235  * @permitted: A pointer to the proposed new permitted capabilities set
236  *
237  * This function validates and applies a proposed mass change to the current
238  * process's capability sets.  The changes are made to the proposed new
239  * credentials, and assuming no error, will be committed by the caller of LSM.
240  */
241 int cap_capset(struct cred *new,
242 	       const struct cred *old,
243 	       const kernel_cap_t *effective,
244 	       const kernel_cap_t *inheritable,
245 	       const kernel_cap_t *permitted)
246 {
247 	if (cap_inh_is_capped() &&
248 	    !cap_issubset(*inheritable,
249 			  cap_combine(old->cap_inheritable,
250 				      old->cap_permitted)))
251 		/* incapable of using this inheritable set */
252 		return -EPERM;
253 
254 	if (!cap_issubset(*inheritable,
255 			  cap_combine(old->cap_inheritable,
256 				      old->cap_bset)))
257 		/* no new pI capabilities outside bounding set */
258 		return -EPERM;
259 
260 	/* verify restrictions on target's new Permitted set */
261 	if (!cap_issubset(*permitted, old->cap_permitted))
262 		return -EPERM;
263 
264 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
265 	if (!cap_issubset(*effective, *permitted))
266 		return -EPERM;
267 
268 	new->cap_effective   = *effective;
269 	new->cap_inheritable = *inheritable;
270 	new->cap_permitted   = *permitted;
271 
272 	/*
273 	 * Mask off ambient bits that are no longer both permitted and
274 	 * inheritable.
275 	 */
276 	new->cap_ambient = cap_intersect(new->cap_ambient,
277 					 cap_intersect(*permitted,
278 						       *inheritable));
279 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
280 		return -EINVAL;
281 	return 0;
282 }
283 
284 /**
285  * cap_inode_need_killpriv - Determine if inode change affects privileges
286  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
287  *
288  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
289  * affects the security markings on that inode, and if it is, should
290  * inode_killpriv() be invoked or the change rejected.
291  *
292  * Returns 1 if security.capability has a value, meaning inode_killpriv()
293  * is required, 0 otherwise, meaning inode_killpriv() is not required.
294  */
295 int cap_inode_need_killpriv(struct dentry *dentry)
296 {
297 	struct inode *inode = d_backing_inode(dentry);
298 	int error;
299 
300 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
301 	return error > 0;
302 }
303 
304 /**
305  * cap_inode_killpriv - Erase the security markings on an inode
306  * @dentry: The inode/dentry to alter
307  *
308  * Erase the privilege-enhancing security markings on an inode.
309  *
310  * Returns 0 if successful, -ve on error.
311  */
312 int cap_inode_killpriv(struct dentry *dentry)
313 {
314 	int error;
315 
316 	error = __vfs_removexattr(dentry, XATTR_NAME_CAPS);
317 	if (error == -EOPNOTSUPP)
318 		error = 0;
319 	return error;
320 }
321 
322 static bool rootid_owns_currentns(kuid_t kroot)
323 {
324 	struct user_namespace *ns;
325 
326 	if (!uid_valid(kroot))
327 		return false;
328 
329 	for (ns = current_user_ns(); ; ns = ns->parent) {
330 		if (from_kuid(ns, kroot) == 0)
331 			return true;
332 		if (ns == &init_user_ns)
333 			break;
334 	}
335 
336 	return false;
337 }
338 
339 static __u32 sansflags(__u32 m)
340 {
341 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
342 }
343 
344 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
345 {
346 	if (size != XATTR_CAPS_SZ_2)
347 		return false;
348 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
349 }
350 
351 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
352 {
353 	if (size != XATTR_CAPS_SZ_3)
354 		return false;
355 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
356 }
357 
358 /*
359  * getsecurity: We are called for security.* before any attempt to read the
360  * xattr from the inode itself.
361  *
362  * This gives us a chance to read the on-disk value and convert it.  If we
363  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
364  *
365  * Note we are not called by vfs_getxattr_alloc(), but that is only called
366  * by the integrity subsystem, which really wants the unconverted values -
367  * so that's good.
