xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_acl.c (revision 842eec28b3c46588c5c38236a54676d763329c1d)
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 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/time.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/resource.h>
32 #include <sys/vfs.h>
33 #include <sys/vnode.h>
34 #include <sys/sid.h>
35 #include <sys/file.h>
36 #include <sys/stat.h>
37 #include <sys/kmem.h>
38 #include <sys/cmn_err.h>
39 #include <sys/errno.h>
40 #include <sys/unistd.h>
41 #include <sys/sdt.h>
42 #include <sys/fs/zfs.h>
43 #include <sys/mode.h>
44 #include <sys/policy.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_fuid.h>
47 #include <sys/zfs_acl.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_vfsops.h>
50 #include <sys/dmu.h>
51 #include <sys/dnode.h>
52 #include <sys/zap.h>
53 #include "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 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 == OWNING_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_acl_valid_ace_type(uint_t type, uint_t flags)
400 {
401 	uint16_t entry_type;
402 
403 	switch (type) {
404 	case ALLOW:
405 	case DENY:
406 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
407 	case ACE_SYSTEM_ALARM_ACE_TYPE:
408 		entry_type = flags & ACE_TYPE_FLAGS;
409 		return (entry_type == ACE_OWNER ||
410 		    entry_type == OWNING_GROUP ||
411 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
412 		    entry_type == ACE_IDENTIFIER_GROUP);
413 	default:
414 		if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
415 			return (B_TRUE);
416 	}
417 	return (B_FALSE);
418 }
419 
420 static boolean_t
421 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
422 {
423 	/*
424 	 * first check type of entry
425 	 */
426 
427 	if (!zfs_acl_valid_ace_type(type, iflags))
428 		return (B_FALSE);
429 
430 	switch (type) {
431 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
432 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
433 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
434 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
435 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
436 			return (B_FALSE);
437 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
438 	}
439 
440 	/*
441 	 * next check inheritance level flags
442 	 */
443 
444 	if (obj_type == VDIR &&
445 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
446 		aclp->z_hints |= ZFS_INHERIT_ACE;
447 
448 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
449 		if ((iflags & (ACE_FILE_INHERIT_ACE|
450 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
451 			return (B_FALSE);
452 		}
453 	}
454 
455 	return (B_TRUE);
456 }
457 
458 static void *
459 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
460     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
461 {
462 	zfs_acl_node_t *aclnode;
463 
464 	if (start == NULL) {
465 		aclnode = list_head(&aclp->z_acl);
466 		if (aclnode == NULL)
467 			return (NULL);
468 
469 		aclp->z_next_ace = aclnode->z_acldata;
470 		aclp->z_curr_node = aclnode;
471 		aclnode->z_ace_idx = 0;
472 	}
473 
474 	aclnode = aclp->z_curr_node;
475 
476 	if (aclnode == NULL)
477 		return (NULL);
478 
479 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
480 		aclnode = list_next(&aclp->z_acl, aclnode);
481 		if (aclnode == NULL)
482 			return (NULL);
483 		else {
484 			aclp->z_curr_node = aclnode;
485 			aclnode->z_ace_idx = 0;
486 			aclp->z_next_ace = aclnode->z_acldata;
487 		}
488 	}
489 
490 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
491 		void *acep = aclp->z_next_ace;
492 		size_t ace_size;
493 
494 		/*
495 		 * Make sure we don't overstep our bounds
496 		 */
497 		ace_size = aclp->z_ops.ace_size(acep);
498 
499 		if (((caddr_t)acep + ace_size) >
500 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
501 			return (NULL);
502 		}
503 
504 		*iflags = aclp->z_ops.ace_flags_get(acep);
505 		*type = aclp->z_ops.ace_type_get(acep);
506 		*access_mask = aclp->z_ops.ace_mask_get(acep);
507 		*who = aclp->z_ops.ace_who_get(acep);
508 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
509 		aclnode->z_ace_idx++;
510 		return ((void *)acep);
511 	}
512 	return (NULL);
513 }
514 
515 /*ARGSUSED*/
516 static uint64_t
517 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
518     uint16_t *flags, uint16_t *type, uint32_t *mask)
519 {
520 	zfs_acl_t *aclp = datap;
521 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
522 	uint64_t who;
523 
524 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
525 	    flags, type);
526 	return ((uint64_t)(uintptr_t)acep);
527 }
528 
529 static zfs_acl_node_t *
530 zfs_acl_curr_node(zfs_acl_t *aclp)
531 {
532 	ASSERT(aclp->z_curr_node);
533 	return (aclp->z_curr_node);
534 }
535 
536 /*
537  * Copy ACE to internal ZFS format.
538  * While processing the ACL each ACE will be validated for correctness.
539  * ACE FUIDs will be created later.
540  */
541 int
542 zfs_copy_ace_2_fuid(vtype_t obj_type, zfs_acl_t *aclp, void *datap,
543     zfs_ace_t *z_acl, int aclcnt, size_t *size)
544 {
545 	int i;
546 	uint16_t entry_type;
547 	zfs_ace_t *aceptr = z_acl;
548 	ace_t *acep = datap;
549 	zfs_object_ace_t *zobjacep;
550 	ace_object_t *aceobjp;
551 
552 	for (i = 0; i != aclcnt; i++) {
553 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
554 		aceptr->z_hdr.z_flags = acep->a_flags;
555 		aceptr->z_hdr.z_type = acep->a_type;
556 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
557 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
558 		    entry_type != ACE_EVERYONE) {
559 			if (!aclp->z_has_fuids)
560 				aclp->z_has_fuids = IS_EPHEMERAL(acep->a_who);
561 			aceptr->z_fuid = (uint64_t)acep->a_who;
562 		}
563 
564 		/*
565 		 * Make sure ACE is valid
566 		 */
567 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
568 		    aceptr->z_hdr.z_flags) != B_TRUE)
569 			return (EINVAL);
570 
571 		switch (acep->a_type) {
572 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
573 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
574 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
575 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
576 			zobjacep = (zfs_object_ace_t *)aceptr;
577 			aceobjp = (ace_object_t *)acep;
578 
579 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
580 			    sizeof (aceobjp->a_obj_type));
581 			bcopy(aceobjp->a_inherit_obj_type,
582 			    zobjacep->z_inherit_type,
583 			    sizeof (aceobjp->a_inherit_obj_type));
