xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_acl.c (revision 2dd2efa5a06a9befe46075cf41e16f57533c9f98)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/vfs.h>
35 #include <sys/vnode.h>
36 #include <sys/sid.h>
37 #include <sys/file.h>
38 #include <sys/stat.h>
39 #include <sys/kmem.h>
40 #include <sys/cmn_err.h>
41 #include <sys/errno.h>
42 #include <sys/unistd.h>
43 #include <sys/sdt.h>
44 #include <sys/fs/zfs.h>
45 #include <sys/mode.h>
46 #include <sys/policy.h>
47 #include <sys/zfs_znode.h>
48 #include <sys/zfs_fuid.h>
49 #include <sys/zfs_acl.h>
50 #include <sys/zfs_dir.h>
51 #include <sys/zfs_vfsops.h>
52 #include <sys/dmu.h>
53 #include <sys/dnode.h>
54 #include <sys/zap.h>
55 #include "fs/fs_subr.h"
56 #include <acl/acl_common.h>
57 
58 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
59 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
60 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
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 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
70 
71 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
72     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
73     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
74     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
75 
76 #define	WRITE_MASK (WRITE_MASK_DATA|ACE_WRITE_ATTRIBUTES|ACE_WRITE_ACL|\
77     ACE_WRITE_OWNER)
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	SECURE_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 static uint16_t
97 zfs_ace_v0_get_type(void *acep)
98 {
99 	return (((zfs_oldace_t *)acep)->z_type);
100 }
101 
102 static uint16_t
103 zfs_ace_v0_get_flags(void *acep)
104 {
105 	return (((zfs_oldace_t *)acep)->z_flags);
106 }
107 
108 static uint32_t
109 zfs_ace_v0_get_mask(void *acep)
110 {
111 	return (((zfs_oldace_t *)acep)->z_access_mask);
112 }
113 
114 static uint64_t
115 zfs_ace_v0_get_who(void *acep)
116 {
117 	return (((zfs_oldace_t *)acep)->z_fuid);
118 }
119 
120 static void
121 zfs_ace_v0_set_type(void *acep, uint16_t type)
122 {
123 	((zfs_oldace_t *)acep)->z_type = type;
124 }
125 
126 static void
127 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
128 {
129 	((zfs_oldace_t *)acep)->z_flags = flags;
130 }
131 
132 static void
133 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
134 {
135 	((zfs_oldace_t *)acep)->z_access_mask = mask;
136 }
137 
138 static void
139 zfs_ace_v0_set_who(void *acep, uint64_t who)
140 {
141 	((zfs_oldace_t *)acep)->z_fuid = who;
142 }
143 
144 /*ARGSUSED*/
145 static size_t
146 zfs_ace_v0_size(void *acep)
147 {
148 	return (sizeof (zfs_oldace_t));
149 }
150 
151 static size_t
152 zfs_ace_v0_abstract_size(void)
153 {
154 	return (sizeof (zfs_oldace_t));
155 }
156 
157 static int
158 zfs_ace_v0_mask_off(void)
159 {
160 	return (offsetof(zfs_oldace_t, z_access_mask));
161 }
162 
163 /*ARGSUSED*/
164 static int
165 zfs_ace_v0_data(void *acep, void **datap)
166 {
167 	*datap = NULL;
168 	return (0);
169 }
170 
171 static acl_ops_t zfs_acl_v0_ops = {
172 	zfs_ace_v0_get_mask,
173 	zfs_ace_v0_set_mask,
174 	zfs_ace_v0_get_flags,
175 	zfs_ace_v0_set_flags,
176 	zfs_ace_v0_get_type,
177 	zfs_ace_v0_set_type,
178 	zfs_ace_v0_get_who,
179 	zfs_ace_v0_set_who,
180 	zfs_ace_v0_size,
181 	zfs_ace_v0_abstract_size,
182 	zfs_ace_v0_mask_off,
183 	zfs_ace_v0_data
184 };
185 
186 static uint16_t
187 zfs_ace_fuid_get_type(void *acep)
188 {
189 	return (((zfs_ace_hdr_t *)acep)->z_type);
190 }
191 
192 static uint16_t
193 zfs_ace_fuid_get_flags(void *acep)
194 {
195 	return (((zfs_ace_hdr_t *)acep)->z_flags);
196 }
197 
198 static uint32_t
199 zfs_ace_fuid_get_mask(void *acep)
200 {
201 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
202 }
203 
204 static uint64_t
205 zfs_ace_fuid_get_who(void *args)
206 {
207 	uint16_t entry_type;
208 	zfs_ace_t *acep = args;
209 
210 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
211 
212 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
213 	    entry_type == ACE_EVERYONE)
214 		return (-1);
215 	return (((zfs_ace_t *)acep)->z_fuid);
216 }
217 
218 static void
219 zfs_ace_fuid_set_type(void *acep, uint16_t type)
220 {
221 	((zfs_ace_hdr_t *)acep)->z_type = type;
222 }
223 
224 static void
225 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
226 {
227 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
228 }
229 
230 static void
231 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
232 {
233 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
234 }
235 
236 static void
237 zfs_ace_fuid_set_who(void *arg, uint64_t who)
238 {
239 	zfs_ace_t *acep = arg;
240 
241 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
242 
243 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
244 	    entry_type == ACE_EVERYONE)
245 		return;
246 	acep->z_fuid = who;
247 }
248 
249 static size_t
250 zfs_ace_fuid_size(void *acep)
251 {
252 	zfs_ace_hdr_t *zacep = acep;
253 	uint16_t entry_type;
254 
255 	switch (zacep->z_type) {
256 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
257 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
258 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
259 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
260 		return (sizeof (zfs_object_ace_t));
261 	case ALLOW:
262 	case DENY:
263 		entry_type =
264 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
265 		if (entry_type == ACE_OWNER ||
266 		    entry_type == (ACE_GROUP | ACE_IDENTIFIER_GROUP) ||
267 		    entry_type == ACE_EVERYONE)
268 			return (sizeof (zfs_ace_hdr_t));
269 		/*FALLTHROUGH*/
270 	default:
271 		return (sizeof (zfs_ace_t));
272 	}
273 }
274 
275 static size_t
276 zfs_ace_fuid_abstract_size(void)
277 {
278 	return (sizeof (zfs_ace_hdr_t));
279 }
280 
281 static int
282 zfs_ace_fuid_mask_off(void)
283 {
284 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
285 }
286 
287 static int
288 zfs_ace_fuid_data(void *acep, void **datap)
289 {
290 	zfs_ace_t *zacep = acep;
291 	zfs_object_ace_t *zobjp;
292 
293 	switch (zacep->z_hdr.z_type) {
294 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
295 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
296 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
297 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
298 		zobjp = acep;
299 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
300 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
301 	default:
302 		*datap = NULL;
303 		return (0);
304 	}
305 }
306 
307 static acl_ops_t zfs_acl_fuid_ops = {
308 	zfs_ace_fuid_get_mask,
309 	zfs_ace_fuid_set_mask,
310 	zfs_ace_fuid_get_flags,
311 	zfs_ace_fuid_set_flags,
312 	zfs_ace_fuid_get_type,
313 	zfs_ace_fuid_set_type,
314 	zfs_ace_fuid_get_who,
315 	zfs_ace_fuid_set_who,
316 	zfs_ace_fuid_size,
317 	zfs_ace_fuid_abstract_size,
318 	zfs_ace_fuid_mask_off,
319 	zfs_ace_fuid_data
320 };
321 
322 static int
323 zfs_acl_version(int version)
324 {
325 	if (version < ZPL_VERSION_FUID)
326 		return (ZFS_ACL_VERSION_INITIAL);
327 	else
328 		return (ZFS_ACL_VERSION_FUID);
329 }
330 
331 static int
332 zfs_acl_version_zp(znode_t *zp)
333 {
334 	return (zfs_acl_version(zp->z_zfsvfs->z_version));
335 }
336 
337 static zfs_acl_t *
338 zfs_acl_alloc(int vers)
339 {
340 	zfs_acl_t *aclp;
341 
342 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
343 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
344 	    offsetof(zfs_acl_node_t, z_next));
345 	aclp->z_version = vers;
346 	if (vers == ZFS_ACL_VERSION_FUID)
347 		aclp->z_ops = zfs_acl_fuid_ops;
348 	else
349 		aclp->z_ops = zfs_acl_v0_ops;
350 	return (aclp);
351 }
352 
353 static zfs_acl_node_t *
354 zfs_acl_node_alloc(size_t bytes)
355 {
356 	zfs_acl_node_t *aclnode;
357 
358 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
359 	if (bytes) {
360 		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
361 		aclnode->z_allocdata = aclnode->z_acldata;
362 		aclnode->z_allocsize = bytes;
363 		aclnode->z_size = bytes;
364 	}
365 
366 	return (aclnode);
367 }
368 
369 static void
370 zfs_acl_node_free(zfs_acl_node_t *aclnode)
371 {
372 	if (aclnode->z_allocsize)
373 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
374 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
375 }
376 
377 static void
378 zfs_acl_release_nodes(zfs_acl_t *aclp)
379 {
380 	zfs_acl_node_t *aclnode;
381 
382 	while (aclnode = list_head(&aclp->z_acl)) {
383 		list_remove(&aclp->z_acl, aclnode);
384 		zfs_acl_node_free(aclnode);
385 	}
386 	aclp->z_acl_count = 0;
387 	aclp->z_acl_bytes = 0;
388 }
389 
390 void
391 zfs_acl_free(zfs_acl_t *aclp)
392 {
393 	zfs_acl_release_nodes(aclp);
394 	list_destroy(&aclp->z_acl);
395 	kmem_free(aclp, sizeof (zfs_acl_t));
396 }
397 
398 static boolean_t
399 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
400 {
401 	/*
402 	 * first check type of entry
403 	 */
404 
405 	switch (iflags & ACE_TYPE_FLAGS) {
406 	case ACE_OWNER:
407 	case (ACE_IDENTIFIER_GROUP | ACE_GROUP):
408 	case ACE_IDENTIFIER_GROUP:
409 	case ACE_EVERYONE:
410 	case 0:	/* User entry */
411 		break;
412 	default:
413 		return (B_FALSE);
414 
415 	}
416 
417 	/*
418 	 * next check inheritance level flags
419 	 */
420 
421 	if (type != ALLOW && type > MAX_ACE_TYPE) {
422 		return (B_FALSE);
423 	}
424 
425 	switch (type) {
426 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
427 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
428 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
429 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
430 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
431 			return (B_FALSE);
432 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
433 	}
434 
435 	/*
436 	 * Only directories should have inheritance flags.
