xref: /openbmc/linux/security/commoncap.c (revision 6abeae2a)
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, const 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 	*ivalue = nscap;
520 	return newsize;
521 }
522 
523 /*
524  * Calculate the new process capability sets from the capability sets attached
525  * to a file.
526  */
527 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
528 					  struct linux_binprm *bprm,
529 					  bool *effective,
530 					  bool *has_fcap)
531 {
532 	struct cred *new = bprm->cred;
533 	unsigned i;
534 	int ret = 0;
535 
536 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
537 		*effective = true;
538 
539 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
540 		*has_fcap = true;
541 
542 	CAP_FOR_EACH_U32(i) {
543 		__u32 permitted = caps->permitted.cap[i];
544 		__u32 inheritable = caps->inheritable.cap[i];
545 
546 		/*
547 		 * pP' = (X & fP) | (pI & fI)
548 		 * The addition of pA' is handled later.
549 		 */
550 		new->cap_permitted.cap[i] =
551 			(new->cap_bset.cap[i] & permitted) |
552 			(new->cap_inheritable.cap[i] & inheritable);
553 
554 		if (permitted & ~new->cap_permitted.cap[i])
555 			/* insufficient to execute correctly */
556 			ret = -EPERM;
557 	}
558 
559 	/*
560 	 * For legacy apps, with no internal support for recognizing they
561 	 * do not have enough capabilities, we return an error if they are
562 	 * missing some "forced" (aka file-permitted) capabilities.
563 	 */
564 	return *effective ? ret : 0;
565 }
566 
567 /*
568  * Extract the on-exec-apply capability sets for an executable file.
569  */
570 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
571 {
572 	struct inode *inode = d_backing_inode(dentry);
573 	__u32 magic_etc;
574 	unsigned tocopy, i;
575 	int size;
576 	struct vfs_ns_cap_data data, *nscaps = &data;
577 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
578 	kuid_t rootkuid;
579 	struct user_namespace *fs_ns;
580 
581 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
582 
583 	if (!inode)
584 		return -ENODATA;
585 
586 	fs_ns = inode->i_sb->s_user_ns;
587 	size = __vfs_getxattr((struct dentry *)dentry, inode,
588 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
589 	if (size == -ENODATA || size == -EOPNOTSUPP)
590 		/* no data, that's ok */
591 		return -ENODATA;
592 
593 	if (size < 0)
594 		return size;
595 
596 	if (size < sizeof(magic_etc))
597 		return -EINVAL;
598 
599 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
600 
601 	rootkuid = make_kuid(fs_ns, 0);
602 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
603 	case VFS_CAP_REVISION_1:
604 		if (size != XATTR_CAPS_SZ_1)
605 			return -EINVAL;
606 		tocopy = VFS_CAP_U32_1;
607 		break;
608 	case VFS_CAP_REVISION_2:
609 		if (size != XATTR_CAPS_SZ_2)
610 			return -EINVAL;
611 		tocopy = VFS_CAP_U32_2;
612 		break;
613 	case VFS_CAP_REVISION_3:
614 		if (size != XATTR_CAPS_SZ_3)
615 			return -EINVAL;
616 		tocopy = VFS_CAP_U32_3;
617 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
618 		break;
619 
620 	default:
621 		return -EINVAL;
622 	}
623 	/* Limit the caps to the mounter of the filesystem
624 	 * or the more limited uid specified in the xattr.
625 	 */
626 	if (!rootid_owns_currentns(rootkuid))
627 		return -ENODATA;
628 
629 	CAP_FOR_EACH_U32(i) {
630 		if (i >= tocopy)
631 			break;
632 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
633 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
634 	}
635 
636 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
637 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
638 
639 	cpu_caps->rootid = rootkuid;
640 
641 	return 0;
642 }
643 
644 /*
645  * Attempt to get the on-exec apply capability sets for an executable file from
646  * its xattrs and, if present, apply them to the proposed credentials being
647  * constructed by execve().
648  */
649 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
650 			 bool *effective, bool *has_fcap)
651 {
652 	int rc = 0;
653 	struct cpu_vfs_cap_data vcaps;
654 
655 	cap_clear(bprm->cred->cap_permitted);
656 
657 	if (!file_caps_enabled)
658 		return 0;
659 
660 	if (!mnt_may_suid(file->f_path.mnt))
661 		return 0;
662 
663 	/*
664 	 * This check is redundant with mnt_may_suid() but is kept to make
665 	 * explicit that capability bits are limited to s_user_ns and its
666 	 * descendants.
