xref: /linux/security/commoncap.c (revision 68550cbc6129159b7a6434796b721e8b66ee12f6)
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  * @targ_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  * Return: 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  *
307  * @mnt_userns:	user namespace of the mount the inode was found from
308  * @dentry:	The inode/dentry to alter
309  *
310  * Erase the privilege-enhancing security markings on an inode.
311  *
312  * If the inode has been found through an idmapped mount the user namespace of
313  * the vfsmount must be passed through @mnt_userns. This function will then
314  * take care to map the inode according to @mnt_userns before checking
315  * permissions. On non-idmapped mounts or if permission checking is to be
316  * performed on the raw inode simply passs init_user_ns.
317  *
318  * Return: 0 if successful, -ve on error.
319  */
320 int cap_inode_killpriv(struct user_namespace *mnt_userns, struct dentry *dentry)
321 {
322 	int error;
323 
324 	error = __vfs_removexattr(mnt_userns, dentry, XATTR_NAME_CAPS);
325 	if (error == -EOPNOTSUPP)
326 		error = 0;
327 	return error;
328 }
329 
330 static bool rootid_owns_currentns(kuid_t kroot)
331 {
332 	struct user_namespace *ns;
333 
334 	if (!uid_valid(kroot))
335 		return false;
336 
337 	for (ns = current_user_ns(); ; ns = ns->parent) {
338 		if (from_kuid(ns, kroot) == 0)
339 			return true;
340 		if (ns == &init_user_ns)
341 			break;
342 	}
343 
344 	return false;
345 }
346 
347 static __u32 sansflags(__u32 m)
348 {
349 	return m & ~VFS_CAP_FLAGS_EFFECTIVE;
350 }
351 
352 static bool is_v2header(size_t size, const struct vfs_cap_data *cap)
353 {
354 	if (size != XATTR_CAPS_SZ_2)
355 		return false;
356 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_2;
357 }
358 
359 static bool is_v3header(size_t size, const struct vfs_cap_data *cap)
360 {
361 	if (size != XATTR_CAPS_SZ_3)
362 		return false;
363 	return sansflags(le32_to_cpu(cap->magic_etc)) == VFS_CAP_REVISION_3;
364 }
365 
366 /*
367  * getsecurity: We are called for security.* before any attempt to read the
368  * xattr from the inode itself.
369  *
370  * This gives us a chance to read the on-disk value and convert it.  If we
371  * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler.
372  *
373  * Note we are not called by vfs_getxattr_alloc(), but that is only called
374  * by the integrity subsystem, which really wants the unconverted values -
375  * so that's good.
376  */
377 int cap_inode_getsecurity(struct user_namespace *mnt_userns,
378 			  struct inode *inode, const char *name, void **buffer,
379 			  bool alloc)
380 {
381 	int size, ret;
382 	kuid_t kroot;
383 	u32 nsmagic, magic;
384 	uid_t root, mappedroot;
385 	char *tmpbuf = NULL;
386 	struct vfs_cap_data *cap;
387 	struct vfs_ns_cap_data *nscap = NULL;
388 	struct dentry *dentry;
389 	struct user_namespace *fs_ns;
390 
391 	if (strcmp(name, "capability") != 0)
392 		return -EOPNOTSUPP;
393 
394 	dentry = d_find_any_alias(inode);
395 	if (!dentry)
396 		return -EINVAL;
397 
398 	size = sizeof(struct vfs_ns_cap_data);
399 	ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
400 				      &tmpbuf, size, GFP_NOFS);
401 	dput(dentry);
402 
403 	if (ret < 0 || !tmpbuf)
404 		return ret;
405 
406 	fs_ns = inode->i_sb->s_user_ns;
407 	cap = (struct vfs_cap_data *) tmpbuf;
408 	if (is_v2header((size_t) ret, cap)) {
409 		root = 0;
410 	} else if (is_v3header((size_t) ret, cap)) {
411 		nscap = (struct vfs_ns_cap_data *) tmpbuf;
412 		root = le32_to_cpu(nscap->rootid);
413 	} else {
414 		size = -EINVAL;
415 		goto out_free;
416 	}
417 
418 	kroot = make_kuid(fs_ns, root);
419 
420 	/* If this is an idmapped mount shift the kuid. */
421 	kroot = kuid_into_mnt(mnt_userns, kroot);
422 
423 	/* If the root kuid maps to a valid uid in current ns, then return
424 	 * this as a nscap. */
425 	mappedroot = from_kuid(current_user_ns(), kroot);
426 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
427 		size = sizeof(struct vfs_ns_cap_data);
428 		if (alloc) {
429 			if (!nscap) {
430 				/* v2 -> v3 conversion */
431 				nscap = kzalloc(size, GFP_ATOMIC);
432 				if (!nscap) {
433 					size = -ENOMEM;
434 					goto out_free;
435 				}
436 				nsmagic = VFS_CAP_REVISION_3;
437 				magic = le32_to_cpu(cap->magic_etc);
438 				if (magic & VFS_CAP_FLAGS_EFFECTIVE)
439 					nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
440 				memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
441 				nscap->magic_etc = cpu_to_le32(nsmagic);
442 			} else {
443 				/* use allocated v3 buffer */
444 				tmpbuf = NULL;
445 			}
446 			nscap->rootid = cpu_to_le32(mappedroot);
447 			*buffer = nscap;
448 		}
449 		goto out_free;