368  */
369 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
370 			  bool alloc)
371 {
372 	int size, ret;
373 	kuid_t kroot;
374 	uid_t root, mappedroot;
375 	char *tmpbuf = NULL;
376 	struct vfs_cap_data *cap;
377 	struct vfs_ns_cap_data *nscap;
378 	struct dentry *dentry;
379 	struct user_namespace *fs_ns;
380 
381 	if (strcmp(name, "capability") != 0)
382 		return -EOPNOTSUPP;
383 
384 	dentry = d_find_any_alias(inode);
385 	if (!dentry)
386 		return -EINVAL;
387 
388 	size = sizeof(struct vfs_ns_cap_data);
389 	ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS,
390 				 &tmpbuf, size, GFP_NOFS);
391 	dput(dentry);
392 
393 	if (ret < 0)
394 		return ret;
395 
396 	fs_ns = inode->i_sb->s_user_ns;
397 	cap = (struct vfs_cap_data *) tmpbuf;
398 	if (is_v2header((size_t) ret, cap)) {
399 		/* If this is sizeof(vfs_cap_data) then we're ok with the
400 		 * on-disk value, so return that.  */
401 		if (alloc)
402 			*buffer = tmpbuf;
403 		else
404 			kfree(tmpbuf);
405 		return ret;
406 	} else if (!is_v3header((size_t) ret, cap)) {
407 		kfree(tmpbuf);
408 		return -EINVAL;
409 	}
410 
411 	nscap = (struct vfs_ns_cap_data *) tmpbuf;
412 	root = le32_to_cpu(nscap->rootid);
413 	kroot = make_kuid(fs_ns, root);
414 
415 	/* If the root kuid maps to a valid uid in current ns, then return
416 	 * this as a nscap. */
417 	mappedroot = from_kuid(current_user_ns(), kroot);
418 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
419 		if (alloc) {
420 			*buffer = tmpbuf;
421 			nscap->rootid = cpu_to_le32(mappedroot);
422 		} else
423 			kfree(tmpbuf);
424 		return size;
425 	}
426 
427 	if (!rootid_owns_currentns(kroot)) {
428 		kfree(tmpbuf);
429 		return -EOPNOTSUPP;
430 	}
431 
432 	/* This comes from a parent namespace.  Return as a v2 capability */
433 	size = sizeof(struct vfs_cap_data);
434 	if (alloc) {
435 		*buffer = kmalloc(size, GFP_ATOMIC);
436 		if (*buffer) {
437 			struct vfs_cap_data *cap = *buffer;
438 			__le32 nsmagic, magic;
439 			magic = VFS_CAP_REVISION_2;
440 			nsmagic = le32_to_cpu(nscap->magic_etc);
441 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
442 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
443 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
444 			cap->magic_etc = cpu_to_le32(magic);
445 		} else {
446 			size = -ENOMEM;
447 		}
448 	}
449 	kfree(tmpbuf);
450 	return size;
451 }
452 
453 static kuid_t rootid_from_xattr(const void *value, size_t size,
454 				struct user_namespace *task_ns)
455 {
456 	const struct vfs_ns_cap_data *nscap = value;
457 	uid_t rootid = 0;
458 
459 	if (size == XATTR_CAPS_SZ_3)
460 		rootid = le32_to_cpu(nscap->rootid);
461 
462 	return make_kuid(task_ns, rootid);
463 }
464 
465 static bool validheader(size_t size, const struct vfs_cap_data *cap)
466 {
467 	return is_v2header(size, cap) || is_v3header(size, cap);
468 }
469 
470 /*
471  * User requested a write of security.capability.  If needed, update the
472  * xattr to change from v2 to v3, or to fixup the v3 rootid.
473  *
474  * If all is ok, we return the new size, on error return < 0.
475  */
476 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size)
477 {
478 	struct vfs_ns_cap_data *nscap;
479 	uid_t nsrootid;
480 	const struct vfs_cap_data *cap = *ivalue;
481 	__u32 magic, nsmagic;
482 	struct inode *inode = d_backing_inode(dentry);
483 	struct user_namespace *task_ns = current_user_ns(),
484 		*fs_ns = inode->i_sb->s_user_ns;
485 	kuid_t rootid;
486 	size_t newsize;
487 
488 	if (!*ivalue)
489 		return -EINVAL;
490 	if (!validheader(size, cap))
491 		return -EINVAL;
492 	if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
493 		return -EPERM;
494 	if (size == XATTR_CAPS_SZ_2)
495 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
496 			/* user is privileged, just write the v2 */
497 			return size;
498 
499 	rootid = rootid_from_xattr(*ivalue, size, task_ns);
500 	if (!uid_valid(rootid))
501 		return -EINVAL;
502 
503 	nsrootid = from_kuid(fs_ns, rootid);
504 	if (nsrootid == -1)
505 		return -EINVAL;
506 
507 	newsize = sizeof(struct vfs_ns_cap_data);
508 	nscap = kmalloc(newsize, GFP_ATOMIC);
509 	if (!nscap)
510 		return -ENOMEM;
511 	nscap->rootid = cpu_to_le32(nsrootid);
512 	nsmagic = VFS_CAP_REVISION_3;
513 	magic = le32_to_cpu(cap->magic_etc);
514 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
515 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
516 	nscap->magic_etc = cpu_to_le32(nsmagic);
517 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
518 
519 	kvfree(*ivalue);
520 	*ivalue = nscap;
521 	return newsize;
522 }
523 
524 /*
525  * Calculate the new process capability sets from the capability sets attached
526  * to a file.