667 	 */
668 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
669 		return 0;
670 
671 	rc = get_vfs_caps_from_disk(file->f_path.dentry, &vcaps);
672 	if (rc < 0) {
673 		if (rc == -EINVAL)
674 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
675 					bprm->filename);
676 		else if (rc == -ENODATA)
677 			rc = 0;
678 		goto out;
679 	}
680 
681 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
682 
683 out:
684 	if (rc)
685 		cap_clear(bprm->cred->cap_permitted);
686 
687 	return rc;
688 }
689 
690 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
691 
692 static inline bool __is_real(kuid_t uid, struct cred *cred)
693 { return uid_eq(cred->uid, uid); }
694 
695 static inline bool __is_eff(kuid_t uid, struct cred *cred)
696 { return uid_eq(cred->euid, uid); }
697 
698 static inline bool __is_suid(kuid_t uid, struct cred *cred)
699 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
700 
701 /*
702  * handle_privileged_root - Handle case of privileged root
703  * @bprm: The execution parameters, including the proposed creds
704  * @has_fcap: Are any file capabilities set?
705  * @effective: Do we have effective root privilege?
706  * @root_uid: This namespace' root UID WRT initial USER namespace
707  *
708  * Handle the case where root is privileged and hasn't been neutered by
709  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
710  * set UID root and nothing is changed.  If we are root, cap_permitted is
711  * updated.  If we have become set UID root, the effective bit is set.
712  */
713 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
714 				   bool *effective, kuid_t root_uid)
715 {
716 	const struct cred *old = current_cred();
717 	struct cred *new = bprm->cred;
718 
719 	if (!root_privileged())
720 		return;
721 	/*
722 	 * If the legacy file capability is set, then don't set privs
723 	 * for a setuid root binary run by a non-root user.  Do set it
724 	 * for a root user just to cause least surprise to an admin.
725 	 */
726 	if (has_fcap && __is_suid(root_uid, new)) {
727 		warn_setuid_and_fcaps_mixed(bprm->filename);
728 		return;
729 	}
730 	/*
731 	 * To support inheritance of root-permissions and suid-root
732 	 * executables under compatibility mode, we override the
733 	 * capability sets for the file.
734 	 */
735 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
736 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
737 		new->cap_permitted = cap_combine(old->cap_bset,
738 						 old->cap_inheritable);
739 	}
740 	/*
741 	 * If only the real uid is 0, we do not set the effective bit.
742 	 */
743 	if (__is_eff(root_uid, new))
744 		*effective = true;
745 }
746 
747 #define __cap_gained(field, target, source) \
748 	!cap_issubset(target->cap_##field, source->cap_##field)
749 #define __cap_grew(target, source, cred) \
750 	!cap_issubset(cred->cap_##target, cred->cap_##source)
751 #define __cap_full(field, cred) \
752 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
753 
754 static inline bool __is_setuid(struct cred *new, const struct cred *old)
755 { return !uid_eq(new->euid, old->uid); }
756 
757 static inline bool __is_setgid(struct cred *new, const struct cred *old)
758 { return !gid_eq(new->egid, old->gid); }
759 
760 /*
761  * 1) Audit candidate if current->cap_effective is set
762  *
763  * We do not bother to audit if 3 things are true:
764  *   1) cap_effective has all caps
765  *   2) we became root *OR* are were already root
766  *   3) root is supposed to have all caps (SECURE_NOROOT)
767  * Since this is just a normal root execing a process.
768  *
769  * Number 1 above might fail if you don't have a full bset, but I think
770  * that is interesting information to audit.
