xref: /openbmc/linux/security/commoncap.c (revision a2818ee4)
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  * @audit: Whether to write an audit message or not
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, int audit)
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 
226 	/* they are so limited unless the current task has the CAP_SETPCAP
227 	 * capability
228 	 */
229 	if (cap_capable(current_cred(), current_cred()->user_ns,
230 			CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
231 		return 0;
232 	return 1;
233 }
234 
235 /**
236  * cap_capset - Validate and apply proposed changes to current's capabilities
237  * @new: The proposed new credentials; alterations should be made here
238  * @old: The current task's current credentials
239  * @effective: A pointer to the proposed new effective capabilities set
240  * @inheritable: A pointer to the proposed new inheritable capabilities set
241  * @permitted: A pointer to the proposed new permitted capabilities set
242  *
243  * This function validates and applies a proposed mass change to the current
244  * process's capability sets.  The changes are made to the proposed new
245  * credentials, and assuming no error, will be committed by the caller of LSM.
246  */
247 int cap_capset(struct cred *new,
248 	       const struct cred *old,
249 	       const kernel_cap_t *effective,
250 	       const kernel_cap_t *inheritable,
251 	       const kernel_cap_t *permitted)
252 {
253 	if (cap_inh_is_capped() &&
254 	    !cap_issubset(*inheritable,
255 			  cap_combine(old->cap_inheritable,
256 				      old->cap_permitted)))
257 		/* incapable of using this inheritable set */
258 		return -EPERM;
259 
260 	if (!cap_issubset(*inheritable,
261 			  cap_combine(old->cap_inheritable,
262 				      old->cap_bset)))
263 		/* no new pI capabilities outside bounding set */
264 		return -EPERM;
265 
266 	/* verify restrictions on target's new Permitted set */
267 	if (!cap_issubset(*permitted, old->cap_permitted))
268 		return -EPERM;
269 
270 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
271 	if (!cap_issubset(*effective, *permitted))
272 		return -EPERM;
273 
274 	new->cap_effective   = *effective;
275 	new->cap_inheritable = *inheritable;
276 	new->cap_permitted   = *permitted;
277 
278 	/*
279 	 * Mask off ambient bits that are no longer both permitted and
280 	 * inheritable.
281 	 */
282 	new->cap_ambient = cap_intersect(new->cap_ambient,
283 					 cap_intersect(*permitted,
284 						       *inheritable));
285 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
286 		return -EINVAL;
287 	return 0;
288 }
289 
290 /**
291  * cap_inode_need_killpriv - Determine if inode change affects privileges
292  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
293  *
294  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
295  * affects the security markings on that inode, and if it is, should
296  * inode_killpriv() be invoked or the change rejected.
297  *
298  * Returns 1 if security.capability has a value, meaning inode_killpriv()
299  * is required, 0 otherwise, meaning inode_killpriv() is not required.
300  */
301 int cap_inode_need_killpriv(struct dentry *dentry)
302 {
303 	struct inode *inode = d_backing_inode(dentry);
304 	int error;
305 
306 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
307 	return error > 0;
308 }
309 
310 /**
311  * cap_inode_killpriv - Erase the security markings on an inode
312  * @dentry: The inode/dentry to alter
313  *
314  * Erase the privilege-enhancing security markings on an inode.
315  *
316  * Returns 0 if successful, -ve on error.
317  */
318 int cap_inode_killpriv(struct dentry *dentry)
319 {
320 	int error;
321 
322 	error = __vfs_removexattr(dentry, XATTR_NAME_CAPS);
323 	if (error == -EOPNOTSUPP)
324 		error = 0;
325 	return error;
326 }
327 
328 static bool rootid_owns_currentns(kuid_t kroot)
329 {
330 	struct user_namespace *ns;
331 
332 	if (!uid_valid(kroot))
333 		return false;
334 
335 	for (ns = current_user_ns(); ; ns = ns->parent) {
336 		if (from_kuid(ns, kroot) == 0)
337 			return true;
338 		if (ns == &init_user_ns)
339 			break;
340 	}
341 
342 	return false;
343 }
344 
345 static __u32 sansflags(__u32 m)
346 {
347 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
348 }
349 
350 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
351 {
352 	if (size != XATTR_CAPS_SZ_2)
353 		return false;
354 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
355 }
356 
357 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
358 {
359 	if (size != XATTR_CAPS_SZ_3)
360 		return false;
361 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
362 }
363 
364 /*
365  * getsecurity: We are called for security.* before any attempt to read the
366  * xattr from the inode itself.
