xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_acl.c (revision efd4c9b63ad77503c101fc6c2ed8ba96c9d52964)
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 #include <sys/types.h>
26 #include <sys/param.h>
27 #include <sys/time.h>
28 #include <sys/systm.h>
29 #include <sys/sysmacros.h>
30 #include <sys/resource.h>
31 #include <sys/vfs.h>
32 #include <sys/vnode.h>
33 #include <sys/sid.h>
34 #include <sys/file.h>
35 #include <sys/stat.h>
36 #include <sys/kmem.h>
37 #include <sys/cmn_err.h>
38 #include <sys/errno.h>
39 #include <sys/unistd.h>
40 #include <sys/sdt.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/mode.h>
43 #include <sys/policy.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_fuid.h>
46 #include <sys/zfs_acl.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_vfsops.h>
49 #include <sys/dmu.h>
50 #include <sys/dnode.h>
51 #include <sys/zap.h>
52 #include <sys/sa.h>
53 #include "fs/fs_subr.h"
54 #include <acl/acl_common.h>
55 
56 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
57 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
58 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
59 #define	MIN_ACE_TYPE	ALLOW
60 
61 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
62 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
63     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
64 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
65     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
66 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
67     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
68 
69 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
70     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
71     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
72     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
73 
74 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
75 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
76     ACE_DELETE|ACE_DELETE_CHILD)
77 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
78 
79 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
80     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
81 
82 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
83     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
84 
85 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
86     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
87 
88 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
89 
90 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
91     ZFS_ACL_PROTECTED)
92 
93 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
94     ZFS_ACL_OBJ_ACE)
95 
96 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
97 
98 static uint16_t
99 zfs_ace_v0_get_type(void *acep)
100 {
101 	return (((zfs_oldace_t *)acep)->z_type);
102 }
103 
104 static uint16_t
105 zfs_ace_v0_get_flags(void *acep)
106 {
107 	return (((zfs_oldace_t *)acep)->z_flags);
108 }
109 
110 static uint32_t
111 zfs_ace_v0_get_mask(void *acep)
112 {
113 	return (((zfs_oldace_t *)acep)->z_access_mask);
114 }
115 
116 static uint64_t
117 zfs_ace_v0_get_who(void *acep)
118 {
119 	return (((zfs_oldace_t *)acep)->z_fuid);
120 }
121 
122 static void
123 zfs_ace_v0_set_type(void *acep, uint16_t type)
124 {
125 	((zfs_oldace_t *)acep)->z_type = type;
126 }
127 
128 static void
129 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
130 {
131 	((zfs_oldace_t *)acep)->z_flags = flags;
132 }
133 
134 static void
135 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
136 {
137 	((zfs_oldace_t *)acep)->z_access_mask = mask;
138 }
139 
140 static void
141 zfs_ace_v0_set_who(void *acep, uint64_t who)
142 {
143 	((zfs_oldace_t *)acep)->z_fuid = who;
144 }
145 
146 /*ARGSUSED*/
147 static size_t
148 zfs_ace_v0_size(void *acep)
149 {
150 	return (sizeof (zfs_oldace_t));
151 }
152 
153 static size_t
154 zfs_ace_v0_abstract_size(void)
155 {
156 	return (sizeof (zfs_oldace_t));
157 }
158 
159 static int
160 zfs_ace_v0_mask_off(void)
161 {
162 	return (offsetof(zfs_oldace_t, z_access_mask));
163 }
164 
165 /*ARGSUSED*/
166 static int
167 zfs_ace_v0_data(void *acep, void **datap)
168 {
169 	*datap = NULL;
170 	return (0);
171 }
172 
173 static acl_ops_t zfs_acl_v0_ops = {
174 	zfs_ace_v0_get_mask,
175 	zfs_ace_v0_set_mask,
176 	zfs_ace_v0_get_flags,
177 	zfs_ace_v0_set_flags,
178 	zfs_ace_v0_get_type,
179 	zfs_ace_v0_set_type,
180 	zfs_ace_v0_get_who,
181 	zfs_ace_v0_set_who,
182 	zfs_ace_v0_size,
183 	zfs_ace_v0_abstract_size,
184 	zfs_ace_v0_mask_off,
185 	zfs_ace_v0_data
186 };
187 
188 static uint16_t
189 zfs_ace_fuid_get_type(void *acep)
190 {
191 	return (((zfs_ace_hdr_t *)acep)->z_type);
192 }
193 
194 static uint16_t
195 zfs_ace_fuid_get_flags(void *acep)
196 {
197 	return (((zfs_ace_hdr_t *)acep)->z_flags);
198 }
199 
200 static uint32_t
201 zfs_ace_fuid_get_mask(void *acep)
202 {
203 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
204 }
205 
206 static uint64_t
207 zfs_ace_fuid_get_who(void *args)
208 {
209 	uint16_t entry_type;
210 	zfs_ace_t *acep = args;
211 
212 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
213 
214 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
215 	    entry_type == ACE_EVERYONE)
216 		return (-1);
217 	return (((zfs_ace_t *)acep)->z_fuid);
218 }
219 
220 static void
221 zfs_ace_fuid_set_type(void *acep, uint16_t type)
222 {
223 	((zfs_ace_hdr_t *)acep)->z_type = type;
224 }
225 
226 static void
227 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
228 {
229 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
230 }
231 
232 static void
233 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
234 {
235 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
236 }
237 
238 static void
239 zfs_ace_fuid_set_who(void *arg, uint64_t who)
240 {
241 	zfs_ace_t *acep = arg;
242 
243 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
244 
245 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
246 	    entry_type == ACE_EVERYONE)
247 		return;
248 	acep->z_fuid = who;
249 }
250 
251 static size_t
252 zfs_ace_fuid_size(void *acep)
253 {
254 	zfs_ace_hdr_t *zacep = acep;
255 	uint16_t entry_type;
256 
257 	switch (zacep->z_type) {
258 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
259 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
260 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
261 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
262 		return (sizeof (zfs_object_ace_t));
263 	case ALLOW:
264 	case DENY:
265 		entry_type =
266 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
267 		if (entry_type == ACE_OWNER ||
268 		    entry_type == OWNING_GROUP ||
269 		    entry_type == ACE_EVERYONE)
270 			return (sizeof (zfs_ace_hdr_t));
271 		/*FALLTHROUGH*/
272 	default:
273 		return (sizeof (zfs_ace_t));
274 	}
275 }
276 
277 static size_t
278 zfs_ace_fuid_abstract_size(void)
279 {
280 	return (sizeof (zfs_ace_hdr_t));
281 }
282 
283 static int
284 zfs_ace_fuid_mask_off(void)
285 {
286 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
287 }
288 
289 static int
290 zfs_ace_fuid_data(void *acep, void **datap)
291 {
292 	zfs_ace_t *zacep = acep;
293 	zfs_object_ace_t *zobjp;
294 
295 	switch (zacep->z_hdr.z_type) {
296 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
297 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
298 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
299 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
300 		zobjp = acep;
301 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
302 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
303 	default:
304 		*datap = NULL;
305 		return (0);
306 	}
307 }
308 
309 static acl_ops_t zfs_acl_fuid_ops = {
310 	zfs_ace_fuid_get_mask,
311 	zfs_ace_fuid_set_mask,
312 	zfs_ace_fuid_get_flags,
313 	zfs_ace_fuid_set_flags,
314 	zfs_ace_fuid_get_type,
315 	zfs_ace_fuid_set_type,
316 	zfs_ace_fuid_get_who,
317 	zfs_ace_fuid_set_who,
318 	zfs_ace_fuid_size,
319 	zfs_ace_fuid_abstract_size,
320 	zfs_ace_fuid_mask_off,
321 	zfs_ace_fuid_data
322 };
323 
324 /*
325  * The following three functions are provided for compatibility with
326  * older ZPL version in order to determine if the file use to have
327  * an external ACL and what version of ACL previously existed on the
328  * file.  Would really be nice to not need this, sigh.
329  */
330 uint64_t
331 zfs_external_acl(znode_t *zp)
332 {
333 	zfs_acl_phys_t acl_phys;
334 	int error;
335 
336 	if (zp->z_is_sa)
337 		return (0);
338 
339 	/*
340 	 * Need to deal with a potential
341 	 * race where zfs_sa_upgrade could cause
342 	 * z_isa_sa to change.
343 	 *
344 	 * If the lookup fails then the state of z_is_sa should have
345 	 * changed.
346 	 */
347 
348 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
349 	    &acl_phys, sizeof (acl_phys))) == 0)
350 		return (acl_phys.z_acl_extern_obj);
351 	else {
352 		/*
353 		 * after upgrade the SA_ZPL_ZNODE_ACL should have been
354 		 * removed
355 		 */
356 		VERIFY(zp->z_is_sa && error == ENOENT);
357 		return (0);
358 	}
359 }
360 
361 /*
362  * Determine size of ACL in bytes
363  *
364  * This is more complicated than it should be since we have to deal
365  * with old external ACLs.
366  */
367 static int
368 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
369     zfs_acl_phys_t *aclphys)
370 {
371 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
372 	uint64_t acl_count;
373 	int size;
374 	int error;
375 
376 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
377 	if (zp->z_is_sa) {
378 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
379 		    &size)) != 0)
380 			return (error);
381 		*aclsize = size;
382 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
383 		    &acl_count, sizeof (acl_count))) != 0)
384 			return (error);
385 		*aclcount = acl_count;
386 	} else {
387 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
388 		    aclphys, sizeof (*aclphys))) != 0)
389 			return (error);
390 
391 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
392 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
393 			*aclcount = aclphys->z_acl_size;
394 		} else {
395 			*aclsize = aclphys->z_acl_size;
396 			*aclcount = aclphys->z_acl_count;
397 		}
398 	}
399 	return (0);
400 }
401 
402 int
403 zfs_znode_acl_version(znode_t *zp)
404 {
405 	zfs_acl_phys_t acl_phys;
406 
407 	if (zp->z_is_sa)
408 		return (ZFS_ACL_VERSION_FUID);
409 	else {
410 		int error;
411 
412 		/*
413 		 * Need to deal with a potential
414 		 * race where zfs_sa_upgrade could cause
415 		 * z_isa_sa to change.
