xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_acl.c (revision 9e37f2b5d225bd6428d62c06655138fc78fd1ac0)
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 http://www.opensolaris.org/os/licensing.
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/time.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/resource.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/unistd.h>
41 #include <sys/sdt.h>
42 #include <sys/fs/zfs.h>
43 #include <sys/mode.h>
44 #include <sys/policy.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_fuid.h>
47 #include <sys/zfs_acl.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_vfsops.h>
50 #include <sys/dmu.h>
51 #include <sys/dnode.h>
52 #include <sys/zap.h>
53 #include <sys/sa.h>
54 #include "fs/fs_subr.h"
55 #include <acl/acl_common.h>
56 
57 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
58 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
59 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
60 #define	MIN_ACE_TYPE	ALLOW
61 
62 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
63 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
64     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
65 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
66     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
67 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
68     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
69 
70 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
71     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
72     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
73     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
74 
75 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
76 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
77     ACE_DELETE|ACE_DELETE_CHILD)
78 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
79 
80 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
81     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
82 
83 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
84     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
85 
86 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
87     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
88 
89 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
90 
91 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
92     ZFS_ACL_PROTECTED)
93 
94 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
95     ZFS_ACL_OBJ_ACE)
96 
97 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
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 /*ARGSUSED*/
148 static size_t
149 zfs_ace_v0_size(void *acep)
150 {
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 /*ARGSUSED*/
167 static int
168 zfs_ace_v0_data(void *acep, void **datap)
169 {
170 	*datap = NULL;
171 	return (0);
172 }
173 
174 static acl_ops_t zfs_acl_v0_ops = {
175 	zfs_ace_v0_get_mask,
176 	zfs_ace_v0_set_mask,
177 	zfs_ace_v0_get_flags,
178 	zfs_ace_v0_set_flags,
179 	zfs_ace_v0_get_type,
180 	zfs_ace_v0_set_type,
181 	zfs_ace_v0_get_who,
182 	zfs_ace_v0_set_who,
183 	zfs_ace_v0_size,
184 	zfs_ace_v0_abstract_size,
185 	zfs_ace_v0_mask_off,
186 	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 		/*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 acl_ops_t zfs_acl_fuid_ops = {
311 	zfs_ace_fuid_get_mask,
312 	zfs_ace_fuid_set_mask,
313 	zfs_ace_fuid_get_flags,
314 	zfs_ace_fuid_set_flags,
315 	zfs_ace_fuid_get_type,
316 	zfs_ace_fuid_set_type,
317 	zfs_ace_fuid_get_who,
318 	zfs_ace_fuid_set_who,
319 	zfs_ace_fuid_size,
320 	zfs_ace_fuid_abstract_size,
321 	zfs_ace_fuid_mask_off,
322 	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 
332 uint64_t
333 zfs_external_acl(znode_t *zp)
334 {
335 	zfs_acl_phys_t acl_phys;
336 
337 	if (zp->z_is_sa)
338 		return (0);
339 
340 	VERIFY(0 == sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
341 	    &acl_phys, sizeof (acl_phys)));
342 
343 	return (acl_phys.z_acl_extern_obj);
344 }
345 
346 /*
347  * Determine size of ACL in bytes
348  *
349  * This is more complicated than it should be since we have to deal
350  * with old external ACLs.
351  */
352 static int
353 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
354     zfs_acl_phys_t *aclphys)
355 {
356 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
357 	uint64_t acl_count;
358 	int size;
359 	int error;
360 
361 	if (zp->z_is_sa) {
362 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
363 		    &size)) != 0)
364 			return (error);
365 		*aclsize = size;
366 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
367 		    &acl_count, sizeof (acl_count))) != 0)
368 			return (error);
369 		*aclcount = acl_count;
370 	} else {
371 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
372 		    aclphys, sizeof (*aclphys))) != 0)
373 			return (error);
374 
375 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
376 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
377 			*aclcount = aclphys->z_acl_size;
378 		} else {
379 			*aclsize = aclphys->z_acl_size;
380 			*aclcount = aclphys->z_acl_count;
381 		}
382 	}
383 	return (0);
384 }
385 
386 int
387 zfs_znode_acl_version(znode_t *zp)
388 {
389 	zfs_acl_phys_t acl_phys;
390 
391 	if (zp->z_is_sa) {
392 		return (ZFS_ACL_VERSION_FUID);
393 	} else {
394 		VERIFY(0 == sa_lookup(zp->z_sa_hdl,
395 		    SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
396 		    &acl_phys, sizeof (acl_phys)));
397 		return (acl_phys.z_acl_version);
398 	}
399 }
400 
401 static int
402 zfs_acl_version(int version)
403 {
404 	if (version < ZPL_VERSION_FUID)
405 		return (ZFS_ACL_VERSION_INITIAL);
406 	else
407 		return (ZFS_ACL_VERSION_FUID);
408 }
409 
410 static int
411 zfs_acl_version_zp(znode_t *zp)
412 {
413 	return (zfs_acl_version(zp->z_zfsvfs->z_version));
414 }
415 
416 zfs_acl_t *
417 zfs_acl_alloc(int vers)
418 {
419 	zfs_acl_t *aclp;
420 
421 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
422 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
423 	    offsetof(zfs_acl_node_t, z_next));
424 	aclp->z_version = vers;
425 	if (vers == ZFS_ACL_VERSION_FUID)
426 		aclp->z_ops = zfs_acl_fuid_ops;
427 	else
428 		aclp->z_ops = zfs_acl_v0_ops;
429 	return (aclp);
430 }
431 
432 zfs_acl_node_t *
433 zfs_acl_node_alloc(size_t bytes)
434 {
435 	zfs_acl_node_t *aclnode;
436 
437 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
438 	if (bytes) {
439 		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
440 		aclnode->z_allocdata = aclnode->z_acldata;
441 		aclnode->z_allocsize = bytes;
442 		aclnode->z_size = bytes;
443 	}
444 
445 	return (aclnode);
446 }
447 
448 static void
449 zfs_acl_node_free(zfs_acl_node_t *aclnode)
450 {
451 	if (aclnode->z_allocsize)
452 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
453 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
454 }
455 
456 static void
457 zfs_acl_release_nodes(zfs_acl_t *aclp)
458 {
459 	zfs_acl_node_t *aclnode;
460 
461 	while (aclnode = list_head(&aclp->z_acl)) {
462 		list_remove(&aclp->z_acl, aclnode);
463 		zfs_acl_node_free(aclnode);
464 	}
465 	aclp->z_acl_count = 0;
466 	aclp->z_acl_bytes = 0;
467 }
468 
469 void
470 zfs_acl_free(zfs_acl_t *aclp)
471 {
472 	zfs_acl_release_nodes(aclp);
473 	list_destroy(&aclp->z_acl);
474 	kmem_free(aclp, sizeof (zfs_acl_t));
475 }
476 
477 static boolean_t
478 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
479 {
480 	uint16_t entry_type;
481 
482 	switch (type) {
483 	case ALLOW:
484 	case DENY:
485 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
486 	case ACE_SYSTEM_ALARM_ACE_TYPE:
487 		entry_type = flags & ACE_TYPE_FLAGS;
488 		return (entry_type == ACE_OWNER ||
489 		    entry_type == OWNING_GROUP ||
490 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
491 		    entry_type == ACE_IDENTIFIER_GROUP);
492 	default:
493 		if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
494 			return (B_TRUE);
495 	}
496 	return (B_FALSE);
497 }
498 
499 static boolean_t
500 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
501 {
502 	/*
503 	 * first check type of entry
504 	 */
505 
506 	if (!zfs_acl_valid_ace_type(type, iflags))
507 		return (B_FALSE);
508 
509 	switch (type) {
510 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
511 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
512 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
513 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
514 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
515 			return (B_FALSE);
516 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
517 	}
518 
519 	/*
520 	 * next check inheritance level flags
521 	 */
522 
523 	if (obj_type == VDIR &&
524 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
525 		aclp->z_hints |= ZFS_INHERIT_ACE;
526 
527 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
528 		if ((iflags & (ACE_FILE_INHERIT_ACE|
529 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
530 			return (B_FALSE);
531 		}
532 	}
533 
534 	return (B_TRUE);
535 }
536 
537 static void *
538 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
539     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
540 {
541 	zfs_acl_node_t *aclnode;
542 
543 	ASSERT(aclp);
544 
545 	if (start == NULL) {
546 		aclnode = list_head(&aclp->z_acl);
547 		if (aclnode == NULL)
548 			return (NULL);
549 
550 		aclp->z_next_ace = aclnode->z_acldata;
551 		aclp->z_curr_node = aclnode;
552 		aclnode->z_ace_idx = 0;
553 	}
554 
555 	aclnode = aclp->z_curr_node;
556 
557 	if (aclnode == NULL)
558 		return (NULL);
559 
560 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
561 		aclnode = list_next(&aclp->z_acl, aclnode);
562 		if (aclnode == NULL)
563 			return (NULL);
564 		else {
565 			aclp->z_curr_node = aclnode;
566 			aclnode->z_ace_idx = 0;
567 			aclp->z_next_ace = aclnode->z_acldata;
568 		}
569 	}
570 
571 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
572 		void *acep = aclp->z_next_ace;
573 		size_t ace_size;
574 
575 		/*
576 		 * Make sure we don't overstep our bounds
577 		 */
578 		ace_size = aclp->z_ops.ace_size(acep);
579 
580 		if (((caddr_t)acep + ace_size) >
581 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
582 			return (NULL);
583 		}
584 
585 		*iflags = aclp->z_ops.ace_flags_get(acep);
586 		*type = aclp->z_ops.ace_type_get(acep);
587 		*access_mask = aclp->z_ops.ace_mask_get(acep);
588 		*who = aclp->z_ops.ace_who_get(acep);
589 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
590 		aclnode->z_ace_idx++;
591 
592 		return ((void *)acep);
593 	}
594 	return (NULL);
595 }
596 
597 /*ARGSUSED*/
598 static uint64_t
599 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
600     uint16_t *flags, uint16_t *type, uint32_t *mask)
601 {
602 	zfs_acl_t *aclp = datap;
603 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
604 	uint64_t who;
605 
606 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
607 	    flags, type);
608 	return ((uint64_t)(uintptr_t)acep);
609 }
610 
611 static zfs_acl_node_t *
612 zfs_acl_curr_node(zfs_acl_t *aclp)
613 {
614 	ASSERT(aclp->z_curr_node);
615 	return (aclp->z_curr_node);
616 }
617 
618 /*
619  * Copy ACE to internal ZFS format.
620  * While processing the ACL each ACE will be validated for correctness.
621  * ACE FUIDs will be created later.