771  *
772  * A number of other conditions require logging:
773  * 2) something prevented setuid root getting all caps
774  * 3) non-setuid root gets fcaps
775  * 4) non-setuid root gets ambient
776  */
777 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
778 				     kuid_t root, bool has_fcap)
779 {
780 	bool ret = false;
781 
782 	if ((__cap_grew(effective, ambient, new) &&
783 	     !(__cap_full(effective, new) &&
784 	       (__is_eff(root, new) || __is_real(root, new)) &&
785 	       root_privileged())) ||
786 	    (root_privileged() &&
787 	     __is_suid(root, new) &&
788 	     !__cap_full(effective, new)) ||
789 	    (!__is_setuid(new, old) &&
790 	     ((has_fcap &&
791 	       __cap_gained(permitted, new, old)) ||
792 	      __cap_gained(ambient, new, old))))
793 
794 		ret = true;
795 
796 	return ret;
797 }
798 
799 /**
800  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
801  * @bprm: The execution parameters, including the proposed creds
802  * @file: The file to pull the credentials from
803  *
804  * Set up the proposed credentials for a new execution context being
805  * constructed by execve().  The proposed creds in @bprm->cred is altered,
806  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
807  */
808 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
809 {
810 	/* Process setpcap binaries and capabilities for uid 0 */
811 	const struct cred *old = current_cred();
812 	struct cred *new = bprm->cred;
813 	bool effective = false, has_fcap = false, is_setid;
814 	int ret;
815 	kuid_t root_uid;
816 
817 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
818 		return -EPERM;
819 
820 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
821 	if (ret < 0)
822 		return ret;
823 
824 	root_uid = make_kuid(new->user_ns, 0);
825 
826 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
827 
828 	/* if we have fs caps, clear dangerous personality flags */
829 	if (__cap_gained(permitted, new, old))
830 		bprm->per_clear |= PER_CLEAR_ON_SETID;
831 
832 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
833 	 * credentials unless they have the appropriate permit.
834 	 *
835 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
836 	 */
837 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
838 
839 	if ((is_setid || __cap_gained(permitted, new, old)) &&
840 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
841 	     !ptracer_capable(current, new->user_ns))) {
842 		/* downgrade; they get no more than they had, and maybe less */
843 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
844 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
845 			new->euid = new->uid;
846 			new->egid = new->gid;
847 		}
848 		new->cap_permitted = cap_intersect(new->cap_permitted,
849 						   old->cap_permitted);
850 	}
851 
852 	new->suid = new->fsuid = new->euid;
853 	new->sgid = new->fsgid = new->egid;
854 
855 	/* File caps or setid cancels ambient. */
856 	if (has_fcap || is_setid)
857 		cap_clear(new->cap_ambient);
858 
859 	/*
860 	 * Now that we've computed pA', update pP' to give:
861 	 *   pP' = (X & fP) | (pI & fI) | pA'
862 	 */
863 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
864 
865 	/*
866 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
867 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
868 	 */
869 	if (effective)
870 		new->cap_effective = new->cap_permitted;
871 	else
872 		new->cap_effective = new->cap_ambient;
873 
874 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
875 		return -EPERM;
876 
877 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
878 		ret = audit_log_bprm_fcaps(bprm, new, old);
879 		if (ret < 0)
880 			return ret;
881 	}
882 
883 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
884 
885 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
886 		return -EPERM;
887 
888 	/* Check for privilege-elevated exec. */
889 	if (is_setid ||
890 	    (!__is_real(root_uid, new) &&
891 	     (effective ||
892 	      __cap_grew(permitted, ambient, new))))
893 		bprm->secureexec = 1;
894 
895 	return 0;
896 }
897 
898 /**
899  * cap_inode_setxattr - Determine whether an xattr may be altered
900  * @dentry: The inode/dentry being altered
901  * @name: The name of the xattr to be changed
902  * @value: The value that the xattr will be changed to
903  * @size: The size of value
904  * @flags: The replacement flag
905  *
906  * Determine whether an xattr may be altered or set on an inode, returning 0 if
907  * permission is granted, -ve if denied.
908  *
909  * This is used to make sure security xattrs don't get updated or set by those
910  * who aren't privileged to do so.
911  */
912 int cap_inode_setxattr(struct dentry *dentry, const char *name,
913 		       const void *value, size_t size, int flags)
914 {
915 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
916 
917 	/* Ignore non-security xattrs */
918 	if (strncmp(name, XATTR_SECURITY_PREFIX,
919 			XATTR_SECURITY_PREFIX_LEN) != 0)
920 		return 0;
921 
922 	/*
923 	 * For XATTR_NAME_CAPS the check will be done in
924 	 * cap_convert_nscap(), called by setxattr()
925 	 */
926 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
927 		return 0;
928 
929 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
930 		return -EPERM;
931 	return 0;
932 }
933 
934 /**
935  * cap_inode_removexattr - Determine whether an xattr may be removed
936  * @dentry: The inode/dentry being altered
937  * @name: The name of the xattr to be changed
938  *
939  * Determine whether an xattr may be removed from an inode, returning 0 if
940  * permission is granted, -ve if denied.
941  *
942  * This is used to make sure security xattrs don't get removed by those who
943  * aren't privileged to remove them.