416 		 *
417 		 * If the lookup fails then the state of z_is_sa should have
418 		 * changed.
419 		 */
420 		if ((error = sa_lookup(zp->z_sa_hdl,
421 		    SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
422 		    &acl_phys, sizeof (acl_phys))) == 0)
423 			return (acl_phys.z_acl_version);
424 		else {
425 			/*
426 			 * After upgrade SA_ZPL_ZNODE_ACL should have
427 			 * been removed.
428 			 */
429 			VERIFY(zp->z_is_sa && error == ENOENT);
430 			return (ZFS_ACL_VERSION_FUID);
431 		}
432 	}
433 }
434 
435 static int
436 zfs_acl_version(int version)
437 {
438 	if (version < ZPL_VERSION_FUID)
439 		return (ZFS_ACL_VERSION_INITIAL);
440 	else
441 		return (ZFS_ACL_VERSION_FUID);
442 }
443 
444 static int
445 zfs_acl_version_zp(znode_t *zp)
446 {
447 	return (zfs_acl_version(zp->z_zfsvfs->z_version));
448 }
449 
450 zfs_acl_t *
451 zfs_acl_alloc(int vers)
452 {
453 	zfs_acl_t *aclp;
454 
455 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
456 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
457 	    offsetof(zfs_acl_node_t, z_next));
458 	aclp->z_version = vers;
459 	if (vers == ZFS_ACL_VERSION_FUID)
460 		aclp->z_ops = zfs_acl_fuid_ops;
461 	else
462 		aclp->z_ops = zfs_acl_v0_ops;
463 	return (aclp);
464 }
465 
466 zfs_acl_node_t *
467 zfs_acl_node_alloc(size_t bytes)
468 {
469 	zfs_acl_node_t *aclnode;
470 
471 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
472 	if (bytes) {
473 		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
474 		aclnode->z_allocdata = aclnode->z_acldata;
475 		aclnode->z_allocsize = bytes;
476 		aclnode->z_size = bytes;
477 	}
478 
479 	return (aclnode);
480 }
481 
482 static void
483 zfs_acl_node_free(zfs_acl_node_t *aclnode)
484 {
485 	if (aclnode->z_allocsize)
486 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
487 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
488 }
489 
490 static void
491 zfs_acl_release_nodes(zfs_acl_t *aclp)
492 {
493 	zfs_acl_node_t *aclnode;
494 
495 	while (aclnode = list_head(&aclp->z_acl)) {
496 		list_remove(&aclp->z_acl, aclnode);
497 		zfs_acl_node_free(aclnode);
498 	}
499 	aclp->z_acl_count = 0;
500 	aclp->z_acl_bytes = 0;
501 }
502 
503 void
504 zfs_acl_free(zfs_acl_t *aclp)
505 {
506 	zfs_acl_release_nodes(aclp);
507 	list_destroy(&aclp->z_acl);
508 	kmem_free(aclp, sizeof (zfs_acl_t));
509 }
510 
511 static boolean_t
512 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
513 {
514 	uint16_t entry_type;
515 
516 	switch (type) {
517 	case ALLOW:
518 	case DENY:
519 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
520 	case ACE_SYSTEM_ALARM_ACE_TYPE:
521 		entry_type = flags & ACE_TYPE_FLAGS;
522 		return (entry_type == ACE_OWNER ||
523 		    entry_type == OWNING_GROUP ||
524 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
525 		    entry_type == ACE_IDENTIFIER_GROUP);
526 	default:
527 		if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
528 			return (B_TRUE);
529 	}
530 	return (B_FALSE);
531 }
532 
533 static boolean_t
534 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
535 {
536 	/*
537 	 * first check type of entry
538 	 */
539 
540 	if (!zfs_acl_valid_ace_type(type, iflags))
541 		return (B_FALSE);
542 
543 	switch (type) {
544 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
545 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
546 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
547 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
548 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
549 			return (B_FALSE);
550 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
551 	}
552 
553 	/*
554 	 * next check inheritance level flags
555 	 */
556 
557 	if (obj_type == VDIR &&
558 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
559 		aclp->z_hints |= ZFS_INHERIT_ACE;
560 
561 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
562 		if ((iflags & (ACE_FILE_INHERIT_ACE|
563 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
564 			return (B_FALSE);
565 		}
566 	}
567 
568 	return (B_TRUE);
569 }
570 
571 static void *
572 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
573     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
574 {
575 	zfs_acl_node_t *aclnode;
576 
577 	ASSERT(aclp);
578 
579 	if (start == NULL) {
580 		aclnode = list_head(&aclp->z_acl);
581 		if (aclnode == NULL)
582 			return (NULL);
583 
584 		aclp->z_next_ace = aclnode->z_acldata;
585 		aclp->z_curr_node = aclnode;
586 		aclnode->z_ace_idx = 0;
587 	}
588 
589 	aclnode = aclp->z_curr_node;
590 
591 	if (aclnode == NULL)
592 		return (NULL);
593 
594 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
595 		aclnode = list_next(&aclp->z_acl, aclnode);
596 		if (aclnode == NULL)
597 			return (NULL);
598 		else {
599 			aclp->z_curr_node = aclnode;
600 			aclnode->z_ace_idx = 0;
601 			aclp->z_next_ace = aclnode->z_acldata;
602 		}
603 	}
604 
605 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
606 		void *acep = aclp->z_next_ace;
607 		size_t ace_size;
608 
609 		/*
610 		 * Make sure we don't overstep our bounds
611 		 */
612 		ace_size = aclp->z_ops.ace_size(acep);
613 
614 		if (((caddr_t)acep + ace_size) >
615 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
616 			return (NULL);
617 		}
618 
619 		*iflags = aclp->z_ops.ace_flags_get(acep);
620 		*type = aclp->z_ops.ace_type_get(acep);
621 		*access_mask = aclp->z_ops.ace_mask_get(acep);
622 		*who = aclp->z_ops.ace_who_get(acep);
623 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
624 		aclnode->z_ace_idx++;
625 
626 		return ((void *)acep);
627 	}
628 	return (NULL);
629 }
630 
631 /*ARGSUSED*/
632 static uint64_t
633 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
634     uint16_t *flags, uint16_t *type, uint32_t *mask)
635 {
636 	zfs_acl_t *aclp = datap;
637 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
638 	uint64_t who;
639 
640 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
641 	    flags, type);
642 	return ((uint64_t)(uintptr_t)acep);
643 }
644 
645 static zfs_acl_node_t *
646 zfs_acl_curr_node(zfs_acl_t *aclp)
647 {
648 	ASSERT(aclp->z_curr_node);
649 	return (aclp->z_curr_node);
650 }
651 
652 /*
653  * Copy ACE to internal ZFS format.
654  * While processing the ACL each ACE will be validated for correctness.
655  * ACE FUIDs will be created later.
656  */
657 int
658 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
659     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
660     zfs_fuid_info_t **fuidp, cred_t *cr)
661 {
662 	int i;
663 	uint16_t entry_type;
664 	zfs_ace_t *aceptr = z_acl;
665 	ace_t *acep = datap;
666 	zfs_object_ace_t *zobjacep;
667 	ace_object_t *aceobjp;
668 
669 	for (i = 0; i != aclcnt; i++) {
670 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
671 		aceptr->z_hdr.z_flags = acep->a_flags;
672 		aceptr->z_hdr.z_type = acep->a_type;
673 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
674 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
675 		    entry_type != ACE_EVERYONE) {
676 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
677 			    cr, (entry_type == 0) ?
678 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
679 		}
680 
681 		/*
682 		 * Make sure ACE is valid
683 		 */
684 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
685 		    aceptr->z_hdr.z_flags) != B_TRUE)
686 			return (EINVAL);
687 
688 		switch (acep->a_type) {
689 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
690 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
691 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
692 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
693 			zobjacep = (zfs_object_ace_t *)aceptr;
694 			aceobjp = (ace_object_t *)acep;
695 
696 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
697 			    sizeof (aceobjp->a_obj_type));
698 			bcopy(aceobjp->a_inherit_obj_type,
699 			    zobjacep->z_inherit_type,
700 			    sizeof (aceobjp->a_inherit_obj_type));
701 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
702 			break;
703 		default:
704 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
705 		}
706 
707 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
708 		    aclp->z_ops.ace_size(aceptr));
709 	}
710 
711 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
712 
713 	return (0);
714 }
715 
716 /*
717  * Copy ZFS ACEs to fixed size ace_t layout
718  */
719 static void
720 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
721     void *datap, int filter)
722 {
723 	uint64_t who;
724 	uint32_t access_mask;
725 	uint16_t iflags, type;
726 	zfs_ace_hdr_t *zacep = NULL;
727 	ace_t *acep = datap;
728 	ace_object_t *objacep;
729 	zfs_object_ace_t *zobjacep;
730 	size_t ace_size;
731 	uint16_t entry_type;
732 
733 	while (zacep = zfs_acl_next_ace(aclp, zacep,
734 	    &who, &access_mask, &iflags, &type)) {
735 
736 		switch (type) {
737 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
738 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
739 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
740 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
741 			if (filter) {
742 				continue;
743 			}
744 			zobjacep = (zfs_object_ace_t *)zacep;
745 			objacep = (ace_object_t *)acep;
746 			bcopy(zobjacep->z_object_type,
747 			    objacep->a_obj_type,
748 			    sizeof (zobjacep->z_object_type));
749 			bcopy(zobjacep->z_inherit_type,
750 			    objacep->a_inherit_obj_type,
751 			    sizeof (zobjacep->z_inherit_type));
752 			ace_size = sizeof (ace_object_t);
753 			break;
754 		default:
755 			ace_size = sizeof (ace_t);
756 			break;
757 		}
758 
759 		entry_type = (iflags & ACE_TYPE_FLAGS);
760 		if ((entry_type != ACE_OWNER &&
761 		    entry_type != OWNING_GROUP &&
762 		    entry_type != ACE_EVERYONE)) {
763 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
764 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
765 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
766 		} else {
767 			acep->a_who = (uid_t)(int64_t)who;
768 		}
769 		acep->a_access_mask = access_mask;
770 		acep->a_flags = iflags;
771 		acep->a_type = type;
772 		acep = (ace_t *)((caddr_t)acep + ace_size);
773 	}
774 }
775 
776 static int
777 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
778     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
779 {
780 	int i;
781 	zfs_oldace_t *aceptr = z_acl;
782 
783 	for (i = 0; i != aclcnt; i++, aceptr++) {
784 		aceptr->z_access_mask = acep[i].a_access_mask;
785 		aceptr->z_type = acep[i].a_type;
786 		aceptr->z_flags = acep[i].a_flags;
787 		aceptr->z_fuid = acep[i].a_who;
788 		/*
789 		 * Make sure ACE is valid
790 		 */
791 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
792 		    aceptr->z_flags) != B_TRUE)
793 			return (EINVAL);
794 	}
795 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
796 	return (0);
797 }
798 
799 /*
800  * convert old ACL format to new
801  */
802 void
803 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
804 {
805 	zfs_oldace_t *oldaclp;
806 	int i;
807 	uint16_t type, iflags;
808 	uint32_t access_mask;
809 	uint64_t who;
810 	void *cookie = NULL;
811 	zfs_acl_node_t *newaclnode;
812 
813 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
814 	/*
815 	 * First create the ACE in a contiguous piece of memory
816 	 * for zfs_copy_ace_2_fuid().