450 	}
451 
452 	if (!rootid_owns_currentns(kroot)) {
453 		size = -EOVERFLOW;
454 		goto out_free;
455 	}
456 
457 	/* This comes from a parent namespace.  Return as a v2 capability */
458 	size = sizeof(struct vfs_cap_data);
459 	if (alloc) {
460 		if (nscap) {
461 			/* v3 -> v2 conversion */
462 			cap = kzalloc(size, GFP_ATOMIC);
463 			if (!cap) {
464 				size = -ENOMEM;
465 				goto out_free;
466 			}
467 			magic = VFS_CAP_REVISION_2;
468 			nsmagic = le32_to_cpu(nscap->magic_etc);
469 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
470 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
471 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
472 			cap->magic_etc = cpu_to_le32(magic);
473 		} else {
474 			/* use unconverted v2 */
475 			tmpbuf = NULL;
476 		}
477 		*buffer = cap;
478 	}
479 out_free:
480 	kfree(tmpbuf);
481 	return size;
482 }
483 
484 /**
485  * rootid_from_xattr - translate root uid of vfs caps
486  *
487  * @value:	vfs caps value which may be modified by this function
488  * @size:	size of @ivalue
489  * @task_ns:	user namespace of the caller
490  * @mnt_userns:	user namespace of the mount the inode was found from
491  *
492  * If the inode has been found through an idmapped mount the user namespace of
493  * the vfsmount must be passed through @mnt_userns. This function will then
494  * take care to map the inode according to @mnt_userns before checking
495  * permissions. On non-idmapped mounts or if permission checking is to be
496  * performed on the raw inode simply passs init_user_ns.
497  */
498 static kuid_t rootid_from_xattr(const void *value, size_t size,
499 				struct user_namespace *task_ns,
500 				struct user_namespace *mnt_userns)
501 {
502 	const struct vfs_ns_cap_data *nscap = value;
503 	kuid_t rootkid;
504 	uid_t rootid = 0;
505 
506 	if (size == XATTR_CAPS_SZ_3)
507 		rootid = le32_to_cpu(nscap->rootid);
508 
509 	rootkid = make_kuid(task_ns, rootid);
510 	return kuid_from_mnt(mnt_userns, rootkid);
511 }
512 
513 static bool validheader(size_t size, const struct vfs_cap_data *cap)
514 {
515 	return is_v2header(size, cap) || is_v3header(size, cap);
516 }
517 
518 /**
519  * cap_convert_nscap - check vfs caps
520  *
521  * @mnt_userns:	user namespace of the mount the inode was found from
522  * @dentry:	used to retrieve inode to check permissions on
523  * @ivalue:	vfs caps value which may be modified by this function
524  * @size:	size of @ivalue
525  *
526  * User requested a write of security.capability.  If needed, update the
527  * xattr to change from v2 to v3, or to fixup the v3 rootid.
528  *
529  * If the inode has been found through an idmapped mount the user namespace of
530  * the vfsmount must be passed through @mnt_userns. This function will then
531  * take care to map the inode according to @mnt_userns before checking
532  * permissions. On non-idmapped mounts or if permission checking is to be
533  * performed on the raw inode simply passs init_user_ns.
534  *
535  * Return: On success, return the new size; on error, return < 0.
536  */
537 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
538 		      const void **ivalue, size_t size)
539 {
540 	struct vfs_ns_cap_data *nscap;
541 	uid_t nsrootid;
542 	const struct vfs_cap_data *cap = *ivalue;
543 	__u32 magic, nsmagic;
544 	struct inode *inode = d_backing_inode(dentry);
545 	struct user_namespace *task_ns = current_user_ns(),
546 		*fs_ns = inode->i_sb->s_user_ns;
547 	kuid_t rootid;
548 	size_t newsize;
549 
550 	if (!*ivalue)
551 		return -EINVAL;
552 	if (!validheader(size, cap))
553 		return -EINVAL;
554 	if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
555 		return -EPERM;
556 	if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
557 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
558 			/* user is privileged, just write the v2 */
559 			return size;
560 
561 	rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
562 	if (!uid_valid(rootid))
563 		return -EINVAL;
564 
565 	nsrootid = from_kuid(fs_ns, rootid);
566 	if (nsrootid == -1)
567 		return -EINVAL;
568 
569 	newsize = sizeof(struct vfs_ns_cap_data);
570 	nscap = kmalloc(newsize, GFP_ATOMIC);
571 	if (!nscap)
572 		return -ENOMEM;
573 	nscap->rootid = cpu_to_le32(nsrootid);
574 	nsmagic = VFS_CAP_REVISION_3;
575 	magic = le32_to_cpu(cap->magic_etc);
576 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
577 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
578 	nscap->magic_etc = cpu_to_le32(nsmagic);
579 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
580 
581 	*ivalue = nscap;
582 	return newsize;
583 }
584 
585 /*
586  * Calculate the new process capability sets from the capability sets attached
587  * to a file.
