xref: /freebsd/sys/contrib/openzfs/module/os/linux/zfs/zfs_acl.c (revision 8ccc0d235c226d84112561d453c49904398d085c)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2013 by Delphix. All rights reserved.
25  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
26  */
27 
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/sysmacros.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/sid.h>
36 #include <sys/file.h>
37 #include <sys/stat.h>
38 #include <sys/kmem.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/policy.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_fuid.h>
45 #include <sys/zfs_acl.h>
46 #include <sys/zfs_dir.h>
47 #include <sys/zfs_quota.h>
48 #include <sys/zfs_vfsops.h>
49 #include <sys/dmu.h>
50 #include <sys/dnode.h>
51 #include <sys/zap.h>
52 #include <sys/sa.h>
53 #include <sys/trace_acl.h>
54 #include <sys/zpl.h>
55 
56 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
57 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
58 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
59 #define	MIN_ACE_TYPE	ALLOW
60 
61 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
62 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
63     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
64 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
65     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
66 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
67     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
68 
69 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
70     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
71     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
72     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
73 
74 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
75 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
76     ACE_DELETE|ACE_DELETE_CHILD)
77 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
78 
79 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
80     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
81 
82 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
83     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
84 
85 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
86     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
87 
88 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
89 
90 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
91     ZFS_ACL_PROTECTED)
92 
93 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
94     ZFS_ACL_OBJ_ACE)
95 
96 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
97 
98 #define	IDMAP_WK_CREATOR_OWNER_UID	2147483648U
99 
100 static uint16_t
101 zfs_ace_v0_get_type(void *acep)
102 {
103 	return (((zfs_oldace_t *)acep)->z_type);
104 }
105 
106 static uint16_t
107 zfs_ace_v0_get_flags(void *acep)
108 {
109 	return (((zfs_oldace_t *)acep)->z_flags);
110 }
111 
112 static uint32_t
113 zfs_ace_v0_get_mask(void *acep)
114 {
115 	return (((zfs_oldace_t *)acep)->z_access_mask);
116 }
117 
118 static uint64_t
119 zfs_ace_v0_get_who(void *acep)
120 {
121 	return (((zfs_oldace_t *)acep)->z_fuid);
122 }
123 
124 static void
125 zfs_ace_v0_set_type(void *acep, uint16_t type)
126 {
127 	((zfs_oldace_t *)acep)->z_type = type;
128 }
129 
130 static void
131 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
132 {
133 	((zfs_oldace_t *)acep)->z_flags = flags;
134 }
135 
136 static void
137 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
138 {
139 	((zfs_oldace_t *)acep)->z_access_mask = mask;
140 }
141 
142 static void
143 zfs_ace_v0_set_who(void *acep, uint64_t who)
144 {
145 	((zfs_oldace_t *)acep)->z_fuid = who;
146 }
147 
148 static size_t
149 zfs_ace_v0_size(void *acep)
150 {
151 	(void) acep;
152 	return (sizeof (zfs_oldace_t));
153 }
154 
155 static size_t
156 zfs_ace_v0_abstract_size(void)
157 {
158 	return (sizeof (zfs_oldace_t));
159 }
160 
161 static int
162 zfs_ace_v0_mask_off(void)
163 {
164 	return (offsetof(zfs_oldace_t, z_access_mask));
165 }
166 
167 static int
168 zfs_ace_v0_data(void *acep, void **datap)
169 {
170 	(void) acep;
171 	*datap = NULL;
172 	return (0);
173 }
174 
175 static const acl_ops_t zfs_acl_v0_ops = {
176 	.ace_mask_get = zfs_ace_v0_get_mask,
177 	.ace_mask_set = zfs_ace_v0_set_mask,
178 	.ace_flags_get = zfs_ace_v0_get_flags,
179 	.ace_flags_set = zfs_ace_v0_set_flags,
180 	.ace_type_get = zfs_ace_v0_get_type,
181 	.ace_type_set = zfs_ace_v0_set_type,
182 	.ace_who_get = zfs_ace_v0_get_who,
183 	.ace_who_set = zfs_ace_v0_set_who,
184 	.ace_size = zfs_ace_v0_size,
185 	.ace_abstract_size = zfs_ace_v0_abstract_size,
186 	.ace_mask_off = zfs_ace_v0_mask_off,
187 	.ace_data = zfs_ace_v0_data
188 };
189 
190 static uint16_t
191 zfs_ace_fuid_get_type(void *acep)
192 {
193 	return (((zfs_ace_hdr_t *)acep)->z_type);
194 }
195 
196 static uint16_t
197 zfs_ace_fuid_get_flags(void *acep)
198 {
199 	return (((zfs_ace_hdr_t *)acep)->z_flags);
200 }
201 
202 static uint32_t
203 zfs_ace_fuid_get_mask(void *acep)
204 {
205 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
206 }
207 
208 static uint64_t
209 zfs_ace_fuid_get_who(void *args)
210 {
211 	uint16_t entry_type;
212 	zfs_ace_t *acep = args;
213 
214 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
215 
216 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
217 	    entry_type == ACE_EVERYONE)
218 		return (-1);
219 	return (((zfs_ace_t *)acep)->z_fuid);
220 }
221 
222 static void
223 zfs_ace_fuid_set_type(void *acep, uint16_t type)
224 {
225 	((zfs_ace_hdr_t *)acep)->z_type = type;
226 }
227 
228 static void
229 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
230 {
231 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
232 }
233 
234 static void
235 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
236 {
237 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
238 }
239 
240 static void
241 zfs_ace_fuid_set_who(void *arg, uint64_t who)
242 {
243 	zfs_ace_t *acep = arg;
244 
245 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
246 
247 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
248 	    entry_type == ACE_EVERYONE)
249 		return;
250 	acep->z_fuid = who;
251 }
252 
253 static size_t
254 zfs_ace_fuid_size(void *acep)
255 {
256 	zfs_ace_hdr_t *zacep = acep;
257 	uint16_t entry_type;
258 
259 	switch (zacep->z_type) {
260 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
261 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
262 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
263 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
264 		return (sizeof (zfs_object_ace_t));
265 	case ALLOW:
266 	case DENY:
267 		entry_type =
268 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
269 		if (entry_type == ACE_OWNER ||
270 		    entry_type == OWNING_GROUP ||
271 		    entry_type == ACE_EVERYONE)
272 			return (sizeof (zfs_ace_hdr_t));
273 		zfs_fallthrough;
274 	default:
275 		return (sizeof (zfs_ace_t));
276 	}
277 }
278 
279 static size_t
280 zfs_ace_fuid_abstract_size(void)
281 {
282 	return (sizeof (zfs_ace_hdr_t));
283 }
284 
285 static int
286 zfs_ace_fuid_mask_off(void)
287 {
288 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
289 }
290 
291 static int
292 zfs_ace_fuid_data(void *acep, void **datap)
293 {
294 	zfs_ace_t *zacep = acep;
295 	zfs_object_ace_t *zobjp;
296 
297 	switch (zacep->z_hdr.z_type) {
298 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
299 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
300 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
301 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
302 		zobjp = acep;
303 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
304 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
305 	default:
306 		*datap = NULL;
307 		return (0);
308 	}
309 }
310 
311 static const acl_ops_t zfs_acl_fuid_ops = {
312 	.ace_mask_get = zfs_ace_fuid_get_mask,
313 	.ace_mask_set = zfs_ace_fuid_set_mask,
314 	.ace_flags_get = zfs_ace_fuid_get_flags,
315 	.ace_flags_set = zfs_ace_fuid_set_flags,
316 	.ace_type_get = zfs_ace_fuid_get_type,
317 	.ace_type_set = zfs_ace_fuid_set_type,
318 	.ace_who_get = zfs_ace_fuid_get_who,
319 	.ace_who_set = zfs_ace_fuid_set_who,
320 	.ace_size = zfs_ace_fuid_size,
321 	.ace_abstract_size = zfs_ace_fuid_abstract_size,
322 	.ace_mask_off = zfs_ace_fuid_mask_off,
323 	.ace_data = zfs_ace_fuid_data
324 };
325 
326 /*
327  * The following three functions are provided for compatibility with
328  * older ZPL version in order to determine if the file use to have
329  * an external ACL and what version of ACL previously existed on the
330  * file.  Would really be nice to not need this, sigh.
331  */
332 uint64_t
333 zfs_external_acl(znode_t *zp)
334 {
335 	zfs_acl_phys_t acl_phys;
336 	int error;
337 
338 	if (zp->z_is_sa)
339 		return (0);
340 
341 	/*
342 	 * Need to deal with a potential
343 	 * race where zfs_sa_upgrade could cause
344 	 * z_isa_sa to change.
345 	 *
346 	 * If the lookup fails then the state of z_is_sa should have
347 	 * changed.
348 	 */
349 
350 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
351 	    &acl_phys, sizeof (acl_phys))) == 0)
352 		return (acl_phys.z_acl_extern_obj);
353 	else {
354 		/*
355 		 * after upgrade the SA_ZPL_ZNODE_ACL should have been
356 		 * removed
357 		 */
358 		VERIFY(zp->z_is_sa && error == ENOENT);
359 		return (0);
360 	}
361 }
362 
363 /*
364  * Determine size of ACL in bytes
365  *
366  * This is more complicated than it should be since we have to deal
367  * with old external ACLs.
368  */
369 static int
370 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
371     zfs_acl_phys_t *aclphys)
372 {
373 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
374 	uint64_t acl_count;
375 	int size;
376 	int error;
377 
378 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
379 	if (zp->z_is_sa) {
380 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
381 		    &size)) != 0)
382 			return (error);
383 		*aclsize = size;
384 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
385 		    &acl_count, sizeof (acl_count))) != 0)
386 			return (error);
387 		*aclcount = acl_count;
388 	} else {
389 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
390 		    aclphys, sizeof (*aclphys))) != 0)
391 			return (error);
392 
393 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
394 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
395 			*aclcount = aclphys->z_acl_size;
396 		} else {
397 			*aclsize = aclphys->z_acl_size;
398 			*aclcount = aclphys->z_acl_count;
399 		}
400 	}
401 	return (0);
402 }
403 
404 int
405 zfs_znode_acl_version(znode_t *zp)
406 {
407 	zfs_acl_phys_t acl_phys;
408 
409 	if (zp->z_is_sa)
410 		return (ZFS_ACL_VERSION_FUID);
411 	else {
412 		int error;
413 
414 		/*
415 		 * Need to deal with a potential
416 		 * race where zfs_sa_upgrade could cause
417 		 * z_isa_sa to change.
418 		 *
419 		 * If the lookup fails then the state of z_is_sa should have
420 		 * changed.
421 		 */
422 		if ((error = sa_lookup(zp->z_sa_hdl,
423 		    SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
424 		    &acl_phys, sizeof (acl_phys))) == 0)
425 			return (acl_phys.z_acl_version);
426 		else {
427 			/*
428 			 * After upgrade SA_ZPL_ZNODE_ACL should have
429 			 * been removed.