622  */
623 int
624 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
625     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
626     zfs_fuid_info_t **fuidp, cred_t *cr)
627 {
628 	int i;
629 	uint16_t entry_type;
630 	zfs_ace_t *aceptr = z_acl;
631 	ace_t *acep = datap;
632 	zfs_object_ace_t *zobjacep;
633 	ace_object_t *aceobjp;
634 
635 	for (i = 0; i != aclcnt; i++) {
636 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
637 		aceptr->z_hdr.z_flags = acep->a_flags;
638 		aceptr->z_hdr.z_type = acep->a_type;
639 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
640 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
641 		    entry_type != ACE_EVERYONE) {
642 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
643 			    cr, (entry_type == 0) ?
644 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
645 		}
646 
647 		/*
648 		 * Make sure ACE is valid
649 		 */
650 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
651 		    aceptr->z_hdr.z_flags) != B_TRUE)
652 			return (EINVAL);
653 
654 		switch (acep->a_type) {
655 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
656 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
657 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
658 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
659 			zobjacep = (zfs_object_ace_t *)aceptr;
660 			aceobjp = (ace_object_t *)acep;
661 
662 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
663 			    sizeof (aceobjp->a_obj_type));
664 			bcopy(aceobjp->a_inherit_obj_type,
665 			    zobjacep->z_inherit_type,
666 			    sizeof (aceobjp->a_inherit_obj_type));
667 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
668 			break;
669 		default:
670 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
671 		}
672 
673 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
674 		    aclp->z_ops.ace_size(aceptr));
675 	}
676 
677 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
678 
679 	return (0);
680 }
681 
682 /*
683  * Copy ZFS ACEs to fixed size ace_t layout
684  */
685 static void
686 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
687     void *datap, int filter)
688 {
689 	uint64_t who;
690 	uint32_t access_mask;
691 	uint16_t iflags, type;
692 	zfs_ace_hdr_t *zacep = NULL;
693 	ace_t *acep = datap;
694 	ace_object_t *objacep;
695 	zfs_object_ace_t *zobjacep;
696 	size_t ace_size;
697 	uint16_t entry_type;
698 
699 	while (zacep = zfs_acl_next_ace(aclp, zacep,
700 	    &who, &access_mask, &iflags, &type)) {
701 
702 		switch (type) {
703 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
704 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
705 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
706 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
707 			if (filter) {
708 				continue;
709 			}
710 			zobjacep = (zfs_object_ace_t *)zacep;
711 			objacep = (ace_object_t *)acep;
712 			bcopy(zobjacep->z_object_type,
713 			    objacep->a_obj_type,
714 			    sizeof (zobjacep->z_object_type));
715 			bcopy(zobjacep->z_inherit_type,
716 			    objacep->a_inherit_obj_type,
717 			    sizeof (zobjacep->z_inherit_type));
718 			ace_size = sizeof (ace_object_t);
719 			break;
720 		default:
721 			ace_size = sizeof (ace_t);
722 			break;
723 		}
724 
725 		entry_type = (iflags & ACE_TYPE_FLAGS);
726 		if ((entry_type != ACE_OWNER &&
727 		    entry_type != OWNING_GROUP &&
728 		    entry_type != ACE_EVERYONE)) {
729 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
730 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
731 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
732 		} else {
733 			acep->a_who = (uid_t)(int64_t)who;
734 		}
735 		acep->a_access_mask = access_mask;
736 		acep->a_flags = iflags;
737 		acep->a_type = type;
738 		acep = (ace_t *)((caddr_t)acep + ace_size);
739 	}
740 }
741 
742 static int
743 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
744     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
745 {
746 	int i;
747 	zfs_oldace_t *aceptr = z_acl;
748 
749 	for (i = 0; i != aclcnt; i++, aceptr++) {
750 		aceptr->z_access_mask = acep[i].a_access_mask;
751 		aceptr->z_type = acep[i].a_type;
752 		aceptr->z_flags = acep[i].a_flags;
753 		aceptr->z_fuid = acep[i].a_who;
754 		/*
755 		 * Make sure ACE is valid
756 		 */
757 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
758 		    aceptr->z_flags) != B_TRUE)
759 			return (EINVAL);
760 	}
761 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
762 	return (0);
763 }
764 
765 /*
766  * convert old ACL format to new
767  */
768 void
769 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
770 {
771 	zfs_oldace_t *oldaclp;
772 	int i;
773 	uint16_t type, iflags;
774 	uint32_t access_mask;
775 	uint64_t who;
776 	void *cookie = NULL;
777 	zfs_acl_node_t *newaclnode;
778 
779 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
780 	/*
781 	 * First create the ACE in a contiguous piece of memory
782 	 * for zfs_copy_ace_2_fuid().
783 	 *
784 	 * We only convert an ACL once, so this won't happen
785 	 * everytime.
786 	 */
787 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
788 	    KM_SLEEP);
789 	i = 0;
790 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
791 	    &access_mask, &iflags, &type)) {
792 		oldaclp[i].z_flags = iflags;
793 		oldaclp[i].z_type = type;
794 		oldaclp[i].z_fuid = who;
795 		oldaclp[i++].z_access_mask = access_mask;
796 	}
797 
798 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
799 	    sizeof (zfs_object_ace_t));
800 	aclp->z_ops = zfs_acl_fuid_ops;
801 	VERIFY(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
802 	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
803 	    &newaclnode->z_size, NULL, cr) == 0);
804 	newaclnode->z_ace_count = aclp->z_acl_count;
805 	aclp->z_version = ZFS_ACL_VERSION;
806 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
807 
808 	/*
809 	 * Release all previous ACL nodes
810 	 */
811 
812 	zfs_acl_release_nodes(aclp);
813 
814 	list_insert_head(&aclp->z_acl, newaclnode);
815 
816 	aclp->z_acl_bytes = newaclnode->z_size;
817 	aclp->z_acl_count = newaclnode->z_ace_count;
818 
819 }
820 
821 /*
822  * Convert unix access mask to v4 access mask
823  */
824 static uint32_t
825 zfs_unix_to_v4(uint32_t access_mask)
826 {
827 	uint32_t new_mask = 0;
828 
829 	if (access_mask & S_IXOTH)
830 		new_mask |= ACE_EXECUTE;
831 	if (access_mask & S_IWOTH)
832 		new_mask |= ACE_WRITE_DATA;
833 	if (access_mask & S_IROTH)
834 		new_mask |= ACE_READ_DATA;
835 	return (new_mask);
836 }
837 
838 static void
839 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
840     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
841 {
842 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
843 
844 	aclp->z_ops.ace_mask_set(acep, access_mask);
845 	aclp->z_ops.ace_type_set(acep, access_type);
846 	aclp->z_ops.ace_flags_set(acep, entry_type);
847 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
848 	    type != ACE_EVERYONE))
849 		aclp->z_ops.ace_who_set(acep, fuid);
850 }
851 
852 /*
853  * Determine mode of file based on ACL.
854  * Also, create FUIDs for any User/Group ACEs
855  */
856 uint64_t
857 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp, uint64_t *pflags)
858 {
859 	int		entry_type;
860 	mode_t		mode;
861 	mode_t		seen = 0;
862 	zfs_ace_hdr_t 	*acep = NULL;
863 	uint64_t	who;
864 	uint16_t	iflags, type;
865 	uint32_t	access_mask;
866 	boolean_t	an_exec_denied = B_FALSE;
867 
868 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
869 
870 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
871 	    &access_mask, &iflags, &type)) {
872 
873 		if (!zfs_acl_valid_ace_type(type, iflags))
874 			continue;
875 
876 		entry_type = (iflags & ACE_TYPE_FLAGS);
877 
878 		/*
879 		 * Skip over owner@, group@ or everyone@ inherit only ACEs
880 		 */
881 		if ((iflags & ACE_INHERIT_ONLY_ACE) &&
882 		    (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
883 		    entry_type == OWNING_GROUP))
884 			continue;
885 
886 		if (entry_type == ACE_OWNER) {
887 			if ((access_mask & ACE_READ_DATA) &&
888 			    (!(seen & S_IRUSR))) {
889 				seen |= S_IRUSR;
890 				if (type == ALLOW) {
891 					mode |= S_IRUSR;
892 				}
893 			}
894 			if ((access_mask & ACE_WRITE_DATA) &&
895 			    (!(seen & S_IWUSR))) {
896 				seen |= S_IWUSR;
897 				if (type == ALLOW) {
898 					mode |= S_IWUSR;
899 				}
900 			}
901 			if ((access_mask & ACE_EXECUTE) &&
902 			    (!(seen & S_IXUSR))) {
903 				seen |= S_IXUSR;
904 				if (type == ALLOW) {
905 					mode |= S_IXUSR;
906 				}
907 			}
908 		} else if (entry_type == OWNING_GROUP) {
909 			if ((access_mask & ACE_READ_DATA) &&
910 			    (!(seen & S_IRGRP))) {
911 				seen |= S_IRGRP;
912 				if (type == ALLOW) {
913 					mode |= S_IRGRP;
914 				}
915 			}
916 			if ((access_mask & ACE_WRITE_DATA) &&
917 			    (!(seen & S_IWGRP))) {
918 				seen |= S_IWGRP;
919 				if (type == ALLOW) {
920 					mode |= S_IWGRP;
921 				}
922 			}
923 			if ((access_mask & ACE_EXECUTE) &&
924 			    (!(seen & S_IXGRP))) {
925 				seen |= S_IXGRP;
926 				if (type == ALLOW) {
927 					mode |= S_IXGRP;
928 				}
929 			}
930 		} else if (entry_type == ACE_EVERYONE) {
931 			if ((access_mask & ACE_READ_DATA)) {
932 				if (!(seen & S_IRUSR)) {
933 					seen |= S_IRUSR;
934 					if (type == ALLOW) {
935 						mode |= S_IRUSR;
936 					}
937 				}
938 				if (!(seen & S_IRGRP)) {
939 					seen |= S_IRGRP;
940 					if (type == ALLOW) {
941 						mode |= S_IRGRP;
942 					}
943 				}
944 				if (!(seen & S_IROTH)) {
945 					seen |= S_IROTH;
946 					if (type == ALLOW) {
947 						mode |= S_IROTH;
948 					}
949 				}
950 			}
951 			if ((access_mask & ACE_WRITE_DATA)) {
952 				if (!(seen & S_IWUSR)) {
953 					seen |= S_IWUSR;
954 					if (type == ALLOW) {
955 						mode |= S_IWUSR;
956 					}
957 				}
958 				if (!(seen & S_IWGRP)) {
959 					seen |= S_IWGRP;
960 					if (type == ALLOW) {
961 						mode |= S_IWGRP;
962 					}
963 				}
964 				if (!(seen & S_IWOTH)) {
965 					seen |= S_IWOTH;
966 					if (type == ALLOW) {
967 						mode |= S_IWOTH;
968 					}
969 				}
970 			}
971 			if ((access_mask & ACE_EXECUTE)) {
972 				if (!(seen & S_IXUSR)) {
973 					seen |= S_IXUSR;
974 					if (type == ALLOW) {
975 						mode |= S_IXUSR;
976 					}
977 				}
978 				if (!(seen & S_IXGRP)) {
979 					seen |= S_IXGRP;
980 					if (type == ALLOW) {
981 						mode |= S_IXGRP;
982 					}
983 				}
984 				if (!(seen & S_IXOTH)) {
985 					seen |= S_IXOTH;
986 					if (type == ALLOW) {
987 						mode |= S_IXOTH;
988 					}
989 				}
990 			}
991 		} else {
992 			/*
993 			 * Only care if this IDENTIFIER_GROUP or
994 			 * USER ACE denies execute access to someone,
995 			 * mode is not affected
996 			 */
997 			if ((access_mask & ACE_EXECUTE) && type == DENY)
998 				an_exec_denied = B_TRUE;
999 		}
1000 	}
1001 
1002 	/*
1003 	 * Failure to allow is effectively a deny, so execute permission
1004 	 * is denied if it was never mentioned or if we explicitly
1005 	 * weren't allowed it.
