xref: /titanic_51/usr/src/uts/common/fs/zfs/zfs_acl.c (revision ded9341448cd6e2573619c7f6fe98909bdd35ec6)
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 
1160 static int
1161 zfs_acl_get_owner_fuids(znode_t *zp, uint64_t *fuid, uint64_t *fgid)
1162 {
1163 	int count = 0;
1164 	sa_bulk_attr_t	bulk[2];
1165 	int error;
1166 
1167 	if (IS_EPHEMERAL(zp->z_uid) || IS_EPHEMERAL(zp->z_gid)) {
1168 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zp->z_zfsvfs), NULL,
1169 		    &fuid, sizeof (fuid));
1170 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zp->z_zfsvfs), NULL,
1171 		    &fgid, sizeof (fuid));
1172 		if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
1173 			return (error);
1174 		}
1175 	} else {
1176 		*fuid = zp->z_uid;
1177 		*fgid = zp->z_gid;
1178 	}
1179 	return (0);
1180 }
1181 
1182 int
1183 zfs_acl_chown_setattr(znode_t *zp)
1184 {
1185 	int error;
1186 	zfs_acl_t *aclp;
1187 	uint64_t fuid, fgid;
1188 
1189 	ASSERT(MUTEX_HELD(&zp->z_lock));
1190 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1191 	if ((error = zfs_acl_get_owner_fuids(zp, &fuid, &fgid)) != 0)
1192 		return (error);
1193 
1194 	if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1195 		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1196 		    &zp->z_pflags, fuid, fgid);
1197 	return (error);
1198 }
1199 
1200 /*
1201  * common code for setting ACLs.
1202  *
1203  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1204  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1205  * already checked the acl and knows whether to inherit.
1206  */
1207 int
1208 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1209 {
1210 	int			error;
1211 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1212 	dmu_object_type_t	otype;
1213 	zfs_acl_locator_cb_t	locate = { 0 };
1214 	uint64_t		mode;
1215 	sa_bulk_attr_t		bulk[5];
1216 	uint64_t		ctime[2];
1217 	int			count = 0;
1218 	uint64_t		fuid, fgid;
1219 
1220 	mode = zp->z_mode;
1221 
1222 	if ((error = zfs_acl_get_owner_fuids(zp, &fuid, &fgid)) != 0)
1223 		return (error);
1224 
1225 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags, fuid, fgid);
1226 
1227 	zp->z_mode = mode;
1228 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1229 	    &mode, sizeof (mode));
1230 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1231 	    &zp->z_pflags, sizeof (zp->z_pflags));
1232 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1233 	    &ctime, sizeof (ctime));
1234 
1235 	if (zp->z_acl_cached) {
1236 		zfs_acl_free(zp->z_acl_cached);
1237 		zp->z_acl_cached = NULL;
1238 	}
1239 
1240 	/*
1241 	 * Upgrade needed?
1242 	 */
1243 	if (!zfsvfs->z_use_fuids) {
1244 		otype = DMU_OT_OLDACL;
1245 	} else {
1246 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1247 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1248 			zfs_acl_xform(zp, aclp, cr);
1249 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1250 		otype = DMU_OT_ACL;
1251 	}
1252 
1253 	/*
1254 	 * Arrgh, we have to handle old on disk format
1255 	 * as well as newer (preferred) SA format.
1256 	 */
1257 
1258 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1259 		locate.cb_aclp = aclp;
1260 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1261 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1262 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1263 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1264 	} else { /* Painful legacy way */
1265 		zfs_acl_node_t *aclnode;
1266 		uint64_t off = 0;
1267 		zfs_acl_phys_t acl_phys;
1268 		uint64_t aoid;
1269 
1270 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1271 		    &acl_phys, sizeof (acl_phys))) != 0)
1272 			return (error);
1273 
1274 		aoid = acl_phys.z_acl_extern_obj;
1275 
1276 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1277 			/*
1278 			 * If ACL was previously external and we are now
1279 			 * converting to new ACL format then release old
1280 			 * ACL object and create a new one.
1281 			 */
1282 			if (aoid &&
1283 			    aclp->z_version != acl_phys.z_acl_version) {
1284 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1285 				if (error)
1286 					return (error);
1287 				aoid = 0;
1288 			}
1289 			if (aoid == 0) {
1290 				aoid = dmu_object_alloc(zfsvfs->z_os,
1291 				    otype, aclp->z_acl_bytes,
1292 				    otype == DMU_OT_ACL ?
1293 				    DMU_OT_SYSACL : DMU_OT_NONE,
1294 				    otype == DMU_OT_ACL ?
1295 				    DN_MAX_BONUSLEN : 0, tx);
1296 			} else {
1297 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1298 				    aoid, aclp->z_acl_bytes, 0, tx);
1299 			}
1300 			acl_phys.z_acl_extern_obj = aoid;
1301 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1302 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1303 				if (aclnode->z_ace_count == 0)
1304 					continue;
1305 				dmu_write(zfsvfs->z_os, aoid, off,
1306 				    aclnode->z_size, aclnode->z_acldata, tx);
1307 				off += aclnode->z_size;
1308 			}
1309 		} else {
1310 			void *start = acl_phys.z_ace_data;
1311 			/*
1312 			 * Migrating back embedded?
1313 			 */
1314 			if (acl_phys.z_acl_extern_obj) {
1315 				error = dmu_object_free(zfsvfs->z_os,
1316 				    acl_phys.z_acl_extern_obj, tx);
1317 				if (error)
1318 					return (error);
1319 				acl_phys.z_acl_extern_obj = 0;
1320 			}
1321 
1322 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1323 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1324 				if (aclnode->z_ace_count == 0)
1325 					continue;
1326 				bcopy(aclnode->z_acldata, start,
1327 				    aclnode->z_size);
1328 				start = (caddr_t)start + aclnode->z_size;
1329 			}
1330 		}
1331 		/*
1332 		 * If Old version then swap count/bytes to match old
1333 		 * layout of znode_acl_phys_t.
1334 		 */
1335 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1336 			acl_phys.z_acl_size = aclp->z_acl_count;
1337 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1338 		} else {
1339 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1340 			acl_phys.z_acl_count = aclp->z_acl_count;
1341 		}
1342 		acl_phys.z_acl_version = aclp->z_version;
1343 
1344 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1345 		    &acl_phys, sizeof (acl_phys));
1346 	}
1347 
1348 	/*
1349 	 * Replace ACL wide bits, but first clear them.