817 	 *
818 	 * We only convert an ACL once, so this won't happen
819 	 * everytime.
820 	 */
821 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
822 	    KM_SLEEP);
823 	i = 0;
824 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
825 	    &access_mask, &iflags, &type)) {
826 		oldaclp[i].z_flags = iflags;
827 		oldaclp[i].z_type = type;
828 		oldaclp[i].z_fuid = who;
829 		oldaclp[i++].z_access_mask = access_mask;
830 	}
831 
832 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
833 	    sizeof (zfs_object_ace_t));
834 	aclp->z_ops = zfs_acl_fuid_ops;
835 	VERIFY(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
836 	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
837 	    &newaclnode->z_size, NULL, cr) == 0);
838 	newaclnode->z_ace_count = aclp->z_acl_count;
839 	aclp->z_version = ZFS_ACL_VERSION;
840 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
841 
842 	/*
843 	 * Release all previous ACL nodes
844 	 */
845 
846 	zfs_acl_release_nodes(aclp);
847 
848 	list_insert_head(&aclp->z_acl, newaclnode);
849 
850 	aclp->z_acl_bytes = newaclnode->z_size;
851 	aclp->z_acl_count = newaclnode->z_ace_count;
852 
853 }
854 
855 /*
856  * Convert unix access mask to v4 access mask
857  */
858 static uint32_t
859 zfs_unix_to_v4(uint32_t access_mask)
860 {
861 	uint32_t new_mask = 0;
862 
863 	if (access_mask & S_IXOTH)
864 		new_mask |= ACE_EXECUTE;
865 	if (access_mask & S_IWOTH)
866 		new_mask |= ACE_WRITE_DATA;
867 	if (access_mask & S_IROTH)
868 		new_mask |= ACE_READ_DATA;
869 	return (new_mask);
870 }
871 
872 static void
873 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
874     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
875 {
876 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
877 
878 	aclp->z_ops.ace_mask_set(acep, access_mask);
879 	aclp->z_ops.ace_type_set(acep, access_type);
880 	aclp->z_ops.ace_flags_set(acep, entry_type);
881 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
882 	    type != ACE_EVERYONE))
883 		aclp->z_ops.ace_who_set(acep, fuid);
884 }
885 
886 /*
887  * Determine mode of file based on ACL.
888  * Also, create FUIDs for any User/Group ACEs
889  */
890 uint64_t
891 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
892     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
893 {
894 	int		entry_type;
895 	mode_t		mode;
896 	mode_t		seen = 0;
897 	zfs_ace_hdr_t 	*acep = NULL;
898 	uint64_t	who;
899 	uint16_t	iflags, type;
900 	uint32_t	access_mask;
901 	boolean_t	an_exec_denied = B_FALSE;
902 
903 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
904 
905 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
906 	    &access_mask, &iflags, &type)) {
907 
908 		if (!zfs_acl_valid_ace_type(type, iflags))
909 			continue;
910 
911 		entry_type = (iflags & ACE_TYPE_FLAGS);
912 
913 		/*
914 		 * Skip over owner@, group@ or everyone@ inherit only ACEs
915 		 */
916 		if ((iflags & ACE_INHERIT_ONLY_ACE) &&
917 		    (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
918 		    entry_type == OWNING_GROUP))
919 			continue;
920 
921 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
922 		    who == fuid)) {
923 			if ((access_mask & ACE_READ_DATA) &&
924 			    (!(seen & S_IRUSR))) {
925 				seen |= S_IRUSR;
926 				if (type == ALLOW) {
927 					mode |= S_IRUSR;
928 				}
929 			}
930 			if ((access_mask & ACE_WRITE_DATA) &&
931 			    (!(seen & S_IWUSR))) {
932 				seen |= S_IWUSR;
933 				if (type == ALLOW) {
934 					mode |= S_IWUSR;
935 				}
936 			}
937 			if ((access_mask & ACE_EXECUTE) &&
938 			    (!(seen & S_IXUSR))) {
939 				seen |= S_IXUSR;
940 				if (type == ALLOW) {
941 					mode |= S_IXUSR;
942 				}
943 			}
944 		} else if (entry_type == OWNING_GROUP ||
945 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
946 			if ((access_mask & ACE_READ_DATA) &&
947 			    (!(seen & S_IRGRP))) {
948 				seen |= S_IRGRP;
949 				if (type == ALLOW) {
950 					mode |= S_IRGRP;
951 				}
952 			}
953 			if ((access_mask & ACE_WRITE_DATA) &&
954 			    (!(seen & S_IWGRP))) {
955 				seen |= S_IWGRP;
956 				if (type == ALLOW) {
957 					mode |= S_IWGRP;
958 				}
959 			}
960 			if ((access_mask & ACE_EXECUTE) &&
961 			    (!(seen & S_IXGRP))) {
962 				seen |= S_IXGRP;
963 				if (type == ALLOW) {
964 					mode |= S_IXGRP;
965 				}
966 			}
967 		} else if (entry_type == ACE_EVERYONE) {
968 			if ((access_mask & ACE_READ_DATA)) {
969 				if (!(seen & S_IRUSR)) {
970 					seen |= S_IRUSR;
971 					if (type == ALLOW) {
972 						mode |= S_IRUSR;
973 					}
974 				}
975 				if (!(seen & S_IRGRP)) {
976 					seen |= S_IRGRP;
977 					if (type == ALLOW) {
978 						mode |= S_IRGRP;
979 					}
980 				}
981 				if (!(seen & S_IROTH)) {
982 					seen |= S_IROTH;
983 					if (type == ALLOW) {
984 						mode |= S_IROTH;
985 					}
986 				}
987 			}
988 			if ((access_mask & ACE_WRITE_DATA)) {
989 				if (!(seen & S_IWUSR)) {
990 					seen |= S_IWUSR;
991 					if (type == ALLOW) {
992 						mode |= S_IWUSR;
993 					}
994 				}
995 				if (!(seen & S_IWGRP)) {
996 					seen |= S_IWGRP;
997 					if (type == ALLOW) {
998 						mode |= S_IWGRP;
999 					}
1000 				}
1001 				if (!(seen & S_IWOTH)) {
1002 					seen |= S_IWOTH;
1003 					if (type == ALLOW) {
1004 						mode |= S_IWOTH;
1005 					}
1006 				}
1007 			}
1008 			if ((access_mask & ACE_EXECUTE)) {
1009 				if (!(seen & S_IXUSR)) {
1010 					seen |= S_IXUSR;
1011 					if (type == ALLOW) {
1012 						mode |= S_IXUSR;
1013 					}
1014 				}
1015 				if (!(seen & S_IXGRP)) {
1016 					seen |= S_IXGRP;
1017 					if (type == ALLOW) {
1018 						mode |= S_IXGRP;
1019 					}
1020 				}
1021 				if (!(seen & S_IXOTH)) {
1022 					seen |= S_IXOTH;
1023 					if (type == ALLOW) {
1024 						mode |= S_IXOTH;
1025 					}
1026 				}
1027 			}
1028 		} else {
1029 			/*
1030 			 * Only care if this IDENTIFIER_GROUP or
1031 			 * USER ACE denies execute access to someone,
1032 			 * mode is not affected
1033 			 */
1034 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1035 				an_exec_denied = B_TRUE;
1036 		}
1037 	}
1038 
1039 	/*
1040 	 * Failure to allow is effectively a deny, so execute permission
1041 	 * is denied if it was never mentioned or if we explicitly
1042 	 * weren't allowed it.
1043 	 */
1044 	if (!an_exec_denied &&
1045 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1046 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1047 		an_exec_denied = B_TRUE;
1048 
1049 	if (an_exec_denied)
1050 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1051 	else
1052 		*pflags |= ZFS_NO_EXECS_DENIED;
1053 
1054 	return (mode);
1055 }
1056 
1057 /*
1058  * Read an external acl object.  If the intent is to modify, always
1059  * create a new acl and leave any cached acl in place.
1060  */
1061 static int
1062 zfs_acl_node_read(znode_t *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1063     boolean_t will_modify)
1064 {
1065 	zfs_acl_t	*aclp;
1066 	int		aclsize;
1067 	int		acl_count;
1068 	zfs_acl_node_t	*aclnode;
1069 	zfs_acl_phys_t	znode_acl;
1070 	int		version;
1071 	int		error;
1072 	boolean_t	drop_lock = B_FALSE;
1073 
1074 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1075 
1076 	if (zp->z_acl_cached && !will_modify) {
1077 		*aclpp = zp->z_acl_cached;
1078 		return (0);
1079 	}
1080 
1081 	/*
1082 	 * close race where znode could be upgrade while trying to
1083 	 * read the znode attributes.