584 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
585 			break;
586 		default:
587 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
588 		}
589 
590 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
591 		    aclp->z_ops.ace_size(aceptr));
592 	}
593 
594 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
595 
596 	return (0);
597 }
598 
599 /*
600  * Copy ZFS ACEs to fixed size ace_t layout
601  */
602 static void
603 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
604     void *datap, int filter)
605 {
606 	uint64_t who;
607 	uint32_t access_mask;
608 	uint16_t iflags, type;
609 	zfs_ace_hdr_t *zacep = NULL;
610 	ace_t *acep = datap;
611 	ace_object_t *objacep;
612 	zfs_object_ace_t *zobjacep;
613 	size_t ace_size;
614 	uint16_t entry_type;
615 
616 	while (zacep = zfs_acl_next_ace(aclp, zacep,
617 	    &who, &access_mask, &iflags, &type)) {
618 
619 		switch (type) {
620 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
621 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
622 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
623 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
624 			if (filter) {
625 				continue;
626 			}
627 			zobjacep = (zfs_object_ace_t *)zacep;
628 			objacep = (ace_object_t *)acep;
629 			bcopy(zobjacep->z_object_type,
630 			    objacep->a_obj_type,
631 			    sizeof (zobjacep->z_object_type));
632 			bcopy(zobjacep->z_inherit_type,
633 			    objacep->a_inherit_obj_type,
634 			    sizeof (zobjacep->z_inherit_type));
635 			ace_size = sizeof (ace_object_t);
636 			break;
637 		default:
638 			ace_size = sizeof (ace_t);
639 			break;
640 		}
641 
642 		entry_type = (iflags & ACE_TYPE_FLAGS);
643 		if ((entry_type != ACE_OWNER &&
644 		    entry_type != OWNING_GROUP &&
645 		    entry_type != ACE_EVERYONE)) {
646 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
647 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
648 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
649 		} else {
650 			acep->a_who = (uid_t)(int64_t)who;
651 		}
652 		acep->a_access_mask = access_mask;
653 		acep->a_flags = iflags;
654 		acep->a_type = type;
655 		acep = (ace_t *)((caddr_t)acep + ace_size);
656 	}
657 }
658 
659 static int
660 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
661     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
662 {
663 	int i;
664 	zfs_oldace_t *aceptr = z_acl;
665 
666 	for (i = 0; i != aclcnt; i++, aceptr++) {
667 		aceptr->z_access_mask = acep[i].a_access_mask;
668 		aceptr->z_type = acep[i].a_type;
669 		aceptr->z_flags = acep[i].a_flags;
670 		aceptr->z_fuid = acep[i].a_who;
671 		/*
672 		 * Make sure ACE is valid
673 		 */
674 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
675 		    aceptr->z_flags) != B_TRUE)
676 			return (EINVAL);
677 	}
678 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
679 	return (0);
680 }
681 
682 /*
683  * convert old ACL format to new
684  */
685 void
686 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp)
687 {
688 	zfs_oldace_t *oldaclp;
689 	int i;
690 	uint16_t type, iflags;
691 	uint32_t access_mask;
692 	uint64_t who;
693 	void *cookie = NULL;
694 	zfs_acl_node_t *newaclnode;
695 
696 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
697 	/*
698 	 * First create the ACE in a contiguous piece of memory
699 	 * for zfs_copy_ace_2_fuid().
700 	 *
701 	 * We only convert an ACL once, so this won't happen
702 	 * everytime.
703 	 */
704 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
705 	    KM_SLEEP);
706 	i = 0;
707 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
708 	    &access_mask, &iflags, &type)) {
709 		oldaclp[i].z_flags = iflags;
710 		oldaclp[i].z_type = type;
711 		oldaclp[i].z_fuid = who;
712 		oldaclp[i++].z_access_mask = access_mask;
713 	}
714 
715 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
716 	    sizeof (zfs_object_ace_t));
717 	aclp->z_ops = zfs_acl_fuid_ops;
718 	VERIFY(zfs_copy_ace_2_fuid(ZTOV(zp)->v_type, aclp, oldaclp,
719 	    newaclnode->z_acldata, aclp->z_acl_count,
720 	    &newaclnode->z_size) == 0);
721 	newaclnode->z_ace_count = aclp->z_acl_count;
722 	aclp->z_version = ZFS_ACL_VERSION;
723 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
724 
725 	/*
726 	 * Release all previous ACL nodes
727 	 */
728 
729 	zfs_acl_release_nodes(aclp);
730 
731 	list_insert_head(&aclp->z_acl, newaclnode);
732 
733 	aclp->z_acl_bytes = newaclnode->z_size;
734 	aclp->z_acl_count = newaclnode->z_ace_count;
735 
736 }
737 
738 /*
739  * Convert unix access mask to v4 access mask
740  */
741 static uint32_t
742 zfs_unix_to_v4(uint32_t access_mask)
743 {
744 	uint32_t new_mask = 0;
745 
746 	if (access_mask & S_IXOTH)
747 		new_mask |= ACE_EXECUTE;
748 	if (access_mask & S_IWOTH)
749 		new_mask |= ACE_WRITE_DATA;
750 	if (access_mask & S_IROTH)
751 		new_mask |= ACE_READ_DATA;
752 	return (new_mask);
753 }
754 
755 static void
756 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
757     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
758 {
759 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
760 
761 	aclp->z_ops.ace_mask_set(acep, access_mask);
762 	aclp->z_ops.ace_type_set(acep, access_type);
763 	aclp->z_ops.ace_flags_set(acep, entry_type);
764 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
765 	    type != ACE_EVERYONE))
766 		aclp->z_ops.ace_who_set(acep, fuid);
767 }
768 
769 /*
770  * Determine mode of file based on ACL.
771  * Also, create FUIDs for any User/Group ACEs
772  */
773 static uint64_t
774 zfs_mode_fuid_compute(znode_t *zp, zfs_acl_t *aclp, cred_t *cr,
775     zfs_fuid_info_t **fuidp, dmu_tx_t *tx)
776 {
777 	int		entry_type;
778 	mode_t		mode;
779 	mode_t		seen = 0;
780 	zfs_ace_hdr_t 	*acep = NULL;
781 	uint64_t	who;
782 	uint16_t	iflags, type;
783 	uint32_t	access_mask;
784 
785 	mode = (zp->z_phys->zp_mode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
786 
787 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
788 	    &access_mask, &iflags, &type)) {
789 
790 		if (!zfs_acl_valid_ace_type(type, iflags))
791 			continue;
792 
793 		entry_type = (iflags & ACE_TYPE_FLAGS);
794 
795 		/*
796 		 * Skip over owner@, group@ or everyone@ inherit only ACEs
797 		 */
798 		if ((iflags & ACE_INHERIT_ONLY_ACE) &&
799 		    (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
800 		    entry_type == OWNING_GROUP))
801 			continue;
802 
803 		if (entry_type == ACE_OWNER) {
804 			if ((access_mask & ACE_READ_DATA) &&
805 			    (!(seen & S_IRUSR))) {
806 				seen |= S_IRUSR;
807 				if (type == ALLOW) {