1006 	 */
1007 	if (!an_exec_denied &&
1008 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1009 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1010 		an_exec_denied = B_TRUE;
1011 
1012 	if (an_exec_denied)
1013 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1014 	else
1015 		*pflags |= ZFS_NO_EXECS_DENIED;
1016 
1017 	return (mode);
1018 }
1019 
1020 /*
1021  * Read an external acl object.  If the intent is to modify, always
1022  * create a new acl and leave any cached acl in place.
1023  */
1024 static int
1025 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
1026 {
1027 	zfs_acl_t	*aclp;
1028 	int		aclsize;
1029 	int		acl_count;
1030 	zfs_acl_node_t	*aclnode;
1031 	zfs_acl_phys_t	znode_acl;
1032 	int		version;
1033 	int		error;
1034 
1035 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1036 
1037 	if (zp->z_acl_cached && !will_modify) {
1038 		*aclpp = zp->z_acl_cached;
1039 		return (0);
1040 	}
1041 
1042 	version = ZNODE_ACL_VERSION(zp);
1043 
1044 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1045 	    &acl_count, &znode_acl)) != 0)
1046 		return (error);
1047 
1048 	aclp = zfs_acl_alloc(version);
1049 
1050 	aclp->z_acl_count = acl_count;
1051 	aclp->z_acl_bytes = aclsize;
1052 
1053 	aclnode = zfs_acl_node_alloc(aclsize);
1054 	aclnode->z_ace_count = aclp->z_acl_count;
1055 	aclnode->z_size = aclsize;
1056 
1057 	if (!zp->z_is_sa) {
1058 		if (znode_acl.z_acl_extern_obj) {
1059 			error = dmu_read(zp->z_zfsvfs->z_os,
1060 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1061 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1062 		} else {
1063 			bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1064 			    aclnode->z_size);
1065 		}
1066 	} else {
1067 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1068 		    aclnode->z_acldata, aclnode->z_size);
1069 	}
1070 
1071 	if (error != 0) {
1072 		zfs_acl_free(aclp);
1073 		zfs_acl_node_free(aclnode);
1074 		/* convert checksum errors into IO errors */
1075 		if (error == ECKSUM)
1076 			error = EIO;
1077 		return (error);
1078 	}
1079 
1080 	list_insert_head(&aclp->z_acl, aclnode);
1081 
1082 	*aclpp = aclp;
1083 	if (!will_modify)
1084 		zp->z_acl_cached = aclp;
1085 	return (0);
1086 }
1087 
1088 /*ARGSUSED*/
1089 void
1090 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1091     boolean_t start, void *userdata)
1092 {
1093 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1094 
1095 	if (start) {
1096 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1097 	} else {
1098 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1099 		    cb->cb_acl_node);
1100 	}
1101 	*dataptr = cb->cb_acl_node->z_acldata;
1102 	*length = cb->cb_acl_node->z_size;
1103 }
1104 
1105 /*
1106  * common code for setting ACLs.
1107  *
1108  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1109  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1110  * already checked the acl and knows whether to inherit.
1111  */
1112 int
1113 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1114 {
1115 	int			error;
1116 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1117 	dmu_object_type_t	otype;
1118 	zfs_acl_locator_cb_t	locate = { 0 };
1119 	uint64_t		mode;
1120 	sa_bulk_attr_t		bulk[5];
1121 	uint64_t		ctime[2];
1122 	int			count = 0;
1123 
1124 	mode = zp->z_mode;
1125 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags);
1126 
1127 	zp->z_mode = mode;
1128 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1129 	    &mode, sizeof (mode));
1130 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1131 	    &zp->z_pflags, sizeof (zp->z_pflags));
1132 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1133 	    &ctime, sizeof (ctime));
1134 
1135 	if (zp->z_acl_cached) {
1136 		zfs_acl_free(zp->z_acl_cached);
1137 		zp->z_acl_cached = NULL;
1138 	}
1139 
1140 	/*
1141 	 * Upgrade needed?
1142 	 */
1143 	if (!zfsvfs->z_use_fuids) {
1144 		otype = DMU_OT_OLDACL;
1145 	} else {
1146 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1147 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1148 			zfs_acl_xform(zp, aclp, cr);
1149 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1150 		otype = DMU_OT_ACL;
1151 	}
1152 
1153 	/*
1154 	 * Arrgh, we have to handle old on disk format
1155 	 * as well as newer (preferred) SA format.
1156 	 */
1157 
1158 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1159 		locate.cb_aclp = aclp;
1160 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1161 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1162 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1163 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1164 	} else { /* Painful legacy way */
1165 		zfs_acl_node_t *aclnode;
1166 		uint64_t off = 0;
1167 		zfs_acl_phys_t acl_phys;
1168 		uint64_t aoid;
1169 
1170 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1171 		    &acl_phys, sizeof (acl_phys))) != 0)
1172 			return (error);
1173 
1174 		aoid = acl_phys.z_acl_extern_obj;
1175 
1176 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1177 			/*
1178 			 * If ACL was previously external and we are now
1179 			 * converting to new ACL format then release old
1180 			 * ACL object and create a new one.
1181 			 */
1182 			if (aoid &&
1183 			    aclp->z_version != acl_phys.z_acl_version) {
1184 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1185 				if (error)
1186 					return (error);
1187 				aoid = 0;
1188 			}
1189 			if (aoid == 0) {
1190 				aoid = dmu_object_alloc(zfsvfs->z_os,
1191 				    otype, aclp->z_acl_bytes,
1192 				    otype == DMU_OT_ACL ?
1193 				    DMU_OT_SYSACL : DMU_OT_NONE,
1194 				    otype == DMU_OT_ACL ?
1195 				    DN_MAX_BONUSLEN : 0, tx);
1196 			} else {
1197 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1198 				    aoid, aclp->z_acl_bytes, 0, tx);
1199 			}
1200 			acl_phys.z_acl_extern_obj = aoid;
1201 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1202 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1203 				if (aclnode->z_ace_count == 0)
1204 					continue;
1205 				dmu_write(zfsvfs->z_os, aoid, off,
1206 				    aclnode->z_size, aclnode->z_acldata, tx);
1207 				off += aclnode->z_size;
1208 			}
1209 		} else {
1210 			void *start = acl_phys.z_ace_data;
1211 			/*
1212 			 * Migrating back embedded?
1213 			 */
1214 			if (acl_phys.z_acl_extern_obj) {
1215 				error = dmu_object_free(zfsvfs->z_os,
1216 				    acl_phys.z_acl_extern_obj, tx);
1217 				if (error)
1218 					return (error);
1219 				acl_phys.z_acl_extern_obj = 0;
1220 			}
1221 
1222 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1223 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1224 				if (aclnode->z_ace_count == 0)
1225 					continue;
1226 				bcopy(aclnode->z_acldata, start,
1227 				    aclnode->z_size);
1228 				start = (caddr_t)start + aclnode->z_size;
1229 			}
1230 		}
1231 		/*
1232 		 * If Old version then swap count/bytes to match old
1233 		 * layout of znode_acl_phys_t.
1234 		 */
1235 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1236 			acl_phys.z_acl_size = aclp->z_acl_count;
1237 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1238 		} else {
1239 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1240 			acl_phys.z_acl_count = aclp->z_acl_count;
1241 		}
1242 		acl_phys.z_acl_version = aclp->z_version;
1243 
1244 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1245 		    &acl_phys, sizeof (acl_phys));
1246 	}
1247 
1248 	/*
1249 	 * Replace ACL wide bits, but first clear them.
1250 	 */
1251 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1252 
1253 	zp->z_pflags |= aclp->z_hints;
1254 
1255 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1256 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1257 
1258 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime, B_TRUE);
1259 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1260 }
1261 
1262 /*
1263  * Update access mask for prepended ACE
1264  *
1265  * This applies the "groupmask" value for aclmode property.
1266  */
1267 static void
1268 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1269     mode_t mode, uint64_t owner)
1270 {
1271 	int	rmask, wmask, xmask;
1272 	int	user_ace;
1273 	uint16_t aceflags;
1274 	uint32_t origmask, acepmask;
1275 	uint64_t fuid;
1276 
1277 	aceflags = aclp->z_ops.ace_flags_get(acep);
1278 	fuid = aclp->z_ops.ace_who_get(acep);
1279 	origmask = aclp->z_ops.ace_mask_get(origacep);
1280 	acepmask = aclp->z_ops.ace_mask_get(acep);
1281 
1282 	user_ace = (!(aceflags &
1283 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1284 
1285 	if (user_ace && (fuid == owner)) {
1286 		rmask = S_IRUSR;
1287 		wmask = S_IWUSR;
1288 		xmask = S_IXUSR;
1289 	} else {
1290 		rmask = S_IRGRP;
1291 		wmask = S_IWGRP;
1292 		xmask = S_IXGRP;
1293 	}
1294 
1295 	if (origmask & ACE_READ_DATA) {
1296 		if (mode & rmask) {
1297 			acepmask &= ~ACE_READ_DATA;
1298 		} else {
1299 			acepmask |= ACE_READ_DATA;
1300 		}
1301 	}
1302 
1303 	if (origmask & ACE_WRITE_DATA) {
1304 		if (mode & wmask) {
1305 			acepmask &= ~ACE_WRITE_DATA;
1306 		} else {
1307 			acepmask |= ACE_WRITE_DATA;
1308 		}
1309 	}
1310 
1311 	if (origmask & ACE_APPEND_DATA) {
1312 		if (mode & wmask) {
1313 			acepmask &= ~ACE_APPEND_DATA;
1314 		} else {
1315 			acepmask |= ACE_APPEND_DATA;
1316 		}
1317 	}
1318 
1319 	if (origmask & ACE_EXECUTE) {
1320 		if (mode & xmask) {
1321 			acepmask &= ~ACE_EXECUTE;
1322 		} else {
1323 			acepmask |= ACE_EXECUTE;
1324 		}
1325 	}
1326 	aclp->z_ops.ace_mask_set(acep, acepmask);
1327 }
1328 
1329 /*
1330  * Apply mode to canonical six ACEs.
