xref: /linux/fs/posix_acl.c (revision 0526b56cbc3c489642bd6a5fe4b718dea7ef0ee8)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2002,2003 by Andreas Gruenbacher <a.gruenbacher@computer.org>
4  *
5  * Fixes from William Schumacher incorporated on 15 March 2001.
6  *    (Reported by Charles Bertsch, <CBertsch@microtest.com>).
7  */
8 
9 /*
10  *  This file contains generic functions for manipulating
11  *  POSIX 1003.1e draft standard 17 ACLs.
12  */
13 
14 #include <linux/kernel.h>
15 #include <linux/slab.h>
16 #include <linux/atomic.h>
17 #include <linux/fs.h>
18 #include <linux/sched.h>
19 #include <linux/cred.h>
20 #include <linux/posix_acl.h>
21 #include <linux/posix_acl_xattr.h>
22 #include <linux/xattr.h>
23 #include <linux/export.h>
24 #include <linux/user_namespace.h>
25 #include <linux/namei.h>
26 #include <linux/mnt_idmapping.h>
27 #include <linux/iversion.h>
28 #include <linux/security.h>
29 #include <linux/evm.h>
30 #include <linux/fsnotify.h>
31 #include <linux/filelock.h>
32 
33 #include "internal.h"
34 
35 static struct posix_acl **acl_by_type(struct inode *inode, int type)
36 {
37 	switch (type) {
38 	case ACL_TYPE_ACCESS:
39 		return &inode->i_acl;
40 	case ACL_TYPE_DEFAULT:
41 		return &inode->i_default_acl;
42 	default:
43 		BUG();
44 	}
45 }
46 
47 struct posix_acl *get_cached_acl(struct inode *inode, int type)
48 {
49 	struct posix_acl **p = acl_by_type(inode, type);
50 	struct posix_acl *acl;
51 
52 	for (;;) {
53 		rcu_read_lock();
54 		acl = rcu_dereference(*p);
55 		if (!acl || is_uncached_acl(acl) ||
56 		    refcount_inc_not_zero(&acl->a_refcount))
57 			break;
58 		rcu_read_unlock();
59 		cpu_relax();
60 	}
61 	rcu_read_unlock();
62 	return acl;
63 }
64 EXPORT_SYMBOL(get_cached_acl);
65 
66 struct posix_acl *get_cached_acl_rcu(struct inode *inode, int type)
67 {
68 	struct posix_acl *acl = rcu_dereference(*acl_by_type(inode, type));
69 
70 	if (acl == ACL_DONT_CACHE) {
71 		struct posix_acl *ret;
72 
73 		ret = inode->i_op->get_inode_acl(inode, type, LOOKUP_RCU);
74 		if (!IS_ERR(ret))
75 			acl = ret;
76 	}
77 
78 	return acl;
79 }
80 EXPORT_SYMBOL(get_cached_acl_rcu);
81 
82 void set_cached_acl(struct inode *inode, int type, struct posix_acl *acl)
83 {
84 	struct posix_acl **p = acl_by_type(inode, type);
85 	struct posix_acl *old;
86 
87 	old = xchg(p, posix_acl_dup(acl));
88 	if (!is_uncached_acl(old))
89 		posix_acl_release(old);
90 }
91 EXPORT_SYMBOL(set_cached_acl);
92 
93 static void __forget_cached_acl(struct posix_acl **p)
94 {
95 	struct posix_acl *old;
96 
97 	old = xchg(p, ACL_NOT_CACHED);
98 	if (!is_uncached_acl(old))
99 		posix_acl_release(old);
100 }
101 
102 void forget_cached_acl(struct inode *inode, int type)
103 {
104 	__forget_cached_acl(acl_by_type(inode, type));
105 }
106 EXPORT_SYMBOL(forget_cached_acl);
107 
108 void forget_all_cached_acls(struct inode *inode)
109 {
110 	__forget_cached_acl(&inode->i_acl);
111 	__forget_cached_acl(&inode->i_default_acl);
112 }
113 EXPORT_SYMBOL(forget_all_cached_acls);
114 
115 static struct posix_acl *__get_acl(struct mnt_idmap *idmap,
116 				   struct dentry *dentry, struct inode *inode,
117 				   int type)
118 {
119 	struct posix_acl *sentinel;
120 	struct posix_acl **p;
121 	struct posix_acl *acl;
122 
123 	/*
124 	 * The sentinel is used to detect when another operation like
125 	 * set_cached_acl() or forget_cached_acl() races with get_inode_acl().
126 	 * It is guaranteed that is_uncached_acl(sentinel) is true.
127 	 */
128 
129 	acl = get_cached_acl(inode, type);
130 	if (!is_uncached_acl(acl))
131 		return acl;
132 
133 	if (!IS_POSIXACL(inode))
134 		return NULL;
135 
136 	sentinel = uncached_acl_sentinel(current);
137 	p = acl_by_type(inode, type);
138 
139 	/*
140 	 * If the ACL isn't being read yet, set our sentinel.  Otherwise, the
141 	 * current value of the ACL will not be ACL_NOT_CACHED and so our own
142 	 * sentinel will not be set; another task will update the cache.  We
143 	 * could wait for that other task to complete its job, but it's easier
144 	 * to just call ->get_inode_acl to fetch the ACL ourself.  (This is
145 	 * going to be an unlikely race.)
146 	 */
147 	cmpxchg(p, ACL_NOT_CACHED, sentinel);
148 
149 	/*
150 	 * Normally, the ACL returned by ->get{_inode}_acl will be cached.
151 	 * A filesystem can prevent that by calling
152 	 * forget_cached_acl(inode, type) in ->get{_inode}_acl.
153 	 *
154 	 * If the filesystem doesn't have a get{_inode}_ acl() function at all,
155 	 * we'll just create the negative cache entry.