527  */
528 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
529 					  struct linux_binprm *bprm,
530 					  bool *effective,
531 					  bool *has_fcap)
532 {
533 	struct cred *new = bprm->cred;
534 	unsigned i;
535 	int ret = 0;
536 
537 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
538 		*effective = true;
539 
540 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
541 		*has_fcap = true;
542 
543 	CAP_FOR_EACH_U32(i) {
544 		__u32 permitted = caps->permitted.cap[i];
545 		__u32 inheritable = caps->inheritable.cap[i];
546 
547 		/*
548 		 * pP' = (X & fP) | (pI & fI)
549 		 * The addition of pA' is handled later.
550 		 */
551 		new->cap_permitted.cap[i] =
552 			(new->cap_bset.cap[i] & permitted) |
553 			(new->cap_inheritable.cap[i] & inheritable);
554 
555 		if (permitted & ~new->cap_permitted.cap[i])
556 			/* insufficient to execute correctly */
557 			ret = -EPERM;
558 	}
559 
560 	/*
561 	 * For legacy apps, with no internal support for recognizing they
562 	 * do not have enough capabilities, we return an error if they are
563 	 * missing some "forced" (aka file-permitted) capabilities.
564 	 */
565 	return *effective ? ret : 0;
566 }
567 
568 /*
569  * Extract the on-exec-apply capability sets for an executable file.
570  */
571 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
572 {
573 	struct inode *inode = d_backing_inode(dentry);
574 	__u32 magic_etc;
575 	unsigned tocopy, i;
576 	int size;
577 	struct vfs_ns_cap_data data, *nscaps = &data;
578 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
579 	kuid_t rootkuid;
580 	struct user_namespace *fs_ns;
581 
582 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
583 
584 	if (!inode)
585 		return -ENODATA;
586 
587 	fs_ns = inode->i_sb->s_user_ns;
588 	size = __vfs_getxattr((struct dentry *)dentry, inode,
589 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
590 	if (size == -ENODATA || size == -EOPNOTSUPP)
591 		/* no data, that's ok */
592 		return -ENODATA;
593 
594 	if (size < 0)
595 		return size;
596 
597 	if (size < sizeof(magic_etc))
598 		return -EINVAL;
599 
600 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
601 
602 	rootkuid = make_kuid(fs_ns, 0);
603 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
604 	case VFS_CAP_REVISION_1:
605 		if (size != XATTR_CAPS_SZ_1)
606 			return -EINVAL;
607 		tocopy = VFS_CAP_U32_1;
608 		break;
609 	case VFS_CAP_REVISION_2:
610 		if (size != XATTR_CAPS_SZ_2)
611 			return -EINVAL;
612 		tocopy = VFS_CAP_U32_2;
613 		break;
614 	case VFS_CAP_REVISION_3:
615 		if (size != XATTR_CAPS_SZ_3)
616 			return -EINVAL;
617 		tocopy = VFS_CAP_U32_3;
618 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
619 		break;
620 
621 	default:
622 		return -EINVAL;
623 	}
624 	/* Limit the caps to the mounter of the filesystem
625 	 * or the more limited uid specified in the xattr.
626 	 */
627 	if (!rootid_owns_currentns(rootkuid))
628 		return -ENODATA;
629 
630 	CAP_FOR_EACH_U32(i) {
631 		if (i >= tocopy)
632 			break;
633 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
634 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
635 	}
636 
637 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
638 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
639 
640 	cpu_caps->rootid = rootkuid;
641 
642 	return 0;
643 }
644 
645 /*
646  * Attempt to get the on-exec apply capability sets for an executable file from
647  * its xattrs and, if present, apply them to the proposed credentials being
648  * constructed by execve().
