xref: /openbmc/linux/security/commoncap.c (revision 6a143a7c)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Common capabilities, needed by capability.o.
3  */
4 
5 #include <linux/capability.h>
6 #include <linux/audit.h>
7 #include <linux/init.h>
8 #include <linux/kernel.h>
9 #include <linux/lsm_hooks.h>
10 #include <linux/file.h>
11 #include <linux/mm.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swap.h>
15 #include <linux/skbuff.h>
16 #include <linux/netlink.h>
17 #include <linux/ptrace.h>
18 #include <linux/xattr.h>
19 #include <linux/hugetlb.h>
20 #include <linux/mount.h>
21 #include <linux/sched.h>
22 #include <linux/prctl.h>
23 #include <linux/securebits.h>
24 #include <linux/user_namespace.h>
25 #include <linux/binfmts.h>
26 #include <linux/personality.h>
27 
28 /*
29  * If a non-root user executes a setuid-root binary in
30  * !secure(SECURE_NOROOT) mode, then we raise capabilities.
31  * However if fE is also set, then the intent is for only
32  * the file capabilities to be applied, and the setuid-root
33  * bit is left on either to change the uid (plausible) or
34  * to get full privilege on a kernel without file capabilities
35  * support.  So in that case we do not raise capabilities.
36  *
37  * Warn if that happens, once per boot.
38  */
39 static void warn_setuid_and_fcaps_mixed(const char *fname)
40 {
41 	static int warned;
42 	if (!warned) {
43 		printk(KERN_INFO "warning: `%s' has both setuid-root and"
44 			" effective capabilities. Therefore not raising all"
45 			" capabilities.\n", fname);
46 		warned = 1;
47 	}
48 }
49 
50 /**
51  * cap_capable - Determine whether a task has a particular effective capability
52  * @cred: The credentials to use
53  * @ns:  The user namespace in which we need the capability
54  * @cap: The capability to check for
55  * @opts: Bitmask of options defined in include/linux/security.h
56  *
57  * Determine whether the nominated task has the specified capability amongst
58  * its effective set, returning 0 if it does, -ve if it does not.
59  *
60  * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable()
61  * and has_capability() functions.  That is, it has the reverse semantics:
62  * cap_has_capability() returns 0 when a task has a capability, but the
63  * kernel's capable() and has_capability() returns 1 for this case.
64  */
65 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns,
66 		int cap, unsigned int opts)
67 {
68 	struct user_namespace *ns = targ_ns;
69 
70 	/* See if cred has the capability in the target user namespace
71 	 * by examining the target user namespace and all of the target
72 	 * user namespace's parents.
73 	 */
74 	for (;;) {
75 		/* Do we have the necessary capabilities? */
76 		if (ns == cred->user_ns)
77 			return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
78 
79 		/*
80 		 * If we're already at a lower level than we're looking for,
81 		 * we're done searching.
82 		 */
83 		if (ns->level <= cred->user_ns->level)
84 			return -EPERM;
85 
86 		/*
87 		 * The owner of the user namespace in the parent of the
88 		 * user namespace has all caps.
89 		 */
90 		if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid))
91 			return 0;
92 
93 		/*
94 		 * If you have a capability in a parent user ns, then you have
95 		 * it over all children user namespaces as well.
96 		 */
97 		ns = ns->parent;
98 	}
99 
100 	/* We never get here */
101 }
102 
103 /**
104  * cap_settime - Determine whether the current process may set the system clock
105  * @ts: The time to set
106  * @tz: The timezone to set
107  *
108  * Determine whether the current process may set the system clock and timezone
109  * information, returning 0 if permission granted, -ve if denied.
110  */
111 int cap_settime(const struct timespec64 *ts, const struct timezone *tz)
112 {
113 	if (!capable(CAP_SYS_TIME))
114 		return -EPERM;
115 	return 0;
116 }
117 
118 /**
119  * cap_ptrace_access_check - Determine whether the current process may access
120  *			   another
121  * @child: The process to be accessed
122  * @mode: The mode of attachment.
123  *
124  * If we are in the same or an ancestor user_ns and have all the target
125  * task's capabilities, then ptrace access is allowed.
126  * If we have the ptrace capability to the target user_ns, then ptrace
127  * access is allowed.
128  * Else denied.
129  *
130  * Determine whether a process may access another, returning 0 if permission
131  * granted, -ve if denied.