1350 	 */
1351 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1352 
1353 	zp->z_pflags |= aclp->z_hints;
1354 
1355 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1356 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1357 
1358 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime, B_TRUE);
1359 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1360 }
1361 
1362 /*
1363  * Update access mask for prepended ACE
1364  *
1365  * This applies the "groupmask" value for aclmode property.
1366  */
1367 static void
1368 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1369     mode_t mode, uint64_t owner)
1370 {
1371 	int	rmask, wmask, xmask;
1372 	int	user_ace;
1373 	uint16_t aceflags;
1374 	uint32_t origmask, acepmask;
1375 	uint64_t fuid;
1376 
1377 	aceflags = aclp->z_ops.ace_flags_get(acep);
1378 	fuid = aclp->z_ops.ace_who_get(acep);
1379 	origmask = aclp->z_ops.ace_mask_get(origacep);
1380 	acepmask = aclp->z_ops.ace_mask_get(acep);
1381 
1382 	user_ace = (!(aceflags &
1383 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1384 
1385 	if (user_ace && (fuid == owner)) {
1386 		rmask = S_IRUSR;
1387 		wmask = S_IWUSR;
1388 		xmask = S_IXUSR;
1389 	} else {
1390 		rmask = S_IRGRP;
1391 		wmask = S_IWGRP;
1392 		xmask = S_IXGRP;
1393 	}
1394 
1395 	if (origmask & ACE_READ_DATA) {
1396 		if (mode & rmask) {
1397 			acepmask &= ~ACE_READ_DATA;
1398 		} else {
1399 			acepmask |= ACE_READ_DATA;
1400 		}
1401 	}
1402 
1403 	if (origmask & ACE_WRITE_DATA) {
1404 		if (mode & wmask) {
1405 			acepmask &= ~ACE_WRITE_DATA;
1406 		} else {
1407 			acepmask |= ACE_WRITE_DATA;
1408 		}
1409 	}
1410 
1411 	if (origmask & ACE_APPEND_DATA) {
1412 		if (mode & wmask) {
1413 			acepmask &= ~ACE_APPEND_DATA;
1414 		} else {
1415 			acepmask |= ACE_APPEND_DATA;
1416 		}
1417 	}
1418 
1419 	if (origmask & ACE_EXECUTE) {
1420 		if (mode & xmask) {
1421 			acepmask &= ~ACE_EXECUTE;
1422 		} else {
1423 			acepmask |= ACE_EXECUTE;
1424 		}
1425 	}
1426 	aclp->z_ops.ace_mask_set(acep, acepmask);
1427 }
1428 
1429 static void
1430 zfs_acl_chmod(zfsvfs_t *zfsvfs, uint64_t mode, zfs_acl_t *aclp)
1431 {
1432 	void		*acep = NULL;
1433 	uint64_t	who;
1434 	int		new_count, new_bytes;
1435 	int		ace_size;
1436 	int 		entry_type;
1437 	uint16_t	iflags, type;
1438 	uint32_t	access_mask;
1439 	zfs_acl_node_t	*newnode;
1440 	size_t 		abstract_size = aclp->z_ops.ace_abstract_size();
1441 	void 		*zacep;
1442 	uint32_t 	owner, group, everyone;
1443 	uint32_t	deny1, deny2, allow0;
1444 
1445 	new_count = new_bytes = 0;
1446 
1447 	acl_trivial_access_masks((mode_t)mode, &allow0, &deny1, &deny2,
1448 	    &owner, &group, &everyone);
1449 
1450 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1451 
1452 	zacep = newnode->z_acldata;
1453 	if (allow0) {
1454 		zfs_set_ace(aclp, zacep, allow0, ALLOW, -1, ACE_OWNER);
1455 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1456 		new_count++;
1457 		new_bytes += abstract_size;
1458 	} if (deny1) {
1459 		zfs_set_ace(aclp, zacep, deny1, DENY, -1, ACE_OWNER);
1460 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1461 		new_count++;
1462 		new_bytes += abstract_size;
1463 	}
1464 	if (deny2) {
1465 		zfs_set_ace(aclp, zacep, deny2, DENY, -1, OWNING_GROUP);
1466 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1467 		new_count++;
1468 		new_bytes += abstract_size;
1469 	}
1470 
1471 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1472 	    &iflags, &type)) {
1473 		uint16_t inherit_flags;
1474 
1475 		entry_type = (iflags & ACE_TYPE_FLAGS);
1476 		inherit_flags = (iflags & ALL_INHERIT);
1477 
1478 		if ((entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1479 		    (entry_type == OWNING_GROUP)) &&
1480 		    ((inherit_flags & ACE_INHERIT_ONLY_ACE) == 0)) {
1481 			continue;
1482 		}
1483 
1484 		if ((type != ALLOW && type != DENY) ||
1485 		    (inherit_flags & ACE_INHERIT_ONLY_ACE)) {
1486 			if (inherit_flags)
1487 				aclp->z_hints |= ZFS_INHERIT_ACE;
1488 			switch (type) {
1489 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1490 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1491 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1492 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1493 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1494 				break;
1495 			}
1496 		} else {
1497 
1498 			/*
1499 			 * Limit permissions to be no greater than
1500 			 * group permissions
1501 			 */
1502 			if (zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) {
1503 				if (!(mode & S_IRGRP))
1504 					access_mask &= ~ACE_READ_DATA;
1505 				if (!(mode & S_IWGRP))
1506 					access_mask &=
1507 					    ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1508 				if (!(mode & S_IXGRP))
1509 					access_mask &= ~ACE_EXECUTE;
1510 				access_mask &=
1511 				    ~(ACE_WRITE_OWNER|ACE_WRITE_ACL|
1512 				    ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS);
1513 			}
1514 		}
1515 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1516 		ace_size = aclp->z_ops.ace_size(acep);
1517 		zacep = (void *)((uintptr_t)zacep + ace_size);
1518 		new_count++;
1519 		new_bytes += ace_size;
1520 	}
1521 	zfs_set_ace(aclp, zacep, owner, 0, -1, ACE_OWNER);
1522 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1523 	zfs_set_ace(aclp, zacep, group, 0, -1, OWNING_GROUP);
1524 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1525 	zfs_set_ace(aclp, zacep, everyone, 0, -1, ACE_EVERYONE);
1526 
1527 	new_count += 3;
1528 	new_bytes += abstract_size * 3;
1529 	zfs_acl_release_nodes(aclp);
1530 	aclp->z_acl_count = new_count;
1531 	aclp->z_acl_bytes = new_bytes;
1532 	newnode->z_ace_count = new_count;
1533 	newnode->z_size = new_bytes;
1534 	list_insert_tail(&aclp->z_acl, newnode);
1535 }
1536 
1537 int
1538 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1539 {
1540 	mutex_enter(&zp->z_acl_lock);
1541 	mutex_enter(&zp->z_lock);
1542 	*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1543 	(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1544 	zfs_acl_chmod(zp->z_zfsvfs, mode, *aclp);
1545 	mutex_exit(&zp->z_lock);
1546 	mutex_exit(&zp->z_acl_lock);
1547 	ASSERT(*aclp);
1548 	return (0);
1549 }
1550 
1551 /*
1552  * strip off write_owner and write_acl
1553  */
1554 static void
1555 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1556 {
1557 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1558 
1559 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1560 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1561 		mask &= ~RESTRICTED_CLEAR;
1562 		aclp->z_ops.ace_mask_set(acep, mask);
1563 	}
1564 }
1565 
1566 /*
1567  * Should ACE be inherited?