437 	 */
438 	if (obj_type != VDIR && (iflags &
439 	    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE|
440 	    ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE))) {
441 		return (B_FALSE);
442 	}
443 
444 	if (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE))
445 		aclp->z_hints |= ZFS_INHERIT_ACE;
446 
447 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
448 		if ((iflags & (ACE_FILE_INHERIT_ACE|
449 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
450 			return (B_FALSE);
451 		}
452 	}
453 
454 	return (B_TRUE);
455 }
456 
457 static void *
458 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
459     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
460 {
461 	zfs_acl_node_t *aclnode;
462 
463 	if (start == NULL) {
464 		aclnode = list_head(&aclp->z_acl);
465 		if (aclnode == NULL)
466 			return (NULL);
467 
468 		aclp->z_next_ace = aclnode->z_acldata;
469 		aclp->z_curr_node = aclnode;
470 		aclnode->z_ace_idx = 0;
471 	}
472 
473 	aclnode = aclp->z_curr_node;
474 
475 	if (aclnode == NULL)
476 		return (NULL);
477 
478 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
479 		aclnode = list_next(&aclp->z_acl, aclnode);
480 		if (aclnode == NULL)
481 			return (NULL);
482 		else {
483 			aclp->z_curr_node = aclnode;
484 			aclnode->z_ace_idx = 0;
485 			aclp->z_next_ace = aclnode->z_acldata;
486 		}
487 	}
488 
489 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
490 		void *acep = aclp->z_next_ace;
491 		*iflags = aclp->z_ops.ace_flags_get(acep);
492 		*type = aclp->z_ops.ace_type_get(acep);
493 		*access_mask = aclp->z_ops.ace_mask_get(acep);
494 		*who = aclp->z_ops.ace_who_get(acep);
495 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace +
496 		    aclp->z_ops.ace_size(acep);
497 		aclnode->z_ace_idx++;
498 		return ((void *)acep);
499 	}
500 	return (NULL);
501 }
502 
503 /*ARGSUSED*/
504 static uint64_t
505 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
506     uint16_t *flags, uint16_t *type, uint32_t *mask)
507 {
508 	zfs_acl_t *aclp = datap;
509 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
510 	uint64_t who;
511 
512 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
513 	    flags, type);
514 	return ((uint64_t)(uintptr_t)acep);
515 }
516 
517 static zfs_acl_node_t *
518 zfs_acl_curr_node(zfs_acl_t *aclp)
519 {
520 	ASSERT(aclp->z_curr_node);
521 	return (aclp->z_curr_node);
522 }
523 
524 /*
525  * Copy ACE to internal ZFS format.
526  * While processing the ACL each ACE will be validated for correctness.
527  * ACE FUIDs will be created later.
528  */
529 int
530 zfs_copy_ace_2_fuid(vtype_t obj_type, zfs_acl_t *aclp, void *datap,
531     zfs_ace_t *z_acl, int aclcnt, size_t *size)
532 {
533 	int i;
534 	uint16_t entry_type;
535 	zfs_ace_t *aceptr = z_acl;
536 	ace_t *acep = datap;
537 	zfs_object_ace_t *zobjacep;
538 	ace_object_t *aceobjp;
539 
540 	for (i = 0; i != aclcnt; i++) {
541 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
542 		aceptr->z_hdr.z_flags = acep->a_flags;
543 		aceptr->z_hdr.z_type = acep->a_type;
544 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
545 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
546 		    entry_type != ACE_EVERYONE) {
547 			if (!aclp->z_has_fuids)
548 				aclp->z_has_fuids = IS_EPHEMERAL(acep->a_who) ?
549 				    B_TRUE : B_FALSE;
550 			aceptr->z_fuid = (uint64_t)acep->a_who;
551 		}
552 
553 		/*
554 		 * Make sure ACE is valid
555 		 */
556 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
557 		    aceptr->z_hdr.z_flags) != B_TRUE)
558 			return (EINVAL);
559 
560 		switch (acep->a_type) {
561 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
562 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
563 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
564 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
565 			zobjacep = (zfs_object_ace_t *)aceptr;
566 			aceobjp = (ace_object_t *)acep;
567 
568 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
569 			    sizeof (aceobjp->a_obj_type));
570 			bcopy(aceobjp->a_inherit_obj_type,
571 			    zobjacep->z_inherit_type,
572 			    sizeof (aceobjp->a_inherit_obj_type));
573 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
574 			break;
575 		default:
576 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
577 		}
578 
579 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
580 		    aclp->z_ops.ace_size(aceptr));
581 	}
582 
583 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
584 
585 	return (0);
586 }
587 
588 /*
589  * Copy ZFS ACEs to fixed size ace_t layout
590  */
591 static void
592 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
593     void *datap, int filter)
594 {
595 	uint64_t who;
596 	uint32_t access_mask;
597 	uint16_t iflags, type;
598 	zfs_ace_hdr_t *zacep = NULL;
599 	ace_t *acep = datap;
600 	ace_object_t *objacep;
601 	zfs_object_ace_t *zobjacep;
602 	zfs_fuid_hdl_t hdl = { 0 };
603 	size_t ace_size;
604 	uint16_t entry_type;
605 
606 	while (zacep = zfs_acl_next_ace(aclp, zacep,
607 	    &who, &access_mask, &iflags, &type)) {
608 
609 		switch (type) {
610 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
611 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
612 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
613 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
614 			if (filter) {
615 				continue;
616 			}
617 			zobjacep = (zfs_object_ace_t *)zacep;
618 			objacep = (ace_object_t *)acep;
619 			bcopy(zobjacep->z_object_type,
620 			    objacep->a_obj_type,
621 			    sizeof (zobjacep->z_object_type));
622 			bcopy(zobjacep->z_inherit_type,
623 			    objacep->a_inherit_obj_type,
624 			    sizeof (zobjacep->z_inherit_type));
625 			ace_size = sizeof (ace_object_t);
626 			break;
627 		default:
628 			ace_size = sizeof (ace_t);
629 			break;
630 		}
631 
632 		entry_type = (iflags & ACE_TYPE_FLAGS);
633 		if ((entry_type != ACE_OWNER &&
634 		    entry_type != (ACE_GROUP | ACE_IDENTIFIER_GROUP) &&
635 		    entry_type != ACE_EVERYONE))
636 			zfs_fuid_queue_map_id(zfsvfs, &hdl, who, cr,
637 			    (entry_type & ACE_IDENTIFIER_GROUP) ?
638 			    ZFS_ACE_GROUP : ZFS_ACE_USER, &acep->a_who);
639 		else
640 			acep->a_who = (uid_t)(int64_t)who;
641 		acep->a_access_mask = access_mask;
642 		acep->a_flags = iflags;
643 		acep->a_type = type;
644 		acep = (ace_t *)((caddr_t)acep + ace_size);
645 	}
646 	zfs_fuid_get_mappings(&hdl);
647 }
648 
649 static int
650 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
651     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
652 {
653 	int i;
654 	zfs_oldace_t *aceptr = z_acl;
655 
656 	for (i = 0; i != aclcnt; i++, aceptr++) {
657 		aceptr->z_access_mask = acep[i].a_access_mask;
658 		aceptr->z_type = acep[i].a_type;
659 		aceptr->z_flags = acep[i].a_flags;
660 		aceptr->z_fuid = acep[i].a_who;
661 		/*
662 		 * Make sure ACE is valid
663 		 */
664 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
665 		    aceptr->z_flags) != B_TRUE)
666 			return (EINVAL);
667 	}
668 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
669 	return (0);
670 }
671 
672 /*
673  * convert old ACL format to new
674  */
675 void
676 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp)
677 {
678 	zfs_oldace_t *oldaclp;
679 	int i;
680 	uint16_t type, iflags;
681 	uint32_t access_mask;
682 	uint64_t who;
683 	void *cookie = NULL;
684 	zfs_acl_node_t *newaclnode;
685 
686 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
687 	/*
688 	 * First create the ACE in a contiguous piece of memory
689 	 * for zfs_copy_ace_2_fuid().
690 	 *
691 	 * We only convert an ACL once, so this won't happen
692 	 * everytime.
693 	 */
694 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
695 	    KM_SLEEP);
696 	i = 0;
697 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
698 	    &access_mask, &iflags, &type)) {
699 		oldaclp[i].z_flags = iflags;
700 		oldaclp[i].z_type = type;
701 		oldaclp[i].z_fuid = who;
702 		oldaclp[i++].z_access_mask = access_mask;
703 	}
704 
705 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
706 	    sizeof (zfs_object_ace_t));
707 	aclp->z_ops = zfs_acl_fuid_ops;
708 	VERIFY(zfs_copy_ace_2_fuid(ZTOV(zp)->v_type, aclp, oldaclp,
709 	    newaclnode->z_acldata, aclp->z_acl_count,
710 	    &newaclnode->z_size) == 0);
711 	newaclnode->z_ace_count = aclp->z_acl_count;
712 	aclp->z_version = ZFS_ACL_VERSION;
713 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
714 
715 	/*
716 	 * Release all previous ACL nodes
717 	 */
718 
719 	zfs_acl_release_nodes(aclp);
720 
721 	list_insert_head(&aclp->z_acl, newaclnode);
722 
723 	aclp->z_acl_bytes = newaclnode->z_size;
724 	aclp->z_acl_count = newaclnode->z_ace_count;
725 
726 }
727 
728 /*
729  * Convert unix access mask to v4 access mask
730  */
731 static uint32_t
732 zfs_unix_to_v4(uint32_t access_mask)
733 {
734 	uint32_t new_mask = 0;
735 
736 	if (access_mask & S_IXOTH)
737 		new_mask |= ACE_EXECUTE;
738 	if (access_mask & S_IWOTH)
739 		new_mask |= ACE_WRITE_DATA;
740 	if (access_mask & S_IROTH)
741 		new_mask |= ACE_READ_DATA;
742 	return (new_mask);
743 }
744 
745 static void
746 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
747     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
748 {
749 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
750 
751 	aclp->z_ops.ace_mask_set(acep, access_mask);
752 	aclp->z_ops.ace_type_set(acep, access_type);
753 	aclp->z_ops.ace_flags_set(acep, entry_type);
754 	if ((type != ACE_OWNER && type != (ACE_GROUP | ACE_IDENTIFIER_GROUP) &&
755 	    type != ACE_EVERYONE))
756 		aclp->z_ops.ace_who_set(acep, fuid);
757 }
758 
759 /*
760  * Determine mode of file based on ACL.
