xref: /linux-6.15/security/commoncap.c (revision 01eadc8d)
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 	uid_t root, mappedroot;
384 	char *tmpbuf = NULL;
385 	struct vfs_cap_data *cap;
386 	struct vfs_ns_cap_data *nscap;
387 	struct dentry *dentry;
388 	struct user_namespace *fs_ns;
389 
390 	if (strcmp(name, "capability") != 0)
391 		return -EOPNOTSUPP;
392 
393 	dentry = d_find_any_alias(inode);
394 	if (!dentry)
395 		return -EINVAL;
396 
397 	size = sizeof(struct vfs_ns_cap_data);
398 	ret = (int)vfs_getxattr_alloc(mnt_userns, dentry, XATTR_NAME_CAPS,
399 				      &tmpbuf, size, GFP_NOFS);
400 	dput(dentry);
401 
402 	if (ret < 0)
403 		return ret;
404 
405 	fs_ns = inode->i_sb->s_user_ns;
406 	cap = (struct vfs_cap_data *) tmpbuf;
407 	if (is_v2header((size_t) ret, cap)) {
408 		/* If this is sizeof(vfs_cap_data) then we're ok with the
409 		 * on-disk value, so return that.  */
410 		if (alloc)
411 			*buffer = tmpbuf;
412 		else
413 			kfree(tmpbuf);
414 		return ret;
415 	} else if (!is_v3header((size_t) ret, cap)) {
416 		kfree(tmpbuf);
417 		return -EINVAL;
418 	}
419 
420 	nscap = (struct vfs_ns_cap_data *) tmpbuf;
421 	root = le32_to_cpu(nscap->rootid);
422 	kroot = make_kuid(fs_ns, root);
423 
424 	/* If this is an idmapped mount shift the kuid. */
425 	kroot = kuid_into_mnt(mnt_userns, kroot);
426 
427 	/* If the root kuid maps to a valid uid in current ns, then return
428 	 * this as a nscap. */
429 	mappedroot = from_kuid(current_user_ns(), kroot);
430 	if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) {
431 		if (alloc) {
432 			*buffer = tmpbuf;
433 			nscap->rootid = cpu_to_le32(mappedroot);
434 		} else
435 			kfree(tmpbuf);
436 		return size;
437 	}
438 
439 	if (!rootid_owns_currentns(kroot)) {
440 		kfree(tmpbuf);
441 		return -EOPNOTSUPP;
442 	}
443 
444 	/* This comes from a parent namespace.  Return as a v2 capability */
445 	size = sizeof(struct vfs_cap_data);
446 	if (alloc) {
447 		*buffer = kmalloc(size, GFP_ATOMIC);
448 		if (*buffer) {
449 			struct vfs_cap_data *cap = *buffer;
450 			__le32 nsmagic, magic;
451 			magic = VFS_CAP_REVISION_2;
452 			nsmagic = le32_to_cpu(nscap->magic_etc);
453 			if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE)
454 				magic |= VFS_CAP_FLAGS_EFFECTIVE;
455 			memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
456 			cap->magic_etc = cpu_to_le32(magic);
457 		} else {
458 			size = -ENOMEM;
459 		}
460 	}
461 	kfree(tmpbuf);
462 	return size;
463 }
464 
465 /**
466  * rootid_from_xattr - translate root uid of vfs caps
467  *
468  * @value:	vfs caps value which may be modified by this function
469  * @size:	size of @ivalue
470  * @task_ns:	user namespace of the caller
471  * @mnt_userns:	user namespace of the mount the inode was found from
472  *
473  * If the inode has been found through an idmapped mount the user namespace of
474  * the vfsmount must be passed through @mnt_userns. This function will then
475  * take care to map the inode according to @mnt_userns before checking
476  * permissions. On non-idmapped mounts or if permission checking is to be
477  * performed on the raw inode simply passs init_user_ns.
478  */
479 static kuid_t rootid_from_xattr(const void *value, size_t size,
480 				struct user_namespace *task_ns,
481 				struct user_namespace *mnt_userns)
482 {
483 	const struct vfs_ns_cap_data *nscap = value;
484 	kuid_t rootkid;
485 	uid_t rootid = 0;
486 
487 	if (size == XATTR_CAPS_SZ_3)
488 		rootid = le32_to_cpu(nscap->rootid);
489 
490 	rootkid = make_kuid(task_ns, rootid);
491 	return kuid_from_mnt(mnt_userns, rootkid);
492 }
493 
494 static bool validheader(size_t size, const struct vfs_cap_data *cap)
495 {
496 	return is_v2header(size, cap) || is_v3header(size, cap);
497 }
498 
499 /**
500  * cap_convert_nscap - check vfs caps
501  *
502  * @mnt_userns:	user namespace of the mount the inode was found from
503  * @dentry:	used to retrieve inode to check permissions on
504  * @ivalue:	vfs caps value which may be modified by this function
505  * @size:	size of @ivalue
506  *
507  * User requested a write of security.capability.  If needed, update the
508  * xattr to change from v2 to v3, or to fixup the v3 rootid.