430 			 */
431 			VERIFY(zp->z_is_sa && error == ENOENT);
432 			return (ZFS_ACL_VERSION_FUID);
433 		}
434 	}
435 }
436 
437 static int
438 zfs_acl_version(int version)
439 {
440 	if (version < ZPL_VERSION_FUID)
441 		return (ZFS_ACL_VERSION_INITIAL);
442 	else
443 		return (ZFS_ACL_VERSION_FUID);
444 }
445 
446 static int
447 zfs_acl_version_zp(znode_t *zp)
448 {
449 	return (zfs_acl_version(ZTOZSB(zp)->z_version));
450 }
451 
452 zfs_acl_t *
453 zfs_acl_alloc(int vers)
454 {
455 	zfs_acl_t *aclp;
456 
457 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
458 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
459 	    offsetof(zfs_acl_node_t, z_next));
460 	aclp->z_version = vers;
461 	if (vers == ZFS_ACL_VERSION_FUID)
462 		aclp->z_ops = &zfs_acl_fuid_ops;
463 	else
464 		aclp->z_ops = &zfs_acl_v0_ops;
465 	return (aclp);
466 }
467 
468 zfs_acl_node_t *
469 zfs_acl_node_alloc(size_t bytes)
470 {
471 	zfs_acl_node_t *aclnode;
472 
473 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
474 	if (bytes) {
475 		aclnode->z_acldata = kmem_zalloc(bytes, KM_SLEEP);
476 		aclnode->z_allocdata = aclnode->z_acldata;
477 		aclnode->z_allocsize = bytes;
478 		aclnode->z_size = bytes;
479 	}
480 
481 	return (aclnode);
482 }
483 
484 static void
485 zfs_acl_node_free(zfs_acl_node_t *aclnode)
486 {
487 	if (aclnode->z_allocsize)
488 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
489 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
490 }
491 
492 static void
493 zfs_acl_release_nodes(zfs_acl_t *aclp)
494 {
495 	zfs_acl_node_t *aclnode;
496 
497 	while ((aclnode = list_remove_head(&aclp->z_acl)))
498 		zfs_acl_node_free(aclnode);
499 	aclp->z_acl_count = 0;
500 	aclp->z_acl_bytes = 0;
501 }
502 
503 void
504 zfs_acl_free(zfs_acl_t *aclp)
505 {
506 	zfs_acl_release_nodes(aclp);
507 	list_destroy(&aclp->z_acl);
508 	kmem_free(aclp, sizeof (zfs_acl_t));
509 }
510 
511 static boolean_t
512 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
513 {
514 	uint16_t entry_type;
515 
516 	switch (type) {
517 	case ALLOW:
518 	case DENY:
519 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
520 	case ACE_SYSTEM_ALARM_ACE_TYPE:
521 		entry_type = flags & ACE_TYPE_FLAGS;
522 		return (entry_type == ACE_OWNER ||
523 		    entry_type == OWNING_GROUP ||
524 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
525 		    entry_type == ACE_IDENTIFIER_GROUP);
526 	default:
527 		if (type <= MAX_ACE_TYPE)
528 			return (B_TRUE);
529 	}
530 	return (B_FALSE);
531 }
532 
533 static boolean_t
534 zfs_ace_valid(umode_t obj_mode, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
535 {
536 	/*
537 	 * first check type of entry
538 	 */
539 
540 	if (!zfs_acl_valid_ace_type(type, iflags))
541 		return (B_FALSE);
542 
543 	switch (type) {
544 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
545 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
546 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
547 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
548 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
549 			return (B_FALSE);
550 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
551 	}
552 
553 	/*
554 	 * next check inheritance level flags
555 	 */
556 
557 	if (S_ISDIR(obj_mode) &&
558 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
559 		aclp->z_hints |= ZFS_INHERIT_ACE;
560 
561 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
562 		if ((iflags & (ACE_FILE_INHERIT_ACE|
563 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
564 			return (B_FALSE);
565 		}
566 	}
567 
568 	return (B_TRUE);
569 }
570 
571 static void *
572 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
573     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
574 {
575 	zfs_acl_node_t *aclnode;
576 
577 	ASSERT(aclp);
578 
579 	if (start == NULL) {
580 		aclnode = list_head(&aclp->z_acl);
581 		if (aclnode == NULL)
582 			return (NULL);
583 
584 		aclp->z_next_ace = aclnode->z_acldata;
585 		aclp->z_curr_node = aclnode;
586 		aclnode->z_ace_idx = 0;
587 	}
588 
589 	aclnode = aclp->z_curr_node;
590 
591 	if (aclnode == NULL)
592 		return (NULL);
593 
594 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
595 		aclnode = list_next(&aclp->z_acl, aclnode);
596 		if (aclnode == NULL)
597 			return (NULL);
598 		else {
599 			aclp->z_curr_node = aclnode;
600 			aclnode->z_ace_idx = 0;
601 			aclp->z_next_ace = aclnode->z_acldata;
602 		}
603 	}
604 
605 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
606 		void *acep = aclp->z_next_ace;
607 		size_t ace_size;
608 
609 		/*
610 		 * Make sure we don't overstep our bounds
611 		 */
612 		ace_size = aclp->z_ops->ace_size(acep);
613 
614 		if (((caddr_t)acep + ace_size) >
615 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
616 			return (NULL);
617 		}
618 
619 		*iflags = aclp->z_ops->ace_flags_get(acep);
620 		*type = aclp->z_ops->ace_type_get(acep);
621 		*access_mask = aclp->z_ops->ace_mask_get(acep);
622 		*who = aclp->z_ops->ace_who_get(acep);
623 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
624 		aclnode->z_ace_idx++;
625 
626 		return ((void *)acep);
627 	}
628 	return (NULL);
629 }
630 
631 static uintptr_t
632 zfs_ace_walk(void *datap, uintptr_t cookie, int aclcnt,
633     uint16_t *flags, uint16_t *type, uint32_t *mask)
634 {
635 	(void) aclcnt;
636 	zfs_acl_t *aclp = datap;
637 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)cookie;
638 	uint64_t who;
639 
640 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
641 	    flags, type);
642 	return ((uintptr_t)acep);
643 }
644 
645 /*
646  * Copy ACE to internal ZFS format.
647  * While processing the ACL each ACE will be validated for correctness.
648  * ACE FUIDs will be created later.
649  */
650 static int
651 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, umode_t obj_mode, zfs_acl_t *aclp,
652     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
653     zfs_fuid_info_t **fuidp, cred_t *cr)
654 {
655 	int i;
656 	uint16_t entry_type;
657 	zfs_ace_t *aceptr = z_acl;
658 	ace_t *acep = datap;
659 	zfs_object_ace_t *zobjacep;
660 	ace_object_t *aceobjp;
661 
662 	for (i = 0; i != aclcnt; i++) {
663 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
664 		aceptr->z_hdr.z_flags = acep->a_flags;
665 		aceptr->z_hdr.z_type = acep->a_type;
666 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
667 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
668 		    entry_type != ACE_EVERYONE) {
669 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
670 			    cr, (entry_type == 0) ?
671 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
672 		}
673 
674 		/*
675 		 * Make sure ACE is valid
676 		 */
677 		if (zfs_ace_valid(obj_mode, aclp, aceptr->z_hdr.z_type,
678 		    aceptr->z_hdr.z_flags) != B_TRUE)
679 			return (SET_ERROR(EINVAL));
680 
681 		switch (acep->a_type) {
682 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
683 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
684 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
685 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
686 			zobjacep = (zfs_object_ace_t *)aceptr;
687 			aceobjp = (ace_object_t *)acep;
688 
689 			memcpy(zobjacep->z_object_type, aceobjp->a_obj_type,
690 			    sizeof (aceobjp->a_obj_type));
691 			memcpy(zobjacep->z_inherit_type,
692 			    aceobjp->a_inherit_obj_type,
693 			    sizeof (aceobjp->a_inherit_obj_type));
694 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
695 			break;
696 		default:
697 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
698 		}
699 
700 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
701 		    aclp->z_ops->ace_size(aceptr));
702 	}
703 
704 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
705 
706 	return (0);
707 }
708 
709 /*
710  * Copy ZFS ACEs to fixed size ace_t layout
711  */
712 static void
713 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
714     void *datap, int filter)
715 {
716 	uint64_t who;
717 	uint32_t access_mask;
718 	uint16_t iflags, type;
719 	zfs_ace_hdr_t *zacep = NULL;
720 	ace_t *acep = datap;
721 	ace_object_t *objacep;
722 	zfs_object_ace_t *zobjacep;
723 	size_t ace_size;
724 	uint16_t entry_type;
725 
726 	while ((zacep = zfs_acl_next_ace(aclp, zacep,
727 	    &who, &access_mask, &iflags, &type))) {
728 
729 		switch (type) {
730 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
731 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
732 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
733 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
734 			if (filter) {
735 				continue;
736 			}
737 			zobjacep = (zfs_object_ace_t *)zacep;
738 			objacep = (ace_object_t *)acep;
739 			memcpy(objacep->a_obj_type,
740 			    zobjacep->z_object_type,
741 			    sizeof (zobjacep->z_object_type));
742 			memcpy(objacep->a_inherit_obj_type,
743 			    zobjacep->z_inherit_type,
744 			    sizeof (zobjacep->z_inherit_type));
745 			ace_size = sizeof (ace_object_t);
746 			break;
747 		default:
748 			ace_size = sizeof (ace_t);
749 			break;
750 		}
751 
752 		entry_type = (iflags & ACE_TYPE_FLAGS);
753 		if ((entry_type != ACE_OWNER &&
754 		    entry_type != OWNING_GROUP &&
755 		    entry_type != ACE_EVERYONE)) {
756 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
757 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
758 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
759 		} else {
760 			acep->a_who = (uid_t)(int64_t)who;
761 		}
762 		acep->a_access_mask = access_mask;
763 		acep->a_flags = iflags;
764 		acep->a_type = type;
765 		acep = (ace_t *)((caddr_t)acep + ace_size);
766 	}
767 }
768 
769 static int
770 zfs_copy_ace_2_oldace(umode_t obj_mode, zfs_acl_t *aclp, ace_t *acep,
771     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
772 {
773 	int i;
774 	zfs_oldace_t *aceptr = z_acl;
775 
776 	for (i = 0; i != aclcnt; i++, aceptr++) {
777 		aceptr->z_access_mask = acep[i].a_access_mask;
778 		aceptr->z_type = acep[i].a_type;
779 		aceptr->z_flags = acep[i].a_flags;
780 		aceptr->z_fuid = acep[i].a_who;
781 		/*
782 		 * Make sure ACE is valid
783 		 */
784 		if (zfs_ace_valid(obj_mode, aclp, aceptr->z_type,
785 		    aceptr->z_flags) != B_TRUE)
786 			return (SET_ERROR(EINVAL));
787 	}
788 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
789 	return (0);
790 }
791 
792 /*
793  * convert old ACL format to new
794  */
795 void
796 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
797 {
798 	zfs_oldace_t *oldaclp;
799 	int i;
800 	uint16_t type, iflags;
801 	uint32_t access_mask;
802 	uint64_t who;
803 	void *cookie = NULL;
804 	zfs_acl_node_t *newaclnode;
805 
806 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
807 	/*
808 	 * First create the ACE in a contiguous piece of memory
809 	 * for zfs_copy_ace_2_fuid().
810 	 *
811 	 * We only convert an ACL once, so this won't happen
812 	 * every time.
813 	 */
814 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
815 	    KM_SLEEP);
816 	i = 0;
817 	while ((cookie = zfs_acl_next_ace(aclp, cookie, &who,
818 	    &access_mask, &iflags, &type))) {
819 		oldaclp[i].z_flags = iflags;
820 		oldaclp[i].z_type = type;
821 		oldaclp[i].z_fuid = who;
822 		oldaclp[i++].z_access_mask = access_mask;
823 	}
824 
825 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
826 	    sizeof (zfs_object_ace_t));
827 	aclp->z_ops = &zfs_acl_fuid_ops;
828 	VERIFY(zfs_copy_ace_2_fuid(ZTOZSB(zp), ZTOI(zp)->i_mode,
829 	    aclp, oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
830 	    &newaclnode->z_size, NULL, cr) == 0);
831 	newaclnode->z_ace_count = aclp->z_acl_count;
832 	aclp->z_version = ZFS_ACL_VERSION;
833 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
834 
835 	/*
836 	 * Release all previous ACL nodes
837 	 */
838 
839 	zfs_acl_release_nodes(aclp);
840 
841 	list_insert_head(&aclp->z_acl, newaclnode);
842 
843 	aclp->z_acl_bytes = newaclnode->z_size;
844 	aclp->z_acl_count = newaclnode->z_ace_count;
845 
846 }
847 
848 /*
849  * Convert unix access mask to v4 access mask
850  */
851 static uint32_t
852 zfs_unix_to_v4(uint32_t access_mask)
853 {
854 	uint32_t new_mask = 0;
855 
856 	if (access_mask & S_IXOTH)
857 		new_mask |= ACE_EXECUTE;
858 	if (access_mask & S_IWOTH)
859 		new_mask |= ACE_WRITE_DATA;
860 	if (access_mask & S_IROTH)
861 		new_mask |= ACE_READ_DATA;
862 	return (new_mask);
863 }
864 
865 
866 static int
867 zfs_v4_to_unix(uint32_t access_mask, int *unmapped)
868 {
869 	int new_mask = 0;
870 
871 	*unmapped = access_mask &
872 	    (ACE_WRITE_OWNER | ACE_WRITE_ACL | ACE_DELETE);
873 
874 	if (access_mask & WRITE_MASK)
875 		new_mask |= S_IWOTH;
876 	if (access_mask & ACE_READ_DATA)
877 		new_mask |= S_IROTH;
878 	if (access_mask & ACE_EXECUTE)
879 		new_mask |= S_IXOTH;
880 
881 	return (new_mask);
882 }
883 
884 
885 static void
886 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
887     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
888 {
889 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
890 
891 	aclp->z_ops->ace_mask_set(acep, access_mask);
892 	aclp->z_ops->ace_type_set(acep, access_type);
893 	aclp->z_ops->ace_flags_set(acep, entry_type);
894 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
895 	    type != ACE_EVERYONE))
896 		aclp->z_ops->ace_who_set(acep, fuid);
897 }
898 
899 /*
900  * Determine mode of file based on ACL.