1331  */
1332 static void
1333 zfs_acl_fixup_canonical_six(zfs_acl_t *aclp, mode_t mode)
1334 {
1335 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1336 	void	*acep;
1337 	int	maskoff = aclp->z_ops.ace_mask_off();
1338 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1339 
1340 	ASSERT(aclnode != NULL);
1341 
1342 	acep = (void *)((caddr_t)aclnode->z_acldata +
1343 	    aclnode->z_size - (abstract_size * 6));
1344 
1345 	/*
1346 	 * Fixup final ACEs to match the mode
1347 	 */
1348 
1349 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1350 	    (mode & 0700) >> 6);	/* owner@ */
1351 
1352 	acep = (caddr_t)acep + (abstract_size * 2);
1353 
1354 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1355 	    (mode & 0070) >> 3);	/* group@ */
1356 
1357 	acep = (caddr_t)acep + (abstract_size * 2);
1358 	adjust_ace_pair_common(acep, maskoff,
1359 	    abstract_size, mode);	/* everyone@ */
1360 }
1361 
1362 
1363 static int
1364 zfs_acl_ace_match(zfs_acl_t *aclp, void *acep, int allow_deny,
1365     int entry_type, int accessmask)
1366 {
1367 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1368 	uint16_t type = aclp->z_ops.ace_type_get(acep);
1369 	uint16_t flags = aclp->z_ops.ace_flags_get(acep);
1370 
1371 	return (mask == accessmask && type == allow_deny &&
1372 	    ((flags & ACE_TYPE_FLAGS) == entry_type));
1373 }
1374 
1375 /*
1376  * Can prepended ACE be reused?
1377  */
1378 static int
1379 zfs_reuse_deny(zfs_acl_t *aclp, void *acep, void *prevacep)
1380 {
1381 	int okay_masks;
1382 	uint16_t prevtype;
1383 	uint16_t prevflags;
1384 	uint16_t flags;
1385 	uint32_t mask, prevmask;
1386 
1387 	if (prevacep == NULL)
1388 		return (B_FALSE);
1389 
1390 	prevtype = aclp->z_ops.ace_type_get(prevacep);
1391 	prevflags = aclp->z_ops.ace_flags_get(prevacep);
1392 	flags = aclp->z_ops.ace_flags_get(acep);
1393 	mask = aclp->z_ops.ace_mask_get(acep);
1394 	prevmask = aclp->z_ops.ace_mask_get(prevacep);
1395 
1396 	if (prevtype != DENY)
1397 		return (B_FALSE);
1398 
1399 	if (prevflags != (flags & ACE_IDENTIFIER_GROUP))
1400 		return (B_FALSE);
1401 
1402 	okay_masks = (mask & OKAY_MASK_BITS);
1403 
1404 	if (prevmask & ~okay_masks)
1405 		return (B_FALSE);
1406 
1407 	return (B_TRUE);
1408 }
1409 
1410 
1411 /*
1412  * Insert new ACL node into chain of zfs_acl_node_t's
1413  *
1414  * This will result in two possible results.
1415  * 1. If the ACL is currently just a single zfs_acl_node and
1416  *    we are prepending the entry then current acl node will have
1417  *    a new node inserted above it.
1418  *
1419  * 2. If we are inserting in the middle of current acl node then
1420  *    the current node will be split in two and new node will be inserted
1421  *    in between the two split nodes.
1422  */
1423 static zfs_acl_node_t *
1424 zfs_acl_ace_insert(zfs_acl_t *aclp, void  *acep)
1425 {
1426 	zfs_acl_node_t 	*newnode;
1427 	zfs_acl_node_t 	*trailernode = NULL;
1428 	zfs_acl_node_t 	*currnode = zfs_acl_curr_node(aclp);
1429 	int		curr_idx = aclp->z_curr_node->z_ace_idx;
1430 	int		trailer_count;
1431 	size_t		oldsize;
1432 
1433 	newnode = zfs_acl_node_alloc(aclp->z_ops.ace_size(acep));
1434 	newnode->z_ace_count = 1;
1435 
1436 	oldsize = currnode->z_size;
1437 
1438 	if (curr_idx != 1) {
1439 		trailernode = zfs_acl_node_alloc(0);
1440 		trailernode->z_acldata = acep;
1441 
1442 		trailer_count = currnode->z_ace_count - curr_idx + 1;
1443 		currnode->z_ace_count = curr_idx - 1;
1444 		currnode->z_size = (caddr_t)acep - (caddr_t)currnode->z_acldata;
1445 		trailernode->z_size = oldsize - currnode->z_size;
1446 		trailernode->z_ace_count = trailer_count;
1447 	}
1448 
1449 	aclp->z_acl_count += 1;
1450 	aclp->z_acl_bytes += aclp->z_ops.ace_size(acep);
1451 
1452 	if (curr_idx == 1)
1453 		list_insert_before(&aclp->z_acl, currnode, newnode);
1454 	else
1455 		list_insert_after(&aclp->z_acl, currnode, newnode);
1456 	if (trailernode) {
1457 		list_insert_after(&aclp->z_acl, newnode, trailernode);
1458 		aclp->z_curr_node = trailernode;
1459 		trailernode->z_ace_idx = 1;
1460 	}
1461 
1462 	return (newnode);
1463 }
1464 
1465 /*
1466  * Prepend deny ACE
1467  */
1468 static void *
1469 zfs_acl_prepend_deny(uint64_t uid, zfs_acl_t *aclp, void *acep,
1470     mode_t mode)
1471 {
1472 	zfs_acl_node_t *aclnode;
1473 	void  *newacep;
1474 	uint64_t fuid;
1475 	uint16_t flags;
1476 
1477 	aclnode = zfs_acl_ace_insert(aclp, acep);
1478 	newacep = aclnode->z_acldata;
1479 	fuid = aclp->z_ops.ace_who_get(acep);
1480 	flags = aclp->z_ops.ace_flags_get(acep);
1481 	zfs_set_ace(aclp, newacep, 0, DENY, fuid, (flags & ACE_TYPE_FLAGS));
1482 	zfs_acl_prepend_fixup(aclp, newacep, acep, mode, uid);
1483 
1484 	return (newacep);
1485 }
1486 
1487 /*
1488  * Split an inherited ACE into inherit_only ACE
1489  * and original ACE with inheritance flags stripped off.
1490  */
1491 static void
1492 zfs_acl_split_ace(zfs_acl_t *aclp, zfs_ace_hdr_t *acep)
1493 {
1494 	zfs_acl_node_t *aclnode;
1495 	zfs_acl_node_t *currnode;
1496 	void  *newacep;
1497 	uint16_t type, flags;
1498 	uint32_t mask;
1499 	uint64_t fuid;
1500 
1501 	type = aclp->z_ops.ace_type_get(acep);
1502 	flags = aclp->z_ops.ace_flags_get(acep);
1503 	mask = aclp->z_ops.ace_mask_get(acep);
1504 	fuid = aclp->z_ops.ace_who_get(acep);
1505 
1506 	aclnode = zfs_acl_ace_insert(aclp, acep);
1507 	newacep = aclnode->z_acldata;
1508 
1509 	aclp->z_ops.ace_type_set(newacep, type);
1510 	aclp->z_ops.ace_flags_set(newacep, flags | ACE_INHERIT_ONLY_ACE);
1511 	aclp->z_ops.ace_mask_set(newacep, mask);
1512 	aclp->z_ops.ace_type_set(newacep, type);
1513 	aclp->z_ops.ace_who_set(newacep, fuid);
1514 	aclp->z_next_ace = acep;
1515 	flags &= ~ALL_INHERIT;
1516 	aclp->z_ops.ace_flags_set(acep, flags);
1517 	currnode = zfs_acl_curr_node(aclp);
1518 	ASSERT(currnode->z_ace_idx >= 1);
1519 	currnode->z_ace_idx -= 1;
1520 }
1521 
1522 /*
1523  * Are ACES started at index i, the canonical six ACES?
1524  */
1525 static int
1526 zfs_have_canonical_six(zfs_acl_t *aclp)
1527 {
1528 	void *acep;
1529 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1530 	int		i = 0;
1531 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1532 
1533 	ASSERT(aclnode != NULL);
1534 
1535 	if (aclnode->z_ace_count < 6)
1536 		return (0);
1537 
1538 	acep = (void *)((caddr_t)aclnode->z_acldata +
1539 	    aclnode->z_size - (aclp->z_ops.ace_abstract_size() * 6));
1540 
1541 	if ((zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1542 	    DENY, ACE_OWNER, 0) &&
1543 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1544 	    ALLOW, ACE_OWNER, OWNER_ALLOW_MASK) &&
1545 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++), DENY,
1546 	    OWNING_GROUP, 0) && zfs_acl_ace_match(aclp, (caddr_t)acep +
1547 	    (abstract_size * i++),
1548 	    ALLOW, OWNING_GROUP, 0) &&
1549 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1550 	    DENY, ACE_EVERYONE, EVERYONE_DENY_MASK) &&
1551 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1552 	    ALLOW, ACE_EVERYONE, EVERYONE_ALLOW_MASK))) {
1553 		return (1);
1554 	} else {
1555 		return (0);
1556 	}
1557 }
1558 
1559 
1560 /*
1561  * Apply step 1g, to group entries
1562  *
1563  * Need to deal with corner case where group may have
1564  * greater permissions than owner.  If so then limit
1565  * group permissions, based on what extra permissions
1566  * group has.
1567  */
1568 static void
1569 zfs_fixup_group_entries(zfs_acl_t *aclp, void *acep, void *prevacep,
1570     mode_t mode)
1571 {
1572 	uint32_t prevmask = aclp->z_ops.ace_mask_get(prevacep);
1573 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1574 	uint16_t prevflags = aclp->z_ops.ace_flags_get(prevacep);
1575 	mode_t extramode = (mode >> 3) & 07;
1576 	mode_t ownermode = (mode >> 6);
1577 
1578 	if (prevflags & ACE_IDENTIFIER_GROUP) {
1579 
1580 		extramode &= ~ownermode;
1581 
1582 		if (extramode) {
1583 			if (extramode & S_IROTH) {
1584 				prevmask &= ~ACE_READ_DATA;
1585 				mask &= ~ACE_READ_DATA;
1586 			}
1587 			if (extramode & S_IWOTH) {
1588 				prevmask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1589 				mask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1590 			}
1591 			if (extramode & S_IXOTH) {
1592 				prevmask  &= ~ACE_EXECUTE;
1593 				mask &= ~ACE_EXECUTE;
1594 			}
1595 		}
1596 	}
1597 	aclp->z_ops.ace_mask_set(acep, mask);
1598 	aclp->z_ops.ace_mask_set(prevacep, prevmask);
1599 }
1600 
1601 /*
1602  * Apply the chmod algorithm as described
1603  * in PSARC/2002/240
1604  */
1605 static void
1606 zfs_acl_chmod(zfsvfs_t *zfsvfs, uint64_t uid,
1607     uint64_t mode, zfs_acl_t *aclp)
1608 {
1609 	void		*acep = NULL, *prevacep = NULL;
1610 	uint64_t	who;
1611 	int 		i;
1612 	int 		entry_type;
1613 	int 		reuse_deny;
1614 	int 		need_canonical_six = 1;
1615 	uint16_t	iflags, type;
1616 	uint32_t	access_mask;
1617 
1618 	/*
1619 	 * If discard then just discard all ACL nodes which
1620 	 * represent the ACEs.