588  */
589 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
590 					  struct linux_binprm *bprm,
591 					  bool *effective,
592 					  bool *has_fcap)
593 {
594 	struct cred *new = bprm->cred;
595 	unsigned i;
596 	int ret = 0;
597 
598 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
599 		*effective = true;
600 
601 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
602 		*has_fcap = true;
603 
604 	CAP_FOR_EACH_U32(i) {
605 		__u32 permitted = caps->permitted.cap[i];
606 		__u32 inheritable = caps->inheritable.cap[i];
607 
608 		/*
609 		 * pP' = (X & fP) | (pI & fI)
610 		 * The addition of pA' is handled later.
611 		 */
612 		new->cap_permitted.cap[i] =
613 			(new->cap_bset.cap[i] & permitted) |
614 			(new->cap_inheritable.cap[i] & inheritable);
615 
616 		if (permitted & ~new->cap_permitted.cap[i])
617 			/* insufficient to execute correctly */
618 			ret = -EPERM;
619 	}
620 
621 	/*
622 	 * For legacy apps, with no internal support for recognizing they
623 	 * do not have enough capabilities, we return an error if they are
624 	 * missing some "forced" (aka file-permitted) capabilities.
625 	 */
626 	return *effective ? ret : 0;
627 }
628 
629 /**
630  * get_vfs_caps_from_disk - retrieve vfs caps from disk
631  *
632  * @mnt_userns:	user namespace of the mount the inode was found from
633  * @dentry:	dentry from which @inode is retrieved
634  * @cpu_caps:	vfs capabilities
635  *
636  * Extract the on-exec-apply capability sets for an executable file.
637  *
638  * If the inode has been found through an idmapped mount the user namespace of
639  * the vfsmount must be passed through @mnt_userns. This function will then
640  * take care to map the inode according to @mnt_userns before checking
641  * permissions. On non-idmapped mounts or if permission checking is to be
642  * performed on the raw inode simply passs init_user_ns.
643  */
644 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
645 			   const struct dentry *dentry,
646 			   struct cpu_vfs_cap_data *cpu_caps)
647 {
648 	struct inode *inode = d_backing_inode(dentry);
649 	__u32 magic_etc;
650 	unsigned tocopy, i;
651 	int size;
652 	struct vfs_ns_cap_data data, *nscaps = &data;
653 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
654 	kuid_t rootkuid;
655 	struct user_namespace *fs_ns;
656 
657 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
658 
659 	if (!inode)
660 		return -ENODATA;
661 
662 	fs_ns = inode->i_sb->s_user_ns;
663 	size = __vfs_getxattr((struct dentry *)dentry, inode,
664 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
665 	if (size == -ENODATA || size == -EOPNOTSUPP)
666 		/* no data, that's ok */
667 		return -ENODATA;
668 
669 	if (size < 0)
670 		return size;
671 
672 	if (size < sizeof(magic_etc))
673 		return -EINVAL;
674 
675 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
676 
677 	rootkuid = make_kuid(fs_ns, 0);
678 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
679 	case VFS_CAP_REVISION_1:
680 		if (size != XATTR_CAPS_SZ_1)
681 			return -EINVAL;
682 		tocopy = VFS_CAP_U32_1;
683 		break;
684 	case VFS_CAP_REVISION_2:
685 		if (size != XATTR_CAPS_SZ_2)
686 			return -EINVAL;
687 		tocopy = VFS_CAP_U32_2;
688 		break;
689 	case VFS_CAP_REVISION_3:
690 		if (size != XATTR_CAPS_SZ_3)
691 			return -EINVAL;
692 		tocopy = VFS_CAP_U32_3;
693 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
694 		break;
695 
696 	default:
697 		return -EINVAL;
698 	}
699 	/* Limit the caps to the mounter of the filesystem
700 	 * or the more limited uid specified in the xattr.