808 					mode |= S_IRUSR;
809 				}
810 			}
811 			if ((access_mask & ACE_WRITE_DATA) &&
812 			    (!(seen & S_IWUSR))) {
813 				seen |= S_IWUSR;
814 				if (type == ALLOW) {
815 					mode |= S_IWUSR;
816 				}
817 			}
818 			if ((access_mask & ACE_EXECUTE) &&
819 			    (!(seen & S_IXUSR))) {
820 				seen |= S_IXUSR;
821 				if (type == ALLOW) {
822 					mode |= S_IXUSR;
823 				}
824 			}
825 		} else if (entry_type == OWNING_GROUP) {
826 			if ((access_mask & ACE_READ_DATA) &&
827 			    (!(seen & S_IRGRP))) {
828 				seen |= S_IRGRP;
829 				if (type == ALLOW) {
830 					mode |= S_IRGRP;
831 				}
832 			}
833 			if ((access_mask & ACE_WRITE_DATA) &&
834 			    (!(seen & S_IWGRP))) {
835 				seen |= S_IWGRP;
836 				if (type == ALLOW) {
837 					mode |= S_IWGRP;
838 				}
839 			}
840 			if ((access_mask & ACE_EXECUTE) &&
841 			    (!(seen & S_IXGRP))) {
842 				seen |= S_IXGRP;
843 				if (type == ALLOW) {
844 					mode |= S_IXGRP;
845 				}
846 			}
847 		} else if (entry_type == ACE_EVERYONE) {
848 			if ((access_mask & ACE_READ_DATA)) {
849 				if (!(seen & S_IRUSR)) {
850 					seen |= S_IRUSR;
851 					if (type == ALLOW) {
852 						mode |= S_IRUSR;
853 					}
854 				}
855 				if (!(seen & S_IRGRP)) {
856 					seen |= S_IRGRP;
857 					if (type == ALLOW) {
858 						mode |= S_IRGRP;
859 					}
860 				}
861 				if (!(seen & S_IROTH)) {
862 					seen |= S_IROTH;
863 					if (type == ALLOW) {
864 						mode |= S_IROTH;
865 					}
866 				}
867 			}
868 			if ((access_mask & ACE_WRITE_DATA)) {
869 				if (!(seen & S_IWUSR)) {
870 					seen |= S_IWUSR;
871 					if (type == ALLOW) {
872 						mode |= S_IWUSR;
873 					}
874 				}
875 				if (!(seen & S_IWGRP)) {
876 					seen |= S_IWGRP;
877 					if (type == ALLOW) {
878 						mode |= S_IWGRP;
879 					}
880 				}
881 				if (!(seen & S_IWOTH)) {
882 					seen |= S_IWOTH;
883 					if (type == ALLOW) {
884 						mode |= S_IWOTH;
885 					}
886 				}
887 			}
888 			if ((access_mask & ACE_EXECUTE)) {
889 				if (!(seen & S_IXUSR)) {
890 					seen |= S_IXUSR;
891 					if (type == ALLOW) {
892 						mode |= S_IXUSR;
893 					}
894 				}
895 				if (!(seen & S_IXGRP)) {
896 					seen |= S_IXGRP;
897 					if (type == ALLOW) {
898 						mode |= S_IXGRP;
899 					}
900 				}
901 				if (!(seen & S_IXOTH)) {
902 					seen |= S_IXOTH;
903 					if (type == ALLOW) {
904 						mode |= S_IXOTH;
905 					}
906 				}
907 			}
908 		}
909 		/*
910 		 * Now handle FUID create for user/group ACEs
911 		 */
912 		if (entry_type == 0 || entry_type == ACE_IDENTIFIER_GROUP) {
913 			aclp->z_ops.ace_who_set(acep,
914 			    zfs_fuid_create(zp->z_zfsvfs, who, cr,
915 			    (entry_type == 0) ? ZFS_ACE_USER : ZFS_ACE_GROUP,
916 			    tx, fuidp));
917 		}
918 	}
919 	return (mode);
920 }
921 
922 static zfs_acl_t *
923 zfs_acl_node_read_internal(znode_t *zp, boolean_t will_modify)
924 {
925 	zfs_acl_t	*aclp;
926 	zfs_acl_node_t	*aclnode;
927 
928 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
929 
930 	/*
931 	 * Version 0 to 1 znode_acl_phys has the size/count fields swapped.
932 	 * Version 0 didn't have a size field, only a count.
933 	 */
934 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
935 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_size;
936 		aclp->z_acl_bytes = ZFS_ACL_SIZE(aclp->z_acl_count);
937 	} else {
938 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_count;
939 		aclp->z_acl_bytes = zp->z_phys->zp_acl.z_acl_size;
940 	}
941 
942 	aclnode = zfs_acl_node_alloc(will_modify ? aclp->z_acl_bytes : 0);
943 	aclnode->z_ace_count = aclp->z_acl_count;
944 	if (will_modify) {
945 		bcopy(zp->z_phys->zp_acl.z_ace_data, aclnode->z_acldata,
946 		    aclp->z_acl_bytes);
947 	} else {
948 		aclnode->z_size = aclp->z_acl_bytes;
949 		aclnode->z_acldata = &zp->z_phys->zp_acl.z_ace_data[0];
950 	}
951 
952 	list_insert_head(&aclp->z_acl, aclnode);
953 
954 	return (aclp);
955 }
956 
957 /*
958  * Read an external acl object.
959  */
960 static int
961 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
962 {
963 	uint64_t extacl = zp->z_phys->zp_acl.z_acl_extern_obj;
964 	zfs_acl_t	*aclp;
965 	size_t		aclsize;
966 	size_t		acl_count;
967 	zfs_acl_node_t	*aclnode;
968 	int error;
969 
970 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
971 
972 	if (zp->z_phys->zp_acl.z_acl_extern_obj == 0) {
973 		*aclpp = zfs_acl_node_read_internal(zp, will_modify);
974 		return (0);
975 	}
976 
977 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
978 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
979 		zfs_acl_phys_v0_t *zacl0 =
980 		    (zfs_acl_phys_v0_t *)&zp->z_phys->zp_acl;
981 
982 		aclsize = ZFS_ACL_SIZE(zacl0->z_acl_count);
983 		acl_count = zacl0->z_acl_count;
984 	} else {
985 		aclsize = zp->z_phys->zp_acl.z_acl_size;
986 		acl_count = zp->z_phys->zp_acl.z_acl_count;
987 		if (aclsize == 0)
988 			aclsize = acl_count * sizeof (zfs_ace_t);
989 	}
990 	aclnode = zfs_acl_node_alloc(aclsize);
991 	list_insert_head(&aclp->z_acl, aclnode);
992 	error = dmu_read(zp->z_zfsvfs->z_os, extacl, 0,
993 	    aclsize, aclnode->z_acldata);
994 	aclnode->z_ace_count = acl_count;
995 	aclp->z_acl_count = acl_count;
996 	aclp->z_acl_bytes = aclsize;
997 
998 	if (error != 0) {
999 		zfs_acl_free(aclp);
1000 		/* convert checksum errors into IO errors */
1001 		if (error == ECKSUM)
1002 			error = EIO;
1003 		return (error);
1004 	}
1005 
1006 	*aclpp = aclp;
1007 	return (0);
1008 }
1009 
1010 /*
1011  * common code for setting ACLs.
1012  *
1013  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1014  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1015  * already checked the acl and knows whether to inherit.
1016  */
1017 int
1018 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr,
1019     zfs_fuid_info_t **fuidp, dmu_tx_t *tx)
1020 {
1021 	int		error;
1022 	znode_phys_t	*zphys = zp->z_phys;
1023 	zfs_acl_phys_t	*zacl = &zphys->zp_acl;
1024 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1025 	uint64_t	aoid = zphys->zp_acl.z_acl_extern_obj;
1026 	uint64_t	off = 0;
1027 	dmu_object_type_t otype;
1028 	zfs_acl_node_t	*aclnode;
1029 
1030 	ASSERT(MUTEX_HELD(&zp->z_lock));
1031 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1032 
1033 	dmu_buf_will_dirty(zp->z_dbuf, tx);
1034 
1035 	zphys->zp_mode = zfs_mode_fuid_compute(zp, aclp, cr, fuidp, tx);
1036 
1037 	/*
1038 	 * Decide which opbject type to use.  If we are forced to
1039 	 * use old ACL format than transform ACL into zfs_oldace_t
1040 	 * layout.