509  *
510  * If the inode has been found through an idmapped mount the user namespace of
511  * the vfsmount must be passed through @mnt_userns. This function will then
512  * take care to map the inode according to @mnt_userns before checking
513  * permissions. On non-idmapped mounts or if permission checking is to be
514  * performed on the raw inode simply passs init_user_ns.
515  *
516  * If all is ok, we return the new size, on error return < 0.
517  */
518 int cap_convert_nscap(struct user_namespace *mnt_userns, struct dentry *dentry,
519 		      const void **ivalue, size_t size)
520 {
521 	struct vfs_ns_cap_data *nscap;
522 	uid_t nsrootid;
523 	const struct vfs_cap_data *cap = *ivalue;
524 	__u32 magic, nsmagic;
525 	struct inode *inode = d_backing_inode(dentry);
526 	struct user_namespace *task_ns = current_user_ns(),
527 		*fs_ns = inode->i_sb->s_user_ns;
528 	kuid_t rootid;
529 	size_t newsize;
530 
531 	if (!*ivalue)
532 		return -EINVAL;
533 	if (!validheader(size, cap))
534 		return -EINVAL;
535 	if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
536 		return -EPERM;
537 	if (size == XATTR_CAPS_SZ_2 && (mnt_userns == &init_user_ns))
538 		if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP))
539 			/* user is privileged, just write the v2 */
540 			return size;
541 
542 	rootid = rootid_from_xattr(*ivalue, size, task_ns, mnt_userns);
543 	if (!uid_valid(rootid))
544 		return -EINVAL;
545 
546 	nsrootid = from_kuid(fs_ns, rootid);
547 	if (nsrootid == -1)
548 		return -EINVAL;
549 
550 	newsize = sizeof(struct vfs_ns_cap_data);
551 	nscap = kmalloc(newsize, GFP_ATOMIC);
552 	if (!nscap)
553 		return -ENOMEM;
554 	nscap->rootid = cpu_to_le32(nsrootid);
555 	nsmagic = VFS_CAP_REVISION_3;
556 	magic = le32_to_cpu(cap->magic_etc);
557 	if (magic & VFS_CAP_FLAGS_EFFECTIVE)
558 		nsmagic |= VFS_CAP_FLAGS_EFFECTIVE;
559 	nscap->magic_etc = cpu_to_le32(nsmagic);
560 	memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32);
561 
562 	*ivalue = nscap;
563 	return newsize;
564 }
565 
566 /*
567  * Calculate the new process capability sets from the capability sets attached
568  * to a file.
569  */
570 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
571 					  struct linux_binprm *bprm,
572 					  bool *effective,
573 					  bool *has_fcap)
574 {
575 	struct cred *new = bprm->cred;
576 	unsigned i;
577 	int ret = 0;
578 
579 	if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
580 		*effective = true;
581 
582 	if (caps->magic_etc & VFS_CAP_REVISION_MASK)
583 		*has_fcap = true;
584 
585 	CAP_FOR_EACH_U32(i) {
586 		__u32 permitted = caps->permitted.cap[i];
587 		__u32 inheritable = caps->inheritable.cap[i];
588 
589 		/*
590 		 * pP' = (X & fP) | (pI & fI)
591 		 * The addition of pA' is handled later.
592 		 */
593 		new->cap_permitted.cap[i] =
594 			(new->cap_bset.cap[i] & permitted) |
595 			(new->cap_inheritable.cap[i] & inheritable);
596 
597 		if (permitted & ~new->cap_permitted.cap[i])
598 			/* insufficient to execute correctly */
599 			ret = -EPERM;
600 	}
601 
602 	/*
603 	 * For legacy apps, with no internal support for recognizing they
604 	 * do not have enough capabilities, we return an error if they are
605 	 * missing some "forced" (aka file-permitted) capabilities.
606 	 */
607 	return *effective ? ret : 0;
608 }
609 
610 /**
611  * get_vfs_caps_from_disk - retrieve vfs caps from disk
612  *
613  * @mnt_userns:	user namespace of the mount the inode was found from
614  * @dentry:	dentry from which @inode is retrieved
615  * @cpu_caps:	vfs capabilities
616  *
617  * Extract the on-exec-apply capability sets for an executable file.
618  *
619  * If the inode has been found through an idmapped mount the user namespace of
620  * the vfsmount must be passed through @mnt_userns. This function will then
621  * take care to map the inode according to @mnt_userns before checking
622  * permissions. On non-idmapped mounts or if permission checking is to be
623  * performed on the raw inode simply passs init_user_ns.