649  */
650 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
651 			 bool *effective, bool *has_fcap)
652 {
653 	int rc = 0;
654 	struct cpu_vfs_cap_data vcaps;
655 
656 	cap_clear(bprm->cred->cap_permitted);
657 
658 	if (!file_caps_enabled)
659 		return 0;
660 
661 	if (!mnt_may_suid(file->f_path.mnt))
662 		return 0;
663 
664 	/*
665 	 * This check is redundant with mnt_may_suid() but is kept to make
666 	 * explicit that capability bits are limited to s_user_ns and its
667 	 * descendants.
668 	 */
669 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
670 		return 0;
671 
672 	rc = get_vfs_caps_from_disk(file->f_path.dentry, &vcaps);
673 	if (rc < 0) {
674 		if (rc == -EINVAL)
675 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
676 					bprm->filename);
677 		else if (rc == -ENODATA)
678 			rc = 0;
679 		goto out;
680 	}
681 
682 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
683 
684 out:
685 	if (rc)
686 		cap_clear(bprm->cred->cap_permitted);
687 
688 	return rc;
689 }
690 
691 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
692 
693 static inline bool __is_real(kuid_t uid, struct cred *cred)
694 { return uid_eq(cred->uid, uid); }
695 
696 static inline bool __is_eff(kuid_t uid, struct cred *cred)
697 { return uid_eq(cred->euid, uid); }
698 
699 static inline bool __is_suid(kuid_t uid, struct cred *cred)
700 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
701 
702 /*
703  * handle_privileged_root - Handle case of privileged root
704  * @bprm: The execution parameters, including the proposed creds
705  * @has_fcap: Are any file capabilities set?
706  * @effective: Do we have effective root privilege?
707  * @root_uid: This namespace' root UID WRT initial USER namespace
708  *
709  * Handle the case where root is privileged and hasn't been neutered by
710  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
711  * set UID root and nothing is changed.  If we are root, cap_permitted is
712  * updated.  If we have become set UID root, the effective bit is set.
713  */
714 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
715 				   bool *effective, kuid_t root_uid)
716 {
717 	const struct cred *old = current_cred();
718 	struct cred *new = bprm->cred;
719 
720 	if (!root_privileged())
721 		return;
722 	/*
723 	 * If the legacy file capability is set, then don't set privs
724 	 * for a setuid root binary run by a non-root user.  Do set it
725 	 * for a root user just to cause least surprise to an admin.
726 	 */
727 	if (has_fcap && __is_suid(root_uid, new)) {
728 		warn_setuid_and_fcaps_mixed(bprm->filename);
729 		return;
730 	}
731 	/*
732 	 * To support inheritance of root-permissions and suid-root
733 	 * executables under compatibility mode, we override the
734 	 * capability sets for the file.
735 	 */
736 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
737 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
738 		new->cap_permitted = cap_combine(old->cap_bset,
739 						 old->cap_inheritable);
740 	}
741 	/*
742 	 * If only the real uid is 0, we do not set the effective bit.
743 	 */
744 	if (__is_eff(root_uid, new))
745 		*effective = true;
746 }
747 
748 #define __cap_gained(field, target, source) \
749 	!cap_issubset(target->cap_##field, source->cap_##field)
750 #define __cap_grew(target, source, cred) \
751 	!cap_issubset(cred->cap_##target, cred->cap_##source)
752 #define __cap_full(field, cred) \
753 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
754 
755 static inline bool __is_setuid(struct cred *new, const struct cred *old)
756 { return !uid_eq(new->euid, old->uid); }
757 
758 static inline bool __is_setgid(struct cred *new, const struct cred *old)
759 { return !gid_eq(new->egid, old->gid); }
760 
761 /*
762  * 1) Audit candidate if current->cap_effective is set
763  *
764  * We do not bother to audit if 3 things are true:
765  *   1) cap_effective has all caps
766  *   2) we became root *OR* are were already root
767  *   3) root is supposed to have all caps (SECURE_NOROOT)
768  * Since this is just a normal root execing a process.
769  *
770  * Number 1 above might fail if you don't have a full bset, but I think
771  * that is interesting information to audit.