1084 	 *
1085 	 * But this could only happen if the file isn't already an SA
1086 	 * znode
1087 	 */
1088 	if (!zp->z_is_sa && !have_lock) {
1089 		mutex_enter(&zp->z_lock);
1090 		drop_lock = B_TRUE;
1091 	}
1092 	version = zfs_znode_acl_version(zp);
1093 
1094 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1095 	    &acl_count, &znode_acl)) != 0) {
1096 		goto done;
1097 	}
1098 
1099 	aclp = zfs_acl_alloc(version);
1100 
1101 	aclp->z_acl_count = acl_count;
1102 	aclp->z_acl_bytes = aclsize;
1103 
1104 	aclnode = zfs_acl_node_alloc(aclsize);
1105 	aclnode->z_ace_count = aclp->z_acl_count;
1106 	aclnode->z_size = aclsize;
1107 
1108 	if (!zp->z_is_sa) {
1109 		if (znode_acl.z_acl_extern_obj) {
1110 			error = dmu_read(zp->z_zfsvfs->z_os,
1111 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1112 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1113 		} else {
1114 			bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1115 			    aclnode->z_size);
1116 		}
1117 	} else {
1118 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1119 		    aclnode->z_acldata, aclnode->z_size);
1120 	}
1121 
1122 	if (error != 0) {
1123 		zfs_acl_free(aclp);
1124 		zfs_acl_node_free(aclnode);
1125 		/* convert checksum errors into IO errors */
1126 		if (error == ECKSUM)
1127 			error = EIO;
1128 		goto done;
1129 	}
1130 
1131 	list_insert_head(&aclp->z_acl, aclnode);
1132 
1133 	*aclpp = aclp;
1134 	if (!will_modify)
1135 		zp->z_acl_cached = aclp;
1136 done:
1137 	if (drop_lock)
1138 		mutex_exit(&zp->z_lock);
1139 	return (error);
1140 }
1141 
1142 /*ARGSUSED*/
1143 void
1144 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1145     boolean_t start, void *userdata)
1146 {
1147 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1148 
1149 	if (start) {
1150 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1151 	} else {
1152 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1153 		    cb->cb_acl_node);
1154 	}
1155 	*dataptr = cb->cb_acl_node->z_acldata;
1156 	*length = cb->cb_acl_node->z_size;
1157 }
1158 
1159 int
1160 zfs_acl_chown_setattr(znode_t *zp)
1161 {
1162 	int error;
1163 	zfs_acl_t *aclp;
1164 
1165 	ASSERT(MUTEX_HELD(&zp->z_lock));
1166 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1167 
1168 	if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1169 		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1170 		    &zp->z_pflags, zp->z_uid, zp->z_gid);
1171 	return (error);
1172 }
1173 
1174 /*
1175  * common code for setting ACLs.
1176  *
1177  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1178  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1179  * already checked the acl and knows whether to inherit.
1180  */
1181 int
1182 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1183 {
1184 	int			error;
1185 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1186 	dmu_object_type_t	otype;
1187 	zfs_acl_locator_cb_t	locate = { 0 };
1188 	uint64_t		mode;
1189 	sa_bulk_attr_t		bulk[5];
1190 	uint64_t		ctime[2];
1191 	int			count = 0;
1192 
1193 	mode = zp->z_mode;
1194 
1195 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1196 	    zp->z_uid, zp->z_gid);
1197 
1198 	zp->z_mode = mode;
1199 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1200 	    &mode, sizeof (mode));
1201 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1202 	    &zp->z_pflags, sizeof (zp->z_pflags));
1203 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1204 	    &ctime, sizeof (ctime));
1205 
1206 	if (zp->z_acl_cached) {
1207 		zfs_acl_free(zp->z_acl_cached);
1208 		zp->z_acl_cached = NULL;
1209 	}
1210 
1211 	/*
1212 	 * Upgrade needed?
1213 	 */
1214 	if (!zfsvfs->z_use_fuids) {
1215 		otype = DMU_OT_OLDACL;
1216 	} else {
1217 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1218 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1219 			zfs_acl_xform(zp, aclp, cr);
1220 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1221 		otype = DMU_OT_ACL;
1222 	}
1223 
1224 	/*
1225 	 * Arrgh, we have to handle old on disk format
1226 	 * as well as newer (preferred) SA format.
1227 	 */
1228 
1229 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1230 		locate.cb_aclp = aclp;
1231 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1232 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1233 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1234 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1235 	} else { /* Painful legacy way */
1236 		zfs_acl_node_t *aclnode;
1237 		uint64_t off = 0;
1238 		zfs_acl_phys_t acl_phys;
1239 		uint64_t aoid;
1240 
1241 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1242 		    &acl_phys, sizeof (acl_phys))) != 0)
1243 			return (error);
1244 
1245 		aoid = acl_phys.z_acl_extern_obj;
1246 
1247 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1248 			/*
1249 			 * If ACL was previously external and we are now
1250 			 * converting to new ACL format then release old
1251 			 * ACL object and create a new one.
1252 			 */
1253 			if (aoid &&
1254 			    aclp->z_version != acl_phys.z_acl_version) {
1255 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1256 				if (error)
1257 					return (error);
1258 				aoid = 0;
1259 			}
1260 			if (aoid == 0) {
1261 				aoid = dmu_object_alloc(zfsvfs->z_os,
1262 				    otype, aclp->z_acl_bytes,
1263 				    otype == DMU_OT_ACL ?
1264 				    DMU_OT_SYSACL : DMU_OT_NONE,
1265 				    otype == DMU_OT_ACL ?
1266 				    DN_MAX_BONUSLEN : 0, tx);
1267 			} else {
1268 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1269 				    aoid, aclp->z_acl_bytes, 0, tx);
1270 			}
1271 			acl_phys.z_acl_extern_obj = aoid;
1272 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1273 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1274 				if (aclnode->z_ace_count == 0)
1275 					continue;
1276 				dmu_write(zfsvfs->z_os, aoid, off,
1277 				    aclnode->z_size, aclnode->z_acldata, tx);
1278 				off += aclnode->z_size;
1279 			}
1280 		} else {
1281 			void *start = acl_phys.z_ace_data;
1282 			/*
1283 			 * Migrating back embedded?
1284 			 */
1285 			if (acl_phys.z_acl_extern_obj) {
1286 				error = dmu_object_free(zfsvfs->z_os,
1287 				    acl_phys.z_acl_extern_obj, tx);
1288 				if (error)
1289 					return (error);
1290 				acl_phys.z_acl_extern_obj = 0;
1291 			}
1292 
1293 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1294 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1295 				if (aclnode->z_ace_count == 0)
1296 					continue;
1297 				bcopy(aclnode->z_acldata, start,
1298 				    aclnode->z_size);
1299 				start = (caddr_t)start + aclnode->z_size;
1300 			}
1301 		}
1302 		/*
1303 		 * If Old version then swap count/bytes to match old
1304 		 * layout of znode_acl_phys_t.
1305 		 */
1306 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1307 			acl_phys.z_acl_size = aclp->z_acl_count;
1308 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1309 		} else {
1310 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1311 			acl_phys.z_acl_count = aclp->z_acl_count;
1312 		}
1313 		acl_phys.z_acl_version = aclp->z_version;
1314 
1315 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1316 		    &acl_phys, sizeof (acl_phys));
1317 	}
1318 
1319 	/*
1320 	 * Replace ACL wide bits, but first clear them.
1321 	 */
1322 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1323 
1324 	zp->z_pflags |= aclp->z_hints;
1325 
1326 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1327 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1328 
1329 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime, B_TRUE);
1330 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1331 }
1332 
1333 /*
1334  * Update access mask for prepended ACE
1335  *
1336  * This applies the "groupmask" value for aclmode property.