1568  */
1569 static int
1570 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1571 {
1572 	int	iflags = (acep_flags & 0xf);
1573 
1574 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1575 		return (1);
1576 	else if (iflags & ACE_FILE_INHERIT_ACE)
1577 		return (!((vtype == VDIR) &&
1578 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1579 	return (0);
1580 }
1581 
1582 /*
1583  * inherit inheritable ACEs from parent
1584  */
1585 static zfs_acl_t *
1586 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1587     uint64_t mode, boolean_t *need_chmod)
1588 {
1589 	void		*pacep;
1590 	void		*acep;
1591 	zfs_acl_node_t  *aclnode;
1592 	zfs_acl_t	*aclp = NULL;
1593 	uint64_t	who;
1594 	uint32_t	access_mask;
1595 	uint16_t	iflags, newflags, type;
1596 	size_t		ace_size;
1597 	void		*data1, *data2;
1598 	size_t		data1sz, data2sz;
1599 	boolean_t	vdir = vtype == VDIR;
1600 	boolean_t	vreg = vtype == VREG;
1601 	boolean_t	passthrough, passthrough_x, noallow;
1602 
1603 	passthrough_x =
1604 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1605 	passthrough = passthrough_x ||
1606 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1607 	noallow =
1608 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1609 
1610 	*need_chmod = B_TRUE;
1611 	pacep = NULL;
1612 	aclp = zfs_acl_alloc(paclp->z_version);
1613 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD || vtype == VLNK)
1614 		return (aclp);
1615 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1616 	    &access_mask, &iflags, &type)) {
1617 
1618 		/*
1619 		 * don't inherit bogus ACEs
1620 		 */
1621 		if (!zfs_acl_valid_ace_type(type, iflags))
1622 			continue;
1623 
1624 		if (noallow && type == ALLOW)
1625 			continue;
1626 
1627 		ace_size = aclp->z_ops.ace_size(pacep);
1628 
1629 		if (!zfs_ace_can_use(vtype, iflags))
1630 			continue;
1631 
1632 		/*
1633 		 * If owner@, group@, or everyone@ inheritable
1634 		 * then zfs_acl_chmod() isn't needed.
1635 		 */
1636 		if (passthrough &&
1637 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1638 		    ((iflags & OWNING_GROUP) ==
1639 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1640 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1641 			*need_chmod = B_FALSE;
1642 		}
1643 
1644 		if (!vdir && passthrough_x &&
1645 		    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1646 			access_mask &= ~ACE_EXECUTE;
1647 		}
1648 
1649 		aclnode = zfs_acl_node_alloc(ace_size);
1650 		list_insert_tail(&aclp->z_acl, aclnode);
1651 		acep = aclnode->z_acldata;
1652 
1653 		zfs_set_ace(aclp, acep, access_mask, type,
1654 		    who, iflags|ACE_INHERITED_ACE);
1655 
1656 		/*
1657 		 * Copy special opaque data if any
1658 		 */
1659 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1660 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1661 			    &data2)) == data1sz);
1662 			bcopy(data1, data2, data2sz);
1663 		}
1664 
1665 		aclp->z_acl_count++;
1666 		aclnode->z_ace_count++;
1667 		aclp->z_acl_bytes += aclnode->z_size;
1668 		newflags = aclp->z_ops.ace_flags_get(acep);
1669 
1670 		if (vdir)
1671 			aclp->z_hints |= ZFS_INHERIT_ACE;
1672 
1673 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1674 			newflags &= ~ALL_INHERIT;
1675 			aclp->z_ops.ace_flags_set(acep,
1676 			    newflags|ACE_INHERITED_ACE);
1677 			zfs_restricted_update(zfsvfs, aclp, acep);
1678 			continue;
1679 		}
1680 
1681 		ASSERT(vdir);
1682 
1683 		/*
1684 		 * If only FILE_INHERIT is set then turn on
1685 		 * inherit_only
1686 		 */
1687 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1688 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1689 			newflags |= ACE_INHERIT_ONLY_ACE;
1690 			aclp->z_ops.ace_flags_set(acep,
1691 			    newflags|ACE_INHERITED_ACE);
1692 		} else {
1693 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1694 			aclp->z_ops.ace_flags_set(acep,
1695 			    newflags|ACE_INHERITED_ACE);
1696 		}
1697 	}
1698 	return (aclp);
1699 }
1700 
1701 /*
1702  * Create file system object initial permissions
1703  * including inheritable ACEs.
1704  */
1705 int
1706 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1707     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1708 {
1709 	int		error;
1710 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1711 	zfs_acl_t	*paclp;
1712 	gid_t		gid;
1713 	boolean_t	need_chmod = B_TRUE;
1714 	boolean_t	inherited = B_FALSE;
1715 	uint64_t	parentgid;
1716 
1717 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1718 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1719 
1720 	if (vsecp)
1721 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1722 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1723 			return (error);
1724 	/*
1725 	 * Determine uid and gid.