772  *
773  * A number of other conditions require logging:
774  * 2) something prevented setuid root getting all caps
775  * 3) non-setuid root gets fcaps
776  * 4) non-setuid root gets ambient
777  */
778 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
779 				     kuid_t root, bool has_fcap)
780 {
781 	bool ret = false;
782 
783 	if ((__cap_grew(effective, ambient, new) &&
784 	     !(__cap_full(effective, new) &&
785 	       (__is_eff(root, new) || __is_real(root, new)) &&
786 	       root_privileged())) ||
787 	    (root_privileged() &&
788 	     __is_suid(root, new) &&
789 	     !__cap_full(effective, new)) ||
790 	    (!__is_setuid(new, old) &&
791 	     ((has_fcap &&
792 	       __cap_gained(permitted, new, old)) ||
793 	      __cap_gained(ambient, new, old))))
794 
795 		ret = true;
796 
797 	return ret;
798 }
799 
800 /**
801  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
802  * @bprm: The execution parameters, including the proposed creds
803  * @file: The file to pull the credentials from
804  *
805  * Set up the proposed credentials for a new execution context being
806  * constructed by execve().  The proposed creds in @bprm->cred is altered,
807  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
808  */
809 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
810 {
811 	/* Process setpcap binaries and capabilities for uid 0 */
812 	const struct cred *old = current_cred();
813 	struct cred *new = bprm->cred;
814 	bool effective = false, has_fcap = false, is_setid;
815 	int ret;
816 	kuid_t root_uid;
817 
818 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
819 		return -EPERM;
820 
821 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
822 	if (ret < 0)
823 		return ret;
824 
825 	root_uid = make_kuid(new->user_ns, 0);
826 
827 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
828 
829 	/* if we have fs caps, clear dangerous personality flags */
830 	if (__cap_gained(permitted, new, old))
831 		bprm->per_clear |= PER_CLEAR_ON_SETID;
832 
833 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
834 	 * credentials unless they have the appropriate permit.
835 	 *
836 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
837 	 */
838 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
839 
840 	if ((is_setid || __cap_gained(permitted, new, old)) &&
841 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
842 	     !ptracer_capable(current, new->user_ns))) {
843 		/* downgrade; they get no more than they had, and maybe less */
844 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
845 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
846 			new->euid = new->uid;
847 			new->egid = new->gid;
848 		}
849 		new->cap_permitted = cap_intersect(new->cap_permitted,
850 						   old->cap_permitted);
851 	}
852 
853 	new->suid = new->fsuid = new->euid;
854 	new->sgid = new->fsgid = new->egid;
855 
856 	/* File caps or setid cancels ambient. */
857 	if (has_fcap || is_setid)
858 		cap_clear(new->cap_ambient);
859 
860 	/*
861 	 * Now that we've computed pA', update pP' to give:
862 	 *   pP' = (X & fP) | (pI & fI) | pA'
863 	 */
864 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
865 
866 	/*
867 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
868 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
869 	 */
870 	if (effective)
871 		new->cap_effective = new->cap_permitted;
872 	else
873 		new->cap_effective = new->cap_ambient;
874 
875 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
876 		return -EPERM;
877 
878 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
879 		ret = audit_log_bprm_fcaps(bprm, new, old);
880 		if (ret < 0)
881 			return ret;
882 	}
883 
884 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
885 
886 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
887 		return -EPERM;
888 
889 	/* Check for privilege-elevated exec. */
890 	if (is_setid ||
891 	    (!__is_real(root_uid, new) &&
892 	     (effective ||
893 	      __cap_grew(permitted, ambient, new))))
894 		bprm->secureexec = 1;
895 
896 	return 0;
897 }
898 
899 /**
900  * cap_inode_setxattr - Determine whether an xattr may be altered
901  * @dentry: The inode/dentry being altered
902  * @name: The name of the xattr to be changed
903  * @value: The value that the xattr will be changed to
904  * @size: The size of value
905  * @flags: The replacement flag
906  *
907  * Determine whether an xattr may be altered or set on an inode, returning 0 if
908  * permission is granted, -ve if denied.
909  *
910  * This is used to make sure security xattrs don't get updated or set by those
911  * who aren't privileged to do so.
912  */
913 int cap_inode_setxattr(struct dentry *dentry, const char *name,
914 		       const void *value, size_t size, int flags)
915 {
916 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
917 
918 	/* Ignore non-security xattrs */
919 	if (strncmp(name, XATTR_SECURITY_PREFIX,
920 			XATTR_SECURITY_PREFIX_LEN) != 0)
921 		return 0;
922 
923 	/*
924 	 * For XATTR_NAME_CAPS the check will be done in
925 	 * cap_convert_nscap(), called by setxattr()
926 	 */
927 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
928 		return 0;
929 
930 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
931 		return -EPERM;
932 	return 0;
933 }
934 
935 /**
936  * cap_inode_removexattr - Determine whether an xattr may be removed
937  * @dentry: The inode/dentry being altered
938  * @name: The name of the xattr to be changed
939  *
940  * Determine whether an xattr may be removed from an inode, returning 0 if
941  * permission is granted, -ve if denied.