761  * Also, create FUIDs for any User/Group ACEs
762  */
763 static uint64_t
764 zfs_mode_fuid_compute(znode_t *zp, zfs_acl_t *aclp, cred_t *cr,
765     zfs_fuid_info_t **fuidp, dmu_tx_t *tx)
766 {
767 	int		entry_type;
768 	mode_t		mode;
769 	mode_t		seen = 0;
770 	zfs_ace_hdr_t 	*acep = NULL;
771 	uint64_t	who;
772 	uint16_t	iflags, type;
773 	uint32_t	access_mask;
774 
775 	mode = (zp->z_phys->zp_mode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
776 
777 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
778 	    &access_mask, &iflags, &type)) {
779 
780 		/*
781 		 * Skip over inherit only ACEs
782 		 */
783 		if (iflags & ACE_INHERIT_ONLY_ACE)
784 			continue;
785 
786 		entry_type = (iflags & ACE_TYPE_FLAGS);
787 
788 		if (entry_type == ACE_OWNER) {
789 			if ((access_mask & ACE_READ_DATA) &&
790 			    (!(seen & S_IRUSR))) {
791 				seen |= S_IRUSR;
792 				if (type == ALLOW) {
793 					mode |= S_IRUSR;
794 				}
795 			}
796 			if ((access_mask & ACE_WRITE_DATA) &&
797 			    (!(seen & S_IWUSR))) {
798 				seen |= S_IWUSR;
799 				if (type == ALLOW) {
800 					mode |= S_IWUSR;
801 				}
802 			}
803 			if ((access_mask & ACE_EXECUTE) &&
804 			    (!(seen & S_IXUSR))) {
805 				seen |= S_IXUSR;
806 				if (type == ALLOW) {
807 					mode |= S_IXUSR;
808 				}
809 			}
810 		} else if (entry_type == OWNING_GROUP) {
811 			if ((access_mask & ACE_READ_DATA) &&
812 			    (!(seen & S_IRGRP))) {
813 				seen |= S_IRGRP;
814 				if (type == ALLOW) {
815 					mode |= S_IRGRP;
816 				}
817 			}
818 			if ((access_mask & ACE_WRITE_DATA) &&
819 			    (!(seen & S_IWGRP))) {
820 				seen |= S_IWGRP;
821 				if (type == ALLOW) {
822 					mode |= S_IWGRP;
823 				}
824 			}
825 			if ((access_mask & ACE_EXECUTE) &&
826 			    (!(seen & S_IXGRP))) {
827 				seen |= S_IXGRP;
828 				if (type == ALLOW) {
829 					mode |= S_IXGRP;
830 				}
831 			}
832 		} else if (entry_type == ACE_EVERYONE) {
833 			if ((access_mask & ACE_READ_DATA)) {
834 				if (!(seen & S_IRUSR)) {
835 					seen |= S_IRUSR;
836 					if (type == ALLOW) {
837 						mode |= S_IRUSR;
838 					}
839 				}
840 				if (!(seen & S_IRGRP)) {
841 					seen |= S_IRGRP;
842 					if (type == ALLOW) {
843 						mode |= S_IRGRP;
844 					}
845 				}
846 				if (!(seen & S_IROTH)) {
847 					seen |= S_IROTH;
848 					if (type == ALLOW) {
849 						mode |= S_IROTH;
850 					}
851 				}
852 			}
853 			if ((access_mask & ACE_WRITE_DATA)) {
854 				if (!(seen & S_IWUSR)) {
855 					seen |= S_IWUSR;
856 					if (type == ALLOW) {
857 						mode |= S_IWUSR;
858 					}
859 				}
860 				if (!(seen & S_IWGRP)) {
861 					seen |= S_IWGRP;
862 					if (type == ALLOW) {
863 						mode |= S_IWGRP;
864 					}
865 				}
866 				if (!(seen & S_IWOTH)) {
867 					seen |= S_IWOTH;
868 					if (type == ALLOW) {
869 						mode |= S_IWOTH;
870 					}
871 				}
872 			}
873 			if ((access_mask & ACE_EXECUTE)) {
874 				if (!(seen & S_IXUSR)) {
875 					seen |= S_IXUSR;
876 					if (type == ALLOW) {
877 						mode |= S_IXUSR;
878 					}
879 				}
880 				if (!(seen & S_IXGRP)) {
881 					seen |= S_IXGRP;
882 					if (type == ALLOW) {
883 						mode |= S_IXGRP;
884 					}
885 				}
886 				if (!(seen & S_IXOTH)) {
887 					seen |= S_IXOTH;
888 					if (type == ALLOW) {
889 						mode |= S_IXOTH;
890 					}
891 				}
892 			}
893 		}
894 		/*
895 		 * Now handle FUID create for user/group ACEs
896 		 */
897 		if (entry_type == 0 || entry_type == ACE_IDENTIFIER_GROUP) {
898 			aclp->z_ops.ace_who_set(acep,
899 			    zfs_fuid_create(zp->z_zfsvfs, who, cr,
900 			    entry_type == 0 ? ZFS_ACE_USER : ZFS_ACE_GROUP, tx,
901 			    fuidp));
902 		}
903 	}
904 	return (mode);
905 }
906 
907 static zfs_acl_t *
908 zfs_acl_node_read_internal(znode_t *zp, boolean_t will_modify)
909 {
910 	zfs_acl_t	*aclp;
911 	zfs_acl_node_t	*aclnode;
912 
913 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
914 
915 	/*
916 	 * Version 0 to 1 znode_acl_phys has the size/count fields swapped.
917 	 * Version 0 didn't have a size field, only a count.
918 	 */
919 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
920 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_size;
921 		aclp->z_acl_bytes = ZFS_ACL_SIZE(aclp->z_acl_count);
922 	} else {
923 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_count;
924 		aclp->z_acl_bytes = zp->z_phys->zp_acl.z_acl_size;
925 	}
926 
927 	aclnode = zfs_acl_node_alloc(will_modify ? aclp->z_acl_bytes : 0);
928 	aclnode->z_ace_count = aclp->z_acl_count;
929 	if (will_modify) {
930 		bcopy(zp->z_phys->zp_acl.z_ace_data, aclnode->z_acldata,
931 		    aclp->z_acl_bytes);
932 	} else {
933 		aclnode->z_size = aclp->z_acl_bytes;
934 		aclnode->z_acldata = &zp->z_phys->zp_acl.z_ace_data[0];
935 	}
936 
937 	list_insert_head(&aclp->z_acl, aclnode);
938 
939 	return (aclp);
940 }
941 
942 /*
943  * Read an external acl object.
944  */
945 static int
946 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
947 {
948 	uint64_t extacl = zp->z_phys->zp_acl.z_acl_extern_obj;
949 	zfs_acl_t	*aclp;
950 	size_t		aclsize;
951 	size_t		acl_count;
952 	zfs_acl_node_t	*aclnode;
953 	int error;
954 
955 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
956 
957 	if (zp->z_phys->zp_acl.z_acl_extern_obj == 0) {
958 		*aclpp = zfs_acl_node_read_internal(zp, will_modify);
959 		return (0);
960 	}
961 
962 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
963 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
964 		zfs_acl_phys_v0_t *zacl0 =
965 		    (zfs_acl_phys_v0_t *)&zp->z_phys->zp_acl;
966 
967 		aclsize = ZFS_ACL_SIZE(zacl0->z_acl_count);
968 		acl_count = zacl0->z_acl_count;
969 	} else {
970 		aclsize = zp->z_phys->zp_acl.z_acl_size;
971 		acl_count = zp->z_phys->zp_acl.z_acl_count;
972 		if (aclsize == 0)
973 			aclsize = acl_count * sizeof (zfs_ace_t);
974 	}
975 	aclnode = zfs_acl_node_alloc(aclsize);
976 	list_insert_head(&aclp->z_acl, aclnode);
977 	error = dmu_read(zp->z_zfsvfs->z_os, extacl, 0,
978 	    aclsize, aclnode->z_acldata);
979 	aclnode->z_ace_count = acl_count;
980 	aclp->z_acl_count = acl_count;
981 	aclp->z_acl_bytes = aclsize;
982 
983 	if (error != 0) {
984 		zfs_acl_free(aclp);
985 		return (error);
986 	}
987 
988 	*aclpp = aclp;
989 	return (0);
990 }
991 
992 /*
993  * common code for setting ACLs.
994  *
995  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
996  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
997  * already checked the acl and knows whether to inherit.
998  */
999 int
1000 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr,
1001     zfs_fuid_info_t **fuidp, dmu_tx_t *tx)
1002 {
1003 	int		error;
1004 	znode_phys_t	*zphys = zp->z_phys;
1005 	zfs_acl_phys_t	*zacl = &zphys->zp_acl;
1006 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1007 	uint64_t	aoid = zphys->zp_acl.z_acl_extern_obj;
1008 	uint64_t	off = 0;
1009 	dmu_object_type_t otype;
1010 	zfs_acl_node_t	*aclnode;
1011 
1012 	ASSERT(MUTEX_HELD(&zp->z_lock));
1013 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1014 
1015 	dmu_buf_will_dirty(zp->z_dbuf, tx);
1016 
1017 	zphys->zp_mode = zfs_mode_fuid_compute(zp, aclp, cr, fuidp, tx);
1018 
1019 	/*
1020 	 * Decide which opbject type to use.  If we are forced to
1021 	 * use old ACL format than transform ACL into zfs_oldace_t
1022 	 * layout.
1023 	 */
1024 	if (!zfsvfs->z_use_fuids) {
1025 		otype = DMU_OT_OLDACL;
1026 	} else {
1027 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1028 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1029 			zfs_acl_xform(zp, aclp);
1030 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1031 		otype = DMU_OT_ACL;
1032 	}
1033 
1034 	if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1035 		/*
1036 		 * If ACL was previously external and we are now
1037 		 * converting to new ACL format then release old
1038 		 * ACL object and create a new one.
1039 		 */
1040 		if (aoid && aclp->z_version != zacl->z_acl_version) {
1041 			error = dmu_object_free(zfsvfs->z_os,
1042 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1043 			if (error)
1044 				return (error);
1045 			aoid = 0;
1046 		}
1047 		if (aoid == 0) {
1048 			aoid = dmu_object_alloc(zfsvfs->z_os,
1049 			    otype, aclp->z_acl_bytes,
1050 			    otype == DMU_OT_ACL ? DMU_OT_SYSACL : DMU_OT_NONE,
1051 			    otype == DMU_OT_ACL ? DN_MAX_BONUSLEN : 0, tx);
1052 		} else {
1053 			(void) dmu_object_set_blocksize(zfsvfs->z_os, aoid,
1054 			    aclp->z_acl_bytes, 0, tx);
1055 		}
1056 		zphys->zp_acl.z_acl_extern_obj = aoid;
1057 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1058 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1059 			if (aclnode->z_ace_count == 0)
1060 				continue;
1061 			dmu_write(zfsvfs->z_os, aoid, off,
1062 			    aclnode->z_size, aclnode->z_acldata, tx);
1063 			off += aclnode->z_size;
1064 		}
1065 	} else {
1066 		void *start = zacl->z_ace_data;
1067 		/*
1068 		 * Migrating back embedded?
