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_head(&aclp->z_acl))) {
497 		list_remove(&aclp->z_acl, aclnode);
498 		zfs_acl_node_free(aclnode);
499 	}
500 	aclp->z_acl_count = 0;
501 	aclp->z_acl_bytes = 0;
502 }
503 
504 void
505 zfs_acl_free(zfs_acl_t *aclp)
506 {
507 	zfs_acl_release_nodes(aclp);
508 	list_destroy(&aclp->z_acl);
509 	kmem_free(aclp, sizeof (zfs_acl_t));
510 }
511 
512 static boolean_t
513 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
514 {
515 	uint16_t entry_type;
516 
517 	switch (type) {
518 	case ALLOW:
519 	case DENY:
520 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
521 	case ACE_SYSTEM_ALARM_ACE_TYPE:
522 		entry_type = flags & ACE_TYPE_FLAGS;
523 		return (entry_type == ACE_OWNER ||
524 		    entry_type == OWNING_GROUP ||
525 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
526 		    entry_type == ACE_IDENTIFIER_GROUP);
527 	default:
528 		if (type <= MAX_ACE_TYPE)
529 			return (B_TRUE);
530 	}
531 	return (B_FALSE);
532 }
533 
534 static boolean_t
535 zfs_ace_valid(umode_t obj_mode, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
536 {
537 	/*
538 	 * first check type of entry
539 	 */
540 
541 	if (!zfs_acl_valid_ace_type(type, iflags))
542 		return (B_FALSE);
543 
544 	switch (type) {
545 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
546 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
547 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
548 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
549 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
550 			return (B_FALSE);
551 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
552 	}
553 
554 	/*
555 	 * next check inheritance level flags
556 	 */
557 
558 	if (S_ISDIR(obj_mode) &&
559 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
560 		aclp->z_hints |= ZFS_INHERIT_ACE;
561 
562 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
563 		if ((iflags & (ACE_FILE_INHERIT_ACE|
564 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
565 			return (B_FALSE);
566 		}
567 	}
568 
569 	return (B_TRUE);
570 }
571 
572 static void *
573 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
574     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
575 {
576 	zfs_acl_node_t *aclnode;
577 
578 	ASSERT(aclp);
579 
580 	if (start == NULL) {
581 		aclnode = list_head(&aclp->z_acl);
582 		if (aclnode == NULL)
583 			return (NULL);
584 
585 		aclp->z_next_ace = aclnode->z_acldata;
586 		aclp->z_curr_node = aclnode;
587 		aclnode->z_ace_idx = 0;
588 	}
589 
590 	aclnode = aclp->z_curr_node;
591 
592 	if (aclnode == NULL)
593 		return (NULL);
594 
595 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
596 		aclnode = list_next(&aclp->z_acl, aclnode);
597 		if (aclnode == NULL)
598 			return (NULL);
599 		else {
600 			aclp->z_curr_node = aclnode;
601 			aclnode->z_ace_idx = 0;
602 			aclp->z_next_ace = aclnode->z_acldata;
603 		}
604 	}
605 
606 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
607 		void *acep = aclp->z_next_ace;
608 		size_t ace_size;
609 
610 		/*
611 		 * Make sure we don't overstep our bounds
612 		 */
613 		ace_size = aclp->z_ops->ace_size(acep);
614 
615 		if (((caddr_t)acep + ace_size) >
616 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
617 			return (NULL);
618 		}
619 
620 		*iflags = aclp->z_ops->ace_flags_get(acep);
621 		*type = aclp->z_ops->ace_type_get(acep);
622 		*access_mask = aclp->z_ops->ace_mask_get(acep);
623 		*who = aclp->z_ops->ace_who_get(acep);
624 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
625 		aclnode->z_ace_idx++;
626 
627 		return ((void *)acep);
628 	}
629 	return (NULL);
630 }
631 
632 static uintptr_t
633 zfs_ace_walk(void *datap, uintptr_t cookie, int aclcnt,
634     uint16_t *flags, uint16_t *type, uint32_t *mask)
635 {
636 	(void) aclcnt;
637 	zfs_acl_t *aclp = datap;
638 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)cookie;
639 	uint64_t who;
640 
641 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
642 	    flags, type);
643 	return ((uintptr_t)acep);
644 }
645 
646 /*
647  * Copy ACE to internal ZFS format.
648  * While processing the ACL each ACE will be validated for correctness.
649  * ACE FUIDs will be created later.
650  */
651 static int
652 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, umode_t obj_mode, zfs_acl_t *aclp,
653     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
654     zfs_fuid_info_t **fuidp, cred_t *cr)
655 {
656 	int i;
657 	uint16_t entry_type;
658 	zfs_ace_t *aceptr = z_acl;
659 	ace_t *acep = datap;
660 	zfs_object_ace_t *zobjacep;
661 	ace_object_t *aceobjp;
662 
663 	for (i = 0; i != aclcnt; i++) {
664 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
665 		aceptr->z_hdr.z_flags = acep->a_flags;
666 		aceptr->z_hdr.z_type = acep->a_type;
667 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
668 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
669 		    entry_type != ACE_EVERYONE) {
670 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
671 			    cr, (entry_type == 0) ?
672 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
673 		}
674 
675 		/*
676 		 * Make sure ACE is valid
677 		 */
678 		if (zfs_ace_valid(obj_mode, aclp, aceptr->z_hdr.z_type,
679 		    aceptr->z_hdr.z_flags) != B_TRUE)
680 			return (SET_ERROR(EINVAL));
681 
682 		switch (acep->a_type) {
683 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
684 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
685 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
686 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
687 			zobjacep = (zfs_object_ace_t *)aceptr;
688 			aceobjp = (ace_object_t *)acep;
689 
690 			memcpy(zobjacep->z_object_type, aceobjp->a_obj_type,
691 			    sizeof (aceobjp->a_obj_type));
692 			memcpy(zobjacep->z_inherit_type,
693 			    aceobjp->a_inherit_obj_type,
694 			    sizeof (aceobjp->a_inherit_obj_type));
695 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
696 			break;
697 		default:
698 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
699 		}
700 
701 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
702 		    aclp->z_ops->ace_size(aceptr));
703 	}
704 
705 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
706 
707 	return (0);
708 }
709 
710 /*
711  * Copy ZFS ACEs to fixed size ace_t layout
712  */
713 static void
714 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
715     void *datap, int filter)
716 {
717 	uint64_t who;
718 	uint32_t access_mask;
719 	uint16_t iflags, type;
720 	zfs_ace_hdr_t *zacep = NULL;
721 	ace_t *acep = datap;
722 	ace_object_t *objacep;
723 	zfs_object_ace_t *zobjacep;
724 	size_t ace_size;
725 	uint16_t entry_type;
726 
727 	while ((zacep = zfs_acl_next_ace(aclp, zacep,
728 	    &who, &access_mask, &iflags, &type))) {
729 
730 		switch (type) {
731 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
732 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
733 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
734 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
735 			if (filter) {
736 				continue;
737 			}
738 			zobjacep = (zfs_object_ace_t *)zacep;
739 			objacep = (ace_object_t *)acep;
740 			memcpy(objacep->a_obj_type,
741 			    zobjacep->z_object_type,
742 			    sizeof (zobjacep->z_object_type));
743 			memcpy(objacep->a_inherit_obj_type,
744 			    zobjacep->z_inherit_type,
745 			    sizeof (zobjacep->z_inherit_type));
746 			ace_size = sizeof (ace_object_t);
747 			break;
748 		default:
749 			ace_size = sizeof (ace_t);
750 			break;
751 		}
752 
753 		entry_type = (iflags & ACE_TYPE_FLAGS);
754 		if ((entry_type != ACE_OWNER &&
755 		    entry_type != OWNING_GROUP &&
756 		    entry_type != ACE_EVERYONE)) {
757 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
758 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
759 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
760 		} else {
761 			acep->a_who = (uid_t)(int64_t)who;
762 		}
763 		acep->a_access_mask = access_mask;
764 		acep->a_flags = iflags;
765 		acep->a_type = type;
766 		acep = (ace_t *)((caddr_t)acep + ace_size);
767 	}
768 }
769 
770 static int
771 zfs_copy_ace_2_oldace(umode_t obj_mode, zfs_acl_t *aclp, ace_t *acep,
772     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
773 {
774 	int i;
775 	zfs_oldace_t *aceptr = z_acl;
776 
777 	for (i = 0; i != aclcnt; i++, aceptr++) {
778 		aceptr->z_access_mask = acep[i].a_access_mask;
779 		aceptr->z_type = acep[i].a_type;
780 		aceptr->z_flags = acep[i].a_flags;
781 		aceptr->z_fuid = acep[i].a_who;
782 		/*
783 		 * Make sure ACE is valid
784 		 */
785 		if (zfs_ace_valid(obj_mode, aclp, aceptr->z_type,
786 		    aceptr->z_flags) != B_TRUE)
787 			return (SET_ERROR(EINVAL));
788 	}
789 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
790 	return (0);
791 }
792 
793 /*
794  * convert old ACL format to new
795  */
796 void
797 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
798 {
799 	zfs_oldace_t *oldaclp;
800 	int i;
801 	uint16_t type, iflags;
802 	uint32_t access_mask;
803 	uint64_t who;
804 	void *cookie = NULL;
805 	zfs_acl_node_t *newaclnode;
806 
807 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
808 	/*
809 	 * First create the ACE in a contiguous piece of memory
810 	 * for zfs_copy_ace_2_fuid().
811 	 *
812 	 * We only convert an ACL once, so this won't happen
813 	 * every time.