132  */
133 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode)
134 {
135 	int ret = 0;
136 	const struct cred *cred, *child_cred;
137 	const kernel_cap_t *caller_caps;
138 
139 	rcu_read_lock();
140 	cred = current_cred();
141 	child_cred = __task_cred(child);
142 	if (mode & PTRACE_MODE_FSCREDS)
143 		caller_caps = &cred->cap_effective;
144 	else
145 		caller_caps = &cred->cap_permitted;
146 	if (cred->user_ns == child_cred->user_ns &&
147 	    cap_issubset(child_cred->cap_permitted, *caller_caps))
148 		goto out;
149 	if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE))
150 		goto out;
151 	ret = -EPERM;
152 out:
153 	rcu_read_unlock();
154 	return ret;
155 }
156 
157 /**
158  * cap_ptrace_traceme - Determine whether another process may trace the current
159  * @parent: The task proposed to be the tracer
160  *
161  * If parent is in the same or an ancestor user_ns and has all current's
162  * capabilities, then ptrace access is allowed.
163  * If parent has the ptrace capability to current's user_ns, then ptrace
164  * access is allowed.
165  * Else denied.
166  *
167  * Determine whether the nominated task is permitted to trace the current
168  * process, returning 0 if permission is granted, -ve if denied.
169  */
170 int cap_ptrace_traceme(struct task_struct *parent)
171 {
172 	int ret = 0;
173 	const struct cred *cred, *child_cred;
174 
175 	rcu_read_lock();
176 	cred = __task_cred(parent);
177 	child_cred = current_cred();
178 	if (cred->user_ns == child_cred->user_ns &&
179 	    cap_issubset(child_cred->cap_permitted, cred->cap_permitted))
180 		goto out;
181 	if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE))
182 		goto out;
183 	ret = -EPERM;
184 out:
185 	rcu_read_unlock();
186 	return ret;
187 }
188 
189 /**
190  * cap_capget - Retrieve a task's capability sets
191  * @target: The task from which to retrieve the capability sets
192  * @effective: The place to record the effective set
193  * @inheritable: The place to record the inheritable set
194  * @permitted: The place to record the permitted set
195  *
196  * This function retrieves the capabilities of the nominated task and returns
197  * them to the caller.
198  */
199 int cap_capget(struct task_struct *target, kernel_cap_t *effective,
200 	       kernel_cap_t *inheritable, kernel_cap_t *permitted)
201 {
202 	const struct cred *cred;
203 
204 	/* Derived from kernel/capability.c:sys_capget. */
205 	rcu_read_lock();
206 	cred = __task_cred(target);
207 	*effective   = cred->cap_effective;
208 	*inheritable = cred->cap_inheritable;
209 	*permitted   = cred->cap_permitted;
210 	rcu_read_unlock();
211 	return 0;
212 }
213 
214 /*
215  * Determine whether the inheritable capabilities are limited to the old
216  * permitted set.  Returns 1 if they are limited, 0 if they are not.
217  */
218 static inline int cap_inh_is_capped(void)
219 {
220 	/* they are so limited unless the current task has the CAP_SETPCAP
221 	 * capability
222 	 */
223 	if (cap_capable(current_cred(), current_cred()->user_ns,
224 			CAP_SETPCAP, CAP_OPT_NONE) == 0)
225 		return 0;
226 	return 1;
227 }
228 
229 /**
230  * cap_capset - Validate and apply proposed changes to current's capabilities
231  * @new: The proposed new credentials; alterations should be made here
232  * @old: The current task's current credentials
233  * @effective: A pointer to the proposed new effective capabilities set
234  * @inheritable: A pointer to the proposed new inheritable capabilities set
235  * @permitted: A pointer to the proposed new permitted capabilities set
236  *
237  * This function validates and applies a proposed mass change to the current
238  * process's capability sets.  The changes are made to the proposed new
239  * credentials, and assuming no error, will be committed by the caller of LSM.
240  */
241 int cap_capset(struct cred *new,
242 	       const struct cred *old,
243 	       const kernel_cap_t *effective,
244 	       const kernel_cap_t *inheritable,
245 	       const kernel_cap_t *permitted)
246 {
247 	if (cap_inh_is_capped() &&
248 	    !cap_issubset(*inheritable,
249 			  cap_combine(old->cap_inheritable,
250 				      old->cap_permitted)))
251 		/* incapable of using this inheritable set */
252 		return -EPERM;
253 
254 	if (!cap_issubset(*inheritable,
255 			  cap_combine(old->cap_inheritable,
256 				      old->cap_bset)))
257 		/* no new pI capabilities outside bounding set */
258 		return -EPERM;
259 
260 	/* verify restrictions on target's new Permitted set */
261 	if (!cap_issubset(*permitted, old->cap_permitted))
262 		return -EPERM;
263 
264 	/* verify the _new_Effective_ is a subset of the _new_Permitted_ */
265 	if (!cap_issubset(*effective, *permitted))
266 		return -EPERM;
267 
268 	new->cap_effective   = *effective;
269 	new->cap_inheritable = *inheritable;
270 	new->cap_permitted   = *permitted;
271 
272 	/*
273 	 * Mask off ambient bits that are no longer both permitted and
274 	 * inheritable.