1069 		 */
1070 		if (zphys->zp_acl.z_acl_extern_obj) {
1071 			error = dmu_object_free(zfsvfs->z_os,
1072 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1073 			if (error)
1074 				return (error);
1075 			zphys->zp_acl.z_acl_extern_obj = 0;
1076 		}
1077 
1078 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1079 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1080 			if (aclnode->z_ace_count == 0)
1081 				continue;
1082 			bcopy(aclnode->z_acldata, start, aclnode->z_size);
1083 			start = (caddr_t)start + aclnode->z_size;
1084 		}
1085 	}
1086 
1087 	/*
1088 	 * If Old version then swap count/bytes to match old
1089 	 * layout of znode_acl_phys_t.
1090 	 */
1091 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1092 		zphys->zp_acl.z_acl_size = aclp->z_acl_count;
1093 		zphys->zp_acl.z_acl_count = aclp->z_acl_bytes;
1094 	} else {
1095 		zphys->zp_acl.z_acl_size = aclp->z_acl_bytes;
1096 		zphys->zp_acl.z_acl_count = aclp->z_acl_count;
1097 	}
1098 
1099 	zphys->zp_acl.z_acl_version = aclp->z_version;
1100 
1101 	/*
1102 	 * Replace ACL wide bits, but first clear them.
1103 	 */
1104 	zp->z_phys->zp_flags &= ~ZFS_ACL_WIDE_FLAGS;
1105 
1106 	zp->z_phys->zp_flags |= aclp->z_hints;
1107 
1108 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1109 		zp->z_phys->zp_flags |= ZFS_ACL_TRIVIAL;
1110 
1111 	zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Update access mask for prepended ACE
1117  *
1118  * This applies the "groupmask" value for aclmode property.
1119  */
1120 static void
1121 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1122     mode_t mode, uint64_t owner)
1123 {
1124 	int	rmask, wmask, xmask;
1125 	int	user_ace;
1126 	uint16_t aceflags;
1127 	uint32_t origmask, acepmask;
1128 	uint64_t fuid;
1129 
1130 	aceflags = aclp->z_ops.ace_flags_get(acep);
1131 	fuid = aclp->z_ops.ace_who_get(acep);
1132 	origmask = aclp->z_ops.ace_mask_get(origacep);
1133 	acepmask = aclp->z_ops.ace_mask_get(acep);
1134 
1135 	user_ace = (!(aceflags &
1136 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1137 
1138 	if (user_ace && (fuid == owner)) {
1139 		rmask = S_IRUSR;
1140 		wmask = S_IWUSR;
1141 		xmask = S_IXUSR;
1142 	} else {
1143 		rmask = S_IRGRP;
1144 		wmask = S_IWGRP;
1145 		xmask = S_IXGRP;
1146 	}
1147 
1148 	if (origmask & ACE_READ_DATA) {
1149 		if (mode & rmask) {
1150 			acepmask &= ~ACE_READ_DATA;
1151 		} else {
1152 			acepmask |= ACE_READ_DATA;
1153 		}
1154 	}
1155 
1156 	if (origmask & ACE_WRITE_DATA) {
1157 		if (mode & wmask) {
1158 			acepmask &= ~ACE_WRITE_DATA;
1159 		} else {
1160 			acepmask |= ACE_WRITE_DATA;
1161 		}
1162 	}
1163 
1164 	if (origmask & ACE_APPEND_DATA) {
1165 		if (mode & wmask) {
1166 			acepmask &= ~ACE_APPEND_DATA;
1167 		} else {
1168 			acepmask |= ACE_APPEND_DATA;
1169 		}
1170 	}
1171 
1172 	if (origmask & ACE_EXECUTE) {
1173 		if (mode & xmask) {
1174 			acepmask &= ~ACE_EXECUTE;
1175 		} else {
1176 			acepmask |= ACE_EXECUTE;
1177 		}
1178 	}
1179 	aclp->z_ops.ace_mask_set(acep, acepmask);
1180 }
1181 
1182 /*
1183  * Apply mode to canonical six ACEs.
1184  */
1185 static void
1186 zfs_acl_fixup_canonical_six(zfs_acl_t *aclp, mode_t mode)
1187 {
1188 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1189 	void	*acep;
1190 	int	maskoff = aclp->z_ops.ace_mask_off();
1191 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1192 
1193 	ASSERT(aclnode != NULL);
1194 
1195 	acep = (void *)((caddr_t)aclnode->z_acldata +
1196 	    aclnode->z_size - (abstract_size * 6));
1197 
1198 	/*
1199 	 * Fixup final ACEs to match the mode
1200 	 */
1201 
1202 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1203 	    (mode & 0700) >> 6);	/* owner@ */
1204 
1205 	acep = (caddr_t)acep + (abstract_size * 2);
1206 
1207 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1208 	    (mode & 0070) >> 3);	/* group@ */
1209 
1210 	acep = (caddr_t)acep + (abstract_size * 2);
1211 	adjust_ace_pair_common(acep, maskoff,
1212 	    abstract_size, mode);	/* everyone@ */
1213 }
1214 
1215 
1216 static int
1217 zfs_acl_ace_match(zfs_acl_t *aclp, void *acep, int allow_deny,
1218     int entry_type, int accessmask)
1219 {
1220 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1221 	uint16_t type = aclp->z_ops.ace_type_get(acep);
1222 	uint16_t flags = aclp->z_ops.ace_flags_get(acep);
1223 
1224 	return (mask == accessmask && type == allow_deny &&
1225 	    ((flags & ACE_TYPE_FLAGS) == entry_type));
1226 }
1227 
1228 /*
1229  * Can prepended ACE be reused?
1230  */
1231 static int
1232 zfs_reuse_deny(zfs_acl_t *aclp, void *acep, void *prevacep)
1233 {
1234 	int okay_masks;
1235 	uint16_t prevtype;
1236 	uint16_t prevflags;
1237 	uint16_t flags;
1238 	uint32_t mask, prevmask;
1239 
1240 	if (prevacep == NULL)
1241 		return (B_FALSE);
1242 
1243 	prevtype = aclp->z_ops.ace_type_get(prevacep);
1244 	prevflags = aclp->z_ops.ace_flags_get(prevacep);
1245 	flags = aclp->z_ops.ace_flags_get(acep);
1246 	mask = aclp->z_ops.ace_mask_get(acep);
1247 	prevmask = aclp->z_ops.ace_mask_get(prevacep);
1248 
1249 	if (prevtype != DENY)
1250 		return (B_FALSE);
1251 
1252 	if (prevflags != (flags & ACE_IDENTIFIER_GROUP))
1253 		return (B_FALSE);
1254 
1255 	okay_masks = (mask & OKAY_MASK_BITS);
1256 
1257 	if (prevmask & ~okay_masks)
1258 		return (B_FALSE);
1259 
1260 	return (B_TRUE);
1261 }
1262 
1263 
1264 /*
1265  * Insert new ACL node into chain of zfs_acl_node_t's
1266  *
1267  * This will result in two possible results.
1268  * 1. If the ACL is currently just a single zfs_acl_node and
1269  *    we are prepending the entry then current acl node will have
1270  *    a new node inserted above it.
1271  *
1272  * 2. If we are inserting in the middle of current acl node then
1273  *    the current node will be split in two and new node will be inserted
1274  *    in between the two split nodes.
1275  */
1276 static zfs_acl_node_t *
1277 zfs_acl_ace_insert(zfs_acl_t *aclp, void  *acep)
1278 {
1279 	zfs_acl_node_t 	*newnode;
1280 	zfs_acl_node_t 	*trailernode = NULL;
1281 	zfs_acl_node_t 	*currnode = zfs_acl_curr_node(aclp);
1282 	int		curr_idx = aclp->z_curr_node->z_ace_idx;
1283 	int		trailer_count;
1284 	size_t		oldsize;
1285 
1286 	newnode = zfs_acl_node_alloc(aclp->z_ops.ace_size(acep));
1287 	newnode->z_ace_count = 1;
1288 
1289 	oldsize = currnode->z_size;
1290 
1291 	if (curr_idx != 1) {
1292 		trailernode = zfs_acl_node_alloc(0);
1293 		trailernode->z_acldata = acep;
1294 
1295 		trailer_count = currnode->z_ace_count - curr_idx + 1;
1296 		currnode->z_ace_count = curr_idx - 1;
1297 		currnode->z_size = (caddr_t)acep - (caddr_t)currnode->z_acldata;
1298 		trailernode->z_size = oldsize - currnode->z_size;
1299 		trailernode->z_ace_count = trailer_count;
1300 	}
1301 
1302 	aclp->z_acl_count += 1;
1303 	aclp->z_acl_bytes += aclp->z_ops.ace_size(acep);
1304 
1305 	if (curr_idx == 1)
1306 		list_insert_before(&aclp->z_acl, currnode, newnode);
1307 	else
1308 		list_insert_after(&aclp->z_acl, currnode, newnode);
1309 	if (trailernode) {
1310 		list_insert_after(&aclp->z_acl, newnode, trailernode);
1311 		aclp->z_curr_node = trailernode;
1312 		trailernode->z_ace_idx = 1;
1313 	}
1314 
1315 	return (newnode);
1316 }
1317 
1318 /*
1319  * Prepend deny ACE
1320  */
1321 static void *
1322 zfs_acl_prepend_deny(znode_t *zp, zfs_acl_t *aclp, void *acep,
1323     mode_t mode)
1324 {
1325 	zfs_acl_node_t *aclnode;
1326 	void  *newacep;
1327 	uint64_t fuid;
1328 	uint16_t flags;
1329 
1330 	aclnode = zfs_acl_ace_insert(aclp, acep);
1331 	newacep = aclnode->z_acldata;
1332 	fuid = aclp->z_ops.ace_who_get(acep);
1333 	flags = aclp->z_ops.ace_flags_get(acep);
1334 	zfs_set_ace(aclp, newacep, 0, DENY, fuid, (flags & ACE_TYPE_FLAGS));
1335 	zfs_acl_prepend_fixup(aclp, newacep, acep, mode, zp->z_phys->zp_uid);
1336 
1337 	return (newacep);
1338 }
1339 
1340 /*
1341  * Split an inherited ACE into inherit_only ACE
1342  * and original ACE with inheritance flags stripped off.