367  *
368  * This gives us a chance to read the on-disk value and convert it.  If we
369  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
370  *
371  * Note we are not called by vfs_getxattr_alloc(), but that is only called
372  * by the integrity subsystem, which really wants the unconverted values -
373  * so that's good.
374  */
375 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer,
376 			  bool alloc)
377 {
378 	int size, ret;
379 	kuid_t kroot;
380 	uid_t root, mappedroot;
381 	char *tmpbuf = NULL;
382 	struct vfs_cap_data *cap;
383 	struct vfs_ns_cap_data *nscap;
384 	struct dentry *dentry;
385 	struct user_namespace *fs_ns;
386 
387 	if (strcmp(name, "capability") != 0)
388 		return -EOPNOTSUPP;
389 
390 	dentry = d_find_any_alias(inode);
391 	if (!dentry)
392 		return -EINVAL;
393 
394 	size = sizeof(struct vfs_ns_cap_data);
395 	ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS,
396 				 &tmpbuf, size, GFP_NOFS);
397 	dput(dentry);
398 
399 	if (ret < 0)
400 		return ret;
401 
402 	fs_ns = inode->i_sb->s_user_ns;
403 	cap = (struct vfs_cap_data *) tmpbuf;
404 	if (is_v2header((size_t) ret, cap)) {
405 		/* If this is sizeof(vfs_cap_data) then we're ok with the
406 		 * on-disk value, so return that.  */
407 		if (alloc)
408 			*buffer = tmpbuf;
409 		else
410 			kfree(tmpbuf);
411 		return ret;
412 	} else if (!is_v3header((size_t) ret, cap)) {
413 		kfree(tmpbuf);
414 		return -EINVAL;
415 	}
416 
417 	nscap = (struct vfs_ns_cap_data *) tmpbuf;
418 	root = le32_to_cpu(nscap->rootid);
419 	kroot = make_kuid(fs_ns, root);
420 
421 	/* If the root kuid maps to a valid uid in current ns, then return
422 	 * this as a nscap. */
423 	mappedroot = from_kuid(current_user_ns(), kroot);
424 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
425 		if (alloc) {
426 			*buffer = tmpbuf;
427 			nscap->rootid = cpu_to_le32(mappedroot);
428 		} else
429 			kfree(tmpbuf);
430 		return size;
431 	}
432 
433 	if (!rootid_owns_currentns(kroot)) {
434 		kfree(tmpbuf);
435 		return -EOPNOTSUPP;
436 	}
437 
438 	/* This comes from a parent namespace.  Return as a v2 capability */
439 	size = sizeof(struct vfs_cap_data);
440 	if (alloc) {
441 		*buffer = kmalloc(size, GFP_ATOMIC);
442 		if (*buffer) {
443 			struct vfs_cap_data *cap = *buffer;
444 			__le32 nsmagic, magic;
445 			magic = VFS_CAP_REVISION_2;
446 			nsmagic = le32_to_cpu(nscap->magic_etc);
447 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
448 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
449 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
450 			cap->magic_etc = cpu_to_le32(magic);
451 		} else {
452 			size = -ENOMEM;
453 		}
454 	}
455 	kfree(tmpbuf);
456 	return size;
457 }
458 
459 static kuid_t rootid_from_xattr(const void *value, size_t size,
460 				struct user_namespace *task_ns)
461 {
462 	const struct vfs_ns_cap_data *nscap = value;
463 	uid_t rootid = 0;
464 
465 	if (size == XATTR_CAPS_SZ_3)
466 		rootid = le32_to_cpu(nscap->rootid);
467 
468 	return make_kuid(task_ns, rootid);
469 }
470 
471 static bool validheader(size_t size, const struct vfs_cap_data *cap)
472 {
473 	return is_v2header(size, cap) || is_v3header(size, cap);
474 }
475 
476 /*
477  * User requested a write of security.capability.  If needed, update the
478  * xattr to change from v2 to v3, or to fixup the v3 rootid.
479  *
480  * If all is ok, we return the new size, on error return < 0.