624  */
625 int get_vfs_caps_from_disk(struct user_namespace *mnt_userns,
626 			   const struct dentry *dentry,
627 			   struct cpu_vfs_cap_data *cpu_caps)
628 {
629 	struct inode *inode = d_backing_inode(dentry);
630 	__u32 magic_etc;
631 	unsigned tocopy, i;
632 	int size;
633 	struct vfs_ns_cap_data data, *nscaps = &data;
634 	struct vfs_cap_data *caps = (struct vfs_cap_data *) &data;
635 	kuid_t rootkuid;
636 	struct user_namespace *fs_ns;
637 
638 	memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
639 
640 	if (!inode)
641 		return -ENODATA;
642 
643 	fs_ns = inode->i_sb->s_user_ns;
644 	size = __vfs_getxattr((struct dentry *)dentry, inode,
645 			      XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ);
646 	if (size == -ENODATA || size == -EOPNOTSUPP)
647 		/* no data, that's ok */
648 		return -ENODATA;
649 
650 	if (size < 0)
651 		return size;
652 
653 	if (size < sizeof(magic_etc))
654 		return -EINVAL;
655 
656 	cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc);
657 
658 	rootkuid = make_kuid(fs_ns, 0);
659 	switch (magic_etc & VFS_CAP_REVISION_MASK) {
660 	case VFS_CAP_REVISION_1:
661 		if (size != XATTR_CAPS_SZ_1)
662 			return -EINVAL;
663 		tocopy = VFS_CAP_U32_1;
664 		break;
665 	case VFS_CAP_REVISION_2:
666 		if (size != XATTR_CAPS_SZ_2)
667 			return -EINVAL;
668 		tocopy = VFS_CAP_U32_2;
669 		break;
670 	case VFS_CAP_REVISION_3:
671 		if (size != XATTR_CAPS_SZ_3)
672 			return -EINVAL;
673 		tocopy = VFS_CAP_U32_3;
674 		rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid));
675 		break;
676 
677 	default:
678 		return -EINVAL;
679 	}
680 	/* Limit the caps to the mounter of the filesystem
681 	 * or the more limited uid specified in the xattr.
682 	 */
683 	rootkuid = kuid_into_mnt(mnt_userns, rootkuid);
684 	if (!rootid_owns_currentns(rootkuid))
685 		return -ENODATA;
686 
687 	CAP_FOR_EACH_U32(i) {
688 		if (i >= tocopy)
689 			break;
690 		cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted);
691 		cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable);
692 	}
693 
694 	cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
695 	cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK;
696 
697 	cpu_caps->rootid = rootkuid;
698 
699 	return 0;
700 }
701 
702 /*
703  * Attempt to get the on-exec apply capability sets for an executable file from
704  * its xattrs and, if present, apply them to the proposed credentials being
705  * constructed by execve().
706  */
707 static int get_file_caps(struct linux_binprm *bprm, struct file *file,
708 			 bool *effective, bool *has_fcap)
709 {
710 	int rc = 0;
711 	struct cpu_vfs_cap_data vcaps;
712 
713 	cap_clear(bprm->cred->cap_permitted);
714 
715 	if (!file_caps_enabled)
716 		return 0;
717 
718 	if (!mnt_may_suid(file->f_path.mnt))
719 		return 0;
720 
721 	/*
722 	 * This check is redundant with mnt_may_suid() but is kept to make
723 	 * explicit that capability bits are limited to s_user_ns and its
724 	 * descendants.
725 	 */
726 	if (!current_in_userns(file->f_path.mnt->mnt_sb->s_user_ns))
727 		return 0;
728 
729 	rc = get_vfs_caps_from_disk(file_mnt_user_ns(file),
730 				    file->f_path.dentry, &vcaps);
731 	if (rc < 0) {
732 		if (rc == -EINVAL)
733 			printk(KERN_NOTICE "Invalid argument reading file caps for %s\n",
734 					bprm->filename);
735 		else if (rc == -ENODATA)
736 			rc = 0;
737 		goto out;
738 	}
739 
740 	rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_fcap);
741 
742 out:
743 	if (rc)
744 		cap_clear(bprm->cred->cap_permitted);
745 
746 	return rc;
747 }
748 
749 static inline bool root_privileged(void) { return !issecure(SECURE_NOROOT); }
750 
751 static inline bool __is_real(kuid_t uid, struct cred *cred)
752 { return uid_eq(cred->uid, uid); }
753 
754 static inline bool __is_eff(kuid_t uid, struct cred *cred)
755 { return uid_eq(cred->euid, uid); }
756 
757 static inline bool __is_suid(kuid_t uid, struct cred *cred)
758 { return !__is_real(uid, cred) && __is_eff(uid, cred); }
759 
760 /*
761  * handle_privileged_root - Handle case of privileged root
762  * @bprm: The execution parameters, including the proposed creds
763  * @has_fcap: Are any file capabilities set?
764  * @effective: Do we have effective root privilege?
765  * @root_uid: This namespace' root UID WRT initial USER namespace
766  *
767  * Handle the case where root is privileged and hasn't been neutered by
768  * SECURE_NOROOT.  If file capabilities are set, they won't be combined with
769  * set UID root and nothing is changed.  If we are root, cap_permitted is
770  * updated.  If we have become set UID root, the effective bit is set.
771  */
772 static void handle_privileged_root(struct linux_binprm *bprm, bool has_fcap,
773 				   bool *effective, kuid_t root_uid)
774 {
775 	const struct cred *old = current_cred();
776 	struct cred *new = bprm->cred;
777 
778 	if (!root_privileged())
779 		return;
780 	/*
781 	 * If the legacy file capability is set, then don't set privs
782 	 * for a setuid root binary run by a non-root user.  Do set it
783 	 * for a root user just to cause least surprise to an admin.