1041 	 */
1042 	if (!zfsvfs->z_use_fuids) {
1043 		otype = DMU_OT_OLDACL;
1044 	} else {
1045 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1046 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1047 			zfs_acl_xform(zp, aclp);
1048 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1049 		otype = DMU_OT_ACL;
1050 	}
1051 
1052 	if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1053 		/*
1054 		 * If ACL was previously external and we are now
1055 		 * converting to new ACL format then release old
1056 		 * ACL object and create a new one.
1057 		 */
1058 		if (aoid && aclp->z_version != zacl->z_acl_version) {
1059 			error = dmu_object_free(zfsvfs->z_os,
1060 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1061 			if (error)
1062 				return (error);
1063 			aoid = 0;
1064 		}
1065 		if (aoid == 0) {
1066 			aoid = dmu_object_alloc(zfsvfs->z_os,
1067 			    otype, aclp->z_acl_bytes,
1068 			    otype == DMU_OT_ACL ? DMU_OT_SYSACL : DMU_OT_NONE,
1069 			    otype == DMU_OT_ACL ? DN_MAX_BONUSLEN : 0, tx);
1070 		} else {
1071 			(void) dmu_object_set_blocksize(zfsvfs->z_os, aoid,
1072 			    aclp->z_acl_bytes, 0, tx);
1073 		}
1074 		zphys->zp_acl.z_acl_extern_obj = aoid;
1075 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1076 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1077 			if (aclnode->z_ace_count == 0)
1078 				continue;
1079 			dmu_write(zfsvfs->z_os, aoid, off,
1080 			    aclnode->z_size, aclnode->z_acldata, tx);
1081 			off += aclnode->z_size;
1082 		}
1083 	} else {
1084 		void *start = zacl->z_ace_data;
1085 		/*
1086 		 * Migrating back embedded?
1087 		 */
1088 		if (zphys->zp_acl.z_acl_extern_obj) {
1089 			error = dmu_object_free(zfsvfs->z_os,
1090 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1091 			if (error)
1092 				return (error);
1093 			zphys->zp_acl.z_acl_extern_obj = 0;
1094 		}
1095 
1096 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1097 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1098 			if (aclnode->z_ace_count == 0)
1099 				continue;
1100 			bcopy(aclnode->z_acldata, start, aclnode->z_size);
1101 			start = (caddr_t)start + aclnode->z_size;
1102 		}
1103 	}
1104 
1105 	/*
1106 	 * If Old version then swap count/bytes to match old
1107 	 * layout of znode_acl_phys_t.
1108 	 */
1109 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1110 		zphys->zp_acl.z_acl_size = aclp->z_acl_count;
1111 		zphys->zp_acl.z_acl_count = aclp->z_acl_bytes;
1112 	} else {
1113 		zphys->zp_acl.z_acl_size = aclp->z_acl_bytes;
1114 		zphys->zp_acl.z_acl_count = aclp->z_acl_count;
1115 	}
1116 
1117 	zphys->zp_acl.z_acl_version = aclp->z_version;
1118 
1119 	/*
1120 	 * Replace ACL wide bits, but first clear them.
1121 	 */
1122 	zp->z_phys->zp_flags &= ~ZFS_ACL_WIDE_FLAGS;
1123 
1124 	zp->z_phys->zp_flags |= aclp->z_hints;
1125 
1126 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1127 		zp->z_phys->zp_flags |= ZFS_ACL_TRIVIAL;
1128 
1129 	zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
1130 	return (0);
1131 }
1132 
1133 /*
1134  * Update access mask for prepended ACE
1135  *
1136  * This applies the "groupmask" value for aclmode property.
1137  */
1138 static void
1139 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1140     mode_t mode, uint64_t owner)
1141 {
1142 	int	rmask, wmask, xmask;
1143 	int	user_ace;
1144 	uint16_t aceflags;
1145 	uint32_t origmask, acepmask;
1146 	uint64_t fuid;
1147 
1148 	aceflags = aclp->z_ops.ace_flags_get(acep);
1149 	fuid = aclp->z_ops.ace_who_get(acep);
1150 	origmask = aclp->z_ops.ace_mask_get(origacep);
1151 	acepmask = aclp->z_ops.ace_mask_get(acep);
1152 
1153 	user_ace = (!(aceflags &
1154 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1155 
1156 	if (user_ace && (fuid == owner)) {
1157 		rmask = S_IRUSR;
1158 		wmask = S_IWUSR;
1159 		xmask = S_IXUSR;
1160 	} else {
1161 		rmask = S_IRGRP;
1162 		wmask = S_IWGRP;
1163 		xmask = S_IXGRP;
1164 	}
1165 
1166 	if (origmask & ACE_READ_DATA) {
1167 		if (mode & rmask) {
1168 			acepmask &= ~ACE_READ_DATA;
1169 		} else {
1170 			acepmask |= ACE_READ_DATA;
1171 		}
1172 	}
1173 
1174 	if (origmask & ACE_WRITE_DATA) {
1175 		if (mode & wmask) {
1176 			acepmask &= ~ACE_WRITE_DATA;
1177 		} else {
1178 			acepmask |= ACE_WRITE_DATA;
1179 		}
1180 	}
1181 
1182 	if (origmask & ACE_APPEND_DATA) {
1183 		if (mode & wmask) {
1184 			acepmask &= ~ACE_APPEND_DATA;
1185 		} else {
1186 			acepmask |= ACE_APPEND_DATA;
1187 		}
1188 	}
1189 
1190 	if (origmask & ACE_EXECUTE) {
1191 		if (mode & xmask) {
1192 			acepmask &= ~ACE_EXECUTE;
1193 		} else {
1194 			acepmask |= ACE_EXECUTE;
1195 		}
1196 	}
1197 	aclp->z_ops.ace_mask_set(acep, acepmask);
1198 }
1199 
1200 /*
1201  * Apply mode to canonical six ACEs.
1202  */
1203 static void
1204 zfs_acl_fixup_canonical_six(zfs_acl_t *aclp, mode_t mode)
1205 {
1206 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1207 	void	*acep;
1208 	int	maskoff = aclp->z_ops.ace_mask_off();
1209 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1210 
1211 	ASSERT(aclnode != NULL);
1212 
1213 	acep = (void *)((caddr_t)aclnode->z_acldata +
1214 	    aclnode->z_size - (abstract_size * 6));
1215 
1216 	/*
1217 	 * Fixup final ACEs to match the mode
1218 	 */
1219 
1220 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1221 	    (mode & 0700) >> 6);	/* owner@ */
1222 
1223 	acep = (caddr_t)acep + (abstract_size * 2);
1224 
1225 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1226 	    (mode & 0070) >> 3);	/* group@ */
1227 
1228 	acep = (caddr_t)acep + (abstract_size * 2);
1229 	adjust_ace_pair_common(acep, maskoff,
1230 	    abstract_size, mode);	/* everyone@ */
1231 }
1232 
1233 
1234 static int
1235 zfs_acl_ace_match(zfs_acl_t *aclp, void *acep, int allow_deny,
1236     int entry_type, int accessmask)
1237 {
1238 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1239 	uint16_t type = aclp->z_ops.ace_type_get(acep);
1240 	uint16_t flags = aclp->z_ops.ace_flags_get(acep);
1241 
1242 	return (mask == accessmask && type == allow_deny &&
1243 	    ((flags & ACE_TYPE_FLAGS) == entry_type));
1244 }
1245 
1246 /*
1247  * Can prepended ACE be reused?