156 	 */
157 	if (dentry && inode->i_op->get_acl) {
158 		acl = inode->i_op->get_acl(idmap, dentry, type);
159 	} else if (inode->i_op->get_inode_acl) {
160 		acl = inode->i_op->get_inode_acl(inode, type, false);
161 	} else {
162 		set_cached_acl(inode, type, NULL);
163 		return NULL;
164 	}
165 	if (IS_ERR(acl)) {
166 		/*
167 		 * Remove our sentinel so that we don't block future attempts
168 		 * to cache the ACL.
169 		 */
170 		cmpxchg(p, sentinel, ACL_NOT_CACHED);
171 		return acl;
172 	}
173 
174 	/*
175 	 * Cache the result, but only if our sentinel is still in place.
176 	 */
177 	posix_acl_dup(acl);
178 	if (unlikely(!try_cmpxchg(p, &sentinel, acl)))
179 		posix_acl_release(acl);
180 	return acl;
181 }
182 
183 struct posix_acl *get_inode_acl(struct inode *inode, int type)
184 {
185 	return __get_acl(&nop_mnt_idmap, NULL, inode, type);
186 }
187 EXPORT_SYMBOL(get_inode_acl);
188 
189 /*
190  * Init a fresh posix_acl
191  */
192 void
193 posix_acl_init(struct posix_acl *acl, int count)
194 {
195 	refcount_set(&acl->a_refcount, 1);
196 	acl->a_count = count;
197 }
198 EXPORT_SYMBOL(posix_acl_init);
199 
200 /*
201  * Allocate a new ACL with the specified number of entries.
202  */
203 struct posix_acl *
204 posix_acl_alloc(int count, gfp_t flags)
205 {
206 	const size_t size = sizeof(struct posix_acl) +
207 	                    count * sizeof(struct posix_acl_entry);
208 	struct posix_acl *acl = kmalloc(size, flags);
209 	if (acl)
210 		posix_acl_init(acl, count);
211 	return acl;
212 }
213 EXPORT_SYMBOL(posix_acl_alloc);
214 
215 /*
216  * Clone an ACL.
217  */
218 struct posix_acl *
219 posix_acl_clone(const struct posix_acl *acl, gfp_t flags)
220 {
221 	struct posix_acl *clone = NULL;
222 
223 	if (acl) {
224 		int size = sizeof(struct posix_acl) + acl->a_count *
225 		           sizeof(struct posix_acl_entry);
226 		clone = kmemdup(acl, size, flags);
227 		if (clone)
228 			refcount_set(&clone->a_refcount, 1);
229 	}
230 	return clone;
231 }
232 EXPORT_SYMBOL_GPL(posix_acl_clone);
233 
234 /*
235  * Check if an acl is valid. Returns 0 if it is, or -E... otherwise.
236  */
237 int
238 posix_acl_valid(struct user_namespace *user_ns, const struct posix_acl *acl)
239 {
240 	const struct posix_acl_entry *pa, *pe;
241 	int state = ACL_USER_OBJ;
242 	int needs_mask = 0;
243 
244 	FOREACH_ACL_ENTRY(pa, acl, pe) {
245 		if (pa->e_perm & ~(ACL_READ|ACL_WRITE|ACL_EXECUTE))
246 			return -EINVAL;
247 		switch (pa->e_tag) {
248 			case ACL_USER_OBJ:
249 				if (state == ACL_USER_OBJ) {
250 					state = ACL_USER;
251 					break;
252 				}
253 				return -EINVAL;
254 
255 			case ACL_USER:
256 				if (state != ACL_USER)
257 					return -EINVAL;
258 				if (!kuid_has_mapping(user_ns, pa->e_uid))
259 					return -EINVAL;
260 				needs_mask = 1;
261 				break;
262 
263 			case ACL_GROUP_OBJ:
264 				if (state == ACL_USER) {
265 					state = ACL_GROUP;
266 					break;
267 				}
268 				return -EINVAL;
269 
270 			case ACL_GROUP:
271 				if (state != ACL_GROUP)
272 					return -EINVAL;
273 				if (!kgid_has_mapping(user_ns, pa->e_gid))
274 					return -EINVAL;
275 				needs_mask = 1;
276 				break;
277 
278 			case ACL_MASK:
279 				if (state != ACL_GROUP)
280 					return -EINVAL;
281 				state = ACL_OTHER;
282 				break;
283 
284 			case ACL_OTHER:
285 				if (state == ACL_OTHER ||
286 				    (state == ACL_GROUP && !needs_mask)) {
287 					state = 0;
288 					break;
289 				}
290 				return -EINVAL;
291 
292 			default:
293 				return -EINVAL;
294 		}
295 	}
296 	if (state == 0)
297 		return 0;
298 	return -EINVAL;
299 }
300 EXPORT_SYMBOL(posix_acl_valid);
301 
302 /*
303  * Returns 0 if the acl can be exactly represented in the traditional
304  * file mode permission bits, or else 1. Returns -E... on error.
305  */
306 int
307 posix_acl_equiv_mode(const struct posix_acl *acl, umode_t *mode_p)
308 {
309 	const struct posix_acl_entry *pa, *pe;
310 	umode_t mode = 0;
311 	int not_equiv = 0;
312 
313 	/*
314 	 * A null ACL can always be presented as mode bits.
315 	 */
316 	if (!acl)
317 		return 0;
318 
319 	FOREACH_ACL_ENTRY(pa, acl, pe) {
320 		switch (pa->e_tag) {
321 			case ACL_USER_OBJ:
322 				mode |= (pa->e_perm & S_IRWXO) << 6;
323 				break;
324 			case ACL_GROUP_OBJ:
325 				mode |= (pa->e_perm & S_IRWXO) << 3;
326 				break;
327 			case ACL_OTHER:
328 				mode |= pa->e_perm & S_IRWXO;
329 				break;
330 			case ACL_MASK:
331 				mode = (mode & ~S_IRWXG) |
332 				       ((pa->e_perm & S_IRWXO) << 3);
333 				not_equiv = 1;
334 				break;
335 			case ACL_USER:
336 			case ACL_GROUP:
337 				not_equiv = 1;
338 				break;
339 			default:
340 				return -EINVAL;
341 		}
342 	}
343         if (mode_p)
344                 *mode_p = (*mode_p & ~S_IRWXUGO) | mode;
345         return not_equiv;
346 }
347 EXPORT_SYMBOL(posix_acl_equiv_mode);
348 
349 /*
350  * Create an ACL representing the file mode permission bits of an inode.
