xref: /openbmc/linux/security/commoncap.c (revision f9a82c48)
1 /* Common capabilities, needed by capability.o.
2  *
3  *	This program is free software; you can redistribute it and/or modify
4  *	it under the terms of the GNU General Public License as published by
5  *	the Free Software Foundation; either version 2 of the License, or
6  *	(at your option) any later version.
7  *
8  */
9 
10 #include <linux/capability.h>
11 #include <linux/audit.h>
12 #include <linux/init.h>
13 #include <linux/kernel.h>
14 #include <linux/lsm_hooks.h>
15 #include <linux/file.h>
16 #include <linux/mm.h>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
20 #include <linux/skbuff.h>
21 #include <linux/netlink.h>
22 #include <linux/ptrace.h>
23 #include <linux/xattr.h>
24 #include <linux/hugetlb.h>
25 #include <linux/mount.h>
26 #include <linux/sched.h>
27 #include <linux/prctl.h>
28 #include <linux/securebits.h>
29 #include <linux/user_namespace.h>
30 #include <linux/binfmts.h>
31 #include <linux/personality.h>
32 
33 /*
34  * If a non-root user executes a setuid-root binary in
35  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
36  * However if fE is also set, then the intent is for only
37  * the file capabilities to be applied, and the setuid-root
38  * bit is left on either to change the uid (plausible) or
39  * to get full privilege on a kernel without file capabilities
40  * support.  So in that case we do not raise capabilities.
41  *
42  * Warn if that happens, once per boot.
43  */
44 static void warn_setuid_and_fcaps_mixed(const char *fname)
45 {
46 	static int warned;
47 	if (!warned) {
48 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
49 			" effective capabilities. Therefore not raising all"
50 			" capabilities.\n", fname);
51 		warned = 1;
52 	}
53 }
54 
55 /**
56  * cap_capable - Determine whether a task has a particular effective capability
57  * @cred: The credentials to use
58  * @ns:  The user namespace in which we need the capability
59  * @cap: The capability to check for
60  * @opts: Bitmask of options defined in include/linux/security.h
61  *
62  * Determine whether the nominated task has the specified capability amongst
63  * its effective set, returning 0 if it does, -ve if it does not.
64  *
65  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
66  * and has_capability() functions.  That is, it has the reverse semantics:
67  * cap_has_capability() returns 0 when a task has a capability, but the
68  * kernel's capable() and has_capability() returns 1 for this case.
69  */
70 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
71 		int cap, unsigned int opts)
72 {
73 	struct user_namespace *ns = targ_ns;
74 
75 	/* See if cred has the capability in the target user namespace
76 	 * by examining the target user namespace and all of the target
77 	 * user namespace's parents.
78 	 */
79 	for (;;) {
80 		/* Do we have the necessary capabilities? */
81 		if (ns == cred->user_ns)
82 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
83 
84 		/*
85 		 * If we're already at a lower level than we're looking for,
86 		 * we're done searching.
87 		 */
88 		if (ns->level <= cred->user_ns->level)
89 			return -EPERM;
90 
91 		/*
92 		 * The owner of the user namespace in the parent of the
93 		 * user namespace has all caps.
94 		 */
95 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
96 			return 0;
97 
98 		/*
99 		 * If you have a capability in a parent user ns, then you have
100 		 * it over all children user namespaces as well.
101 		 */
102 		ns = ns->parent;
103 	}
104 
105 	/* We never get here */
106 }
107 
108 /**
109  * cap_settime - Determine whether the current process may set the system clock
110  * @ts: The time to set
111  * @tz: The timezone to set
112  *
113  * Determine whether the current process may set the system clock and timezone
114  * information, returning 0 if permission granted, -ve if denied.
115  */
116 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
117 {
118 	if (!capable(CAP_SYS_TIME))
119 		return -EPERM;
120 	return 0;
121 }
122 
123 /**
124  * cap_ptrace_access_check - Determine whether the current process may access
125  *			   another
126  * @child: The process to be accessed
127  * @mode: The mode of attachment.
128  *
129  * If we are in the same or an ancestor user_ns and have all the target
130  * task's capabilities, then ptrace access is allowed.
131  * If we have the ptrace capability to the target user_ns, then ptrace
132  * access is allowed.
133  * Else denied.
134  *
135  * Determine whether a process may access another, returning 0 if permission
136  * granted, -ve if denied.
137  */
138 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
139 {
140 	int ret = 0;
141 	const struct cred *cred, *child_cred;
142 	const kernel_cap_t *caller_caps;
143 
144 	rcu_read_lock();
145 	cred = current_cred();
146 	child_cred = __task_cred(child);
147 	if (mode & PTRACE_MODE_FSCREDS)
148 		caller_caps = &cred->cap_effective;
149 	else
150 		caller_caps = &cred->cap_permitted;
151 	if (cred->user_ns == child_cred->user_ns &&
152 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
153 		goto out;
154 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
155 		goto out;
156 	ret = -EPERM;
157 out:
158 	rcu_read_unlock();
159 	return ret;
160 }
161 
162 /**
163  * cap_ptrace_traceme - Determine whether another process may trace the current
164  * @parent: The task proposed to be the tracer
165  *
166  * If parent is in the same or an ancestor user_ns and has all current's
167  * capabilities, then ptrace access is allowed.
168  * If parent has the ptrace capability to current's user_ns, then ptrace
169  * access is allowed.
170  * Else denied.