944  */
945 int cap_inode_removexattr(struct dentry *dentry, const char *name)
946 {
947 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
948 
949 	/* Ignore non-security xattrs */
950 	if (strncmp(name, XATTR_SECURITY_PREFIX,
951 			XATTR_SECURITY_PREFIX_LEN) != 0)
952 		return 0;
953 
954 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
955 		/* security.capability gets namespaced */
956 		struct inode *inode = d_backing_inode(dentry);
957 		if (!inode)
958 			return -EINVAL;
959 		if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
960 			return -EPERM;
961 		return 0;
962 	}
963 
964 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
965 		return -EPERM;
966 	return 0;
967 }
968 
969 /*
970  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
971  * a process after a call to setuid, setreuid, or setresuid.
972  *
973  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
974  *  {r,e,s}uid != 0, the permitted and effective capabilities are
975  *  cleared.
976  *
977  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
978  *  capabilities of the process are cleared.
979  *
980  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
981  *  capabilities are set to the permitted capabilities.
982  *
983  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
984  *  never happen.
985  *
986  *  -astor
987  *
988  * cevans - New behaviour, Oct '99
989  * A process may, via prctl(), elect to keep its capabilities when it
990  * calls setuid() and switches away from uid==0. Both permitted and
991  * effective sets will be retained.
992  * Without this change, it was impossible for a daemon to drop only some
993  * of its privilege. The call to setuid(!=0) would drop all privileges!
994  * Keeping uid 0 is not an option because uid 0 owns too many vital
995  * files..
996  * Thanks to Olaf Kirch and Peter Benie for spotting this.
997  */
998 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
999 {
1000 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1001 
1002 	if ((uid_eq(old->uid, root_uid) ||
1003 	     uid_eq(old->euid, root_uid) ||
1004 	     uid_eq(old->suid, root_uid)) &&
1005 	    (!uid_eq(new->uid, root_uid) &&
1006 	     !uid_eq(new->euid, root_uid) &&
1007 	     !uid_eq(new->suid, root_uid))) {
1008 		if (!issecure(SECURE_KEEP_CAPS)) {
1009 			cap_clear(new->cap_permitted);
1010 			cap_clear(new->cap_effective);
1011 		}
1012 
1013 		/*
1014 		 * Pre-ambient programs expect setresuid to nonroot followed
1015 		 * by exec to drop capabilities.  We should make sure that
1016 		 * this remains the case.
1017 		 */
1018 		cap_clear(new->cap_ambient);
1019 	}
1020 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1021 		cap_clear(new->cap_effective);
1022 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1023 		new->cap_effective = new->cap_permitted;
1024 }
1025 
1026 /**
1027  * cap_task_fix_setuid - Fix up the results of setuid() call
1028  * @new: The proposed credentials
1029  * @old: The current task's current credentials
1030  * @flags: Indications of what has changed
1031  *
1032  * Fix up the results of setuid() call before the credential changes are
1033  * actually applied, returning 0 to grant the changes, -ve to deny them.
1034  */
1035 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1036 {
1037 	switch (flags) {
1038 	case LSM_SETID_RE:
1039 	case LSM_SETID_ID:
1040 	case LSM_SETID_RES:
1041 		/* juggle the capabilities to follow [RES]UID changes unless
1042 		 * otherwise suppressed */
1043 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1044 			cap_emulate_setxuid(new, old);
1045 		break;
1046 
1047 	case LSM_SETID_FS:
1048 		/* juggle the capabilties to follow FSUID changes, unless
1049 		 * otherwise suppressed
1050 		 *
1051 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1052 		 *          if not, we might be a bit too harsh here.
1053 		 */
1054 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1055 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1056 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1057 				new->cap_effective =
1058 					cap_drop_fs_set(new->cap_effective);
1059 
1060 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1061 				new->cap_effective =
1062 					cap_raise_fs_set(new->cap_effective,
1063 							 new->cap_permitted);
1064 		}
1065 		break;
1066 
1067 	default:
1068 		return -EINVAL;
1069 	}
1070 
1071 	return 0;
1072 }
1073 
1074 /*
1075  * Rationale: code calling task_setscheduler, task_setioprio, and
1076  * task_setnice, assumes that
1077  *   . if capable(cap_sys_nice), then those actions should be allowed
1078  *   . if not capable(cap_sys_nice), but acting on your own processes,
1079  *   	then those actions should be allowed
1080  * This is insufficient now since you can call code without suid, but
1081  * yet with increased caps.