275 	 */
276 	new->cap_ambient = cap_intersect(new->cap_ambient,
277 					 cap_intersect(*permitted,
278 						       *inheritable));
279 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
280 		return -EINVAL;
281 	return 0;
282 }
283 
284 /**
285  * cap_inode_need_killpriv - Determine if inode change affects privileges
286  * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV
287  *
288  * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV
289  * affects the security markings on that inode, and if it is, should
290  * inode_killpriv() be invoked or the change rejected.
291  *
292  * Returns 1 if security.capability has a value, meaning inode_killpriv()
293  * is required, 0 otherwise, meaning inode_killpriv() is not required.
294  */
295 int cap_inode_need_killpriv(struct dentry *dentry)
296 {
297 	struct inode *inode = d_backing_inode(dentry);
298 	int error;
299 
300 	error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0);
301 	return error > 0;
302 }
303 
304 /**
305  * cap_inode_killpriv - Erase the security markings on an inode
306  *
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  * Returns 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)
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  * If all is ok, we 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 		*ancestor;
548 	kuid_t rootid;
549 	size_t newsize;
550 
551 	if (!*ivalue)
552 		return -EINVAL;
553 	if (!validheader(size, cap))
554 		return -EINVAL;
555 	if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
556 		return -EPERM;
557 	if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
558 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
559 			/* user is privileged, just write the v2 */
560 			return size;
561 
562 	rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
563 	if (!uid_valid(rootid))
564 		return -EINVAL;
565 
566 	nsrootid = from_kuid(fs_ns, rootid);
567 	if (nsrootid == -1)
568 		return -EINVAL;
569 
570 	/*
571 	 * Do not allow allow adding a v3 filesystem capability xattr
572 	 * if the rootid field is ambiguous.
573 	 */
574 	for (ancestor = task_ns->parent; ancestor; ancestor = ancestor->parent) {
575 		if (from_kuid(ancestor, rootid) == 0)
576 			return -EINVAL;
577 	}
578 
579 	newsize = sizeof(struct vfs_ns_cap_data);
580 	nscap = kmalloc(newsize, GFP_ATOMIC);
581 	if (!nscap)
582 		return -ENOMEM;
583 	nscap->rootid = cpu_to_le32(nsrootid);
584 	nsmagic = VFS_CAP_REVISION_3;
585 	magic = le32_to_cpu(cap->magic_etc);
586 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
587 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
588 	nscap->magic_etc = cpu_to_le32(nsmagic);
589 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
590 
591 	*ivalue = nscap;
592 	return newsize;
593 }
594 
595 /*
596  * Calculate the new process capability sets from the capability sets attached
597  * to a file.
598  */
599 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
600 					  struct linux_binprm *bprm,
601 					  bool *effective,
602 					  bool *has_fcap)
603 {
604 	struct cred *new = bprm->cred;
605 	unsigned i;
606 	int ret = 0;
607 
608 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
609 		*effective = true;
610 
611 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
612 		*has_fcap = true;
613 
614 	CAP_FOR_EACH_U32(i) {
615 		__u32 permitted = caps->permitted.cap[i];
616 		__u32 inheritable = caps->inheritable.cap[i];
617 
618 		/*
619 		 * pP' = (X & fP) | (pI & fI)
620 		 * The addition of pA' is handled later.
621 		 */
622 		new->cap_permitted.cap[i] =
623 			(new->cap_bset.cap[i] & permitted) |
624 			(new->cap_inheritable.cap[i] & inheritable);
625 
626 		if (permitted & ~new->cap_permitted.cap[i])
627 			/* insufficient to execute correctly */
628 			ret = -EPERM;
629 	}
630 
631 	/*
632 	 * For legacy apps, with no internal support for recognizing they
633 	 * do not have enough capabilities, we return an error if they are
634 	 * missing some "forced" (aka file-permitted) capabilities.
635 	 */
636 	return *effective ? ret : 0;
637 }
638 
639 /**
640  * get_vfs_caps_from_disk - retrieve vfs caps from disk
641  *
642  * @mnt_userns:	user namespace of the mount the inode was found from
643  * @dentry:	dentry from which @inode is retrieved
644  * @cpu_caps:	vfs capabilities
645  *
646  * Extract the on-exec-apply capability sets for an executable file.
647  *
648  * If the inode has been found through an idmapped mount the user namespace of
649  * the vfsmount must be passed through @mnt_userns. This function will then
650  * take care to map the inode according to @mnt_userns before checking
651  * permissions. On non-idmapped mounts or if permission checking is to be
652  * performed on the raw inode simply passs init_user_ns.