481  */
482 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size)
483 {
484 	struct vfs_ns_cap_data *nscap;
485 	uid_t nsrootid;
486 	const struct vfs_cap_data *cap = *ivalue;
487 	__u32 magic, nsmagic;
488 	struct inode *inode = d_backing_inode(dentry);
489 	struct user_namespace *task_ns = current_user_ns(),
490 		*fs_ns = inode->i_sb->s_user_ns;
491 	kuid_t rootid;
492 	size_t newsize;
493 
494 	if (!*ivalue)
495 		return -EINVAL;
496 	if (!validheader(size, cap))
497 		return -EINVAL;
498 	if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
499 		return -EPERM;
500 	if (size == XATTR_CAPS_SZ_2)
501 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
502 			/* user is privileged, just write the v2 */
503 			return size;
504 
505 	rootid = rootid_from_xattr(*ivalue, size, task_ns);
506 	if (!uid_valid(rootid))
507 		return -EINVAL;
508 
509 	nsrootid = from_kuid(fs_ns, rootid);
510 	if (nsrootid == -1)
511 		return -EINVAL;
512 
513 	newsize = sizeof(struct vfs_ns_cap_data);
514 	nscap = kmalloc(newsize, GFP_ATOMIC);
515 	if (!nscap)
516 		return -ENOMEM;
517 	nscap->rootid = cpu_to_le32(nsrootid);
518 	nsmagic = VFS_CAP_REVISION_3;
519 	magic = le32_to_cpu(cap->magic_etc);
520 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
521 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
522 	nscap->magic_etc = cpu_to_le32(nsmagic);
523 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
524 
525 	kvfree(*ivalue);
526 	*ivalue = nscap;
527 	return newsize;
528 }
529 
530 /*
531  * Calculate the new process capability sets from the capability sets attached
532  * to a file.
533  */
534 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
535 					  struct linux_binprm *bprm,
536 					  bool *effective,
537 					  bool *has_fcap)
538 {
539 	struct cred *new = bprm->cred;
540 	unsigned i;
541 	int ret = 0;
542 
543 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
544 		*effective = true;
545 
546 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
547 		*has_fcap = true;
548 
549 	CAP_FOR_EACH_U32(i) {
550 		__u32 permitted = caps->permitted.cap[i];
551 		__u32 inheritable = caps->inheritable.cap[i];
552 
553 		/*
554 		 * pP' = (X & fP) | (pI & fI)
555 		 * The addition of pA' is handled later.
556 		 */
557 		new->cap_permitted.cap[i] =
558 			(new->cap_bset.cap[i] & permitted) |
559 			(new->cap_inheritable.cap[i] & inheritable);
560 
561 		if (permitted & ~new->cap_permitted.cap[i])
562 			/* insufficient to execute correctly */
563 			ret = -EPERM;
564 	}
565 
566 	/*
567 	 * For legacy apps, with no internal support for recognizing they
568 	 * do not have enough capabilities, we return an error if they are
569 	 * missing some "forced" (aka file-permitted) capabilities.
570 	 */
571 	return *effective ? ret : 0;
572 }
573 
574 /*
575  * Extract the on-exec-apply capability sets for an executable file.
576  */
577 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
578 {
579 	struct inode *inode = d_backing_inode(dentry);
580 	__u32 magic_etc;
581 	unsigned tocopy, i;
582 	int size;
583 	struct vfs_ns_cap_data data, *nscaps = &data;
584 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
585 	kuid_t rootkuid;
586 	struct user_namespace *fs_ns;
587 
588 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
589 
590 	if (!inode)
591 		return -ENODATA;
592 
593 	fs_ns = inode->i_sb->s_user_ns;
594 	size = __vfs_getxattr((struct dentry *)dentry, inode,
595 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
596 	if (size == -ENODATA || size == -EOPNOTSUPP)
597 		/* no data, that's ok */
598 		return -ENODATA;
599 
600 	if (size < 0)
601 		return size;
602 
603 	if (size < sizeof(magic_etc))
604 		return -EINVAL;
605 
606 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
607 
608 	rootkuid = make_kuid(fs_ns, 0);
609 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
610 	case VFS_CAP_REVISION_1:
611 		if (size != XATTR_CAPS_SZ_1)
612 			return -EINVAL;
613 		tocopy = VFS_CAP_U32_1;
614 		break;
615 	case VFS_CAP_REVISION_2:
616 		if (size != XATTR_CAPS_SZ_2)
617 			return -EINVAL;
618 		tocopy = VFS_CAP_U32_2;
619 		break;
620 	case VFS_CAP_REVISION_3:
621 		if (size != XATTR_CAPS_SZ_3)
622 			return -EINVAL;
623 		tocopy = VFS_CAP_U32_3;
624 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
625 		break;
626 
627 	default:
628 		return -EINVAL;
629 	}
630 	/* Limit the caps to the mounter of the filesystem
631 	 * or the more limited uid specified in the xattr.