901  */
902 uint64_t
903 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
904     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
905 {
906 	int		entry_type;
907 	mode_t		mode;
908 	mode_t		seen = 0;
909 	zfs_ace_hdr_t 	*acep = NULL;
910 	uint64_t	who;
911 	uint16_t	iflags, type;
912 	uint32_t	access_mask;
913 	boolean_t	an_exec_denied = B_FALSE;
914 
915 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
916 
917 	while ((acep = zfs_acl_next_ace(aclp, acep, &who,
918 	    &access_mask, &iflags, &type))) {
919 
920 		if (!zfs_acl_valid_ace_type(type, iflags))
921 			continue;
922 
923 		entry_type = (iflags & ACE_TYPE_FLAGS);
924 
925 		/*
926 		 * Skip over any inherit_only ACEs
927 		 */
928 		if (iflags & ACE_INHERIT_ONLY_ACE)
929 			continue;
930 
931 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
932 		    who == fuid)) {
933 			if ((access_mask & ACE_READ_DATA) &&
934 			    (!(seen & S_IRUSR))) {
935 				seen |= S_IRUSR;
936 				if (type == ALLOW) {
937 					mode |= S_IRUSR;
938 				}
939 			}
940 			if ((access_mask & ACE_WRITE_DATA) &&
941 			    (!(seen & S_IWUSR))) {
942 				seen |= S_IWUSR;
943 				if (type == ALLOW) {
944 					mode |= S_IWUSR;
945 				}
946 			}
947 			if ((access_mask & ACE_EXECUTE) &&
948 			    (!(seen & S_IXUSR))) {
949 				seen |= S_IXUSR;
950 				if (type == ALLOW) {
951 					mode |= S_IXUSR;
952 				}
953 			}
954 		} else if (entry_type == OWNING_GROUP ||
955 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
956 			if ((access_mask & ACE_READ_DATA) &&
957 			    (!(seen & S_IRGRP))) {
958 				seen |= S_IRGRP;
959 				if (type == ALLOW) {
960 					mode |= S_IRGRP;
961 				}
962 			}
963 			if ((access_mask & ACE_WRITE_DATA) &&
964 			    (!(seen & S_IWGRP))) {
965 				seen |= S_IWGRP;
966 				if (type == ALLOW) {
967 					mode |= S_IWGRP;
968 				}
969 			}
970 			if ((access_mask & ACE_EXECUTE) &&
971 			    (!(seen & S_IXGRP))) {
972 				seen |= S_IXGRP;
973 				if (type == ALLOW) {
974 					mode |= S_IXGRP;
975 				}
976 			}
977 		} else if (entry_type == ACE_EVERYONE) {
978 			if ((access_mask & ACE_READ_DATA)) {
979 				if (!(seen & S_IRUSR)) {
980 					seen |= S_IRUSR;
981 					if (type == ALLOW) {
982 						mode |= S_IRUSR;
983 					}
984 				}
985 				if (!(seen & S_IRGRP)) {
986 					seen |= S_IRGRP;
987 					if (type == ALLOW) {
988 						mode |= S_IRGRP;
989 					}
990 				}
991 				if (!(seen & S_IROTH)) {
992 					seen |= S_IROTH;
993 					if (type == ALLOW) {
994 						mode |= S_IROTH;
995 					}
996 				}
997 			}
998 			if ((access_mask & ACE_WRITE_DATA)) {
999 				if (!(seen & S_IWUSR)) {
1000 					seen |= S_IWUSR;
1001 					if (type == ALLOW) {
1002 						mode |= S_IWUSR;
1003 					}
1004 				}
1005 				if (!(seen & S_IWGRP)) {
1006 					seen |= S_IWGRP;
1007 					if (type == ALLOW) {
1008 						mode |= S_IWGRP;
1009 					}
1010 				}
1011 				if (!(seen & S_IWOTH)) {
1012 					seen |= S_IWOTH;
1013 					if (type == ALLOW) {
1014 						mode |= S_IWOTH;
1015 					}
1016 				}
1017 			}
1018 			if ((access_mask & ACE_EXECUTE)) {
1019 				if (!(seen & S_IXUSR)) {
1020 					seen |= S_IXUSR;
1021 					if (type == ALLOW) {
1022 						mode |= S_IXUSR;
1023 					}
1024 				}
1025 				if (!(seen & S_IXGRP)) {
1026 					seen |= S_IXGRP;
1027 					if (type == ALLOW) {
1028 						mode |= S_IXGRP;
1029 					}
1030 				}
1031 				if (!(seen & S_IXOTH)) {
1032 					seen |= S_IXOTH;
1033 					if (type == ALLOW) {
1034 						mode |= S_IXOTH;
1035 					}
1036 				}
1037 			}
1038 		} else {
1039 			/*
1040 			 * Only care if this IDENTIFIER_GROUP or
1041 			 * USER ACE denies execute access to someone,
1042 			 * mode is not affected
1043 			 */
1044 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1045 				an_exec_denied = B_TRUE;
1046 		}
1047 	}
1048 
1049 	/*
1050 	 * Failure to allow is effectively a deny, so execute permission
1051 	 * is denied if it was never mentioned or if we explicitly
1052 	 * weren't allowed it.
1053 	 */
1054 	if (!an_exec_denied &&
1055 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1056 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1057 		an_exec_denied = B_TRUE;
1058 
1059 	if (an_exec_denied)
1060 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1061 	else
1062 		*pflags |= ZFS_NO_EXECS_DENIED;
1063 
1064 	return (mode);
1065 }
1066 
1067 /*
1068  * Read an external acl object.  If the intent is to modify, always
1069  * create a new acl and leave any cached acl in place.
1070  */
1071 int
1072 zfs_acl_node_read(struct znode *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1073     boolean_t will_modify)
1074 {
1075 	zfs_acl_t	*aclp;
1076 	int		aclsize = 0;
1077 	int		acl_count = 0;
1078 	zfs_acl_node_t	*aclnode;
1079 	zfs_acl_phys_t	znode_acl;
1080 	int		version;
1081 	int		error;
1082 	boolean_t	drop_lock = B_FALSE;
1083 
1084 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1085 
1086 	if (zp->z_acl_cached && !will_modify) {
1087 		*aclpp = zp->z_acl_cached;
1088 		return (0);
1089 	}
1090 
1091 	/*
1092 	 * close race where znode could be upgrade while trying to
1093 	 * read the znode attributes.
1094 	 *
1095 	 * But this could only happen if the file isn't already an SA
1096 	 * znode
1097 	 */
1098 	if (!zp->z_is_sa && !have_lock) {
1099 		mutex_enter(&zp->z_lock);
1100 		drop_lock = B_TRUE;
1101 	}
1102 	version = zfs_znode_acl_version(zp);
1103 
1104 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1105 	    &acl_count, &znode_acl)) != 0) {
1106 		goto done;
1107 	}
1108 
1109 	aclp = zfs_acl_alloc(version);
1110 
1111 	aclp->z_acl_count = acl_count;
1112 	aclp->z_acl_bytes = aclsize;
1113 
1114 	aclnode = zfs_acl_node_alloc(aclsize);
1115 	aclnode->z_ace_count = aclp->z_acl_count;
1116 	aclnode->z_size = aclsize;
1117 
1118 	if (!zp->z_is_sa) {
1119 		if (znode_acl.z_acl_extern_obj) {
1120 			error = dmu_read(ZTOZSB(zp)->z_os,
1121 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1122 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1123 		} else {
1124 			memcpy(aclnode->z_acldata, znode_acl.z_ace_data,
1125 			    aclnode->z_size);
1126 		}
1127 	} else {
1128 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(ZTOZSB(zp)),
1129 		    aclnode->z_acldata, aclnode->z_size);
1130 	}
1131 
1132 	if (error != 0) {
1133 		zfs_acl_free(aclp);
1134 		zfs_acl_node_free(aclnode);
1135 		/* convert checksum errors into IO errors */
1136 		if (error == ECKSUM)
1137 			error = SET_ERROR(EIO);
1138 		goto done;
1139 	}
1140 
1141 	list_insert_head(&aclp->z_acl, aclnode);
1142 
1143 	*aclpp = aclp;
1144 	if (!will_modify)
1145 		zp->z_acl_cached = aclp;
1146 done:
1147 	if (drop_lock)
1148 		mutex_exit(&zp->z_lock);
1149 	return (error);
1150 }
1151 
1152 void
1153 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1154     boolean_t start, void *userdata)
1155 {
1156 	(void) buflen;
1157 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1158 
1159 	if (start) {
1160 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1161 	} else {
1162 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1163 		    cb->cb_acl_node);
1164 	}
1165 	ASSERT3P(cb->cb_acl_node, !=, NULL);
1166 	*dataptr = cb->cb_acl_node->z_acldata;
1167 	*length = cb->cb_acl_node->z_size;
1168 }
1169 
1170 int
1171 zfs_acl_chown_setattr(znode_t *zp)
1172 {
1173 	int error;
1174 	zfs_acl_t *aclp;
1175 
1176 	if (ZTOZSB(zp)->z_acl_type == ZFS_ACLTYPE_POSIX)
1177 		return (0);
1178 
1179 	ASSERT(MUTEX_HELD(&zp->z_lock));
1180 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1181 
1182 	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1183 	if (error == 0 && aclp->z_acl_count > 0)
1184 		zp->z_mode = ZTOI(zp)->i_mode =
1185 		    zfs_mode_compute(zp->z_mode, aclp,
1186 		    &zp->z_pflags, KUID_TO_SUID(ZTOI(zp)->i_uid),
1187 		    KGID_TO_SGID(ZTOI(zp)->i_gid));
1188 
1189 	/*
1190 	 * Some ZFS implementations (ZEVO) create neither a ZNODE_ACL
1191 	 * nor a DACL_ACES SA in which case ENOENT is returned from
1192 	 * zfs_acl_node_read() when the SA can't be located.
1193 	 * Allow chown/chgrp to succeed in these cases rather than
1194 	 * returning an error that makes no sense in the context of
1195 	 * the caller.
1196 	 */
1197 	if (error == ENOENT)
1198 		return (0);
1199 
1200 	return (error);
1201 }
1202 
1203 typedef struct trivial_acl {
1204 	uint32_t	allow0;		/* allow mask for bits only in owner */
1205 	uint32_t	deny1;		/* deny mask for bits not in owner */
1206 	uint32_t	deny2;		/* deny mask for bits not in group */
1207 	uint32_t	owner;		/* allow mask matching mode */
1208 	uint32_t	group;		/* allow mask matching mode */
1209 	uint32_t	everyone;	/* allow mask matching mode */
1210 } trivial_acl_t;
1211 
1212 static void
1213 acl_trivial_access_masks(mode_t mode, boolean_t isdir, trivial_acl_t *masks)
1214 {
1215 	uint32_t read_mask = ACE_READ_DATA;
1216 	uint32_t write_mask = ACE_WRITE_DATA|ACE_APPEND_DATA;
1217 	uint32_t execute_mask = ACE_EXECUTE;
1218 
1219 	if (isdir)
1220 		write_mask |= ACE_DELETE_CHILD;
1221 
1222 	masks->deny1 = 0;
1223 
1224 	if (!(mode & S_IRUSR) && (mode & (S_IRGRP|S_IROTH)))
1225 		masks->deny1 |= read_mask;
1226 	if (!(mode & S_IWUSR) && (mode & (S_IWGRP|S_IWOTH)))
1227 		masks->deny1 |= write_mask;
1228 	if (!(mode & S_IXUSR) && (mode & (S_IXGRP|S_IXOTH)))
1229 		masks->deny1 |= execute_mask;
1230 
1231 	masks->deny2 = 0;
1232 	if (!(mode & S_IRGRP) && (mode & S_IROTH))
1233 		masks->deny2 |= read_mask;
1234 	if (!(mode & S_IWGRP) && (mode & S_IWOTH))
1235 		masks->deny2 |= write_mask;
1236 	if (!(mode & S_IXGRP) && (mode & S_IXOTH))
1237 		masks->deny2 |= execute_mask;
1238 
1239 	masks->allow0 = 0;
1240 	if ((mode & S_IRUSR) && (!(mode & S_IRGRP) && (mode & S_IROTH)))
1241 		masks->allow0 |= read_mask;
1242 	if ((mode & S_IWUSR) && (!(mode & S_IWGRP) && (mode & S_IWOTH)))
1243 		masks->allow0 |= write_mask;
1244 	if ((mode & S_IXUSR) && (!(mode & S_IXGRP) && (mode & S_IXOTH)))
1245 		masks->allow0 |= execute_mask;
1246 
1247 	masks->owner = ACE_WRITE_ATTRIBUTES|ACE_WRITE_OWNER|ACE_WRITE_ACL|
1248 	    ACE_WRITE_NAMED_ATTRS|ACE_READ_ACL|ACE_READ_ATTRIBUTES|
1249 	    ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE;
1250 	if (mode & S_IRUSR)
1251 		masks->owner |= read_mask;
1252 	if (mode & S_IWUSR)
1253 		masks->owner |= write_mask;
1254 	if (mode & S_IXUSR)
1255 		masks->owner |= execute_mask;
1256 
1257 	masks->group = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1258 	    ACE_SYNCHRONIZE;
1259 	if (mode & S_IRGRP)
1260 		masks->group |= read_mask;
1261 	if (mode & S_IWGRP)
1262 		masks->group |= write_mask;
1263 	if (mode & S_IXGRP)
1264 		masks->group |= execute_mask;
1265 
1266 	masks->everyone = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1267 	    ACE_SYNCHRONIZE;
1268 	if (mode & S_IROTH)
1269 		masks->everyone |= read_mask;
1270 	if (mode & S_IWOTH)
1271 		masks->everyone |= write_mask;
1272 	if (mode & S_IXOTH)
1273 		masks->everyone |= execute_mask;
1274 }
1275 
1276 /*
1277  * ace_trivial:
1278  * determine whether an ace_t acl is trivial
1279  *
1280  * Trivialness implies that the acl is composed of only
1281  * owner, group, everyone entries.  ACL can't
1282  * have read_acl denied, and write_owner/write_acl/write_attributes
1283  * can only be owner@ entry.