171  *
172  * Determine whether the nominated task is permitted to trace the current
173  * process, returning 0 if permission is granted, -ve if denied.
174  */
175 int cap_ptrace_traceme(struct task_struct *parent)
176 {
177 	int ret = 0;
178 	const struct cred *cred, *child_cred;
179 
180 	rcu_read_lock();
181 	cred = __task_cred(parent);
182 	child_cred = current_cred();
183 	if (cred->user_ns == child_cred->user_ns &&
184 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
185 		goto out;
186 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
187 		goto out;
188 	ret = -EPERM;
189 out:
190 	rcu_read_unlock();
191 	return ret;
192 }
193 
194 /**
195  * cap_capget - Retrieve a task's capability sets
196  * @target: The task from which to retrieve the capability sets
197  * @effective: The place to record the effective set
198  * @inheritable: The place to record the inheritable set
199  * @permitted: The place to record the permitted set
200  *
201  * This function retrieves the capabilities of the nominated task and returns
202  * them to the caller.
203  */
204 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
205 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
206 {
207 	const struct cred *cred;
208 
209 	/* Derived from kernel/capability.c:sys_capget. */
210 	rcu_read_lock();
211 	cred = __task_cred(target);
212 	*effective   = cred->cap_effective;
213 	*inheritable = cred->cap_inheritable;
214 	*permitted   = cred->cap_permitted;
215 	rcu_read_unlock();
216 	return 0;
217 }
218 
219 /*
220  * Determine whether the inheritable capabilities are limited to the old
221  * permitted set.  Returns 1 if they are limited, 0 if they are not.
222  */
223 static inline int cap_inh_is_capped(void)
224 {
225 	/* they are so limited unless the current task has the CAP_SETPCAP
226 	 * capability
227 	 */
228 	if (cap_capable(current_cred(), current_cred()->user_ns,
229 			CAP_SETPCAP, CAP_OPT_NONE) == 0)
230 		return 0;
231 	return 1;
232 }
233 
234 /**
235  * cap_capset - Validate and apply proposed changes to current's capabilities
236  * @new: The proposed new credentials; alterations should be made here
237  * @old: The current task's current credentials
238  * @effective: A pointer to the proposed new effective capabilities set
239  * @inheritable: A pointer to the proposed new inheritable capabilities set
240  * @permitted: A pointer to the proposed new permitted capabilities set
241  *
242  * This function validates and applies a proposed mass change to the current
243  * process's capability sets.  The changes are made to the proposed new
244  * credentials, and assuming no error, will be committed by the caller of LSM.
245  */
246 int cap_capset(struct cred *new,
247 	       const struct cred *old,
248 	       const kernel_cap_t *effective,
249 	       const kernel_cap_t *inheritable,
250 	       const kernel_cap_t *permitted)
251 {
252 	if (cap_inh_is_capped() &&
253 	    !cap_issubset(*inheritable,
254 			  cap_combine(old->cap_inheritable,
255 				      old->cap_permitted)))
256 		/* incapable of using this inheritable set */
257 		return -EPERM;
258 
259 	if (!cap_issubset(*inheritable,
260 			  cap_combine(old->cap_inheritable,
261 				      old->cap_bset)))
262 		/* no new pI capabilities outside bounding set */
263 		return -EPERM;
264 
265 	/* verify restrictions on target's new Permitted set */
266 	if (!cap_issubset(*permitted, old->cap_permitted))
267 		return -EPERM;
268 
269 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
270 	if (!cap_issubset(*effective, *permitted))
271 		return -EPERM;
272 
273 	new->cap_effective   = *effective;
274 	new->cap_inheritable = *inheritable;
275 	new->cap_permitted   = *permitted;
276 
277 	/*
278 	 * Mask off ambient bits that are no longer both permitted and
279 	 * inheritable.
280 	 */
281 	new->cap_ambient = cap_intersect(new->cap_ambient,
282 					 cap_intersect(*permitted,
283 						       *inheritable));
284 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
285 		return -EINVAL;
286 	return 0;
287 }
288 
289 /**
290  * cap_inode_need_killpriv - Determine if inode change affects privileges
291  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
292  *
293  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
294  * affects the security markings on that inode, and if it is, should
295  * inode_killpriv() be invoked or the change rejected.
296  *
297  * Returns 1 if security.capability has a value, meaning inode_killpriv()
298  * is required, 0 otherwise, meaning inode_killpriv() is not required.
299  */
300 int cap_inode_need_killpriv(struct dentry *dentry)
301 {
302 	struct inode *inode = d_backing_inode(dentry);
303 	int error;
304 
305 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
306 	return error > 0;
307 }
308 
309 /**
310  * cap_inode_killpriv - Erase the security markings on an inode
311  * @dentry: The inode/dentry to alter
312  *
313  * Erase the privilege-enhancing security markings on an inode.
314  *
315  * Returns 0 if successful, -ve on error.