632 	 */
633 	if (!rootid_owns_currentns(rootkuid))
634 		return -ENODATA;
635 
636 	CAP_FOR_EACH_U32(i) {
637 		if (i >= tocopy)
638 			break;
639 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
640 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
641 	}
642 
643 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
644 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
645 
646 	return 0;
647 }
648 
649 /*
650  * Attempt to get the on-exec apply capability sets for an executable file from
651  * its xattrs and, if present, apply them to the proposed credentials being
652  * constructed by execve().
653  */
654 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_fcap)
655 {
656 	int rc = 0;
657 	struct cpu_vfs_cap_data vcaps;
658 
659 	cap_clear(bprm->cred->cap_permitted);
660 
661 	if (!file_caps_enabled)
662 		return 0;
663 
664 	if (!mnt_may_suid(bprm->file->f_path.mnt))
665 		return 0;
666 
667 	/*
668 	 * This check is redundant with mnt_may_suid() but is kept to make
669 	 * explicit that capability bits are limited to s_user_ns and its
670 	 * descendants.
671 	 */
672 	if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns))
673 		return 0;
674 
675 	rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps);
676 	if (rc < 0) {
677 		if (rc == -EINVAL)
678 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
679 					bprm->filename);
680 		else if (rc == -ENODATA)
681 			rc = 0;
682 		goto out;
683 	}
684 
685 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
686 
687 out:
688 	if (rc)
689 		cap_clear(bprm->cred->cap_permitted);
690 
691 	return rc;
692 }
693 
694 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
695 
696 static inline bool __is_real(kuid_t uid, struct cred *cred)
697 { return uid_eq(cred->uid, uid); }
698 
699 static inline bool __is_eff(kuid_t uid, struct cred *cred)
700 { return uid_eq(cred->euid, uid); }
701 
702 static inline bool __is_suid(kuid_t uid, struct cred *cred)
703 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
704 
705 /*
706  * handle_privileged_root - Handle case of privileged root
707  * @bprm: The execution parameters, including the proposed creds
708  * @has_fcap: Are any file capabilities set?
709  * @effective: Do we have effective root privilege?
710  * @root_uid: This namespace' root UID WRT initial USER namespace
711  *
712  * Handle the case where root is privileged and hasn't been neutered by
713  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
714  * set UID root and nothing is changed.  If we are root, cap_permitted is
715  * updated.  If we have become set UID root, the effective bit is set.
716  */
717 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
718 				   bool *effective, kuid_t root_uid)
719 {
720 	const struct cred *old = current_cred();
721 	struct cred *new = bprm->cred;
722 
723 	if (!root_privileged())
724 		return;
725 	/*
726 	 * If the legacy file capability is set, then don't set privs
727 	 * for a setuid root binary run by a non-root user.  Do set it
728 	 * for a root user just to cause least surprise to an admin.
729 	 */
730 	if (has_fcap && __is_suid(root_uid, new)) {
731 		warn_setuid_and_fcaps_mixed(bprm->filename);
732 		return;
733 	}
734 	/*
735 	 * To support inheritance of root-permissions and suid-root
736 	 * executables under compatibility mode, we override the
737 	 * capability sets for the file.
738 	 */
739 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
740 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
741 		new->cap_permitted = cap_combine(old->cap_bset,
742 						 old->cap_inheritable);
743 	}
744 	/*
745 	 * If only the real uid is 0, we do not set the effective bit.
746 	 */
747 	if (__is_eff(root_uid, new))
748 		*effective = true;
749 }
750 
751 #define __cap_gained(field, target, source) \
752 	!cap_issubset(target->cap_##field, source->cap_##field)
753 #define __cap_grew(target, source, cred) \
754 	!cap_issubset(cred->cap_##target, cred->cap_##source)
755 #define __cap_full(field, cred) \
756 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
757 
758 static inline bool __is_setuid(struct cred *new, const struct cred *old)
759 { return !uid_eq(new->euid, old->uid); }
760 
761 static inline bool __is_setgid(struct cred *new, const struct cred *old)
762 { return !gid_eq(new->egid, old->gid); }
763 
764 /*
765  * 1) Audit candidate if current->cap_effective is set
766  *
767  * We do not bother to audit if 3 things are true:
768  *   1) cap_effective has all caps
769  *   2) we became root *OR* are were already root
770  *   3) root is supposed to have all caps (SECURE_NOROOT)
771  * Since this is just a normal root execing a process.