784 	 */
785 	if (has_fcap && __is_suid(root_uid, new)) {
786 		warn_setuid_and_fcaps_mixed(bprm->filename);
787 		return;
788 	}
789 	/*
790 	 * To support inheritance of root-permissions and suid-root
791 	 * executables under compatibility mode, we override the
792 	 * capability sets for the file.
793 	 */
794 	if (__is_eff(root_uid, new) || __is_real(root_uid, new)) {
795 		/* pP' = (cap_bset & ~0) | (pI & ~0) */
796 		new->cap_permitted = cap_combine(old->cap_bset,
797 						 old->cap_inheritable);
798 	}
799 	/*
800 	 * If only the real uid is 0, we do not set the effective bit.
801 	 */
802 	if (__is_eff(root_uid, new))
803 		*effective = true;
804 }
805 
806 #define __cap_gained(field, target, source) \
807 	!cap_issubset(target->cap_##field, source->cap_##field)
808 #define __cap_grew(target, source, cred) \
809 	!cap_issubset(cred->cap_##target, cred->cap_##source)
810 #define __cap_full(field, cred) \
811 	cap_issubset(CAP_FULL_SET, cred->cap_##field)
812 
813 static inline bool __is_setuid(struct cred *new, const struct cred *old)
814 { return !uid_eq(new->euid, old->uid); }
815 
816 static inline bool __is_setgid(struct cred *new, const struct cred *old)
817 { return !gid_eq(new->egid, old->gid); }
818 
819 /*
820  * 1) Audit candidate if current->cap_effective is set
821  *
822  * We do not bother to audit if 3 things are true:
823  *   1) cap_effective has all caps
824  *   2) we became root *OR* are were already root
825  *   3) root is supposed to have all caps (SECURE_NOROOT)
826  * Since this is just a normal root execing a process.
827  *
828  * Number 1 above might fail if you don't have a full bset, but I think
829  * that is interesting information to audit.
830  *
831  * A number of other conditions require logging:
832  * 2) something prevented setuid root getting all caps
833  * 3) non-setuid root gets fcaps
834  * 4) non-setuid root gets ambient
835  */
836 static inline bool nonroot_raised_pE(struct cred *new, const struct cred *old,
837 				     kuid_t root, bool has_fcap)
838 {
839 	bool ret = false;
840 
841 	if ((__cap_grew(effective, ambient, new) &&
842 	     !(__cap_full(effective, new) &&
843 	       (__is_eff(root, new) || __is_real(root, new)) &&
844 	       root_privileged())) ||
845 	    (root_privileged() &&
846 	     __is_suid(root, new) &&
847 	     !__cap_full(effective, new)) ||
848 	    (!__is_setuid(new, old) &&
849 	     ((has_fcap &&
850 	       __cap_gained(permitted, new, old)) ||
851 	      __cap_gained(ambient, new, old))))
852 
853 		ret = true;
854 
855 	return ret;
856 }
857 
858 /**
859  * cap_bprm_creds_from_file - Set up the proposed credentials for execve().
860  * @bprm: The execution parameters, including the proposed creds
861  * @file: The file to pull the credentials from
862  *
863  * Set up the proposed credentials for a new execution context being
864  * constructed by execve().  The proposed creds in @bprm->cred is altered,
865  * which won't take effect immediately.  Returns 0 if successful, -ve on error.
866  */
867 int cap_bprm_creds_from_file(struct linux_binprm *bprm, struct file *file)
868 {
869 	/* Process setpcap binaries and capabilities for uid 0 */
870 	const struct cred *old = current_cred();
871 	struct cred *new = bprm->cred;
872 	bool effective = false, has_fcap = false, is_setid;
873 	int ret;
874 	kuid_t root_uid;
875 
876 	if (WARN_ON(!cap_ambient_invariant_ok(old)))
877 		return -EPERM;
878 
879 	ret = get_file_caps(bprm, file, &effective, &has_fcap);
880 	if (ret < 0)
881 		return ret;
882 
883 	root_uid = make_kuid(new->user_ns, 0);
884 
885 	handle_privileged_root(bprm, has_fcap, &effective, root_uid);
886 
887 	/* if we have fs caps, clear dangerous personality flags */
888 	if (__cap_gained(permitted, new, old))
889 		bprm->per_clear |= PER_CLEAR_ON_SETID;
890 
891 	/* Don't let someone trace a set[ug]id/setpcap binary with the revised
892 	 * credentials unless they have the appropriate permit.
893 	 *
894 	 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs.