814 	 */
815 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
816 	    KM_SLEEP);
817 	i = 0;
818 	while ((cookie = zfs_acl_next_ace(aclp, cookie, &who,
819 	    &access_mask, &iflags, &type))) {
820 		oldaclp[i].z_flags = iflags;
821 		oldaclp[i].z_type = type;
822 		oldaclp[i].z_fuid = who;
823 		oldaclp[i++].z_access_mask = access_mask;
824 	}
825 
826 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
827 	    sizeof (zfs_object_ace_t));
828 	aclp->z_ops = &zfs_acl_fuid_ops;
829 	VERIFY(zfs_copy_ace_2_fuid(ZTOZSB(zp), ZTOI(zp)->i_mode,
830 	    aclp, oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
831 	    &newaclnode->z_size, NULL, cr) == 0);
832 	newaclnode->z_ace_count = aclp->z_acl_count;
833 	aclp->z_version = ZFS_ACL_VERSION;
834 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
835 
836 	/*
837 	 * Release all previous ACL nodes
838 	 */
839 
840 	zfs_acl_release_nodes(aclp);
841 
842 	list_insert_head(&aclp->z_acl, newaclnode);
843 
844 	aclp->z_acl_bytes = newaclnode->z_size;
845 	aclp->z_acl_count = newaclnode->z_ace_count;
846 
847 }
848 
849 /*
850  * Convert unix access mask to v4 access mask
851  */
852 static uint32_t
853 zfs_unix_to_v4(uint32_t access_mask)
854 {
855 	uint32_t new_mask = 0;
856 
857 	if (access_mask & S_IXOTH)
858 		new_mask |= ACE_EXECUTE;
859 	if (access_mask & S_IWOTH)
860 		new_mask |= ACE_WRITE_DATA;
861 	if (access_mask & S_IROTH)
862 		new_mask |= ACE_READ_DATA;
863 	return (new_mask);
864 }
865 
866 
867 static int
868 zfs_v4_to_unix(uint32_t access_mask, int *unmapped)
869 {
870 	int new_mask = 0;
871 
872 	*unmapped = access_mask &
873 	    (ACE_WRITE_OWNER | ACE_WRITE_ACL | ACE_DELETE);
874 
875 	if (access_mask & WRITE_MASK)
876 		new_mask |= S_IWOTH;
877 	if (access_mask & ACE_READ_DATA)
878 		new_mask |= S_IROTH;
879 	if (access_mask & ACE_EXECUTE)
880 		new_mask |= S_IXOTH;
881 
882 	return (new_mask);
883 }
884 
885 
886 static void
887 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
888     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
889 {
890 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
891 
892 	aclp->z_ops->ace_mask_set(acep, access_mask);
893 	aclp->z_ops->ace_type_set(acep, access_type);
894 	aclp->z_ops->ace_flags_set(acep, entry_type);
895 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
896 	    type != ACE_EVERYONE))
897 		aclp->z_ops->ace_who_set(acep, fuid);
898 }
899 
900 /*
901  * Determine mode of file based on ACL.
902  */
903 uint64_t
904 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
905     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
906 {
907 	int		entry_type;
908 	mode_t		mode;
909 	mode_t		seen = 0;
910 	zfs_ace_hdr_t 	*acep = NULL;
911 	uint64_t	who;
912 	uint16_t	iflags, type;
913 	uint32_t	access_mask;
914 	boolean_t	an_exec_denied = B_FALSE;
915 
916 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
917 
918 	while ((acep = zfs_acl_next_ace(aclp, acep, &who,
919 	    &access_mask, &iflags, &type))) {
920 
921 		if (!zfs_acl_valid_ace_type(type, iflags))
922 			continue;
923 
924 		entry_type = (iflags & ACE_TYPE_FLAGS);
925 
926 		/*
927 		 * Skip over any inherit_only ACEs
928 		 */
929 		if (iflags & ACE_INHERIT_ONLY_ACE)
930 			continue;
931 
932 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
933 		    who == fuid)) {
934 			if ((access_mask & ACE_READ_DATA) &&
935 			    (!(seen & S_IRUSR))) {
936 				seen |= S_IRUSR;
937 				if (type == ALLOW) {
938 					mode |= S_IRUSR;
939 				}
940 			}
941 			if ((access_mask & ACE_WRITE_DATA) &&
942 			    (!(seen & S_IWUSR))) {
943 				seen |= S_IWUSR;
944 				if (type == ALLOW) {
945 					mode |= S_IWUSR;
946 				}
947 			}
948 			if ((access_mask & ACE_EXECUTE) &&
949 			    (!(seen & S_IXUSR))) {
950 				seen |= S_IXUSR;
951 				if (type == ALLOW) {
952 					mode |= S_IXUSR;
953 				}
954 			}
955 		} else if (entry_type == OWNING_GROUP ||
956 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
957 			if ((access_mask & ACE_READ_DATA) &&
958 			    (!(seen & S_IRGRP))) {
959 				seen |= S_IRGRP;
960 				if (type == ALLOW) {
961 					mode |= S_IRGRP;
962 				}
963 			}
964 			if ((access_mask & ACE_WRITE_DATA) &&
965 			    (!(seen & S_IWGRP))) {
966 				seen |= S_IWGRP;
967 				if (type == ALLOW) {
968 					mode |= S_IWGRP;
969 				}
970 			}
971 			if ((access_mask & ACE_EXECUTE) &&
972 			    (!(seen & S_IXGRP))) {
973 				seen |= S_IXGRP;
974 				if (type == ALLOW) {
975 					mode |= S_IXGRP;
976 				}
977 			}
978 		} else if (entry_type == ACE_EVERYONE) {
979 			if ((access_mask & ACE_READ_DATA)) {
980 				if (!(seen & S_IRUSR)) {
981 					seen |= S_IRUSR;
982 					if (type == ALLOW) {
983 						mode |= S_IRUSR;
984 					}
985 				}
986 				if (!(seen & S_IRGRP)) {
987 					seen |= S_IRGRP;
988 					if (type == ALLOW) {
989 						mode |= S_IRGRP;
990 					}
991 				}
992 				if (!(seen & S_IROTH)) {
993 					seen |= S_IROTH;
994 					if (type == ALLOW) {
995 						mode |= S_IROTH;
996 					}
997 				}
998 			}
999 			if ((access_mask & ACE_WRITE_DATA)) {
1000 				if (!(seen & S_IWUSR)) {
1001 					seen |= S_IWUSR;
1002 					if (type == ALLOW) {
1003 						mode |= S_IWUSR;
1004 					}
1005 				}
1006 				if (!(seen & S_IWGRP)) {
1007 					seen |= S_IWGRP;
1008 					if (type == ALLOW) {
1009 						mode |= S_IWGRP;
1010 					}
1011 				}
1012 				if (!(seen & S_IWOTH)) {
1013 					seen |= S_IWOTH;
1014 					if (type == ALLOW) {
1015 						mode |= S_IWOTH;
1016 					}
1017 				}
1018 			}
1019 			if ((access_mask & ACE_EXECUTE)) {
1020 				if (!(seen & S_IXUSR)) {
1021 					seen |= S_IXUSR;
1022 					if (type == ALLOW) {
1023 						mode |= S_IXUSR;
1024 					}
1025 				}
1026 				if (!(seen & S_IXGRP)) {
1027 					seen |= S_IXGRP;
1028 					if (type == ALLOW) {
1029 						mode |= S_IXGRP;
1030 					}
1031 				}
1032 				if (!(seen & S_IXOTH)) {
1033 					seen |= S_IXOTH;
1034 					if (type == ALLOW) {
1035 						mode |= S_IXOTH;
1036 					}
1037 				}
1038 			}
1039 		} else {
1040 			/*
1041 			 * Only care if this IDENTIFIER_GROUP or
1042 			 * USER ACE denies execute access to someone,
1043 			 * mode is not affected
1044 			 */
1045 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1046 				an_exec_denied = B_TRUE;
1047 		}
1048 	}
1049 
1050 	/*
1051 	 * Failure to allow is effectively a deny, so execute permission
1052 	 * is denied if it was never mentioned or if we explicitly
1053 	 * weren't allowed it.
1054 	 */
1055 	if (!an_exec_denied &&
1056 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1057 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1058 		an_exec_denied = B_TRUE;
1059 
1060 	if (an_exec_denied)
1061 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1062 	else
1063 		*pflags |= ZFS_NO_EXECS_DENIED;
1064 
1065 	return (mode);
1066 }
1067 
1068 /*
1069  * Read an external acl object.  If the intent is to modify, always
1070  * create a new acl and leave any cached acl in place.
1071  */
1072 int
1073 zfs_acl_node_read(struct znode *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1074     boolean_t will_modify)
1075 {
1076 	zfs_acl_t	*aclp;
1077 	int		aclsize = 0;
1078 	int		acl_count = 0;
1079 	zfs_acl_node_t	*aclnode;
1080 	zfs_acl_phys_t	znode_acl;
1081 	int		version;
1082 	int		error;
1083 	boolean_t	drop_lock = B_FALSE;
1084 
1085 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1086 
1087 	if (zp->z_acl_cached && !will_modify) {
1088 		*aclpp = zp->z_acl_cached;
1089 		return (0);
1090 	}
1091 
1092 	/*
1093 	 * close race where znode could be upgrade while trying to
1094 	 * read the znode attributes.
1095 	 *
1096 	 * But this could only happen if the file isn't already an SA
1097 	 * znode
1098 	 */
1099 	if (!zp->z_is_sa && !have_lock) {
1100 		mutex_enter(&zp->z_lock);
1101 		drop_lock = B_TRUE;
1102 	}
1103 	version = zfs_znode_acl_version(zp);
1104 
1105 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1106 	    &acl_count, &znode_acl)) != 0) {
1107 		goto done;
1108 	}
1109 
1110 	aclp = zfs_acl_alloc(version);
1111 
1112 	aclp->z_acl_count = acl_count;
1113 	aclp->z_acl_bytes = aclsize;
1114 
1115 	aclnode = zfs_acl_node_alloc(aclsize);
1116 	aclnode->z_ace_count = aclp->z_acl_count;
1117 	aclnode->z_size = aclsize;
1118 
1119 	if (!zp->z_is_sa) {
1120 		if (znode_acl.z_acl_extern_obj) {
1121 			error = dmu_read(ZTOZSB(zp)->z_os,
1122 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1123 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1124 		} else {
1125 			memcpy(aclnode->z_acldata, znode_acl.z_ace_data,
1126 			    aclnode->z_size);
1127 		}
1128 	} else {
1129 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(ZTOZSB(zp)),
1130 		    aclnode->z_acldata, aclnode->z_size);
1131 	}
1132 
1133 	if (error != 0) {
1134 		zfs_acl_free(aclp);
1135 		zfs_acl_node_free(aclnode);
1136 		/* convert checksum errors into IO errors */
1137 		if (error == ECKSUM)
1138 			error = SET_ERROR(EIO);
1139 		goto done;
1140 	}
1141 
1142 	list_insert_head(&aclp->z_acl, aclnode);
1143 
1144 	*aclpp = aclp;
1145 	if (!will_modify)
1146 		zp->z_acl_cached = aclp;
1147 done:
1148 	if (drop_lock)
1149 		mutex_exit(&zp->z_lock);
1150 	return (error);
1151 }
1152 
1153 void
1154 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1155     boolean_t start, void *userdata)
1156 {
1157 	(void) buflen;
1158 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1159 
1160 	if (start) {
1161 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1162 	} else {
1163 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1164 		    cb->cb_acl_node);
1165 	}
1166 	ASSERT3P(cb->cb_acl_node, !=, NULL);
1167 	*dataptr = cb->cb_acl_node->z_acldata;
1168 	*length = cb->cb_acl_node->z_size;
1169 }
1170 
1171 int
1172 zfs_acl_chown_setattr(znode_t *zp)
1173 {
1174 	int error;
1175 	zfs_acl_t *aclp;
1176 
1177 	if (ZTOZSB(zp)->z_acl_type == ZFS_ACLTYPE_POSIX)
1178 		return (0);
1179 
1180 	ASSERT(MUTEX_HELD(&zp->z_lock));
1181 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1182 
1183 	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1184 	if (error == 0 && aclp->z_acl_count > 0)
1185 		zp->z_mode = ZTOI(zp)->i_mode =
1186 		    zfs_mode_compute(zp->z_mode, aclp,
1187 		    &zp->z_pflags, KUID_TO_SUID(ZTOI(zp)->i_uid),
1188 		    KGID_TO_SGID(ZTOI(zp)->i_gid));
1189 
1190 	/*
1191 	 * Some ZFS implementations (ZEVO) create neither a ZNODE_ACL
1192 	 * nor a DACL_ACES SA in which case ENOENT is returned from
1193 	 * zfs_acl_node_read() when the SA can't be located.