701 	 */
702 	rootkuid = kuid_into_mnt(mnt_userns, rootkuid);
703 	if (!rootid_owns_currentns(rootkuid))
704 		return -ENODATA;
705 
706 	CAP_FOR_EACH_U32(i) {
707 		if (i >= tocopy)
708 			break;
709 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
710 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
711 	}
712 
713 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
714 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
715 
716 	cpu_caps->rootid = rootkuid;
717 
718 	return 0;
719 }
720 
721 /*
722  * Attempt to get the on-exec apply capability sets for an executable file from
723  * its xattrs and, if present, apply them to the proposed credentials being
724  * constructed by execve().
725  */
726 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
727 			 bool *effective, bool *has_fcap)
728 {
729 	int rc = 0;
730 	struct cpu_vfs_cap_data vcaps;
731 
732 	cap_clear(bprm->cred->cap_permitted);
733 
734 	if (!file_caps_enabled)
735 		return 0;
736 
737 	if (!mnt_may_suid(file->f_path.mnt))
738 		return 0;
739 
740 	/*
741 	 * This check is redundant with mnt_may_suid() but is kept to make
742 	 * explicit that capability bits are limited to s_user_ns and its
743 	 * descendants.
744 	 */
745 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
746 		return 0;
747 
748 	rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
749 				    file->f_path.dentry, &vcaps);
750 	if (rc < 0) {
751 		if (rc == -EINVAL)
752 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
753 					bprm->filename);
754 		else if (rc == -ENODATA)
755 			rc = 0;
756 		goto out;
757 	}
758 
759 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
760 
761 out:
762 	if (rc)
763 		cap_clear(bprm->cred->cap_permitted);
764 
765 	return rc;
766 }
767 
768 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
769 
770 static inline bool __is_real(kuid_t uid, struct cred *cred)
771 { return uid_eq(cred->uid, uid); }
772 
773 static inline bool __is_eff(kuid_t uid, struct cred *cred)
774 { return uid_eq(cred->euid, uid); }
775 
776 static inline bool __is_suid(kuid_t uid, struct cred *cred)
777 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
778 
779 /*
780  * handle_privileged_root - Handle case of privileged root
781  * @bprm: The execution parameters, including the proposed creds
782  * @has_fcap: Are any file capabilities set?
783  * @effective: Do we have effective root privilege?
784  * @root_uid: This namespace' root UID WRT initial USER namespace
785  *
786  * Handle the case where root is privileged and hasn't been neutered by
787  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
788  * set UID root and nothing is changed.  If we are root, cap_permitted is
789  * updated.  If we have become set UID root, the effective bit is set.
790  */
791 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
792 				   bool *effective, kuid_t root_uid)
793 {
794 	const struct cred *old = current_cred();
795 	struct cred *new = bprm->cred;
796 
797 	if (!root_privileged())
798 		return;
799 	/*
800 	 * If the legacy file capability is set, then don't set privs
801 	 * for a setuid root binary run by a non-root user.  Do set it
802 	 * for a root user just to cause least surprise to an admin.
803 	 */
804 	if (has_fcap && __is_suid(root_uid, new)) {
805 		warn_setuid_and_fcaps_mixed(bprm->filename);
806 		return;
807 	}
808 	/*
809 	 * To support inheritance of root-permissions and suid-root
810 	 * executables under compatibility mode, we override the
811 	 * capability sets for the file.
812 	 */
813 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
814 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
815 		new->cap_permitted = cap_combine(old->cap_bset,
816 						 old->cap_inheritable);
817 	}
818 	/*
819 	 * If only the real uid is 0, we do not set the effective bit.
820 	 */
821 	if (__is_eff(root_uid, new))
822 		*effective = true;
823 }
824 
825 #define __cap_gained(field, target, source) \
826 	!cap_issubset(target->cap_##field, source->cap_##field)
827 #define __cap_grew(target, source, cred) \
828 	!cap_issubset(cred->cap_##target, cred->cap_##source)
829 #define __cap_full(field, cred) \
830 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
831 
832 static inline bool __is_setuid(struct cred *new, const struct cred *old)
833 { return !uid_eq(new->euid, old->uid); }
834 
835 static inline bool __is_setgid(struct cred *new, const struct cred *old)
836 { return !gid_eq(new->egid, old->gid); }
837 
838 /*
839  * 1) Audit candidate if current->cap_effective is set
840  *
841  * We do not bother to audit if 3 things are true:
842  *   1) cap_effective has all caps
843  *   2) we became root *OR* are were already root
844  *   3) root is supposed to have all caps (SECURE_NOROOT)
845  * Since this is just a normal root execing a process.
846  *
847  * Number 1 above might fail if you don't have a full bset, but I think
848  * that is interesting information to audit.