1621 	 *
1622 	 * New owner@/group@/everone@ ACEs will be added
1623 	 * later.
1624 	 */
1625 	if (zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1626 		zfs_acl_release_nodes(aclp);
1627 
1628 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1629 	    &iflags, &type)) {
1630 
1631 		entry_type = (iflags & ACE_TYPE_FLAGS);
1632 		iflags = (iflags & ALL_INHERIT);
1633 
1634 		if ((type != ALLOW && type != DENY) ||
1635 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1636 			if (iflags)
1637 				aclp->z_hints |= ZFS_INHERIT_ACE;
1638 			switch (type) {
1639 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1640 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1641 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1642 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1643 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1644 				break;
1645 			}
1646 			goto nextace;
1647 		}
1648 
1649 		/*
1650 		 * Need to split ace into two?
1651 		 */
1652 		if ((iflags & (ACE_FILE_INHERIT_ACE|
1653 		    ACE_DIRECTORY_INHERIT_ACE)) &&
1654 		    (!(iflags & ACE_INHERIT_ONLY_ACE))) {
1655 			zfs_acl_split_ace(aclp, acep);
1656 			aclp->z_hints |= ZFS_INHERIT_ACE;
1657 			goto nextace;
1658 		}
1659 
1660 		if (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1661 		    (entry_type == OWNING_GROUP)) {
1662 			access_mask &= ~OGE_CLEAR;
1663 			aclp->z_ops.ace_mask_set(acep, access_mask);
1664 			goto nextace;
1665 		} else {
1666 			reuse_deny = B_TRUE;
1667 			if (type == ALLOW) {
1668 
1669 				/*
1670 				 * Check preceding ACE if any, to see
1671 				 * if we need to prepend a DENY ACE.
1672 				 * This is only applicable when the acl_mode
1673 				 * property == groupmask.
1674 				 */
1675 				if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK) {
1676 
1677 					reuse_deny = zfs_reuse_deny(aclp, acep,
1678 					    prevacep);
1679 
1680 					if (!reuse_deny) {
1681 						prevacep =
1682 						    zfs_acl_prepend_deny(uid,
1683 						    aclp, acep, mode);
1684 					} else {
1685 						zfs_acl_prepend_fixup(
1686 						    aclp, prevacep,
1687 						    acep, mode, uid);
1688 					}
1689 					zfs_fixup_group_entries(aclp, acep,
1690 					    prevacep, mode);
1691 				}
1692 			}
1693 		}
1694 nextace:
1695 		prevacep = acep;
1696 	}
1697 
1698 	/*
1699 	 * Check out last six aces, if we have six.
1700 	 */
1701 
1702 	if (aclp->z_acl_count >= 6) {
1703 		if (zfs_have_canonical_six(aclp)) {
1704 			need_canonical_six = 0;
1705 		}
1706 	}
1707 
1708 	if (need_canonical_six) {
1709 		size_t abstract_size = aclp->z_ops.ace_abstract_size();
1710 		void *zacep;
1711 		zfs_acl_node_t *aclnode =
1712 		    zfs_acl_node_alloc(abstract_size * 6);
1713 
1714 		aclnode->z_size = abstract_size * 6;
1715 		aclnode->z_ace_count = 6;
1716 		aclp->z_acl_bytes += aclnode->z_size;
1717 		list_insert_tail(&aclp->z_acl, aclnode);
1718 
1719 		zacep = aclnode->z_acldata;
1720 
1721 		i = 0;
1722 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1723 		    0, DENY, -1, ACE_OWNER);
1724 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1725 		    OWNER_ALLOW_MASK, ALLOW, -1, ACE_OWNER);
1726 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1727 		    DENY, -1, OWNING_GROUP);
1728 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1729 		    ALLOW, -1, OWNING_GROUP);
1730 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1731 		    EVERYONE_DENY_MASK, DENY, -1, ACE_EVERYONE);
1732 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1733 		    EVERYONE_ALLOW_MASK, ALLOW, -1, ACE_EVERYONE);
1734 		aclp->z_acl_count += 6;
1735 	}
1736 
1737 	zfs_acl_fixup_canonical_six(aclp, mode);
1738 }
1739 
1740 int
1741 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1742 {
1743 	int error;
1744 
1745 	mutex_enter(&zp->z_lock);
1746 	mutex_enter(&zp->z_acl_lock);
1747 
1748 	*aclp = NULL;
1749 	error = zfs_acl_node_read(zp, aclp, B_TRUE);
1750 	if (error == 0) {
1751 		uint64_t owner;
1752 		if (IS_EPHEMERAL(zp->z_uid))
1753 			if ((error = sa_lookup(zp->z_sa_hdl,
1754 			    SA_ZPL_UID(zp->z_zfsvfs),
1755 			    &owner, sizeof (owner))) != 0)
1756 				return (error);
1757 		else
1758 			owner = (uint64_t)zp->z_uid;
1759 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1760 		zfs_acl_chmod(zp->z_zfsvfs, owner, mode, *aclp);
1761 	}
1762 	mutex_exit(&zp->z_acl_lock);
1763 	mutex_exit(&zp->z_lock);
1764 	return (error);
1765 }
1766 
1767 /*
1768  * strip off write_owner and write_acl
1769  */
1770 static void
1771 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1772 {
1773 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1774 
1775 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1776 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1777 		mask &= ~RESTRICTED_CLEAR;
1778 		aclp->z_ops.ace_mask_set(acep, mask);
1779 	}
1780 }
1781 
1782 /*
1783  * Should ACE be inherited?
1784  */
1785 static int
1786 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1787 {
1788 	int	iflags = (acep_flags & 0xf);
1789 
1790 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1791 		return (1);
1792 	else if (iflags & ACE_FILE_INHERIT_ACE)
1793 		return (!((vtype == VDIR) &&
1794 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1795 	return (0);
1796 }
1797 
1798 /*
1799  * inherit inheritable ACEs from parent
1800  */
1801 static zfs_acl_t *
1802 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1803     uint64_t mode, boolean_t *need_chmod)
1804 {
1805 	void		*pacep;
1806 	void		*acep, *acep2;
1807 	zfs_acl_node_t  *aclnode, *aclnode2;
1808 	zfs_acl_t	*aclp = NULL;
1809 	uint64_t	who;
1810 	uint32_t	access_mask;
1811 	uint16_t	iflags, newflags, type;
1812 	size_t		ace_size;
1813 	void		*data1, *data2;
1814 	size_t		data1sz, data2sz;
1815 	boolean_t	vdir = vtype == VDIR;
1816 	boolean_t	vreg = vtype == VREG;
1817 	boolean_t	passthrough, passthrough_x, noallow;
1818 
1819 	passthrough_x =
1820 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1821 	passthrough = passthrough_x ||
1822 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1823 	noallow =
1824 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1825 
1826 	*need_chmod = B_TRUE;
1827 	pacep = NULL;
1828 	aclp = zfs_acl_alloc(paclp->z_version);
1829 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD || vtype == VLNK)
1830 		return (aclp);
1831 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1832 	    &access_mask, &iflags, &type)) {
1833 
1834 		/*
1835 		 * don't inherit bogus ACEs
1836 		 */
1837 		if (!zfs_acl_valid_ace_type(type, iflags))
1838 			continue;
1839 
1840 		if (noallow && type == ALLOW)
1841 			continue;
1842 
1843 		ace_size = aclp->z_ops.ace_size(pacep);
1844 
1845 		if (!zfs_ace_can_use(vtype, iflags))
1846 			continue;
1847 
1848 		/*
1849 		 * If owner@, group@, or everyone@ inheritable
1850 		 * then zfs_acl_chmod() isn't needed.
1851 		 */
1852 		if (passthrough &&
1853 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1854 		    ((iflags & OWNING_GROUP) ==
1855 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1856 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1857 			*need_chmod = B_FALSE;
1858 		}
1859 
1860 		if (!vdir && passthrough_x &&
1861 		    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1862 			access_mask &= ~ACE_EXECUTE;
1863 		}
1864 
1865 		aclnode = zfs_acl_node_alloc(ace_size);
1866 		list_insert_tail(&aclp->z_acl, aclnode);
1867 		acep = aclnode->z_acldata;
1868 
1869 		zfs_set_ace(aclp, acep, access_mask, type,
1870 		    who, iflags|ACE_INHERITED_ACE);
1871 
1872 		/*
1873 		 * Copy special opaque data if any
1874 		 */
1875 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1876 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1877 			    &data2)) == data1sz);
1878 			bcopy(data1, data2, data2sz);
1879 		}
1880 		aclp->z_acl_count++;
1881 		aclnode->z_ace_count++;
1882 		aclp->z_acl_bytes += aclnode->z_size;
1883 		newflags = aclp->z_ops.ace_flags_get(acep);
1884 
1885 		if (vdir)
1886 			aclp->z_hints |= ZFS_INHERIT_ACE;
1887 
1888 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1889 			newflags &= ~ALL_INHERIT;
1890 			aclp->z_ops.ace_flags_set(acep,
1891 			    newflags|ACE_INHERITED_ACE);
1892 			zfs_restricted_update(zfsvfs, aclp, acep);
1893 			continue;
1894 		}
1895 
1896 		ASSERT(vdir);
1897 
1898 		newflags = aclp->z_ops.ace_flags_get(acep);
1899 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1900 		    ACE_DIRECTORY_INHERIT_ACE)) !=
1901 		    ACE_FILE_INHERIT_ACE) {
1902 			aclnode2 = zfs_acl_node_alloc(ace_size);
1903 			list_insert_tail(&aclp->z_acl, aclnode2);
1904 			acep2 = aclnode2->z_acldata;
1905 			zfs_set_ace(aclp, acep2,
1906 			    access_mask, type, who,
1907 			    iflags|ACE_INHERITED_ACE);
1908 			newflags |= ACE_INHERIT_ONLY_ACE;
1909 			aclp->z_ops.ace_flags_set(acep, newflags);
1910 			newflags &= ~ALL_INHERIT;
1911 			aclp->z_ops.ace_flags_set(acep2,
1912 			    newflags|ACE_INHERITED_ACE);
1913 
1914 			/*
1915 			 * Copy special opaque data if any
1916 			 */
1917 			if ((data1sz = aclp->z_ops.ace_data(acep,
1918 			    &data1)) != 0) {
1919 				VERIFY((data2sz =
1920 				    aclp->z_ops.ace_data(acep2,
1921 				    &data2)) == data1sz);
1922 				bcopy(data1, data2, data1sz);
1923 			}
1924 			aclp->z_acl_count++;
1925 			aclnode2->z_ace_count++;
1926 			aclp->z_acl_bytes += aclnode->z_size;
1927 			zfs_restricted_update(zfsvfs, aclp, acep2);
1928 		} else {
1929 			newflags |= ACE_INHERIT_ONLY_ACE;
1930 			aclp->z_ops.ace_flags_set(acep,
1931 			    newflags|ACE_INHERITED_ACE);
1932 		}
1933 	}
1934 	return (aclp);
1935 }
1936 
1937 /*
1938  * Create file system object initial permissions
1939  * including inheritable ACEs.