1343  */
1344 static void
1345 zfs_acl_split_ace(zfs_acl_t *aclp, zfs_ace_hdr_t *acep)
1346 {
1347 	zfs_acl_node_t *aclnode;
1348 	zfs_acl_node_t *currnode;
1349 	void  *newacep;
1350 	uint16_t type, flags;
1351 	uint32_t mask;
1352 	uint64_t fuid;
1353 
1354 	type = aclp->z_ops.ace_type_get(acep);
1355 	flags = aclp->z_ops.ace_flags_get(acep);
1356 	mask = aclp->z_ops.ace_mask_get(acep);
1357 	fuid = aclp->z_ops.ace_who_get(acep);
1358 
1359 	aclnode = zfs_acl_ace_insert(aclp, acep);
1360 	newacep = aclnode->z_acldata;
1361 
1362 	aclp->z_ops.ace_type_set(newacep, type);
1363 	aclp->z_ops.ace_flags_set(newacep, flags | ACE_INHERIT_ONLY_ACE);
1364 	aclp->z_ops.ace_mask_set(newacep, mask);
1365 	aclp->z_ops.ace_type_set(newacep, type);
1366 	aclp->z_ops.ace_who_set(newacep, fuid);
1367 	aclp->z_next_ace = acep;
1368 	flags &= ~ALL_INHERIT;
1369 	aclp->z_ops.ace_flags_set(acep, flags);
1370 	currnode = zfs_acl_curr_node(aclp);
1371 	ASSERT(currnode->z_ace_idx >= 1);
1372 	currnode->z_ace_idx -= 1;
1373 }
1374 
1375 /*
1376  * Are ACES started at index i, the canonical six ACES?
1377  */
1378 static int
1379 zfs_have_canonical_six(zfs_acl_t *aclp)
1380 {
1381 	void *acep;
1382 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1383 	int		i = 0;
1384 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1385 
1386 	ASSERT(aclnode != NULL);
1387 
1388 	if (aclnode->z_ace_count < 6)
1389 		return (0);
1390 
1391 	acep = (void *)((caddr_t)aclnode->z_acldata +
1392 	    aclnode->z_size - (aclp->z_ops.ace_abstract_size() * 6));
1393 
1394 	if ((zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1395 	    DENY, ACE_OWNER, 0) &&
1396 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1397 	    ALLOW, ACE_OWNER, OWNER_ALLOW_MASK) &&
1398 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++), DENY,
1399 	    OWNING_GROUP, 0) && zfs_acl_ace_match(aclp, (caddr_t)acep +
1400 	    (abstract_size * i++),
1401 	    ALLOW, OWNING_GROUP, 0) &&
1402 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1403 	    DENY, ACE_EVERYONE, EVERYONE_DENY_MASK) &&
1404 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1405 	    ALLOW, ACE_EVERYONE, EVERYONE_ALLOW_MASK))) {
1406 		return (1);
1407 	} else {
1408 		return (0);
1409 	}
1410 }
1411 
1412 
1413 /*
1414  * Apply step 1g, to group entries
1415  *
1416  * Need to deal with corner case where group may have
1417  * greater permissions than owner.  If so then limit
1418  * group permissions, based on what extra permissions
1419  * group has.
1420  */
1421 static void
1422 zfs_fixup_group_entries(zfs_acl_t *aclp, void *acep, void *prevacep,
1423     mode_t mode)
1424 {
1425 	uint32_t prevmask = aclp->z_ops.ace_mask_get(prevacep);
1426 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1427 	uint16_t prevflags = aclp->z_ops.ace_flags_get(prevacep);
1428 	mode_t extramode = (mode >> 3) & 07;
1429 	mode_t ownermode = (mode >> 6);
1430 
1431 	if (prevflags & ACE_IDENTIFIER_GROUP) {
1432 
1433 		extramode &= ~ownermode;
1434 
1435 		if (extramode) {
1436 			if (extramode & S_IROTH) {
1437 				prevmask &= ~ACE_READ_DATA;
1438 				mask &= ~ACE_READ_DATA;
1439 			}
1440 			if (extramode & S_IWOTH) {
1441 				prevmask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1442 				mask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1443 			}
1444 			if (extramode & S_IXOTH) {
1445 				prevmask  &= ~ACE_EXECUTE;
1446 				mask &= ~ACE_EXECUTE;
1447 			}
1448 		}
1449 	}
1450 	aclp->z_ops.ace_mask_set(acep, mask);
1451 	aclp->z_ops.ace_mask_set(prevacep, prevmask);
1452 }
1453 
1454 /*
1455  * Apply the chmod algorithm as described
1456  * in PSARC/2002/240
1457  */
1458 static void
1459 zfs_acl_chmod(znode_t *zp, uint64_t mode, zfs_acl_t *aclp)
1460 {
1461 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1462 	void		*acep = NULL, *prevacep = NULL;
1463 	uint64_t	who;
1464 	int 		i;
1465 	int 		entry_type;
1466 	int 		reuse_deny;
1467 	int 		need_canonical_six = 1;
1468 	uint16_t	iflags, type;
1469 	uint32_t	access_mask;
1470 
1471 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1472 	ASSERT(MUTEX_HELD(&zp->z_lock));
1473 
1474 	aclp->z_hints = (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
1475 
1476 	/*
1477 	 * If discard then just discard all ACL nodes which
1478 	 * represent the ACEs.
1479 	 *
1480 	 * New owner@/group@/everone@ ACEs will be added
1481 	 * later.
1482 	 */
1483 	if (zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1484 		zfs_acl_release_nodes(aclp);
1485 
1486 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1487 	    &iflags, &type)) {
1488 
1489 		entry_type = (iflags & ACE_TYPE_FLAGS);
1490 		iflags = (iflags & ALL_INHERIT);
1491 
1492 		if ((type != ALLOW && type != DENY) ||
1493 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1494 			if (iflags)
1495 				aclp->z_hints |= ZFS_INHERIT_ACE;
1496 			switch (type) {
1497 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1498 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1499 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1500 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1501 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1502 				break;
1503 			}
1504 			goto nextace;
1505 		}
1506 
1507 		/*
1508 		 * Need to split ace into two?
1509 		 */
1510 		if ((iflags & (ACE_FILE_INHERIT_ACE|
1511 		    ACE_DIRECTORY_INHERIT_ACE)) &&
1512 		    (!(iflags & ACE_INHERIT_ONLY_ACE))) {
1513 			zfs_acl_split_ace(aclp, acep);
1514 			aclp->z_hints |= ZFS_INHERIT_ACE;
1515 			goto nextace;
1516 		}
1517 
1518 		if (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1519 		    (entry_type == OWNING_GROUP)) {
1520 			access_mask &= ~OGE_CLEAR;
1521 			aclp->z_ops.ace_mask_set(acep, access_mask);
1522 			goto nextace;
1523 		} else {
1524 			reuse_deny = B_TRUE;
1525 			if (type == ALLOW) {
1526 
1527 				/*
1528 				 * Check preceding ACE if any, to see
1529 				 * if we need to prepend a DENY ACE.
1530 				 * This is only applicable when the acl_mode
1531 				 * property == groupmask.
1532 				 */
1533 				if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK) {
1534 
1535 					reuse_deny = zfs_reuse_deny(aclp, acep,
1536 					    prevacep);
1537 
1538 					if (reuse_deny == B_FALSE) {
1539 						prevacep =
1540 						    zfs_acl_prepend_deny(zp,
1541 						    aclp, acep, mode);
1542 					} else {
1543 						zfs_acl_prepend_fixup(
1544 						    aclp, prevacep,
1545 						    acep, mode,
1546 						    zp->z_phys->zp_uid);
1547 					}
1548 					zfs_fixup_group_entries(aclp, acep,
1549 					    prevacep, mode);
1550 
1551 				}
1552 			}
1553 		}
1554 nextace:
1555 		prevacep = acep;
1556 	}
1557 
1558 	/*
1559 	 * Check out last six aces, if we have six.
1560 	 */
1561 
1562 	if (aclp->z_acl_count >= 6) {
1563 		if (zfs_have_canonical_six(aclp)) {
1564 			need_canonical_six = 0;
1565 		}
1566 	}
1567 
1568 	if (need_canonical_six) {
1569 		size_t abstract_size = aclp->z_ops.ace_abstract_size();
1570 		void *zacep;
1571 		zfs_acl_node_t *aclnode =
1572 		    zfs_acl_node_alloc(abstract_size * 6);
1573 
1574 		aclnode->z_size = abstract_size * 6;
1575 		aclnode->z_ace_count = 6;
1576 		aclp->z_acl_bytes += aclnode->z_size;
1577 		list_insert_tail(&aclp->z_acl, aclnode);
1578 
1579 		zacep = aclnode->z_acldata;
1580 
1581 		i = 0;
1582 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1583 		    0, DENY, -1, ACE_OWNER);
1584 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1585 		    OWNER_ALLOW_MASK, ALLOW, -1, ACE_OWNER);
1586 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1587 		    DENY, -1, OWNING_GROUP);
1588 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1589 		    ALLOW, -1, OWNING_GROUP);
1590 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1591 		    EVERYONE_DENY_MASK, DENY, -1, ACE_EVERYONE);
1592 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1593 		    EVERYONE_ALLOW_MASK, ALLOW, -1, ACE_EVERYONE);
1594 		aclp->z_acl_count += 6;
1595 	}
1596 
1597 	zfs_acl_fixup_canonical_six(aclp, mode);
1598 }
1599 
1600 int
1601 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1602 {
1603 	int error;
1604 
1605 	mutex_enter(&zp->z_lock);
1606 	mutex_enter(&zp->z_acl_lock);
1607 	*aclp = NULL;
1608 	error = zfs_acl_node_read(zp, aclp, B_TRUE);
1609 	if (error == 0)
1610 		zfs_acl_chmod(zp, mode, *aclp);
1611 	mutex_exit(&zp->z_acl_lock);
1612 	mutex_exit(&zp->z_lock);
1613 	return (error);
1614 }
1615 
1616 /*
1617  * strip off write_owner and write_acl
1618  */
1619 static void
1620 zfs_securemode_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1621 {
1622 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1623 
1624 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_SECURE) &&
1625 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1626 		mask &= ~SECURE_CLEAR;
1627 		aclp->z_ops.ace_mask_set(acep, mask);
1628 	}
1629 }
1630 
1631 /*
1632  * Should ACE be inherited?