849  *
850  * A number of other conditions require logging:
851  * 2) something prevented setuid root getting all caps
852  * 3) non-setuid root gets fcaps
853  * 4) non-setuid root gets ambient
854  */
855 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
856 				     kuid_t root, bool has_fcap)
857 {
858 	bool ret = false;
859 
860 	if ((__cap_grew(effective, ambient, new) &&
861 	     !(__cap_full(effective, new) &&
862 	       (__is_eff(root, new) || __is_real(root, new)) &&
863 	       root_privileged())) ||
864 	    (root_privileged() &&
865 	     __is_suid(root, new) &&
866 	     !__cap_full(effective, new)) ||
867 	    (!__is_setuid(new, old) &&
868 	     ((has_fcap &&
869 	       __cap_gained(permitted, new, old)) ||
870 	      __cap_gained(ambient, new, old))))
871 
872 		ret = true;
873 
874 	return ret;
875 }
876 
877 /**
878  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
879  * @bprm: The execution parameters, including the proposed creds
880  * @file: The file to pull the credentials from
881  *
882  * Set up the proposed credentials for a new execution context being
883  * constructed by execve().  The proposed creds in @bprm->cred is altered,
884  * which won't take effect immediately.
885  *
886  * Return: 0 if successful, -ve on error.
887  */
888 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
889 {
890 	/* Process setpcap binaries and capabilities for uid 0 */
891 	const struct cred *old = current_cred();
892 	struct cred *new = bprm->cred;
893 	bool effective = false, has_fcap = false, is_setid;
894 	int ret;
895 	kuid_t root_uid;
896 
897 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
898 		return -EPERM;
899 
900 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
901 	if (ret < 0)
902 		return ret;
903 
904 	root_uid = make_kuid(new->user_ns, 0);
905 
906 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
907 
908 	/* if we have fs caps, clear dangerous personality flags */
909 	if (__cap_gained(permitted, new, old))
910 		bprm->per_clear |= PER_CLEAR_ON_SETID;
911 
912 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
913 	 * credentials unless they have the appropriate permit.
914 	 *
915 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
916 	 */
917 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
918 
919 	if ((is_setid || __cap_gained(permitted, new, old)) &&
920 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
921 	     !ptracer_capable(current, new->user_ns))) {
922 		/* downgrade; they get no more than they had, and maybe less */
923 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
924 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
925 			new->euid = new->uid;
926 			new->egid = new->gid;
927 		}
928 		new->cap_permitted = cap_intersect(new->cap_permitted,
929 						   old->cap_permitted);
930 	}
931 
932 	new->suid = new->fsuid = new->euid;
933 	new->sgid = new->fsgid = new->egid;
934 
935 	/* File caps or setid cancels ambient. */
936 	if (has_fcap || is_setid)
937 		cap_clear(new->cap_ambient);
938 
939 	/*
940 	 * Now that we've computed pA', update pP' to give:
941 	 *   pP' = (X & fP) | (pI & fI) | pA'
942 	 */
943 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
944 
945 	/*
946 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
947 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
948 	 */
949 	if (effective)
950 		new->cap_effective = new->cap_permitted;
951 	else
952 		new->cap_effective = new->cap_ambient;
953 
954 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
955 		return -EPERM;
956 
957 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
958 		ret = audit_log_bprm_fcaps(bprm, new, old);
959 		if (ret < 0)
960 			return ret;
961 	}
962 
963 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
964 
965 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
966 		return -EPERM;
967 
968 	/* Check for privilege-elevated exec. */
969 	if (is_setid ||
970 	    (!__is_real(root_uid, new) &&
971 	     (effective ||
972 	      __cap_grew(permitted, ambient, new))))
973 		bprm->secureexec = 1;
974 
975 	return 0;
976 }
977 
978 /**
979  * cap_inode_setxattr - Determine whether an xattr may be altered
980  * @dentry: The inode/dentry being altered
981  * @name: The name of the xattr to be changed
982  * @value: The value that the xattr will be changed to
983  * @size: The size of value
984  * @flags: The replacement flag
985  *
986  * Determine whether an xattr may be altered or set on an inode, returning 0 if
987  * permission is granted, -ve if denied.
988  *
989  * This is used to make sure security xattrs don't get updated or set by those
990  * who aren't privileged to do so.
991  */
992 int cap_inode_setxattr(struct dentry *dentry, const char *name,
993 		       const void *value, size_t size, int flags)
994 {
995 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
996 
997 	/* Ignore non-security xattrs */
998 	if (strncmp(name, XATTR_SECURITY_PREFIX,
999 			XATTR_SECURITY_PREFIX_LEN) != 0)
1000 		return 0;
1001 
1002 	/*
1003 	 * For XATTR_NAME_CAPS the check will be done in
1004 	 * cap_convert_nscap(), called by setxattr()
1005 	 */
1006 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
1007 		return 0;
1008 
1009 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1010 		return -EPERM;
1011 	return 0;
1012 }
1013 
1014 /**
1015  * cap_inode_removexattr - Determine whether an xattr may be removed
1016  *
1017  * @mnt_userns:	User namespace of the mount the inode was found from
1018  * @dentry:	The inode/dentry being altered
1019  * @name:	The name of the xattr to be changed
1020  *
1021  * Determine whether an xattr may be removed from an inode, returning 0 if
1022  * permission is granted, -ve if denied.