1284  */
1285 static int
1286 ace_trivial_common(void *acep, int aclcnt,
1287     uintptr_t (*walk)(void *, uintptr_t, int,
1288     uint16_t *, uint16_t *, uint32_t *))
1289 {
1290 	uint16_t flags;
1291 	uint32_t mask;
1292 	uint16_t type;
1293 	uint64_t cookie = 0;
1294 
1295 	while ((cookie = walk(acep, cookie, aclcnt, &flags, &type, &mask))) {
1296 		switch (flags & ACE_TYPE_FLAGS) {
1297 		case ACE_OWNER:
1298 		case ACE_GROUP|ACE_IDENTIFIER_GROUP:
1299 		case ACE_EVERYONE:
1300 			break;
1301 		default:
1302 			return (1);
1303 		}
1304 
1305 		if (flags & (ACE_FILE_INHERIT_ACE|
1306 		    ACE_DIRECTORY_INHERIT_ACE|ACE_NO_PROPAGATE_INHERIT_ACE|
1307 		    ACE_INHERIT_ONLY_ACE))
1308 			return (1);
1309 
1310 		/*
1311 		 * Special check for some special bits
1312 		 *
1313 		 * Don't allow anybody to deny reading basic
1314 		 * attributes or a files ACL.
1315 		 */
1316 		if ((mask & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
1317 		    (type == ACE_ACCESS_DENIED_ACE_TYPE))
1318 			return (1);
1319 
1320 		/*
1321 		 * Delete permission is never set by default
1322 		 */
1323 		if (mask & ACE_DELETE)
1324 			return (1);
1325 
1326 		/*
1327 		 * Child delete permission should be accompanied by write
1328 		 */
1329 		if ((mask & ACE_DELETE_CHILD) && !(mask & ACE_WRITE_DATA))
1330 			return (1);
1331 
1332 		/*
1333 		 * only allow owner@ to have
1334 		 * write_acl/write_owner/write_attributes/write_xattr/
1335 		 */
1336 		if (type == ACE_ACCESS_ALLOWED_ACE_TYPE &&
1337 		    (!(flags & ACE_OWNER) && (mask &
1338 		    (ACE_WRITE_OWNER|ACE_WRITE_ACL| ACE_WRITE_ATTRIBUTES|
1339 		    ACE_WRITE_NAMED_ATTRS))))
1340 			return (1);
1341 
1342 	}
1343 
1344 	return (0);
1345 }
1346 
1347 /*
1348  * common code for setting ACLs.
1349  *
1350  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1351  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1352  * already checked the acl and knows whether to inherit.
1353  */
1354 int
1355 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1356 {
1357 	int			error;
1358 	zfsvfs_t		*zfsvfs = ZTOZSB(zp);
1359 	dmu_object_type_t	otype;
1360 	zfs_acl_locator_cb_t	locate = { 0 };
1361 	uint64_t		mode;
1362 	sa_bulk_attr_t		bulk[5];
1363 	uint64_t		ctime[2];
1364 	int			count = 0;
1365 	zfs_acl_phys_t		acl_phys;
1366 
1367 	mode = zp->z_mode;
1368 
1369 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1370 	    KUID_TO_SUID(ZTOI(zp)->i_uid), KGID_TO_SGID(ZTOI(zp)->i_gid));
1371 
1372 	zp->z_mode = ZTOI(zp)->i_mode = mode;
1373 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1374 	    &mode, sizeof (mode));
1375 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1376 	    &zp->z_pflags, sizeof (zp->z_pflags));
1377 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1378 	    &ctime, sizeof (ctime));
1379 
1380 	if (zp->z_acl_cached) {
1381 		zfs_acl_free(zp->z_acl_cached);
1382 		zp->z_acl_cached = NULL;
1383 	}
1384 
1385 	/*
1386 	 * Upgrade needed?
1387 	 */
1388 	if (!zfsvfs->z_use_fuids) {
1389 		otype = DMU_OT_OLDACL;
1390 	} else {
1391 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1392 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1393 			zfs_acl_xform(zp, aclp, cr);
1394 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1395 		otype = DMU_OT_ACL;
1396 	}
1397 
1398 	/*
1399 	 * Arrgh, we have to handle old on disk format
1400 	 * as well as newer (preferred) SA format.
1401 	 */
1402 
1403 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1404 		locate.cb_aclp = aclp;
1405 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1406 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1407 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1408 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1409 	} else { /* Painful legacy way */
1410 		zfs_acl_node_t *aclnode;
1411 		uint64_t off = 0;
1412 		uint64_t aoid;
1413 
1414 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1415 		    &acl_phys, sizeof (acl_phys))) != 0)
1416 			return (error);
1417 
1418 		aoid = acl_phys.z_acl_extern_obj;
1419 
1420 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1421 			/*
1422 			 * If ACL was previously external and we are now
1423 			 * converting to new ACL format then release old
1424 			 * ACL object and create a new one.
1425 			 */
1426 			if (aoid &&
1427 			    aclp->z_version != acl_phys.z_acl_version) {
1428 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1429 				if (error)
1430 					return (error);
1431 				aoid = 0;
1432 			}
1433 			if (aoid == 0) {
1434 				aoid = dmu_object_alloc(zfsvfs->z_os,
1435 				    otype, aclp->z_acl_bytes,
1436 				    otype == DMU_OT_ACL ?
1437 				    DMU_OT_SYSACL : DMU_OT_NONE,
1438 				    otype == DMU_OT_ACL ?
1439 				    DN_OLD_MAX_BONUSLEN : 0, tx);
1440 			} else {
1441 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1442 				    aoid, aclp->z_acl_bytes, 0, tx);
1443 			}
1444 			acl_phys.z_acl_extern_obj = aoid;
1445 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1446 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1447 				if (aclnode->z_ace_count == 0)
1448 					continue;
1449 				dmu_write(zfsvfs->z_os, aoid, off,
1450 				    aclnode->z_size, aclnode->z_acldata, tx,
1451 				    DMU_READ_NO_PREFETCH);
1452 				off += aclnode->z_size;
1453 			}
1454 		} else {
1455 			void *start = acl_phys.z_ace_data;
1456 			/*
1457 			 * Migrating back embedded?
1458 			 */
1459 			if (acl_phys.z_acl_extern_obj) {
1460 				error = dmu_object_free(zfsvfs->z_os,
1461 				    acl_phys.z_acl_extern_obj, tx);
1462 				if (error)
1463 					return (error);
1464 				acl_phys.z_acl_extern_obj = 0;
1465 			}
1466 
1467 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1468 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1469 				if (aclnode->z_ace_count == 0)
1470 					continue;
1471 				memcpy(start, aclnode->z_acldata,
1472 				    aclnode->z_size);
1473 				start = (caddr_t)start + aclnode->z_size;
1474 			}
1475 		}
1476 		/*
1477 		 * If Old version then swap count/bytes to match old
1478 		 * layout of znode_acl_phys_t.
1479 		 */
1480 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1481 			acl_phys.z_acl_size = aclp->z_acl_count;
1482 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1483 		} else {
1484 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1485 			acl_phys.z_acl_count = aclp->z_acl_count;
1486 		}
1487 		acl_phys.z_acl_version = aclp->z_version;
1488 
1489 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1490 		    &acl_phys, sizeof (acl_phys));
1491 	}
1492 
1493 	/*
1494 	 * Replace ACL wide bits, but first clear them.
1495 	 */
1496 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1497 
1498 	zp->z_pflags |= aclp->z_hints;
1499 
1500 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1501 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1502 
1503 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1504 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1505 }
1506 
1507 static void
1508 zfs_acl_chmod(boolean_t isdir, uint64_t mode, boolean_t split, boolean_t trim,
1509     zfs_acl_t *aclp)
1510 {
1511 	void		*acep = NULL;
1512 	uint64_t	who;
1513 	int		new_count, new_bytes;
1514 	int		ace_size;
1515 	int		entry_type;
1516 	uint16_t	iflags, type;
1517 	uint32_t	access_mask;
1518 	zfs_acl_node_t	*newnode;
1519 	size_t		abstract_size = aclp->z_ops->ace_abstract_size();
1520 	void		*zacep;
1521 	trivial_acl_t	masks;
1522 
1523 	new_count = new_bytes = 0;
1524 
1525 	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1526 
1527 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1528 
1529 	zacep = newnode->z_acldata;
1530 	if (masks.allow0) {
1531 		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1532 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1533 		new_count++;
1534 		new_bytes += abstract_size;
1535 	}
1536 	if (masks.deny1) {
1537 		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1538 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1539 		new_count++;
1540 		new_bytes += abstract_size;
1541 	}
1542 	if (masks.deny2) {
1543 		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1544 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1545 		new_count++;
1546 		new_bytes += abstract_size;
1547 	}
1548 
1549 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1550 	    &iflags, &type))) {
1551 		entry_type = (iflags & ACE_TYPE_FLAGS);
1552 		/*
1553 		 * ACEs used to represent the file mode may be divided
1554 		 * into an equivalent pair of inherit-only and regular
1555 		 * ACEs, if they are inheritable.
1556 		 * Skip regular ACEs, which are replaced by the new mode.
1557 		 */
1558 		if (split && (entry_type == ACE_OWNER ||
1559 		    entry_type == OWNING_GROUP ||
1560 		    entry_type == ACE_EVERYONE)) {
1561 			if (!isdir || !(iflags &
1562 			    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1563 				continue;
1564 			/*
1565 			 * We preserve owner@, group@, or @everyone
1566 			 * permissions, if they are inheritable, by
1567 			 * copying them to inherit_only ACEs. This
1568 			 * prevents inheritable permissions from being
1569 			 * altered along with the file mode.
1570 			 */
1571 			iflags |= ACE_INHERIT_ONLY_ACE;
1572 		}
1573 
1574 		/*
1575 		 * If this ACL has any inheritable ACEs, mark that in
1576 		 * the hints (which are later masked into the pflags)
1577 		 * so create knows to do inheritance.
1578 		 */
1579 		if (isdir && (iflags &
1580 		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1581 			aclp->z_hints |= ZFS_INHERIT_ACE;
1582 
1583 		if ((type != ALLOW && type != DENY) ||
1584 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1585 			switch (type) {
1586 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1587 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1588 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1589 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1590 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1591 				break;
1592 			}
1593 		} else {
1594 			/*
1595 			 * Limit permissions to be no greater than
1596 			 * group permissions.
1597 			 * The "aclinherit" and "aclmode" properties
1598 			 * affect policy for create and chmod(2),
1599 			 * respectively.