1194 	 * Allow chown/chgrp to succeed in these cases rather than
1195 	 * returning an error that makes no sense in the context of
1196 	 * the caller.
1197 	 */
1198 	if (error == ENOENT)
1199 		return (0);
1200 
1201 	return (error);
1202 }
1203 
1204 typedef struct trivial_acl {
1205 	uint32_t	allow0;		/* allow mask for bits only in owner */
1206 	uint32_t	deny1;		/* deny mask for bits not in owner */
1207 	uint32_t	deny2;		/* deny mask for bits not in group */
1208 	uint32_t	owner;		/* allow mask matching mode */
1209 	uint32_t	group;		/* allow mask matching mode */
1210 	uint32_t	everyone;	/* allow mask matching mode */
1211 } trivial_acl_t;
1212 
1213 static void
1214 acl_trivial_access_masks(mode_t mode, boolean_t isdir, trivial_acl_t *masks)
1215 {
1216 	uint32_t read_mask = ACE_READ_DATA;
1217 	uint32_t write_mask = ACE_WRITE_DATA|ACE_APPEND_DATA;
1218 	uint32_t execute_mask = ACE_EXECUTE;
1219 
1220 	if (isdir)
1221 		write_mask |= ACE_DELETE_CHILD;
1222 
1223 	masks->deny1 = 0;
1224 
1225 	if (!(mode & S_IRUSR) && (mode & (S_IRGRP|S_IROTH)))
1226 		masks->deny1 |= read_mask;
1227 	if (!(mode & S_IWUSR) && (mode & (S_IWGRP|S_IWOTH)))
1228 		masks->deny1 |= write_mask;
1229 	if (!(mode & S_IXUSR) && (mode & (S_IXGRP|S_IXOTH)))
1230 		masks->deny1 |= execute_mask;
1231 
1232 	masks->deny2 = 0;
1233 	if (!(mode & S_IRGRP) && (mode & S_IROTH))
1234 		masks->deny2 |= read_mask;
1235 	if (!(mode & S_IWGRP) && (mode & S_IWOTH))
1236 		masks->deny2 |= write_mask;
1237 	if (!(mode & S_IXGRP) && (mode & S_IXOTH))
1238 		masks->deny2 |= execute_mask;
1239 
1240 	masks->allow0 = 0;
1241 	if ((mode & S_IRUSR) && (!(mode & S_IRGRP) && (mode & S_IROTH)))
1242 		masks->allow0 |= read_mask;
1243 	if ((mode & S_IWUSR) && (!(mode & S_IWGRP) && (mode & S_IWOTH)))
1244 		masks->allow0 |= write_mask;
1245 	if ((mode & S_IXUSR) && (!(mode & S_IXGRP) && (mode & S_IXOTH)))
1246 		masks->allow0 |= execute_mask;
1247 
1248 	masks->owner = ACE_WRITE_ATTRIBUTES|ACE_WRITE_OWNER|ACE_WRITE_ACL|
1249 	    ACE_WRITE_NAMED_ATTRS|ACE_READ_ACL|ACE_READ_ATTRIBUTES|
1250 	    ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE;
1251 	if (mode & S_IRUSR)
1252 		masks->owner |= read_mask;
1253 	if (mode & S_IWUSR)
1254 		masks->owner |= write_mask;
1255 	if (mode & S_IXUSR)
1256 		masks->owner |= execute_mask;
1257 
1258 	masks->group = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1259 	    ACE_SYNCHRONIZE;
1260 	if (mode & S_IRGRP)
1261 		masks->group |= read_mask;
1262 	if (mode & S_IWGRP)
1263 		masks->group |= write_mask;
1264 	if (mode & S_IXGRP)
1265 		masks->group |= execute_mask;
1266 
1267 	masks->everyone = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1268 	    ACE_SYNCHRONIZE;
1269 	if (mode & S_IROTH)
1270 		masks->everyone |= read_mask;
1271 	if (mode & S_IWOTH)
1272 		masks->everyone |= write_mask;
1273 	if (mode & S_IXOTH)
1274 		masks->everyone |= execute_mask;
1275 }
1276 
1277 /*
1278  * ace_trivial:
1279  * determine whether an ace_t acl is trivial
1280  *
1281  * Trivialness implies that the acl is composed of only
1282  * owner, group, everyone entries.  ACL can't
1283  * have read_acl denied, and write_owner/write_acl/write_attributes
1284  * can only be owner@ entry.
1285  */
1286 static int
1287 ace_trivial_common(void *acep, int aclcnt,
1288     uintptr_t (*walk)(void *, uintptr_t, int,
1289     uint16_t *, uint16_t *, uint32_t *))
1290 {
1291 	uint16_t flags;
1292 	uint32_t mask;
1293 	uint16_t type;
1294 	uint64_t cookie = 0;
1295 
1296 	while ((cookie = walk(acep, cookie, aclcnt, &flags, &type, &mask))) {
1297 		switch (flags & ACE_TYPE_FLAGS) {
1298 		case ACE_OWNER:
1299 		case ACE_GROUP|ACE_IDENTIFIER_GROUP:
1300 		case ACE_EVERYONE:
1301 			break;
1302 		default:
1303 			return (1);
1304 		}
1305 
1306 		if (flags & (ACE_FILE_INHERIT_ACE|
1307 		    ACE_DIRECTORY_INHERIT_ACE|ACE_NO_PROPAGATE_INHERIT_ACE|
1308 		    ACE_INHERIT_ONLY_ACE))
1309 			return (1);
1310 
1311 		/*
1312 		 * Special check for some special bits
1313 		 *
1314 		 * Don't allow anybody to deny reading basic
1315 		 * attributes or a files ACL.
1316 		 */
1317 		if ((mask & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
1318 		    (type == ACE_ACCESS_DENIED_ACE_TYPE))
1319 			return (1);
1320 
1321 		/*
1322 		 * Delete permission is never set by default
1323 		 */
1324 		if (mask & ACE_DELETE)
1325 			return (1);
1326 
1327 		/*
1328 		 * Child delete permission should be accompanied by write
1329 		 */
1330 		if ((mask & ACE_DELETE_CHILD) && !(mask & ACE_WRITE_DATA))
1331 			return (1);
1332 
1333 		/*
1334 		 * only allow owner@ to have
1335 		 * write_acl/write_owner/write_attributes/write_xattr/
1336 		 */
1337 		if (type == ACE_ACCESS_ALLOWED_ACE_TYPE &&
1338 		    (!(flags & ACE_OWNER) && (mask &
1339 		    (ACE_WRITE_OWNER|ACE_WRITE_ACL| ACE_WRITE_ATTRIBUTES|
1340 		    ACE_WRITE_NAMED_ATTRS))))
1341 			return (1);
1342 
1343 	}
1344 
1345 	return (0);
1346 }
1347 
1348 /*
1349  * common code for setting ACLs.
1350  *
1351  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1352  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1353  * already checked the acl and knows whether to inherit.
1354  */
1355 int
1356 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1357 {
1358 	int			error;
1359 	zfsvfs_t		*zfsvfs = ZTOZSB(zp);
1360 	dmu_object_type_t	otype;
1361 	zfs_acl_locator_cb_t	locate = { 0 };
1362 	uint64_t		mode;
1363 	sa_bulk_attr_t		bulk[5];
1364 	uint64_t		ctime[2];
1365 	int			count = 0;
1366 	zfs_acl_phys_t		acl_phys;
1367 
1368 	mode = zp->z_mode;
1369 
1370 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1371 	    KUID_TO_SUID(ZTOI(zp)->i_uid), KGID_TO_SGID(ZTOI(zp)->i_gid));
1372 
1373 	zp->z_mode = ZTOI(zp)->i_mode = mode;
1374 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1375 	    &mode, sizeof (mode));
1376 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1377 	    &zp->z_pflags, sizeof (zp->z_pflags));
1378 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1379 	    &ctime, sizeof (ctime));
1380 
1381 	if (zp->z_acl_cached) {
1382 		zfs_acl_free(zp->z_acl_cached);
1383 		zp->z_acl_cached = NULL;
1384 	}
1385 
1386 	/*
1387 	 * Upgrade needed?
1388 	 */
1389 	if (!zfsvfs->z_use_fuids) {
1390 		otype = DMU_OT_OLDACL;
1391 	} else {
1392 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1393 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1394 			zfs_acl_xform(zp, aclp, cr);
1395 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1396 		otype = DMU_OT_ACL;
1397 	}
1398 
1399 	/*
1400 	 * Arrgh, we have to handle old on disk format
1401 	 * as well as newer (preferred) SA format.
1402 	 */
1403 
1404 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1405 		locate.cb_aclp = aclp;
1406 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1407 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1408 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1409 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1410 	} else { /* Painful legacy way */
1411 		zfs_acl_node_t *aclnode;
1412 		uint64_t off = 0;
1413 		uint64_t aoid;
1414 
1415 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1416 		    &acl_phys, sizeof (acl_phys))) != 0)
1417 			return (error);
1418 
1419 		aoid = acl_phys.z_acl_extern_obj;
1420 
1421 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1422 			/*
1423 			 * If ACL was previously external and we are now
1424 			 * converting to new ACL format then release old
1425 			 * ACL object and create a new one.
1426 			 */
1427 			if (aoid &&
1428 			    aclp->z_version != acl_phys.z_acl_version) {
1429 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1430 				if (error)
1431 					return (error);
1432 				aoid = 0;
1433 			}
1434 			if (aoid == 0) {
1435 				aoid = dmu_object_alloc(zfsvfs->z_os,
1436 				    otype, aclp->z_acl_bytes,
1437 				    otype == DMU_OT_ACL ?
1438 				    DMU_OT_SYSACL : DMU_OT_NONE,
1439 				    otype == DMU_OT_ACL ?
1440 				    DN_OLD_MAX_BONUSLEN : 0, tx);
1441 			} else {
1442 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1443 				    aoid, aclp->z_acl_bytes, 0, tx);
1444 			}
1445 			acl_phys.z_acl_extern_obj = aoid;
1446 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1447 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1448 				if (aclnode->z_ace_count == 0)
1449 					continue;
1450 				dmu_write(zfsvfs->z_os, aoid, off,
1451 				    aclnode->z_size, aclnode->z_acldata, tx);
1452 				off += aclnode->z_size;
1453 			}
1454 		} else {
1455 			void *start = acl_phys.z_ace_data;
1456 			/*
1457 			 * Migrating back embedded?