895 	 */
896 	is_setid = __is_setuid(new, old) || __is_setgid(new, old);
897 
898 	if ((is_setid || __cap_gained(permitted, new, old)) &&
899 	    ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) ||
900 	     !ptracer_capable(current, new->user_ns))) {
901 		/* downgrade; they get no more than they had, and maybe less */
902 		if (!ns_capable(new->user_ns, CAP_SETUID) ||
903 		    (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) {
904 			new->euid = new->uid;
905 			new->egid = new->gid;
906 		}
907 		new->cap_permitted = cap_intersect(new->cap_permitted,
908 						   old->cap_permitted);
909 	}
910 
911 	new->suid = new->fsuid = new->euid;
912 	new->sgid = new->fsgid = new->egid;
913 
914 	/* File caps or setid cancels ambient. */
915 	if (has_fcap || is_setid)
916 		cap_clear(new->cap_ambient);
917 
918 	/*
919 	 * Now that we've computed pA', update pP' to give:
920 	 *   pP' = (X & fP) | (pI & fI) | pA'
921 	 */
922 	new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient);
923 
924 	/*
925 	 * Set pE' = (fE ? pP' : pA').  Because pA' is zero if fE is set,
926 	 * this is the same as pE' = (fE ? pP' : 0) | pA'.
927 	 */
928 	if (effective)
929 		new->cap_effective = new->cap_permitted;
930 	else
931 		new->cap_effective = new->cap_ambient;
932 
933 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
934 		return -EPERM;
935 
936 	if (nonroot_raised_pE(new, old, root_uid, has_fcap)) {
937 		ret = audit_log_bprm_fcaps(bprm, new, old);
938 		if (ret < 0)
939 			return ret;
940 	}
941 
942 	new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
943 
944 	if (WARN_ON(!cap_ambient_invariant_ok(new)))
945 		return -EPERM;
946 
947 	/* Check for privilege-elevated exec. */
948 	if (is_setid ||
949 	    (!__is_real(root_uid, new) &&
950 	     (effective ||
951 	      __cap_grew(permitted, ambient, new))))
952 		bprm->secureexec = 1;
953 
954 	return 0;
955 }
956 
957 /**
958  * cap_inode_setxattr - Determine whether an xattr may be altered
959  * @dentry: The inode/dentry being altered
960  * @name: The name of the xattr to be changed
961  * @value: The value that the xattr will be changed to
962  * @size: The size of value
963  * @flags: The replacement flag
964  *
965  * Determine whether an xattr may be altered or set on an inode, returning 0 if
966  * permission is granted, -ve if denied.
967  *
968  * This is used to make sure security xattrs don't get updated or set by those
969  * who aren't privileged to do so.
970  */
971 int cap_inode_setxattr(struct dentry *dentry, const char *name,
972 		       const void *value, size_t size, int flags)
973 {
974 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
975 
976 	/* Ignore non-security xattrs */
977 	if (strncmp(name, XATTR_SECURITY_PREFIX,
978 			XATTR_SECURITY_PREFIX_LEN) != 0)
979 		return 0;
980 
981 	/*
982 	 * For XATTR_NAME_CAPS the check will be done in
983 	 * cap_convert_nscap(), called by setxattr()
984 	 */
985 	if (strcmp(name, XATTR_NAME_CAPS) == 0)
986 		return 0;
987 
988 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
989 		return -EPERM;
990 	return 0;
991 }
992 
993 /**
994  * cap_inode_removexattr - Determine whether an xattr may be removed
995  *
996  * @mnt_userns:	User namespace of the mount the inode was found from
997  * @dentry:	The inode/dentry being altered
998  * @name:	The name of the xattr to be changed
999  *
1000  * Determine whether an xattr may be removed from an inode, returning 0 if
1001  * permission is granted, -ve if denied.
1002  *
1003  * If the inode has been found through an idmapped mount the user namespace of
1004  * the vfsmount must be passed through @mnt_userns. This function will then
1005  * take care to map the inode according to @mnt_userns before checking
1006  * permissions. On non-idmapped mounts or if permission checking is to be
1007  * performed on the raw inode simply passs init_user_ns.
1008  *
1009  * This is used to make sure security xattrs don't get removed by those who
1010  * aren't privileged to remove them.
1011  */
1012 int cap_inode_removexattr(struct user_namespace *mnt_userns,
1013 			  struct dentry *dentry, const char *name)
1014 {
1015 	struct user_namespace *user_ns = dentry->d_sb->s_user_ns;
1016 
1017 	/* Ignore non-security xattrs */
1018 	if (strncmp(name, XATTR_SECURITY_PREFIX,
1019 			XATTR_SECURITY_PREFIX_LEN) != 0)
1020 		return 0;
1021 
1022 	if (strcmp(name, XATTR_NAME_CAPS) == 0) {
1023 		/* security.capability gets namespaced */
1024 		struct inode *inode = d_backing_inode(dentry);
1025 		if (!inode)
1026 			return -EINVAL;
1027 		if (!capable_wrt_inode_uidgid(mnt_userns, inode, CAP_SETFCAP))
1028 			return -EPERM;
1029 		return 0;
1030 	}
1031 
1032 	if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1033 		return -EPERM;
1034 	return 0;
1035 }
1036 
1037 /*
1038  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
1039  * a process after a call to setuid, setreuid, or setresuid.
1040  *
1041  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
1042  *  {r,e,s}uid != 0, the permitted and effective capabilities are
1043  *  cleared.
1044  *
1045  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
1046  *  capabilities of the process are cleared.