316  */
317 int cap_inode_killpriv(struct dentry *dentry)
318 {
319 	int error;
320 
321 	error = __vfs_removexattr(dentry, XATTR_NAME_CAPS);
322 	if (error == -EOPNOTSUPP)
323 		error = 0;
324 	return error;
325 }
326 
327 static bool rootid_owns_currentns(kuid_t kroot)
328 {
329 	struct user_namespace *ns;
330 
331 	if (!uid_valid(kroot))
332 		return false;
333 
334 	for (ns = current_user_ns(); ; ns = ns->parent) {
335 		if (from_kuid(ns, kroot) == 0)
336 			return true;
337 		if (ns == &init_user_ns)
338 			break;
339 	}
340 
341 	return false;
342 }
343 
344 static __u32 sansflags(__u32 m)
345 {
346 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
347 }
348 
349 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
350 {
351 	if (size != XATTR_CAPS_SZ_2)
352 		return false;
353 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
354 }
355 
356 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
357 {
358 	if (size != XATTR_CAPS_SZ_3)
359 		return false;
360 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
361 }
362 
363 /*
364  * getsecurity: We are called for security.* before any attempt to read the
365  * xattr from the inode itself.
366  *
367  * This gives us a chance to read the on-disk value and convert it.  If we
368  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
369  *
370  * Note we are not called by vfs_getxattr_alloc(), but that is only called
371  * by the integrity subsystem, which really wants the unconverted values -
372  * so that's good.
373  */
374 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
375 			  bool alloc)
376 {
377 	int size, ret;
378 	kuid_t kroot;
379 	uid_t root, mappedroot;
380 	char *tmpbuf = NULL;
381 	struct vfs_cap_data *cap;
382 	struct vfs_ns_cap_data *nscap;
383 	struct dentry *dentry;
384 	struct user_namespace *fs_ns;
385 
386 	if (strcmp(name, "capability") != 0)
387 		return -EOPNOTSUPP;
388 
389 	dentry = d_find_any_alias(inode);
390 	if (!dentry)
391 		return -EINVAL;
392 
393 	size = sizeof(struct vfs_ns_cap_data);
394 	ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS,
395 				 &tmpbuf, size, GFP_NOFS);
396 	dput(dentry);
397 
398 	if (ret < 0)
399 		return ret;
400 
401 	fs_ns = inode->i_sb->s_user_ns;
402 	cap = (struct vfs_cap_data *) tmpbuf;
403 	if (is_v2header((size_t) ret, cap)) {
404 		/* If this is sizeof(vfs_cap_data) then we're ok with the
405 		 * on-disk value, so return that.  */
406 		if (alloc)
407 			*buffer = tmpbuf;
408 		else
409 			kfree(tmpbuf);
410 		return ret;
411 	} else if (!is_v3header((size_t) ret, cap)) {
412 		kfree(tmpbuf);
413 		return -EINVAL;
414 	}
415 
416 	nscap = (struct vfs_ns_cap_data *) tmpbuf;
417 	root = le32_to_cpu(nscap->rootid);
418 	kroot = make_kuid(fs_ns, root);
419 
420 	/* If the root kuid maps to a valid uid in current ns, then return
421 	 * this as a nscap. */
422 	mappedroot = from_kuid(current_user_ns(), kroot);
423 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
424 		if (alloc) {
425 			*buffer = tmpbuf;
426 			nscap->rootid = cpu_to_le32(mappedroot);
427 		} else
428 			kfree(tmpbuf);
429 		return size;
430 	}
431 
432 	if (!rootid_owns_currentns(kroot)) {
433 		kfree(tmpbuf);
434 		return -EOPNOTSUPP;
435 	}
436 
437 	/* This comes from a parent namespace.  Return as a v2 capability */
438 	size = sizeof(struct vfs_cap_data);
439 	if (alloc) {
440 		*buffer = kmalloc(size, GFP_ATOMIC);
441 		if (*buffer) {
442 			struct vfs_cap_data *cap = *buffer;
443 			__le32 nsmagic, magic;
444 			magic = VFS_CAP_REVISION_2;
445 			nsmagic = le32_to_cpu(nscap->magic_etc);
446 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
447 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
448 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
449 			cap->magic_etc = cpu_to_le32(magic);
450 		} else {
451 			size = -ENOMEM;
452 		}
453 	}
454 	kfree(tmpbuf);
455 	return size;
456 }
457 
458 static kuid_t rootid_from_xattr(const void *value, size_t size,
459 				struct user_namespace *task_ns)
460 {
461 	const struct vfs_ns_cap_data *nscap = value;
462 	uid_t rootid = 0;
463 
464 	if (size == XATTR_CAPS_SZ_3)
465 		rootid = le32_to_cpu(nscap->rootid);
466 
467 	return make_kuid(task_ns, rootid);
468 }
469 
470 static bool validheader(size_t size, const struct vfs_cap_data *cap)
471 {
472 	return is_v2header(size, cap) || is_v3header(size, cap);
473 }
474 
475 /*
476  * User requested a write of security.capability.  If needed, update the
477  * xattr to change from v2 to v3, or to fixup the v3 rootid.
478  *
479  * If all is ok, we return the new size, on error return < 0.