1726 	 */
1727 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1728 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1729 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1730 		    (uint64_t)vap->va_uid, cr,
1731 		    ZFS_OWNER, &acl_ids->z_fuidp);
1732 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1733 		    (uint64_t)vap->va_gid, cr,
1734 		    ZFS_GROUP, &acl_ids->z_fuidp);
1735 		gid = vap->va_gid;
1736 	} else {
1737 		if (IS_EPHEMERAL(dzp->z_gid))
1738 			VERIFY(0 == sa_lookup(dzp->z_sa_hdl, SA_ZPL_GID(zfsvfs),
1739 			    &parentgid, sizeof (parentgid)));
1740 		else
1741 			parentgid = (uint64_t)dzp->z_gid;
1742 
1743 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1744 		    cr, &acl_ids->z_fuidp);
1745 		acl_ids->z_fgid = 0;
1746 		if (vap->va_mask & AT_GID)  {
1747 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1748 			    (uint64_t)vap->va_gid,
1749 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1750 			gid = vap->va_gid;
1751 			if (acl_ids->z_fgid != parentgid &&
1752 			    !groupmember(vap->va_gid, cr) &&
1753 			    secpolicy_vnode_create_gid(cr) != 0)
1754 				acl_ids->z_fgid = 0;
1755 		}
1756 		if (acl_ids->z_fgid == 0) {
1757 			if (dzp->z_mode & S_ISGID) {
1758 				char		*domain;
1759 				uint32_t	rid;
1760 
1761 				acl_ids->z_fgid = parentgid;
1762 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1763 				    cr, ZFS_GROUP);
1764 
1765 				if (zfsvfs->z_use_fuids &&
1766 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1767 					domain = zfs_fuid_idx_domain(
1768 					    &zfsvfs->z_fuid_idx,
1769 					    FUID_INDEX(acl_ids->z_fgid));
1770 					rid = FUID_RID(acl_ids->z_fgid);
1771 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1772 					    domain, rid,
1773 					    FUID_INDEX(acl_ids->z_fgid),
1774 					    acl_ids->z_fgid, ZFS_GROUP);
1775 				}
1776 			} else {
1777 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1778 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1779 				gid = crgetgid(cr);
1780 			}
1781 		}
1782 	}
1783 
1784 	/*
1785 	 * If we're creating a directory, and the parent directory has the
1786 	 * set-GID bit set, set in on the new directory.
1787 	 * Otherwise, if the user is neither privileged nor a member of the
1788 	 * file's new group, clear the file's set-GID bit.
1789 	 */
1790 
1791 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1792 	    (vap->va_type == VDIR)) {
1793 		acl_ids->z_mode |= S_ISGID;
1794 	} else {
1795 		if ((acl_ids->z_mode & S_ISGID) &&
1796 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1797 			acl_ids->z_mode &= ~S_ISGID;
1798 	}
1799 
1800 	if (acl_ids->z_aclp == NULL) {
1801 		mutex_enter(&dzp->z_acl_lock);
1802 		mutex_enter(&dzp->z_lock);
1803 		if (!(flag & IS_ROOT_NODE) && (ZTOV(dzp)->v_type == VDIR &&
1804 		    (dzp->z_pflags & ZFS_INHERIT_ACE)) &&
1805 		    !(dzp->z_pflags & ZFS_XATTR)) {
1806 			VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1807 			    &paclp, B_FALSE));
1808 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1809 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1810 			inherited = B_TRUE;
1811 		} else {
1812 			acl_ids->z_aclp =
1813 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1814 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1815 		}
1816 		mutex_exit(&dzp->z_lock);
1817 		mutex_exit(&dzp->z_acl_lock);
1818 		if (need_chmod) {
1819 			acl_ids->z_aclp->z_hints |= (vap->va_type == VDIR) ?
1820 			    ZFS_ACL_AUTO_INHERIT : 0;
1821 			zfs_acl_chmod(zfsvfs, acl_ids->z_mode, acl_ids->z_aclp);
1822 		}
1823 	}
1824 
1825 	if (inherited || vsecp) {
1826 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1827 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1828 		    acl_ids->z_fuid, acl_ids->z_fgid);
1829 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1830 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1831 	}
1832 
1833 	return (0);
1834 }
1835 
1836 /*
1837  * Free ACL and fuid_infop, but not the acl_ids structure
1838  */
1839 void
1840 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1841 {
1842 	if (acl_ids->z_aclp)
1843 		zfs_acl_free(acl_ids->z_aclp);
1844 	if (acl_ids->z_fuidp)
1845 		zfs_fuid_info_free(acl_ids->z_fuidp);
1846 	acl_ids->z_aclp = NULL;
1847 	acl_ids->z_fuidp = NULL;
1848 }
1849 
1850 boolean_t
1851 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1852 {
1853 	return (zfs_fuid_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1854 	    zfs_fuid_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1855 }
1856 
1857 /*
1858  * Retrieve a files ACL
1859  */
1860 int
1861 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1862 {
1863 	zfs_acl_t	*aclp;
1864 	ulong_t		mask;
1865 	int		error;
1866 	int 		count = 0;
1867 	int		largeace = 0;
1868 
1869 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1870 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1871 
1872 	if (mask == 0)
1873 		return (ENOSYS);
1874 
1875 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1876 		return (error);
1877 
1878 	mutex_enter(&zp->z_acl_lock);
1879 
1880 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1881 	if (error != 0) {
1882 		mutex_exit(&zp->z_acl_lock);
1883 		return (error);
1884 	}
1885 
1886 	/*
1887 	 * Scan ACL to determine number of ACEs
1888 	 */
1889 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1890 		void *zacep = NULL;
1891 		uint64_t who;
1892 		uint32_t access_mask;
1893 		uint16_t type, iflags;
1894 
1895 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1896 		    &who, &access_mask, &iflags, &type)) {
1897 			switch (type) {
1898 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1899 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1900 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1901 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1902 				largeace++;
1903 				continue;
1904 			default:
1905 				count++;
1906 			}
1907 		}
1908 		vsecp->vsa_aclcnt = count;
1909 	} else
1910 		count = (int)aclp->z_acl_count;
1911 
1912 	if (mask & VSA_ACECNT) {
1913 		vsecp->vsa_aclcnt = count;
1914 	}
1915 
1916 	if (mask & VSA_ACE) {
1917 		size_t aclsz;
1918 
1919 		aclsz = count * sizeof (ace_t) +
1920 		    sizeof (ace_object_t) * largeace;
1921 
1922 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1923 		vsecp->vsa_aclentsz = aclsz;