351  */
352 struct posix_acl *
353 posix_acl_from_mode(umode_t mode, gfp_t flags)
354 {
355 	struct posix_acl *acl = posix_acl_alloc(3, flags);
356 	if (!acl)
357 		return ERR_PTR(-ENOMEM);
358 
359 	acl->a_entries[0].e_tag  = ACL_USER_OBJ;
360 	acl->a_entries[0].e_perm = (mode & S_IRWXU) >> 6;
361 
362 	acl->a_entries[1].e_tag  = ACL_GROUP_OBJ;
363 	acl->a_entries[1].e_perm = (mode & S_IRWXG) >> 3;
364 
365 	acl->a_entries[2].e_tag  = ACL_OTHER;
366 	acl->a_entries[2].e_perm = (mode & S_IRWXO);
367 	return acl;
368 }
369 EXPORT_SYMBOL(posix_acl_from_mode);
370 
371 /*
372  * Return 0 if current is granted want access to the inode
373  * by the acl. Returns -E... otherwise.
374  */
375 int
376 posix_acl_permission(struct mnt_idmap *idmap, struct inode *inode,
377 		     const struct posix_acl *acl, int want)
378 {
379 	const struct posix_acl_entry *pa, *pe, *mask_obj;
380 	struct user_namespace *fs_userns = i_user_ns(inode);
381 	int found = 0;
382 	vfsuid_t vfsuid;
383 	vfsgid_t vfsgid;
384 
385 	want &= MAY_READ | MAY_WRITE | MAY_EXEC;
386 
387 	FOREACH_ACL_ENTRY(pa, acl, pe) {
388                 switch(pa->e_tag) {
389                         case ACL_USER_OBJ:
390 				/* (May have been checked already) */
391 				vfsuid = i_uid_into_vfsuid(idmap, inode);
392 				if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
393                                         goto check_perm;
394                                 break;
395                         case ACL_USER:
396 				vfsuid = make_vfsuid(idmap, fs_userns,
397 						     pa->e_uid);
398 				if (vfsuid_eq_kuid(vfsuid, current_fsuid()))
399                                         goto mask;
400 				break;
401                         case ACL_GROUP_OBJ:
402 				vfsgid = i_gid_into_vfsgid(idmap, inode);
403 				if (vfsgid_in_group_p(vfsgid)) {
404 					found = 1;
405 					if ((pa->e_perm & want) == want)
406 						goto mask;
407                                 }
408 				break;
409                         case ACL_GROUP:
410 				vfsgid = make_vfsgid(idmap, fs_userns,
411 						     pa->e_gid);
412 				if (vfsgid_in_group_p(vfsgid)) {
413 					found = 1;
414 					if ((pa->e_perm & want) == want)
415 						goto mask;
416                                 }
417                                 break;
418                         case ACL_MASK:
419                                 break;
420                         case ACL_OTHER:
421 				if (found)
422 					return -EACCES;
423 				else
424 					goto check_perm;
425 			default:
426 				return -EIO;
427                 }
428         }
429 	return -EIO;
430 
431 mask:
432 	for (mask_obj = pa+1; mask_obj != pe; mask_obj++) {
433 		if (mask_obj->e_tag == ACL_MASK) {
434 			if ((pa->e_perm & mask_obj->e_perm & want) == want)
435 				return 0;
436 			return -EACCES;
437 		}
438 	}
439 
440 check_perm:
441 	if ((pa->e_perm & want) == want)
442 		return 0;
443 	return -EACCES;
444 }
445 
446 /*
447  * Modify acl when creating a new inode. The caller must ensure the acl is
448  * only referenced once.
449  *
450  * mode_p initially must contain the mode parameter to the open() / creat()
451  * system calls. All permissions that are not granted by the acl are removed.
452  * The permissions in the acl are changed to reflect the mode_p parameter.
453  */
454 static int posix_acl_create_masq(struct posix_acl *acl, umode_t *mode_p)
455 {
456 	struct posix_acl_entry *pa, *pe;
457 	struct posix_acl_entry *group_obj = NULL, *mask_obj = NULL;
458 	umode_t mode = *mode_p;
459 	int not_equiv = 0;
460 
461 	/* assert(atomic_read(acl->a_refcount) == 1); */
462 
463 	FOREACH_ACL_ENTRY(pa, acl, pe) {
464                 switch(pa->e_tag) {
465                         case ACL_USER_OBJ:
466 				pa->e_perm &= (mode >> 6) | ~S_IRWXO;
467 				mode &= (pa->e_perm << 6) | ~S_IRWXU;
468 				break;
469 
470 			case ACL_USER:
471 			case ACL_GROUP:
472 				not_equiv = 1;
473 				break;
474 
475                         case ACL_GROUP_OBJ:
476 				group_obj = pa;
477                                 break;
478 
479                         case ACL_OTHER:
480 				pa->e_perm &= mode | ~S_IRWXO;
481 				mode &= pa->e_perm | ~S_IRWXO;
482                                 break;
483 
484                         case ACL_MASK:
485 				mask_obj = pa;
486 				not_equiv = 1;
487                                 break;
488 
489 			default:
490 				return -EIO;
491                 }
492         }
493 
494 	if (mask_obj) {
495 		mask_obj->e_perm &= (mode >> 3) | ~S_IRWXO;
496 		mode &= (mask_obj->e_perm << 3) | ~S_IRWXG;
497 	} else {
498 		if (!group_obj)
499 			return -EIO;
500 		group_obj->e_perm &= (mode >> 3) | ~S_IRWXO;
501 		mode &= (group_obj->e_perm << 3) | ~S_IRWXG;
502 	}
503 
504 	*mode_p = (*mode_p & ~S_IRWXUGO) | mode;
505         return not_equiv;
506 }
507 
508 /*
509  * Modify the ACL for the chmod syscall.