480  */
481 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size)
482 {
483 	struct vfs_ns_cap_data *nscap;
484 	uid_t nsrootid;
485 	const struct vfs_cap_data *cap = *ivalue;
486 	__u32 magic, nsmagic;
487 	struct inode *inode = d_backing_inode(dentry);
488 	struct user_namespace *task_ns = current_user_ns(),
489 		*fs_ns = inode->i_sb->s_user_ns;
490 	kuid_t rootid;
491 	size_t newsize;
492 
493 	if (!*ivalue)
494 		return -EINVAL;
495 	if (!validheader(size, cap))
496 		return -EINVAL;
497 	if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
498 		return -EPERM;
499 	if (size == XATTR_CAPS_SZ_2)
500 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
501 			/* user is privileged, just write the v2 */
502 			return size;
503 
504 	rootid = rootid_from_xattr(*ivalue, size, task_ns);
505 	if (!uid_valid(rootid))
506 		return -EINVAL;
507 
508 	nsrootid = from_kuid(fs_ns, rootid);
509 	if (nsrootid == -1)
510 		return -EINVAL;
511 
512 	newsize = sizeof(struct vfs_ns_cap_data);
513 	nscap = kmalloc(newsize, GFP_ATOMIC);
514 	if (!nscap)
515 		return -ENOMEM;
516 	nscap->rootid = cpu_to_le32(nsrootid);
517 	nsmagic = VFS_CAP_REVISION_3;
518 	magic = le32_to_cpu(cap->magic_etc);
519 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
520 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
521 	nscap->magic_etc = cpu_to_le32(nsmagic);
522 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
523 
524 	kvfree(*ivalue);
525 	*ivalue = nscap;
526 	return newsize;
527 }
528 
529 /*
530  * Calculate the new process capability sets from the capability sets attached
531  * to a file.
532  */
533 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
534 					  struct linux_binprm *bprm,
535 					  bool *effective,
536 					  bool *has_fcap)
537 {
538 	struct cred *new = bprm->cred;
539 	unsigned i;
540 	int ret = 0;
541 
542 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
543 		*effective = true;
544 
545 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
546 		*has_fcap = true;
547 
548 	CAP_FOR_EACH_U32(i) {
549 		__u32 permitted = caps->permitted.cap[i];
550 		__u32 inheritable = caps->inheritable.cap[i];
551 
552 		/*
553 		 * pP' = (X & fP) | (pI & fI)
554 		 * The addition of pA' is handled later.
555 		 */
556 		new->cap_permitted.cap[i] =
557 			(new->cap_bset.cap[i] & permitted) |
558 			(new->cap_inheritable.cap[i] & inheritable);
559 
560 		if (permitted & ~new->cap_permitted.cap[i])
561 			/* insufficient to execute correctly */
562 			ret = -EPERM;
563 	}
564 
565 	/*
566 	 * For legacy apps, with no internal support for recognizing they
567 	 * do not have enough capabilities, we return an error if they are
568 	 * missing some "forced" (aka file-permitted) capabilities.
569 	 */
570 	return *effective ? ret : 0;
571 }
572 
573 /*
574  * Extract the on-exec-apply capability sets for an executable file.
575  */
576 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
577 {
578 	struct inode *inode = d_backing_inode(dentry);
579 	__u32 magic_etc;
580 	unsigned tocopy, i;
581 	int size;
582 	struct vfs_ns_cap_data data, *nscaps = &data;
583 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
584 	kuid_t rootkuid;
585 	struct user_namespace *fs_ns;
586 
587 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
588 
589 	if (!inode)
590 		return -ENODATA;
591 
592 	fs_ns = inode->i_sb->s_user_ns;
593 	size = __vfs_getxattr((struct dentry *)dentry, inode,
594 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
595 	if (size == -ENODATA || size == -EOPNOTSUPP)
596 		/* no data, that's ok */
597 		return -ENODATA;
598 
599 	if (size < 0)
600 		return size;
601 
602 	if (size < sizeof(magic_etc))
603 		return -EINVAL;
604 
605 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
606 
607 	rootkuid = make_kuid(fs_ns, 0);
608 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
609 	case VFS_CAP_REVISION_1:
610 		if (size != XATTR_CAPS_SZ_1)
611 			return -EINVAL;
612 		tocopy = VFS_CAP_U32_1;
613 		break;
614 	case VFS_CAP_REVISION_2:
615 		if (size != XATTR_CAPS_SZ_2)
616 			return -EINVAL;
617 		tocopy = VFS_CAP_U32_2;
618 		break;
619 	case VFS_CAP_REVISION_3:
620 		if (size != XATTR_CAPS_SZ_3)
621 			return -EINVAL;
622 		tocopy = VFS_CAP_U32_3;
623 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
624 		break;
625 
626 	default:
627 		return -EINVAL;
628 	}
629 	/* Limit the caps to the mounter of the filesystem
630 	 * or the more limited uid specified in the xattr.
631 	 */
632 	if (!rootid_owns_currentns(rootkuid))
633 		return -ENODATA;
634 
635 	CAP_FOR_EACH_U32(i) {
636 		if (i >= tocopy)
637 			break;
638 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
639 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
640 	}
641 
642 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
643 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
644 
645 	cpu_caps->rootid = rootkuid;
646 
647 	return 0;
648 }
649 
650 /*
651  * Attempt to get the on-exec apply capability sets for an executable file from
652  * its xattrs and, if present, apply them to the proposed credentials being
653  * constructed by execve().
654  */
655 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_fcap)
656 {
657 	int rc = 0;
658 	struct cpu_vfs_cap_data vcaps;
659 
660 	cap_clear(bprm->cred->cap_permitted);
661 
662 	if (!file_caps_enabled)
663 		return 0;
664 
665 	if (!mnt_may_suid(bprm->file->f_path.mnt))
666 		return 0;
667 
668 	/*
669 	 * This check is redundant with mnt_may_suid() but is kept to make
670 	 * explicit that capability bits are limited to s_user_ns and its
671 	 * descendants.