1600 			 */
1601 			if ((type == ALLOW) && trim)
1602 				access_mask &= masks.group;
1603 		}
1604 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1605 		ace_size = aclp->z_ops->ace_size(acep);
1606 		zacep = (void *)((uintptr_t)zacep + ace_size);
1607 		new_count++;
1608 		new_bytes += ace_size;
1609 	}
1610 	zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1611 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1612 	zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1613 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1614 	zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1615 
1616 	new_count += 3;
1617 	new_bytes += abstract_size * 3;
1618 	zfs_acl_release_nodes(aclp);
1619 	aclp->z_acl_count = new_count;
1620 	aclp->z_acl_bytes = new_bytes;
1621 	newnode->z_ace_count = new_count;
1622 	newnode->z_size = new_bytes;
1623 	list_insert_tail(&aclp->z_acl, newnode);
1624 }
1625 
1626 int
1627 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1628 {
1629 	int error = 0;
1630 
1631 	mutex_enter(&zp->z_acl_lock);
1632 	mutex_enter(&zp->z_lock);
1633 	if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_DISCARD)
1634 		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1635 	else
1636 		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1637 
1638 	if (error == 0) {
1639 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1640 		zfs_acl_chmod(S_ISDIR(ZTOI(zp)->i_mode), mode, B_TRUE,
1641 		    (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1642 	}
1643 	mutex_exit(&zp->z_lock);
1644 	mutex_exit(&zp->z_acl_lock);
1645 
1646 	return (error);
1647 }
1648 
1649 /*
1650  * Should ACE be inherited?
1651  */
1652 static int
1653 zfs_ace_can_use(umode_t obj_mode, uint16_t acep_flags)
1654 {
1655 	int	iflags = (acep_flags & 0xf);
1656 
1657 	if (S_ISDIR(obj_mode) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1658 		return (1);
1659 	else if (iflags & ACE_FILE_INHERIT_ACE)
1660 		return (!(S_ISDIR(obj_mode) &&
1661 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1662 	return (0);
1663 }
1664 
1665 /*
1666  * inherit inheritable ACEs from parent
1667  */
1668 static zfs_acl_t *
1669 zfs_acl_inherit(zfsvfs_t *zfsvfs, umode_t va_mode, zfs_acl_t *paclp,
1670     uint64_t mode, boolean_t *need_chmod)
1671 {
1672 	void		*pacep = NULL;
1673 	void		*acep;
1674 	zfs_acl_node_t  *aclnode;
1675 	zfs_acl_t	*aclp = NULL;
1676 	uint64_t	who;
1677 	uint32_t	access_mask;
1678 	uint16_t	iflags, newflags, type;
1679 	size_t		ace_size;
1680 	void		*data1, *data2;
1681 	size_t		data1sz, data2sz;
1682 	uint_t		aclinherit;
1683 	boolean_t	isdir = S_ISDIR(va_mode);
1684 	boolean_t	isreg = S_ISREG(va_mode);
1685 
1686 	*need_chmod = B_TRUE;
1687 
1688 	aclp = zfs_acl_alloc(paclp->z_version);
1689 	aclinherit = zfsvfs->z_acl_inherit;
1690 	if (aclinherit == ZFS_ACL_DISCARD || S_ISLNK(va_mode))
1691 		return (aclp);
1692 
1693 	while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1694 	    &access_mask, &iflags, &type))) {
1695 
1696 		/*
1697 		 * don't inherit bogus ACEs
1698 		 */
1699 		if (!zfs_acl_valid_ace_type(type, iflags))
1700 			continue;
1701 
1702 		/*
1703 		 * Check if ACE is inheritable by this vnode
1704 		 */
1705 		if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1706 		    !zfs_ace_can_use(va_mode, iflags))
1707 			continue;
1708 
1709 		/*
1710 		 * If owner@, group@, or everyone@ inheritable
1711 		 * then zfs_acl_chmod() isn't needed.
1712 		 */
1713 		if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1714 		    aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1715 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1716 		    ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1717 		    (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1718 			*need_chmod = B_FALSE;
1719 
1720 		/*
1721 		 * Strip inherited execute permission from file if
1722 		 * not in mode
1723 		 */
1724 		if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1725 		    !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1726 			access_mask &= ~ACE_EXECUTE;
1727 		}
1728 
1729 		/*
1730 		 * Strip write_acl and write_owner from permissions
1731 		 * when inheriting an ACE
1732 		 */
1733 		if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1734 			access_mask &= ~RESTRICTED_CLEAR;
1735 		}
1736 
1737 		ace_size = aclp->z_ops->ace_size(pacep);
1738 		aclnode = zfs_acl_node_alloc(ace_size);
1739 		list_insert_tail(&aclp->z_acl, aclnode);
1740 		acep = aclnode->z_acldata;
1741 
1742 		zfs_set_ace(aclp, acep, access_mask, type,
1743 		    who, iflags|ACE_INHERITED_ACE);
1744 
1745 		/*
1746 		 * Copy special opaque data if any
1747 		 */
1748 		if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1749 			VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1750 			    &data2)) == data1sz);
1751 			memcpy(data2, data1, data2sz);
1752 		}
1753 
1754 		aclp->z_acl_count++;
1755 		aclnode->z_ace_count++;
1756 		aclp->z_acl_bytes += aclnode->z_size;
1757 		newflags = aclp->z_ops->ace_flags_get(acep);
1758 
1759 		/*
1760 		 * If ACE is not to be inherited further, or if the vnode is
1761 		 * not a directory, remove all inheritance flags
1762 		 */
1763 		if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1764 			newflags &= ~ALL_INHERIT;
1765 			aclp->z_ops->ace_flags_set(acep,
1766 			    newflags|ACE_INHERITED_ACE);
1767 			continue;
1768 		}
1769 
1770 		/*
1771 		 * This directory has an inheritable ACE
1772 		 */
1773 		aclp->z_hints |= ZFS_INHERIT_ACE;
1774 
1775 		/*
1776 		 * If only FILE_INHERIT is set then turn on
1777 		 * inherit_only
1778 		 */
1779 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1780 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1781 			newflags |= ACE_INHERIT_ONLY_ACE;
1782 			aclp->z_ops->ace_flags_set(acep,
1783 			    newflags|ACE_INHERITED_ACE);
1784 		} else {
1785 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1786 			aclp->z_ops->ace_flags_set(acep,
1787 			    newflags|ACE_INHERITED_ACE);
1788 		}
1789 	}
1790 	if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1791 	    aclp->z_acl_count != 0) {
1792 		*need_chmod = B_FALSE;
1793 	}
1794 
1795 	return (aclp);
1796 }
1797 
1798 /*
1799  * Create file system object initial permissions
1800  * including inheritable ACEs.
1801  * Also, create FUIDs for owner and group.
1802  */
1803 int
1804 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1805     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids, zidmap_t *mnt_ns)
1806 {
1807 	int		error;
1808 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
1809 	zfs_acl_t	*paclp;
1810 	gid_t		gid = vap->va_gid;
1811 	boolean_t	need_chmod = B_TRUE;
1812 	boolean_t	trim = B_FALSE;
1813 	boolean_t	inherited = B_FALSE;
1814 
1815 	memset(acl_ids, 0, sizeof (zfs_acl_ids_t));
1816 	acl_ids->z_mode = vap->va_mode;
1817 
1818 	if (vsecp)
1819 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_mode, vsecp,
1820 		    cr, &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1821 			return (error);
1822 
1823 	acl_ids->z_fuid = vap->va_uid;
1824 	acl_ids->z_fgid = vap->va_gid;
1825 #ifdef HAVE_KSID
1826 	/*
1827 	 * Determine uid and gid.
1828 	 */
1829 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1830 	    ((flag & IS_XATTR) && (S_ISDIR(vap->va_mode)))) {
1831 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_uid,
1832 		    cr, ZFS_OWNER, &acl_ids->z_fuidp);
1833 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
1834 		    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1835 		gid = vap->va_gid;
1836 	} else {
1837 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1838 		    cr, &acl_ids->z_fuidp);
1839 		acl_ids->z_fgid = 0;
1840 		if (vap->va_mask & AT_GID)  {
1841 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1842 			    (uint64_t)vap->va_gid,
1843 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1844 			gid = vap->va_gid;
1845 			if (acl_ids->z_fgid != KGID_TO_SGID(ZTOI(dzp)->i_gid) &&
1846 			    !groupmember(vap->va_gid, cr) &&
1847 			    secpolicy_vnode_create_gid(cr) != 0)
1848 				acl_ids->z_fgid = 0;
1849 		}
1850 		if (acl_ids->z_fgid == 0) {
1851 			if (dzp->z_mode & S_ISGID) {
1852 				char		*domain;
1853 				uint32_t	rid;
1854 
1855 				acl_ids->z_fgid = KGID_TO_SGID(
1856 				    ZTOI(dzp)->i_gid);
1857 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1858 				    cr, ZFS_GROUP);
1859 
1860 				if (zfsvfs->z_use_fuids &&
1861 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1862 					domain = zfs_fuid_idx_domain(
1863 					    &zfsvfs->z_fuid_idx,
1864 					    FUID_INDEX(acl_ids->z_fgid));
1865 					rid = FUID_RID(acl_ids->z_fgid);
1866 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1867 					    domain, rid,
1868 					    FUID_INDEX(acl_ids->z_fgid),
1869 					    acl_ids->z_fgid, ZFS_GROUP);
1870 				}
1871 			} else {
1872 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1873 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1874 				gid = crgetgid(cr);
1875 			}
1876 		}
1877 	}
1878 #endif /* HAVE_KSID */
1879 
1880 	/*
1881 	 * If we're creating a directory, and the parent directory has the
1882 	 * set-GID bit set, set in on the new directory.
1883 	 * Otherwise, if the user is neither privileged nor a member of the
1884 	 * file's new group, clear the file's set-GID bit.