510  */
511 static int __posix_acl_chmod_masq(struct posix_acl *acl, umode_t mode)
512 {
513 	struct posix_acl_entry *group_obj = NULL, *mask_obj = NULL;
514 	struct posix_acl_entry *pa, *pe;
515 
516 	/* assert(atomic_read(acl->a_refcount) == 1); */
517 
518 	FOREACH_ACL_ENTRY(pa, acl, pe) {
519 		switch(pa->e_tag) {
520 			case ACL_USER_OBJ:
521 				pa->e_perm = (mode & S_IRWXU) >> 6;
522 				break;
523 
524 			case ACL_USER:
525 			case ACL_GROUP:
526 				break;
527 
528 			case ACL_GROUP_OBJ:
529 				group_obj = pa;
530 				break;
531 
532 			case ACL_MASK:
533 				mask_obj = pa;
534 				break;
535 
536 			case ACL_OTHER:
537 				pa->e_perm = (mode & S_IRWXO);
538 				break;
539 
540 			default:
541 				return -EIO;
542 		}
543 	}
544 
545 	if (mask_obj) {
546 		mask_obj->e_perm = (mode & S_IRWXG) >> 3;
547 	} else {
548 		if (!group_obj)
549 			return -EIO;
550 		group_obj->e_perm = (mode & S_IRWXG) >> 3;
551 	}
552 
553 	return 0;
554 }
555 
556 int
557 __posix_acl_create(struct posix_acl **acl, gfp_t gfp, umode_t *mode_p)
558 {
559 	struct posix_acl *clone = posix_acl_clone(*acl, gfp);
560 	int err = -ENOMEM;
561 	if (clone) {
562 		err = posix_acl_create_masq(clone, mode_p);
563 		if (err < 0) {
564 			posix_acl_release(clone);
565 			clone = NULL;
566 		}
567 	}
568 	posix_acl_release(*acl);
569 	*acl = clone;
570 	return err;
571 }
572 EXPORT_SYMBOL(__posix_acl_create);
573 
574 int
575 __posix_acl_chmod(struct posix_acl **acl, gfp_t gfp, umode_t mode)
576 {
577 	struct posix_acl *clone = posix_acl_clone(*acl, gfp);
578 	int err = -ENOMEM;
579 	if (clone) {
580 		err = __posix_acl_chmod_masq(clone, mode);
581 		if (err) {
582 			posix_acl_release(clone);
583 			clone = NULL;
584 		}
585 	}
586 	posix_acl_release(*acl);
587 	*acl = clone;
588 	return err;
589 }
590 EXPORT_SYMBOL(__posix_acl_chmod);
591 
592 /**
593  * posix_acl_chmod - chmod a posix acl
594  *
595  * @idmap:	idmap of the mount @inode was found from
596  * @dentry:	dentry to check permissions on
597  * @mode:	the new mode of @inode
598  *
599  * If the dentry has been found through an idmapped mount the idmap of
600  * the vfsmount must be passed through @idmap. This function will then
601  * take care to map the inode according to @idmap before checking
602  * permissions. On non-idmapped mounts or if permission checking is to be
603  * performed on the raw inode simply passs @nop_mnt_idmap.
604  */
605 int
606  posix_acl_chmod(struct mnt_idmap *idmap, struct dentry *dentry,
607 		    umode_t mode)
608 {
609 	struct inode *inode = d_inode(dentry);
610 	struct posix_acl *acl;
611 	int ret = 0;
612 
613 	if (!IS_POSIXACL(inode))
614 		return 0;
615 	if (!inode->i_op->set_acl)
616 		return -EOPNOTSUPP;
617 
618 	acl = get_inode_acl(inode, ACL_TYPE_ACCESS);
619 	if (IS_ERR_OR_NULL(acl)) {
620 		if (acl == ERR_PTR(-EOPNOTSUPP))
621 			return 0;
622 		return PTR_ERR(acl);
623 	}
624 
625 	ret = __posix_acl_chmod(&acl, GFP_KERNEL, mode);
626 	if (ret)
627 		return ret;
628 	ret = inode->i_op->set_acl(idmap, dentry, acl, ACL_TYPE_ACCESS);
629 	posix_acl_release(acl);
630 	return ret;
631 }
632 EXPORT_SYMBOL(posix_acl_chmod);
633 
634 int
635 posix_acl_create(struct inode *dir, umode_t *mode,
636 		struct posix_acl **default_acl, struct posix_acl **acl)
637 {
638 	struct posix_acl *p;
639 	struct posix_acl *clone;
640 	int ret;
641 
642 	*acl = NULL;
643 	*default_acl = NULL;
644 
645 	if (S_ISLNK(*mode) || !IS_POSIXACL(dir))
646 		return 0;
647 
648 	p = get_inode_acl(dir, ACL_TYPE_DEFAULT);
649 	if (!p || p == ERR_PTR(-EOPNOTSUPP)) {
650 		*mode &= ~current_umask();
651 		return 0;
652 	}
653 	if (IS_ERR(p))
654 		return PTR_ERR(p);
655 
656 	ret = -ENOMEM;
657 	clone = posix_acl_clone(p, GFP_NOFS);
658 	if (!clone)
659 		goto err_release;
660 
661 	ret = posix_acl_create_masq(clone, mode);
662 	if (ret < 0)
663 		goto err_release_clone;
664 
665 	if (ret == 0)
666 		posix_acl_release(clone);
667 	else
668 		*acl = clone;
669 
670 	if (!S_ISDIR(*mode))
671 		posix_acl_release(p);
672 	else
673 		*default_acl = p;
674 
675 	return 0;
676 
677 err_release_clone:
678 	posix_acl_release(clone);
679 err_release:
680 	posix_acl_release(p);
681 	return ret;
682 }
683 EXPORT_SYMBOL_GPL(posix_acl_create);
684 
685 /**
686  * posix_acl_update_mode  -  update mode in set_acl
687  * @idmap:	idmap of the mount @inode was found from
688  * @inode:	target inode
689  * @mode_p:	mode (pointer) for update
690  * @acl:	acl pointer
691  *
692  * Update the file mode when setting an ACL: compute the new file permission
693  * bits based on the ACL.  In addition, if the ACL is equivalent to the new
694  * file mode, set *@acl to NULL to indicate that no ACL should be set.