1458 			 */
1459 			if (acl_phys.z_acl_extern_obj) {
1460 				error = dmu_object_free(zfsvfs->z_os,
1461 				    acl_phys.z_acl_extern_obj, tx);
1462 				if (error)
1463 					return (error);
1464 				acl_phys.z_acl_extern_obj = 0;
1465 			}
1466 
1467 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1468 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1469 				if (aclnode->z_ace_count == 0)
1470 					continue;
1471 				memcpy(start, aclnode->z_acldata,
1472 				    aclnode->z_size);
1473 				start = (caddr_t)start + aclnode->z_size;
1474 			}
1475 		}
1476 		/*
1477 		 * If Old version then swap count/bytes to match old
1478 		 * layout of znode_acl_phys_t.
1479 		 */
1480 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1481 			acl_phys.z_acl_size = aclp->z_acl_count;
1482 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1483 		} else {
1484 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1485 			acl_phys.z_acl_count = aclp->z_acl_count;
1486 		}
1487 		acl_phys.z_acl_version = aclp->z_version;
1488 
1489 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1490 		    &acl_phys, sizeof (acl_phys));
1491 	}
1492 
1493 	/*
1494 	 * Replace ACL wide bits, but first clear them.
1495 	 */
1496 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1497 
1498 	zp->z_pflags |= aclp->z_hints;
1499 
1500 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1501 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1502 
1503 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1504 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1505 }
1506 
1507 static void
1508 zfs_acl_chmod(boolean_t isdir, uint64_t mode, boolean_t split, boolean_t trim,
1509     zfs_acl_t *aclp)
1510 {
1511 	void		*acep = NULL;
1512 	uint64_t	who;
1513 	int		new_count, new_bytes;
1514 	int		ace_size;
1515 	int		entry_type;
1516 	uint16_t	iflags, type;
1517 	uint32_t	access_mask;
1518 	zfs_acl_node_t	*newnode;
1519 	size_t		abstract_size = aclp->z_ops->ace_abstract_size();
1520 	void		*zacep;
1521 	trivial_acl_t	masks;
1522 
1523 	new_count = new_bytes = 0;
1524 
1525 	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1526 
1527 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1528 
1529 	zacep = newnode->z_acldata;
1530 	if (masks.allow0) {
1531 		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1532 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1533 		new_count++;
1534 		new_bytes += abstract_size;
1535 	}
1536 	if (masks.deny1) {
1537 		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1538 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1539 		new_count++;
1540 		new_bytes += abstract_size;
1541 	}
1542 	if (masks.deny2) {
1543 		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1544 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1545 		new_count++;
1546 		new_bytes += abstract_size;
1547 	}
1548 
1549 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1550 	    &iflags, &type))) {
1551 		entry_type = (iflags & ACE_TYPE_FLAGS);
1552 		/*
1553 		 * ACEs used to represent the file mode may be divided
1554 		 * into an equivalent pair of inherit-only and regular
1555 		 * ACEs, if they are inheritable.
1556 		 * Skip regular ACEs, which are replaced by the new mode.
1557 		 */
1558 		if (split && (entry_type == ACE_OWNER ||
1559 		    entry_type == OWNING_GROUP ||
1560 		    entry_type == ACE_EVERYONE)) {
1561 			if (!isdir || !(iflags &
1562 			    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1563 				continue;
1564 			/*
1565 			 * We preserve owner@, group@, or @everyone
1566 			 * permissions, if they are inheritable, by
1567 			 * copying them to inherit_only ACEs. This
1568 			 * prevents inheritable permissions from being
1569 			 * altered along with the file mode.
1570 			 */
1571 			iflags |= ACE_INHERIT_ONLY_ACE;
1572 		}
1573 
1574 		/*
1575 		 * If this ACL has any inheritable ACEs, mark that in
1576 		 * the hints (which are later masked into the pflags)
1577 		 * so create knows to do inheritance.
1578 		 */
1579 		if (isdir && (iflags &
1580 		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1581 			aclp->z_hints |= ZFS_INHERIT_ACE;
1582 
1583 		if ((type != ALLOW && type != DENY) ||
1584 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1585 			switch (type) {
1586 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1587 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1588 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1589 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1590 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1591 				break;
1592 			}
1593 		} else {
1594 			/*
1595 			 * Limit permissions to be no greater than
1596 			 * group permissions.
1597 			 * The "aclinherit" and "aclmode" properties
1598 			 * affect policy for create and chmod(2),
1599 			 * respectively.
1600 			 */
1601 			if ((type == ALLOW) && trim)
1602 				access_mask &= masks.group;
1603 		}
1604 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1605 		ace_size = aclp->z_ops->ace_size(acep);
1606 		zacep = (void *)((uintptr_t)zacep + ace_size);
1607 		new_count++;
1608 		new_bytes += ace_size;
1609 	}
1610 	zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1611 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1612 	zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1613 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1614 	zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1615 
1616 	new_count += 3;
1617 	new_bytes += abstract_size * 3;
1618 	zfs_acl_release_nodes(aclp);
1619 	aclp->z_acl_count = new_count;
1620 	aclp->z_acl_bytes = new_bytes;
1621 	newnode->z_ace_count = new_count;
1622 	newnode->z_size = new_bytes;
1623 	list_insert_tail(&aclp->z_acl, newnode);
1624 }
1625 
1626 int
1627 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1628 {
1629 	int error = 0;
1630 
1631 	mutex_enter(&zp->z_acl_lock);
1632 	mutex_enter(&zp->z_lock);
1633 	if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_DISCARD)
1634 		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1635 	else
1636 		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1637 
1638 	if (error == 0) {
1639 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1640 		zfs_acl_chmod(S_ISDIR(ZTOI(zp)->i_mode), mode, B_TRUE,
1641 		    (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1642 	}
1643 	mutex_exit(&zp->z_lock);
1644 	mutex_exit(&zp->z_acl_lock);
1645 
1646 	return (error);
1647 }
1648 
1649 /*
1650  * Should ACE be inherited?
1651  */
1652 static int
1653 zfs_ace_can_use(umode_t obj_mode, uint16_t acep_flags)
1654 {
1655 	int	iflags = (acep_flags & 0xf);
1656 
1657 	if (S_ISDIR(obj_mode) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1658 		return (1);
1659 	else if (iflags & ACE_FILE_INHERIT_ACE)
1660 		return (!(S_ISDIR(obj_mode) &&
1661 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1662 	return (0);
1663 }
1664 
1665 /*
1666  * inherit inheritable ACEs from parent
1667  */
1668 static zfs_acl_t *
1669 zfs_acl_inherit(zfsvfs_t *zfsvfs, umode_t va_mode, zfs_acl_t *paclp,
1670     uint64_t mode, boolean_t *need_chmod)
1671 {
1672 	void		*pacep = NULL;
1673 	void		*acep;
1674 	zfs_acl_node_t  *aclnode;
1675 	zfs_acl_t	*aclp = NULL;
1676 	uint64_t	who;
1677 	uint32_t	access_mask;
1678 	uint16_t	iflags, newflags, type;
1679 	size_t		ace_size;
1680 	void		*data1, *data2;
1681 	size_t		data1sz, data2sz;
1682 	uint_t		aclinherit;
1683 	boolean_t	isdir = S_ISDIR(va_mode);
1684 	boolean_t	isreg = S_ISREG(va_mode);
1685 
1686 	*need_chmod = B_TRUE;
1687 
1688 	aclp = zfs_acl_alloc(paclp->z_version);
1689 	aclinherit = zfsvfs->z_acl_inherit;
1690 	if (aclinherit == ZFS_ACL_DISCARD || S_ISLNK(va_mode))
1691 		return (aclp);
1692 
1693 	while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1694 	    &access_mask, &iflags, &type))) {
1695 
1696 		/*
1697 		 * don't inherit bogus ACEs
1698 		 */
1699 		if (!zfs_acl_valid_ace_type(type, iflags))
1700 			continue;
1701 
1702 		/*
1703 		 * Check if ACE is inheritable by this vnode
1704 		 */
1705 		if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1706 		    !zfs_ace_can_use(va_mode, iflags))
1707 			continue;
1708 
1709 		/*
1710 		 * If owner@, group@, or everyone@ inheritable
1711 		 * then zfs_acl_chmod() isn't needed.
1712 		 */
1713 		if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1714 		    aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1715 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1716 		    ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1717 		    (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1718 			*need_chmod = B_FALSE;
1719 
1720 		/*
1721 		 * Strip inherited execute permission from file if
1722 		 * not in mode
1723 		 */
1724 		if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1725 		    !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1726 			access_mask &= ~ACE_EXECUTE;
1727 		}
1728 
1729 		/*
1730 		 * Strip write_acl and write_owner from permissions
1731 		 * when inheriting an ACE
1732 		 */
1733 		if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1734 			access_mask &= ~RESTRICTED_CLEAR;
1735 		}
1736 
1737 		ace_size = aclp->z_ops->ace_size(pacep);
1738 		aclnode = zfs_acl_node_alloc(ace_size);
1739 		list_insert_tail(&aclp->z_acl, aclnode);
1740 		acep = aclnode->z_acldata;
1741 
1742 		zfs_set_ace(aclp, acep, access_mask, type,
1743 		    who, iflags|ACE_INHERITED_ACE);
1744 
1745 		/*
1746 		 * Copy special opaque data if any
1747 		 */
1748 		if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1749 			VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1750 			    &data2)) == data1sz);
1751 			memcpy(data2, data1, data2sz);
1752 		}
1753 
1754 		aclp->z_acl_count++;
1755 		aclnode->z_ace_count++;
1756 		aclp->z_acl_bytes += aclnode->z_size;
1757 		newflags = aclp->z_ops->ace_flags_get(acep);
1758 
1759 		/*
1760 		 * If ACE is not to be inherited further, or if the vnode is
1761 		 * not a directory, remove all inheritance flags
1762 		 */
1763 		if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1764 			newflags &= ~ALL_INHERIT;
1765 			aclp->z_ops->ace_flags_set(acep,
1766 			    newflags|ACE_INHERITED_ACE);
1767 			continue;
1768 		}
1769 
1770 		/*
1771 		 * This directory has an inheritable ACE
1772 		 */
1773 		aclp->z_hints |= ZFS_INHERIT_ACE;
1774 
1775 		/*
1776 		 * If only FILE_INHERIT is set then turn on
1777 		 * inherit_only
1778 		 */
1779 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1780 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1781 			newflags |= ACE_INHERIT_ONLY_ACE;
1782 			aclp->z_ops->ace_flags_set(acep,
1783 			    newflags|ACE_INHERITED_ACE);
1784 		} else {
1785 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1786 			aclp->z_ops->ace_flags_set(acep,
1787 			    newflags|ACE_INHERITED_ACE);
1788 		}
1789 	}
1790 	if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1791 	    aclp->z_acl_count != 0) {
1792 		*need_chmod = B_FALSE;
1793 	}
1794 
1795 	return (aclp);
1796 }
1797 
1798 /*
1799  * Create file system object initial permissions
1800  * including inheritable ACEs.