653  */
654 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
655 			   const struct dentry *dentry,
656 			   struct cpu_vfs_cap_data *cpu_caps)
657 {
658 	struct inode *inode = d_backing_inode(dentry);
659 	__u32 magic_etc;
660 	unsigned tocopy, i;
661 	int size;
662 	struct vfs_ns_cap_data data, *nscaps = &data;
663 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
664 	kuid_t rootkuid;
665 	struct user_namespace *fs_ns;
666 
667 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
668 
669 	if (!inode)
670 		return -ENODATA;
671 
672 	fs_ns = inode->i_sb->s_user_ns;
673 	size = __vfs_getxattr((struct dentry *)dentry, inode,
674 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
675 	if (size == -ENODATA || size == -EOPNOTSUPP)
676 		/* no data, that's ok */
677 		return -ENODATA;
678 
679 	if (size < 0)
680 		return size;
681 
682 	if (size < sizeof(magic_etc))
683 		return -EINVAL;
684 
685 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
686 
687 	rootkuid = make_kuid(fs_ns, 0);
688 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
689 	case VFS_CAP_REVISION_1:
690 		if (size != XATTR_CAPS_SZ_1)
691 			return -EINVAL;
692 		tocopy = VFS_CAP_U32_1;
693 		break;
694 	case VFS_CAP_REVISION_2:
695 		if (size != XATTR_CAPS_SZ_2)
696 			return -EINVAL;
697 		tocopy = VFS_CAP_U32_2;
698 		break;
699 	case VFS_CAP_REVISION_3:
700 		if (size != XATTR_CAPS_SZ_3)
701 			return -EINVAL;
702 		tocopy = VFS_CAP_U32_3;
703 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
704 		break;
705 
706 	default:
707 		return -EINVAL;
708 	}
709 	/* Limit the caps to the mounter of the filesystem
710 	 * or the more limited uid specified in the xattr.
711 	 */
712 	rootkuid = kuid_into_mnt(mnt_userns, rootkuid);
713 	if (!rootid_owns_currentns(rootkuid))
714 		return -ENODATA;
715 
716 	CAP_FOR_EACH_U32(i) {
717 		if (i >= tocopy)
718 			break;
719 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
720 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
721 	}
722 
723 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
724 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
725 
726 	cpu_caps->rootid = rootkuid;
727 
728 	return 0;
729 }
730 
731 /*
732  * Attempt to get the on-exec apply capability sets for an executable file from
733  * its xattrs and, if present, apply them to the proposed credentials being
734  * constructed by execve().
735  */
736 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
737 			 bool *effective, bool *has_fcap)
738 {
739 	int rc = 0;
740 	struct cpu_vfs_cap_data vcaps;
741 
742 	cap_clear(bprm->cred->cap_permitted);
743 
744 	if (!file_caps_enabled)
745 		return 0;
746 
747 	if (!mnt_may_suid(file->f_path.mnt))
748 		return 0;
749 
750 	/*
751 	 * This check is redundant with mnt_may_suid() but is kept to make
752 	 * explicit that capability bits are limited to s_user_ns and its
753 	 * descendants.
754 	 */
755 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
756 		return 0;
757 
758 	rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
759 				    file->f_path.dentry, &vcaps);
760 	if (rc < 0) {
761 		if (rc == -EINVAL)
762 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
763 					bprm->filename);
764 		else if (rc == -ENODATA)
765 			rc = 0;
766 		goto out;
767 	}
768 
769 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
770 
771 out:
772 	if (rc)
773 		cap_clear(bprm->cred->cap_permitted);
774 
775 	return rc;
776 }
777 
778 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
779 
780 static inline bool __is_real(kuid_t uid, struct cred *cred)
781 { return uid_eq(cred->uid, uid); }
782 
783 static inline bool __is_eff(kuid_t uid, struct cred *cred)
784 { return uid_eq(cred->euid, uid); }
785 
786 static inline bool __is_suid(kuid_t uid, struct cred *cred)
787 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
788 
789 /*
790  * handle_privileged_root - Handle case of privileged root
791  * @bprm: The execution parameters, including the proposed creds
792  * @has_fcap: Are any file capabilities set?
793  * @effective: Do we have effective root privilege?
794  * @root_uid: This namespace' root UID WRT initial USER namespace
795  *
796  * Handle the case where root is privileged and hasn't been neutered by
797  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
798  * set UID root and nothing is changed.  If we are root, cap_permitted is
799  * updated.  If we have become set UID root, the effective bit is set.
800  */
801 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
802 				   bool *effective, kuid_t root_uid)
803 {
804 	const struct cred *old = current_cred();
805 	struct cred *new = bprm->cred;
806 
807 	if (!root_privileged())
808 		return;
809 	/*
810 	 * If the legacy file capability is set, then don't set privs
811 	 * for a setuid root binary run by a non-root user.  Do set it
812 	 * for a root user just to cause least surprise to an admin.
813 	 */
814 	if (has_fcap && __is_suid(root_uid, new)) {
815 		warn_setuid_and_fcaps_mixed(bprm->filename);
816 		return;
817 	}
818 	/*
819 	 * To support inheritance of root-permissions and suid-root
820 	 * executables under compatibility mode, we override the
821 	 * capability sets for the file.