695  *
696  * As with chmod, clear the setgid bit if the caller is not in the owning group
697  * or capable of CAP_FSETID (see inode_change_ok).
698  *
699  * If the inode has been found through an idmapped mount the idmap of
700  * the vfsmount must be passed through @idmap. This function will then
701  * take care to map the inode according to @idmap before checking
702  * permissions. On non-idmapped mounts or if permission checking is to be
703  * performed on the raw inode simply passs @nop_mnt_idmap.
704  *
705  * Called from set_acl inode operations.
706  */
707 int posix_acl_update_mode(struct mnt_idmap *idmap,
708 			  struct inode *inode, umode_t *mode_p,
709 			  struct posix_acl **acl)
710 {
711 	umode_t mode = inode->i_mode;
712 	int error;
713 
714 	error = posix_acl_equiv_mode(*acl, &mode);
715 	if (error < 0)
716 		return error;
717 	if (error == 0)
718 		*acl = NULL;
719 	if (!vfsgid_in_group_p(i_gid_into_vfsgid(idmap, inode)) &&
720 	    !capable_wrt_inode_uidgid(idmap, inode, CAP_FSETID))
721 		mode &= ~S_ISGID;
722 	*mode_p = mode;
723 	return 0;
724 }
725 EXPORT_SYMBOL(posix_acl_update_mode);
726 
727 /*
728  * Fix up the uids and gids in posix acl extended attributes in place.
729  */
730 static int posix_acl_fix_xattr_common(const void *value, size_t size)
731 {
732 	const struct posix_acl_xattr_header *header = value;
733 	int count;
734 
735 	if (!header)
736 		return -EINVAL;
737 	if (size < sizeof(struct posix_acl_xattr_header))
738 		return -EINVAL;
739 	if (header->a_version != cpu_to_le32(POSIX_ACL_XATTR_VERSION))
740 		return -EOPNOTSUPP;
741 
742 	count = posix_acl_xattr_count(size);
743 	if (count < 0)
744 		return -EINVAL;
745 	if (count == 0)
746 		return 0;
747 
748 	return count;
749 }
750 
751 /**
752  * posix_acl_from_xattr - convert POSIX ACLs from backing store to VFS format
753  * @userns: the filesystem's idmapping
754  * @value: the uapi representation of POSIX ACLs
755  * @size: the size of @void
756  *
757  * Filesystems that store POSIX ACLs in the unaltered uapi format should use
758  * posix_acl_from_xattr() when reading them from the backing store and
759  * converting them into the struct posix_acl VFS format. The helper is
760  * specifically intended to be called from the acl inode operation.
761  *
762  * The posix_acl_from_xattr() function will map the raw {g,u}id values stored
763  * in ACL_{GROUP,USER} entries into idmapping in @userns.
764  *
765  * Note that posix_acl_from_xattr() does not take idmapped mounts into account.
766  * If it did it calling it from the get acl inode operation would return POSIX
767  * ACLs mapped according to an idmapped mount which would mean that the value
768  * couldn't be cached for the filesystem. Idmapped mounts are taken into
769  * account on the fly during permission checking or right at the VFS -
770  * userspace boundary before reporting them to the user.
771  *
772  * Return: Allocated struct posix_acl on success, NULL for a valid header but
773  *         without actual POSIX ACL entries, or ERR_PTR() encoded error code.
774  */
775 struct posix_acl *posix_acl_from_xattr(struct user_namespace *userns,
776 				       const void *value, size_t size)
777 {
778 	const struct posix_acl_xattr_header *header = value;
779 	const struct posix_acl_xattr_entry *entry = (const void *)(header + 1), *end;
780 	int count;
781 	struct posix_acl *acl;
782 	struct posix_acl_entry *acl_e;
783 
784 	count = posix_acl_fix_xattr_common(value, size);
785 	if (count < 0)
786 		return ERR_PTR(count);
787 	if (count == 0)
788 		return NULL;
789 
790 	acl = posix_acl_alloc(count, GFP_NOFS);
791 	if (!acl)
792 		return ERR_PTR(-ENOMEM);
793 	acl_e = acl->a_entries;
794 
795 	for (end = entry + count; entry != end; acl_e++, entry++) {
796 		acl_e->e_tag  = le16_to_cpu(entry->e_tag);
797 		acl_e->e_perm = le16_to_cpu(entry->e_perm);
798 
799 		switch(acl_e->e_tag) {
800 			case ACL_USER_OBJ:
801 			case ACL_GROUP_OBJ:
802 			case ACL_MASK:
803 			case ACL_OTHER:
804 				break;
805 
806 			case ACL_USER:
807 				acl_e->e_uid = make_kuid(userns,
808 						le32_to_cpu(entry->e_id));
809 				if (!uid_valid(acl_e->e_uid))
810 					goto fail;
811 				break;
812 			case ACL_GROUP:
813 				acl_e->e_gid = make_kgid(userns,
814 						le32_to_cpu(entry->e_id));
815 				if (!gid_valid(acl_e->e_gid))
816 					goto fail;
817 				break;
818 
819 			default:
820 				goto fail;
821 		}
822 	}
823 	return acl;
824 
825 fail:
826 	posix_acl_release(acl);
827 	return ERR_PTR(-EINVAL);
828 }
829 EXPORT_SYMBOL (posix_acl_from_xattr);
830 
831 /*
832  * Convert from in-memory to extended attribute representation.