1082  * So we check for increased caps on the target process.
1083  */
1084 static int cap_safe_nice(struct task_struct *p)
1085 {
1086 	int is_subset, ret = 0;
1087 
1088 	rcu_read_lock();
1089 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1090 				 current_cred()->cap_permitted);
1091 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1092 		ret = -EPERM;
1093 	rcu_read_unlock();
1094 
1095 	return ret;
1096 }
1097 
1098 /**
1099  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1100  * @p: The task to affect
1101  *
1102  * Detemine if the requested scheduler policy change is permitted for the
1103  * specified task, returning 0 if permission is granted, -ve if denied.
1104  */
1105 int cap_task_setscheduler(struct task_struct *p)
1106 {
1107 	return cap_safe_nice(p);
1108 }
1109 
1110 /**
1111  * cap_task_ioprio - Detemine if I/O priority change is permitted
1112  * @p: The task to affect
1113  * @ioprio: The I/O priority to set
1114  *
1115  * Detemine if the requested I/O priority change is permitted for the specified
1116  * task, returning 0 if permission is granted, -ve if denied.
1117  */
1118 int cap_task_setioprio(struct task_struct *p, int ioprio)
1119 {
1120 	return cap_safe_nice(p);
1121 }
1122 
1123 /**
1124  * cap_task_ioprio - Detemine if task priority change is permitted
1125  * @p: The task to affect
1126  * @nice: The nice value to set
1127  *
1128  * Detemine if the requested task priority change is permitted for the
1129  * specified task, returning 0 if permission is granted, -ve if denied.
1130  */
1131 int cap_task_setnice(struct task_struct *p, int nice)
1132 {
1133 	return cap_safe_nice(p);
1134 }
1135 
1136 /*
1137  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1138  * the current task's bounding set.  Returns 0 on success, -ve on error.
1139  */
1140 static int cap_prctl_drop(unsigned long cap)
1141 {
1142 	struct cred *new;
1143 
1144 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1145 		return -EPERM;
1146 	if (!cap_valid(cap))
1147 		return -EINVAL;
1148 
1149 	new = prepare_creds();
1150 	if (!new)
1151 		return -ENOMEM;
1152 	cap_lower(new->cap_bset, cap);
1153 	return commit_creds(new);
1154 }
1155 
1156 /**
1157  * cap_task_prctl - Implement process control functions for this security module
1158  * @option: The process control function requested
1159  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1160  *
1161  * Allow process control functions (sys_prctl()) to alter capabilities; may
1162  * also deny access to other functions not otherwise implemented here.
1163  *
1164  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1165  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1166  * modules will consider performing the function.
1167  */
1168 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1169 		   unsigned long arg4, unsigned long arg5)
1170 {
1171 	const struct cred *old = current_cred();
1172 	struct cred *new;
1173 
1174 	switch (option) {
1175 	case PR_CAPBSET_READ:
1176 		if (!cap_valid(arg2))
1177 			return -EINVAL;
1178 		return !!cap_raised(old->cap_bset, arg2);
1179 
1180 	case PR_CAPBSET_DROP:
1181 		return cap_prctl_drop(arg2);
1182 
1183 	/*
1184 	 * The next four prctl's remain to assist with transitioning a
1185 	 * system from legacy UID=0 based privilege (when filesystem
1186 	 * capabilities are not in use) to a system using filesystem
1187 	 * capabilities only - as the POSIX.1e draft intended.
1188 	 *
1189 	 * Note:
1190 	 *
1191 	 *  PR_SET_SECUREBITS =
1192 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1193 	 *    | issecure_mask(SECURE_NOROOT)
1194 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1195 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1196 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1197 	 *
1198 	 * will ensure that the current process and all of its
1199 	 * children will be locked into a pure
1200 	 * capability-based-privilege environment.