942  *
943  * This is used to make sure security xattrs don't get removed by those who
944  * aren't privileged to remove them.
945  */
946 int cap_inode_removexattr(struct dentry *dentry, const char *name)
947 {
948 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
949 
950 	/* Ignore non-security xattrs */
951 	if (strncmp(name, XATTR_SECURITY_PREFIX,
952 			XATTR_SECURITY_PREFIX_LEN) != 0)
953 		return 0;
954 
955 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
956 		/* security.capability gets namespaced */
957 		struct inode *inode = d_backing_inode(dentry);
958 		if (!inode)
959 			return -EINVAL;
960 		if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
961 			return -EPERM;
962 		return 0;
963 	}
964 
965 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
966 		return -EPERM;
967 	return 0;
968 }
969 
970 /*
971  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
972  * a process after a call to setuid, setreuid, or setresuid.
973  *
974  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
975  *  {r,e,s}uid != 0, the permitted and effective capabilities are
976  *  cleared.
977  *
978  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
979  *  capabilities of the process are cleared.
980  *
981  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
982  *  capabilities are set to the permitted capabilities.
983  *
984  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
985  *  never happen.
986  *
987  *  -astor
988  *
989  * cevans - New behaviour, Oct '99
990  * A process may, via prctl(), elect to keep its capabilities when it
991  * calls setuid() and switches away from uid==0. Both permitted and
992  * effective sets will be retained.
993  * Without this change, it was impossible for a daemon to drop only some
994  * of its privilege. The call to setuid(!=0) would drop all privileges!
995  * Keeping uid 0 is not an option because uid 0 owns too many vital
996  * files..
997  * Thanks to Olaf Kirch and Peter Benie for spotting this.
998  */
999 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1000 {
1001 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1002 
1003 	if ((uid_eq(old->uid, root_uid) ||
1004 	     uid_eq(old->euid, root_uid) ||
1005 	     uid_eq(old->suid, root_uid)) &&
1006 	    (!uid_eq(new->uid, root_uid) &&
1007 	     !uid_eq(new->euid, root_uid) &&
1008 	     !uid_eq(new->suid, root_uid))) {
1009 		if (!issecure(SECURE_KEEP_CAPS)) {
1010 			cap_clear(new->cap_permitted);
1011 			cap_clear(new->cap_effective);
1012 		}
1013 
1014 		/*
1015 		 * Pre-ambient programs expect setresuid to nonroot followed
1016 		 * by exec to drop capabilities.  We should make sure that
1017 		 * this remains the case.
1018 		 */
1019 		cap_clear(new->cap_ambient);
1020 	}
1021 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1022 		cap_clear(new->cap_effective);
1023 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1024 		new->cap_effective = new->cap_permitted;
1025 }
1026 
1027 /**
1028  * cap_task_fix_setuid - Fix up the results of setuid() call
1029  * @new: The proposed credentials
1030  * @old: The current task's current credentials
1031  * @flags: Indications of what has changed
1032  *
1033  * Fix up the results of setuid() call before the credential changes are
1034  * actually applied, returning 0 to grant the changes, -ve to deny them.
1035  */
1036 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1037 {
1038 	switch (flags) {
1039 	case LSM_SETID_RE:
1040 	case LSM_SETID_ID:
1041 	case LSM_SETID_RES:
1042 		/* juggle the capabilities to follow [RES]UID changes unless
1043 		 * otherwise suppressed */
1044 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1045 			cap_emulate_setxuid(new, old);
1046 		break;
1047 
1048 	case LSM_SETID_FS:
1049 		/* juggle the capabilties to follow FSUID changes, unless
1050 		 * otherwise suppressed
1051 		 *
1052 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1053 		 *          if not, we might be a bit too harsh here.
1054 		 */
1055 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1056 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1057 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1058 				new->cap_effective =
1059 					cap_drop_fs_set(new->cap_effective);
1060 
1061 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1062 				new->cap_effective =
1063 					cap_raise_fs_set(new->cap_effective,
1064 							 new->cap_permitted);
1065 		}
1066 		break;
1067 
1068 	default:
1069 		return -EINVAL;
1070 	}
1071 
1072 	return 0;
1073 }
1074 
1075 /*
1076  * Rationale: code calling task_setscheduler, task_setioprio, and
1077  * task_setnice, assumes that
1078  *   . if capable(cap_sys_nice), then those actions should be allowed
1079  *   . if not capable(cap_sys_nice), but acting on your own processes,
1080  *   	then those actions should be allowed
1081  * This is insufficient now since you can call code without suid, but
1082  * yet with increased caps.