1633  */
1634 static int
1635 zfs_ace_can_use(znode_t *zp, uint16_t acep_flags)
1636 {
1637 	int vtype = ZTOV(zp)->v_type;
1638 	int	iflags = (acep_flags & 0xf);
1639 
1640 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1641 		return (1);
1642 	else if (iflags & ACE_FILE_INHERIT_ACE)
1643 		return (!((vtype == VDIR) &&
1644 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1645 	return (0);
1646 }
1647 
1648 /*
1649  * inherit inheritable ACEs from parent
1650  */
1651 static zfs_acl_t *
1652 zfs_acl_inherit(znode_t *zp, zfs_acl_t *paclp)
1653 {
1654 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1655 	void		*pacep;
1656 	void		*acep, *acep2;
1657 	zfs_acl_node_t  *aclnode, *aclnode2;
1658 	zfs_acl_t	*aclp = NULL;
1659 	uint64_t	who;
1660 	uint32_t	access_mask;
1661 	uint16_t	iflags, newflags, type;
1662 	size_t		ace_size;
1663 	void		*data1, *data2;
1664 	size_t		data1sz, data2sz;
1665 
1666 	pacep = NULL;
1667 	aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1668 	if (zfsvfs->z_acl_inherit != ZFS_ACL_DISCARD) {
1669 		while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1670 		    &access_mask, &iflags, &type)) {
1671 
1672 			if (zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW &&
1673 			    type == ALLOW)
1674 				continue;
1675 
1676 			ace_size = aclp->z_ops.ace_size(pacep);
1677 
1678 			if (zfs_ace_can_use(zp, iflags)) {
1679 				aclnode =
1680 				    zfs_acl_node_alloc(ace_size);
1681 
1682 				list_insert_tail(&aclp->z_acl, aclnode);
1683 				acep = aclnode->z_acldata;
1684 				zfs_set_ace(aclp, acep, access_mask, type,
1685 				    who, iflags|ACE_INHERITED_ACE);
1686 
1687 				/*
1688 				 * Copy special opaque data if any
1689 				 */
1690 				if ((data1sz = paclp->z_ops.ace_data(pacep,
1691 				    &data1)) != 0) {
1692 					VERIFY((data2sz =
1693 					    aclp->z_ops.ace_data(acep,
1694 					    &data2)) == data1sz);
1695 					bcopy(data1, data2, data2sz);
1696 				}
1697 				aclp->z_acl_count++;
1698 				aclnode->z_ace_count++;
1699 				aclp->z_acl_bytes += aclnode->z_size;
1700 				newflags = aclp->z_ops.ace_flags_get(acep);
1701 				if ((iflags &
1702 				    ACE_NO_PROPAGATE_INHERIT_ACE) ||
1703 				    (ZTOV(zp)->v_type != VDIR)) {
1704 					newflags &= ~ALL_INHERIT;
1705 					aclp->z_ops.ace_flags_set(acep,
1706 					    newflags|ACE_INHERITED_ACE);
1707 					zfs_securemode_update(zfsvfs,
1708 					    aclp, acep);
1709 					continue;
1710 				}
1711 
1712 				ASSERT(ZTOV(zp)->v_type == VDIR);
1713 
1714 				newflags = aclp->z_ops.ace_flags_get(acep);
1715 				if ((iflags & (ACE_FILE_INHERIT_ACE |
1716 				    ACE_DIRECTORY_INHERIT_ACE)) !=
1717 				    ACE_FILE_INHERIT_ACE) {
1718 					aclnode2 = zfs_acl_node_alloc(ace_size);
1719 					list_insert_tail(&aclp->z_acl,
1720 					    aclnode2);
1721 					acep2 = aclnode2->z_acldata;
1722 					zfs_set_ace(aclp, acep2,
1723 					    access_mask, type, who,
1724 					    iflags|ACE_INHERITED_ACE);
1725 					newflags |= ACE_INHERIT_ONLY_ACE;
1726 					aclp->z_ops.ace_flags_set(acep,
1727 					    newflags);
1728 					newflags &= ~ALL_INHERIT;
1729 					aclp->z_ops.ace_flags_set(acep2,
1730 					    newflags|ACE_INHERITED_ACE);
1731 
1732 					/*
1733 					 * Copy special opaque data if any
1734 					 */
1735 					if ((data1sz =
1736 					    aclp->z_ops.ace_data(acep,
1737 					    &data1)) != 0) {
1738 						VERIFY((data2sz =
1739 						    aclp->z_ops.ace_data(acep2,
1740 						    &data2)) == data1sz);
1741 						bcopy(data1, data2, data1sz);
1742 					}
1743 					aclp->z_acl_count++;
1744 					aclnode2->z_ace_count++;
1745 					aclp->z_acl_bytes += aclnode->z_size;
1746 					zfs_securemode_update(zfsvfs,
1747 					    aclp, acep2);
1748 				} else {
1749 					newflags |= ACE_INHERIT_ONLY_ACE;
1750 					aclp->z_ops.ace_flags_set(acep,
1751 					    newflags|ACE_INHERITED_ACE);
1752 				}
1753 
1754 			}
1755 		}
1756 	}
1757 	return (aclp);
1758 }
1759 
1760 /*
1761  * Create file system object initial permissions
1762  * including inheritable ACEs.
1763  */
1764 void
1765 zfs_perm_init(znode_t *zp, znode_t *parent, int flag,
1766     vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
1767     zfs_acl_t *setaclp, zfs_fuid_info_t **fuidp)
1768 {
1769 	uint64_t	mode;
1770 	uint64_t	uid;
1771 	uint64_t	gid;
1772 	int		error;
1773 	int		pull_down;
1774 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1775 	zfs_acl_t	*aclp = NULL;
1776 	zfs_acl_t	*paclp;
1777 	xvattr_t	*xvap = (xvattr_t *)vap;
1778 
1779 	if (setaclp)
1780 		aclp = setaclp;
1781 
1782 	mode = MAKEIMODE(vap->va_type, vap->va_mode);
1783 
1784 	/*
1785 	 * Determine uid and gid.
1786 	 */
1787 	if ((flag & (IS_ROOT_NODE | IS_REPLAY)) ||
1788 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1789 		uid = zfs_fuid_create(zfsvfs, vap->va_uid, cr,
1790 		    ZFS_OWNER, tx, fuidp);
1791 		gid = zfs_fuid_create(zfsvfs, vap->va_gid, cr,
1792 		    ZFS_GROUP, tx, fuidp);
1793 	} else {
1794 		uid = zfs_fuid_create_cred(zfsvfs, crgetuid(cr),
1795 		    ZFS_OWNER, tx, cr, fuidp);
1796 		if ((vap->va_mask & AT_GID) &&
1797 		    ((vap->va_gid == parent->z_phys->zp_gid) ||
1798 		    groupmember(vap->va_gid, cr) ||
1799 		    secpolicy_vnode_create_gid(cr) == 0)) {
1800 			gid = zfs_fuid_create_cred(zfsvfs, vap->va_gid,
1801 			    ZFS_GROUP, tx, cr, fuidp);
1802 		} else {
1803 			gid = (parent->z_phys->zp_mode & S_ISGID) ?
1804 			    parent->z_phys->zp_gid : crgetgid(cr);
1805 			gid = zfs_fuid_create_cred(zfsvfs, gid,
1806 			    ZFS_GROUP, tx, cr, fuidp);
1807 		}
1808 	}
1809 
1810 	/*
1811 	 * If we're creating a directory, and the parent directory has the
1812 	 * set-GID bit set, set in on the new directory.
1813 	 * Otherwise, if the user is neither privileged nor a member of the
1814 	 * file's new group, clear the file's set-GID bit.
1815 	 */
1816 
1817 	if ((parent->z_phys->zp_mode & S_ISGID) && (vap->va_type == VDIR))
1818 		mode |= S_ISGID;
1819 	else {
1820 		if ((mode & S_ISGID) &&
1821 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1822 			mode &= ~S_ISGID;
1823 	}
1824 
1825 	zp->z_phys->zp_uid = uid;
1826 	zp->z_phys->zp_gid = gid;
1827 	zp->z_phys->zp_mode = mode;
1828 
1829 	if (aclp == NULL) {
1830 		mutex_enter(&parent->z_lock);
1831 		pull_down = (parent->z_phys->zp_flags & ZFS_INHERIT_ACE);
1832 		if (pull_down) {
1833 			mutex_enter(&parent->z_acl_lock);
1834 			VERIFY(0 == zfs_acl_node_read(parent, &paclp, B_FALSE));
1835 			mutex_exit(&parent->z_acl_lock);
1836 			aclp = zfs_acl_inherit(zp, paclp);
1837 			zfs_acl_free(paclp);
1838 		} else {
1839 			aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1840 		}
1841 		mutex_exit(&parent->z_lock);
1842 		mutex_enter(&zp->z_lock);
1843 		mutex_enter(&zp->z_acl_lock);
1844 		zfs_acl_chmod(zp, mode, aclp);
1845 	} else {
1846 		mutex_enter(&zp->z_lock);
1847 		mutex_enter(&zp->z_acl_lock);
1848 	}
1849 
1850 	/* Force auto_inherit on all new directory objects */
1851 	if (vap->va_type == VDIR)
1852 		aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1853 
1854 	error = zfs_aclset_common(zp, aclp, cr, fuidp, tx);
1855 
1856 	/* Set optional attributes if any */
1857 	if (vap->va_mask & AT_XVATTR)
1858 		zfs_xvattr_set(zp, xvap);
1859 
1860 	mutex_exit(&zp->z_lock);
1861 	mutex_exit(&zp->z_acl_lock);
1862 	ASSERT3U(error, ==, 0);
1863 
1864 	if (aclp != setaclp) {
1865 		zfs_acl_free(aclp);
1866 	}
1867 }
1868 
1869 /*
1870  * Retrieve a files ACL
1871  */
1872 int
1873 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1874 {
1875 	zfs_acl_t	*aclp;
1876 	ulong_t		mask;
1877 	int		error;
1878 	int 		count = 0;
1879 	int		largeace = 0;
1880 
1881 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1882 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1883 
1884 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1885 		return (error);
1886 
1887 	if (mask == 0)
1888 		return (ENOSYS);
1889 
1890 	mutex_enter(&zp->z_acl_lock);
1891 
1892 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
1893 	if (error != 0) {
1894 		mutex_exit(&zp->z_acl_lock);
1895 		return (error);
1896 	}
1897 
1898 	/*
1899 	 * Scan ACL to determine number of ACEs
1900 	 */
1901 	if ((zp->z_phys->zp_flags & ZFS_ACL_OBJ_ACE) &&
1902 	    !(mask & VSA_ACE_ALLTYPES)) {
1903 		void *zacep = NULL;
1904 		uint64_t who;
1905 		uint32_t access_mask;
1906 		uint16_t type, iflags;
1907 
1908 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1909 		    &who, &access_mask, &iflags, &type)) {
1910 			switch (type) {
1911 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1912 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1913 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1914 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1915 				largeace++;
1916 				continue;
1917 			default:
1918 				count++;
1919 			}
1920 		}
1921 		vsecp->vsa_aclcnt = count;
1922 	} else
1923 		count = aclp->z_acl_count;
1924 
1925 	if (mask & VSA_ACECNT) {
1926 		vsecp->vsa_aclcnt = count;
1927 	}
1928 
1929 	if (mask & VSA_ACE) {
1930 		size_t aclsz;
1931 
1932 		zfs_acl_node_t *aclnode = list_head(&aclp->z_acl);
1933 
1934 		aclsz = count * sizeof (ace_t) +
1935 		    sizeof (ace_object_t) * largeace;
1936 
1937 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1938 		vsecp->vsa_aclentsz = aclsz;
1939 
1940 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1941 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1942 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1943 		else {
1944 			bcopy(aclnode->z_acldata, vsecp->vsa_aclentp,
1945 			    count * sizeof (ace_t));
1946 		}
1947 	}
1948 	if (mask & VSA_ACE_ACLFLAGS) {
1949 		vsecp->vsa_aclflags = 0;
1950 		if (zp->z_phys->zp_flags & ZFS_ACL_DEFAULTED)
1951 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1952 		if (zp->z_phys->zp_flags & ZFS_ACL_PROTECTED)