1924 
1925 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1926 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1927 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1928 		else {
1929 			zfs_acl_node_t *aclnode;
1930 			void *start = vsecp->vsa_aclentp;
1931 
1932 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1933 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1934 				bcopy(aclnode->z_acldata, start,
1935 				    aclnode->z_size);
1936 				start = (caddr_t)start + aclnode->z_size;
1937 			}
1938 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1939 			    aclp->z_acl_bytes);
1940 		}
1941 	}
1942 	if (mask & VSA_ACE_ACLFLAGS) {
1943 		vsecp->vsa_aclflags = 0;
1944 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1945 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1946 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
1947 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1948 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1949 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1950 	}
1951 
1952 	mutex_exit(&zp->z_acl_lock);
1953 
1954 	return (0);
1955 }
1956 
1957 int
1958 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
1959     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1960 {
1961 	zfs_acl_t *aclp;
1962 	zfs_acl_node_t *aclnode;
1963 	int aclcnt = vsecp->vsa_aclcnt;
1964 	int error;
1965 
1966 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1967 		return (EINVAL);
1968 
1969 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1970 
1971 	aclp->z_hints = 0;
1972 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1973 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1974 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1975 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1976 		    aclcnt, &aclnode->z_size)) != 0) {
1977 			zfs_acl_free(aclp);
1978 			zfs_acl_node_free(aclnode);
1979 			return (error);
1980 		}
1981 	} else {
1982 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1983 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1984 		    &aclnode->z_size, fuidp, cr)) != 0) {
1985 			zfs_acl_free(aclp);
1986 			zfs_acl_node_free(aclnode);
1987 			return (error);
1988 		}
1989 	}
1990 	aclp->z_acl_bytes = aclnode->z_size;
1991 	aclnode->z_ace_count = aclcnt;
1992 	aclp->z_acl_count = aclcnt;
1993 	list_insert_head(&aclp->z_acl, aclnode);
1994 
1995 	/*
1996 	 * If flags are being set then add them to z_hints
1997 	 */
1998 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1999 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2000 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2001 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2002 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2003 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2004 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2005 	}
2006 
2007 	*zaclp = aclp;
2008 
2009 	return (0);
2010 }
2011 
2012 /*
2013  * Set a files ACL
2014  */
2015 int
2016 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2017 {
2018 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2019 	zilog_t		*zilog = zfsvfs->z_log;
2020 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2021 	dmu_tx_t	*tx;
2022 	int		error;
2023 	zfs_acl_t	*aclp;
2024 	zfs_fuid_info_t	*fuidp = NULL;
2025 	boolean_t	fuid_dirtied;
2026 	uint64_t	acl_obj;
2027 
2028 	if (mask == 0)
2029 		return (ENOSYS);
2030 
2031 	if (zp->z_pflags & ZFS_IMMUTABLE)
2032 		return (EPERM);
2033 
2034 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
2035 		return (error);
2036 
2037 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
2038 	    &aclp);
2039 	if (error)
2040 		return (error);
2041 
2042 	/*
2043 	 * If ACL wide flags aren't being set then preserve any
2044 	 * existing flags.
2045 	 */
2046 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2047 		aclp->z_hints |=
2048 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2049 	}
2050 top:
2051 	mutex_enter(&zp->z_acl_lock);
2052 	mutex_enter(&zp->z_lock);
2053 
2054 	tx = dmu_tx_create(zfsvfs->z_os);
2055 
2056 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2057 
2058 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2059 	if (fuid_dirtied)
2060 		zfs_fuid_txhold(zfsvfs, tx);
2061 
2062 	/*
2063 	 * If old version and ACL won't fit in bonus and we aren't
2064 	 * upgrading then take out necessary DMU holds
2065 	 */
2066 
2067 	if ((acl_obj = zfs_external_acl(zp)) != 0) {
2068 		if (zfsvfs->z_version <= ZPL_VERSION_SA &&
2069 		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2070 			dmu_tx_hold_free(tx, acl_obj, 0,
2071 			    DMU_OBJECT_END);
2072 		} else {
2073 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2074 		}
2075 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2076 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2077 	}
2078 
2079 	zfs_sa_upgrade_txholds(tx, zp);
2080 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2081 	if (error) {
2082 		mutex_exit(&zp->z_acl_lock);
2083 		mutex_exit(&zp->z_lock);
2084 
2085 		if (error == ERESTART) {
2086 			dmu_tx_wait(tx);
2087 			dmu_tx_abort(tx);
2088 			goto top;
2089 		}
2090 		dmu_tx_abort(tx);
2091 		zfs_acl_free(aclp);
2092 		return (error);
2093 	}
2094 
2095 	error = zfs_aclset_common(zp, aclp, cr, tx);
2096 	ASSERT(error == 0);
2097 	zp->z_acl_cached = aclp;
2098 
2099 	if (fuid_dirtied)
2100 		zfs_fuid_sync(zfsvfs, tx);
2101 
2102 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2103 
2104 	if (fuidp)
2105 		zfs_fuid_info_free(fuidp);
2106 	dmu_tx_commit(tx);
2107 done:
2108 	mutex_exit(&zp->z_lock);
2109 	mutex_exit(&zp->z_acl_lock);
2110 
2111 	return (error);
2112 }
2113 
2114 /*
2115  * Check accesses of interest (AoI) against attributes of the dataset
2116  * such as read-only.  Returns zero if no AoI conflict with dataset
2117  * attributes, otherwise an appropriate errno is returned.
2118  */
2119 static int
2120 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2121 {
2122 	if ((v4_mode & WRITE_MASK) &&
2123 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2124 	    (!IS_DEVVP(ZTOV(zp)) ||
2125 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2126 		return (EROFS);
2127 	}
2128 
2129 	/*
2130 	 * Only check for READONLY on non-directories.