1337  */
1338 static void
1339 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1340     mode_t mode, uint64_t owner)
1341 {
1342 	int	rmask, wmask, xmask;
1343 	int	user_ace;
1344 	uint16_t aceflags;
1345 	uint32_t origmask, acepmask;
1346 	uint64_t fuid;
1347 
1348 	aceflags = aclp->z_ops.ace_flags_get(acep);
1349 	fuid = aclp->z_ops.ace_who_get(acep);
1350 	origmask = aclp->z_ops.ace_mask_get(origacep);
1351 	acepmask = aclp->z_ops.ace_mask_get(acep);
1352 
1353 	user_ace = (!(aceflags &
1354 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1355 
1356 	if (user_ace && (fuid == owner)) {
1357 		rmask = S_IRUSR;
1358 		wmask = S_IWUSR;
1359 		xmask = S_IXUSR;
1360 	} else {
1361 		rmask = S_IRGRP;
1362 		wmask = S_IWGRP;
1363 		xmask = S_IXGRP;
1364 	}
1365 
1366 	if (origmask & ACE_READ_DATA) {
1367 		if (mode & rmask) {
1368 			acepmask &= ~ACE_READ_DATA;
1369 		} else {
1370 			acepmask |= ACE_READ_DATA;
1371 		}
1372 	}
1373 
1374 	if (origmask & ACE_WRITE_DATA) {
1375 		if (mode & wmask) {
1376 			acepmask &= ~ACE_WRITE_DATA;
1377 		} else {
1378 			acepmask |= ACE_WRITE_DATA;
1379 		}
1380 	}
1381 
1382 	if (origmask & ACE_APPEND_DATA) {
1383 		if (mode & wmask) {
1384 			acepmask &= ~ACE_APPEND_DATA;
1385 		} else {
1386 			acepmask |= ACE_APPEND_DATA;
1387 		}
1388 	}
1389 
1390 	if (origmask & ACE_EXECUTE) {
1391 		if (mode & xmask) {
1392 			acepmask &= ~ACE_EXECUTE;
1393 		} else {
1394 			acepmask |= ACE_EXECUTE;
1395 		}
1396 	}
1397 	aclp->z_ops.ace_mask_set(acep, acepmask);
1398 }
1399 
1400 static void
1401 zfs_acl_chmod(zfsvfs_t *zfsvfs, uint64_t mode, zfs_acl_t *aclp)
1402 {
1403 	void		*acep = NULL;
1404 	uint64_t	who;
1405 	int		new_count, new_bytes;
1406 	int		ace_size;
1407 	int 		entry_type;
1408 	uint16_t	iflags, type;
1409 	uint32_t	access_mask;
1410 	zfs_acl_node_t	*newnode;
1411 	size_t 		abstract_size = aclp->z_ops.ace_abstract_size();
1412 	void 		*zacep;
1413 	uint32_t 	owner, group, everyone;
1414 	uint32_t	deny1, deny2, allow0;
1415 
1416 	new_count = new_bytes = 0;
1417 
1418 	acl_trivial_access_masks((mode_t)mode, &allow0, &deny1, &deny2,
1419 	    &owner, &group, &everyone);
1420 
1421 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1422 
1423 	zacep = newnode->z_acldata;
1424 	if (allow0) {
1425 		zfs_set_ace(aclp, zacep, allow0, ALLOW, -1, ACE_OWNER);
1426 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1427 		new_count++;
1428 		new_bytes += abstract_size;
1429 	} if (deny1) {
1430 		zfs_set_ace(aclp, zacep, deny1, DENY, -1, ACE_OWNER);
1431 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1432 		new_count++;
1433 		new_bytes += abstract_size;
1434 	}
1435 	if (deny2) {
1436 		zfs_set_ace(aclp, zacep, deny2, DENY, -1, OWNING_GROUP);
1437 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1438 		new_count++;
1439 		new_bytes += abstract_size;
1440 	}
1441 
1442 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1443 	    &iflags, &type)) {
1444 		uint16_t inherit_flags;
1445 
1446 		entry_type = (iflags & ACE_TYPE_FLAGS);
1447 		inherit_flags = (iflags & ALL_INHERIT);
1448 
1449 		if ((entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1450 		    (entry_type == OWNING_GROUP)) &&
1451 		    ((inherit_flags & ACE_INHERIT_ONLY_ACE) == 0)) {
1452 			continue;
1453 		}
1454 
1455 		if ((type != ALLOW && type != DENY) ||
1456 		    (inherit_flags & ACE_INHERIT_ONLY_ACE)) {
1457 			if (inherit_flags)
1458 				aclp->z_hints |= ZFS_INHERIT_ACE;
1459 			switch (type) {
1460 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1461 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1462 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1463 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1464 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1465 				break;
1466 			}
1467 		} else {
1468 
1469 			/*
1470 			 * Limit permissions to be no greater than
1471 			 * group permissions
1472 			 */
1473 			if (zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) {
1474 				if (!(mode & S_IRGRP))
1475 					access_mask &= ~ACE_READ_DATA;
1476 				if (!(mode & S_IWGRP))
1477 					access_mask &=
1478 					    ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1479 				if (!(mode & S_IXGRP))
1480 					access_mask &= ~ACE_EXECUTE;
1481 				access_mask &=
1482 				    ~(ACE_WRITE_OWNER|ACE_WRITE_ACL|
1483 				    ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS);
1484 			}
1485 		}
1486 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1487 		ace_size = aclp->z_ops.ace_size(acep);
1488 		zacep = (void *)((uintptr_t)zacep + ace_size);
1489 		new_count++;
1490 		new_bytes += ace_size;
1491 	}
1492 	zfs_set_ace(aclp, zacep, owner, 0, -1, ACE_OWNER);
1493 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1494 	zfs_set_ace(aclp, zacep, group, 0, -1, OWNING_GROUP);
1495 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1496 	zfs_set_ace(aclp, zacep, everyone, 0, -1, ACE_EVERYONE);
1497 
1498 	new_count += 3;
1499 	new_bytes += abstract_size * 3;
1500 	zfs_acl_release_nodes(aclp);
1501 	aclp->z_acl_count = new_count;
1502 	aclp->z_acl_bytes = new_bytes;
1503 	newnode->z_ace_count = new_count;
1504 	newnode->z_size = new_bytes;
1505 	list_insert_tail(&aclp->z_acl, newnode);
1506 }
1507 
1508 void
1509 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1510 {
1511 	mutex_enter(&zp->z_acl_lock);
1512 	mutex_enter(&zp->z_lock);
1513 	*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1514 	(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1515 	zfs_acl_chmod(zp->z_zfsvfs, mode, *aclp);
1516 	mutex_exit(&zp->z_lock);
1517 	mutex_exit(&zp->z_acl_lock);
1518 	ASSERT(*aclp);
1519 }
1520 
1521 /*
1522  * strip off write_owner and write_acl
1523  */
1524 static void
1525 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1526 {
1527 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1528 
1529 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1530 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1531 		mask &= ~RESTRICTED_CLEAR;
1532 		aclp->z_ops.ace_mask_set(acep, mask);
1533 	}
1534 }
1535 
1536 /*
1537  * Should ACE be inherited?
1538  */
1539 static int
1540 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1541 {
1542 	int	iflags = (acep_flags & 0xf);
1543 
1544 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1545 		return (1);
1546 	else if (iflags & ACE_FILE_INHERIT_ACE)
1547 		return (!((vtype == VDIR) &&
1548 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1549 	return (0);
1550 }
1551 
1552 /*
1553  * inherit inheritable ACEs from parent
1554  */
1555 static zfs_acl_t *
1556 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1557     uint64_t mode, boolean_t *need_chmod)
1558 {
1559 	void		*pacep;
1560 	void		*acep;
1561 	zfs_acl_node_t  *aclnode;
1562 	zfs_acl_t	*aclp = NULL;
1563 	uint64_t	who;
1564 	uint32_t	access_mask;
1565 	uint16_t	iflags, newflags, type;
1566 	size_t		ace_size;
1567 	void		*data1, *data2;
1568 	size_t		data1sz, data2sz;
1569 	boolean_t	vdir = vtype == VDIR;
1570 	boolean_t	vreg = vtype == VREG;
1571 	boolean_t	passthrough, passthrough_x, noallow;
1572 
1573 	passthrough_x =
1574 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1575 	passthrough = passthrough_x ||
1576 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1577 	noallow =
1578 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1579 
1580 	*need_chmod = B_TRUE;
1581 	pacep = NULL;
1582 	aclp = zfs_acl_alloc(paclp->z_version);
1583 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD || vtype == VLNK)
1584 		return (aclp);
1585 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1586 	    &access_mask, &iflags, &type)) {
1587 
1588 		/*
1589 		 * don't inherit bogus ACEs
1590 		 */
1591 		if (!zfs_acl_valid_ace_type(type, iflags))
1592 			continue;
1593 
1594 		if (noallow && type == ALLOW)
1595 			continue;
1596 
1597 		ace_size = aclp->z_ops.ace_size(pacep);
1598 
1599 		if (!zfs_ace_can_use(vtype, iflags))
1600 			continue;
1601 
1602 		/*
1603 		 * If owner@, group@, or everyone@ inheritable
1604 		 * then zfs_acl_chmod() isn't needed.
1605 		 */
1606 		if (passthrough &&
1607 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1608 		    ((iflags & OWNING_GROUP) ==
1609 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1610 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1611 			*need_chmod = B_FALSE;
1612 		}
1613 
1614 		if (!vdir && passthrough_x &&
1615 		    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1616 			access_mask &= ~ACE_EXECUTE;
1617 		}
1618 
1619 		aclnode = zfs_acl_node_alloc(ace_size);
1620 		list_insert_tail(&aclp->z_acl, aclnode);
1621 		acep = aclnode->z_acldata;
1622 
1623 		zfs_set_ace(aclp, acep, access_mask, type,
1624 		    who, iflags|ACE_INHERITED_ACE);
1625 
1626 		/*
1627 		 * Copy special opaque data if any
1628 		 */
1629 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1630 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1631 			    &data2)) == data1sz);
1632 			bcopy(data1, data2, data2sz);
1633 		}
1634 
1635 		aclp->z_acl_count++;
1636 		aclnode->z_ace_count++;
1637 		aclp->z_acl_bytes += aclnode->z_size;
1638 		newflags = aclp->z_ops.ace_flags_get(acep);
1639 
1640 		if (vdir)
1641 			aclp->z_hints |= ZFS_INHERIT_ACE;
1642 
1643 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1644 			newflags &= ~ALL_INHERIT;
1645 			aclp->z_ops.ace_flags_set(acep,
1646 			    newflags|ACE_INHERITED_ACE);
1647 			zfs_restricted_update(zfsvfs, aclp, acep);
1648 			continue;
1649 		}
1650 
1651 		ASSERT(vdir);
1652 
1653 		/*
1654 		 * If only FILE_INHERIT is set then turn on
1655 		 * inherit_only
1656 		 */
1657 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1658 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1659 			newflags |= ACE_INHERIT_ONLY_ACE;
1660 			aclp->z_ops.ace_flags_set(acep,
1661 			    newflags|ACE_INHERITED_ACE);
1662 		} else {
1663 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1664 			aclp->z_ops.ace_flags_set(acep,
1665 			    newflags|ACE_INHERITED_ACE);
1666 		}
1667 	}
1668 	return (aclp);
1669 }
1670 
1671 /*
1672  * Create file system object initial permissions
1673  * including inheritable ACEs.
1674  */
1675 int
1676 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1677     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1678 {
1679 	int		error;
1680 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1681 	zfs_acl_t	*paclp;
1682 	gid_t		gid;
1683 	boolean_t	need_chmod = B_TRUE;
1684 	boolean_t	inherited = B_FALSE;
1685 
1686 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1687 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1688 
1689 	if (vsecp)
1690 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1691 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1692 			return (error);
1693 	/*
1694 	 * Determine uid and gid.