1940  */
1941 int
1942 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1943     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1944 {
1945 	int		error;
1946 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1947 	zfs_acl_t	*paclp;
1948 	gid_t		gid;
1949 	boolean_t	need_chmod = B_TRUE;
1950 	boolean_t	inherited = B_FALSE;
1951 	uint64_t	parentgid;
1952 
1953 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1954 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1955 
1956 	if (vsecp)
1957 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1958 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1959 			return (error);
1960 	/*
1961 	 * Determine uid and gid.
1962 	 */
1963 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1964 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1965 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1966 		    (uint64_t)vap->va_uid, cr,
1967 		    ZFS_OWNER, &acl_ids->z_fuidp);
1968 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1969 		    (uint64_t)vap->va_gid, cr,
1970 		    ZFS_GROUP, &acl_ids->z_fuidp);
1971 		gid = vap->va_gid;
1972 	} else {
1973 		if (IS_EPHEMERAL(dzp->z_gid))
1974 			VERIFY(0 == sa_lookup(dzp->z_sa_hdl, SA_ZPL_GID(zfsvfs),
1975 			    &parentgid, sizeof (parentgid)));
1976 		else
1977 			parentgid = (uint64_t)dzp->z_gid;
1978 
1979 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1980 		    cr, &acl_ids->z_fuidp);
1981 		acl_ids->z_fgid = 0;
1982 		if (vap->va_mask & AT_GID)  {
1983 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1984 			    (uint64_t)vap->va_gid,
1985 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1986 			gid = vap->va_gid;
1987 			if (acl_ids->z_fgid != parentgid &&
1988 			    !groupmember(vap->va_gid, cr) &&
1989 			    secpolicy_vnode_create_gid(cr) != 0)
1990 				acl_ids->z_fgid = 0;
1991 		}
1992 		if (acl_ids->z_fgid == 0) {
1993 			if (dzp->z_mode & S_ISGID) {
1994 				char		*domain;
1995 				uint32_t	rid;
1996 
1997 				acl_ids->z_fgid = parentgid;
1998 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1999 				    cr, ZFS_GROUP);
2000 
2001 				if (zfsvfs->z_use_fuids &&
2002 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
2003 					domain = zfs_fuid_idx_domain(
2004 					    &zfsvfs->z_fuid_idx,
2005 					    FUID_INDEX(acl_ids->z_fgid));
2006 					rid = FUID_RID(acl_ids->z_fgid);
2007 					zfs_fuid_node_add(&acl_ids->z_fuidp,
2008 					    domain, rid,
2009 					    FUID_INDEX(acl_ids->z_fgid),
2010 					    acl_ids->z_fgid, ZFS_GROUP);
2011 				}
2012 			} else {
2013 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
2014 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
2015 				gid = crgetgid(cr);
2016 			}
2017 		}
2018 	}
2019 
2020 	/*
2021 	 * If we're creating a directory, and the parent directory has the
2022 	 * set-GID bit set, set in on the new directory.
2023 	 * Otherwise, if the user is neither privileged nor a member of the
2024 	 * file's new group, clear the file's set-GID bit.
2025 	 */
2026 
2027 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
2028 	    (vap->va_type == VDIR)) {
2029 		acl_ids->z_mode |= S_ISGID;
2030 	} else {
2031 		if ((acl_ids->z_mode & S_ISGID) &&
2032 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
2033 			acl_ids->z_mode &= ~S_ISGID;
2034 	}
2035 
2036 	if (acl_ids->z_aclp == NULL) {
2037 		mutex_enter(&dzp->z_lock);
2038 		if (!(flag & IS_ROOT_NODE) && (ZTOV(dzp)->v_type == VDIR &&
2039 		    (dzp->z_pflags & ZFS_INHERIT_ACE)) &&
2040 		    !(dzp->z_pflags & ZFS_XATTR)) {
2041 			mutex_enter(&dzp->z_acl_lock);
2042 			VERIFY(0 == zfs_acl_node_read(dzp, &paclp, B_FALSE));
2043 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
2044 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
2045 			mutex_exit(&dzp->z_acl_lock);
2046 			inherited = B_TRUE;
2047 		} else {
2048 			acl_ids->z_aclp =
2049 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
2050 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
2051 		}
2052 		mutex_exit(&dzp->z_lock);
2053 		if (need_chmod) {
2054 			acl_ids->z_aclp->z_hints |= (vap->va_type == VDIR) ?
2055 			    ZFS_ACL_AUTO_INHERIT : 0;
2056 			zfs_acl_chmod(zfsvfs, acl_ids->z_fuid,
2057 			    acl_ids->z_mode, acl_ids->z_aclp);
2058 		}
2059 	}
2060 
2061 	if (inherited || vsecp) {
2062 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
2063 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints);
2064 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
2065 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
2066 	}
2067 
2068 	return (0);
2069 }
2070 
2071 /*
2072  * Free ACL and fuid_infop, but not the acl_ids structure
2073  */
2074 void
2075 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
2076 {
2077 	if (acl_ids->z_aclp)
2078 		zfs_acl_free(acl_ids->z_aclp);
2079 	if (acl_ids->z_fuidp)
2080 		zfs_fuid_info_free(acl_ids->z_fuidp);
2081 	acl_ids->z_aclp = NULL;
2082 	acl_ids->z_fuidp = NULL;
2083 }
2084 
2085 boolean_t
2086 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
2087 {
2088 	return (zfs_fuid_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
2089 	    zfs_fuid_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
2090 }
2091 
2092 /*
2093  * Retrieve a files ACL
2094  */
2095 int
2096 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2097 {
2098 	zfs_acl_t	*aclp;
2099 	ulong_t		mask;
2100 	int		error;
2101 	int 		count = 0;
2102 	int		largeace = 0;
2103 
2104 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
2105 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
2106 
2107 	if (mask == 0)
2108 		return (ENOSYS);
2109 
2110 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
2111 		return (error);
2112 
2113 	mutex_enter(&zp->z_acl_lock);
2114 
2115 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2116 	if (error != 0) {
2117 		mutex_exit(&zp->z_acl_lock);
2118 		return (error);
2119 	}
2120 
2121 	/*
2122 	 * Scan ACL to determine number of ACEs
2123 	 */
2124 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
2125 		void *zacep = NULL;
2126 		uint64_t who;
2127 		uint32_t access_mask;
2128 		uint16_t type, iflags;
2129 
2130 		while (zacep = zfs_acl_next_ace(aclp, zacep,
2131 		    &who, &access_mask, &iflags, &type)) {
2132 			switch (type) {
2133 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
2134 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
2135 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
2136 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
2137 				largeace++;
2138 				continue;
2139 			default:
2140 				count++;
2141 			}
2142 		}
2143 		vsecp->vsa_aclcnt = count;
2144 	} else
2145 		count = (int)aclp->z_acl_count;
2146 
2147 	if (mask & VSA_ACECNT) {
2148 		vsecp->vsa_aclcnt = count;
2149 	}
2150 
2151 	if (mask & VSA_ACE) {
2152 		size_t aclsz;
2153 
2154 		aclsz = count * sizeof (ace_t) +
2155 		    sizeof (ace_object_t) * largeace;
2156 
2157 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2158 		vsecp->vsa_aclentsz = aclsz;
2159 
2160 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2161 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
2162 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2163 		else {
2164 			zfs_acl_node_t *aclnode;
2165 			void *start = vsecp->vsa_aclentp;
2166 
2167 			for (aclnode = list_head(&aclp->z_acl); aclnode;
2168 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
2169 				bcopy(aclnode->z_acldata, start,
2170 				    aclnode->z_size);
2171 				start = (caddr_t)start + aclnode->z_size;
2172 			}
2173 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
2174 			    aclp->z_acl_bytes);
2175 		}
2176 	}
2177 	if (mask & VSA_ACE_ACLFLAGS) {
2178 		vsecp->vsa_aclflags = 0;
2179 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
2180 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
2181 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
2182 			vsecp->vsa_aclflags |= ACL_PROTECTED;
2183 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
2184 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2185 	}
2186 
2187 	mutex_exit(&zp->z_acl_lock);
2188 
2189 	return (0);
2190 }
2191 
2192 int
2193 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
2194     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2195 {
2196 	zfs_acl_t *aclp;
2197 	zfs_acl_node_t *aclnode;
2198 	int aclcnt = vsecp->vsa_aclcnt;
2199 	int error;
2200 
2201 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2202 		return (EINVAL);
2203 
2204 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2205 
2206 	aclp->z_hints = 0;
2207 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2208 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2209 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
2210 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2211 		    aclcnt, &aclnode->z_size)) != 0) {
2212 			zfs_acl_free(aclp);
2213 			zfs_acl_node_free(aclnode);
2214 			return (error);
2215 		}
2216 	} else {
2217 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
2218 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2219 		    &aclnode->z_size, fuidp, cr)) != 0) {
2220 			zfs_acl_free(aclp);
2221 			zfs_acl_node_free(aclnode);
2222 			return (error);
2223 		}
2224 	}
2225 	aclp->z_acl_bytes = aclnode->z_size;
2226 	aclnode->z_ace_count = aclcnt;
2227 	aclp->z_acl_count = aclcnt;
2228 	list_insert_head(&aclp->z_acl, aclnode);
2229 
2230 	/*
2231 	 * If flags are being set then add them to z_hints
2232 	 */
2233 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2234 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2235 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2236 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2237 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2238 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2239 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2240 	}
2241 
2242 	*zaclp = aclp;
2243 
2244 	return (0);
2245 }
2246 
2247 /*
2248  * Set a files ACL
2249  */
2250 int
2251 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2252 {
2253 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2254 	zilog_t		*zilog = zfsvfs->z_log;
2255 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2256 	dmu_tx_t	*tx;
2257 	int		error;
2258 	zfs_acl_t	*aclp;
2259 	zfs_fuid_info_t	*fuidp = NULL;
2260 	boolean_t	fuid_dirtied;
2261 
2262 	if (mask == 0)
2263 		return (ENOSYS);
2264 
2265 	if (zp->z_pflags & ZFS_IMMUTABLE)
2266 		return (EPERM);
2267 
2268 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
2269 		return (error);
2270 
2271 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
2272 	    &aclp);
2273 	if (error)
2274 		return (error);
2275 
2276 	/*
2277 	 * If ACL wide flags aren't being set then preserve any
2278 	 * existing flags.