1248  */
1249 static int
1250 zfs_reuse_deny(zfs_acl_t *aclp, void *acep, void *prevacep)
1251 {
1252 	int okay_masks;
1253 	uint16_t prevtype;
1254 	uint16_t prevflags;
1255 	uint16_t flags;
1256 	uint32_t mask, prevmask;
1257 
1258 	if (prevacep == NULL)
1259 		return (B_FALSE);
1260 
1261 	prevtype = aclp->z_ops.ace_type_get(prevacep);
1262 	prevflags = aclp->z_ops.ace_flags_get(prevacep);
1263 	flags = aclp->z_ops.ace_flags_get(acep);
1264 	mask = aclp->z_ops.ace_mask_get(acep);
1265 	prevmask = aclp->z_ops.ace_mask_get(prevacep);
1266 
1267 	if (prevtype != DENY)
1268 		return (B_FALSE);
1269 
1270 	if (prevflags != (flags & ACE_IDENTIFIER_GROUP))
1271 		return (B_FALSE);
1272 
1273 	okay_masks = (mask & OKAY_MASK_BITS);
1274 
1275 	if (prevmask & ~okay_masks)
1276 		return (B_FALSE);
1277 
1278 	return (B_TRUE);
1279 }
1280 
1281 
1282 /*
1283  * Insert new ACL node into chain of zfs_acl_node_t's
1284  *
1285  * This will result in two possible results.
1286  * 1. If the ACL is currently just a single zfs_acl_node and
1287  *    we are prepending the entry then current acl node will have
1288  *    a new node inserted above it.
1289  *
1290  * 2. If we are inserting in the middle of current acl node then
1291  *    the current node will be split in two and new node will be inserted
1292  *    in between the two split nodes.
1293  */
1294 static zfs_acl_node_t *
1295 zfs_acl_ace_insert(zfs_acl_t *aclp, void  *acep)
1296 {
1297 	zfs_acl_node_t 	*newnode;
1298 	zfs_acl_node_t 	*trailernode = NULL;
1299 	zfs_acl_node_t 	*currnode = zfs_acl_curr_node(aclp);
1300 	int		curr_idx = aclp->z_curr_node->z_ace_idx;
1301 	int		trailer_count;
1302 	size_t		oldsize;
1303 
1304 	newnode = zfs_acl_node_alloc(aclp->z_ops.ace_size(acep));
1305 	newnode->z_ace_count = 1;
1306 
1307 	oldsize = currnode->z_size;
1308 
1309 	if (curr_idx != 1) {
1310 		trailernode = zfs_acl_node_alloc(0);
1311 		trailernode->z_acldata = acep;
1312 
1313 		trailer_count = currnode->z_ace_count - curr_idx + 1;
1314 		currnode->z_ace_count = curr_idx - 1;
1315 		currnode->z_size = (caddr_t)acep - (caddr_t)currnode->z_acldata;
1316 		trailernode->z_size = oldsize - currnode->z_size;
1317 		trailernode->z_ace_count = trailer_count;
1318 	}
1319 
1320 	aclp->z_acl_count += 1;
1321 	aclp->z_acl_bytes += aclp->z_ops.ace_size(acep);
1322 
1323 	if (curr_idx == 1)
1324 		list_insert_before(&aclp->z_acl, currnode, newnode);
1325 	else
1326 		list_insert_after(&aclp->z_acl, currnode, newnode);
1327 	if (trailernode) {
1328 		list_insert_after(&aclp->z_acl, newnode, trailernode);
1329 		aclp->z_curr_node = trailernode;
1330 		trailernode->z_ace_idx = 1;
1331 	}
1332 
1333 	return (newnode);
1334 }
1335 
1336 /*
1337  * Prepend deny ACE
1338  */
1339 static void *
1340 zfs_acl_prepend_deny(znode_t *zp, zfs_acl_t *aclp, void *acep,
1341     mode_t mode)
1342 {
1343 	zfs_acl_node_t *aclnode;
1344 	void  *newacep;
1345 	uint64_t fuid;
1346 	uint16_t flags;
1347 
1348 	aclnode = zfs_acl_ace_insert(aclp, acep);
1349 	newacep = aclnode->z_acldata;
1350 	fuid = aclp->z_ops.ace_who_get(acep);
1351 	flags = aclp->z_ops.ace_flags_get(acep);
1352 	zfs_set_ace(aclp, newacep, 0, DENY, fuid, (flags & ACE_TYPE_FLAGS));
1353 	zfs_acl_prepend_fixup(aclp, newacep, acep, mode, zp->z_phys->zp_uid);
1354 
1355 	return (newacep);
1356 }
1357 
1358 /*
1359  * Split an inherited ACE into inherit_only ACE
1360  * and original ACE with inheritance flags stripped off.
1361  */
1362 static void
1363 zfs_acl_split_ace(zfs_acl_t *aclp, zfs_ace_hdr_t *acep)
1364 {
1365 	zfs_acl_node_t *aclnode;
1366 	zfs_acl_node_t *currnode;
1367 	void  *newacep;
1368 	uint16_t type, flags;
1369 	uint32_t mask;
1370 	uint64_t fuid;
1371 
1372 	type = aclp->z_ops.ace_type_get(acep);
1373 	flags = aclp->z_ops.ace_flags_get(acep);
1374 	mask = aclp->z_ops.ace_mask_get(acep);
1375 	fuid = aclp->z_ops.ace_who_get(acep);
1376 
1377 	aclnode = zfs_acl_ace_insert(aclp, acep);
1378 	newacep = aclnode->z_acldata;
1379 
1380 	aclp->z_ops.ace_type_set(newacep, type);
1381 	aclp->z_ops.ace_flags_set(newacep, flags | ACE_INHERIT_ONLY_ACE);
1382 	aclp->z_ops.ace_mask_set(newacep, mask);
1383 	aclp->z_ops.ace_type_set(newacep, type);
1384 	aclp->z_ops.ace_who_set(newacep, fuid);
1385 	aclp->z_next_ace = acep;
1386 	flags &= ~ALL_INHERIT;
1387 	aclp->z_ops.ace_flags_set(acep, flags);
1388 	currnode = zfs_acl_curr_node(aclp);
1389 	ASSERT(currnode->z_ace_idx >= 1);
1390 	currnode->z_ace_idx -= 1;
1391 }
1392 
1393 /*
1394  * Are ACES started at index i, the canonical six ACES?