833  */
834 int
835 posix_acl_to_xattr(struct user_namespace *user_ns, const struct posix_acl *acl,
836 		   void *buffer, size_t size)
837 {
838 	struct posix_acl_xattr_header *ext_acl = buffer;
839 	struct posix_acl_xattr_entry *ext_entry;
840 	int real_size, n;
841 
842 	real_size = posix_acl_xattr_size(acl->a_count);
843 	if (!buffer)
844 		return real_size;
845 	if (real_size > size)
846 		return -ERANGE;
847 
848 	ext_entry = (void *)(ext_acl + 1);
849 	ext_acl->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION);
850 
851 	for (n=0; n < acl->a_count; n++, ext_entry++) {
852 		const struct posix_acl_entry *acl_e = &acl->a_entries[n];
853 		ext_entry->e_tag  = cpu_to_le16(acl_e->e_tag);
854 		ext_entry->e_perm = cpu_to_le16(acl_e->e_perm);
855 		switch(acl_e->e_tag) {
856 		case ACL_USER:
857 			ext_entry->e_id =
858 				cpu_to_le32(from_kuid(user_ns, acl_e->e_uid));
859 			break;
860 		case ACL_GROUP:
861 			ext_entry->e_id =
862 				cpu_to_le32(from_kgid(user_ns, acl_e->e_gid));
863 			break;
864 		default:
865 			ext_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
866 			break;
867 		}
868 	}
869 	return real_size;
870 }
871 EXPORT_SYMBOL (posix_acl_to_xattr);
872 
873 /**
874  * vfs_posix_acl_to_xattr - convert from kernel to userspace representation
875  * @idmap: idmap of the mount
876  * @inode: inode the posix acls are set on
877  * @acl: the posix acls as represented by the vfs
878  * @buffer: the buffer into which to convert @acl
879  * @size: size of @buffer
880  *
881  * This converts @acl from the VFS representation in the filesystem idmapping
882  * to the uapi form reportable to userspace. And mount and caller idmappings
883  * are handled appropriately.
884  *
885  * Return: On success, the size of the stored uapi posix acls, on error a
886  * negative errno.
887  */
888 static ssize_t vfs_posix_acl_to_xattr(struct mnt_idmap *idmap,
889 				      struct inode *inode,
890 				      const struct posix_acl *acl, void *buffer,
891 				      size_t size)
892 
893 {
894 	struct posix_acl_xattr_header *ext_acl = buffer;
895 	struct posix_acl_xattr_entry *ext_entry;
896 	struct user_namespace *fs_userns, *caller_userns;
897 	ssize_t real_size, n;
898 	vfsuid_t vfsuid;
899 	vfsgid_t vfsgid;
900 
901 	real_size = posix_acl_xattr_size(acl->a_count);
902 	if (!buffer)
903 		return real_size;
904 	if (real_size > size)
905 		return -ERANGE;
906 
907 	ext_entry = (void *)(ext_acl + 1);
908 	ext_acl->a_version = cpu_to_le32(POSIX_ACL_XATTR_VERSION);
909 
910 	fs_userns = i_user_ns(inode);
911 	caller_userns = current_user_ns();
912 	for (n=0; n < acl->a_count; n++, ext_entry++) {
913 		const struct posix_acl_entry *acl_e = &acl->a_entries[n];
914 		ext_entry->e_tag  = cpu_to_le16(acl_e->e_tag);
915 		ext_entry->e_perm = cpu_to_le16(acl_e->e_perm);
916 		switch(acl_e->e_tag) {
917 		case ACL_USER:
918 			vfsuid = make_vfsuid(idmap, fs_userns, acl_e->e_uid);
919 			ext_entry->e_id = cpu_to_le32(from_kuid(
920 				caller_userns, vfsuid_into_kuid(vfsuid)));
921 			break;
922 		case ACL_GROUP:
923 			vfsgid = make_vfsgid(idmap, fs_userns, acl_e->e_gid);
924 			ext_entry->e_id = cpu_to_le32(from_kgid(
925 				caller_userns, vfsgid_into_kgid(vfsgid)));
926 			break;
927 		default:
928 			ext_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
929 			break;
930 		}
931 	}
932 	return real_size;
933 }
934 
935 int
936 set_posix_acl(struct mnt_idmap *idmap, struct dentry *dentry,
937 	      int type, struct posix_acl *acl)
938 {
939 	struct inode *inode = d_inode(dentry);
940 
941 	if (!IS_POSIXACL(inode))
942 		return -EOPNOTSUPP;
943 	if (!inode->i_op->set_acl)
944 		return -EOPNOTSUPP;
945 
946 	if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode))
947 		return acl ? -EACCES : 0;
948 	if (!inode_owner_or_capable(idmap, inode))
949 		return -EPERM;
950 
951 	if (acl) {
952 		int ret = posix_acl_valid(inode->i_sb->s_user_ns, acl);
953 		if (ret)
954 			return ret;
955 	}
956 	return inode->i_op->set_acl(idmap, dentry, acl, type);
957 }
958 EXPORT_SYMBOL(set_posix_acl);
959 
960 int posix_acl_listxattr(struct inode *inode, char **buffer,
961 			ssize_t *remaining_size)
962 {
963 	int err;
964 
965 	if (!IS_POSIXACL(inode))
966 		return 0;
967 
968 	if (inode->i_acl) {
969 		err = xattr_list_one(buffer, remaining_size,
970 				     XATTR_NAME_POSIX_ACL_ACCESS);
971 		if (err)
972 			return err;
973 	}
974 
975 	if (inode->i_default_acl) {
976 		err = xattr_list_one(buffer, remaining_size,
977 				     XATTR_NAME_POSIX_ACL_DEFAULT);
978 		if (err)
979 			return err;
980 	}
981 
982 	return 0;
983 }
984 
985 static bool
986 posix_acl_xattr_list(struct dentry *dentry)
987 {
988 	return IS_POSIXACL(d_backing_inode(dentry));
989 }
990 
991 /*
992  * nop_posix_acl_access - legacy xattr handler for access POSIX ACLs
993  *
994  * This is the legacy POSIX ACL access xattr handler. It is used by some
995  * filesystems to implement their ->listxattr() inode operation. New code
996  * should never use them.