1047  *
1048  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
1049  *  capabilities are set to the permitted capabilities.
1050  *
1051  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
1052  *  never happen.
1053  *
1054  *  -astor
1055  *
1056  * cevans - New behaviour, Oct '99
1057  * A process may, via prctl(), elect to keep its capabilities when it
1058  * calls setuid() and switches away from uid==0. Both permitted and
1059  * effective sets will be retained.
1060  * Without this change, it was impossible for a daemon to drop only some
1061  * of its privilege. The call to setuid(!=0) would drop all privileges!
1062  * Keeping uid 0 is not an option because uid 0 owns too many vital
1063  * files..
1064  * Thanks to Olaf Kirch and Peter Benie for spotting this.
1065  */
1066 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old)
1067 {
1068 	kuid_t root_uid = make_kuid(old->user_ns, 0);
1069 
1070 	if ((uid_eq(old->uid, root_uid) ||
1071 	     uid_eq(old->euid, root_uid) ||
1072 	     uid_eq(old->suid, root_uid)) &&
1073 	    (!uid_eq(new->uid, root_uid) &&
1074 	     !uid_eq(new->euid, root_uid) &&
1075 	     !uid_eq(new->suid, root_uid))) {
1076 		if (!issecure(SECURE_KEEP_CAPS)) {
1077 			cap_clear(new->cap_permitted);
1078 			cap_clear(new->cap_effective);
1079 		}
1080 
1081 		/*
1082 		 * Pre-ambient programs expect setresuid to nonroot followed
1083 		 * by exec to drop capabilities.  We should make sure that
1084 		 * this remains the case.
1085 		 */
1086 		cap_clear(new->cap_ambient);
1087 	}
1088 	if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid))
1089 		cap_clear(new->cap_effective);
1090 	if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid))
1091 		new->cap_effective = new->cap_permitted;
1092 }
1093 
1094 /**
1095  * cap_task_fix_setuid - Fix up the results of setuid() call
1096  * @new: The proposed credentials
1097  * @old: The current task's current credentials
1098  * @flags: Indications of what has changed
1099  *
1100  * Fix up the results of setuid() call before the credential changes are
1101  * actually applied, returning 0 to grant the changes, -ve to deny them.
1102  */
1103 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags)
1104 {
1105 	switch (flags) {
1106 	case LSM_SETID_RE:
1107 	case LSM_SETID_ID:
1108 	case LSM_SETID_RES:
1109 		/* juggle the capabilities to follow [RES]UID changes unless
1110 		 * otherwise suppressed */
1111 		if (!issecure(SECURE_NO_SETUID_FIXUP))
1112 			cap_emulate_setxuid(new, old);
1113 		break;
1114 
1115 	case LSM_SETID_FS:
1116 		/* juggle the capabilties to follow FSUID changes, unless
1117 		 * otherwise suppressed
1118 		 *
1119 		 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
1120 		 *          if not, we might be a bit too harsh here.
1121 		 */
1122 		if (!issecure(SECURE_NO_SETUID_FIXUP)) {
1123 			kuid_t root_uid = make_kuid(old->user_ns, 0);
1124 			if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid))
1125 				new->cap_effective =
1126 					cap_drop_fs_set(new->cap_effective);
1127 
1128 			if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid))
1129 				new->cap_effective =
1130 					cap_raise_fs_set(new->cap_effective,
1131 							 new->cap_permitted);
1132 		}
1133 		break;
1134 
1135 	default:
1136 		return -EINVAL;
1137 	}
1138 
1139 	return 0;
1140 }
1141 
1142 /*
1143  * Rationale: code calling task_setscheduler, task_setioprio, and
1144  * task_setnice, assumes that
1145  *   . if capable(cap_sys_nice), then those actions should be allowed
1146  *   . if not capable(cap_sys_nice), but acting on your own processes,
1147  *   	then those actions should be allowed
1148  * This is insufficient now since you can call code without suid, but
1149  * yet with increased caps.
1150  * So we check for increased caps on the target process.
1151  */
1152 static int cap_safe_nice(struct task_struct *p)
1153 {
1154 	int is_subset, ret = 0;
1155 
1156 	rcu_read_lock();
1157 	is_subset = cap_issubset(__task_cred(p)->cap_permitted,
1158 				 current_cred()->cap_permitted);
1159 	if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE))
1160 		ret = -EPERM;
1161 	rcu_read_unlock();
1162 
1163 	return ret;
1164 }
1165 
1166 /**
1167  * cap_task_setscheduler - Detemine if scheduler policy change is permitted
1168  * @p: The task to affect
1169  *
1170  * Detemine if the requested scheduler policy change is permitted for the
1171  * specified task, returning 0 if permission is granted, -ve if denied.
1172  */
1173 int cap_task_setscheduler(struct task_struct *p)
1174 {
1175 	return cap_safe_nice(p);
1176 }
1177 
1178 /**
1179  * cap_task_ioprio - Detemine if I/O priority change is permitted
1180  * @p: The task to affect
1181  * @ioprio: The I/O priority to set
1182  *
1183  * Detemine if the requested I/O priority change is permitted for the specified
1184  * task, returning 0 if permission is granted, -ve if denied.