1801  * Also, create FUIDs for owner and group.
1802  */
1803 int
1804 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1805     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids, zuserns_t *mnt_ns)
1806 {
1807 	int		error;
1808 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
1809 	zfs_acl_t	*paclp;
1810 	gid_t		gid = vap->va_gid;
1811 	boolean_t	need_chmod = B_TRUE;
1812 	boolean_t	trim = B_FALSE;
1813 	boolean_t	inherited = B_FALSE;
1814 
1815 	memset(acl_ids, 0, sizeof (zfs_acl_ids_t));
1816 	acl_ids->z_mode = vap->va_mode;
1817 
1818 	if (vsecp)
1819 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_mode, vsecp,
1820 		    cr, &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1821 			return (error);
1822 
1823 	acl_ids->z_fuid = vap->va_uid;
1824 	acl_ids->z_fgid = vap->va_gid;
1825 #ifdef HAVE_KSID
1826 	/*
1827 	 * Determine uid and gid.
1828 	 */
1829 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1830 	    ((flag & IS_XATTR) && (S_ISDIR(vap->va_mode)))) {
1831 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_uid,
1832 		    cr, ZFS_OWNER, &acl_ids->z_fuidp);
1833 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
1834 		    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1835 		gid = vap->va_gid;
1836 	} else {
1837 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1838 		    cr, &acl_ids->z_fuidp);
1839 		acl_ids->z_fgid = 0;
1840 		if (vap->va_mask & AT_GID)  {
1841 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1842 			    (uint64_t)vap->va_gid,
1843 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1844 			gid = vap->va_gid;
1845 			if (acl_ids->z_fgid != KGID_TO_SGID(ZTOI(dzp)->i_gid) &&
1846 			    !groupmember(vap->va_gid, cr) &&
1847 			    secpolicy_vnode_create_gid(cr) != 0)
1848 				acl_ids->z_fgid = 0;
1849 		}
1850 		if (acl_ids->z_fgid == 0) {
1851 			if (dzp->z_mode & S_ISGID) {
1852 				char		*domain;
1853 				uint32_t	rid;
1854 
1855 				acl_ids->z_fgid = KGID_TO_SGID(
1856 				    ZTOI(dzp)->i_gid);
1857 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1858 				    cr, ZFS_GROUP);
1859 
1860 				if (zfsvfs->z_use_fuids &&
1861 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1862 					domain = zfs_fuid_idx_domain(
1863 					    &zfsvfs->z_fuid_idx,
1864 					    FUID_INDEX(acl_ids->z_fgid));
1865 					rid = FUID_RID(acl_ids->z_fgid);
1866 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1867 					    domain, rid,
1868 					    FUID_INDEX(acl_ids->z_fgid),
1869 					    acl_ids->z_fgid, ZFS_GROUP);
1870 				}
1871 			} else {
1872 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1873 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1874 				gid = crgetgid(cr);
1875 			}
1876 		}
1877 	}
1878 #endif /* HAVE_KSID */
1879 
1880 	/*
1881 	 * If we're creating a directory, and the parent directory has the
1882 	 * set-GID bit set, set in on the new directory.
1883 	 * Otherwise, if the user is neither privileged nor a member of the
1884 	 * file's new group, clear the file's set-GID bit.
1885 	 */
1886 
1887 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1888 	    (S_ISDIR(vap->va_mode))) {
1889 		acl_ids->z_mode |= S_ISGID;
1890 	} else {
1891 		if ((acl_ids->z_mode & S_ISGID) &&
1892 		    secpolicy_vnode_setids_setgids(cr, gid, mnt_ns,
1893 		    zfs_i_user_ns(ZTOI(dzp))) != 0) {
1894 			acl_ids->z_mode &= ~S_ISGID;
1895 		}
1896 	}
1897 
1898 	if (acl_ids->z_aclp == NULL) {
1899 		mutex_enter(&dzp->z_acl_lock);
1900 		mutex_enter(&dzp->z_lock);
1901 		if (!(flag & IS_ROOT_NODE) &&
1902 		    (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1903 		    !(dzp->z_pflags & ZFS_XATTR)) {
1904 			VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1905 			    &paclp, B_FALSE));
1906 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1907 			    vap->va_mode, paclp, acl_ids->z_mode, &need_chmod);
1908 			inherited = B_TRUE;
1909 		} else {
1910 			acl_ids->z_aclp =
1911 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1912 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1913 		}
1914 		mutex_exit(&dzp->z_lock);
1915 		mutex_exit(&dzp->z_acl_lock);
1916 
1917 		if (need_chmod) {
1918 			if (S_ISDIR(vap->va_mode))
1919 				acl_ids->z_aclp->z_hints |=
1920 				    ZFS_ACL_AUTO_INHERIT;
1921 
1922 			if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1923 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1924 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1925 				trim = B_TRUE;
1926 			zfs_acl_chmod(vap->va_mode, acl_ids->z_mode, B_FALSE,
1927 			    trim, acl_ids->z_aclp);
1928 		}
1929 	}
1930 
1931 	if (inherited || vsecp) {
1932 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1933 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1934 		    acl_ids->z_fuid, acl_ids->z_fgid);
1935 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1936 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1937 	}
1938 
1939 	return (0);
1940 }
1941 
1942 /*
1943  * Free ACL and fuid_infop, but not the acl_ids structure
1944  */
1945 void
1946 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1947 {
1948 	if (acl_ids->z_aclp)
1949 		zfs_acl_free(acl_ids->z_aclp);
1950 	if (acl_ids->z_fuidp)
1951 		zfs_fuid_info_free(acl_ids->z_fuidp);
1952 	acl_ids->z_aclp = NULL;
1953 	acl_ids->z_fuidp = NULL;
1954 }
1955 
1956 boolean_t
1957 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1958 {
1959 	return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1960 	    zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1961 	    (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1962 	    zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1963 }
1964 
1965 /*
1966  * Retrieve a file's ACL
1967  */
1968 int
1969 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1970 {
1971 	zfs_acl_t	*aclp;
1972 	ulong_t		mask;
1973 	int		error;
1974 	int 		count = 0;
1975 	int		largeace = 0;
1976 
1977 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1978 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1979 
1980 	if (mask == 0)
1981 		return (SET_ERROR(ENOSYS));
1982 
1983 	if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr,
1984 	    kcred->user_ns)))
1985 		return (error);
1986 
1987 	mutex_enter(&zp->z_acl_lock);
1988 
1989 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1990 	if (error != 0) {
1991 		mutex_exit(&zp->z_acl_lock);
1992 		return (error);
1993 	}
1994 
1995 	/*
1996 	 * Scan ACL to determine number of ACEs
1997 	 */
1998 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1999 		void *zacep = NULL;
2000 		uint64_t who;
2001 		uint32_t access_mask;
2002 		uint16_t type, iflags;
2003 
2004 		while ((zacep = zfs_acl_next_ace(aclp, zacep,
2005 		    &who, &access_mask, &iflags, &type))) {
2006 			switch (type) {
2007 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
2008 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
2009 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
2010 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
2011 				largeace++;
2012 				continue;
2013 			default:
2014 				count++;
2015 			}
2016 		}
2017 		vsecp->vsa_aclcnt = count;
2018 	} else
2019 		count = (int)aclp->z_acl_count;
2020 
2021 	if (mask & VSA_ACECNT) {
2022 		vsecp->vsa_aclcnt = count;
2023 	}
2024 
2025 	if (mask & VSA_ACE) {
2026 		size_t aclsz;
2027 
2028 		aclsz = count * sizeof (ace_t) +
2029 		    sizeof (ace_object_t) * largeace;
2030 
2031 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2032 		vsecp->vsa_aclentsz = aclsz;
2033 
2034 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2035 			zfs_copy_fuid_2_ace(ZTOZSB(zp), aclp, cr,
2036 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2037 		else {
2038 			zfs_acl_node_t *aclnode;
2039 			void *start = vsecp->vsa_aclentp;
2040 
2041 			for (aclnode = list_head(&aclp->z_acl); aclnode;
2042 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
2043 				memcpy(start, aclnode->z_acldata,
2044 				    aclnode->z_size);
2045 				start = (caddr_t)start + aclnode->z_size;
2046 			}
2047 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
2048 			    aclp->z_acl_bytes);
2049 		}
2050 	}
2051 	if (mask & VSA_ACE_ACLFLAGS) {
2052 		vsecp->vsa_aclflags = 0;
2053 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
2054 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
2055 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
2056 			vsecp->vsa_aclflags |= ACL_PROTECTED;
2057 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
2058 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2059 	}
2060 
2061 	mutex_exit(&zp->z_acl_lock);
2062 
2063 	return (0);
2064 }
2065 
2066 int
2067 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_mode,
2068     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2069 {
2070 	zfs_acl_t *aclp;
2071 	zfs_acl_node_t *aclnode;
2072 	int aclcnt = vsecp->vsa_aclcnt;
2073 	int error;
2074 
2075 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2076 		return (SET_ERROR(EINVAL));
2077 
2078 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2079 
2080 	aclp->z_hints = 0;
2081 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2082 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2083 		if ((error = zfs_copy_ace_2_oldace(obj_mode, aclp,
2084 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2085 		    aclcnt, &aclnode->z_size)) != 0) {
2086 			zfs_acl_free(aclp);
2087 			zfs_acl_node_free(aclnode);
2088 			return (error);
2089 		}
2090 	} else {
2091 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_mode, aclp,
2092 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2093 		    &aclnode->z_size, fuidp, cr)) != 0) {
2094 			zfs_acl_free(aclp);
2095 			zfs_acl_node_free(aclnode);
2096 			return (error);
2097 		}
2098 	}
2099 	aclp->z_acl_bytes = aclnode->z_size;
2100 	aclnode->z_ace_count = aclcnt;
2101 	aclp->z_acl_count = aclcnt;
2102 	list_insert_head(&aclp->z_acl, aclnode);
2103 
2104 	/*
2105 	 * If flags are being set then add them to z_hints
2106 	 */
2107 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2108 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2109 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2110 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2111 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2112 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2113 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2114 	}
2115 
2116 	*zaclp = aclp;
2117 
2118 	return (0);
2119 }
2120 
2121 /*
2122  * Set a file's ACL
2123  */
2124 int
2125 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2126 {
2127 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
2128 	zilog_t		*zilog = zfsvfs->z_log;
2129 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2130 	dmu_tx_t	*tx;
2131 	int		error;
2132 	zfs_acl_t	*aclp;
2133 	zfs_fuid_info_t	*fuidp = NULL;
2134 	boolean_t	fuid_dirtied;
2135 	uint64_t	acl_obj;
2136 
2137 	if (mask == 0)
2138 		return (SET_ERROR(ENOSYS));
2139 
2140 	if (zp->z_pflags & ZFS_IMMUTABLE)
2141 		return (SET_ERROR(EPERM));
2142 
2143 	if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr,
2144 	    kcred->user_ns)))
2145 		return (error);
2146 
2147 	error = zfs_vsec_2_aclp(zfsvfs, ZTOI(zp)->i_mode, vsecp, cr, &fuidp,
2148 	    &aclp);
2149 	if (error)
2150 		return (error);
2151 
2152 	/*
2153 	 * If ACL wide flags aren't being set then preserve any
2154 	 * existing flags.