1695 	 */
1696 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1697 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1698 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1699 		    (uint64_t)vap->va_uid, cr,
1700 		    ZFS_OWNER, &acl_ids->z_fuidp);
1701 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1702 		    (uint64_t)vap->va_gid, cr,
1703 		    ZFS_GROUP, &acl_ids->z_fuidp);
1704 		gid = vap->va_gid;
1705 	} else {
1706 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1707 		    cr, &acl_ids->z_fuidp);
1708 		acl_ids->z_fgid = 0;
1709 		if (vap->va_mask & AT_GID)  {
1710 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1711 			    (uint64_t)vap->va_gid,
1712 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1713 			gid = vap->va_gid;
1714 			if (acl_ids->z_fgid != dzp->z_gid &&
1715 			    !groupmember(vap->va_gid, cr) &&
1716 			    secpolicy_vnode_create_gid(cr) != 0)
1717 				acl_ids->z_fgid = 0;
1718 		}
1719 		if (acl_ids->z_fgid == 0) {
1720 			if (dzp->z_mode & S_ISGID) {
1721 				char		*domain;
1722 				uint32_t	rid;
1723 
1724 				acl_ids->z_fgid = dzp->z_gid;
1725 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1726 				    cr, ZFS_GROUP);
1727 
1728 				if (zfsvfs->z_use_fuids &&
1729 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1730 					domain = zfs_fuid_idx_domain(
1731 					    &zfsvfs->z_fuid_idx,
1732 					    FUID_INDEX(acl_ids->z_fgid));
1733 					rid = FUID_RID(acl_ids->z_fgid);
1734 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1735 					    domain, rid,
1736 					    FUID_INDEX(acl_ids->z_fgid),
1737 					    acl_ids->z_fgid, ZFS_GROUP);
1738 				}
1739 			} else {
1740 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1741 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1742 				gid = crgetgid(cr);
1743 			}
1744 		}
1745 	}
1746 
1747 	/*
1748 	 * If we're creating a directory, and the parent directory has the
1749 	 * set-GID bit set, set in on the new directory.
1750 	 * Otherwise, if the user is neither privileged nor a member of the
1751 	 * file's new group, clear the file's set-GID bit.
1752 	 */
1753 
1754 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1755 	    (vap->va_type == VDIR)) {
1756 		acl_ids->z_mode |= S_ISGID;
1757 	} else {
1758 		if ((acl_ids->z_mode & S_ISGID) &&
1759 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1760 			acl_ids->z_mode &= ~S_ISGID;
1761 	}
1762 
1763 	if (acl_ids->z_aclp == NULL) {
1764 		mutex_enter(&dzp->z_acl_lock);
1765 		mutex_enter(&dzp->z_lock);
1766 		if (!(flag & IS_ROOT_NODE) && (ZTOV(dzp)->v_type == VDIR &&
1767 		    (dzp->z_pflags & ZFS_INHERIT_ACE)) &&
1768 		    !(dzp->z_pflags & ZFS_XATTR)) {
1769 			VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1770 			    &paclp, B_FALSE));
1771 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1772 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1773 			inherited = B_TRUE;
1774 		} else {
1775 			acl_ids->z_aclp =
1776 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1777 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1778 		}
1779 		mutex_exit(&dzp->z_lock);
1780 		mutex_exit(&dzp->z_acl_lock);
1781 		if (need_chmod) {
1782 			acl_ids->z_aclp->z_hints |= (vap->va_type == VDIR) ?
1783 			    ZFS_ACL_AUTO_INHERIT : 0;
1784 			zfs_acl_chmod(zfsvfs, acl_ids->z_mode, acl_ids->z_aclp);
1785 		}
1786 	}
1787 
1788 	if (inherited || vsecp) {
1789 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1790 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1791 		    acl_ids->z_fuid, acl_ids->z_fgid);
1792 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1793 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1794 	}
1795 
1796 	return (0);
1797 }
1798 
1799 /*
1800  * Free ACL and fuid_infop, but not the acl_ids structure
1801  */
1802 void
1803 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1804 {
1805 	if (acl_ids->z_aclp)
1806 		zfs_acl_free(acl_ids->z_aclp);
1807 	if (acl_ids->z_fuidp)
1808 		zfs_fuid_info_free(acl_ids->z_fuidp);
1809 	acl_ids->z_aclp = NULL;
1810 	acl_ids->z_fuidp = NULL;
1811 }
1812 
1813 boolean_t
1814 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1815 {
1816 	return (zfs_fuid_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1817 	    zfs_fuid_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1818 }
1819 
1820 /*
1821  * Retrieve a files ACL
1822  */
1823 int
1824 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1825 {
1826 	zfs_acl_t	*aclp;
1827 	ulong_t		mask;
1828 	int		error;
1829 	int 		count = 0;
1830 	int		largeace = 0;
1831 
1832 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1833 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1834 
1835 	if (mask == 0)
1836 		return (ENOSYS);
1837 
1838 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1839 		return (error);
1840 
1841 	mutex_enter(&zp->z_acl_lock);
1842 
1843 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1844 	if (error != 0) {
1845 		mutex_exit(&zp->z_acl_lock);
1846 		return (error);
1847 	}
1848 
1849 	/*
1850 	 * Scan ACL to determine number of ACEs
1851 	 */
1852 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1853 		void *zacep = NULL;
1854 		uint64_t who;
1855 		uint32_t access_mask;
1856 		uint16_t type, iflags;
1857 
1858 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1859 		    &who, &access_mask, &iflags, &type)) {
1860 			switch (type) {
1861 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1862 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1863 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1864 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1865 				largeace++;
1866 				continue;
1867 			default:
1868 				count++;
1869 			}
1870 		}
1871 		vsecp->vsa_aclcnt = count;
1872 	} else
1873 		count = (int)aclp->z_acl_count;
1874 
1875 	if (mask & VSA_ACECNT) {
1876 		vsecp->vsa_aclcnt = count;
1877 	}
1878 
1879 	if (mask & VSA_ACE) {
1880 		size_t aclsz;
1881 
1882 		aclsz = count * sizeof (ace_t) +
1883 		    sizeof (ace_object_t) * largeace;
1884 
1885 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1886 		vsecp->vsa_aclentsz = aclsz;
1887 
1888 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1889 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1890 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1891 		else {
1892 			zfs_acl_node_t *aclnode;
1893 			void *start = vsecp->vsa_aclentp;
1894 
1895 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1896 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1897 				bcopy(aclnode->z_acldata, start,
1898 				    aclnode->z_size);
1899 				start = (caddr_t)start + aclnode->z_size;
1900 			}
1901 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1902 			    aclp->z_acl_bytes);
1903 		}
1904 	}
1905 	if (mask & VSA_ACE_ACLFLAGS) {
1906 		vsecp->vsa_aclflags = 0;
1907 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1908 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1909 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
1910 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1911 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1912 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1913 	}
1914 
1915 	mutex_exit(&zp->z_acl_lock);
1916 
1917 	return (0);
1918 }
1919 
1920 int
1921 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
1922     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1923 {
1924 	zfs_acl_t *aclp;
1925 	zfs_acl_node_t *aclnode;
1926 	int aclcnt = vsecp->vsa_aclcnt;
1927 	int error;
1928 
1929 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1930 		return (EINVAL);
1931 
1932 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1933 
1934 	aclp->z_hints = 0;
1935 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1936 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1937 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1938 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1939 		    aclcnt, &aclnode->z_size)) != 0) {
1940 			zfs_acl_free(aclp);
1941 			zfs_acl_node_free(aclnode);
1942 			return (error);
1943 		}
1944 	} else {
1945 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1946 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1947 		    &aclnode->z_size, fuidp, cr)) != 0) {
1948 			zfs_acl_free(aclp);
1949 			zfs_acl_node_free(aclnode);
1950 			return (error);
1951 		}
1952 	}
1953 	aclp->z_acl_bytes = aclnode->z_size;
1954 	aclnode->z_ace_count = aclcnt;
1955 	aclp->z_acl_count = aclcnt;
1956 	list_insert_head(&aclp->z_acl, aclnode);
1957 
1958 	/*
1959 	 * If flags are being set then add them to z_hints
1960 	 */
1961 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1962 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
1963 			aclp->z_hints |= ZFS_ACL_PROTECTED;
1964 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1965 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
1966 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1967 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1968 	}
1969 
1970 	*zaclp = aclp;
1971 
1972 	return (0);
1973 }
1974 
1975 /*
1976  * Set a files ACL
1977  */
1978 int
1979 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1980 {
1981 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1982 	zilog_t		*zilog = zfsvfs->z_log;
1983 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1984 	dmu_tx_t	*tx;
1985 	int		error;
1986 	zfs_acl_t	*aclp;
1987 	zfs_fuid_info_t	*fuidp = NULL;
1988 	boolean_t	fuid_dirtied;
1989 	uint64_t	acl_obj;
1990 
1991 	if (mask == 0)
1992 		return (ENOSYS);
1993 
1994 	if (zp->z_pflags & ZFS_IMMUTABLE)
1995 		return (EPERM);
1996 
1997 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
1998 		return (error);
1999 
2000 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
2001 	    &aclp);
2002 	if (error)
2003 		return (error);
2004 
2005 	/*
2006 	 * If ACL wide flags aren't being set then preserve any
2007 	 * existing flags.
2008 	 */
2009 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2010 		aclp->z_hints |=
2011 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2012 	}
2013 top:
2014 	mutex_enter(&zp->z_acl_lock);
2015 	mutex_enter(&zp->z_lock);
2016 
2017 	tx = dmu_tx_create(zfsvfs->z_os);
2018 
2019 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2020 
2021 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2022 	if (fuid_dirtied)
2023 		zfs_fuid_txhold(zfsvfs, tx);
2024 
2025 	/*
2026 	 * If old version and ACL won't fit in bonus and we aren't
2027 	 * upgrading then take out necessary DMU holds
2028 	 */
2029 
2030 	if ((acl_obj = zfs_external_acl(zp)) != 0) {
2031 		if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2032 		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2033 			dmu_tx_hold_free(tx, acl_obj, 0,
2034 			    DMU_OBJECT_END);
2035 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2036 			    aclp->z_acl_bytes);
2037 		} else {
2038 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2039 		}
2040 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2041 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2042 	}
2043 
2044 	zfs_sa_upgrade_txholds(tx, zp);
2045 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2046 	if (error) {
2047 		mutex_exit(&zp->z_acl_lock);
2048 		mutex_exit(&zp->z_lock);
2049 
2050 		if (error == ERESTART) {
2051 			dmu_tx_wait(tx);
2052 			dmu_tx_abort(tx);
2053 			goto top;
2054 		}
2055 		dmu_tx_abort(tx);
2056 		zfs_acl_free(aclp);
2057 		return (error);
2058 	}
2059 
2060 	error = zfs_aclset_common(zp, aclp, cr, tx);
2061 	ASSERT(error == 0);
2062 	ASSERT(zp->z_acl_cached == NULL);
2063 	zp->z_acl_cached = aclp;
2064 
2065 	if (fuid_dirtied)
2066 		zfs_fuid_sync(zfsvfs, tx);
2067 
2068 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2069 
2070 	if (fuidp)
2071 		zfs_fuid_info_free(fuidp);
2072 	dmu_tx_commit(tx);
2073 done:
2074 	mutex_exit(&zp->z_lock);
2075 	mutex_exit(&zp->z_acl_lock);
2076 
2077 	return (error);
2078 }
2079 
2080 /*
2081  * Check accesses of interest (AoI) against attributes of the dataset
2082  * such as read-only.  Returns zero if no AoI conflict with dataset
2083  * attributes, otherwise an appropriate errno is returned.