822 	 */
823 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
824 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
825 		new->cap_permitted = cap_combine(old->cap_bset,
826 						 old->cap_inheritable);
827 	}
828 	/*
829 	 * If only the real uid is 0, we do not set the effective bit.
830 	 */
831 	if (__is_eff(root_uid, new))
832 		*effective = true;
833 }
834 
835 #define __cap_gained(field, target, source) \
836 	!cap_issubset(target->cap_##field, source->cap_##field)
837 #define __cap_grew(target, source, cred) \
838 	!cap_issubset(cred->cap_##target, cred->cap_##source)
839 #define __cap_full(field, cred) \
840 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
841 
842 static inline bool __is_setuid(struct cred *new, const struct cred *old)
843 { return !uid_eq(new->euid, old->uid); }
844 
845 static inline bool __is_setgid(struct cred *new, const struct cred *old)
846 { return !gid_eq(new->egid, old->gid); }
847 
848 /*
849  * 1) Audit candidate if current->cap_effective is set
850  *
851  * We do not bother to audit if 3 things are true:
852  *   1) cap_effective has all caps
853  *   2) we became root *OR* are were already root
854  *   3) root is supposed to have all caps (SECURE_NOROOT)
855  * Since this is just a normal root execing a process.
856  *
857  * Number 1 above might fail if you don't have a full bset, but I think
858  * that is interesting information to audit.
859  *
860  * A number of other conditions require logging:
861  * 2) something prevented setuid root getting all caps
862  * 3) non-setuid root gets fcaps
863  * 4) non-setuid root gets ambient
864  */
865 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
866 				     kuid_t root, bool has_fcap)
867 {
868 	bool ret = false;
869 
870 	if ((__cap_grew(effective, ambient, new) &&
871 	     !(__cap_full(effective, new) &&
872 	       (__is_eff(root, new) || __is_real(root, new)) &&
873 	       root_privileged())) ||
874 	    (root_privileged() &&
875 	     __is_suid(root, new) &&
876 	     !__cap_full(effective, new)) ||
877 	    (!__is_setuid(new, old) &&
878 	     ((has_fcap &&
879 	       __cap_gained(permitted, new, old)) ||
880 	      __cap_gained(ambient, new, old))))
881 
882 		ret = true;
883 
884 	return ret;
885 }
886 
887 /**
888  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
889  * @bprm: The execution parameters, including the proposed creds
890  * @file: The file to pull the credentials from
891  *
892  * Set up the proposed credentials for a new execution context being
893  * constructed by execve().  The proposed creds in @bprm->cred is altered,
894  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
895  */
896 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
897 {
898 	/* Process setpcap binaries and capabilities for uid 0 */
899 	const struct cred *old = current_cred();
900 	struct cred *new = bprm->cred;
901 	bool effective = false, has_fcap = false, is_setid;
902 	int ret;
903 	kuid_t root_uid;
904 
905 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
906 		return -EPERM;
907 
908 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
909 	if (ret < 0)
910 		return ret;
911 
912 	root_uid = make_kuid(new->user_ns, 0);
913 
914 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
915 
916 	/* if we have fs caps, clear dangerous personality flags */
917 	if (__cap_gained(permitted, new, old))
918 		bprm->per_clear |= PER_CLEAR_ON_SETID;
919 
920 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
921 	 * credentials unless they have the appropriate permit.
922 	 *
923 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
924 	 */
925 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
926 
927 	if ((is_setid || __cap_gained(permitted, new, old)) &&
928 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
929 	     !ptracer_capable(current, new->user_ns))) {
930 		/* downgrade; they get no more than they had, and maybe less */
931 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
932 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
933 			new->euid = new->uid;
934 			new->egid = new->gid;
935 		}
936 		new->cap_permitted = cap_intersect(new->cap_permitted,
937 						   old->cap_permitted);
938 	}
939 
940 	new->suid = new->fsuid = new->euid;
941 	new->sgid = new->fsgid = new->egid;
942 
943 	/* File caps or setid cancels ambient. */
944 	if (has_fcap || is_setid)
945 		cap_clear(new->cap_ambient);
946 
947 	/*
948 	 * Now that we've computed pA', update pP' to give:
949 	 *   pP' = (X & fP) | (pI & fI) | pA'
950 	 */
951 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
952 
953 	/*
954 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
955 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
956 	 */
957 	if (effective)
958 		new->cap_effective = new->cap_permitted;
959 	else
960 		new->cap_effective = new->cap_ambient;
961 
962 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
963 		return -EPERM;
964 
965 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
966 		ret = audit_log_bprm_fcaps(bprm, new, old);
967 		if (ret < 0)
968 			return ret;
969 	}
970 
971 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
972 
973 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
974 		return -EPERM;
975 
976 	/* Check for privilege-elevated exec. */
977 	if (is_setid ||
978 	    (!__is_real(root_uid, new) &&
979 	     (effective ||
980 	      __cap_grew(permitted, ambient, new))))
981 		bprm->secureexec = 1;
982 
983 	return 0;
984 }
985 
986 /**
987  * cap_inode_setxattr - Determine whether an xattr may be altered
988  * @dentry: The inode/dentry being altered
989  * @name: The name of the xattr to be changed
990  * @value: The value that the xattr will be changed to
991  * @size: The size of value
992  * @flags: The replacement flag
993  *
994  * Determine whether an xattr may be altered or set on an inode, returning 0 if
995  * permission is granted, -ve if denied.