997  */
998 const struct xattr_handler nop_posix_acl_access = {
999 	.name = XATTR_NAME_POSIX_ACL_ACCESS,
1000 	.list = posix_acl_xattr_list,
1001 };
1002 EXPORT_SYMBOL_GPL(nop_posix_acl_access);
1003 
1004 /*
1005  * nop_posix_acl_default - legacy xattr handler for default POSIX ACLs
1006  *
1007  * This is the legacy POSIX ACL default xattr handler. It is used by some
1008  * filesystems to implement their ->listxattr() inode operation. New code
1009  * should never use them.
1010  */
1011 const struct xattr_handler nop_posix_acl_default = {
1012 	.name = XATTR_NAME_POSIX_ACL_DEFAULT,
1013 	.list = posix_acl_xattr_list,
1014 };
1015 EXPORT_SYMBOL_GPL(nop_posix_acl_default);
1016 
1017 int simple_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1018 		   struct posix_acl *acl, int type)
1019 {
1020 	int error;
1021 	struct inode *inode = d_inode(dentry);
1022 
1023 	if (type == ACL_TYPE_ACCESS) {
1024 		error = posix_acl_update_mode(idmap, inode,
1025 				&inode->i_mode, &acl);
1026 		if (error)
1027 			return error;
1028 	}
1029 
1030 	inode->i_ctime = current_time(inode);
1031 	if (IS_I_VERSION(inode))
1032 		inode_inc_iversion(inode);
1033 	set_cached_acl(inode, type, acl);
1034 	return 0;
1035 }
1036 
1037 int simple_acl_create(struct inode *dir, struct inode *inode)
1038 {
1039 	struct posix_acl *default_acl, *acl;
1040 	int error;
1041 
1042 	error = posix_acl_create(dir, &inode->i_mode, &default_acl, &acl);
1043 	if (error)
1044 		return error;
1045 
1046 	set_cached_acl(inode, ACL_TYPE_DEFAULT, default_acl);
1047 	set_cached_acl(inode, ACL_TYPE_ACCESS, acl);
1048 
1049 	if (default_acl)
1050 		posix_acl_release(default_acl);
1051 	if (acl)
1052 		posix_acl_release(acl);
1053 	return 0;
1054 }
1055 
1056 static int vfs_set_acl_idmapped_mnt(struct mnt_idmap *idmap,
1057 				    struct user_namespace *fs_userns,
1058 				    struct posix_acl *acl)
1059 {
1060 	for (int n = 0; n < acl->a_count; n++) {
1061 		struct posix_acl_entry *acl_e = &acl->a_entries[n];
1062 
1063 		switch (acl_e->e_tag) {
1064 		case ACL_USER:
1065 			acl_e->e_uid = from_vfsuid(idmap, fs_userns,
1066 						   VFSUIDT_INIT(acl_e->e_uid));
1067 			break;
1068 		case ACL_GROUP:
1069 			acl_e->e_gid = from_vfsgid(idmap, fs_userns,
1070 						   VFSGIDT_INIT(acl_e->e_gid));
1071 			break;
1072 		}
1073 	}
1074 
1075 	return 0;
1076 }
1077 
1078 /**
1079  * vfs_set_acl - set posix acls
1080  * @idmap: idmap of the mount
1081  * @dentry: the dentry based on which to set the posix acls
1082  * @acl_name: the name of the posix acl
1083  * @kacl: the posix acls in the appropriate VFS format
1084  *
1085  * This function sets @kacl. The caller must all posix_acl_release() on @kacl
1086  * afterwards.
1087  *
1088  * Return: On success 0, on error negative errno.
1089  */
1090 int vfs_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1091 		const char *acl_name, struct posix_acl *kacl)
1092 {
1093 	int acl_type;
1094 	int error;
1095 	struct inode *inode = d_inode(dentry);
1096 	struct inode *delegated_inode = NULL;
1097 
1098 	acl_type = posix_acl_type(acl_name);
1099 	if (acl_type < 0)
1100 		return -EINVAL;
1101 
1102 	if (kacl) {
1103 		/*
1104 		 * If we're on an idmapped mount translate from mount specific
1105 		 * vfs{g,u}id_t into global filesystem k{g,u}id_t.
1106 		 * Afterwards we can cache the POSIX ACLs filesystem wide and -
1107 		 * if this is a filesystem with a backing store - ultimately
1108 		 * translate them to backing store values.
1109 		 */
1110 		error = vfs_set_acl_idmapped_mnt(idmap, i_user_ns(inode), kacl);
1111 		if (error)
1112 			return error;
1113 	}
1114 
1115 retry_deleg:
1116 	inode_lock(inode);
1117 
1118 	/*
1119 	 * We only care about restrictions the inode struct itself places upon
1120 	 * us otherwise POSIX ACLs aren't subject to any VFS restrictions.