1885 	 */
1886 
1887 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1888 	    (S_ISDIR(vap->va_mode))) {
1889 		acl_ids->z_mode |= S_ISGID;
1890 	} else {
1891 		if ((acl_ids->z_mode & S_ISGID) &&
1892 		    secpolicy_vnode_setids_setgids(cr, gid, mnt_ns,
1893 		    zfs_i_user_ns(ZTOI(dzp))) != 0) {
1894 			acl_ids->z_mode &= ~S_ISGID;
1895 		}
1896 	}
1897 
1898 	if (acl_ids->z_aclp == NULL) {
1899 		mutex_enter(&dzp->z_acl_lock);
1900 		mutex_enter(&dzp->z_lock);
1901 		if (!(flag & IS_ROOT_NODE) &&
1902 		    (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1903 		    !(dzp->z_pflags & ZFS_XATTR)) {
1904 			VERIFY0(zfs_acl_node_read(dzp, B_TRUE,
1905 			    &paclp, B_FALSE));
1906 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1907 			    vap->va_mode, paclp, acl_ids->z_mode, &need_chmod);
1908 			inherited = B_TRUE;
1909 		} else {
1910 			acl_ids->z_aclp =
1911 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1912 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1913 		}
1914 		mutex_exit(&dzp->z_lock);
1915 		mutex_exit(&dzp->z_acl_lock);
1916 
1917 		if (need_chmod) {
1918 			if (S_ISDIR(vap->va_mode))
1919 				acl_ids->z_aclp->z_hints |=
1920 				    ZFS_ACL_AUTO_INHERIT;
1921 
1922 			if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1923 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1924 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1925 				trim = B_TRUE;
1926 			zfs_acl_chmod(S_ISDIR(vap->va_mode), acl_ids->z_mode,
1927 			    B_FALSE, trim, acl_ids->z_aclp);
1928 		}
1929 	}
1930 
1931 	if (inherited || vsecp) {
1932 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1933 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1934 		    acl_ids->z_fuid, acl_ids->z_fgid);
1935 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1936 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1937 	}
1938 
1939 	return (0);
1940 }
1941 
1942 /*
1943  * Free ACL and fuid_infop, but not the acl_ids structure
1944  */
1945 void
1946 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1947 {
1948 	if (acl_ids->z_aclp)
1949 		zfs_acl_free(acl_ids->z_aclp);
1950 	if (acl_ids->z_fuidp)
1951 		zfs_fuid_info_free(acl_ids->z_fuidp);
1952 	acl_ids->z_aclp = NULL;
1953 	acl_ids->z_fuidp = NULL;
1954 }
1955 
1956 boolean_t
1957 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1958 {
1959 	return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1960 	    zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1961 	    (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1962 	    zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1963 }
1964 
1965 /*
1966  * Retrieve a file's ACL
1967  */
1968 int
1969 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1970 {
1971 	zfs_acl_t	*aclp;
1972 	ulong_t		mask;
1973 	int		error;
1974 	int 		count = 0;
1975 	int		largeace = 0;
1976 
1977 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1978 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1979 
1980 	if (mask == 0)
1981 		return (SET_ERROR(ENOSYS));
1982 
1983 	if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr,
1984 	    zfs_init_idmap)))
1985 		return (error);
1986 
1987 	mutex_enter(&zp->z_acl_lock);
1988 
1989 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1990 	if (error != 0) {
1991 		mutex_exit(&zp->z_acl_lock);
1992 		return (error);
1993 	}
1994 
1995 	/*
1996 	 * Scan ACL to determine number of ACEs
1997 	 */
1998 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1999 		void *zacep = NULL;
2000 		uint64_t who;
2001 		uint32_t access_mask;
2002 		uint16_t type, iflags;
2003 
2004 		while ((zacep = zfs_acl_next_ace(aclp, zacep,
2005 		    &who, &access_mask, &iflags, &type))) {
2006 			switch (type) {
2007 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
2008 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
2009 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
2010 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
2011 				largeace++;
2012 				continue;
2013 			default:
2014 				count++;
2015 			}
2016 		}
2017 		vsecp->vsa_aclcnt = count;
2018 	} else
2019 		count = (int)aclp->z_acl_count;
2020 
2021 	if (mask & VSA_ACECNT) {
2022 		vsecp->vsa_aclcnt = count;
2023 	}
2024 
2025 	if (mask & VSA_ACE) {
2026 		size_t aclsz;
2027 
2028 		aclsz = count * sizeof (ace_t) +
2029 		    sizeof (ace_object_t) * largeace;
2030 
2031 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2032 		vsecp->vsa_aclentsz = aclsz;
2033 
2034 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2035 			zfs_copy_fuid_2_ace(ZTOZSB(zp), aclp, cr,
2036 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2037 		else {
2038 			zfs_acl_node_t *aclnode;
2039 			void *start = vsecp->vsa_aclentp;
2040 
2041 			for (aclnode = list_head(&aclp->z_acl); aclnode;
2042 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
2043 				memcpy(start, aclnode->z_acldata,
2044 				    aclnode->z_size);
2045 				start = (caddr_t)start + aclnode->z_size;
2046 			}
2047 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
2048 			    aclp->z_acl_bytes);
2049 		}
2050 	}
2051 	if (mask & VSA_ACE_ACLFLAGS) {
2052 		vsecp->vsa_aclflags = 0;
2053 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
2054 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
2055 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
2056 			vsecp->vsa_aclflags |= ACL_PROTECTED;
2057 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
2058 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2059 	}
2060 
2061 	mutex_exit(&zp->z_acl_lock);
2062 
2063 	return (0);
2064 }
2065 
2066 int
2067 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_mode,
2068     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2069 {
2070 	zfs_acl_t *aclp;
2071 	zfs_acl_node_t *aclnode;
2072 	int aclcnt = vsecp->vsa_aclcnt;
2073 	int error;
2074 
2075 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2076 		return (SET_ERROR(EINVAL));
2077 
2078 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2079 
2080 	aclp->z_hints = 0;
2081 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2082 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2083 		if ((error = zfs_copy_ace_2_oldace(obj_mode, aclp,
2084 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2085 		    aclcnt, &aclnode->z_size)) != 0) {
2086 			zfs_acl_free(aclp);
2087 			zfs_acl_node_free(aclnode);
2088 			return (error);
2089 		}
2090 	} else {
2091 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_mode, aclp,
2092 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2093 		    &aclnode->z_size, fuidp, cr)) != 0) {
2094 			zfs_acl_free(aclp);
2095 			zfs_acl_node_free(aclnode);
2096 			return (error);
2097 		}
2098 	}
2099 	aclp->z_acl_bytes = aclnode->z_size;
2100 	aclnode->z_ace_count = aclcnt;
2101 	aclp->z_acl_count = aclcnt;
2102 	list_insert_head(&aclp->z_acl, aclnode);
2103 
2104 	/*
2105 	 * If flags are being set then add them to z_hints
2106 	 */
2107 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2108 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2109 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2110 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2111 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2112 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2113 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2114 	}
2115 
2116 	*zaclp = aclp;
2117 
2118 	return (0);
2119 }
2120 
2121 /*
2122  * Set a file's ACL
2123  */
2124 int
2125 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2126 {
2127 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
2128 	zilog_t		*zilog = zfsvfs->z_log;
2129 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2130 	dmu_tx_t	*tx;
2131 	int		error;
2132 	zfs_acl_t	*aclp;
2133 	zfs_fuid_info_t	*fuidp = NULL;
2134 	boolean_t	fuid_dirtied;
2135 	uint64_t	acl_obj;
2136 
2137 	if (mask == 0)
2138 		return (SET_ERROR(ENOSYS));
2139 
2140 	if (zp->z_pflags & ZFS_IMMUTABLE)
2141 		return (SET_ERROR(EPERM));
2142 
2143 	if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr,
2144 	    zfs_init_idmap)))
2145 		return (error);
2146 
2147 	error = zfs_vsec_2_aclp(zfsvfs, ZTOI(zp)->i_mode, vsecp, cr, &fuidp,
2148 	    &aclp);
2149 	if (error)
2150 		return (error);
2151 
2152 	/*
2153 	 * If ACL wide flags aren't being set then preserve any
2154 	 * existing flags.
2155 	 */
2156 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2157 		aclp->z_hints |=
2158 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2159 	}
2160 top:
2161 	mutex_enter(&zp->z_acl_lock);
2162 	mutex_enter(&zp->z_lock);
2163 
2164 	tx = dmu_tx_create(zfsvfs->z_os);
2165 
2166 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2167 
2168 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2169 	if (fuid_dirtied)
2170 		zfs_fuid_txhold(zfsvfs, tx);
2171 
2172 	/*
2173 	 * If old version and ACL won't fit in bonus and we aren't
2174 	 * upgrading then take out necessary DMU holds
2175 	 */
2176 
2177 	if ((acl_obj = zfs_external_acl(zp)) != 0) {
2178 		if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2179 		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2180 			dmu_tx_hold_free(tx, acl_obj, 0,
2181 			    DMU_OBJECT_END);
2182 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2183 			    aclp->z_acl_bytes);
2184 		} else {
2185 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2186 		}
2187 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2188 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2189 	}
2190 
2191 	zfs_sa_upgrade_txholds(tx, zp);
2192 	error = dmu_tx_assign(tx, DMU_TX_NOWAIT);
2193 	if (error) {
2194 		mutex_exit(&zp->z_acl_lock);
2195 		mutex_exit(&zp->z_lock);
2196 
2197 		if (error == ERESTART) {
2198 			dmu_tx_wait(tx);
2199 			dmu_tx_abort(tx);
2200 			goto top;
2201 		}
2202 		dmu_tx_abort(tx);
2203 		zfs_acl_free(aclp);
2204 		return (error);
2205 	}
2206 
2207 	error = zfs_aclset_common(zp, aclp, cr, tx);
2208 	ASSERT0(error);
2209 	ASSERT0P(zp->z_acl_cached);
2210 	zp->z_acl_cached = aclp;
2211 
2212 	if (fuid_dirtied)
2213 		zfs_fuid_sync(zfsvfs, tx);
2214 
2215 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2216 
2217 	if (fuidp)
2218 		zfs_fuid_info_free(fuidp);
2219 	dmu_tx_commit(tx);
2220 
2221 	mutex_exit(&zp->z_lock);
2222 	mutex_exit(&zp->z_acl_lock);
2223 
2224 	return (error);
2225 }
2226 
2227 /*
2228  * Check accesses of interest (AoI) against attributes of the dataset
2229  * such as read-only.  Returns zero if no AoI conflict with dataset
2230  * attributes, otherwise an appropriate errno is returned.
2231  */
2232 static int
2233 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2234 {
2235 	if ((v4_mode & WRITE_MASK) && (zfs_is_readonly(ZTOZSB(zp))) &&
2236 	    (!Z_ISDEV(ZTOI(zp)->i_mode) || (v4_mode & WRITE_MASK_ATTRS))) {
2237 		return (SET_ERROR(EROFS));
2238 	}
2239 
2240 	/*
2241 	 * Intentionally allow ZFS_READONLY through here.
2242 	 * See zfs_zaccess_common().
2243 	 */
2244 	if ((v4_mode & WRITE_MASK_DATA) &&
2245 	    (zp->z_pflags & ZFS_IMMUTABLE)) {
2246 		return (SET_ERROR(EPERM));
2247 	}
2248 
2249 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2250 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2251 		return (SET_ERROR(EPERM));
2252 	}
2253 
2254 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2255 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2256 		return (SET_ERROR(EACCES));
2257 	}
2258 
2259 	return (0);
2260 }
2261 
2262 /*
2263  * The primary usage of this function is to loop through all of the
2264  * ACEs in the znode, determining what accesses of interest (AoI) to
2265  * the caller are allowed or denied.  The AoI are expressed as bits in
2266  * the working_mode parameter.  As each ACE is processed, bits covered
2267  * by that ACE are removed from the working_mode.  This removal
2268  * facilitates two things.  The first is that when the working mode is
2269  * empty (= 0), we know we've looked at all the AoI. The second is
2270  * that the ACE interpretation rules don't allow a later ACE to undo
2271  * something granted or denied by an earlier ACE.  Removing the
2272  * discovered access or denial enforces this rule.  At the end of
2273  * processing the ACEs, all AoI that were found to be denied are
2274  * placed into the working_mode, giving the caller a mask of denied
2275  * accesses.  Returns:
2276  *	0		if all AoI granted
2277  *	EACCES 		if the denied mask is non-zero
2278  *	other error	if abnormal failure (e.g., IO error)
2279  *
2280  * A secondary usage of the function is to determine if any of the
2281  * AoI are granted.  If an ACE grants any access in
2282  * the working_mode, we immediately short circuit out of the function.
2283  * This mode is chosen by setting anyaccess to B_TRUE.  The
2284  * working_mode is not a denied access mask upon exit if the function
2285  * is used in this manner.
2286  */
2287 static int
2288 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2289     boolean_t anyaccess, cred_t *cr, zidmap_t *mnt_ns)
2290 {
2291 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
2292 	zfs_acl_t	*aclp;
2293 	int		error;
2294 	uid_t		uid = crgetuid(cr);
2295 	uint64_t	who;
2296 	uint16_t	type, iflags;
2297 	uint16_t	entry_type;
2298 	uint32_t	access_mask;
2299 	uint32_t	deny_mask = 0;
2300 	zfs_ace_hdr_t	*acep = NULL;
2301 	boolean_t	checkit;
2302 	uid_t		gowner;
2303 	uid_t		fowner;
2304 
2305 	if (mnt_ns) {
2306 		fowner = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2307 		    KUID_TO_SUID(ZTOI(zp)->i_uid));
2308 		gowner = zfs_gid_to_vfsgid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2309 		    KGID_TO_SGID(ZTOI(zp)->i_gid));
2310 	} else
2311 		zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2312 
2313 	mutex_enter(&zp->z_acl_lock);
2314 
2315 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2316 	if (error != 0) {
2317 		mutex_exit(&zp->z_acl_lock);
2318 		return (error);
2319 	}
2320 
2321 	ASSERT(zp->z_acl_cached);
2322 
2323 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2324 	    &iflags, &type))) {
2325 		uint32_t mask_matched;
2326 
2327 		if (!zfs_acl_valid_ace_type(type, iflags))
2328 			continue;
2329 
2330 		if (S_ISDIR(ZTOI(zp)->i_mode) &&
2331 		    (iflags & ACE_INHERIT_ONLY_ACE))
2332 			continue;
2333 
2334 		/* Skip ACE if it does not affect any AoI */
2335 		mask_matched = (access_mask & *working_mode);
2336 		if (!mask_matched)
2337 			continue;
2338 
2339 		entry_type = (iflags & ACE_TYPE_FLAGS);
2340 
2341 		checkit = B_FALSE;
2342 
2343 		switch (entry_type) {
2344 		case ACE_OWNER:
2345 			if (uid == fowner)
2346 				checkit = B_TRUE;
2347 			break;
2348 		case OWNING_GROUP:
2349 			who = gowner;
2350 			zfs_fallthrough;
2351 		case ACE_IDENTIFIER_GROUP:
2352 			checkit = zfs_groupmember(zfsvfs, who, cr);
2353 			break;
2354 		case ACE_EVERYONE:
2355 			checkit = B_TRUE;
2356 			break;
2357 
2358 		/* USER Entry */
2359 		default:
2360 			if (entry_type == 0) {
2361 				uid_t newid;
2362 
2363 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2364 				    ZFS_ACE_USER);
2365 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2366 				    uid == newid)
2367 					checkit = B_TRUE;
2368 				break;
2369 			} else {
2370 				mutex_exit(&zp->z_acl_lock);
2371 				return (SET_ERROR(EIO));
2372 			}
2373 		}
2374 
2375 		if (checkit) {
2376 			if (type == DENY) {
2377 				DTRACE_PROBE3(zfs__ace__denies,
2378 				    znode_t *, zp,
2379 				    zfs_ace_hdr_t *, acep,
2380 				    uint32_t, mask_matched);
2381 				deny_mask |= mask_matched;
2382 			} else {
2383 				DTRACE_PROBE3(zfs__ace__allows,
2384 				    znode_t *, zp,
2385 				    zfs_ace_hdr_t *, acep,
2386 				    uint32_t, mask_matched);
2387 				if (anyaccess) {
2388 					mutex_exit(&zp->z_acl_lock);
2389 					return (0);
2390 				}
2391 			}
2392 			*working_mode &= ~mask_matched;
2393 		}
2394 
2395 		/* Are we done? */
2396 		if (*working_mode == 0)
2397 			break;
2398 	}
2399 
2400 	mutex_exit(&zp->z_acl_lock);
2401 
2402 	/* Put the found 'denies' back on the working mode */
2403 	if (deny_mask) {
2404 		*working_mode |= deny_mask;
2405 		return (SET_ERROR(EACCES));
2406 	} else if (*working_mode) {
2407 		return (-1);
2408 	}
2409 
2410 	return (0);
2411 }
2412 
2413 /*
2414  * Return true if any access whatsoever granted, we don't actually
2415  * care what access is granted.