672 	 */
673 	if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
674 		return 0;
675 
676 	rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
677 	if (rc < 0) {
678 		if (rc == -EINVAL)
679 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
680 					bprm->filename);
681 		else if (rc == -ENODATA)
682 			rc = 0;
683 		goto out;
684 	}
685 
686 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
687 
688 out:
689 	if (rc)
690 		cap_clear(bprm->cred->cap_permitted);
691 
692 	return rc;
693 }
694 
695 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
696 
697 static inline bool __is_real(kuid_t uid, struct cred *cred)
698 { return uid_eq(cred->uid, uid); }
699 
700 static inline bool __is_eff(kuid_t uid, struct cred *cred)
701 { return uid_eq(cred->euid, uid); }
702 
703 static inline bool __is_suid(kuid_t uid, struct cred *cred)
704 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
705 
706 /*
707  * handle_privileged_root - Handle case of privileged root
708  * @bprm: The execution parameters, including the proposed creds
709  * @has_fcap: Are any file capabilities set?
710  * @effective: Do we have effective root privilege?
711  * @root_uid: This namespace' root UID WRT initial USER namespace
712  *
713  * Handle the case where root is privileged and hasn't been neutered by
714  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
715  * set UID root and nothing is changed.  If we are root, cap_permitted is
716  * updated.  If we have become set UID root, the effective bit is set.
717  */
718 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
719 				   bool *effective, kuid_t root_uid)
720 {
721 	const struct cred *old = current_cred();
722 	struct cred *new = bprm->cred;
723 
724 	if (!root_privileged())
725 		return;
726 	/*
727 	 * If the legacy file capability is set, then don't set privs
728 	 * for a setuid root binary run by a non-root user.  Do set it
729 	 * for a root user just to cause least surprise to an admin.
730 	 */
731 	if (has_fcap && __is_suid(root_uid, new)) {
732 		warn_setuid_and_fcaps_mixed(bprm->filename);
733 		return;
734 	}
735 	/*
736 	 * To support inheritance of root-permissions and suid-root
737 	 * executables under compatibility mode, we override the
738 	 * capability sets for the file.
739 	 */
740 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
741 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
742 		new->cap_permitted = cap_combine(old->cap_bset,
743 						 old->cap_inheritable);
744 	}
745 	/*
746 	 * If only the real uid is 0, we do not set the effective bit.
747 	 */
748 	if (__is_eff(root_uid, new))
749 		*effective = true;
750 }
751 
752 #define __cap_gained(field, target, source) \
753 	!cap_issubset(target->cap_##field, source->cap_##field)
754 #define __cap_grew(target, source, cred) \
755 	!cap_issubset(cred->cap_##target, cred->cap_##source)
756 #define __cap_full(field, cred) \
757 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
758 
759 static inline bool __is_setuid(struct cred *new, const struct cred *old)
760 { return !uid_eq(new->euid, old->uid); }
761 
762 static inline bool __is_setgid(struct cred *new, const struct cred *old)
763 { return !gid_eq(new->egid, old->gid); }
764 
765 /*
766  * 1) Audit candidate if current->cap_effective is set
767  *
768  * We do not bother to audit if 3 things are true:
769  *   1) cap_effective has all caps
770  *   2) we became root *OR* are were already root
771  *   3) root is supposed to have all caps (SECURE_NOROOT)
772  * Since this is just a normal root execing a process.
773  *
774  * Number 1 above might fail if you don't have a full bset, but I think
775  * that is interesting information to audit.
776  *
777  * A number of other conditions require logging:
778  * 2) something prevented setuid root getting all caps
779  * 3) non-setuid root gets fcaps
780  * 4) non-setuid root gets ambient
781  */
782 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
783 				     kuid_t root, bool has_fcap)
784 {
785 	bool ret = false;
786 
787 	if ((__cap_grew(effective, ambient, new) &&
788 	     !(__cap_full(effective, new) &&
789 	       (__is_eff(root, new) || __is_real(root, new)) &&
790 	       root_privileged())) ||
791 	    (root_privileged() &&
792 	     __is_suid(root, new) &&
793 	     !__cap_full(effective, new)) ||
794 	    (!__is_setuid(new, old) &&
795 	     ((has_fcap &&
796 	       __cap_gained(permitted, new, old)) ||
797 	      __cap_gained(ambient, new, old))))
798 
799 		ret = true;
800 
801 	return ret;
802 }
803 
804 /**
805  * cap_bprm_set_creds - Set up the proposed credentials for execve().
806  * @bprm: The execution parameters, including the proposed creds
807  *
808  * Set up the proposed credentials for a new execution context being
809  * constructed by execve().  The proposed creds in @bprm->cred is altered,
810  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
811  */
812 int cap_bprm_set_creds(struct linux_binprm *bprm)
813 {
814 	const struct cred *old = current_cred();
815 	struct cred *new = bprm->cred;
816 	bool effective = false, has_fcap = false, is_setid;
817 	int ret;
818 	kuid_t root_uid;
819 
820 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
821 		return -EPERM;
822 
823 	ret = get_file_caps(bprm, &effective, &has_fcap);
824 	if (ret < 0)
825 		return ret;
826 
827 	root_uid = make_kuid(new->user_ns, 0);
828 
829 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
830 
831 	/* if we have fs caps, clear dangerous personality flags */
832 	if (__cap_gained(permitted, new, old))
833 		bprm->per_clear |= PER_CLEAR_ON_SETID;
834 
835 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
836 	 * credentials unless they have the appropriate permit.