2131 	 */
2132 	if ((v4_mode & WRITE_MASK_DATA) &&
2133 	    (((ZTOV(zp)->v_type != VDIR) &&
2134 	    (zp->z_pflags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2135 	    (ZTOV(zp)->v_type == VDIR &&
2136 	    (zp->z_pflags & ZFS_IMMUTABLE)))) {
2137 		return (EPERM);
2138 	}
2139 
2140 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2141 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2142 		return (EPERM);
2143 	}
2144 
2145 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2146 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2147 		return (EACCES);
2148 	}
2149 
2150 	return (0);
2151 }
2152 
2153 /*
2154  * The primary usage of this function is to loop through all of the
2155  * ACEs in the znode, determining what accesses of interest (AoI) to
2156  * the caller are allowed or denied.  The AoI are expressed as bits in
2157  * the working_mode parameter.  As each ACE is processed, bits covered
2158  * by that ACE are removed from the working_mode.  This removal
2159  * facilitates two things.  The first is that when the working mode is
2160  * empty (= 0), we know we've looked at all the AoI. The second is
2161  * that the ACE interpretation rules don't allow a later ACE to undo
2162  * something granted or denied by an earlier ACE.  Removing the
2163  * discovered access or denial enforces this rule.  At the end of
2164  * processing the ACEs, all AoI that were found to be denied are
2165  * placed into the working_mode, giving the caller a mask of denied
2166  * accesses.  Returns:
2167  *	0		if all AoI granted
2168  *	EACCESS 	if the denied mask is non-zero
2169  *	other error	if abnormal failure (e.g., IO error)
2170  *
2171  * A secondary usage of the function is to determine if any of the
2172  * AoI are granted.  If an ACE grants any access in
2173  * the working_mode, we immediately short circuit out of the function.
2174  * This mode is chosen by setting anyaccess to B_TRUE.  The
2175  * working_mode is not a denied access mask upon exit if the function
2176  * is used in this manner.
2177  */
2178 static int
2179 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2180     boolean_t anyaccess, cred_t *cr)
2181 {
2182 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2183 	zfs_acl_t	*aclp;
2184 	int		error;
2185 	uid_t		uid = crgetuid(cr);
2186 	uint64_t 	who;
2187 	uint16_t	type, iflags;
2188 	uint16_t	entry_type;
2189 	uint32_t	access_mask;
2190 	uint32_t	deny_mask = 0;
2191 	zfs_ace_hdr_t	*acep = NULL;
2192 	boolean_t	checkit;
2193 	uint64_t	gowner;
2194 
2195 	mutex_enter(&zp->z_acl_lock);
2196 
2197 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2198 	if (error != 0) {
2199 		mutex_exit(&zp->z_acl_lock);
2200 		return (error);
2201 	}
2202 
2203 	ASSERT(zp->z_acl_cached);
2204 
2205 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GID(zfsvfs),
2206 	    &gowner, sizeof (gowner))) != 0) {
2207 		mutex_exit(&zp->z_acl_lock);
2208 		return (error);
2209 	}
2210 
2211 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2212 	    &iflags, &type)) {
2213 		uint32_t mask_matched;
2214 
2215 		if (!zfs_acl_valid_ace_type(type, iflags))
2216 			continue;
2217 
2218 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2219 			continue;
2220 
2221 		/* Skip ACE if it does not affect any AoI */
2222 		mask_matched = (access_mask & *working_mode);
2223 		if (!mask_matched)
2224 			continue;
2225 
2226 		entry_type = (iflags & ACE_TYPE_FLAGS);
2227 
2228 		checkit = B_FALSE;
2229 
2230 		switch (entry_type) {
2231 		case ACE_OWNER:
2232 			if (uid == zp->z_uid)
2233 				checkit = B_TRUE;
2234 			break;
2235 		case OWNING_GROUP:
2236 			who = gowner;
2237 			/*FALLTHROUGH*/
2238 		case ACE_IDENTIFIER_GROUP:
2239 			checkit = zfs_groupmember(zfsvfs, who, cr);
2240 			break;
2241 		case ACE_EVERYONE:
2242 			checkit = B_TRUE;
2243 			break;
2244 
2245 		/* USER Entry */
2246 		default:
2247 			if (entry_type == 0) {
2248 				uid_t newid;
2249 
2250 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2251 				    ZFS_ACE_USER);
2252 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2253 				    uid == newid)
2254 					checkit = B_TRUE;
2255 				break;
2256 			} else {
2257 				mutex_exit(&zp->z_acl_lock);
2258 				return (EIO);
2259 			}
2260 		}
2261 
2262 		if (checkit) {
2263 			if (type == DENY) {
2264 				DTRACE_PROBE3(zfs__ace__denies,
2265 				    znode_t *, zp,
2266 				    zfs_ace_hdr_t *, acep,
2267 				    uint32_t, mask_matched);
2268 				deny_mask |= mask_matched;
2269 			} else {
2270 				DTRACE_PROBE3(zfs__ace__allows,
2271 				    znode_t *, zp,
2272 				    zfs_ace_hdr_t *, acep,
2273 				    uint32_t, mask_matched);
2274 				if (anyaccess) {
2275 					mutex_exit(&zp->z_acl_lock);
2276 					return (0);
2277 				}
2278 			}
2279 			*working_mode &= ~mask_matched;
2280 		}
2281 
2282 		/* Are we done? */
2283 		if (*working_mode == 0)
2284 			break;
2285 	}
2286 
2287 	mutex_exit(&zp->z_acl_lock);
2288 
2289 	/* Put the found 'denies' back on the working mode */
2290 	if (deny_mask) {
2291 		*working_mode |= deny_mask;
2292 		return (EACCES);
2293 	} else if (*working_mode) {
2294 		return (-1);
2295 	}
2296 
2297 	return (0);
2298 }
2299 
2300 /*
2301  * Return true if any access whatsoever granted, we don't actually
2302  * care what access is granted.
2303  */
2304 boolean_t
2305 zfs_has_access(znode_t *zp, cred_t *cr)
2306 {
2307 	uint32_t have = ACE_ALL_PERMS;
2308 
2309 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2310 		return (secpolicy_vnode_any_access(cr, ZTOV(zp),
2311 		    zp->z_uid) == 0);
2312 	}
2313 	return (B_TRUE);
2314 }
2315 
2316 static int
2317 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2318     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2319 {
2320 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2321 	int err;
2322 
2323 	*working_mode = v4_mode;
2324 	*check_privs = B_TRUE;
2325 
2326 	/*
2327 	 * Short circuit empty requests
2328 	 */
2329 	if (v4_mode == 0 || zfsvfs->z_replay) {
2330 		*working_mode = 0;
2331 		return (0);
2332 	}
2333 
2334 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2335 		*check_privs = B_FALSE;
2336 		return (err);
2337 	}
2338 
2339 	/*
2340 	 * The caller requested that the ACL check be skipped.  This
2341 	 * would only happen if the caller checked VOP_ACCESS() with a
2342 	 * 32 bit ACE mask and already had the appropriate permissions.