1953 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1954 		if (zp->z_phys->zp_flags & ZFS_ACL_AUTO_INHERIT)
1955 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1956 	}
1957 
1958 	mutex_exit(&zp->z_acl_lock);
1959 
1960 	zfs_acl_free(aclp);
1961 
1962 	return (0);
1963 }
1964 
1965 int
1966 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
1967     vsecattr_t *vsecp, zfs_acl_t **zaclp)
1968 {
1969 	zfs_acl_t *aclp;
1970 	zfs_acl_node_t *aclnode;
1971 	int aclcnt = vsecp->vsa_aclcnt;
1972 	int error;
1973 
1974 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1975 		return (EINVAL);
1976 
1977 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1978 
1979 	aclp->z_hints = 0;
1980 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1981 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1982 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1983 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1984 		    aclcnt, &aclnode->z_size)) != 0) {
1985 			zfs_acl_free(aclp);
1986 			zfs_acl_node_free(aclnode);
1987 			return (error);
1988 		}
1989 	} else {
1990 		if ((error = zfs_copy_ace_2_fuid(obj_type, aclp,
1991 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1992 		    &aclnode->z_size)) != 0) {
1993 			zfs_acl_free(aclp);
1994 			zfs_acl_node_free(aclnode);
1995 			return (error);
1996 		}
1997 	}
1998 	aclp->z_acl_bytes = aclnode->z_size;
1999 	aclnode->z_ace_count = aclcnt;
2000 	aclp->z_acl_count = aclcnt;
2001 	list_insert_head(&aclp->z_acl, aclnode);
2002 
2003 	/*
2004 	 * If flags are being set then add them to z_hints
2005 	 */
2006 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2007 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2008 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2009 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2010 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2011 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2012 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2013 	}
2014 
2015 	*zaclp = aclp;
2016 
2017 	return (0);
2018 }
2019 
2020 /*
2021  * Set a files ACL
2022  */
2023 int
2024 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2025 {
2026 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2027 	zilog_t		*zilog = zfsvfs->z_log;
2028 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2029 	dmu_tx_t	*tx;
2030 	int		error;
2031 	zfs_acl_t	*aclp;
2032 	zfs_fuid_info_t	*fuidp = NULL;
2033 
2034 	if (mask == 0)
2035 		return (ENOSYS);
2036 
2037 	if (zp->z_phys->zp_flags & ZFS_IMMUTABLE)
2038 		return (EPERM);
2039 
2040 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
2041 		return (error);
2042 
2043 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, &aclp);
2044 	if (error)
2045 		return (error);
2046 
2047 	/*
2048 	 * If ACL wide flags aren't being set then preserve any
2049 	 * existing flags.
2050 	 */
2051 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2052 		aclp->z_hints |= (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
2053 	}
2054 top:
2055 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)) {
2056 		zfs_acl_free(aclp);
2057 		return (error);
2058 	}
2059 
2060 	mutex_enter(&zp->z_lock);
2061 	mutex_enter(&zp->z_acl_lock);
2062 
2063 	tx = dmu_tx_create(zfsvfs->z_os);
2064 	dmu_tx_hold_bonus(tx, zp->z_id);
2065 
2066 	if (zp->z_phys->zp_acl.z_acl_extern_obj) {
2067 		/* Are we upgrading ACL? */
2068 		if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2069 		    zp->z_phys->zp_acl.z_acl_version ==
2070 		    ZFS_ACL_VERSION_INITIAL) {
2071 			dmu_tx_hold_free(tx,
2072 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2073 			    0, DMU_OBJECT_END);
2074 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2075 			    0, aclp->z_acl_bytes);
2076 		} else {
2077 			dmu_tx_hold_write(tx,
2078 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2079 			    0, aclp->z_acl_bytes);
2080 		}
2081 	} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2082 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2083 	}
2084 	if (aclp->z_has_fuids) {
2085 		if (zfsvfs->z_fuid_obj == 0) {
2086 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2087 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2088 			    FUID_SIZE_ESTIMATE(zfsvfs));
2089 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2090 		} else {
2091 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2092 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2093 			    FUID_SIZE_ESTIMATE(zfsvfs));
2094 		}
2095 	}
2096 
2097 	error = dmu_tx_assign(tx, zfsvfs->z_assign);
2098 	if (error) {
2099 		mutex_exit(&zp->z_acl_lock);
2100 		mutex_exit(&zp->z_lock);
2101 
2102 		if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
2103 			dmu_tx_wait(tx);
2104 			dmu_tx_abort(tx);
2105 			goto top;
2106 		}
2107 		dmu_tx_abort(tx);
2108 		zfs_acl_free(aclp);
2109 		return (error);
2110 	}
2111 
2112 	error = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2113 	ASSERT(error == 0);
2114 
2115 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2116 
2117 	if (fuidp)
2118 		zfs_fuid_info_free(fuidp);
2119 	zfs_acl_free(aclp);
2120 	dmu_tx_commit(tx);
2121 done:
2122 	mutex_exit(&zp->z_acl_lock);
2123 	mutex_exit(&zp->z_lock);
2124 
2125 	return (error);
2126 }
2127 
2128 /*
2129  * working_mode returns the permissions that were not granted
2130  */
2131 static int
2132 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2133     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2134 {
2135 	zfs_acl_t	*aclp;
2136 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2137 	int		error;
2138 	int		access_deny = ACCESS_UNDETERMINED;
2139 	uid_t		uid = crgetuid(cr);
2140 	uint64_t 	who;
2141 	uint16_t	type, iflags;
2142 	uint16_t	entry_type;
2143 	uint32_t	access_mask;
2144 	zfs_ace_hdr_t	*acep = NULL;
2145 	boolean_t	checkit;
2146 	uid_t		fowner;
2147 	uid_t		gowner;
2148 
2149 	/*
2150 	 * Short circuit empty requests
2151 	 */
2152 	if (v4_mode == 0)
2153 		return (0);
2154 
2155 	*check_privs = B_TRUE;
2156 
2157 	if (zfsvfs->z_assign >= TXG_INITIAL) {		/* ZIL replay */
2158 		*working_mode = 0;
2159 		return (0);
2160 	}
2161 
2162 	*working_mode = v4_mode;
2163 
2164 	if ((v4_mode & WRITE_MASK) &&
2165 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2166 	    (!IS_DEVVP(ZTOV(zp)))) {
2167 		*check_privs = B_FALSE;
2168 		return (EROFS);
2169 	}
2170 
2171 	/*
2172 	 * Only check for READONLY on non-directories.
2173 	 */
2174 	if ((v4_mode & WRITE_MASK_DATA) &&
2175 	    (((ZTOV(zp)->v_type != VDIR) &&
2176 	    (zp->z_phys->zp_flags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2177 	    (ZTOV(zp)->v_type == VDIR &&
2178 	    (zp->z_phys->zp_flags & ZFS_IMMUTABLE)))) {
2179 		*check_privs = B_FALSE;
2180 		return (EPERM);
2181 	}
2182 
2183 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2184 	    (zp->z_phys->zp_flags & ZFS_NOUNLINK)) {
2185 		*check_privs = B_FALSE;
2186 		return (EPERM);
2187 	}
2188 
2189 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2190 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED))) {
2191 		*check_privs = B_FALSE;
2192 		return (EACCES);
2193 	}
2194 
2195 	/*
2196 	 * The caller requested that the ACL check be skipped.  This
2197 	 * would only happen if the caller checked VOP_ACCESS() with a
2198 	 * 32 bit ACE mask and already had the appropriate permissions.
2199 	 */
2200 	if (skipaclchk) {
2201 		*working_mode = 0;
2202 		return (0);
2203 	}
2204 
2205 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2206 
2207 	mutex_enter(&zp->z_acl_lock);
2208 
2209 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2210 	if (error != 0) {
2211 		mutex_exit(&zp->z_acl_lock);
2212 		return (error);
2213 	}
2214 
2215 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2216 	    &iflags, &type)) {
2217 
2218 		if (iflags & ACE_INHERIT_ONLY_ACE)
2219 			continue;
2220 
2221 		entry_type = (iflags & ACE_TYPE_FLAGS);
2222 
2223 		checkit = B_FALSE;
2224 
2225 		switch (entry_type) {
2226 		case ACE_OWNER:
2227 			if (uid == fowner)
2228 				checkit = B_TRUE;
2229 			break;
2230 		case OWNING_GROUP:
2231 			who = gowner;
2232 			/*FALLTHROUGH*/
2233 		case ACE_IDENTIFIER_GROUP:
2234 			checkit = zfs_groupmember(zfsvfs, who, cr);
2235 			break;
2236 		case ACE_EVERYONE:
2237 			checkit = B_TRUE;
2238 			break;
2239 
2240 		/* USER Entry */
2241 		default:
2242 			if (entry_type == 0) {
2243 				uid_t newid;
2244 
2245 				zfs_fuid_map_id(zfsvfs, who, cr,
2246 				    ZFS_ACE_USER, &newid);
2247 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2248 				    uid == newid)
2249 					checkit = B_TRUE;
2250 				break;
2251 			} else {
2252 				zfs_acl_free(aclp);
2253 				mutex_exit(&zp->z_acl_lock);
2254 				return (EIO);
2255 			}
2256 		}
2257 
2258 		if (checkit) {
2259 			if (access_mask & *working_mode) {
2260 				if (type == ALLOW) {
2261 					*working_mode &=
2262 					    ~(*working_mode & access_mask);
2263 					if (*working_mode == 0) {
2264 						access_deny = 0;
2265 					}
2266 				} else if (type == DENY) {
2267 					access_deny = EACCES;
2268 				}
2269 			}
2270 		}
2271 
2272 		if (access_deny != ACCESS_UNDETERMINED)
2273 			break;
2274 	}
2275 
2276 	mutex_exit(&zp->z_acl_lock);
2277 	zfs_acl_free(aclp);
2278 out:
2279 	return (access_deny);
2280 }
2281 
2282 static int
2283 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2284     cred_t *cr)
2285 {
2286 	if (*working_mode != ACE_WRITE_DATA)
2287 		return (EACCES);
2288 
2289 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2290 	    check_privs, B_FALSE, cr));
2291 }
2292 
2293 /*
2294  * Determine whether Access should be granted/denied, invoking least
2295  * priv subsytem when a deny is determined.