1083  * So we check for increased caps on the target process.
1084  */
1085 static int cap_safe_nice(struct task_struct *p)
1086 {
1087 	int is_subset, ret = 0;
1088 
1089 	rcu_read_lock();
1090 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1091 				 current_cred()->cap_permitted);
1092 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1093 		ret = -EPERM;
1094 	rcu_read_unlock();
1095 
1096 	return ret;
1097 }
1098 
1099 /**
1100  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1101  * @p: The task to affect
1102  *
1103  * Detemine if the requested scheduler policy change is permitted for the
1104  * specified task, returning 0 if permission is granted, -ve if denied.
1105  */
1106 int cap_task_setscheduler(struct task_struct *p)
1107 {
1108 	return cap_safe_nice(p);
1109 }
1110 
1111 /**
1112  * cap_task_ioprio - Detemine if I/O priority change is permitted
1113  * @p: The task to affect
1114  * @ioprio: The I/O priority to set
1115  *
1116  * Detemine if the requested I/O priority change is permitted for the specified
1117  * task, returning 0 if permission is granted, -ve if denied.
1118  */
1119 int cap_task_setioprio(struct task_struct *p, int ioprio)
1120 {
1121 	return cap_safe_nice(p);
1122 }
1123 
1124 /**
1125  * cap_task_ioprio - Detemine if task priority change is permitted
1126  * @p: The task to affect
1127  * @nice: The nice value to set
1128  *
1129  * Detemine if the requested task priority change is permitted for the
1130  * specified task, returning 0 if permission is granted, -ve if denied.
1131  */
1132 int cap_task_setnice(struct task_struct *p, int nice)
1133 {
1134 	return cap_safe_nice(p);
1135 }
1136 
1137 /*
1138  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1139  * the current task's bounding set.  Returns 0 on success, -ve on error.
1140  */
1141 static int cap_prctl_drop(unsigned long cap)
1142 {
1143 	struct cred *new;
1144 
1145 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1146 		return -EPERM;
1147 	if (!cap_valid(cap))
1148 		return -EINVAL;
1149 
1150 	new = prepare_creds();
1151 	if (!new)
1152 		return -ENOMEM;
1153 	cap_lower(new->cap_bset, cap);
1154 	return commit_creds(new);
1155 }
1156 
1157 /**
1158  * cap_task_prctl - Implement process control functions for this security module
1159  * @option: The process control function requested
1160  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1161  *
1162  * Allow process control functions (sys_prctl()) to alter capabilities; may
1163  * also deny access to other functions not otherwise implemented here.
1164  *
1165  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1166  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1167  * modules will consider performing the function.
1168  */
1169 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1170 		   unsigned long arg4, unsigned long arg5)
1171 {
1172 	const struct cred *old = current_cred();
1173 	struct cred *new;
1174 
1175 	switch (option) {
1176 	case PR_CAPBSET_READ:
1177 		if (!cap_valid(arg2))
1178 			return -EINVAL;
1179 		return !!cap_raised(old->cap_bset, arg2);
1180 
1181 	case PR_CAPBSET_DROP:
1182 		return cap_prctl_drop(arg2);
1183 
1184 	/*
1185 	 * The next four prctl's remain to assist with transitioning a
1186 	 * system from legacy UID=0 based privilege (when filesystem
1187 	 * capabilities are not in use) to a system using filesystem
1188 	 * capabilities only - as the POSIX.1e draft intended.
1189 	 *
1190 	 * Note:
1191 	 *
1192 	 *  PR_SET_SECUREBITS =
1193 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1194 	 *    | issecure_mask(SECURE_NOROOT)
1195 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1196 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1197 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1198 	 *
1199 	 * will ensure that the current process and all of its
1200 	 * children will be locked into a pure
1201 	 * capability-based-privilege environment.