2084  */
2085 static int
2086 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2087 {
2088 	if ((v4_mode & WRITE_MASK) &&
2089 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2090 	    (!IS_DEVVP(ZTOV(zp)) ||
2091 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2092 		return (EROFS);
2093 	}
2094 
2095 	/*
2096 	 * Only check for READONLY on non-directories.
2097 	 */
2098 	if ((v4_mode & WRITE_MASK_DATA) &&
2099 	    (((ZTOV(zp)->v_type != VDIR) &&
2100 	    (zp->z_pflags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2101 	    (ZTOV(zp)->v_type == VDIR &&
2102 	    (zp->z_pflags & ZFS_IMMUTABLE)))) {
2103 		return (EPERM);
2104 	}
2105 
2106 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2107 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2108 		return (EPERM);
2109 	}
2110 
2111 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2112 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2113 		return (EACCES);
2114 	}
2115 
2116 	return (0);
2117 }
2118 
2119 /*
2120  * The primary usage of this function is to loop through all of the
2121  * ACEs in the znode, determining what accesses of interest (AoI) to
2122  * the caller are allowed or denied.  The AoI are expressed as bits in
2123  * the working_mode parameter.  As each ACE is processed, bits covered
2124  * by that ACE are removed from the working_mode.  This removal
2125  * facilitates two things.  The first is that when the working mode is
2126  * empty (= 0), we know we've looked at all the AoI. The second is
2127  * that the ACE interpretation rules don't allow a later ACE to undo
2128  * something granted or denied by an earlier ACE.  Removing the
2129  * discovered access or denial enforces this rule.  At the end of
2130  * processing the ACEs, all AoI that were found to be denied are
2131  * placed into the working_mode, giving the caller a mask of denied
2132  * accesses.  Returns:
2133  *	0		if all AoI granted
2134  *	EACCESS 	if the denied mask is non-zero
2135  *	other error	if abnormal failure (e.g., IO error)
2136  *
2137  * A secondary usage of the function is to determine if any of the
2138  * AoI are granted.  If an ACE grants any access in
2139  * the working_mode, we immediately short circuit out of the function.
2140  * This mode is chosen by setting anyaccess to B_TRUE.  The
2141  * working_mode is not a denied access mask upon exit if the function
2142  * is used in this manner.
2143  */
2144 static int
2145 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2146     boolean_t anyaccess, cred_t *cr)
2147 {
2148 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2149 	zfs_acl_t	*aclp;
2150 	int		error;
2151 	uid_t		uid = crgetuid(cr);
2152 	uint64_t 	who;
2153 	uint16_t	type, iflags;
2154 	uint16_t	entry_type;
2155 	uint32_t	access_mask;
2156 	uint32_t	deny_mask = 0;
2157 	zfs_ace_hdr_t	*acep = NULL;
2158 	boolean_t	checkit;
2159 	uid_t		gowner;
2160 	uid_t		fowner;
2161 
2162 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2163 
2164 	mutex_enter(&zp->z_acl_lock);
2165 
2166 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2167 	if (error != 0) {
2168 		mutex_exit(&zp->z_acl_lock);
2169 		return (error);
2170 	}
2171 
2172 	ASSERT(zp->z_acl_cached);
2173 
2174 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2175 	    &iflags, &type)) {
2176 		uint32_t mask_matched;
2177 
2178 		if (!zfs_acl_valid_ace_type(type, iflags))
2179 			continue;
2180 
2181 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2182 			continue;
2183 
2184 		/* Skip ACE if it does not affect any AoI */
2185 		mask_matched = (access_mask & *working_mode);
2186 		if (!mask_matched)
2187 			continue;
2188 
2189 		entry_type = (iflags & ACE_TYPE_FLAGS);
2190 
2191 		checkit = B_FALSE;
2192 
2193 		switch (entry_type) {
2194 		case ACE_OWNER:
2195 			if (uid == fowner)
2196 				checkit = B_TRUE;
2197 			break;
2198 		case OWNING_GROUP:
2199 			who = gowner;
2200 			/*FALLTHROUGH*/
2201 		case ACE_IDENTIFIER_GROUP:
2202 			checkit = zfs_groupmember(zfsvfs, who, cr);
2203 			break;
2204 		case ACE_EVERYONE:
2205 			checkit = B_TRUE;
2206 			break;
2207 
2208 		/* USER Entry */
2209 		default:
2210 			if (entry_type == 0) {
2211 				uid_t newid;
2212 
2213 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2214 				    ZFS_ACE_USER);
2215 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2216 				    uid == newid)
2217 					checkit = B_TRUE;
2218 				break;
2219 			} else {
2220 				mutex_exit(&zp->z_acl_lock);
2221 				return (EIO);
2222 			}
2223 		}
2224 
2225 		if (checkit) {
2226 			if (type == DENY) {
2227 				DTRACE_PROBE3(zfs__ace__denies,
2228 				    znode_t *, zp,
2229 				    zfs_ace_hdr_t *, acep,
2230 				    uint32_t, mask_matched);
2231 				deny_mask |= mask_matched;
2232 			} else {
2233 				DTRACE_PROBE3(zfs__ace__allows,
2234 				    znode_t *, zp,
2235 				    zfs_ace_hdr_t *, acep,
2236 				    uint32_t, mask_matched);
2237 				if (anyaccess) {
2238 					mutex_exit(&zp->z_acl_lock);
2239 					return (0);
2240 				}
2241 			}
2242 			*working_mode &= ~mask_matched;
2243 		}
2244 
2245 		/* Are we done? */
2246 		if (*working_mode == 0)
2247 			break;
2248 	}
2249 
2250 	mutex_exit(&zp->z_acl_lock);
2251 
2252 	/* Put the found 'denies' back on the working mode */
2253 	if (deny_mask) {
2254 		*working_mode |= deny_mask;
2255 		return (EACCES);
2256 	} else if (*working_mode) {
2257 		return (-1);
2258 	}
2259 
2260 	return (0);
2261 }
2262 
2263 /*
2264  * Return true if any access whatsoever granted, we don't actually
2265  * care what access is granted.
2266  */
2267 boolean_t
2268 zfs_has_access(znode_t *zp, cred_t *cr)
2269 {
2270 	uint32_t have = ACE_ALL_PERMS;
2271 
2272 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2273 		uid_t owner;
2274 
2275 		owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2276 		return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2277 	}
2278 	return (B_TRUE);
2279 }
2280 
2281 static int
2282 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2283     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2284 {
2285 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2286 	int err;
2287 
2288 	*working_mode = v4_mode;
2289 	*check_privs = B_TRUE;
2290 
2291 	/*
2292 	 * Short circuit empty requests
2293 	 */
2294 	if (v4_mode == 0 || zfsvfs->z_replay) {
2295 		*working_mode = 0;
2296 		return (0);
2297 	}
2298 
2299 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2300 		*check_privs = B_FALSE;
2301 		return (err);
2302 	}
2303 
2304 	/*
2305 	 * The caller requested that the ACL check be skipped.  This
2306 	 * would only happen if the caller checked VOP_ACCESS() with a
2307 	 * 32 bit ACE mask and already had the appropriate permissions.
2308 	 */
2309 	if (skipaclchk) {
2310 		*working_mode = 0;
2311 		return (0);
2312 	}
2313 
2314 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2315 }
2316 
2317 static int
2318 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2319     cred_t *cr)
2320 {
2321 	if (*working_mode != ACE_WRITE_DATA)
2322 		return (EACCES);
2323 
2324 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2325 	    check_privs, B_FALSE, cr));
2326 }
2327 
2328 int
2329 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2330 {
2331 	boolean_t owner = B_FALSE;
2332 	boolean_t groupmbr = B_FALSE;
2333 	boolean_t is_attr;
2334 	uid_t uid = crgetuid(cr);
2335 	int error;
2336 
2337 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2338 		return (EACCES);
2339 
2340 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2341 	    (ZTOV(zdp)->v_type == VDIR));
2342 	if (is_attr)
2343 		goto slow;
2344 
2345 
2346 	mutex_enter(&zdp->z_acl_lock);
2347 
2348 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2349 		mutex_exit(&zdp->z_acl_lock);
2350 		return (0);
2351 	}
2352 
2353 	if (FUID_INDEX(zdp->z_uid) != 0 || FUID_INDEX(zdp->z_gid) != 0) {
2354 		mutex_exit(&zdp->z_acl_lock);
2355 		goto slow;
2356 	}
2357 
2358 	if (uid == zdp->z_uid) {
2359 		owner = B_TRUE;
2360 		if (zdp->z_mode & S_IXUSR) {
2361 			mutex_exit(&zdp->z_acl_lock);
2362 			return (0);
2363 		} else {
2364 			mutex_exit(&zdp->z_acl_lock);
2365 			goto slow;
2366 		}
2367 	}
2368 	if (groupmember(zdp->z_gid, cr)) {
2369 		groupmbr = B_TRUE;
2370 		if (zdp->z_mode & S_IXGRP) {
2371 			mutex_exit(&zdp->z_acl_lock);
2372 			return (0);
2373 		} else {
2374 			mutex_exit(&zdp->z_acl_lock);
2375 			goto slow;
2376 		}
2377 	}
2378 	if (!owner && !groupmbr) {
2379 		if (zdp->z_mode & S_IXOTH) {
2380 			mutex_exit(&zdp->z_acl_lock);
2381 			return (0);
2382 		}
2383 	}
2384 
2385 	mutex_exit(&zdp->z_acl_lock);
2386 
2387 slow:
2388 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2389 	ZFS_ENTER(zdp->z_zfsvfs);
2390 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2391 	ZFS_EXIT(zdp->z_zfsvfs);
2392 	return (error);
2393 }
2394 
2395 /*
2396  * Determine whether Access should be granted/denied.