2416  */
2417 boolean_t
2418 zfs_has_access(znode_t *zp, cred_t *cr)
2419 {
2420 	uint32_t have = ACE_ALL_PERMS;
2421 
2422 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr,
2423 	    zfs_init_idmap) != 0) {
2424 		uid_t owner;
2425 
2426 		owner = zfs_fuid_map_id(ZTOZSB(zp),
2427 		    KUID_TO_SUID(ZTOI(zp)->i_uid), cr, ZFS_OWNER);
2428 		return (secpolicy_vnode_any_access(cr, ZTOI(zp), owner) == 0);
2429 	}
2430 	return (B_TRUE);
2431 }
2432 
2433 /*
2434  * Simplified access check for case where ACL is known to not contain
2435  * information beyond what is defined in the mode. In this case, we
2436  * can pass along to the kernel / vfs generic_permission() check, which
2437  * evaluates the mode and POSIX ACL.
2438  *
2439  * NFSv4 ACLs allow granting permissions that are usually relegated only
2440  * to the file owner or superuser. Examples are ACE_WRITE_OWNER (chown),
2441  * ACE_WRITE_ACL(chmod), and ACE_DELETE. ACE_DELETE requests must fail
2442  * because with conventional posix permissions, right to delete file
2443  * is determined by write bit on the parent dir.
2444  *
2445  * If unmappable perms are requested, then we must return EPERM
2446  * and include those bits in the working_mode so that the caller of
2447  * zfs_zaccess_common() can decide whether to perform additional
2448  * policy / capability checks. EACCES is used in zfs_zaccess_aces_check()
2449  * to indicate access check failed due to explicit DENY entry, and so
2450  * we want to avoid that here.
2451  */
2452 static int
2453 zfs_zaccess_trivial(znode_t *zp, uint32_t *working_mode, cred_t *cr,
2454     zidmap_t *mnt_ns)
2455 {
2456 	int err, mask;
2457 	int unmapped = 0;
2458 
2459 	ASSERT(zp->z_pflags & ZFS_ACL_TRIVIAL);
2460 
2461 	mask = zfs_v4_to_unix(*working_mode, &unmapped);
2462 	if (mask == 0 || unmapped) {
2463 		*working_mode = unmapped;
2464 		return (unmapped ? SET_ERROR(EPERM) : 0);
2465 	}
2466 
2467 #if (defined(HAVE_IOPS_PERMISSION_USERNS) || \
2468 	defined(HAVE_IOPS_PERMISSION_IDMAP))
2469 	err = generic_permission(mnt_ns, ZTOI(zp), mask);
2470 #else
2471 	err = generic_permission(ZTOI(zp), mask);
2472 #endif
2473 	if (err != 0) {
2474 		return (SET_ERROR(EPERM));
2475 	}
2476 
2477 	*working_mode = unmapped;
2478 
2479 	return (0);
2480 }
2481 
2482 static int
2483 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2484     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr, zidmap_t *mnt_ns)
2485 {
2486 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
2487 	int err;
2488 
2489 	*working_mode = v4_mode;
2490 	*check_privs = B_TRUE;
2491 
2492 	/*
2493 	 * Short circuit empty requests
2494 	 */
2495 	if (v4_mode == 0 || zfsvfs->z_replay) {
2496 		*working_mode = 0;
2497 		return (0);
2498 	}
2499 
2500 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2501 		*check_privs = B_FALSE;
2502 		return (err);
2503 	}
2504 
2505 	/*
2506 	 * The caller requested that the ACL check be skipped.  This
2507 	 * would only happen if the caller checked VOP_ACCESS() with a
2508 	 * 32 bit ACE mask and already had the appropriate permissions.
2509 	 */
2510 	if (skipaclchk) {
2511 		*working_mode = 0;
2512 		return (0);
2513 	}
2514 
2515 	/*
2516 	 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2517 	 * When that flag is set, we should behave as if write access
2518 	 * were not granted by anything in the ACL.  In particular:
2519 	 * We _must_ allow writes after opening the file r/w, then
2520 	 * setting the DOS R/O attribute, and writing some more.
2521 	 * (Similar to how you can write after fchmod(fd, 0444).)
2522 	 *
2523 	 * Therefore ZFS_READONLY is ignored in the dataset check
2524 	 * above, and checked here as if part of the ACL check.
2525 	 * Also note: DOS R/O is ignored for directories.
2526 	 */
2527 	if ((v4_mode & WRITE_MASK_DATA) &&
2528 	    !S_ISDIR(ZTOI(zp)->i_mode) &&
2529 	    (zp->z_pflags & ZFS_READONLY)) {
2530 		return (SET_ERROR(EPERM));
2531 	}
2532 
2533 	if (zp->z_pflags & ZFS_ACL_TRIVIAL)
2534 		return (zfs_zaccess_trivial(zp, working_mode, cr, mnt_ns));
2535 
2536 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr, mnt_ns));
2537 }
2538 
2539 static int
2540 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2541     cred_t *cr, zidmap_t *mnt_ns)
2542 {
2543 	if (*working_mode != ACE_WRITE_DATA)
2544 		return (SET_ERROR(EACCES));
2545 
2546 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2547 	    check_privs, B_FALSE, cr, mnt_ns));
2548 }
2549 
2550 int
2551 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2552 {
2553 	boolean_t owner = B_FALSE;
2554 	boolean_t groupmbr = B_FALSE;
2555 	boolean_t is_attr;
2556 	uid_t uid = crgetuid(cr);
2557 	int error;
2558 
2559 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2560 		return (SET_ERROR(EACCES));
2561 
2562 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2563 	    (S_ISDIR(ZTOI(zdp)->i_mode)));
2564 	if (is_attr)
2565 		goto slow;
2566 
2567 
2568 	mutex_enter(&zdp->z_acl_lock);
2569 
2570 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2571 		mutex_exit(&zdp->z_acl_lock);
2572 		return (0);
2573 	}
2574 
2575 	if (KUID_TO_SUID(ZTOI(zdp)->i_uid) != 0 ||
2576 	    KGID_TO_SGID(ZTOI(zdp)->i_gid) != 0) {
2577 		mutex_exit(&zdp->z_acl_lock);
2578 		goto slow;
2579 	}
2580 
2581 	if (uid == KUID_TO_SUID(ZTOI(zdp)->i_uid)) {
2582 		if (zdp->z_mode & S_IXUSR) {
2583 			mutex_exit(&zdp->z_acl_lock);
2584 			return (0);
2585 		} else {
2586 			mutex_exit(&zdp->z_acl_lock);
2587 			goto slow;
2588 		}
2589 	}
2590 	if (groupmember(KGID_TO_SGID(ZTOI(zdp)->i_gid), cr)) {
2591 		if (zdp->z_mode & S_IXGRP) {
2592 			mutex_exit(&zdp->z_acl_lock);
2593 			return (0);
2594 		} else {
2595 			mutex_exit(&zdp->z_acl_lock);
2596 			goto slow;
2597 		}
2598 	}
2599 	if (!owner && !groupmbr) {
2600 		if (zdp->z_mode & S_IXOTH) {
2601 			mutex_exit(&zdp->z_acl_lock);
2602 			return (0);
2603 		}
2604 	}
2605 
2606 	mutex_exit(&zdp->z_acl_lock);
2607 
2608 slow:
2609 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2610 	if ((error = zfs_enter(ZTOZSB(zdp), FTAG)) != 0)
2611 		return (error);
2612 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr,
2613 	    zfs_init_idmap);
2614 	zfs_exit(ZTOZSB(zdp), FTAG);
2615 	return (error);
2616 }
2617 
2618 /*
2619  * Determine whether Access should be granted/denied.
2620  *
2621  * The least priv subsystem is always consulted as a basic privilege
2622  * can define any form of access.
2623  */
2624 int
2625 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr,
2626     zidmap_t *mnt_ns)
2627 {
2628 	uint32_t	working_mode;
2629 	int		error;
2630 	int		is_attr;
2631 	boolean_t 	check_privs;
2632 	znode_t		*xzp;
2633 	znode_t 	*check_zp = zp;
2634 	mode_t		needed_bits;
2635 	uid_t		owner;
2636 
2637 	is_attr = ((zp->z_pflags & ZFS_XATTR) && S_ISDIR(ZTOI(zp)->i_mode));
2638 
2639 	/*
2640 	 * If attribute then validate against base file
2641 	 */
2642 	if (is_attr) {
2643 		if ((error = zfs_zget(ZTOZSB(zp),
2644 		    zp->z_xattr_parent, &xzp)) != 0) {
2645 			return (error);
2646 		}
2647 
2648 		check_zp = xzp;
2649 
2650 		/*
2651 		 * fixup mode to map to xattr perms
2652 		 */
2653 
2654 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2655 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2656 			mode |= ACE_WRITE_NAMED_ATTRS;
2657 		}
2658 
2659 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2660 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2661 			mode |= ACE_READ_NAMED_ATTRS;
2662 		}
2663 	}
2664 
2665 	owner = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2666 	    KUID_TO_SUID(ZTOI(zp)->i_uid));
2667 	owner = zfs_fuid_map_id(ZTOZSB(zp), owner, cr, ZFS_OWNER);
2668 
2669 	/*
2670 	 * Map the bits required to the standard inode flags
2671 	 * S_IRUSR|S_IWUSR|S_IXUSR in the needed_bits.  Map the bits
2672 	 * mapped by working_mode (currently missing) in missing_bits.
2673 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2674 	 * needed_bits.