2279 	 */
2280 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2281 		aclp->z_hints |=
2282 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2283 	}
2284 top:
2285 	mutex_enter(&zp->z_lock);
2286 	mutex_enter(&zp->z_acl_lock);
2287 
2288 	tx = dmu_tx_create(zfsvfs->z_os);
2289 
2290 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2291 
2292 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2293 	if (fuid_dirtied)
2294 		zfs_fuid_txhold(zfsvfs, tx);
2295 
2296 	/*
2297 	 * If old version and ACL won't fit in bonus and we aren't
2298 	 * upgrading then take out necessary DMU holds
2299 	 */
2300 
2301 	if (ZFS_EXTERNAL_ACL(zp)) {
2302 		if (zfsvfs->z_version <= ZPL_VERSION_SA &&
2303 		    ZNODE_ACL_VERSION(zp) <= ZFS_ACL_VERSION_INITIAL) {
2304 			dmu_tx_hold_free(tx, ZFS_EXTERNAL_ACL(zp), 0,
2305 			    DMU_OBJECT_END);
2306 		} else {
2307 			dmu_tx_hold_write(tx, ZFS_EXTERNAL_ACL(zp),
2308 			    0, aclp->z_acl_bytes);
2309 		}
2310 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2311 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2312 	}
2313 
2314 	zfs_sa_upgrade_txholds(tx, zp);
2315 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2316 	if (error) {
2317 		mutex_exit(&zp->z_acl_lock);
2318 		mutex_exit(&zp->z_lock);
2319 
2320 		if (error == ERESTART) {
2321 			dmu_tx_wait(tx);
2322 			dmu_tx_abort(tx);
2323 			goto top;
2324 		}
2325 		dmu_tx_abort(tx);
2326 		zfs_acl_free(aclp);
2327 		return (error);
2328 	}
2329 
2330 	error = zfs_aclset_common(zp, aclp, cr, tx);
2331 	ASSERT(error == 0);
2332 	zp->z_acl_cached = aclp;
2333 
2334 	if (fuid_dirtied)
2335 		zfs_fuid_sync(zfsvfs, tx);
2336 
2337 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2338 
2339 	if (fuidp)
2340 		zfs_fuid_info_free(fuidp);
2341 	dmu_tx_commit(tx);
2342 done:
2343 	mutex_exit(&zp->z_acl_lock);
2344 	mutex_exit(&zp->z_lock);
2345 
2346 	return (error);
2347 }
2348 
2349 /*
2350  * Check accesses of interest (AoI) against attributes of the dataset
2351  * such as read-only.  Returns zero if no AoI conflict with dataset
2352  * attributes, otherwise an appropriate errno is returned.
2353  */
2354 static int
2355 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2356 {
2357 	if ((v4_mode & WRITE_MASK) &&
2358 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2359 	    (!IS_DEVVP(ZTOV(zp)) ||
2360 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2361 		return (EROFS);
2362 	}
2363 
2364 	/*
2365 	 * Only check for READONLY on non-directories.
2366 	 */
2367 	if ((v4_mode & WRITE_MASK_DATA) &&
2368 	    (((ZTOV(zp)->v_type != VDIR) &&
2369 	    (zp->z_pflags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2370 	    (ZTOV(zp)->v_type == VDIR &&
2371 	    (zp->z_pflags & ZFS_IMMUTABLE)))) {
2372 		return (EPERM);
2373 	}
2374 
2375 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2376 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2377 		return (EPERM);
2378 	}
2379 
2380 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2381 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2382 		return (EACCES);
2383 	}
2384 
2385 	return (0);
2386 }
2387 
2388 /*
2389  * The primary usage of this function is to loop through all of the
2390  * ACEs in the znode, determining what accesses of interest (AoI) to
2391  * the caller are allowed or denied.  The AoI are expressed as bits in
2392  * the working_mode parameter.  As each ACE is processed, bits covered
2393  * by that ACE are removed from the working_mode.  This removal
2394  * facilitates two things.  The first is that when the working mode is
2395  * empty (= 0), we know we've looked at all the AoI. The second is
2396  * that the ACE interpretation rules don't allow a later ACE to undo
2397  * something granted or denied by an earlier ACE.  Removing the
2398  * discovered access or denial enforces this rule.  At the end of
2399  * processing the ACEs, all AoI that were found to be denied are
2400  * placed into the working_mode, giving the caller a mask of denied
2401  * accesses.  Returns:
2402  *	0		if all AoI granted
2403  *	EACCESS 	if the denied mask is non-zero
2404  *	other error	if abnormal failure (e.g., IO error)
2405  *
2406  * A secondary usage of the function is to determine if any of the
2407  * AoI are granted.  If an ACE grants any access in
2408  * the working_mode, we immediately short circuit out of the function.
2409  * This mode is chosen by setting anyaccess to B_TRUE.  The
2410  * working_mode is not a denied access mask upon exit if the function
2411  * is used in this manner.
2412  */
2413 static int
2414 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2415     boolean_t anyaccess, cred_t *cr)
2416 {
2417 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2418 	zfs_acl_t	*aclp;
2419 	int		error;
2420 	uid_t		uid = crgetuid(cr);
2421 	uint64_t 	who;
2422 	uint16_t	type, iflags;
2423 	uint16_t	entry_type;
2424 	uint32_t	access_mask;
2425 	uint32_t	deny_mask = 0;
2426 	zfs_ace_hdr_t	*acep = NULL;
2427 	boolean_t	checkit;
2428 	uint64_t	gowner;
2429 
2430 	mutex_enter(&zp->z_acl_lock);
2431 
2432 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2433 	if (error != 0) {
2434 		mutex_exit(&zp->z_acl_lock);
2435 		return (error);
2436 	}
2437 
2438 	ASSERT(zp->z_acl_cached);
2439 
2440 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GID(zfsvfs),
2441 	    &gowner, sizeof (gowner))) != 0)
2442 		return (error);
2443 
2444 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2445 	    &iflags, &type)) {
2446 		uint32_t mask_matched;
2447 
2448 		if (!zfs_acl_valid_ace_type(type, iflags))
2449 			continue;
2450 
2451 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2452 			continue;
2453 
2454 		/* Skip ACE if it does not affect any AoI */
2455 		mask_matched = (access_mask & *working_mode);
2456 		if (!mask_matched)
2457 			continue;
2458 
2459 		entry_type = (iflags & ACE_TYPE_FLAGS);
2460 
2461 		checkit = B_FALSE;
2462 
2463 		switch (entry_type) {
2464 		case ACE_OWNER:
2465 			if (uid == zp->z_uid)
2466 				checkit = B_TRUE;
2467 			break;
2468 		case OWNING_GROUP:
2469 			who = gowner;
2470 			/*FALLTHROUGH*/
2471 		case ACE_IDENTIFIER_GROUP:
2472 			checkit = zfs_groupmember(zfsvfs, who, cr);
2473 			break;
2474 		case ACE_EVERYONE:
2475 			checkit = B_TRUE;
2476 			break;
2477 
2478 		/* USER Entry */
2479 		default:
2480 			if (entry_type == 0) {
2481 				uid_t newid;
2482 
2483 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2484 				    ZFS_ACE_USER);
2485 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2486 				    uid == newid)
2487 					checkit = B_TRUE;
2488 				break;
2489 			} else {
2490 				mutex_exit(&zp->z_acl_lock);
2491 				return (EIO);
2492 			}
2493 		}
2494 
2495 		if (checkit) {
2496 			if (type == DENY) {
2497 				DTRACE_PROBE3(zfs__ace__denies,
2498 				    znode_t *, zp,
2499 				    zfs_ace_hdr_t *, acep,
2500 				    uint32_t, mask_matched);
2501 				deny_mask |= mask_matched;
2502 			} else {
2503 				DTRACE_PROBE3(zfs__ace__allows,
2504 				    znode_t *, zp,
2505 				    zfs_ace_hdr_t *, acep,
2506 				    uint32_t, mask_matched);
2507 				if (anyaccess) {
2508 					mutex_exit(&zp->z_acl_lock);
2509 					return (0);
2510 				}
2511 			}
2512 			*working_mode &= ~mask_matched;
2513 		}
2514 
2515 		/* Are we done? */
2516 		if (*working_mode == 0)
2517 			break;
2518 	}
2519 
2520 	mutex_exit(&zp->z_acl_lock);
2521 
2522 	/* Put the found 'denies' back on the working mode */
2523 	if (deny_mask) {
2524 		*working_mode |= deny_mask;
2525 		return (EACCES);
2526 	} else if (*working_mode) {
2527 		return (-1);
2528 	}
2529 
2530 	return (0);
2531 }
2532 
2533 /*
2534  * Return true if any access whatsoever granted, we don't actually
2535  * care what access is granted.
2536  */
2537 boolean_t
2538 zfs_has_access(znode_t *zp, cred_t *cr)
2539 {
2540 	uint32_t have = ACE_ALL_PERMS;
2541 
2542 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2543 		return (
2544 		    secpolicy_vnode_access(cr, ZTOV(zp),
2545 		    zp->z_uid, VREAD) == 0 || secpolicy_vnode_access(cr,
2546 		    ZTOV(zp), zp->z_uid, VWRITE) == 0 ||
2547 		    secpolicy_vnode_access(cr, ZTOV(zp),
2548 		    zp->z_uid, VEXEC) == 0 ||
2549 		    secpolicy_vnode_chown(cr, zp->z_uid) == 0 ||
2550 		    secpolicy_vnode_setdac(cr, zp->z_uid) == 0 ||
2551 		    secpolicy_vnode_remove(cr) == 0);
2552 	}
2553 	return (B_TRUE);
2554 }
2555 
2556 static int
2557 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2558     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2559 {
2560 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2561 	int err;
2562 
2563 	*working_mode = v4_mode;
2564 	*check_privs = B_TRUE;
2565 
2566 	/*
2567 	 * Short circuit empty requests
2568 	 */
2569 	if (v4_mode == 0 || zfsvfs->z_replay) {
2570 		*working_mode = 0;
2571 		return (0);
2572 	}
2573 
2574 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2575 		*check_privs = B_FALSE;
2576 		return (err);
2577 	}
2578 
2579 	/*
2580 	 * The caller requested that the ACL check be skipped.  This
2581 	 * would only happen if the caller checked VOP_ACCESS() with a
2582 	 * 32 bit ACE mask and already had the appropriate permissions.