2397  * The least priv subsytem is always consulted as a basic privilege
2398  * can define any form of access.
2399  */
2400 int
2401 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2402 {
2403 	uint32_t	working_mode;
2404 	int		error;
2405 	int		is_attr;
2406 	boolean_t 	check_privs;
2407 	znode_t		*xzp;
2408 	znode_t 	*check_zp = zp;
2409 	mode_t		needed_bits;
2410 	uid_t		owner;
2411 
2412 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2413 
2414 	/*
2415 	 * If attribute then validate against base file
2416 	 */
2417 	if (is_attr) {
2418 		uint64_t	parent;
2419 
2420 		if ((error = sa_lookup(zp->z_sa_hdl,
2421 		    SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2422 		    sizeof (parent))) != 0)
2423 			return (error);
2424 
2425 		if ((error = zfs_zget(zp->z_zfsvfs,
2426 		    parent, &xzp)) != 0)	{
2427 			return (error);
2428 		}
2429 
2430 		check_zp = xzp;
2431 
2432 		/*
2433 		 * fixup mode to map to xattr perms
2434 		 */
2435 
2436 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2437 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2438 			mode |= ACE_WRITE_NAMED_ATTRS;
2439 		}
2440 
2441 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2442 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2443 			mode |= ACE_READ_NAMED_ATTRS;
2444 		}
2445 	}
2446 
2447 	owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2448 	/*
2449 	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2450 	 * in needed_bits.  Map the bits mapped by working_mode (currently
2451 	 * missing) in missing_bits.
2452 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2453 	 * needed_bits.
2454 	 */
2455 	needed_bits = 0;
2456 
2457 	working_mode = mode;
2458 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2459 	    owner == crgetuid(cr))
2460 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2461 
2462 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2463 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2464 		needed_bits |= VREAD;
2465 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2466 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2467 		needed_bits |= VWRITE;
2468 	if (working_mode & ACE_EXECUTE)
2469 		needed_bits |= VEXEC;
2470 
2471 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2472 	    &check_privs, skipaclchk, cr)) == 0) {
2473 		if (is_attr)
2474 			VN_RELE(ZTOV(xzp));
2475 		return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2476 		    needed_bits, needed_bits));
2477 	}
2478 
2479 	if (error && !check_privs) {
2480 		if (is_attr)
2481 			VN_RELE(ZTOV(xzp));
2482 		return (error);
2483 	}
2484 
2485 	if (error && (flags & V_APPEND)) {
2486 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2487 	}
2488 
2489 	if (error && check_privs) {
2490 		mode_t		checkmode = 0;
2491 
2492 		/*
2493 		 * First check for implicit owner permission on
2494 		 * read_acl/read_attributes
2495 		 */
2496 
2497 		error = 0;
2498 		ASSERT(working_mode != 0);
2499 
2500 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2501 		    owner == crgetuid(cr)))
2502 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2503 
2504 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2505 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2506 			checkmode |= VREAD;
2507 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2508 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2509 			checkmode |= VWRITE;
2510 		if (working_mode & ACE_EXECUTE)
2511 			checkmode |= VEXEC;
2512 
2513 		error = secpolicy_vnode_access2(cr, ZTOV(check_zp), owner,
2514 		    needed_bits & ~checkmode, needed_bits);
2515 
2516 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2517 			error = secpolicy_vnode_chown(cr, owner);
2518 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2519 			error = secpolicy_vnode_setdac(cr, owner);
2520 
2521 		if (error == 0 && (working_mode &
2522 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2523 			error = secpolicy_vnode_remove(cr);
2524 
2525 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2526 			error = secpolicy_vnode_chown(cr, owner);
2527 		}
2528 		if (error == 0) {
2529 			/*
2530 			 * See if any bits other than those already checked
2531 			 * for are still present.  If so then return EACCES
2532 			 */
2533 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2534 				error = EACCES;
2535 			}
2536 		}
2537 	} else if (error == 0) {
2538 		error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2539 		    needed_bits, needed_bits);
2540 	}
2541 
2542 
2543 	if (is_attr)
2544 		VN_RELE(ZTOV(xzp));
2545 
2546 	return (error);
2547 }
2548 
2549 /*
2550  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2551  * native ACL format and call zfs_zaccess()
2552  */
2553 int
2554 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2555 {
2556 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2557 }
2558 
2559 /*
2560  * Access function for secpolicy_vnode_setattr
2561  */
2562 int
2563 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2564 {
2565 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2566 
2567 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2568 }
2569 
2570 static int
2571 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2572     mode_t available_perms, cred_t *cr)
2573 {
2574 	int error;
2575 	uid_t downer;
2576 
2577 	downer = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr, ZFS_OWNER);
2578 
2579 	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2580 	    downer, available_perms, VWRITE|VEXEC);
2581 
2582 	if (error == 0)
2583 		error = zfs_sticky_remove_access(dzp, zp, cr);
2584 
2585 	return (error);
2586 }
2587 
2588 /*
2589  * Determine whether Access should be granted/deny, without
2590  * consulting least priv subsystem.
2591  *
2592  *
2593  * The following chart is the recommended NFSv4 enforcement for
2594  * ability to delete an object.
2595  *
2596  *      -------------------------------------------------------
2597  *      |   Parent Dir  |           Target Object Permissions |
2598  *      |  permissions  |                                     |
2599  *      -------------------------------------------------------
2600  *      |               | ACL Allows | ACL Denies| Delete     |
2601  *      |               |  Delete    |  Delete   | unspecified|
2602  *      -------------------------------------------------------
2603  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2604  *      |  DELETE_CHILD |                                     |
2605  *      -------------------------------------------------------
2606  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2607  *      |  DELETE_CHILD |            |           |            |
2608  *      -------------------------------------------------------
2609  *      | ACL specifies |            |           |            |
2610  *      | only allow    | Permit     | Permit    | Permit     |
2611  *      | write and     |            |           |            |
2612  *      | execute       |            |           |            |
2613  *      -------------------------------------------------------
2614  *      | ACL denies    |            |           |            |
2615  *      | write and     | Permit     | Deny      | Deny       |
2616  *      | execute       |            |           |            |
2617  *      -------------------------------------------------------
2618  *         ^
2619  *         |
2620  *         No search privilege, can't even look up file?
2621  *
2622  */
2623 int
2624 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2625 {
2626 	uint32_t dzp_working_mode = 0;
2627 	uint32_t zp_working_mode = 0;
2628 	int dzp_error, zp_error;
2629 	mode_t available_perms;
2630 	boolean_t dzpcheck_privs = B_TRUE;
2631 	boolean_t zpcheck_privs = B_TRUE;
2632 
2633 	/*
2634 	 * We want specific DELETE permissions to
2635 	 * take precedence over WRITE/EXECUTE.  We don't
2636 	 * want an ACL such as this to mess us up.
2637 	 * user:joe:write_data:deny,user:joe:delete:allow
2638 	 *
2639 	 * However, deny permissions may ultimately be overridden
2640 	 * by secpolicy_vnode_access().
2641 	 *
2642 	 * We will ask for all of the necessary permissions and then
2643 	 * look at the working modes from the directory and target object
2644 	 * to determine what was found.
2645 	 */
2646 
2647 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2648 		return (EPERM);
2649 
2650 	/*
2651 	 * First row
2652 	 * If the directory permissions allow the delete, we are done.
2653 	 */
2654 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2655 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2656 		return (0);
2657 
2658 	/*
2659 	 * If target object has delete permission then we are done
2660 	 */
2661 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2662 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2663 		return (0);
2664 
2665 	ASSERT(dzp_error && zp_error);
2666 
2667 	if (!dzpcheck_privs)
2668 		return (dzp_error);
2669 	if (!zpcheck_privs)
2670 		return (zp_error);
2671 
2672 	/*
2673 	 * Second row
2674 	 *
2675 	 * If directory returns EACCES then delete_child was denied
2676 	 * due to deny delete_child.  In this case send the request through
2677 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2678 	 * since that *could* allow the delete based on write/execute permission
2679 	 * and we want delete permissions to override write/execute.
2680 	 */
2681 
2682 	if (dzp_error == EACCES)
2683 		return (secpolicy_vnode_remove(cr));
2684 
2685 	/*
2686 	 * Third Row
2687 	 * only need to see if we have write/execute on directory.
2688 	 */
2689 
2690 	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2691 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2692 
2693 	if (dzp_error != 0 && !dzpcheck_privs)
2694 		return (dzp_error);
2695 
2696 	/*
2697 	 * Fourth row
2698 	 */
2699 
2700 	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2701 	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2702 
2703 	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2704 
2705 }
2706 
2707 int
2708 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2709     znode_t *tzp, cred_t *cr)
2710 {
2711 	int add_perm;
2712 	int error;
2713 
2714 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2715 		return (EACCES);
2716 
2717 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2718 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2719 
2720 	/*
2721 	 * Rename permissions are combination of delete permission +
2722 	 * add file/subdir permission.
2723 	 */
2724 
2725 	/*
2726 	 * first make sure we do the delete portion.
2727 	 *
2728 	 * If that succeeds then check for add_file/add_subdir permissions
2729 	 */
2730 
2731 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2732 		return (error);
2733 
2734 	/*
2735 	 * If we have a tzp, see if we can delete it?
2736 	 */
2737 	if (tzp) {
2738 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2739 			return (error);
2740 	}
2741 
2742 	/*
2743 	 * Now check for add permissions
2744 	 */
2745 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2746 
2747 	return (error);
2748 }
2749