837 	 *
838 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
839 	 */
840 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
841 
842 	if ((is_setid || __cap_gained(permitted, new, old)) &&
843 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
844 	     !ptracer_capable(current, new->user_ns))) {
845 		/* downgrade; they get no more than they had, and maybe less */
846 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
847 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
848 			new->euid = new->uid;
849 			new->egid = new->gid;
850 		}
851 		new->cap_permitted = cap_intersect(new->cap_permitted,
852 						   old->cap_permitted);
853 	}
854 
855 	new->suid = new->fsuid = new->euid;
856 	new->sgid = new->fsgid = new->egid;
857 
858 	/* File caps or setid cancels ambient. */
859 	if (has_fcap || is_setid)
860 		cap_clear(new->cap_ambient);
861 
862 	/*
863 	 * Now that we've computed pA', update pP' to give:
864 	 *   pP' = (X & fP) | (pI & fI) | pA'
865 	 */
866 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
867 
868 	/*
869 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
870 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
871 	 */
872 	if (effective)
873 		new->cap_effective = new->cap_permitted;
874 	else
875 		new->cap_effective = new->cap_ambient;
876 
877 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
878 		return -EPERM;
879 
880 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
881 		ret = audit_log_bprm_fcaps(bprm, new, old);
882 		if (ret < 0)
883 			return ret;
884 	}
885 
886 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
887 
888 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
889 		return -EPERM;
890 
891 	/* Check for privilege-elevated exec. */
892 	bprm->cap_elevated = 0;
893 	if (is_setid ||
894 	    (!__is_real(root_uid, new) &&
895 	     (effective ||
896 	      __cap_grew(permitted, ambient, new))))
897 		bprm->cap_elevated = 1;
898 
899 	return 0;
900 }
901 
902 /**
903  * cap_inode_setxattr - Determine whether an xattr may be altered
904  * @dentry: The inode/dentry being altered
905  * @name: The name of the xattr to be changed
906  * @value: The value that the xattr will be changed to
907  * @size: The size of value
908  * @flags: The replacement flag
909  *
910  * Determine whether an xattr may be altered or set on an inode, returning 0 if
911  * permission is granted, -ve if denied.
912  *
913  * This is used to make sure security xattrs don't get updated or set by those
914  * who aren't privileged to do so.
915  */
916 int cap_inode_setxattr(struct dentry *dentry, const char *name,
917 		       const void *value, size_t size, int flags)
918 {
919 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
920 
921 	/* Ignore non-security xattrs */
922 	if (strncmp(name, XATTR_SECURITY_PREFIX,
923 			sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
924 		return 0;
925 
926 	/*
927 	 * For XATTR_NAME_CAPS the check will be done in
928 	 * cap_convert_nscap(), called by setxattr()
929 	 */
930 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
931 		return 0;
932 
933 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
934 		return -EPERM;
935 	return 0;
936 }
937 
938 /**
939  * cap_inode_removexattr - Determine whether an xattr may be removed
940  * @dentry: The inode/dentry being altered
941  * @name: The name of the xattr to be changed
942  *
943  * Determine whether an xattr may be removed from an inode, returning 0 if
944  * permission is granted, -ve if denied.
945  *
946  * This is used to make sure security xattrs don't get removed by those who
947  * aren't privileged to remove them.
948  */
949 int cap_inode_removexattr(struct dentry *dentry, const char *name)
950 {
951 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
952 
953 	/* Ignore non-security xattrs */
954 	if (strncmp(name, XATTR_SECURITY_PREFIX,
955 			sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
956 		return 0;
957 
958 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
959 		/* security.capability gets namespaced */
960 		struct inode *inode = d_backing_inode(dentry);
961 		if (!inode)
962 			return -EINVAL;
963 		if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
964 			return -EPERM;
965 		return 0;
966 	}
967 
968 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
969 		return -EPERM;
970 	return 0;
971 }
972 
973 /*
974  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
975  * a process after a call to setuid, setreuid, or setresuid.
976  *
977  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
978  *  {r,e,s}uid != 0, the permitted and effective capabilities are
979  *  cleared.
980  *
981  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
982  *  capabilities of the process are cleared.
983  *
984  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
985  *  capabilities are set to the permitted capabilities.
986  *
987  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
988  *  never happen.
989  *
990  *  -astor
991  *
992  * cevans - New behaviour, Oct '99
993  * A process may, via prctl(), elect to keep its capabilities when it
994  * calls setuid() and switches away from uid==0. Both permitted and
995  * effective sets will be retained.
996  * Without this change, it was impossible for a daemon to drop only some
997  * of its privilege. The call to setuid(!=0) would drop all privileges!
998  * Keeping uid 0 is not an option because uid 0 owns too many vital
999  * files..
1000  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1001  */
1002 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1003 {
1004 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1005 
1006 	if ((uid_eq(old->uid, root_uid) ||
1007 	     uid_eq(old->euid, root_uid) ||
1008 	     uid_eq(old->suid, root_uid)) &&
1009 	    (!uid_eq(new->uid, root_uid) &&
1010 	     !uid_eq(new->euid, root_uid) &&
1011 	     !uid_eq(new->suid, root_uid))) {
1012 		if (!issecure(SECURE_KEEP_CAPS)) {
1013 			cap_clear(new->cap_permitted);
1014 			cap_clear(new->cap_effective);
1015 		}
1016 
1017 		/*
1018 		 * Pre-ambient programs expect setresuid to nonroot followed
1019 		 * by exec to drop capabilities.  We should make sure that
1020 		 * this remains the case.