2343 	 */
2344 	if (skipaclchk) {
2345 		*working_mode = 0;
2346 		return (0);
2347 	}
2348 
2349 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2350 }
2351 
2352 static int
2353 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2354     cred_t *cr)
2355 {
2356 	if (*working_mode != ACE_WRITE_DATA)
2357 		return (EACCES);
2358 
2359 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2360 	    check_privs, B_FALSE, cr));
2361 }
2362 
2363 int
2364 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2365 {
2366 	boolean_t owner = B_FALSE;
2367 	boolean_t groupmbr = B_FALSE;
2368 	boolean_t is_attr;
2369 	uid_t uid = crgetuid(cr);
2370 	int error;
2371 
2372 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2373 		return (EACCES);
2374 
2375 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2376 	    (ZTOV(zdp)->v_type == VDIR));
2377 	if (is_attr)
2378 		goto slow;
2379 
2380 
2381 	mutex_enter(&zdp->z_acl_lock);
2382 
2383 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2384 		mutex_exit(&zdp->z_acl_lock);
2385 		return (0);
2386 	}
2387 
2388 	if (IS_EPHEMERAL(zdp->z_uid) != 0 || IS_EPHEMERAL(zdp->z_gid) != 0) {
2389 		mutex_exit(&zdp->z_acl_lock);
2390 		goto slow;
2391 	}
2392 
2393 	if (uid == zdp->z_uid) {
2394 		owner = B_TRUE;
2395 		if (zdp->z_mode & S_IXUSR) {
2396 			mutex_exit(&zdp->z_acl_lock);
2397 			return (0);
2398 		} else {
2399 			mutex_exit(&zdp->z_acl_lock);
2400 			goto slow;
2401 		}
2402 	}
2403 	if (groupmember(zdp->z_gid, cr)) {
2404 		groupmbr = B_TRUE;
2405 		if (zdp->z_mode & S_IXGRP) {
2406 			mutex_exit(&zdp->z_acl_lock);
2407 			return (0);
2408 		} else {
2409 			mutex_exit(&zdp->z_acl_lock);
2410 			goto slow;
2411 		}
2412 	}
2413 	if (!owner && !groupmbr) {
2414 		if (zdp->z_mode & S_IXOTH) {
2415 			mutex_exit(&zdp->z_acl_lock);
2416 			return (0);
2417 		}
2418 	}
2419 
2420 	mutex_exit(&zdp->z_acl_lock);
2421 
2422 slow:
2423 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2424 	ZFS_ENTER(zdp->z_zfsvfs);
2425 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2426 	ZFS_EXIT(zdp->z_zfsvfs);
2427 	return (error);
2428 }
2429 
2430 /*
2431  * Determine whether Access should be granted/denied.
2432  * The least priv subsytem is always consulted as a basic privilege
2433  * can define any form of access.
2434  */
2435 int
2436 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2437 {
2438 	uint32_t	working_mode;
2439 	int		error;
2440 	int		is_attr;
2441 	boolean_t 	check_privs;
2442 	znode_t		*xzp;
2443 	znode_t 	*check_zp = zp;
2444 	mode_t		needed_bits;
2445 
2446 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2447 
2448 	/*
2449 	 * If attribute then validate against base file
2450 	 */
2451 	if (is_attr) {
2452 		uint64_t	parent;
2453 
2454 		if ((error = sa_lookup(zp->z_sa_hdl,
2455 		    SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2456 		    sizeof (parent))) != 0)
2457 			return (error);
2458 
2459 		if ((error = zfs_zget(zp->z_zfsvfs,
2460 		    parent, &xzp)) != 0)	{
2461 			return (error);
2462 		}
2463 
2464 		check_zp = xzp;
2465 
2466 		/*
2467 		 * fixup mode to map to xattr perms
2468 		 */
2469 
2470 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2471 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2472 			mode |= ACE_WRITE_NAMED_ATTRS;
2473 		}
2474 
2475 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2476 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2477 			mode |= ACE_READ_NAMED_ATTRS;
2478 		}
2479 	}
2480 
2481 	/*
2482 	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2483 	 * in needed_bits.  Map the bits mapped by working_mode (currently
2484 	 * missing) in missing_bits.
2485 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2486 	 * needed_bits.
2487 	 */
2488 	needed_bits = 0;
2489 
2490 	working_mode = mode;
2491 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2492 	    zp->z_uid == crgetuid(cr))
2493 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2494 
2495 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2496 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2497 		needed_bits |= VREAD;
2498 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2499 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2500 		needed_bits |= VWRITE;
2501 	if (working_mode & ACE_EXECUTE)
2502 		needed_bits |= VEXEC;
2503 
2504 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2505 	    &check_privs, skipaclchk, cr)) == 0) {
2506 		if (is_attr)
2507 			VN_RELE(ZTOV(xzp));
2508 		return (secpolicy_vnode_access2(cr, ZTOV(zp), zp->z_uid,
2509 		    needed_bits, needed_bits));
2510 	}
2511 
2512 	if (error && !check_privs) {
2513 		if (is_attr)
2514 			VN_RELE(ZTOV(xzp));
2515 		return (error);
2516 	}
2517 
2518 	if (error && (flags & V_APPEND)) {
2519 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2520 	}
2521 
2522 	if (error && check_privs) {
2523 		mode_t		checkmode = 0;
2524 
2525 		/*
2526 		 * First check for implicit owner permission on
2527 		 * read_acl/read_attributes
2528 		 */
2529 
2530 		error = 0;
2531 		ASSERT(working_mode != 0);
2532 
2533 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2534 		    zp->z_uid == crgetuid(cr)))
2535 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2536 
2537 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2538 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2539 			checkmode |= VREAD;
2540 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2541 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2542 			checkmode |= VWRITE;
2543 		if (working_mode & ACE_EXECUTE)
2544 			checkmode |= VEXEC;
2545 
2546 		error = secpolicy_vnode_access2(cr, ZTOV(check_zp), zp->z_uid,
2547 		    needed_bits & ~checkmode, needed_bits);
2548 
2549 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2550 			error = secpolicy_vnode_chown(cr, zp->z_uid);
2551 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2552 			error = secpolicy_vnode_setdac(cr, zp->z_uid);
2553 
2554 		if (error == 0 && (working_mode &
2555 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2556 			error = secpolicy_vnode_remove(cr);
2557 
2558 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2559 			error = secpolicy_vnode_chown(cr, zp->z_uid);
2560 		}
2561 		if (error == 0) {
2562 			/*
2563 			 * See if any bits other than those already checked
2564 			 * for are still present.  If so then return EACCES
2565 			 */
2566 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2567 				error = EACCES;
2568 			}
2569 		}
2570 	} else if (error == 0) {
2571 		error = secpolicy_vnode_access2(cr, ZTOV(zp), zp->z_uid,
2572 		    needed_bits, needed_bits);
2573 	}
2574 
2575 
2576 	if (is_attr)
2577 		VN_RELE(ZTOV(xzp));
2578 
2579 	return (error);
2580 }
2581 
2582 /*
2583  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2584  * native ACL format and call zfs_zaccess()
2585  */
2586 int
2587 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2588 {
2589 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2590 }
2591 
2592 /*
2593  * Access function for secpolicy_vnode_setattr
2594  */
2595 int
2596 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2597 {
2598 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2599 
2600 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2601 }
2602 
2603 static int
2604 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2605     mode_t available_perms, cred_t *cr)
2606 {
2607 	int error;
2608 
2609 	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2610 	    dzp->z_uid, available_perms, VWRITE|VEXEC);
2611 
2612 	if (error == 0)
2613 		error = zfs_sticky_remove_access(dzp, zp, cr);
2614 
2615 	return (error);
2616 }
2617 
2618 /*
2619  * Determine whether Access should be granted/deny, without
2620  * consulting least priv subsystem.