1395  */
1396 static int
1397 zfs_have_canonical_six(zfs_acl_t *aclp)
1398 {
1399 	void *acep;
1400 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1401 	int		i = 0;
1402 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1403 
1404 	ASSERT(aclnode != NULL);
1405 
1406 	if (aclnode->z_ace_count < 6)
1407 		return (0);
1408 
1409 	acep = (void *)((caddr_t)aclnode->z_acldata +
1410 	    aclnode->z_size - (aclp->z_ops.ace_abstract_size() * 6));
1411 
1412 	if ((zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1413 	    DENY, ACE_OWNER, 0) &&
1414 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1415 	    ALLOW, ACE_OWNER, OWNER_ALLOW_MASK) &&
1416 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++), DENY,
1417 	    OWNING_GROUP, 0) && zfs_acl_ace_match(aclp, (caddr_t)acep +
1418 	    (abstract_size * i++),
1419 	    ALLOW, OWNING_GROUP, 0) &&
1420 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1421 	    DENY, ACE_EVERYONE, EVERYONE_DENY_MASK) &&
1422 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1423 	    ALLOW, ACE_EVERYONE, EVERYONE_ALLOW_MASK))) {
1424 		return (1);
1425 	} else {
1426 		return (0);
1427 	}
1428 }
1429 
1430 
1431 /*
1432  * Apply step 1g, to group entries
1433  *
1434  * Need to deal with corner case where group may have
1435  * greater permissions than owner.  If so then limit
1436  * group permissions, based on what extra permissions
1437  * group has.
1438  */
1439 static void
1440 zfs_fixup_group_entries(zfs_acl_t *aclp, void *acep, void *prevacep,
1441     mode_t mode)
1442 {
1443 	uint32_t prevmask = aclp->z_ops.ace_mask_get(prevacep);
1444 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1445 	uint16_t prevflags = aclp->z_ops.ace_flags_get(prevacep);
1446 	mode_t extramode = (mode >> 3) & 07;
1447 	mode_t ownermode = (mode >> 6);
1448 
1449 	if (prevflags & ACE_IDENTIFIER_GROUP) {
1450 
1451 		extramode &= ~ownermode;
1452 
1453 		if (extramode) {
1454 			if (extramode & S_IROTH) {
1455 				prevmask &= ~ACE_READ_DATA;
1456 				mask &= ~ACE_READ_DATA;
1457 			}
1458 			if (extramode & S_IWOTH) {
1459 				prevmask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1460 				mask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1461 			}
1462 			if (extramode & S_IXOTH) {
1463 				prevmask  &= ~ACE_EXECUTE;
1464 				mask &= ~ACE_EXECUTE;
1465 			}
1466 		}
1467 	}
1468 	aclp->z_ops.ace_mask_set(acep, mask);
1469 	aclp->z_ops.ace_mask_set(prevacep, prevmask);
1470 }
1471 
1472 /*
1473  * Apply the chmod algorithm as described
1474  * in PSARC/2002/240
1475  */
1476 static void
1477 zfs_acl_chmod(znode_t *zp, uint64_t mode, zfs_acl_t *aclp)
1478 {
1479 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1480 	void		*acep = NULL, *prevacep = NULL;
1481 	uint64_t	who;
1482 	int 		i;
1483 	int 		entry_type;
1484 	int 		reuse_deny;
1485 	int 		need_canonical_six = 1;
1486 	uint16_t	iflags, type;
1487 	uint32_t	access_mask;
1488 
1489 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1490 	ASSERT(MUTEX_HELD(&zp->z_lock));
1491 
1492 	aclp->z_hints = (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
1493 
1494 	/*
1495 	 * If discard then just discard all ACL nodes which
1496 	 * represent the ACEs.
1497 	 *
1498 	 * New owner@/group@/everone@ ACEs will be added
1499 	 * later.
1500 	 */
1501 	if (zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1502 		zfs_acl_release_nodes(aclp);
1503 
1504 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1505 	    &iflags, &type)) {
1506 
1507 		entry_type = (iflags & ACE_TYPE_FLAGS);
1508 		iflags = (iflags & ALL_INHERIT);
1509 
1510 		if ((type != ALLOW && type != DENY) ||
1511 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1512 			if (iflags)
1513 				aclp->z_hints |= ZFS_INHERIT_ACE;
1514 			switch (type) {
1515 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1516 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1517 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1518 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1519 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1520 				break;
1521 			}
1522 			goto nextace;
1523 		}
1524 
1525 		/*
1526 		 * Need to split ace into two?
1527 		 */
1528 		if ((iflags & (ACE_FILE_INHERIT_ACE|
1529 		    ACE_DIRECTORY_INHERIT_ACE)) &&
1530 		    (!(iflags & ACE_INHERIT_ONLY_ACE))) {
1531 			zfs_acl_split_ace(aclp, acep);
1532 			aclp->z_hints |= ZFS_INHERIT_ACE;
1533 			goto nextace;
1534 		}
1535 
1536 		if (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1537 		    (entry_type == OWNING_GROUP)) {
1538 			access_mask &= ~OGE_CLEAR;
1539 			aclp->z_ops.ace_mask_set(acep, access_mask);
1540 			goto nextace;
1541 		} else {
1542 			reuse_deny = B_TRUE;
1543 			if (type == ALLOW) {
1544 
1545 				/*
1546 				 * Check preceding ACE if any, to see
1547 				 * if we need to prepend a DENY ACE.
1548 				 * This is only applicable when the acl_mode
1549 				 * property == groupmask.
1550 				 */
1551 				if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK) {
1552 
1553 					reuse_deny = zfs_reuse_deny(aclp, acep,
1554 					    prevacep);
1555 
1556 					if (!reuse_deny) {
1557 						prevacep =
1558 						    zfs_acl_prepend_deny(zp,
1559 						    aclp, acep, mode);
1560 					} else {
1561 						zfs_acl_prepend_fixup(
1562 						    aclp, prevacep,
1563 						    acep, mode,
1564 						    zp->z_phys->zp_uid);
1565 					}
1566 					zfs_fixup_group_entries(aclp, acep,
1567 					    prevacep, mode);
1568 
1569 				}
1570 			}
1571 		}
1572 nextace:
1573 		prevacep = acep;
1574 	}
1575 
1576 	/*
1577 	 * Check out last six aces, if we have six.