2155 	 */
2156 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2157 		aclp->z_hints |=
2158 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2159 	}
2160 top:
2161 	mutex_enter(&zp->z_acl_lock);
2162 	mutex_enter(&zp->z_lock);
2163 
2164 	tx = dmu_tx_create(zfsvfs->z_os);
2165 
2166 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2167 
2168 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2169 	if (fuid_dirtied)
2170 		zfs_fuid_txhold(zfsvfs, tx);
2171 
2172 	/*
2173 	 * If old version and ACL won't fit in bonus and we aren't
2174 	 * upgrading then take out necessary DMU holds
2175 	 */
2176 
2177 	if ((acl_obj = zfs_external_acl(zp)) != 0) {
2178 		if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2179 		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2180 			dmu_tx_hold_free(tx, acl_obj, 0,
2181 			    DMU_OBJECT_END);
2182 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2183 			    aclp->z_acl_bytes);
2184 		} else {
2185 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2186 		}
2187 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2188 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2189 	}
2190 
2191 	zfs_sa_upgrade_txholds(tx, zp);
2192 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2193 	if (error) {
2194 		mutex_exit(&zp->z_acl_lock);
2195 		mutex_exit(&zp->z_lock);
2196 
2197 		if (error == ERESTART) {
2198 			dmu_tx_wait(tx);
2199 			dmu_tx_abort(tx);
2200 			goto top;
2201 		}
2202 		dmu_tx_abort(tx);
2203 		zfs_acl_free(aclp);
2204 		return (error);
2205 	}
2206 
2207 	error = zfs_aclset_common(zp, aclp, cr, tx);
2208 	ASSERT(error == 0);
2209 	ASSERT(zp->z_acl_cached == NULL);
2210 	zp->z_acl_cached = aclp;
2211 
2212 	if (fuid_dirtied)
2213 		zfs_fuid_sync(zfsvfs, tx);
2214 
2215 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2216 
2217 	if (fuidp)
2218 		zfs_fuid_info_free(fuidp);
2219 	dmu_tx_commit(tx);
2220 
2221 	mutex_exit(&zp->z_lock);
2222 	mutex_exit(&zp->z_acl_lock);
2223 
2224 	return (error);
2225 }
2226 
2227 /*
2228  * Check accesses of interest (AoI) against attributes of the dataset
2229  * such as read-only.  Returns zero if no AoI conflict with dataset
2230  * attributes, otherwise an appropriate errno is returned.
2231  */
2232 static int
2233 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2234 {
2235 	if ((v4_mode & WRITE_MASK) && (zfs_is_readonly(ZTOZSB(zp))) &&
2236 	    (!Z_ISDEV(ZTOI(zp)->i_mode) || (v4_mode & WRITE_MASK_ATTRS))) {
2237 		return (SET_ERROR(EROFS));
2238 	}
2239 
2240 	/*
2241 	 * Intentionally allow ZFS_READONLY through here.
2242 	 * See zfs_zaccess_common().
2243 	 */
2244 	if ((v4_mode & WRITE_MASK_DATA) &&
2245 	    (zp->z_pflags & ZFS_IMMUTABLE)) {
2246 		return (SET_ERROR(EPERM));
2247 	}
2248 
2249 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2250 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2251 		return (SET_ERROR(EPERM));
2252 	}
2253 
2254 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2255 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2256 		return (SET_ERROR(EACCES));
2257 	}
2258 
2259 	return (0);
2260 }
2261 
2262 /*
2263  * The primary usage of this function is to loop through all of the
2264  * ACEs in the znode, determining what accesses of interest (AoI) to
2265  * the caller are allowed or denied.  The AoI are expressed as bits in
2266  * the working_mode parameter.  As each ACE is processed, bits covered
2267  * by that ACE are removed from the working_mode.  This removal
2268  * facilitates two things.  The first is that when the working mode is
2269  * empty (= 0), we know we've looked at all the AoI. The second is
2270  * that the ACE interpretation rules don't allow a later ACE to undo
2271  * something granted or denied by an earlier ACE.  Removing the
2272  * discovered access or denial enforces this rule.  At the end of
2273  * processing the ACEs, all AoI that were found to be denied are
2274  * placed into the working_mode, giving the caller a mask of denied
2275  * accesses.  Returns:
2276  *	0		if all AoI granted
2277  *	EACCES 		if the denied mask is non-zero
2278  *	other error	if abnormal failure (e.g., IO error)
2279  *
2280  * A secondary usage of the function is to determine if any of the
2281  * AoI are granted.  If an ACE grants any access in
2282  * the working_mode, we immediately short circuit out of the function.
2283  * This mode is chosen by setting anyaccess to B_TRUE.  The
2284  * working_mode is not a denied access mask upon exit if the function
2285  * is used in this manner.
2286  */
2287 static int
2288 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2289     boolean_t anyaccess, cred_t *cr, zuserns_t *mnt_ns)
2290 {
2291 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
2292 	zfs_acl_t	*aclp;
2293 	int		error;
2294 	uid_t		uid = crgetuid(cr);
2295 	uint64_t	who;
2296 	uint16_t	type, iflags;
2297 	uint16_t	entry_type;
2298 	uint32_t	access_mask;
2299 	uint32_t	deny_mask = 0;
2300 	zfs_ace_hdr_t	*acep = NULL;
2301 	boolean_t	checkit;
2302 	uid_t		gowner;
2303 	uid_t		fowner;
2304 
2305 	if (mnt_ns) {
2306 		fowner = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2307 		    KUID_TO_SUID(ZTOI(zp)->i_uid));
2308 		gowner = zfs_gid_to_vfsgid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2309 		    KGID_TO_SGID(ZTOI(zp)->i_gid));
2310 	} else
2311 		zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2312 
2313 	mutex_enter(&zp->z_acl_lock);
2314 
2315 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2316 	if (error != 0) {
2317 		mutex_exit(&zp->z_acl_lock);
2318 		return (error);
2319 	}
2320 
2321 	ASSERT(zp->z_acl_cached);
2322 
2323 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2324 	    &iflags, &type))) {
2325 		uint32_t mask_matched;
2326 
2327 		if (!zfs_acl_valid_ace_type(type, iflags))
2328 			continue;
2329 
2330 		if (S_ISDIR(ZTOI(zp)->i_mode) &&
2331 		    (iflags & ACE_INHERIT_ONLY_ACE))
2332 			continue;
2333 
2334 		/* Skip ACE if it does not affect any AoI */
2335 		mask_matched = (access_mask & *working_mode);
2336 		if (!mask_matched)
2337 			continue;
2338 
2339 		entry_type = (iflags & ACE_TYPE_FLAGS);
2340 
2341 		checkit = B_FALSE;
2342 
2343 		switch (entry_type) {
2344 		case ACE_OWNER:
2345 			if (uid == fowner)
2346 				checkit = B_TRUE;
2347 			break;
2348 		case OWNING_GROUP:
2349 			who = gowner;
2350 			zfs_fallthrough;
2351 		case ACE_IDENTIFIER_GROUP:
2352 			checkit = zfs_groupmember(zfsvfs, who, cr);
2353 			break;
2354 		case ACE_EVERYONE:
2355 			checkit = B_TRUE;
2356 			break;
2357 
2358 		/* USER Entry */
2359 		default:
2360 			if (entry_type == 0) {
2361 				uid_t newid;
2362 
2363 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2364 				    ZFS_ACE_USER);
2365 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2366 				    uid == newid)
2367 					checkit = B_TRUE;
2368 				break;
2369 			} else {
2370 				mutex_exit(&zp->z_acl_lock);
2371 				return (SET_ERROR(EIO));
2372 			}
2373 		}
2374 
2375 		if (checkit) {
2376 			if (type == DENY) {
2377 				DTRACE_PROBE3(zfs__ace__denies,
2378 				    znode_t *, zp,
2379 				    zfs_ace_hdr_t *, acep,
2380 				    uint32_t, mask_matched);
2381 				deny_mask |= mask_matched;
2382 			} else {
2383 				DTRACE_PROBE3(zfs__ace__allows,
2384 				    znode_t *, zp,
2385 				    zfs_ace_hdr_t *, acep,
2386 				    uint32_t, mask_matched);
2387 				if (anyaccess) {
2388 					mutex_exit(&zp->z_acl_lock);
2389 					return (0);
2390 				}
2391 			}
2392 			*working_mode &= ~mask_matched;
2393 		}
2394 
2395 		/* Are we done? */
2396 		if (*working_mode == 0)
2397 			break;
2398 	}
2399 
2400 	mutex_exit(&zp->z_acl_lock);
2401 
2402 	/* Put the found 'denies' back on the working mode */
2403 	if (deny_mask) {
2404 		*working_mode |= deny_mask;
2405 		return (SET_ERROR(EACCES));
2406 	} else if (*working_mode) {
2407 		return (-1);
2408 	}
2409 
2410 	return (0);
2411 }
2412 
2413 /*
2414  * Return true if any access whatsoever granted, we don't actually
2415  * care what access is granted.
2416  */
2417 boolean_t
2418 zfs_has_access(znode_t *zp, cred_t *cr)
2419 {
2420 	uint32_t have = ACE_ALL_PERMS;
2421 
2422 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr,
2423 	    kcred->user_ns) != 0) {
2424 		uid_t owner;
2425 
2426 		owner = zfs_fuid_map_id(ZTOZSB(zp),
2427 		    KUID_TO_SUID(ZTOI(zp)->i_uid), cr, ZFS_OWNER);
2428 		return (secpolicy_vnode_any_access(cr, ZTOI(zp), owner) == 0);
2429 	}
2430 	return (B_TRUE);
2431 }
2432 
2433 /*
2434  * Simplified access check for case where ACL is known to not contain
2435  * information beyond what is defined in the mode. In this case, we
2436  * can pass along to the kernel / vfs generic_permission() check, which
2437  * evaluates the mode and POSIX ACL.
2438  *
2439  * NFSv4 ACLs allow granting permissions that are usually relegated only
2440  * to the file owner or superuser. Examples are ACE_WRITE_OWNER (chown),
2441  * ACE_WRITE_ACL(chmod), and ACE_DELETE. ACE_DELETE requests must fail
2442  * because with conventional posix permissions, right to delete file
2443  * is determined by write bit on the parent dir.