772  *
773  * Number 1 above might fail if you don't have a full bset, but I think
774  * that is interesting information to audit.
775  *
776  * A number of other conditions require logging:
777  * 2) something prevented setuid root getting all caps
778  * 3) non-setuid root gets fcaps
779  * 4) non-setuid root gets ambient
780  */
781 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
782 				     kuid_t root, bool has_fcap)
783 {
784 	bool ret = false;
785 
786 	if ((__cap_grew(effective, ambient, new) &&
787 	     !(__cap_full(effective, new) &&
788 	       (__is_eff(root, new) || __is_real(root, new)) &&
789 	       root_privileged())) ||
790 	    (root_privileged() &&
791 	     __is_suid(root, new) &&
792 	     !__cap_full(effective, new)) ||
793 	    (!__is_setuid(new, old) &&
794 	     ((has_fcap &&
795 	       __cap_gained(permitted, new, old)) ||
796 	      __cap_gained(ambient, new, old))))
797 
798 		ret = true;
799 
800 	return ret;
801 }
802 
803 /**
804  * cap_bprm_set_creds - Set up the proposed credentials for execve().
805  * @bprm: The execution parameters, including the proposed creds
806  *
807  * Set up the proposed credentials for a new execution context being
808  * constructed by execve().  The proposed creds in @bprm->cred is altered,
809  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
810  */
811 int cap_bprm_set_creds(struct linux_binprm *bprm)
812 {
813 	const struct cred *old = current_cred();
814 	struct cred *new = bprm->cred;
815 	bool effective = false, has_fcap = false, is_setid;
816 	int ret;
817 	kuid_t root_uid;
818 
819 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
820 		return -EPERM;
821 
822 	ret = get_file_caps(bprm, &effective, &has_fcap);
823 	if (ret < 0)
824 		return ret;
825 
826 	root_uid = make_kuid(new->user_ns, 0);
827 
828 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
829 
830 	/* if we have fs caps, clear dangerous personality flags */
831 	if (__cap_gained(permitted, new, old))
832 		bprm->per_clear |= PER_CLEAR_ON_SETID;
833 
834 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
835 	 * credentials unless they have the appropriate permit.
836 	 *
837 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
838 	 */
839 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
840 
841 	if ((is_setid || __cap_gained(permitted, new, old)) &&
842 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
843 	     !ptracer_capable(current, new->user_ns))) {
844 		/* downgrade; they get no more than they had, and maybe less */
845 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
846 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
847 			new->euid = new->uid;
848 			new->egid = new->gid;
849 		}
850 		new->cap_permitted = cap_intersect(new->cap_permitted,
851 						   old->cap_permitted);
852 	}
853 
854 	new->suid = new->fsuid = new->euid;
855 	new->sgid = new->fsgid = new->egid;
856 
857 	/* File caps or setid cancels ambient. */
858 	if (has_fcap || is_setid)
859 		cap_clear(new->cap_ambient);
860 
861 	/*
862 	 * Now that we've computed pA', update pP' to give:
863 	 *   pP' = (X & fP) | (pI & fI) | pA'
864 	 */
865 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
866 
867 	/*
868 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
869 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
870 	 */
871 	if (effective)
872 		new->cap_effective = new->cap_permitted;
873 	else
874 		new->cap_effective = new->cap_ambient;
875 
876 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
877 		return -EPERM;
878 
879 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
880 		ret = audit_log_bprm_fcaps(bprm, new, old);
881 		if (ret < 0)
882 			return ret;
883 	}
884 
885 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
886 
887 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
888 		return -EPERM;
889 
890 	/* Check for privilege-elevated exec. */
891 	bprm->cap_elevated = 0;
892 	if (is_setid ||
893 	    (!__is_real(root_uid, new) &&
894 	     (effective ||
895 	      __cap_grew(permitted, ambient, new))))
896 		bprm->cap_elevated = 1;
897 
898 	return 0;
899 }
900 
901 /**
902  * cap_inode_setxattr - Determine whether an xattr may be altered
903  * @dentry: The inode/dentry being altered
904  * @name: The name of the xattr to be changed
905  * @value: The value that the xattr will be changed to
906  * @size: The size of value
907  * @flags: The replacement flag
908  *
909  * Determine whether an xattr may be altered or set on an inode, returning 0 if
910  * permission is granted, -ve if denied.