1201 	 */
1202 	case PR_SET_SECUREBITS:
1203 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1204 		     & (old->securebits ^ arg2))			/*[1]*/
1205 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1206 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1207 		    || (cap_capable(current_cred(),
1208 				    current_cred()->user_ns,
1209 				    CAP_SETPCAP,
1210 				    CAP_OPT_NONE) != 0)			/*[4]*/
1211 			/*
1212 			 * [1] no changing of bits that are locked
1213 			 * [2] no unlocking of locks
1214 			 * [3] no setting of unsupported bits
1215 			 * [4] doing anything requires privilege (go read about
1216 			 *     the "sendmail capabilities bug")
1217 			 */
1218 		    )
1219 			/* cannot change a locked bit */
1220 			return -EPERM;
1221 
1222 		new = prepare_creds();
1223 		if (!new)
1224 			return -ENOMEM;
1225 		new->securebits = arg2;
1226 		return commit_creds(new);
1227 
1228 	case PR_GET_SECUREBITS:
1229 		return old->securebits;
1230 
1231 	case PR_GET_KEEPCAPS:
1232 		return !!issecure(SECURE_KEEP_CAPS);
1233 
1234 	case PR_SET_KEEPCAPS:
1235 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1236 			return -EINVAL;
1237 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1238 			return -EPERM;
1239 
1240 		new = prepare_creds();
1241 		if (!new)
1242 			return -ENOMEM;
1243 		if (arg2)
1244 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1245 		else
1246 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1247 		return commit_creds(new);
1248 
1249 	case PR_CAP_AMBIENT:
1250 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1251 			if (arg3 | arg4 | arg5)
1252 				return -EINVAL;
1253 
1254 			new = prepare_creds();
1255 			if (!new)
1256 				return -ENOMEM;
1257 			cap_clear(new->cap_ambient);
1258 			return commit_creds(new);
1259 		}
1260 
1261 		if (((!cap_valid(arg3)) | arg4 | arg5))
1262 			return -EINVAL;
1263 
1264 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1265 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1266 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1267 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1268 			return -EINVAL;
1269 		} else {
1270 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1271 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1272 			     !cap_raised(current_cred()->cap_inheritable,
1273 					 arg3) ||
1274 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1275 				return -EPERM;
1276 
1277 			new = prepare_creds();
1278 			if (!new)
1279 				return -ENOMEM;
1280 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1281 				cap_raise(new->cap_ambient, arg3);
1282 			else
1283 				cap_lower(new->cap_ambient, arg3);
1284 			return commit_creds(new);
1285 		}
1286 
1287 	default:
1288 		/* No functionality available - continue with default */
1289 		return -ENOSYS;
1290 	}
1291 }
1292 
1293 /**
1294  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1295  * @mm: The VM space in which the new mapping is to be made
1296  * @pages: The size of the mapping
1297  *
1298  * Determine whether the allocation of a new virtual mapping by the current
1299  * task is permitted, returning 1 if permission is granted, 0 if not.
1300  */
1301 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1302 {
1303 	int cap_sys_admin = 0;
1304 
1305 	if (cap_capable(current_cred(), &init_user_ns,
1306 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1307 		cap_sys_admin = 1;
1308 
1309 	return cap_sys_admin;
1310 }
1311 
1312 /*
1313  * cap_mmap_addr - check if able to map given addr
1314  * @addr: address attempting to be mapped
1315  *
1316  * If the process is attempting to map memory below dac_mmap_min_addr they need
1317  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1318  * capability security module.  Returns 0 if this mapping should be allowed
1319  * -EPERM if not.
1320  */
1321 int cap_mmap_addr(unsigned long addr)
1322 {
1323 	int ret = 0;
1324 
1325 	if (addr < dac_mmap_min_addr) {
1326 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1327 				  CAP_OPT_NONE);
1328 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1329 		if (ret == 0)
1330 			current->flags |= PF_SUPERPRIV;
1331 	}
1332 	return ret;
1333 }
1334 
1335 int cap_mmap_file(struct file *file, unsigned long reqprot,
1336 		  unsigned long prot, unsigned long flags)
1337 {
1338 	return 0;
1339 }
1340 
1341 #ifdef CONFIG_SECURITY
1342 
1343 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1344 	LSM_HOOK_INIT(capable, cap_capable),
1345 	LSM_HOOK_INIT(settime, cap_settime),
1346 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1347 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1348 	LSM_HOOK_INIT(capget, cap_capget),
1349 	LSM_HOOK_INIT(capset, cap_capset),
1350 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1351 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1352 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1353 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1354 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1355 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1356 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1357 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1358 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1359 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1360 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1361 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1362 };
1363 
1364 static int __init capability_init(void)
1365 {
1366 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1367 				"capability");
1368 	return 0;
1369 }
1370 
1371 DEFINE_LSM(capability) = {
1372 	.name = "capability",
1373 	.order = LSM_ORDER_FIRST,
1374 	.init = capability_init,
1375 };
1376 
1377 #endif /* CONFIG_SECURITY */
1378