1185  */
1186 int cap_task_setioprio(struct task_struct *p, int ioprio)
1187 {
1188 	return cap_safe_nice(p);
1189 }
1190 
1191 /**
1192  * cap_task_ioprio - Detemine if task priority change is permitted
1193  * @p: The task to affect
1194  * @nice: The nice value to set
1195  *
1196  * Detemine if the requested task priority change is permitted for the
1197  * specified task, returning 0 if permission is granted, -ve if denied.
1198  */
1199 int cap_task_setnice(struct task_struct *p, int nice)
1200 {
1201 	return cap_safe_nice(p);
1202 }
1203 
1204 /*
1205  * Implement PR_CAPBSET_DROP.  Attempt to remove the specified capability from
1206  * the current task's bounding set.  Returns 0 on success, -ve on error.
1207  */
1208 static int cap_prctl_drop(unsigned long cap)
1209 {
1210 	struct cred *new;
1211 
1212 	if (!ns_capable(current_user_ns(), CAP_SETPCAP))
1213 		return -EPERM;
1214 	if (!cap_valid(cap))
1215 		return -EINVAL;
1216 
1217 	new = prepare_creds();
1218 	if (!new)
1219 		return -ENOMEM;
1220 	cap_lower(new->cap_bset, cap);
1221 	return commit_creds(new);
1222 }
1223 
1224 /**
1225  * cap_task_prctl - Implement process control functions for this security module
1226  * @option: The process control function requested
1227  * @arg2, @arg3, @arg4, @arg5: The argument data for this function
1228  *
1229  * Allow process control functions (sys_prctl()) to alter capabilities; may
1230  * also deny access to other functions not otherwise implemented here.
1231  *
1232  * Returns 0 or +ve on success, -ENOSYS if this function is not implemented
1233  * here, other -ve on error.  If -ENOSYS is returned, sys_prctl() and other LSM
1234  * modules will consider performing the function.
1235  */
1236 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
1237 		   unsigned long arg4, unsigned long arg5)
1238 {
1239 	const struct cred *old = current_cred();
1240 	struct cred *new;
1241 
1242 	switch (option) {
1243 	case PR_CAPBSET_READ:
1244 		if (!cap_valid(arg2))
1245 			return -EINVAL;
1246 		return !!cap_raised(old->cap_bset, arg2);
1247 
1248 	case PR_CAPBSET_DROP:
1249 		return cap_prctl_drop(arg2);
1250 
1251 	/*
1252 	 * The next four prctl's remain to assist with transitioning a
1253 	 * system from legacy UID=0 based privilege (when filesystem
1254 	 * capabilities are not in use) to a system using filesystem
1255 	 * capabilities only - as the POSIX.1e draft intended.
1256 	 *
1257 	 * Note:
1258 	 *
1259 	 *  PR_SET_SECUREBITS =
1260 	 *      issecure_mask(SECURE_KEEP_CAPS_LOCKED)
1261 	 *    | issecure_mask(SECURE_NOROOT)
1262 	 *    | issecure_mask(SECURE_NOROOT_LOCKED)
1263 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP)
1264 	 *    | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
1265 	 *
1266 	 * will ensure that the current process and all of its
1267 	 * children will be locked into a pure
1268 	 * capability-based-privilege environment.
1269 	 */
1270 	case PR_SET_SECUREBITS:
1271 		if ((((old->securebits & SECURE_ALL_LOCKS) >> 1)
1272 		     & (old->securebits ^ arg2))			/*[1]*/
1273 		    || ((old->securebits & SECURE_ALL_LOCKS & ~arg2))	/*[2]*/
1274 		    || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))	/*[3]*/
1275 		    || (cap_capable(current_cred(),
1276 				    current_cred()->user_ns,
1277 				    CAP_SETPCAP,
1278 				    CAP_OPT_NONE) != 0)			/*[4]*/
1279 			/*
1280 			 * [1] no changing of bits that are locked
1281 			 * [2] no unlocking of locks
1282 			 * [3] no setting of unsupported bits
1283 			 * [4] doing anything requires privilege (go read about
1284 			 *     the "sendmail capabilities bug")
1285 			 */
1286 		    )
1287 			/* cannot change a locked bit */
1288 			return -EPERM;
1289 
1290 		new = prepare_creds();
1291 		if (!new)
1292 			return -ENOMEM;
1293 		new->securebits = arg2;
1294 		return commit_creds(new);
1295 
1296 	case PR_GET_SECUREBITS:
1297 		return old->securebits;
1298 
1299 	case PR_GET_KEEPCAPS:
1300 		return !!issecure(SECURE_KEEP_CAPS);
1301 
1302 	case PR_SET_KEEPCAPS:
1303 		if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
1304 			return -EINVAL;
1305 		if (issecure(SECURE_KEEP_CAPS_LOCKED))
1306 			return -EPERM;
1307 
1308 		new = prepare_creds();
1309 		if (!new)
1310 			return -ENOMEM;
1311 		if (arg2)
1312 			new->securebits |= issecure_mask(SECURE_KEEP_CAPS);
1313 		else
1314 			new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1315 		return commit_creds(new);