1021 		 */
1022 		cap_clear(new->cap_ambient);
1023 	}
1024 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1025 		cap_clear(new->cap_effective);
1026 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1027 		new->cap_effective = new->cap_permitted;
1028 }
1029 
1030 /**
1031  * cap_task_fix_setuid - Fix up the results of setuid() call
1032  * @new: The proposed credentials
1033  * @old: The current task's current credentials
1034  * @flags: Indications of what has changed
1035  *
1036  * Fix up the results of setuid() call before the credential changes are
1037  * actually applied, returning 0 to grant the changes, -ve to deny them.
1038  */
1039 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1040 {
1041 	switch (flags) {
1042 	case LSM_SETID_RE:
1043 	case LSM_SETID_ID:
1044 	case LSM_SETID_RES:
1045 		/* juggle the capabilities to follow [RES]UID changes unless
1046 		 * otherwise suppressed */
1047 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1048 			cap_emulate_setxuid(new, old);
1049 		break;
1050 
1051 	case LSM_SETID_FS:
1052 		/* juggle the capabilties to follow FSUID changes, unless
1053 		 * otherwise suppressed
1054 		 *
1055 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1056 		 *          if not, we might be a bit too harsh here.
1057 		 */
1058 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1059 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1060 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1061 				new->cap_effective =
1062 					cap_drop_fs_set(new->cap_effective);
1063 
1064 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1065 				new->cap_effective =
1066 					cap_raise_fs_set(new->cap_effective,
1067 							 new->cap_permitted);
1068 		}
1069 		break;
1070 
1071 	default:
1072 		return -EINVAL;
1073 	}
1074 
1075 	return 0;
1076 }
1077 
1078 /*
1079  * Rationale: code calling task_setscheduler, task_setioprio, and
1080  * task_setnice, assumes that
1081  *   . if capable(cap_sys_nice), then those actions should be allowed
1082  *   . if not capable(cap_sys_nice), but acting on your own processes,
1083  *   	then those actions should be allowed
1084  * This is insufficient now since you can call code without suid, but
1085  * yet with increased caps.
1086  * So we check for increased caps on the target process.
1087  */
1088 static int cap_safe_nice(struct task_struct *p)
1089 {
1090 	int is_subset, ret = 0;
1091 
1092 	rcu_read_lock();
1093 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1094 				 current_cred()->cap_permitted);
1095 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1096 		ret = -EPERM;
1097 	rcu_read_unlock();
1098 
1099 	return ret;
1100 }
1101 
1102 /**
1103  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1104  * @p: The task to affect
1105  *
1106  * Detemine if the requested scheduler policy change is permitted for the
1107  * specified task, returning 0 if permission is granted, -ve if denied.
1108  */
1109 int cap_task_setscheduler(struct task_struct *p)
1110 {
1111 	return cap_safe_nice(p);
1112 }
1113 
1114 /**
1115  * cap_task_ioprio - Detemine if I/O priority change is permitted
1116  * @p: The task to affect
1117  * @ioprio: The I/O priority to set
1118  *
1119  * Detemine if the requested I/O priority change is permitted for the specified
1120  * task, returning 0 if permission is granted, -ve if denied.
1121  */
1122 int cap_task_setioprio(struct task_struct *p, int ioprio)
1123 {
1124 	return cap_safe_nice(p);
1125 }
1126 
1127 /**
1128  * cap_task_ioprio - Detemine if task priority change is permitted
1129  * @p: The task to affect
1130  * @nice: The nice value to set
1131  *
1132  * Detemine if the requested task priority change is permitted for the
1133  * specified task, returning 0 if permission is granted, -ve if denied.
1134  */
1135 int cap_task_setnice(struct task_struct *p, int nice)
1136 {
1137 	return cap_safe_nice(p);
1138 }
1139 
1140 /*
1141  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1142  * the current task's bounding set.  Returns 0 on success, -ve on error.
1143  */
1144 static int cap_prctl_drop(unsigned long cap)
1145 {
1146 	struct cred *new;
1147 
1148 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1149 		return -EPERM;
1150 	if (!cap_valid(cap))
1151 		return -EINVAL;
1152 
1153 	new = prepare_creds();
1154 	if (!new)
1155 		return -ENOMEM;
1156 	cap_lower(new->cap_bset, cap);
1157 	return commit_creds(new);
1158 }
1159 
1160 /**
1161  * cap_task_prctl - Implement process control functions for this security module
1162  * @option: The process control function requested
1163  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1164  *
1165  * Allow process control functions (sys_prctl()) to alter capabilities; may
1166  * also deny access to other functions not otherwise implemented here.
1167  *
1168  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1169  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1170  * modules will consider performing the function.
1171  */
1172 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1173 		   unsigned long arg4, unsigned long arg5)
1174 {
1175 	const struct cred *old = current_cred();
1176 	struct cred *new;
1177 
1178 	switch (option) {
1179 	case PR_CAPBSET_READ:
1180 		if (!cap_valid(arg2))
1181 			return -EINVAL;
1182 		return !!cap_raised(old->cap_bset, arg2);
1183 
1184 	case PR_CAPBSET_DROP:
1185 		return cap_prctl_drop(arg2);
1186 
1187 	/*
1188 	 * The next four prctl's remain to assist with transitioning a
1189 	 * system from legacy UID=0 based privilege (when filesystem
1190 	 * capabilities are not in use) to a system using filesystem
1191 	 * capabilities only - as the POSIX.1e draft intended.