1023  *
1024  * If the inode has been found through an idmapped mount the user namespace of
1025  * the vfsmount must be passed through @mnt_userns. This function will then
1026  * take care to map the inode according to @mnt_userns before checking
1027  * permissions. On non-idmapped mounts or if permission checking is to be
1028  * performed on the raw inode simply passs init_user_ns.
1029  *
1030  * This is used to make sure security xattrs don't get removed by those who
1031  * aren't privileged to remove them.
1032  */
1033 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1034 			  struct dentry *dentry, const char *name)
1035 {
1036 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1037 
1038 	/* Ignore non-security xattrs */
1039 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1040 			XATTR_SECURITY_PREFIX_LEN) != 0)
1041 		return 0;
1042 
1043 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1044 		/* security.capability gets namespaced */
1045 		struct inode *inode = d_backing_inode(dentry);
1046 		if (!inode)
1047 			return -EINVAL;
1048 		if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1049 			return -EPERM;
1050 		return 0;
1051 	}
1052 
1053 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1054 		return -EPERM;
1055 	return 0;
1056 }
1057 
1058 /*
1059  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1060  * a process after a call to setuid, setreuid, or setresuid.
1061  *
1062  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1063  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1064  *  cleared.
1065  *
1066  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1067  *  capabilities of the process are cleared.
1068  *
1069  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1070  *  capabilities are set to the permitted capabilities.
1071  *
1072  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1073  *  never happen.
1074  *
1075  *  -astor
1076  *
1077  * cevans - New behaviour, Oct '99
1078  * A process may, via prctl(), elect to keep its capabilities when it
1079  * calls setuid() and switches away from uid==0. Both permitted and
1080  * effective sets will be retained.
1081  * Without this change, it was impossible for a daemon to drop only some
1082  * of its privilege. The call to setuid(!=0) would drop all privileges!
1083  * Keeping uid 0 is not an option because uid 0 owns too many vital
1084  * files..
1085  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1086  */
1087 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1088 {
1089 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1090 
1091 	if ((uid_eq(old->uid, root_uid) ||
1092 	     uid_eq(old->euid, root_uid) ||
1093 	     uid_eq(old->suid, root_uid)) &&
1094 	    (!uid_eq(new->uid, root_uid) &&
1095 	     !uid_eq(new->euid, root_uid) &&
1096 	     !uid_eq(new->suid, root_uid))) {
1097 		if (!issecure(SECURE_KEEP_CAPS)) {
1098 			cap_clear(new->cap_permitted);
1099 			cap_clear(new->cap_effective);
1100 		}
1101 
1102 		/*
1103 		 * Pre-ambient programs expect setresuid to nonroot followed
1104 		 * by exec to drop capabilities.  We should make sure that
1105 		 * this remains the case.
1106 		 */
1107 		cap_clear(new->cap_ambient);
1108 	}
1109 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1110 		cap_clear(new->cap_effective);
1111 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1112 		new->cap_effective = new->cap_permitted;
1113 }
1114 
1115 /**
1116  * cap_task_fix_setuid - Fix up the results of setuid() call
1117  * @new: The proposed credentials
1118  * @old: The current task's current credentials
1119  * @flags: Indications of what has changed
1120  *
1121  * Fix up the results of setuid() call before the credential changes are
1122  * actually applied.
1123  *
1124  * Return: 0 to grant the changes, -ve to deny them.
1125  */
1126 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1127 {
1128 	switch (flags) {
1129 	case LSM_SETID_RE:
1130 	case LSM_SETID_ID:
1131 	case LSM_SETID_RES:
1132 		/* juggle the capabilities to follow [RES]UID changes unless
1133 		 * otherwise suppressed */
1134 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1135 			cap_emulate_setxuid(new, old);
1136 		break;
1137 
1138 	case LSM_SETID_FS:
1139 		/* juggle the capabilties to follow FSUID changes, unless
1140 		 * otherwise suppressed
1141 		 *
1142 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1143 		 *          if not, we might be a bit too harsh here.
1144 		 */
1145 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1146 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1147 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1148 				new->cap_effective =
1149 					cap_drop_fs_set(new->cap_effective);
1150 
1151 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1152 				new->cap_effective =
1153 					cap_raise_fs_set(new->cap_effective,
1154 							 new->cap_permitted);
1155 		}
1156 		break;
1157 
1158 	default:
1159 		return -EINVAL;
1160 	}
1161 
1162 	return 0;
1163 }
1164 
1165 /*
1166  * Rationale: code calling task_setscheduler, task_setioprio, and
1167  * task_setnice, assumes that
1168  *   . if capable(cap_sys_nice), then those actions should be allowed
1169  *   . if not capable(cap_sys_nice), but acting on your own processes,
1170  *   	then those actions should be allowed
1171  * This is insufficient now since you can call code without suid, but
1172  * yet with increased caps.