1316 
1317 	case PR_CAP_AMBIENT:
1318 		if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) {
1319 			if (arg3 | arg4 | arg5)
1320 				return -EINVAL;
1321 
1322 			new = prepare_creds();
1323 			if (!new)
1324 				return -ENOMEM;
1325 			cap_clear(new->cap_ambient);
1326 			return commit_creds(new);
1327 		}
1328 
1329 		if (((!cap_valid(arg3)) | arg4 | arg5))
1330 			return -EINVAL;
1331 
1332 		if (arg2 == PR_CAP_AMBIENT_IS_SET) {
1333 			return !!cap_raised(current_cred()->cap_ambient, arg3);
1334 		} else if (arg2 != PR_CAP_AMBIENT_RAISE &&
1335 			   arg2 != PR_CAP_AMBIENT_LOWER) {
1336 			return -EINVAL;
1337 		} else {
1338 			if (arg2 == PR_CAP_AMBIENT_RAISE &&
1339 			    (!cap_raised(current_cred()->cap_permitted, arg3) ||
1340 			     !cap_raised(current_cred()->cap_inheritable,
1341 					 arg3) ||
1342 			     issecure(SECURE_NO_CAP_AMBIENT_RAISE)))
1343 				return -EPERM;
1344 
1345 			new = prepare_creds();
1346 			if (!new)
1347 				return -ENOMEM;
1348 			if (arg2 == PR_CAP_AMBIENT_RAISE)
1349 				cap_raise(new->cap_ambient, arg3);
1350 			else
1351 				cap_lower(new->cap_ambient, arg3);
1352 			return commit_creds(new);
1353 		}
1354 
1355 	default:
1356 		/* No functionality available - continue with default */
1357 		return -ENOSYS;
1358 	}
1359 }
1360 
1361 /**
1362  * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted
1363  * @mm: The VM space in which the new mapping is to be made
1364  * @pages: The size of the mapping
1365  *
1366  * Determine whether the allocation of a new virtual mapping by the current
1367  * task is permitted, returning 1 if permission is granted, 0 if not.
1368  */
1369 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1370 {
1371 	int cap_sys_admin = 0;
1372 
1373 	if (cap_capable(current_cred(), &init_user_ns,
1374 				CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) == 0)
1375 		cap_sys_admin = 1;
1376 
1377 	return cap_sys_admin;
1378 }
1379 
1380 /*
1381  * cap_mmap_addr - check if able to map given addr
1382  * @addr: address attempting to be mapped
1383  *
1384  * If the process is attempting to map memory below dac_mmap_min_addr they need
1385  * CAP_SYS_RAWIO.  The other parameters to this function are unused by the
1386  * capability security module.  Returns 0 if this mapping should be allowed
1387  * -EPERM if not.
1388  */
1389 int cap_mmap_addr(unsigned long addr)
1390 {
1391 	int ret = 0;
1392 
1393 	if (addr < dac_mmap_min_addr) {
1394 		ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO,
1395 				  CAP_OPT_NONE);
1396 		/* set PF_SUPERPRIV if it turns out we allow the low mmap */
1397 		if (ret == 0)
1398 			current->flags |= PF_SUPERPRIV;
1399 	}
1400 	return ret;
1401 }
1402 
1403 int cap_mmap_file(struct file *file, unsigned long reqprot,
1404 		  unsigned long prot, unsigned long flags)
1405 {
1406 	return 0;
1407 }
1408 
1409 #ifdef CONFIG_SECURITY
1410 
1411 static struct security_hook_list capability_hooks[] __lsm_ro_after_init = {
1412 	LSM_HOOK_INIT(capable, cap_capable),
1413 	LSM_HOOK_INIT(settime, cap_settime),
1414 	LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check),
1415 	LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme),
1416 	LSM_HOOK_INIT(capget, cap_capget),
1417 	LSM_HOOK_INIT(capset, cap_capset),
1418 	LSM_HOOK_INIT(bprm_creds_from_file, cap_bprm_creds_from_file),
1419 	LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv),
1420 	LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv),
1421 	LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity),
1422 	LSM_HOOK_INIT(mmap_addr, cap_mmap_addr),
1423 	LSM_HOOK_INIT(mmap_file, cap_mmap_file),
1424 	LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid),
1425 	LSM_HOOK_INIT(task_prctl, cap_task_prctl),
1426 	LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler),
1427 	LSM_HOOK_INIT(task_setioprio, cap_task_setioprio),
1428 	LSM_HOOK_INIT(task_setnice, cap_task_setnice),
1429 	LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory),
1430 };
1431 
1432 static int __init capability_init(void)
1433 {
1434 	security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks),
1435 				"capability");
1436 	return 0;
1437 }
1438 
1439 DEFINE_LSM(capability) = {
1440 	.name = "capability",
1441 	.order = LSM_ORDER_FIRST,
1442 	.init = capability_init,
1443 };
1444 
1445 #endif /* CONFIG_SECURITY */
1446