911  *
912  * This is used to make sure security xattrs don't get updated or set by those
913  * who aren't privileged to do so.
914  */
915 int cap_inode_setxattr(struct dentry *dentry, const char *name,
916 		       const void *value, size_t size, int flags)
917 {
918 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
919 
920 	/* Ignore non-security xattrs */
921 	if (strncmp(name, XATTR_SECURITY_PREFIX,
922 			sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
923 		return 0;
924 
925 	/*
926 	 * For XATTR_NAME_CAPS the check will be done in
927 	 * cap_convert_nscap(), called by setxattr()
928 	 */
929 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
930 		return 0;
931 
932 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
933 		return -EPERM;
934 	return 0;
935 }
936 
937 /**
938  * cap_inode_removexattr - Determine whether an xattr may be removed
939  * @dentry: The inode/dentry being altered
940  * @name: The name of the xattr to be changed
941  *
942  * Determine whether an xattr may be removed from an inode, returning 0 if
943  * permission is granted, -ve if denied.
944  *
945  * This is used to make sure security xattrs don't get removed by those who
946  * aren't privileged to remove them.
947  */
948 int cap_inode_removexattr(struct dentry *dentry, const char *name)
949 {
950 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
951 
952 	/* Ignore non-security xattrs */
953 	if (strncmp(name, XATTR_SECURITY_PREFIX,
954 			sizeof(XATTR_SECURITY_PREFIX) - 1) != 0)
955 		return 0;
956 
957 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
958 		/* security.capability gets namespaced */
959 		struct inode *inode = d_backing_inode(dentry);
960 		if (!inode)
961 			return -EINVAL;
962 		if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP))
963 			return -EPERM;
964 		return 0;
965 	}
966 
967 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
968 		return -EPERM;
969 	return 0;
970 }
971 
972 /*
973  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
974  * a process after a call to setuid, setreuid, or setresuid.
975  *
976  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
977  *  {r,e,s}uid != 0, the permitted and effective capabilities are
978  *  cleared.
979  *
980  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
981  *  capabilities of the process are cleared.
982  *
983  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
984  *  capabilities are set to the permitted capabilities.
985  *
986  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
987  *  never happen.
988  *
989  *  -astor
990  *
991  * cevans - New behaviour, Oct '99
992  * A process may, via prctl(), elect to keep its capabilities when it
993  * calls setuid() and switches away from uid==0. Both permitted and
994  * effective sets will be retained.
995  * Without this change, it was impossible for a daemon to drop only some
996  * of its privilege. The call to setuid(!=0) would drop all privileges!
997  * Keeping uid 0 is not an option because uid 0 owns too many vital
998  * files..
999  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1000  */
1001 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1002 {
1003 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1004 
1005 	if ((uid_eq(old->uid, root_uid) ||
1006 	     uid_eq(old->euid, root_uid) ||
1007 	     uid_eq(old->suid, root_uid)) &&
1008 	    (!uid_eq(new->uid, root_uid) &&
1009 	     !uid_eq(new->euid, root_uid) &&
1010 	     !uid_eq(new->suid, root_uid))) {
1011 		if (!issecure(SECURE_KEEP_CAPS)) {
1012 			cap_clear(new->cap_permitted);
1013 			cap_clear(new->cap_effective);
1014 		}
1015 
1016 		/*
1017 		 * Pre-ambient programs expect setresuid to nonroot followed
1018 		 * by exec to drop capabilities.  We should make sure that
1019 		 * this remains the case.
1020 		 */
1021 		cap_clear(new->cap_ambient);
1022 	}
1023 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1024 		cap_clear(new->cap_effective);
1025 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1026 		new->cap_effective = new->cap_permitted;
1027 }
1028 
1029 /**
1030  * cap_task_fix_setuid - Fix up the results of setuid() call
1031  * @new: The proposed credentials
1032  * @old: The current task's current credentials
1033  * @flags: Indications of what has changed
1034  *
1035  * Fix up the results of setuid() call before the credential changes are
1036  * actually applied, returning 0 to grant the changes, -ve to deny them.