1173  * So we check for increased caps on the target process.
1174  */
1175 static int cap_safe_nice(struct task_struct *p)
1176 {
1177 	int is_subset, ret = 0;
1178 
1179 	rcu_read_lock();
1180 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1181 				 current_cred()->cap_permitted);
1182 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1183 		ret = -EPERM;
1184 	rcu_read_unlock();
1185 
1186 	return ret;
1187 }
1188 
1189 /**
1190  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1191  * @p: The task to affect
1192  *
1193  * Detemine if the requested scheduler policy change is permitted for the
1194  * specified task.
1195  *
1196  * Return: 0 if permission is granted, -ve if denied.
1197  */
1198 int cap_task_setscheduler(struct task_struct *p)
1199 {
1200 	return cap_safe_nice(p);
1201 }
1202 
1203 /**
1204  * cap_task_setioprio - Detemine if I/O priority change is permitted
1205  * @p: The task to affect
1206  * @ioprio: The I/O priority to set
1207  *
1208  * Detemine if the requested I/O priority change is permitted for the specified
1209  * task.
1210  *
1211  * Return: 0 if permission is granted, -ve if denied.
1212  */
1213 int cap_task_setioprio(struct task_struct *p, int ioprio)
1214 {
1215 	return cap_safe_nice(p);
1216 }
1217 
1218 /**
1219  * cap_task_setnice - Detemine if task priority change is permitted
1220  * @p: The task to affect
1221  * @nice: The nice value to set
1222  *
1223  * Detemine if the requested task priority change is permitted for the
1224  * specified task.
1225  *
1226  * Return: 0 if permission is granted, -ve if denied.
1227  */
1228 int cap_task_setnice(struct task_struct *p, int nice)
1229 {
1230 	return cap_safe_nice(p);
1231 }
1232 
1233 /*
1234  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1235  * the current task's bounding set.  Returns 0 on success, -ve on error.
1236  */
1237 static int cap_prctl_drop(unsigned long cap)
1238 {
1239 	struct cred *new;
1240 
1241 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1242 		return -EPERM;
1243 	if (!cap_valid(cap))
1244 		return -EINVAL;
1245 
1246 	new = prepare_creds();
1247 	if (!new)
1248 		return -ENOMEM;
1249 	cap_lower(new->cap_bset, cap);
1250 	return commit_creds(new);
1251 }
1252 
1253 /**
1254  * cap_task_prctl - Implement process control functions for this security module
1255  * @option: The process control function requested
1256  * @arg2: The argument data for this function
1257  * @arg3: The argument data for this function
1258  * @arg4: The argument data for this function
1259  * @arg5: The argument data for this function
1260  *
1261  * Allow process control functions (sys_prctl()) to alter capabilities; may
1262  * also deny access to other functions not otherwise implemented here.
1263  *
1264  * Return: 0 or +ve on success, -ENOSYS if this function is not implemented
1265  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1266  * modules will consider performing the function.
1267  */
1268 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1269 		   unsigned long arg4, unsigned long arg5)
1270 {
1271 	const struct cred *old = current_cred();
1272 	struct cred *new;
1273 
1274 	switch (option) {
1275 	case PR_CAPBSET_READ:
1276 		if (!cap_valid(arg2))
1277 			return -EINVAL;
1278 		return !!cap_raised(old->cap_bset, arg2);
1279 
1280 	case PR_CAPBSET_DROP:
1281 		return cap_prctl_drop(arg2);
1282 
1283 	/*
1284 	 * The next four prctl's remain to assist with transitioning a
1285 	 * system from legacy UID=0 based privilege (when filesystem
1286 	 * capabilities are not in use) to a system using filesystem
1287 	 * capabilities only - as the POSIX.1e draft intended.
1288 	 *
1289 	 * Note:
1290 	 *
1291 	 *  PR_SET_SECUREBITS =
1292 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1293 	 *    | issecure_mask(SECURE_NOROOT)
1294 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1295 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1296 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1297 	 *
1298 	 * will ensure that the current process and all of its
1299 	 * children will be locked into a pure
1300 	 * capability-based-privilege environment.