2444  *
2445  * If unmappable perms are requested, then we must return EPERM
2446  * and include those bits in the working_mode so that the caller of
2447  * zfs_zaccess_common() can decide whether to perform additional
2448  * policy / capability checks. EACCES is used in zfs_zaccess_aces_check()
2449  * to indicate access check failed due to explicit DENY entry, and so
2450  * we want to avoid that here.
2451  */
2452 static int
2453 zfs_zaccess_trivial(znode_t *zp, uint32_t *working_mode, cred_t *cr,
2454     zuserns_t *mnt_ns)
2455 {
2456 	int err, mask;
2457 	int unmapped = 0;
2458 
2459 	ASSERT(zp->z_pflags & ZFS_ACL_TRIVIAL);
2460 
2461 	mask = zfs_v4_to_unix(*working_mode, &unmapped);
2462 	if (mask == 0 || unmapped) {
2463 		*working_mode = unmapped;
2464 		return (unmapped ? SET_ERROR(EPERM) : 0);
2465 	}
2466 
2467 #if defined(HAVE_IOPS_PERMISSION_USERNS)
2468 	if (mnt_ns)
2469 		err = generic_permission(mnt_ns, ZTOI(zp), mask);
2470 	else
2471 		err = generic_permission(cr->user_ns, ZTOI(zp), mask);
2472 #else
2473 	err = generic_permission(ZTOI(zp), mask);
2474 #endif
2475 	if (err != 0) {
2476 		return (SET_ERROR(EPERM));
2477 	}
2478 
2479 	*working_mode = unmapped;
2480 
2481 	return (0);
2482 }
2483 
2484 static int
2485 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2486     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr, zuserns_t *mnt_ns)
2487 {
2488 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
2489 	int err;
2490 
2491 	*working_mode = v4_mode;
2492 	*check_privs = B_TRUE;
2493 
2494 	/*
2495 	 * Short circuit empty requests
2496 	 */
2497 	if (v4_mode == 0 || zfsvfs->z_replay) {
2498 		*working_mode = 0;
2499 		return (0);
2500 	}
2501 
2502 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2503 		*check_privs = B_FALSE;
2504 		return (err);
2505 	}
2506 
2507 	/*
2508 	 * The caller requested that the ACL check be skipped.  This
2509 	 * would only happen if the caller checked VOP_ACCESS() with a
2510 	 * 32 bit ACE mask and already had the appropriate permissions.
2511 	 */
2512 	if (skipaclchk) {
2513 		*working_mode = 0;
2514 		return (0);
2515 	}
2516 
2517 	/*
2518 	 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2519 	 * When that flag is set, we should behave as if write access
2520 	 * were not granted by anything in the ACL.  In particular:
2521 	 * We _must_ allow writes after opening the file r/w, then
2522 	 * setting the DOS R/O attribute, and writing some more.
2523 	 * (Similar to how you can write after fchmod(fd, 0444).)
2524 	 *
2525 	 * Therefore ZFS_READONLY is ignored in the dataset check
2526 	 * above, and checked here as if part of the ACL check.
2527 	 * Also note: DOS R/O is ignored for directories.
2528 	 */
2529 	if ((v4_mode & WRITE_MASK_DATA) &&
2530 	    S_ISDIR(ZTOI(zp)->i_mode) &&
2531 	    (zp->z_pflags & ZFS_READONLY)) {
2532 		return (SET_ERROR(EPERM));
2533 	}
2534 
2535 	if (zp->z_pflags & ZFS_ACL_TRIVIAL)
2536 		return (zfs_zaccess_trivial(zp, working_mode, cr, mnt_ns));
2537 
2538 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr, mnt_ns));
2539 }
2540 
2541 static int
2542 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2543     cred_t *cr, zuserns_t *mnt_ns)
2544 {
2545 	if (*working_mode != ACE_WRITE_DATA)
2546 		return (SET_ERROR(EACCES));
2547 
2548 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2549 	    check_privs, B_FALSE, cr, mnt_ns));
2550 }
2551 
2552 int
2553 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2554 {
2555 	boolean_t owner = B_FALSE;
2556 	boolean_t groupmbr = B_FALSE;
2557 	boolean_t is_attr;
2558 	uid_t uid = crgetuid(cr);
2559 	int error;
2560 
2561 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2562 		return (SET_ERROR(EACCES));
2563 
2564 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2565 	    (S_ISDIR(ZTOI(zdp)->i_mode)));
2566 	if (is_attr)
2567 		goto slow;
2568 
2569 
2570 	mutex_enter(&zdp->z_acl_lock);
2571 
2572 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2573 		mutex_exit(&zdp->z_acl_lock);
2574 		return (0);
2575 	}
2576 
2577 	if (KUID_TO_SUID(ZTOI(zdp)->i_uid) != 0 ||
2578 	    KGID_TO_SGID(ZTOI(zdp)->i_gid) != 0) {
2579 		mutex_exit(&zdp->z_acl_lock);
2580 		goto slow;
2581 	}
2582 
2583 	if (uid == KUID_TO_SUID(ZTOI(zdp)->i_uid)) {
2584 		if (zdp->z_mode & S_IXUSR) {
2585 			mutex_exit(&zdp->z_acl_lock);
2586 			return (0);
2587 		} else {
2588 			mutex_exit(&zdp->z_acl_lock);
2589 			goto slow;
2590 		}
2591 	}
2592 	if (groupmember(KGID_TO_SGID(ZTOI(zdp)->i_gid), cr)) {
2593 		if (zdp->z_mode & S_IXGRP) {
2594 			mutex_exit(&zdp->z_acl_lock);
2595 			return (0);
2596 		} else {
2597 			mutex_exit(&zdp->z_acl_lock);
2598 			goto slow;
2599 		}
2600 	}
2601 	if (!owner && !groupmbr) {
2602 		if (zdp->z_mode & S_IXOTH) {
2603 			mutex_exit(&zdp->z_acl_lock);
2604 			return (0);
2605 		}
2606 	}
2607 
2608 	mutex_exit(&zdp->z_acl_lock);
2609 
2610 slow:
2611 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2612 	if ((error = zfs_enter(ZTOZSB(zdp), FTAG)) != 0)
2613 		return (error);
2614 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr,
2615 	    kcred->user_ns);
2616 	zfs_exit(ZTOZSB(zdp), FTAG);
2617 	return (error);
2618 }
2619 
2620 /*
2621  * Determine whether Access should be granted/denied.
2622  *
2623  * The least priv subsystem is always consulted as a basic privilege
2624  * can define any form of access.
2625  */
2626 int
2627 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr,
2628     zuserns_t *mnt_ns)
2629 {
2630 	uint32_t	working_mode;
2631 	int		error;
2632 	int		is_attr;
2633 	boolean_t 	check_privs;
2634 	znode_t		*xzp;
2635 	znode_t 	*check_zp = zp;
2636 	mode_t		needed_bits;
2637 	uid_t		owner;
2638 
2639 	is_attr = ((zp->z_pflags & ZFS_XATTR) && S_ISDIR(ZTOI(zp)->i_mode));
2640 
2641 	/*
2642 	 * If attribute then validate against base file
2643 	 */
2644 	if (is_attr) {
2645 		if ((error = zfs_zget(ZTOZSB(zp),
2646 		    zp->z_xattr_parent, &xzp)) != 0) {
2647 			return (error);
2648 		}
2649 
2650 		check_zp = xzp;
2651 
2652 		/*
2653 		 * fixup mode to map to xattr perms
2654 		 */
2655 
2656 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2657 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2658 			mode |= ACE_WRITE_NAMED_ATTRS;
2659 		}
2660 
2661 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2662 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2663 			mode |= ACE_READ_NAMED_ATTRS;
2664 		}
2665 	}
2666 
2667 	owner = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2668 	    KUID_TO_SUID(ZTOI(zp)->i_uid));
2669 	owner = zfs_fuid_map_id(ZTOZSB(zp), owner, cr, ZFS_OWNER);
2670 
2671 	/*
2672 	 * Map the bits required to the standard inode flags
2673 	 * S_IRUSR|S_IWUSR|S_IXUSR in the needed_bits.  Map the bits
2674 	 * mapped by working_mode (currently missing) in missing_bits.
2675 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2676 	 * needed_bits.
2677 	 */
2678 	needed_bits = 0;
2679 
2680 	working_mode = mode;
2681 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2682 	    owner == crgetuid(cr))
2683 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2684 
2685 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2686 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2687 		needed_bits |= S_IRUSR;
2688 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2689 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2690 		needed_bits |= S_IWUSR;
2691 	if (working_mode & ACE_EXECUTE)
2692 		needed_bits |= S_IXUSR;
2693 
2694 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2695 	    &check_privs, skipaclchk, cr, mnt_ns)) == 0) {
2696 		if (is_attr)
2697 			zrele(xzp);
2698 		return (secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2699 		    needed_bits, needed_bits));
2700 	}
2701 
2702 	if (error && !check_privs) {
2703 		if (is_attr)
2704 			zrele(xzp);
2705 		return (error);
2706 	}
2707 
2708 	if (error && (flags & V_APPEND)) {
2709 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr,
2710 		    mnt_ns);
2711 	}
2712 
2713 	if (error && check_privs) {
2714 		mode_t		checkmode = 0;
2715 
2716 		/*
2717 		 * First check for implicit owner permission on
2718 		 * read_acl/read_attributes
2719 		 */
2720 
2721 		ASSERT(working_mode != 0);
2722 
2723 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2724 		    owner == crgetuid(cr)))
2725 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2726 
2727 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2728 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2729 			checkmode |= S_IRUSR;
2730 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2731 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2732 			checkmode |= S_IWUSR;
2733 		if (working_mode & ACE_EXECUTE)
2734 			checkmode |= S_IXUSR;
2735 
2736 		error = secpolicy_vnode_access2(cr, ZTOI(check_zp), owner,
2737 		    needed_bits & ~checkmode, needed_bits);
2738 
2739 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2740 			error = secpolicy_vnode_chown(cr, owner);
2741 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2742 			error = secpolicy_vnode_setdac(cr, owner);
2743 
2744 		if (error == 0 && (working_mode &
2745 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2746 			error = secpolicy_vnode_remove(cr);
2747 
2748 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2749 			error = secpolicy_vnode_chown(cr, owner);
2750 		}
2751 		if (error == 0) {
2752 			/*
2753 			 * See if any bits other than those already checked
2754 			 * for are still present.  If so then return EACCES
2755 			 */
2756 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2757 				error = SET_ERROR(EACCES);
2758 			}
2759 		}
2760 	} else if (error == 0) {
2761 		error = secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2762 		    needed_bits, needed_bits);
2763 	}
2764 
2765 	if (is_attr)
2766 		zrele(xzp);
2767 
2768 	return (error);
2769 }
2770 
2771 /*
2772  * Translate traditional unix S_IRUSR/S_IWUSR/S_IXUSR mode into
2773  * NFSv4-style ZFS ACL format and call zfs_zaccess()
2774  */
2775 int
2776 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr,
2777     zuserns_t *mnt_ns)
2778 {
2779 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr,
2780 	    mnt_ns));
2781 }
2782 
2783 /*
2784  * Access function for secpolicy_vnode_setattr
2785  */
2786 int
2787 zfs_zaccess_unix(void *zp, int mode, cred_t *cr)
2788 {
2789 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2790 
2791 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr, kcred->user_ns));
2792 }
2793 
2794 /* See zfs_zaccess_delete() */
2795 static const boolean_t zfs_write_implies_delete_child = B_TRUE;
2796 
2797 /*
2798  * Determine whether delete access should be granted.