1037  */
1038 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1039 {
1040 	switch (flags) {
1041 	case LSM_SETID_RE:
1042 	case LSM_SETID_ID:
1043 	case LSM_SETID_RES:
1044 		/* juggle the capabilities to follow [RES]UID changes unless
1045 		 * otherwise suppressed */
1046 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1047 			cap_emulate_setxuid(new, old);
1048 		break;
1049 
1050 	case LSM_SETID_FS:
1051 		/* juggle the capabilties to follow FSUID changes, unless
1052 		 * otherwise suppressed
1053 		 *
1054 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1055 		 *          if not, we might be a bit too harsh here.
1056 		 */
1057 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1058 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1059 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1060 				new->cap_effective =
1061 					cap_drop_fs_set(new->cap_effective);
1062 
1063 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1064 				new->cap_effective =
1065 					cap_raise_fs_set(new->cap_effective,
1066 							 new->cap_permitted);
1067 		}
1068 		break;
1069 
1070 	default:
1071 		return -EINVAL;
1072 	}
1073 
1074 	return 0;
1075 }
1076 
1077 /*
1078  * Rationale: code calling task_setscheduler, task_setioprio, and
1079  * task_setnice, assumes that
1080  *   . if capable(cap_sys_nice), then those actions should be allowed
1081  *   . if not capable(cap_sys_nice), but acting on your own processes,
1082  *   	then those actions should be allowed
1083  * This is insufficient now since you can call code without suid, but
1084  * yet with increased caps.
1085  * So we check for increased caps on the target process.
1086  */
1087 static int cap_safe_nice(struct task_struct *p)
1088 {
1089 	int is_subset, ret = 0;
1090 
1091 	rcu_read_lock();
1092 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1093 				 current_cred()->cap_permitted);
1094 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1095 		ret = -EPERM;
1096 	rcu_read_unlock();
1097 
1098 	return ret;
1099 }
1100 
1101 /**
1102  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1103  * @p: The task to affect
1104  *
1105  * Detemine if the requested scheduler policy change is permitted for the
1106  * specified task, returning 0 if permission is granted, -ve if denied.
1107  */
1108 int cap_task_setscheduler(struct task_struct *p)
1109 {
1110 	return cap_safe_nice(p);
1111 }
1112 
1113 /**
1114  * cap_task_ioprio - Detemine if I/O priority change is permitted
1115  * @p: The task to affect
1116  * @ioprio: The I/O priority to set
1117  *
1118  * Detemine if the requested I/O priority change is permitted for the specified
1119  * task, returning 0 if permission is granted, -ve if denied.
1120  */
1121 int cap_task_setioprio(struct task_struct *p, int ioprio)
1122 {
1123 	return cap_safe_nice(p);
1124 }
1125 
1126 /**
1127  * cap_task_ioprio - Detemine if task priority change is permitted
1128  * @p: The task to affect
1129  * @nice: The nice value to set
1130  *
1131  * Detemine if the requested task priority change is permitted for the
1132  * specified task, returning 0 if permission is granted, -ve if denied.
1133  */
1134 int cap_task_setnice(struct task_struct *p, int nice)
1135 {
1136 	return cap_safe_nice(p);
1137 }
1138 
1139 /*
1140  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1141  * the current task's bounding set.  Returns 0 on success, -ve on error.
1142  */
1143 static int cap_prctl_drop(unsigned long cap)
1144 {
1145 	struct cred *new;
1146 
1147 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1148 		return -EPERM;
1149 	if (!cap_valid(cap))
1150 		return -EINVAL;
1151 
1152 	new = prepare_creds();
1153 	if (!new)
1154 		return -ENOMEM;
1155 	cap_lower(new->cap_bset, cap);
1156 	return commit_creds(new);
1157 }
1158 
1159 /**
1160  * cap_task_prctl - Implement process control functions for this security module
1161  * @option: The process control function requested
1162  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1163  *
1164  * Allow process control functions (sys_prctl()) to alter capabilities; may
1165  * also deny access to other functions not otherwise implemented here.
1166  *
1167  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1168  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1169  * modules will consider performing the function.