2583 	 */
2584 	if (skipaclchk) {
2585 		*working_mode = 0;
2586 		return (0);
2587 	}
2588 
2589 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2590 }
2591 
2592 static int
2593 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2594     cred_t *cr)
2595 {
2596 	if (*working_mode != ACE_WRITE_DATA)
2597 		return (EACCES);
2598 
2599 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2600 	    check_privs, B_FALSE, cr));
2601 }
2602 
2603 int
2604 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2605 {
2606 	boolean_t owner = B_FALSE;
2607 	boolean_t groupmbr = B_FALSE;
2608 	boolean_t is_attr;
2609 	uid_t uid = crgetuid(cr);
2610 	int error;
2611 
2612 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2613 		return (EACCES);
2614 
2615 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2616 	    (ZTOV(zdp)->v_type == VDIR));
2617 	if (is_attr)
2618 		goto slow;
2619 
2620 
2621 	mutex_enter(&zdp->z_acl_lock);
2622 
2623 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2624 		mutex_exit(&zdp->z_acl_lock);
2625 		return (0);
2626 	}
2627 
2628 	if (IS_EPHEMERAL(zdp->z_uid) != 0 || IS_EPHEMERAL(zdp->z_gid) != 0) {
2629 		mutex_exit(&zdp->z_acl_lock);
2630 		goto slow;
2631 	}
2632 
2633 	if (uid == zdp->z_uid) {
2634 		owner = B_TRUE;
2635 		if (zdp->z_mode & S_IXUSR) {
2636 			mutex_exit(&zdp->z_acl_lock);
2637 			return (0);
2638 		} else {
2639 			mutex_exit(&zdp->z_acl_lock);
2640 			goto slow;
2641 		}
2642 	}
2643 	if (groupmember(zdp->z_gid, cr)) {
2644 		groupmbr = B_TRUE;
2645 		if (zdp->z_mode & S_IXGRP) {
2646 			mutex_exit(&zdp->z_acl_lock);
2647 			return (0);
2648 		} else {
2649 			mutex_exit(&zdp->z_acl_lock);
2650 			goto slow;
2651 		}
2652 	}
2653 	if (!owner && !groupmbr) {
2654 		if (zdp->z_mode & S_IXOTH) {
2655 			mutex_exit(&zdp->z_acl_lock);
2656 			return (0);
2657 		}
2658 	}
2659 
2660 	mutex_exit(&zdp->z_acl_lock);
2661 
2662 slow:
2663 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2664 	ZFS_ENTER(zdp->z_zfsvfs);
2665 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2666 	ZFS_EXIT(zdp->z_zfsvfs);
2667 	return (error);
2668 }
2669 
2670 /*
2671  * Determine whether Access should be granted/denied, invoking least
2672  * priv subsytem when a deny is determined.
2673  */
2674 int
2675 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2676 {
2677 	uint32_t	working_mode;
2678 	int		error;
2679 	int		is_attr;
2680 	boolean_t 	check_privs;
2681 	znode_t		*xzp;
2682 	znode_t 	*check_zp = zp;
2683 
2684 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2685 
2686 	/*
2687 	 * If attribute then validate against base file
2688 	 */
2689 	if (is_attr) {
2690 		uint64_t	parent;
2691 
2692 		if ((error = sa_lookup(zp->z_sa_hdl,
2693 		    SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2694 		    sizeof (parent))) != 0)
2695 			return (error);
2696 
2697 		if ((error = zfs_zget(zp->z_zfsvfs,
2698 		    parent, &xzp)) != 0)	{
2699 			return (error);
2700 		}
2701 
2702 		check_zp = xzp;
2703 
2704 		/*
2705 		 * fixup mode to map to xattr perms
2706 		 */
2707 
2708 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2709 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2710 			mode |= ACE_WRITE_NAMED_ATTRS;
2711 		}
2712 
2713 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2714 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2715 			mode |= ACE_READ_NAMED_ATTRS;
2716 		}
2717 	}
2718 
2719 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2720 	    &check_privs, skipaclchk, cr)) == 0) {
2721 		if (is_attr)
2722 			VN_RELE(ZTOV(xzp));
2723 		return (0);
2724 	}
2725 
2726 	if (error && !check_privs) {
2727 		if (is_attr)
2728 			VN_RELE(ZTOV(xzp));
2729 		return (error);
2730 	}
2731 
2732 	if (error && (flags & V_APPEND)) {
2733 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2734 	}
2735 
2736 	if (error && check_privs) {
2737 		mode_t		checkmode = 0;
2738 
2739 		/*
2740 		 * First check for implicit owner permission on
2741 		 * read_acl/read_attributes
2742 		 */
2743 
2744 		error = 0;
2745 		ASSERT(working_mode != 0);
2746 
2747 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2748 		    zp->z_uid == crgetuid(cr)))
2749 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2750 
2751 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2752 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2753 			checkmode |= VREAD;
2754 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2755 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2756 			checkmode |= VWRITE;
2757 		if (working_mode & ACE_EXECUTE)
2758 			checkmode |= VEXEC;
2759 
2760 		if (checkmode)
2761 			error = secpolicy_vnode_access(cr, ZTOV(check_zp),
2762 			    zp->z_uid, checkmode);
2763 
2764 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2765 			error = secpolicy_vnode_chown(cr, zp->z_uid);
2766 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2767 			error = secpolicy_vnode_setdac(cr, zp->z_uid);
2768 
2769 		if (error == 0 && (working_mode &
2770 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2771 			error = secpolicy_vnode_remove(cr);
2772 
2773 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2774 			error = secpolicy_vnode_chown(cr, zp->z_uid);
2775 		}
2776 		if (error == 0) {
2777 			/*
2778 			 * See if any bits other than those already checked
2779 			 * for are still present.  If so then return EACCES
2780 			 */
2781 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2782 				error = EACCES;
2783 			}
2784 		}
2785 	}
2786 
2787 	if (is_attr)
2788 		VN_RELE(ZTOV(xzp));
2789 
2790 	return (error);
2791 }
2792 
2793 /*
2794  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2795  * native ACL format and call zfs_zaccess()
2796  */
2797 int
2798 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2799 {
2800 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2801 }
2802 
2803 /*
2804  * Access function for secpolicy_vnode_setattr
2805  */
2806 int
2807 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2808 {
2809 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2810 
2811 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2812 }
2813 
2814 static int
2815 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2816     mode_t missing_perms, cred_t *cr)
2817 {
2818 	int error;
2819 
2820 	error = secpolicy_vnode_access(cr, ZTOV(dzp),
2821 	    dzp->z_uid, missing_perms);
2822 
2823 	if (error == 0)
2824 		error = zfs_sticky_remove_access(dzp, zp, cr);
2825 
2826 	return (error);
2827 }
2828 
2829 /*
2830  * Determine whether Access should be granted/deny, without
2831  * consulting least priv subsystem.
2832  *
2833  *
2834  * The following chart is the recommended NFSv4 enforcement for
2835  * ability to delete an object.
2836  *
2837  *      -------------------------------------------------------
2838  *      |   Parent Dir  |           Target Object Permissions |
2839  *      |  permissions  |                                     |
2840  *      -------------------------------------------------------
2841  *      |               | ACL Allows | ACL Denies| Delete     |
2842  *      |               |  Delete    |  Delete   | unspecified|
2843  *      -------------------------------------------------------
2844  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2845  *      |  DELETE_CHILD |                                     |
2846  *      -------------------------------------------------------
2847  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2848  *      |  DELETE_CHILD |            |           |            |
2849  *      -------------------------------------------------------
2850  *      | ACL specifies |            |           |            |
2851  *      | only allow    | Permit     | Permit    | Permit     |
2852  *      | write and     |            |           |            |
2853  *      | execute       |            |           |            |
2854  *      -------------------------------------------------------
2855  *      | ACL denies    |            |           |            |
2856  *      | write and     | Permit     | Deny      | Deny       |
2857  *      | execute       |            |           |            |
2858  *      -------------------------------------------------------
2859  *         ^
2860  *         |
2861  *         No search privilege, can't even look up file?
2862  *
2863  */
2864 int
2865 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2866 {
2867 	uint32_t dzp_working_mode = 0;
2868 	uint32_t zp_working_mode = 0;
2869 	int dzp_error, zp_error;
2870 	mode_t missing_perms;
2871 	boolean_t dzpcheck_privs = B_TRUE;
2872 	boolean_t zpcheck_privs = B_TRUE;
2873 
2874 	/*
2875 	 * We want specific DELETE permissions to
2876 	 * take precedence over WRITE/EXECUTE.  We don't
2877 	 * want an ACL such as this to mess us up.
2878 	 * user:joe:write_data:deny,user:joe:delete:allow
2879 	 *
2880 	 * However, deny permissions may ultimately be overridden
2881 	 * by secpolicy_vnode_access().
2882 	 *
2883 	 * We will ask for all of the necessary permissions and then
2884 	 * look at the working modes from the directory and target object
2885 	 * to determine what was found.
2886 	 */
2887 
2888 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2889 		return (EPERM);
2890 
2891 	/*
2892 	 * First row
2893 	 * If the directory permissions allow the delete, we are done.
2894 	 */
2895 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2896 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2897 		return (0);
2898 
2899 	/*
2900 	 * If target object has delete permission then we are done
2901 	 */
2902 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2903 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2904 		return (0);
2905 
2906 	ASSERT(dzp_error && zp_error);
2907 
2908 	if (!dzpcheck_privs)
2909 		return (dzp_error);
2910 	if (!zpcheck_privs)
2911 		return (zp_error);
2912 
2913 	/*
2914 	 * Second row
2915 	 *
2916 	 * If directory returns EACCES then delete_child was denied
2917 	 * due to deny delete_child.  In this case send the request through
2918 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2919 	 * since that *could* allow the delete based on write/execute permission
2920 	 * and we want delete permissions to override write/execute.
2921 	 */
2922 
2923 	if (dzp_error == EACCES)
2924 		return (secpolicy_vnode_remove(cr));
2925 
2926 	/*
2927 	 * Third Row
2928 	 * only need to see if we have write/execute on directory.
2929 	 */
2930 
2931 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2932 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2933 		return (zfs_sticky_remove_access(dzp, zp, cr));
2934 
2935 	if (!dzpcheck_privs)
2936 		return (dzp_error);
2937 
2938 	/*
2939 	 * Fourth row
2940 	 */
2941 
2942 	missing_perms = (dzp_working_mode & ACE_WRITE_DATA) ? VWRITE : 0;
2943 	missing_perms |= (dzp_working_mode & ACE_EXECUTE) ? VEXEC : 0;
2944 
2945 	ASSERT(missing_perms);
2946 
2947 	return (zfs_delete_final_check(zp, dzp, missing_perms, cr));
2948 
2949 }
2950 
2951 int
2952 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2953     znode_t *tzp, cred_t *cr)
2954 {
2955 	int add_perm;
2956 	int error;
2957 
2958 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2959 		return (EACCES);
2960 
2961 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2962 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2963 
2964 	/*
2965 	 * Rename permissions are combination of delete permission +
2966 	 * add file/subdir permission.
2967 	 */
2968 
2969 	/*
2970 	 * first make sure we do the delete portion.
2971 	 *
2972 	 * If that succeeds then check for add_file/add_subdir permissions
2973 	 */
2974 
2975 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2976 		return (error);
2977 
2978 	/*
2979 	 * If we have a tzp, see if we can delete it?
2980 	 */
2981 	if (tzp) {
2982 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2983 			return (error);
2984 	}
2985 
2986 	/*
2987 	 * Now check for add permissions
2988 	 */
2989 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2990 
2991 	return (error);
2992 }
2993