2296  */
2297 int
2298 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2299 {
2300 	uint32_t	working_mode;
2301 	int		error;
2302 	int		is_attr;
2303 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2304 	boolean_t 	check_privs;
2305 	znode_t		*xzp;
2306 	znode_t 	*check_zp = zp;
2307 
2308 	is_attr = ((zp->z_phys->zp_flags & ZFS_XATTR) &&
2309 	    (ZTOV(zp)->v_type == VDIR));
2310 
2311 	/*
2312 	 * If attribute then validate against base file
2313 	 */
2314 	if (is_attr) {
2315 		if ((error = zfs_zget(zp->z_zfsvfs,
2316 		    zp->z_phys->zp_parent, &xzp)) != 0)	{
2317 			return (error);
2318 		}
2319 
2320 		check_zp = xzp;
2321 
2322 		/*
2323 		 * fixup mode to map to xattr perms
2324 		 */
2325 
2326 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2327 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2328 			mode |= ACE_WRITE_NAMED_ATTRS;
2329 		}
2330 
2331 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2332 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2333 			mode |= ACE_READ_NAMED_ATTRS;
2334 		}
2335 	}
2336 
2337 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2338 	    &check_privs, skipaclchk, cr)) == 0) {
2339 		if (is_attr)
2340 			VN_RELE(ZTOV(xzp));
2341 		return (0);
2342 	}
2343 
2344 	if (error && check_privs == B_FALSE) {
2345 		if (is_attr)
2346 			VN_RELE(ZTOV(xzp));
2347 		return (error);
2348 	}
2349 
2350 	if (error && (flags & V_APPEND)) {
2351 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2352 	}
2353 
2354 	if (error && check_privs) {
2355 		uid_t		owner;
2356 		mode_t		checkmode = 0;
2357 
2358 		zfs_fuid_map_id(zfsvfs, check_zp->z_phys->zp_uid, cr,
2359 		    ZFS_OWNER, &owner);
2360 
2361 		/*
2362 		 * First check for implicit owner permission on
2363 		 * read_acl/read_attributes
2364 		 */
2365 
2366 		error = 0;
2367 		ASSERT(working_mode != 0);
2368 
2369 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2370 		    owner == crgetuid(cr)))
2371 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2372 
2373 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2374 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES))
2375 			checkmode |= VREAD;
2376 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2377 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES))
2378 			checkmode |= VWRITE;
2379 		if (working_mode & ACE_EXECUTE)
2380 			checkmode |= VEXEC;
2381 
2382 		if (checkmode)
2383 			error = secpolicy_vnode_access(cr, ZTOV(check_zp),
2384 			    owner, checkmode);
2385 
2386 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2387 			error = secpolicy_vnode_create_gid(cr);
2388 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2389 			error = secpolicy_vnode_setdac(cr, owner);
2390 
2391 		if (error == 0 && (working_mode &
2392 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2393 			error = secpolicy_vnode_remove(cr);
2394 
2395 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE))
2396 			error = secpolicy_vnode_owner(cr, owner);
2397 
2398 		if (error == 0) {
2399 			/*
2400 			 * See if any bits other than those already checked
2401 			 * for are still present.  If so then return EACCES
2402 			 */
2403 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2404 				error = EACCES;
2405 			}
2406 		}
2407 	}
2408 
2409 	if (is_attr)
2410 		VN_RELE(ZTOV(xzp));
2411 
2412 	return (error);
2413 }
2414 
2415 /*
2416  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2417  * native ACL format and call zfs_zaccess()
2418  */
2419 int
2420 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2421 {
2422 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2423 }
2424 
2425 /*
2426  * Access function for secpolicy_vnode_setattr
2427  */
2428 int
2429 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2430 {
2431 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2432 
2433 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2434 }
2435 
2436 static int
2437 zfs_delete_final_check(znode_t *zp, znode_t *dzp, cred_t *cr)
2438 {
2439 	int error;
2440 	uid_t downer;
2441 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2442 
2443 	zfs_fuid_map_id(zfsvfs, dzp->z_phys->zp_uid, cr, ZFS_OWNER, &downer);
2444 
2445 	error = secpolicy_vnode_access(cr, ZTOV(zp), downer, S_IWRITE|S_IEXEC);
2446 
2447 	if (error == 0)
2448 		error = zfs_sticky_remove_access(dzp, zp, cr);
2449 
2450 	return (error);
2451 }
2452 
2453 /*
2454  * Determine whether Access should be granted/deny, without
2455  * consulting least priv subsystem.
2456  *
2457  *
2458  * The following chart is the recommended NFSv4 enforcement for
2459  * ability to delete an object.
2460  *
2461  *      -------------------------------------------------------
2462  *      |   Parent Dir  |           Target Object Permissions |
2463  *      |  permissions  |                                     |
2464  *      -------------------------------------------------------
2465  *      |               | ACL Allows | ACL Denies| Delete     |
2466  *      |               |  Delete    |  Delete   | unspecified|
2467  *      -------------------------------------------------------
2468  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2469  *      |  DELETE_CHILD |                                     |
2470  *      -------------------------------------------------------
2471  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2472  *      |  DELETE_CHILD |            |           |            |
2473  *      -------------------------------------------------------
2474  *      | ACL specifies |            |           |            |
2475  *      | only allow    | Permit     | Permit    | Permit     |
2476  *      | write and     |            |           |            |
2477  *      | execute       |            |           |            |
2478  *      -------------------------------------------------------
2479  *      | ACL denies    |            |           |            |
2480  *      | write and     | Permit     | Deny      | Deny       |
2481  *      | execute       |            |           |            |
2482  *      -------------------------------------------------------
2483  *         ^
2484  *         |
2485  *         No search privilege, can't even look up file?
2486  *
2487  */
2488 int
2489 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2490 {
2491 	uint32_t dzp_working_mode = 0;
2492 	uint32_t zp_working_mode = 0;
2493 	int dzp_error, zp_error;
2494 	boolean_t dzpcheck_privs = B_TRUE;
2495 	boolean_t zpcheck_privs = B_TRUE;
2496 
2497 	/*
2498 	 * Arghh, this check is going to require a couple of questions
2499 	 * to be asked.  We want specific DELETE permissions to
2500 	 * take precedence over WRITE/EXECUTE.  We don't
2501 	 * want an ACL such as this to mess us up.
2502 	 * user:joe:write_data:deny,user:joe:delete:allow
2503 	 *
2504 	 * However, deny permissions may ultimately be overridden
2505 	 * by secpolicy_vnode_access().
2506 	 */
2507 
2508 	if (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2509 		return (EPERM);
2510 
2511 	dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2512 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2513 	zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2514 	    &zpcheck_privs, B_FALSE, cr);
2515 
2516 	if ((dzp_error && dzpcheck_privs == B_FALSE) ||
2517 	    (zp_error && zpcheck_privs == B_FALSE))
2518 		return (dzp_error);
2519 
2520 	/*
2521 	 * First check the first row.
2522 	 * We only need to see if parent Allows delete_child
2523 	 */
2524 	if ((dzp_working_mode & ACE_DELETE_CHILD) == 0)
2525 		return (0);
2526 
2527 	/*
2528 	 * Second row
2529 	 * we already have the necessary information in
2530 	 * zp_working_mode, zp_error and dzp_error.
2531 	 */
2532 
2533 	if ((zp_working_mode & ACE_DELETE) == 0)
2534 		return (0);
2535 
2536 	/*
2537 	 * Now zp_error should either be EACCES which indicates
2538 	 * a "deny" delete entry or ACCESS_UNDETERMINED if the "delete"
2539 	 * entry exists on the target.
2540 	 *
2541 	 * dzp_error should be either EACCES which indicates a "deny"
2542 	 * entry for delete_child or ACCESS_UNDETERMINED if no delete_child
2543 	 * entry exists.  If value is EACCES then we are done
2544 	 * and zfs_delete_final_check() will make the final decision
2545 	 * regarding to allow the delete.
2546 	 */
2547 
2548 	ASSERT(zp_error != 0 && dzp_error != 0);
2549 	if (dzp_error == EACCES)
2550 		return (zfs_delete_final_check(zp, dzp, cr));
2551 
2552 	/*
2553 	 * Third Row
2554 	 * Only need to check for write/execute on parent
2555 	 */
2556 
2557 	dzp_error = zfs_zaccess_common(dzp, ACE_WRITE_DATA|ACE_EXECUTE,
2558 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2559 
2560 	if (dzp_error && dzpcheck_privs == B_FALSE)
2561 		return (dzp_error);
2562 
2563 	if ((dzp_working_mode & (ACE_WRITE_DATA|ACE_EXECUTE)) == 0)
2564 		return (zfs_sticky_remove_access(dzp, zp, cr));
2565 
2566 	/*
2567 	 * Fourth Row
2568 	 */
2569 
2570 	if (((dzp_working_mode & (ACE_WRITE_DATA|ACE_EXECUTE)) != 0) &&
2571 	    ((zp_working_mode & ACE_DELETE) == 0))
2572 		return (zfs_sticky_remove_access(dzp, zp, cr));
2573 
2574 	return (zfs_delete_final_check(zp, dzp, cr));
2575 }
2576 
2577 int
2578 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2579     znode_t *tzp, cred_t *cr)
2580 {
2581 	int add_perm;
2582 	int error;
2583 
2584 	if (szp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
2585 		return (EACCES);
2586 
2587 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2588 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2589 
2590 	/*
2591 	 * Rename permissions are combination of delete permission +
2592 	 * add file/subdir permission.
2593 	 */
2594 
2595 	/*
2596 	 * first make sure we do the delete portion.
2597 	 *
2598 	 * If that succeeds then check for add_file/add_subdir permissions
2599 	 */
2600 
2601 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2602 		return (error);
2603 
2604 	/*
2605 	 * If we have a tzp, see if we can delete it?
2606 	 */
2607 	if (tzp) {
2608 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2609 			return (error);
2610 	}
2611 
2612 	/*
2613 	 * Now check for add permissions
2614 	 */
2615 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2616 
2617 	return (error);
2618 }
2619