2675 	 */
2676 	needed_bits = 0;
2677 
2678 	working_mode = mode;
2679 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2680 	    owner == crgetuid(cr))
2681 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2682 
2683 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2684 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2685 		needed_bits |= S_IRUSR;
2686 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2687 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2688 		needed_bits |= S_IWUSR;
2689 	if (working_mode & ACE_EXECUTE)
2690 		needed_bits |= S_IXUSR;
2691 
2692 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2693 	    &check_privs, skipaclchk, cr, mnt_ns)) == 0) {
2694 		if (is_attr)
2695 			zrele(xzp);
2696 		return (secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2697 		    needed_bits, needed_bits));
2698 	}
2699 
2700 	if (error && !check_privs) {
2701 		if (is_attr)
2702 			zrele(xzp);
2703 		return (error);
2704 	}
2705 
2706 	if (error && (flags & V_APPEND)) {
2707 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr,
2708 		    mnt_ns);
2709 	}
2710 
2711 	if (error && check_privs) {
2712 		mode_t		checkmode = 0;
2713 
2714 		/*
2715 		 * First check for implicit owner permission on
2716 		 * read_acl/read_attributes
2717 		 */
2718 
2719 		ASSERT(working_mode != 0);
2720 
2721 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2722 		    owner == crgetuid(cr)))
2723 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2724 
2725 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2726 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2727 			checkmode |= S_IRUSR;
2728 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2729 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2730 			checkmode |= S_IWUSR;
2731 		if (working_mode & ACE_EXECUTE)
2732 			checkmode |= S_IXUSR;
2733 
2734 		error = secpolicy_vnode_access2(cr, ZTOI(check_zp), owner,
2735 		    needed_bits & ~checkmode, needed_bits);
2736 
2737 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2738 			error = secpolicy_vnode_chown(cr, owner);
2739 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2740 			error = secpolicy_vnode_setdac(cr, owner);
2741 
2742 		if (error == 0 && (working_mode &
2743 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2744 			error = secpolicy_vnode_remove(cr);
2745 
2746 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2747 			error = secpolicy_vnode_chown(cr, owner);
2748 		}
2749 		if (error == 0) {
2750 			/*
2751 			 * See if any bits other than those already checked
2752 			 * for are still present.  If so then return EACCES
2753 			 */
2754 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2755 				error = SET_ERROR(EACCES);
2756 			}
2757 		}
2758 	} else if (error == 0) {
2759 		error = secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2760 		    needed_bits, needed_bits);
2761 	}
2762 
2763 	if (is_attr)
2764 		zrele(xzp);
2765 
2766 	return (error);
2767 }
2768 
2769 /*
2770  * Translate traditional unix S_IRUSR/S_IWUSR/S_IXUSR mode into
2771  * NFSv4-style ZFS ACL format and call zfs_zaccess()
2772  */
2773 int
2774 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr,
2775     zidmap_t *mnt_ns)
2776 {
2777 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr,
2778 	    mnt_ns));
2779 }
2780 
2781 /*
2782  * Access function for secpolicy_vnode_setattr
2783  */
2784 int
2785 zfs_zaccess_unix(void *zp, int mode, cred_t *cr)
2786 {
2787 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2788 
2789 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr, zfs_init_idmap));
2790 }
2791 
2792 /* See zfs_zaccess_delete() */
2793 static const boolean_t zfs_write_implies_delete_child = B_TRUE;
2794 
2795 /*
2796  * Determine whether delete access should be granted.
2797  *
2798  * The following chart outlines how we handle delete permissions which is
2799  * how recent versions of windows (Windows 2008) handles it.  The efficiency
2800  * comes from not having to check the parent ACL where the object itself grants
2801  * delete:
2802  *
2803  *      -------------------------------------------------------
2804  *      |   Parent Dir  |      Target Object Permissions      |
2805  *      |  permissions  |                                     |
2806  *      -------------------------------------------------------
2807  *      |               | ACL Allows | ACL Denies| Delete     |
2808  *      |               |  Delete    |  Delete   | unspecified|
2809  *      -------------------------------------------------------
2810  *      | ACL Allows    | Permit     | Deny *    | Permit     |
2811  *      | DELETE_CHILD  |            |           |            |
2812  *      -------------------------------------------------------
2813  *      | ACL Denies    | Permit     | Deny      | Deny       |
2814  *      | DELETE_CHILD  |            |           |            |
2815  *      -------------------------------------------------------
2816  *      | ACL specifies |            |           |            |
2817  *      | only allow    | Permit     | Deny *    | Permit     |
2818  *      | write and     |            |           |            |
2819  *      | execute       |            |           |            |
2820  *      -------------------------------------------------------
2821  *      | ACL denies    |            |           |            |
2822  *      | write and     | Permit     | Deny      | Deny       |
2823  *      | execute       |            |           |            |
2824  *      -------------------------------------------------------
2825  *         ^
2826  *         |
2827  *         Re. execute permission on the directory:  if that's missing,
2828  *	   the vnode lookup of the target will fail before we get here.
2829  *
2830  * Re [*] in the table above:  NFSv4 would normally Permit delete for
2831  * these two cells of the matrix.
2832  * See acl.h for notes on which ACE_... flags should be checked for which
2833  * operations.  Specifically, the NFSv4 committee recommendation is in
2834  * conflict with the Windows interpretation of DENY ACEs, where DENY ACEs
2835  * should take precedence ahead of ALLOW ACEs.
2836  *
2837  * This implementation always consults the target object's ACL first.
2838  * If a DENY ACE is present on the target object that specifies ACE_DELETE,
2839  * delete access is denied.  If an ALLOW ACE with ACE_DELETE is present on
2840  * the target object, access is allowed.  If and only if no entries with
2841  * ACE_DELETE are present in the object's ACL, check the container's ACL
2842  * for entries with ACE_DELETE_CHILD.
2843  *
2844  * A summary of the logic implemented from the table above is as follows:
2845  *
2846  * First check for DENY ACEs that apply.
2847  * If either target or container has a deny, EACCES.
2848  *
2849  * Delete access can then be summarized as follows:
2850  * 1: The object to be deleted grants ACE_DELETE, or
2851  * 2: The containing directory grants ACE_DELETE_CHILD.
2852  * In a Windows system, that would be the end of the story.
2853  * In this system, (2) has some complications...
2854  * 2a: "sticky" bit on a directory adds restrictions, and
2855  * 2b: existing ACEs from previous versions of ZFS may
2856  * not carry ACE_DELETE_CHILD where they should, so we
2857  * also allow delete when ACE_WRITE_DATA is granted.
2858  *
2859  * Note: 2b is technically a work-around for a prior bug,
2860  * which hopefully can go away some day.  For those who
2861  * no longer need the work around, and for testing, this
2862  * work-around is made conditional via the tunable:
2863  * zfs_write_implies_delete_child
2864  */
2865 int
2866 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr, zidmap_t *mnt_ns)
2867 {
2868 	uint32_t wanted_dirperms;
2869 	uint32_t dzp_working_mode = 0;
2870 	uint32_t zp_working_mode = 0;
2871 	int dzp_error, zp_error;
2872 	boolean_t dzpcheck_privs;
2873 	boolean_t zpcheck_privs;
2874 
2875 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2876 		return (SET_ERROR(EPERM));
2877 
2878 	/*
2879 	 * Case 1:
2880 	 * If target object grants ACE_DELETE then we are done.  This is
2881 	 * indicated by a return value of 0.  For this case we don't worry
2882 	 * about the sticky bit because sticky only applies to the parent
2883 	 * directory and this is the child access result.
2884 	 *
2885 	 * If we encounter a DENY ACE here, we're also done (EACCES).
2886 	 * Note that if we hit a DENY ACE here (on the target) it should
2887 	 * take precedence over a DENY ACE on the container, so that when
2888 	 * we have more complete auditing support we will be able to
2889 	 * report an access failure against the specific target.
2890 	 * (This is part of why we're checking the target first.)
2891 	 */
2892 	zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2893 	    &zpcheck_privs, B_FALSE, cr, mnt_ns);
2894 	if (zp_error == EACCES) {
2895 		/* We hit a DENY ACE. */
2896 		if (!zpcheck_privs)
2897 			return (SET_ERROR(zp_error));
2898 		return (secpolicy_vnode_remove(cr));
2899 
2900 	}
2901 	if (zp_error == 0)
2902 		return (0);
2903 
2904 	/*
2905 	 * Case 2:
2906 	 * If the containing directory grants ACE_DELETE_CHILD,
2907 	 * or we're in backward compatibility mode and the
2908 	 * containing directory has ACE_WRITE_DATA, allow.
2909 	 * Case 2b is handled with wanted_dirperms.
2910 	 */
2911 	wanted_dirperms = ACE_DELETE_CHILD;
2912 	if (zfs_write_implies_delete_child)
2913 		wanted_dirperms |= ACE_WRITE_DATA;
2914 	dzp_error = zfs_zaccess_common(dzp, wanted_dirperms,
2915 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr, mnt_ns);
2916 	if (dzp_error == EACCES) {
2917 		/* We hit a DENY ACE. */
2918 		if (!dzpcheck_privs)
2919 			return (SET_ERROR(dzp_error));
2920 		return (secpolicy_vnode_remove(cr));
2921 	}
2922 
2923 	/*
2924 	 * Cases 2a, 2b (continued)
2925 	 *
2926 	 * Note: dzp_working_mode now contains any permissions
2927 	 * that were NOT granted.  Therefore, if any of the
2928 	 * wanted_dirperms WERE granted, we will have:
2929 	 *   dzp_working_mode != wanted_dirperms
2930 	 * We're really asking if ANY of those permissions
2931 	 * were granted, and if so, grant delete access.
2932 	 */
2933 	if (dzp_working_mode != wanted_dirperms)
2934 		dzp_error = 0;
2935 
2936 	/*
2937 	 * dzp_error is 0 if the container granted us permissions to "modify".
2938 	 * If we do not have permission via one or more ACEs, our current
2939 	 * privileges may still permit us to modify the container.
2940 	 *
2941 	 * dzpcheck_privs is false when i.e. the FS is read-only.
2942 	 * Otherwise, do privilege checks for the container.
2943 	 */
2944 	if (dzp_error != 0 && dzpcheck_privs) {
2945 		uid_t owner;
2946 
2947 		/*
2948 		 * The secpolicy call needs the requested access and
2949 		 * the current access mode of the container, but it
2950 		 * only knows about Unix-style modes (VEXEC, VWRITE),
2951 		 * so this must condense the fine-grained ACE bits into
2952 		 * Unix modes.
2953 		 *
2954 		 * The VEXEC flag is easy, because we know that has
2955 		 * always been checked before we get here (during the
2956 		 * lookup of the target vnode).  The container has not
2957 		 * granted us permissions to "modify", so we do not set
2958 		 * the VWRITE flag in the current access mode.
2959 		 */
2960 		owner = zfs_fuid_map_id(ZTOZSB(dzp),
2961 		    KUID_TO_SUID(ZTOI(dzp)->i_uid), cr, ZFS_OWNER);
2962 		dzp_error = secpolicy_vnode_access2(cr, ZTOI(dzp),
2963 		    owner, S_IXUSR, S_IWUSR|S_IXUSR);
2964 	}
2965 	if (dzp_error != 0) {
2966 		/*
2967 		 * Note: We may have dzp_error = -1 here (from
2968 		 * zfs_zacess_common).  Don't return that.
2969 		 */
2970 		return (SET_ERROR(EACCES));
2971 	}
2972 
2973 
2974 	/*
2975 	 * At this point, we know that the directory permissions allow
2976 	 * us to modify, but we still need to check for the additional
2977 	 * restrictions that apply when the "sticky bit" is set.
2978 	 *
2979 	 * Yes, zfs_sticky_remove_access() also checks this bit, but
2980 	 * checking it here and skipping the call below is nice when
2981 	 * you're watching all of this with dtrace.
2982 	 */
2983 	if ((dzp->z_mode & S_ISVTX) == 0)
2984 		return (0);
2985 
2986 	/*
2987 	 * zfs_sticky_remove_access will succeed if:
2988 	 * 1. The sticky bit is absent.
2989 	 * 2. We pass the sticky bit restrictions.
2990 	 * 3. We have privileges that always allow file removal.
2991 	 */
2992 	return (zfs_sticky_remove_access(dzp, zp, cr));
2993 }
2994 
2995 int
2996 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2997     znode_t *tzp, cred_t *cr, zidmap_t *mnt_ns)
2998 {
2999 	int add_perm;
3000 	int error;
3001 
3002 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
3003 		return (SET_ERROR(EACCES));
3004 
3005 	add_perm = S_ISDIR(ZTOI(szp)->i_mode) ?
3006 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
3007 
3008 	/*
3009 	 * Rename permissions are combination of delete permission +
3010 	 * add file/subdir permission.
3011 	 */
3012 
3013 	/*
3014 	 * first make sure we do the delete portion.
3015 	 *
3016 	 * If that succeeds then check for add_file/add_subdir permissions
3017 	 */
3018 
3019 	if ((error = zfs_zaccess_delete(sdzp, szp, cr, mnt_ns)))
3020 		return (error);
3021 
3022 	/*
3023 	 * If we have a tzp, see if we can delete it?
3024 	 */
3025 	if (tzp) {
3026 		if ((error = zfs_zaccess_delete(tdzp, tzp, cr, mnt_ns)))
3027 			return (error);
3028 	}
3029 
3030 	/*
3031 	 * Now check for add permissions
3032 	 */
3033 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr, mnt_ns);
3034 
3035 	return (error);
3036 }
3037