1578 	 */
1579 
1580 	if (aclp->z_acl_count >= 6) {
1581 		if (zfs_have_canonical_six(aclp)) {
1582 			need_canonical_six = 0;
1583 		}
1584 	}
1585 
1586 	if (need_canonical_six) {
1587 		size_t abstract_size = aclp->z_ops.ace_abstract_size();
1588 		void *zacep;
1589 		zfs_acl_node_t *aclnode =
1590 		    zfs_acl_node_alloc(abstract_size * 6);
1591 
1592 		aclnode->z_size = abstract_size * 6;
1593 		aclnode->z_ace_count = 6;
1594 		aclp->z_acl_bytes += aclnode->z_size;
1595 		list_insert_tail(&aclp->z_acl, aclnode);
1596 
1597 		zacep = aclnode->z_acldata;
1598 
1599 		i = 0;
1600 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1601 		    0, DENY, -1, ACE_OWNER);
1602 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1603 		    OWNER_ALLOW_MASK, ALLOW, -1, ACE_OWNER);
1604 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1605 		    DENY, -1, OWNING_GROUP);
1606 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1607 		    ALLOW, -1, OWNING_GROUP);
1608 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1609 		    EVERYONE_DENY_MASK, DENY, -1, ACE_EVERYONE);
1610 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1611 		    EVERYONE_ALLOW_MASK, ALLOW, -1, ACE_EVERYONE);
1612 		aclp->z_acl_count += 6;
1613 	}
1614 
1615 	zfs_acl_fixup_canonical_six(aclp, mode);
1616 }
1617 
1618 int
1619 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1620 {
1621 	int error;
1622 
1623 	mutex_enter(&zp->z_lock);
1624 	mutex_enter(&zp->z_acl_lock);
1625 	*aclp = NULL;
1626 	error = zfs_acl_node_read(zp, aclp, B_TRUE);
1627 	if (error == 0)
1628 		zfs_acl_chmod(zp, mode, *aclp);
1629 	mutex_exit(&zp->z_acl_lock);
1630 	mutex_exit(&zp->z_lock);
1631 	return (error);
1632 }
1633 
1634 /*
1635  * strip off write_owner and write_acl
1636  */
1637 static void
1638 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1639 {
1640 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1641 
1642 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1643 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1644 		mask &= ~RESTRICTED_CLEAR;
1645 		aclp->z_ops.ace_mask_set(acep, mask);
1646 	}
1647 }
1648 
1649 /*
1650  * Should ACE be inherited?
1651  */
1652 static int
1653 zfs_ace_can_use(znode_t *zp, uint16_t acep_flags)
1654 {
1655 	int vtype = ZTOV(zp)->v_type;
1656 	int	iflags = (acep_flags & 0xf);
1657 
1658 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1659 		return (1);
1660 	else if (iflags & ACE_FILE_INHERIT_ACE)
1661 		return (!((vtype == VDIR) &&
1662 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1663 	return (0);
1664 }
1665 
1666 /*
1667  * inherit inheritable ACEs from parent
1668  */
1669 static zfs_acl_t *
1670 zfs_acl_inherit(znode_t *zp, zfs_acl_t *paclp, uint64_t mode,
1671     boolean_t *need_chmod)
1672 {
1673 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1674 	void		*pacep;
1675 	void		*acep, *acep2;
1676 	zfs_acl_node_t  *aclnode, *aclnode2;
1677 	zfs_acl_t	*aclp = NULL;
1678 	uint64_t	who;
1679 	uint32_t	access_mask;
1680 	uint16_t	iflags, newflags, type;
1681 	size_t		ace_size;
1682 	void		*data1, *data2;
1683 	size_t		data1sz, data2sz;
1684 	boolean_t	vdir = ZTOV(zp)->v_type == VDIR;
1685 	boolean_t	vreg = ZTOV(zp)->v_type == VREG;
1686 	boolean_t	passthrough, passthrough_x, noallow;
1687 
1688 	passthrough_x =
1689 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1690 	passthrough = passthrough_x ||
1691 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1692 	noallow =
1693 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1694 
1695 	*need_chmod = B_TRUE;
1696 	pacep = NULL;
1697 	aclp = zfs_acl_alloc(paclp->z_version);
1698 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD)
1699 		return (aclp);
1700 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1701 	    &access_mask, &iflags, &type)) {
1702 
1703 		/*
1704 		 * don't inherit bogus ACEs
1705 		 */
1706 		if (!zfs_acl_valid_ace_type(type, iflags))
1707 			continue;
1708 
1709 		if (noallow && type == ALLOW)
1710 			continue;
1711 
1712 		ace_size = aclp->z_ops.ace_size(pacep);
1713 
1714 		if (!zfs_ace_can_use(zp, iflags))
1715 			continue;
1716 
1717 		/*
1718 		 * If owner@, group@, or everyone@ inheritable
1719 		 * then zfs_acl_chmod() isn't needed.
1720 		 */
1721 		if (passthrough &&
1722 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1723 		    ((iflags & OWNING_GROUP) ==
1724 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1725 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1726 			*need_chmod = B_FALSE;
1727 
1728 			if (!vdir && passthrough_x &&
1729 			    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1730 				access_mask &= ~ACE_EXECUTE;
1731 			}
1732 		}
1733 
1734 		aclnode = zfs_acl_node_alloc(ace_size);
1735 		list_insert_tail(&aclp->z_acl, aclnode);
1736 		acep = aclnode->z_acldata;
1737 
1738 		zfs_set_ace(aclp, acep, access_mask, type,
1739 		    who, iflags|ACE_INHERITED_ACE);
1740 
1741 		/*
1742 		 * Copy special opaque data if any
1743 		 */
1744 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1745 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1746 			    &data2)) == data1sz);
1747 			bcopy(data1, data2, data2sz);
1748 		}
1749 		aclp->z_acl_count++;
1750 		aclnode->z_ace_count++;
1751 		aclp->z_acl_bytes += aclnode->z_size;
1752 		newflags = aclp->z_ops.ace_flags_get(acep);
1753 
1754 		if (vdir)
1755 			aclp->z_hints |= ZFS_INHERIT_ACE;
1756 
1757 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1758 			newflags &= ~ALL_INHERIT;
1759 			aclp->z_ops.ace_flags_set(acep,
1760 			    newflags|ACE_INHERITED_ACE);
1761 			zfs_restricted_update(zfsvfs, aclp, acep);
1762 			continue;
1763 		}
1764 
1765 		ASSERT(vdir);
1766 
1767 		newflags = aclp->z_ops.ace_flags_get(acep);
1768 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1769 		    ACE_DIRECTORY_INHERIT_ACE)) !=
1770 		    ACE_FILE_INHERIT_ACE) {
1771 			aclnode2 = zfs_acl_node_alloc(ace_size);
1772 			list_insert_tail(&aclp->z_acl, aclnode2);
1773 			acep2 = aclnode2->z_acldata;
1774 			zfs_set_ace(aclp, acep2,
1775 			    access_mask, type, who,
1776 			    iflags|ACE_INHERITED_ACE);
1777 			newflags |= ACE_INHERIT_ONLY_ACE;
1778 			aclp->z_ops.ace_flags_set(acep, newflags);
1779 			newflags &= ~ALL_INHERIT;
1780 			aclp->z_ops.ace_flags_set(acep2,
1781 			    newflags|ACE_INHERITED_ACE);
1782 
1783 			/*
1784 			 * Copy special opaque data if any
1785 			 */
1786 			if ((data1sz = aclp->z_ops.ace_data(acep,
1787 			    &data1)) != 0) {
1788 				VERIFY((data2sz =
1789 				    aclp->z_ops.ace_data(acep2,
1790 				    &data2)) == data1sz);
1791 				bcopy(data1, data2, data1sz);
1792 			}
1793 			aclp->z_acl_count++;
1794 			aclnode2->z_ace_count++;
1795 			aclp->z_acl_bytes += aclnode->z_size;
1796 			zfs_restricted_update(zfsvfs, aclp, acep2);
1797 		} else {
1798 			newflags |= ACE_INHERIT_ONLY_ACE;
1799 			aclp->z_ops.ace_flags_set(acep,
1800 			    newflags|ACE_INHERITED_ACE);
1801 		}
1802 	}
1803 	return (aclp);
1804 }
1805 
1806 /*
1807  * Create file system object initial permissions
1808  * including inheritable ACEs.