2799  *
2800  * The following chart outlines how we handle delete permissions which is
2801  * how recent versions of windows (Windows 2008) handles it.  The efficiency
2802  * comes from not having to check the parent ACL where the object itself grants
2803  * delete:
2804  *
2805  *      -------------------------------------------------------
2806  *      |   Parent Dir  |      Target Object Permissions      |
2807  *      |  permissions  |                                     |
2808  *      -------------------------------------------------------
2809  *      |               | ACL Allows | ACL Denies| Delete     |
2810  *      |               |  Delete    |  Delete   | unspecified|
2811  *      -------------------------------------------------------
2812  *      | ACL Allows    | Permit     | Deny *    | Permit     |
2813  *      | DELETE_CHILD  |            |           |            |
2814  *      -------------------------------------------------------
2815  *      | ACL Denies    | Permit     | Deny      | Deny       |
2816  *      | DELETE_CHILD  |            |           |            |
2817  *      -------------------------------------------------------
2818  *      | ACL specifies |            |           |            |
2819  *      | only allow    | Permit     | Deny *    | Permit     |
2820  *      | write and     |            |           |            |
2821  *      | execute       |            |           |            |
2822  *      -------------------------------------------------------
2823  *      | ACL denies    |            |           |            |
2824  *      | write and     | Permit     | Deny      | Deny       |
2825  *      | execute       |            |           |            |
2826  *      -------------------------------------------------------
2827  *         ^
2828  *         |
2829  *         Re. execute permission on the directory:  if that's missing,
2830  *	   the vnode lookup of the target will fail before we get here.
2831  *
2832  * Re [*] in the table above:  NFSv4 would normally Permit delete for
2833  * these two cells of the matrix.
2834  * See acl.h for notes on which ACE_... flags should be checked for which
2835  * operations.  Specifically, the NFSv4 committee recommendation is in
2836  * conflict with the Windows interpretation of DENY ACEs, where DENY ACEs
2837  * should take precedence ahead of ALLOW ACEs.
2838  *
2839  * This implementation always consults the target object's ACL first.
2840  * If a DENY ACE is present on the target object that specifies ACE_DELETE,
2841  * delete access is denied.  If an ALLOW ACE with ACE_DELETE is present on
2842  * the target object, access is allowed.  If and only if no entries with
2843  * ACE_DELETE are present in the object's ACL, check the container's ACL
2844  * for entries with ACE_DELETE_CHILD.
2845  *
2846  * A summary of the logic implemented from the table above is as follows:
2847  *
2848  * First check for DENY ACEs that apply.
2849  * If either target or container has a deny, EACCES.
2850  *
2851  * Delete access can then be summarized as follows:
2852  * 1: The object to be deleted grants ACE_DELETE, or
2853  * 2: The containing directory grants ACE_DELETE_CHILD.
2854  * In a Windows system, that would be the end of the story.
2855  * In this system, (2) has some complications...
2856  * 2a: "sticky" bit on a directory adds restrictions, and
2857  * 2b: existing ACEs from previous versions of ZFS may
2858  * not carry ACE_DELETE_CHILD where they should, so we
2859  * also allow delete when ACE_WRITE_DATA is granted.
2860  *
2861  * Note: 2b is technically a work-around for a prior bug,
2862  * which hopefully can go away some day.  For those who
2863  * no longer need the work around, and for testing, this
2864  * work-around is made conditional via the tunable:
2865  * zfs_write_implies_delete_child
2866  */
2867 int
2868 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr, zuserns_t *mnt_ns)
2869 {
2870 	uint32_t wanted_dirperms;
2871 	uint32_t dzp_working_mode = 0;
2872 	uint32_t zp_working_mode = 0;
2873 	int dzp_error, zp_error;
2874 	boolean_t dzpcheck_privs;
2875 	boolean_t zpcheck_privs;
2876 
2877 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2878 		return (SET_ERROR(EPERM));
2879 
2880 	/*
2881 	 * Case 1:
2882 	 * If target object grants ACE_DELETE then we are done.  This is
2883 	 * indicated by a return value of 0.  For this case we don't worry
2884 	 * about the sticky bit because sticky only applies to the parent
2885 	 * directory and this is the child access result.
2886 	 *
2887 	 * If we encounter a DENY ACE here, we're also done (EACCES).
2888 	 * Note that if we hit a DENY ACE here (on the target) it should
2889 	 * take precedence over a DENY ACE on the container, so that when
2890 	 * we have more complete auditing support we will be able to
2891 	 * report an access failure against the specific target.
2892 	 * (This is part of why we're checking the target first.)
2893 	 */
2894 	zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2895 	    &zpcheck_privs, B_FALSE, cr, mnt_ns);
2896 	if (zp_error == EACCES) {
2897 		/* We hit a DENY ACE. */
2898 		if (!zpcheck_privs)
2899 			return (SET_ERROR(zp_error));
2900 		return (secpolicy_vnode_remove(cr));
2901 
2902 	}
2903 	if (zp_error == 0)
2904 		return (0);
2905 
2906 	/*
2907 	 * Case 2:
2908 	 * If the containing directory grants ACE_DELETE_CHILD,
2909 	 * or we're in backward compatibility mode and the
2910 	 * containing directory has ACE_WRITE_DATA, allow.
2911 	 * Case 2b is handled with wanted_dirperms.
2912 	 */
2913 	wanted_dirperms = ACE_DELETE_CHILD;
2914 	if (zfs_write_implies_delete_child)
2915 		wanted_dirperms |= ACE_WRITE_DATA;
2916 	dzp_error = zfs_zaccess_common(dzp, wanted_dirperms,
2917 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr, mnt_ns);
2918 	if (dzp_error == EACCES) {
2919 		/* We hit a DENY ACE. */
2920 		if (!dzpcheck_privs)
2921 			return (SET_ERROR(dzp_error));
2922 		return (secpolicy_vnode_remove(cr));
2923 	}
2924 
2925 	/*
2926 	 * Cases 2a, 2b (continued)
2927 	 *
2928 	 * Note: dzp_working_mode now contains any permissions
2929 	 * that were NOT granted.  Therefore, if any of the
2930 	 * wanted_dirperms WERE granted, we will have:
2931 	 *   dzp_working_mode != wanted_dirperms
2932 	 * We're really asking if ANY of those permissions
2933 	 * were granted, and if so, grant delete access.
2934 	 */
2935 	if (dzp_working_mode != wanted_dirperms)
2936 		dzp_error = 0;
2937 
2938 	/*
2939 	 * dzp_error is 0 if the container granted us permissions to "modify".
2940 	 * If we do not have permission via one or more ACEs, our current
2941 	 * privileges may still permit us to modify the container.
2942 	 *
2943 	 * dzpcheck_privs is false when i.e. the FS is read-only.
2944 	 * Otherwise, do privilege checks for the container.
2945 	 */
2946 	if (dzp_error != 0 && dzpcheck_privs) {
2947 		uid_t owner;
2948 
2949 		/*
2950 		 * The secpolicy call needs the requested access and
2951 		 * the current access mode of the container, but it
2952 		 * only knows about Unix-style modes (VEXEC, VWRITE),
2953 		 * so this must condense the fine-grained ACE bits into
2954 		 * Unix modes.
2955 		 *
2956 		 * The VEXEC flag is easy, because we know that has
2957 		 * always been checked before we get here (during the
2958 		 * lookup of the target vnode).  The container has not
2959 		 * granted us permissions to "modify", so we do not set
2960 		 * the VWRITE flag in the current access mode.
2961 		 */
2962 		owner = zfs_fuid_map_id(ZTOZSB(dzp),
2963 		    KUID_TO_SUID(ZTOI(dzp)->i_uid), cr, ZFS_OWNER);
2964 		dzp_error = secpolicy_vnode_access2(cr, ZTOI(dzp),
2965 		    owner, S_IXUSR, S_IWUSR|S_IXUSR);
2966 	}
2967 	if (dzp_error != 0) {
2968 		/*
2969 		 * Note: We may have dzp_error = -1 here (from
2970 		 * zfs_zacess_common).  Don't return that.
2971 		 */
2972 		return (SET_ERROR(EACCES));
2973 	}
2974 
2975 
2976 	/*
2977 	 * At this point, we know that the directory permissions allow
2978 	 * us to modify, but we still need to check for the additional
2979 	 * restrictions that apply when the "sticky bit" is set.
2980 	 *
2981 	 * Yes, zfs_sticky_remove_access() also checks this bit, but
2982 	 * checking it here and skipping the call below is nice when
2983 	 * you're watching all of this with dtrace.
2984 	 */
2985 	if ((dzp->z_mode & S_ISVTX) == 0)
2986 		return (0);
2987 
2988 	/*
2989 	 * zfs_sticky_remove_access will succeed if:
2990 	 * 1. The sticky bit is absent.
2991 	 * 2. We pass the sticky bit restrictions.
2992 	 * 3. We have privileges that always allow file removal.
2993 	 */
2994 	return (zfs_sticky_remove_access(dzp, zp, cr));
2995 }
2996 
2997 int
2998 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2999     znode_t *tzp, cred_t *cr, zuserns_t *mnt_ns)
3000 {
3001 	int add_perm;
3002 	int error;
3003 
3004 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
3005 		return (SET_ERROR(EACCES));
3006 
3007 	add_perm = S_ISDIR(ZTOI(szp)->i_mode) ?
3008 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
3009 
3010 	/*
3011 	 * Rename permissions are combination of delete permission +
3012 	 * add file/subdir permission.
3013 	 */
3014 
3015 	/*
3016 	 * first make sure we do the delete portion.
3017 	 *
3018 	 * If that succeeds then check for add_file/add_subdir permissions
3019 	 */
3020 
3021 	if ((error = zfs_zaccess_delete(sdzp, szp, cr, mnt_ns)))
3022 		return (error);
3023 
3024 	/*
3025 	 * If we have a tzp, see if we can delete it?
3026 	 */
3027 	if (tzp) {
3028 		if ((error = zfs_zaccess_delete(tdzp, tzp, cr, mnt_ns)))
3029 			return (error);
3030 	}
3031 
3032 	/*
3033 	 * Now check for add permissions
3034 	 */
3035 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr, mnt_ns);
3036 
3037 	return (error);
3038 }
3039