1202 	 */
1203 	case PR_SET_SECUREBITS:
1204 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1205 		     & (old->securebits ^ arg2))			/*[1]*/
1206 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1207 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1208 		    || (cap_capable(current_cred(),
1209 				    current_cred()->user_ns,
1210 				    CAP_SETPCAP,
1211 				    CAP_OPT_NONE) != 0)			/*[4]*/
1212 			/*
1213 			 * [1] no changing of bits that are locked
1214 			 * [2] no unlocking of locks
1215 			 * [3] no setting of unsupported bits
1216 			 * [4] doing anything requires privilege (go read about
1217 			 *     the "sendmail capabilities bug")
1218 			 */
1219 		    )
1220 			/* cannot change a locked bit */
1221 			return -EPERM;
1222 
1223 		new = prepare_creds();
1224 		if (!new)
1225 			return -ENOMEM;
1226 		new->securebits = arg2;
1227 		return commit_creds(new);
1228 
1229 	case PR_GET_SECUREBITS:
1230 		return old->securebits;
1231 
1232 	case PR_GET_KEEPCAPS:
1233 		return !!issecure(SECURE_KEEP_CAPS);
1234 
1235 	case PR_SET_KEEPCAPS:
1236 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1237 			return -EINVAL;
1238 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1239 			return -EPERM;
1240 
1241 		new = prepare_creds();
1242 		if (!new)
1243 			return -ENOMEM;
1244 		if (arg2)
1245 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1246 		else
1247 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1248 		return commit_creds(new);
1249 
1250 	case PR_CAP_AMBIENT:
1251 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1252 			if (arg3 | arg4 | arg5)
1253 				return -EINVAL;
1254 
1255 			new = prepare_creds();
1256 			if (!new)
1257 				return -ENOMEM;
1258 			cap_clear(new->cap_ambient);
1259 			return commit_creds(new);
1260 		}
1261 
1262 		if (((!cap_valid(arg3)) | arg4 | arg5))
1263 			return -EINVAL;
1264 
1265 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1266 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1267 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1268 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1269 			return -EINVAL;
1270 		} else {
1271 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1272 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1273 			     !cap_raised(current_cred()->cap_inheritable,
1274 					 arg3) ||
1275 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1276 				return -EPERM;
1277 
1278 			new = prepare_creds();
1279 			if (!new)
1280 				return -ENOMEM;
1281 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1282 				cap_raise(new->cap_ambient, arg3);
1283 			else
1284 				cap_lower(new->cap_ambient, arg3);
1285 			return commit_creds(new);
1286 		}
1287 
1288 	default:
1289 		/* No functionality available - continue with default */
1290 		return -ENOSYS;
1291 	}
1292 }
1293 
1294 /**
1295  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1296  * @mm: The VM space in which the new mapping is to be made
1297  * @pages: The size of the mapping
1298  *
1299  * Determine whether the allocation of a new virtual mapping by the current
1300  * task is permitted, returning 1 if permission is granted, 0 if not.
1301  */
1302 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1303 {
1304 	int cap_sys_admin = 0;
1305 
1306 	if (cap_capable(current_cred(), &init_user_ns,
1307 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1308 		cap_sys_admin = 1;
1309 
1310 	return cap_sys_admin;
1311 }
1312 
1313 /*
1314  * cap_mmap_addr - check if able to map given addr
1315  * @addr: address attempting to be mapped
1316  *
1317  * If the process is attempting to map memory below dac_mmap_min_addr they need
1318  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1319  * capability security module.  Returns 0 if this mapping should be allowed
1320  * -EPERM if not.
1321  */
1322 int cap_mmap_addr(unsigned long addr)
1323 {
1324 	int ret = 0;
1325 
1326 	if (addr < dac_mmap_min_addr) {
1327 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1328 				  CAP_OPT_NONE);
1329 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1330 		if (ret == 0)
1331 			current->flags |= PF_SUPERPRIV;
1332 	}
1333 	return ret;
1334 }
1335 
1336 int cap_mmap_file(struct file *file, unsigned long reqprot,
1337 		  unsigned long prot, unsigned long flags)
1338 {
1339 	return 0;
1340 }
1341 
1342 #ifdef CONFIG_SECURITY
1343 
1344 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1345 	LSM_HOOK_INIT(capable, cap_capable),
1346 	LSM_HOOK_INIT(settime, cap_settime),
1347 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1348 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1349 	LSM_HOOK_INIT(capget, cap_capget),
1350 	LSM_HOOK_INIT(capset, cap_capset),
1351 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1352 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1353 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1354 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1355 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1356 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1357 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1358 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1359 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1360 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1361 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1362 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1363 };
1364 
1365 static int __init capability_init(void)
1366 {
1367 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1368 				"capability");
1369 	return 0;
1370 }
1371 
1372 DEFINE_LSM(capability) = {
1373 	.name = "capability",
1374 	.order = LSM_ORDER_FIRST,
1375 	.init = capability_init,
1376 };
1377 
1378 #endif /* CONFIG_SECURITY */
1379