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