1170  */
1171 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1172 		   unsigned long arg4, unsigned long arg5)
1173 {
1174 	const struct cred *old = current_cred();
1175 	struct cred *new;
1176 
1177 	switch (option) {
1178 	case PR_CAPBSET_READ:
1179 		if (!cap_valid(arg2))
1180 			return -EINVAL;
1181 		return !!cap_raised(old->cap_bset, arg2);
1182 
1183 	case PR_CAPBSET_DROP:
1184 		return cap_prctl_drop(arg2);
1185 
1186 	/*
1187 	 * The next four prctl's remain to assist with transitioning a
1188 	 * system from legacy UID=0 based privilege (when filesystem
1189 	 * capabilities are not in use) to a system using filesystem
1190 	 * capabilities only - as the POSIX.1e draft intended.
1191 	 *
1192 	 * Note:
1193 	 *
1194 	 *  PR_SET_SECUREBITS =
1195 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1196 	 *    | issecure_mask(SECURE_NOROOT)
1197 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1198 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1199 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1200 	 *
1201 	 * will ensure that the current process and all of its
1202 	 * children will be locked into a pure
1203 	 * capability-based-privilege environment.
1204 	 */
1205 	case PR_SET_SECUREBITS:
1206 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1207 		     & (old->securebits ^ arg2))			/*[1]*/
1208 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1209 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1210 		    || (cap_capable(current_cred(),
1211 				    current_cred()->user_ns, CAP_SETPCAP,
1212 				    SECURITY_CAP_AUDIT) != 0)		/*[4]*/
1213 			/*
1214 			 * [1] no changing of bits that are locked
1215 			 * [2] no unlocking of locks
1216 			 * [3] no setting of unsupported bits
1217 			 * [4] doing anything requires privilege (go read about
1218 			 *     the "sendmail capabilities bug")
1219 			 */
1220 		    )
1221 			/* cannot change a locked bit */
1222 			return -EPERM;
1223 
1224 		new = prepare_creds();
1225 		if (!new)
1226 			return -ENOMEM;
1227 		new->securebits = arg2;
1228 		return commit_creds(new);
1229 
1230 	case PR_GET_SECUREBITS:
1231 		return old->securebits;
1232 
1233 	case PR_GET_KEEPCAPS:
1234 		return !!issecure(SECURE_KEEP_CAPS);
1235 
1236 	case PR_SET_KEEPCAPS:
1237 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1238 			return -EINVAL;
1239 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1240 			return -EPERM;
1241 
1242 		new = prepare_creds();
1243 		if (!new)
1244 			return -ENOMEM;
1245 		if (arg2)
1246 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1247 		else
1248 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1249 		return commit_creds(new);
1250 
1251 	case PR_CAP_AMBIENT:
1252 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1253 			if (arg3 | arg4 | arg5)
1254 				return -EINVAL;
1255 
1256 			new = prepare_creds();
1257 			if (!new)
1258 				return -ENOMEM;
1259 			cap_clear(new->cap_ambient);
1260 			return commit_creds(new);
1261 		}
1262 
1263 		if (((!cap_valid(arg3)) | arg4 | arg5))
1264 			return -EINVAL;
1265 
1266 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1267 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1268 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1269 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1270 			return -EINVAL;
1271 		} else {
1272 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1273 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1274 			     !cap_raised(current_cred()->cap_inheritable,
1275 					 arg3) ||
1276 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1277 				return -EPERM;
1278 
1279 			new = prepare_creds();
1280 			if (!new)
1281 				return -ENOMEM;
1282 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1283 				cap_raise(new->cap_ambient, arg3);
1284 			else
1285 				cap_lower(new->cap_ambient, arg3);
1286 			return commit_creds(new);
1287 		}
1288 
1289 	default:
1290 		/* No functionality available - continue with default */
1291 		return -ENOSYS;
1292 	}
1293 }
1294 
1295 /**
1296  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1297  * @mm: The VM space in which the new mapping is to be made
1298  * @pages: The size of the mapping
1299  *
1300  * Determine whether the allocation of a new virtual mapping by the current
1301  * task is permitted, returning 1 if permission is granted, 0 if not.
1302  */
1303 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1304 {
1305 	int cap_sys_admin = 0;
1306 
1307 	if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1308 			SECURITY_CAP_NOAUDIT) == 0)
1309 		cap_sys_admin = 1;
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 				  SECURITY_CAP_AUDIT);
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 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_set_creds, cap_bprm_set_creds),
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 void __init capability_add_hooks(void)
1366 {
1367 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1368 				"capability");
1369 }
1370 
1371 #endif /* CONFIG_SECURITY */
1372