996  *
997  * This is used to make sure security xattrs don't get updated or set by those
998  * who aren't privileged to do so.
999  */
1000 int cap_inode_setxattr(struct dentry *dentry, const char *name,
1001 		       const void *value, size_t size, int flags)
1002 {
1003 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1004 
1005 	/* Ignore non-security xattrs */
1006 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1007 			XATTR_SECURITY_PREFIX_LEN) != 0)
1008 		return 0;
1009 
1010 	/*
1011 	 * For XATTR_NAME_CAPS the check will be done in
1012 	 * cap_convert_nscap(), called by setxattr()
1013 	 */
1014 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
1015 		return 0;
1016 
1017 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1018 		return -EPERM;
1019 	return 0;
1020 }
1021 
1022 /**
1023  * cap_inode_removexattr - Determine whether an xattr may be removed
1024  *
1025  * @mnt_userns:	User namespace of the mount the inode was found from
1026  * @dentry:	The inode/dentry being altered
1027  * @name:	The name of the xattr to be changed
1028  *
1029  * Determine whether an xattr may be removed from an inode, returning 0 if
1030  * permission is granted, -ve if denied.
1031  *
1032  * If the inode has been found through an idmapped mount the user namespace of
1033  * the vfsmount must be passed through @mnt_userns. This function will then
1034  * take care to map the inode according to @mnt_userns before checking
1035  * permissions. On non-idmapped mounts or if permission checking is to be
1036  * performed on the raw inode simply passs init_user_ns.
1037  *
1038  * This is used to make sure security xattrs don't get removed by those who
1039  * aren't privileged to remove them.
1040  */
1041 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1042 			  struct dentry *dentry, const char *name)
1043 {
1044 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1045 
1046 	/* Ignore non-security xattrs */
1047 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1048 			XATTR_SECURITY_PREFIX_LEN) != 0)
1049 		return 0;
1050 
1051 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1052 		/* security.capability gets namespaced */
1053 		struct inode *inode = d_backing_inode(dentry);
1054 		if (!inode)
1055 			return -EINVAL;
1056 		if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1057 			return -EPERM;
1058 		return 0;
1059 	}
1060 
1061 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1062 		return -EPERM;
1063 	return 0;
1064 }
1065 
1066 /*
1067  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1068  * a process after a call to setuid, setreuid, or setresuid.
1069  *
1070  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1071  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1072  *  cleared.
1073  *
1074  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1075  *  capabilities of the process are cleared.
1076  *
1077  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1078  *  capabilities are set to the permitted capabilities.
1079  *
1080  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1081  *  never happen.
1082  *
1083  *  -astor
1084  *
1085  * cevans - New behaviour, Oct '99
1086  * A process may, via prctl(), elect to keep its capabilities when it
1087  * calls setuid() and switches away from uid==0. Both permitted and
1088  * effective sets will be retained.
1089  * Without this change, it was impossible for a daemon to drop only some
1090  * of its privilege. The call to setuid(!=0) would drop all privileges!
1091  * Keeping uid 0 is not an option because uid 0 owns too many vital
1092  * files..
1093  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1094  */
1095 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1096 {
1097 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1098 
1099 	if ((uid_eq(old->uid, root_uid) ||
1100 	     uid_eq(old->euid, root_uid) ||
1101 	     uid_eq(old->suid, root_uid)) &&
1102 	    (!uid_eq(new->uid, root_uid) &&
1103 	     !uid_eq(new->euid, root_uid) &&
1104 	     !uid_eq(new->suid, root_uid))) {
1105 		if (!issecure(SECURE_KEEP_CAPS)) {
1106 			cap_clear(new->cap_permitted);
1107 			cap_clear(new->cap_effective);
1108 		}
1109 
1110 		/*
1111 		 * Pre-ambient programs expect setresuid to nonroot followed
1112 		 * by exec to drop capabilities.  We should make sure that
1113 		 * this remains the case.
1114 		 */
1115 		cap_clear(new->cap_ambient);
1116 	}
1117 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1118 		cap_clear(new->cap_effective);
1119 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1120 		new->cap_effective = new->cap_permitted;
1121 }
1122 
1123 /**
1124  * cap_task_fix_setuid - Fix up the results of setuid() call
1125  * @new: The proposed credentials
1126  * @old: The current task's current credentials
1127  * @flags: Indications of what has changed
1128  *
1129  * Fix up the results of setuid() call before the credential changes are
1130  * actually applied, returning 0 to grant the changes, -ve to deny them.