1301 	 */
1302 	case PR_SET_SECUREBITS:
1303 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1304 		     & (old->securebits ^ arg2))			/*[1]*/
1305 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1306 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1307 		    || (cap_capable(current_cred(),
1308 				    current_cred()->user_ns,
1309 				    CAP_SETPCAP,
1310 				    CAP_OPT_NONE) != 0)			/*[4]*/
1311 			/*
1312 			 * [1] no changing of bits that are locked
1313 			 * [2] no unlocking of locks
1314 			 * [3] no setting of unsupported bits
1315 			 * [4] doing anything requires privilege (go read about
1316 			 *     the "sendmail capabilities bug")
1317 			 */
1318 		    )
1319 			/* cannot change a locked bit */
1320 			return -EPERM;
1321 
1322 		new = prepare_creds();
1323 		if (!new)
1324 			return -ENOMEM;
1325 		new->securebits = arg2;
1326 		return commit_creds(new);
1327 
1328 	case PR_GET_SECUREBITS:
1329 		return old->securebits;
1330 
1331 	case PR_GET_KEEPCAPS:
1332 		return !!issecure(SECURE_KEEP_CAPS);
1333 
1334 	case PR_SET_KEEPCAPS:
1335 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1336 			return -EINVAL;
1337 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1338 			return -EPERM;
1339 
1340 		new = prepare_creds();
1341 		if (!new)
1342 			return -ENOMEM;
1343 		if (arg2)
1344 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1345 		else
1346 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1347 		return commit_creds(new);
1348 
1349 	case PR_CAP_AMBIENT:
1350 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1351 			if (arg3 | arg4 | arg5)
1352 				return -EINVAL;
1353 
1354 			new = prepare_creds();
1355 			if (!new)
1356 				return -ENOMEM;
1357 			cap_clear(new->cap_ambient);
1358 			return commit_creds(new);
1359 		}
1360 
1361 		if (((!cap_valid(arg3)) | arg4 | arg5))
1362 			return -EINVAL;
1363 
1364 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1365 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1366 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1367 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1368 			return -EINVAL;
1369 		} else {
1370 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1371 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1372 			     !cap_raised(current_cred()->cap_inheritable,
1373 					 arg3) ||
1374 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1375 				return -EPERM;
1376 
1377 			new = prepare_creds();
1378 			if (!new)
1379 				return -ENOMEM;
1380 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1381 				cap_raise(new->cap_ambient, arg3);
1382 			else
1383 				cap_lower(new->cap_ambient, arg3);
1384 			return commit_creds(new);
1385 		}
1386 
1387 	default:
1388 		/* No functionality available - continue with default */
1389 		return -ENOSYS;
1390 	}
1391 }
1392 
1393 /**
1394  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1395  * @mm: The VM space in which the new mapping is to be made
1396  * @pages: The size of the mapping
1397  *
1398  * Determine whether the allocation of a new virtual mapping by the current
1399  * task is permitted.
1400  *
1401  * Return: 1 if permission is granted, 0 if not.
1402  */
1403 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1404 {
1405 	int cap_sys_admin = 0;
1406 
1407 	if (cap_capable(current_cred(), &init_user_ns,
1408 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1409 		cap_sys_admin = 1;
1410 
1411 	return cap_sys_admin;
1412 }
1413 
1414 /**
1415  * cap_mmap_addr - check if able to map given addr
1416  * @addr: address attempting to be mapped
1417  *
1418  * If the process is attempting to map memory below dac_mmap_min_addr they need
1419  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1420  * capability security module.
1421  *
1422  * Return: 0 if this mapping should be allowed or -EPERM if not.
1423  */
1424 int cap_mmap_addr(unsigned long addr)
1425 {
1426 	int ret = 0;
1427 
1428 	if (addr < dac_mmap_min_addr) {
1429 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1430 				  CAP_OPT_NONE);
1431 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1432 		if (ret == 0)
1433 			current->flags |= PF_SUPERPRIV;
1434 	}
1435 	return ret;
1436 }
1437 
1438 int cap_mmap_file(struct file *file, unsigned long reqprot,
1439 		  unsigned long prot, unsigned long flags)
1440 {
1441 	return 0;
1442 }
1443 
1444 #ifdef CONFIG_SECURITY
1445 
1446 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1447 	LSM_HOOK_INIT(capable, cap_capable),
1448 	LSM_HOOK_INIT(settime, cap_settime),
1449 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1450 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1451 	LSM_HOOK_INIT(capget, cap_capget),
1452 	LSM_HOOK_INIT(capset, cap_capset),
1453 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1454 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1455 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1456 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1457 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1458 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1459 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1460 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1461 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1462 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1463 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1464 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1465 };
1466 
1467 static int __init capability_init(void)
1468 {
1469 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1470 				"capability");
1471 	return 0;
1472 }
1473 
1474 DEFINE_LSM(capability) = {
1475 	.name = "capability",
1476 	.order = LSM_ORDER_FIRST,
1477 	.init = capability_init,
1478 };
1479 
1480 #endif /* CONFIG_SECURITY */
1481