2621  *
2622  *
2623  * The following chart is the recommended NFSv4 enforcement for
2624  * ability to delete an object.
2625  *
2626  *      -------------------------------------------------------
2627  *      |   Parent Dir  |           Target Object Permissions |
2628  *      |  permissions  |                                     |
2629  *      -------------------------------------------------------
2630  *      |               | ACL Allows | ACL Denies| Delete     |
2631  *      |               |  Delete    |  Delete   | unspecified|
2632  *      -------------------------------------------------------
2633  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2634  *      |  DELETE_CHILD |                                     |
2635  *      -------------------------------------------------------
2636  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2637  *      |  DELETE_CHILD |            |           |            |
2638  *      -------------------------------------------------------
2639  *      | ACL specifies |            |           |            |
2640  *      | only allow    | Permit     | Permit    | Permit     |
2641  *      | write and     |            |           |            |
2642  *      | execute       |            |           |            |
2643  *      -------------------------------------------------------
2644  *      | ACL denies    |            |           |            |
2645  *      | write and     | Permit     | Deny      | Deny       |
2646  *      | execute       |            |           |            |
2647  *      -------------------------------------------------------
2648  *         ^
2649  *         |
2650  *         No search privilege, can't even look up file?
2651  *
2652  */
2653 int
2654 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2655 {
2656 	uint32_t dzp_working_mode = 0;
2657 	uint32_t zp_working_mode = 0;
2658 	int dzp_error, zp_error;
2659 	mode_t available_perms;
2660 	boolean_t dzpcheck_privs = B_TRUE;
2661 	boolean_t zpcheck_privs = B_TRUE;
2662 
2663 	/*
2664 	 * We want specific DELETE permissions to
2665 	 * take precedence over WRITE/EXECUTE.  We don't
2666 	 * want an ACL such as this to mess us up.
2667 	 * user:joe:write_data:deny,user:joe:delete:allow
2668 	 *
2669 	 * However, deny permissions may ultimately be overridden
2670 	 * by secpolicy_vnode_access().
2671 	 *
2672 	 * We will ask for all of the necessary permissions and then
2673 	 * look at the working modes from the directory and target object
2674 	 * to determine what was found.
2675 	 */
2676 
2677 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2678 		return (EPERM);
2679 
2680 	/*
2681 	 * First row
2682 	 * If the directory permissions allow the delete, we are done.
2683 	 */
2684 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2685 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2686 		return (0);
2687 
2688 	/*
2689 	 * If target object has delete permission then we are done
2690 	 */
2691 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2692 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2693 		return (0);
2694 
2695 	ASSERT(dzp_error && zp_error);
2696 
2697 	if (!dzpcheck_privs)
2698 		return (dzp_error);
2699 	if (!zpcheck_privs)
2700 		return (zp_error);
2701 
2702 	/*
2703 	 * Second row
2704 	 *
2705 	 * If directory returns EACCES then delete_child was denied
2706 	 * due to deny delete_child.  In this case send the request through
2707 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2708 	 * since that *could* allow the delete based on write/execute permission
2709 	 * and we want delete permissions to override write/execute.
2710 	 */
2711 
2712 	if (dzp_error == EACCES)
2713 		return (secpolicy_vnode_remove(cr));
2714 
2715 	/*
2716 	 * Third Row
2717 	 * only need to see if we have write/execute on directory.
2718 	 */
2719 
2720 	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2721 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2722 
2723 	if (dzp_error != 0 && !dzpcheck_privs)
2724 		return (dzp_error);
2725 
2726 	/*
2727 	 * Fourth row
2728 	 */
2729 
2730 	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2731 	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2732 
2733 	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2734 
2735 }
2736 
2737 int
2738 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2739     znode_t *tzp, cred_t *cr)
2740 {
2741 	int add_perm;
2742 	int error;
2743 
2744 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2745 		return (EACCES);
2746 
2747 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2748 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2749 
2750 	/*
2751 	 * Rename permissions are combination of delete permission +
2752 	 * add file/subdir permission.
2753 	 */
2754 
2755 	/*
2756 	 * first make sure we do the delete portion.
2757 	 *
2758 	 * If that succeeds then check for add_file/add_subdir permissions
2759 	 */
2760 
2761 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2762 		return (error);
2763 
2764 	/*
2765 	 * If we have a tzp, see if we can delete it?
2766 	 */
2767 	if (tzp) {
2768 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2769 			return (error);
2770 	}
2771 
2772 	/*
2773 	 * Now check for add permissions
2774 	 */
2775 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2776 
2777 	return (error);
2778 }
2779