1131  */
1132 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1133 {
1134 	switch (flags) {
1135 	case LSM_SETID_RE:
1136 	case LSM_SETID_ID:
1137 	case LSM_SETID_RES:
1138 		/* juggle the capabilities to follow [RES]UID changes unless
1139 		 * otherwise suppressed */
1140 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1141 			cap_emulate_setxuid(new, old);
1142 		break;
1143 
1144 	case LSM_SETID_FS:
1145 		/* juggle the capabilties to follow FSUID changes, unless
1146 		 * otherwise suppressed
1147 		 *
1148 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1149 		 *          if not, we might be a bit too harsh here.
1150 		 */
1151 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1152 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1153 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1154 				new->cap_effective =
1155 					cap_drop_fs_set(new->cap_effective);
1156 
1157 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1158 				new->cap_effective =
1159 					cap_raise_fs_set(new->cap_effective,
1160 							 new->cap_permitted);
1161 		}
1162 		break;
1163 
1164 	default:
1165 		return -EINVAL;
1166 	}
1167 
1168 	return 0;
1169 }
1170 
1171 /*
1172  * Rationale: code calling task_setscheduler, task_setioprio, and
1173  * task_setnice, assumes that
1174  *   . if capable(cap_sys_nice), then those actions should be allowed
1175  *   . if not capable(cap_sys_nice), but acting on your own processes,
1176  *   	then those actions should be allowed
1177  * This is insufficient now since you can call code without suid, but
1178  * yet with increased caps.
1179  * So we check for increased caps on the target process.
1180  */
1181 static int cap_safe_nice(struct task_struct *p)
1182 {
1183 	int is_subset, ret = 0;
1184 
1185 	rcu_read_lock();
1186 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1187 				 current_cred()->cap_permitted);
1188 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1189 		ret = -EPERM;
1190 	rcu_read_unlock();
1191 
1192 	return ret;
1193 }
1194 
1195 /**
1196  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1197  * @p: The task to affect
1198  *
1199  * Detemine if the requested scheduler policy change is permitted for the
1200  * specified task, returning 0 if permission is granted, -ve if denied.
1201  */
1202 int cap_task_setscheduler(struct task_struct *p)
1203 {
1204 	return cap_safe_nice(p);
1205 }
1206 
1207 /**
1208  * cap_task_ioprio - Detemine if I/O priority change is permitted
1209  * @p: The task to affect
1210  * @ioprio: The I/O priority to set
1211  *
1212  * Detemine if the requested I/O priority change is permitted for the specified
1213  * task, returning 0 if permission is granted, -ve if denied.
1214  */
1215 int cap_task_setioprio(struct task_struct *p, int ioprio)
1216 {
1217 	return cap_safe_nice(p);
1218 }
1219 
1220 /**
1221  * cap_task_ioprio - Detemine if task priority change is permitted
1222  * @p: The task to affect
1223  * @nice: The nice value to set
1224  *
1225  * Detemine if the requested task priority change is permitted for the
1226  * specified task, returning 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, @arg3, @arg4, @arg5: The argument data for this function
1257  *
1258  * Allow process control functions (sys_prctl()) to alter capabilities; may
1259  * also deny access to other functions not otherwise implemented here.
1260  *
1261  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1262  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1263  * modules will consider performing the function.
1264  */
1265 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1266 		   unsigned long arg4, unsigned long arg5)
1267 {
1268 	const struct cred *old = current_cred();
1269 	struct cred *new;
1270 
1271 	switch (option) {
1272 	case PR_CAPBSET_READ:
1273 		if (!cap_valid(arg2))
1274 			return -EINVAL;
1275 		return !!cap_raised(old->cap_bset, arg2);
1276 
1277 	case PR_CAPBSET_DROP:
1278 		return cap_prctl_drop(arg2);
1279 
1280 	/*
1281 	 * The next four prctl's remain to assist with transitioning a
1282 	 * system from legacy UID=0 based privilege (when filesystem
1283 	 * capabilities are not in use) to a system using filesystem
1284 	 * capabilities only - as the POSIX.1e draft intended.
1285 	 *
1286 	 * Note:
1287 	 *
1288 	 *  PR_SET_SECUREBITS =
1289 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1290 	 *    | issecure_mask(SECURE_NOROOT)
1291 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1292 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1293 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1294 	 *
1295 	 * will ensure that the current process and all of its
1296 	 * children will be locked into a pure
1297 	 * capability-based-privilege environment.