1192 	 *
1193 	 * Note:
1194 	 *
1195 	 *  PR_SET_SECUREBITS =
1196 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1197 	 *    | issecure_mask(SECURE_NOROOT)
1198 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1199 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1200 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1201 	 *
1202 	 * will ensure that the current process and all of its
1203 	 * children will be locked into a pure
1204 	 * capability-based-privilege environment.
1205 	 */
1206 	case PR_SET_SECUREBITS:
1207 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1208 		     & (old->securebits ^ arg2))			/*[1]*/
1209 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1210 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1211 		    || (cap_capable(current_cred(),
1212 				    current_cred()->user_ns,
1213 				    CAP_SETPCAP,
1214 				    CAP_OPT_NONE) != 0)			/*[4]*/
1215 			/*
1216 			 * [1] no changing of bits that are locked
1217 			 * [2] no unlocking of locks
1218 			 * [3] no setting of unsupported bits
1219 			 * [4] doing anything requires privilege (go read about
1220 			 *     the "sendmail capabilities bug")
1221 			 */
1222 		    )
1223 			/* cannot change a locked bit */
1224 			return -EPERM;
1225 
1226 		new = prepare_creds();
1227 		if (!new)
1228 			return -ENOMEM;
1229 		new->securebits = arg2;
1230 		return commit_creds(new);
1231 
1232 	case PR_GET_SECUREBITS:
1233 		return old->securebits;
1234 
1235 	case PR_GET_KEEPCAPS:
1236 		return !!issecure(SECURE_KEEP_CAPS);
1237 
1238 	case PR_SET_KEEPCAPS:
1239 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1240 			return -EINVAL;
1241 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1242 			return -EPERM;
1243 
1244 		new = prepare_creds();
1245 		if (!new)
1246 			return -ENOMEM;
1247 		if (arg2)
1248 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1249 		else
1250 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1251 		return commit_creds(new);
1252 
1253 	case PR_CAP_AMBIENT:
1254 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1255 			if (arg3 | arg4 | arg5)
1256 				return -EINVAL;
1257 
1258 			new = prepare_creds();
1259 			if (!new)
1260 				return -ENOMEM;
1261 			cap_clear(new->cap_ambient);
1262 			return commit_creds(new);
1263 		}
1264 
1265 		if (((!cap_valid(arg3)) | arg4 | arg5))
1266 			return -EINVAL;
1267 
1268 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1269 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1270 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1271 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1272 			return -EINVAL;
1273 		} else {
1274 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1275 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1276 			     !cap_raised(current_cred()->cap_inheritable,
1277 					 arg3) ||
1278 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1279 				return -EPERM;
1280 
1281 			new = prepare_creds();
1282 			if (!new)
1283 				return -ENOMEM;
1284 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1285 				cap_raise(new->cap_ambient, arg3);
1286 			else
1287 				cap_lower(new->cap_ambient, arg3);
1288 			return commit_creds(new);
1289 		}
1290 
1291 	default:
1292 		/* No functionality available - continue with default */
1293 		return -ENOSYS;
1294 	}
1295 }
1296 
1297 /**
1298  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1299  * @mm: The VM space in which the new mapping is to be made
1300  * @pages: The size of the mapping
1301  *
1302  * Determine whether the allocation of a new virtual mapping by the current
1303  * task is permitted, returning 1 if permission is granted, 0 if not.
1304  */
1305 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1306 {
1307 	int cap_sys_admin = 0;
1308 
1309 	if (cap_capable(current_cred(), &init_user_ns,
1310 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1311 		cap_sys_admin = 1;
1312 
1313 	return cap_sys_admin;
1314 }
1315 
1316 /*
1317  * cap_mmap_addr - check if able to map given addr
1318  * @addr: address attempting to be mapped
1319  *
1320  * If the process is attempting to map memory below dac_mmap_min_addr they need
1321  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1322  * capability security module.  Returns 0 if this mapping should be allowed
1323  * -EPERM if not.
1324  */
1325 int cap_mmap_addr(unsigned long addr)
1326 {
1327 	int ret = 0;
1328 
1329 	if (addr < dac_mmap_min_addr) {
1330 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1331 				  CAP_OPT_NONE);
1332 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1333 		if (ret == 0)
1334 			current->flags |= PF_SUPERPRIV;
1335 	}
1336 	return ret;
1337 }
1338 
1339 int cap_mmap_file(struct file *file, unsigned long reqprot,
1340 		  unsigned long prot, unsigned long flags)
1341 {
1342 	return 0;
1343 }
1344 
1345 #ifdef CONFIG_SECURITY
1346 
1347 struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1348 	LSM_HOOK_INIT(capable, cap_capable),
1349 	LSM_HOOK_INIT(settime, cap_settime),
1350 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1351 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1352 	LSM_HOOK_INIT(capget, cap_capget),
1353 	LSM_HOOK_INIT(capset, cap_capset),
1354 	LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds),
1355 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1356 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1357 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1358 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1359 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1360 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1361 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1362 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1363 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1364 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1365 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1366 };
1367 
1368 static int __init capability_init(void)
1369 {
1370 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1371 				"capability");
1372 	return 0;
1373 }
1374 
1375 DEFINE_LSM(capability) = {
1376 	.name = "capability",
1377 	.order = LSM_ORDER_FIRST,
1378 	.init = capability_init,
1379 };
1380 
1381 #endif /* CONFIG_SECURITY */
1382