1121 	 */
1122 	error = may_write_xattr(idmap, inode);
1123 	if (error)
1124 		goto out_inode_unlock;
1125 
1126 	error = security_inode_set_acl(idmap, dentry, acl_name, kacl);
1127 	if (error)
1128 		goto out_inode_unlock;
1129 
1130 	error = try_break_deleg(inode, &delegated_inode);
1131 	if (error)
1132 		goto out_inode_unlock;
1133 
1134 	if (likely(!is_bad_inode(inode)))
1135 		error = set_posix_acl(idmap, dentry, acl_type, kacl);
1136 	else
1137 		error = -EIO;
1138 	if (!error) {
1139 		fsnotify_xattr(dentry);
1140 		evm_inode_post_set_acl(dentry, acl_name, kacl);
1141 	}
1142 
1143 out_inode_unlock:
1144 	inode_unlock(inode);
1145 
1146 	if (delegated_inode) {
1147 		error = break_deleg_wait(&delegated_inode);
1148 		if (!error)
1149 			goto retry_deleg;
1150 	}
1151 
1152 	return error;
1153 }
1154 EXPORT_SYMBOL_GPL(vfs_set_acl);
1155 
1156 /**
1157  * vfs_get_acl - get posix acls
1158  * @idmap: idmap of the mount
1159  * @dentry: the dentry based on which to retrieve the posix acls
1160  * @acl_name: the name of the posix acl
1161  *
1162  * This function retrieves @kacl from the filesystem. The caller must all
1163  * posix_acl_release() on @kacl.
1164  *
1165  * Return: On success POSIX ACLs in VFS format, on error negative errno.
1166  */
1167 struct posix_acl *vfs_get_acl(struct mnt_idmap *idmap,
1168 			      struct dentry *dentry, const char *acl_name)
1169 {
1170 	struct inode *inode = d_inode(dentry);
1171 	struct posix_acl *acl;
1172 	int acl_type, error;
1173 
1174 	acl_type = posix_acl_type(acl_name);
1175 	if (acl_type < 0)
1176 		return ERR_PTR(-EINVAL);
1177 
1178 	/*
1179 	 * The VFS has no restrictions on reading POSIX ACLs so calling
1180 	 * something like xattr_permission() isn't needed. Only LSMs get a say.
1181 	 */
1182 	error = security_inode_get_acl(idmap, dentry, acl_name);
1183 	if (error)
1184 		return ERR_PTR(error);
1185 
1186 	if (!IS_POSIXACL(inode))
1187 		return ERR_PTR(-EOPNOTSUPP);
1188 	if (S_ISLNK(inode->i_mode))
1189 		return ERR_PTR(-EOPNOTSUPP);
1190 
1191 	acl = __get_acl(idmap, dentry, inode, acl_type);
1192 	if (IS_ERR(acl))
1193 		return acl;
1194 	if (!acl)
1195 		return ERR_PTR(-ENODATA);
1196 
1197 	return acl;
1198 }
1199 EXPORT_SYMBOL_GPL(vfs_get_acl);
1200 
1201 /**
1202  * vfs_remove_acl - remove posix acls
1203  * @idmap: idmap of the mount
1204  * @dentry: the dentry based on which to retrieve the posix acls
1205  * @acl_name: the name of the posix acl
1206  *
1207  * This function removes posix acls.
1208  *
1209  * Return: On success 0, on error negative errno.
1210  */
1211 int vfs_remove_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1212 		   const char *acl_name)
1213 {
1214 	int acl_type;
1215 	int error;
1216 	struct inode *inode = d_inode(dentry);
1217 	struct inode *delegated_inode = NULL;
1218 
1219 	acl_type = posix_acl_type(acl_name);
1220 	if (acl_type < 0)
1221 		return -EINVAL;
1222 
1223 retry_deleg:
1224 	inode_lock(inode);
1225 
1226 	/*
1227 	 * We only care about restrictions the inode struct itself places upon
1228 	 * us otherwise POSIX ACLs aren't subject to any VFS restrictions.
1229 	 */
1230 	error = may_write_xattr(idmap, inode);
1231 	if (error)
1232 		goto out_inode_unlock;
1233 
1234 	error = security_inode_remove_acl(idmap, dentry, acl_name);
1235 	if (error)
1236 		goto out_inode_unlock;
1237 
1238 	error = try_break_deleg(inode, &delegated_inode);
1239 	if (error)
1240 		goto out_inode_unlock;
1241 
1242 	if (likely(!is_bad_inode(inode)))
1243 		error = set_posix_acl(idmap, dentry, acl_type, NULL);
1244 	else
1245 		error = -EIO;
1246 	if (!error) {
1247 		fsnotify_xattr(dentry);
1248 		evm_inode_post_remove_acl(idmap, dentry, acl_name);
1249 	}
1250 
1251 out_inode_unlock:
1252 	inode_unlock(inode);
1253 
1254 	if (delegated_inode) {
1255 		error = break_deleg_wait(&delegated_inode);
1256 		if (!error)
1257 			goto retry_deleg;
1258 	}
1259 
1260 	return error;
1261 }
1262 EXPORT_SYMBOL_GPL(vfs_remove_acl);
1263 
1264 int do_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1265 	       const char *acl_name, const void *kvalue, size_t size)
1266 {
1267 	int error;
1268 	struct posix_acl *acl = NULL;
1269 
1270 	if (size) {
1271 		/*
1272 		 * Note that posix_acl_from_xattr() uses GFP_NOFS when it
1273 		 * probably doesn't need to here.
1274 		 */
1275 		acl = posix_acl_from_xattr(current_user_ns(), kvalue, size);
1276 		if (IS_ERR(acl))
1277 			return PTR_ERR(acl);
1278 	}
1279 
1280 	error = vfs_set_acl(idmap, dentry, acl_name, acl);
1281 	posix_acl_release(acl);
1282 	return error;
1283 }
1284 
1285 ssize_t do_get_acl(struct mnt_idmap *idmap, struct dentry *dentry,
1286 		   const char *acl_name, void *kvalue, size_t size)
1287 {
1288 	ssize_t error;
1289 	struct posix_acl *acl;
1290 
1291 	acl = vfs_get_acl(idmap, dentry, acl_name);
1292 	if (IS_ERR(acl))
1293 		return PTR_ERR(acl);
1294 
1295 	error = vfs_posix_acl_to_xattr(idmap, d_inode(dentry),
1296 				       acl, kvalue, size);
1297 	posix_acl_release(acl);
1298 	return error;
1299 }
1300