1298 	 */
1299 	case PR_SET_SECUREBITS:
1300 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1301 		     & (old->securebits ^ arg2))			/*[1]*/
1302 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1303 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1304 		    || (cap_capable(current_cred(),
1305 				    current_cred()->user_ns,
1306 				    CAP_SETPCAP,
1307 				    CAP_OPT_NONE) != 0)			/*[4]*/
1308 			/*
1309 			 * [1] no changing of bits that are locked
1310 			 * [2] no unlocking of locks
1311 			 * [3] no setting of unsupported bits
1312 			 * [4] doing anything requires privilege (go read about
1313 			 *     the "sendmail capabilities bug")
1314 			 */
1315 		    )
1316 			/* cannot change a locked bit */
1317 			return -EPERM;
1318 
1319 		new = prepare_creds();
1320 		if (!new)
1321 			return -ENOMEM;
1322 		new->securebits = arg2;
1323 		return commit_creds(new);
1324 
1325 	case PR_GET_SECUREBITS:
1326 		return old->securebits;
1327 
1328 	case PR_GET_KEEPCAPS:
1329 		return !!issecure(SECURE_KEEP_CAPS);
1330 
1331 	case PR_SET_KEEPCAPS:
1332 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1333 			return -EINVAL;
1334 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1335 			return -EPERM;
1336 
1337 		new = prepare_creds();
1338 		if (!new)
1339 			return -ENOMEM;
1340 		if (arg2)
1341 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1342 		else
1343 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1344 		return commit_creds(new);
1345 
1346 	case PR_CAP_AMBIENT:
1347 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1348 			if (arg3 | arg4 | arg5)
1349 				return -EINVAL;
1350 
1351 			new = prepare_creds();
1352 			if (!new)
1353 				return -ENOMEM;
1354 			cap_clear(new->cap_ambient);
1355 			return commit_creds(new);
1356 		}
1357 
1358 		if (((!cap_valid(arg3)) | arg4 | arg5))
1359 			return -EINVAL;
1360 
1361 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1362 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1363 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1364 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1365 			return -EINVAL;
1366 		} else {
1367 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1368 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1369 			     !cap_raised(current_cred()->cap_inheritable,
1370 					 arg3) ||
1371 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1372 				return -EPERM;
1373 
1374 			new = prepare_creds();
1375 			if (!new)
1376 				return -ENOMEM;
1377 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1378 				cap_raise(new->cap_ambient, arg3);
1379 			else
1380 				cap_lower(new->cap_ambient, arg3);
1381 			return commit_creds(new);
1382 		}
1383 
1384 	default:
1385 		/* No functionality available - continue with default */
1386 		return -ENOSYS;
1387 	}
1388 }
1389 
1390 /**
1391  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1392  * @mm: The VM space in which the new mapping is to be made
1393  * @pages: The size of the mapping
1394  *
1395  * Determine whether the allocation of a new virtual mapping by the current
1396  * task is permitted, returning 1 if permission is granted, 0 if not.
1397  */
1398 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1399 {
1400 	int cap_sys_admin = 0;
1401 
1402 	if (cap_capable(current_cred(), &init_user_ns,
1403 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1404 		cap_sys_admin = 1;
1405 
1406 	return cap_sys_admin;
1407 }
1408 
1409 /*
1410  * cap_mmap_addr - check if able to map given addr
1411  * @addr: address attempting to be mapped
1412  *
1413  * If the process is attempting to map memory below dac_mmap_min_addr they need
1414  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1415  * capability security module.  Returns 0 if this mapping should be allowed
1416  * -EPERM if not.
1417  */
1418 int cap_mmap_addr(unsigned long addr)
1419 {
1420 	int ret = 0;
1421 
1422 	if (addr < dac_mmap_min_addr) {
1423 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1424 				  CAP_OPT_NONE);
1425 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1426 		if (ret == 0)
1427 			current->flags |= PF_SUPERPRIV;
1428 	}
1429 	return ret;
1430 }
1431 
1432 int cap_mmap_file(struct file *file, unsigned long reqprot,
1433 		  unsigned long prot, unsigned long flags)
1434 {
1435 	return 0;
1436 }
1437 
1438 #ifdef CONFIG_SECURITY
1439 
1440 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1441 	LSM_HOOK_INIT(capable, cap_capable),
1442 	LSM_HOOK_INIT(settime, cap_settime),
1443 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1444 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1445 	LSM_HOOK_INIT(capget, cap_capget),
1446 	LSM_HOOK_INIT(capset, cap_capset),
1447 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1448 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1449 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1450 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1451 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1452 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1453 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1454 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1455 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1456 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1457 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1458 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1459 };
1460 
1461 static int __init capability_init(void)
1462 {
1463 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1464 				"capability");
1465 	return 0;
1466 }
1467 
1468 DEFINE_LSM(capability) = {
1469 	.name = "capability",
1470 	.order = LSM_ORDER_FIRST,
1471 	.init = capability_init,
1472 };
1473 
1474 #endif /* CONFIG_SECURITY */
1475