1809  */
1810 void
1811 zfs_perm_init(znode_t *zp, znode_t *parent, int flag,
1812     vattr_t *vap, dmu_tx_t *tx, cred_t *cr,
1813     zfs_acl_t *setaclp, zfs_fuid_info_t **fuidp)
1814 {
1815 	uint64_t	mode, fuid, fgid;
1816 	int		error;
1817 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1818 	zfs_acl_t	*aclp = NULL;
1819 	zfs_acl_t	*paclp;
1820 	xvattr_t	*xvap = (xvattr_t *)vap;
1821 	gid_t		gid;
1822 	boolean_t	need_chmod = B_TRUE;
1823 
1824 	if (setaclp)
1825 		aclp = setaclp;
1826 
1827 	mode = MAKEIMODE(vap->va_type, vap->va_mode);
1828 
1829 	/*
1830 	 * Determine uid and gid.
1831 	 */
1832 	if ((flag & (IS_ROOT_NODE | IS_REPLAY)) ||
1833 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1834 		fuid = zfs_fuid_create(zfsvfs, vap->va_uid, cr,
1835 		    ZFS_OWNER, tx, fuidp);
1836 		fgid = zfs_fuid_create(zfsvfs, vap->va_gid, cr,
1837 		    ZFS_GROUP, tx, fuidp);
1838 		gid = vap->va_gid;
1839 	} else {
1840 		fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER, tx, cr, fuidp);
1841 		fgid = 0;
1842 		if (vap->va_mask & AT_GID)  {
1843 			fgid = zfs_fuid_create(zfsvfs, vap->va_gid, cr,
1844 			    ZFS_GROUP, tx, fuidp);
1845 			gid = vap->va_gid;
1846 			if (fgid != parent->z_phys->zp_gid &&
1847 			    !groupmember(vap->va_gid, cr) &&
1848 			    secpolicy_vnode_create_gid(cr) != 0)
1849 				fgid = 0;
1850 		}
1851 		if (fgid == 0) {
1852 			if (parent->z_phys->zp_mode & S_ISGID) {
1853 				fgid = parent->z_phys->zp_gid;
1854 				gid = zfs_fuid_map_id(zfsvfs, fgid,
1855 				    cr, ZFS_GROUP);
1856 			} else {
1857 				fgid = zfs_fuid_create_cred(zfsvfs,
1858 				    ZFS_GROUP, tx, cr, fuidp);
1859 				gid = crgetgid(cr);
1860 			}
1861 		}
1862 	}
1863 
1864 	/*
1865 	 * If we're creating a directory, and the parent directory has the
1866 	 * set-GID bit set, set in on the new directory.
1867 	 * Otherwise, if the user is neither privileged nor a member of the
1868 	 * file's new group, clear the file's set-GID bit.
1869 	 */
1870 
1871 	if ((parent->z_phys->zp_mode & S_ISGID) && (vap->va_type == VDIR)) {
1872 		mode |= S_ISGID;
1873 	} else {
1874 		if ((mode & S_ISGID) &&
1875 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1876 			mode &= ~S_ISGID;
1877 	}
1878 
1879 	zp->z_phys->zp_uid = fuid;
1880 	zp->z_phys->zp_gid = fgid;
1881 	zp->z_phys->zp_mode = mode;
1882 
1883 	if (aclp == NULL) {
1884 		mutex_enter(&parent->z_lock);
1885 		if ((ZTOV(parent)->v_type == VDIR &&
1886 		    (parent->z_phys->zp_flags & ZFS_INHERIT_ACE)) &&
1887 		    !(zp->z_phys->zp_flags & ZFS_XATTR)) {
1888 			mutex_enter(&parent->z_acl_lock);
1889 			VERIFY(0 == zfs_acl_node_read(parent, &paclp, B_FALSE));
1890 			mutex_exit(&parent->z_acl_lock);
1891 			aclp = zfs_acl_inherit(zp, paclp, mode, &need_chmod);
1892 			zfs_acl_free(paclp);
1893 		} else {
1894 			aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1895 		}
1896 		mutex_exit(&parent->z_lock);
1897 		mutex_enter(&zp->z_lock);
1898 		mutex_enter(&zp->z_acl_lock);
1899 		if (need_chmod)
1900 			zfs_acl_chmod(zp, mode, aclp);
1901 	} else {
1902 		mutex_enter(&zp->z_lock);
1903 		mutex_enter(&zp->z_acl_lock);
1904 	}
1905 
1906 	/* Force auto_inherit on all new directory objects */
1907 	if (vap->va_type == VDIR)
1908 		aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1909 
1910 	error = zfs_aclset_common(zp, aclp, cr, fuidp, tx);
1911 
1912 	/* Set optional attributes if any */
1913 	if (vap->va_mask & AT_XVATTR)
1914 		zfs_xvattr_set(zp, xvap);
1915 
1916 	mutex_exit(&zp->z_lock);
1917 	mutex_exit(&zp->z_acl_lock);
1918 	ASSERT3U(error, ==, 0);
1919 
1920 	if (aclp != setaclp)
1921 		zfs_acl_free(aclp);
1922 }
1923 
1924 /*
1925  * Retrieve a files ACL
1926  */
1927 int
1928 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1929 {
1930 	zfs_acl_t	*aclp;
1931 	ulong_t		mask;
1932 	int		error;
1933 	int 		count = 0;
1934 	int		largeace = 0;
1935 
1936 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1937 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1938 
1939 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1940 		return (error);
1941 
1942 	if (mask == 0)
1943 		return (ENOSYS);
1944 
1945 	mutex_enter(&zp->z_acl_lock);
1946 
1947 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
1948 	if (error != 0) {
1949 		mutex_exit(&zp->z_acl_lock);
1950 		return (error);
1951 	}
1952 
1953 	/*
1954 	 * Scan ACL to determine number of ACEs
1955 	 */
1956 	if ((zp->z_phys->zp_flags & ZFS_ACL_OBJ_ACE) &&
1957 	    !(mask & VSA_ACE_ALLTYPES)) {
1958 		void *zacep = NULL;
1959 		uint64_t who;
1960 		uint32_t access_mask;
1961 		uint16_t type, iflags;
1962 
1963 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1964 		    &who, &access_mask, &iflags, &type)) {
1965 			switch (type) {
1966 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1967 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1968 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1969 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1970 				largeace++;
1971 				continue;
1972 			default:
1973 				count++;
1974 			}
1975 		}
1976 		vsecp->vsa_aclcnt = count;
1977 	} else
1978 		count = aclp->z_acl_count;
1979 
1980 	if (mask & VSA_ACECNT) {
1981 		vsecp->vsa_aclcnt = count;
1982 	}
1983 
1984 	if (mask & VSA_ACE) {
1985 		size_t aclsz;
1986 
1987 		zfs_acl_node_t *aclnode = list_head(&aclp->z_acl);
1988 
1989 		aclsz = count * sizeof (ace_t) +
1990 		    sizeof (ace_object_t) * largeace;
1991 
1992 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1993 		vsecp->vsa_aclentsz = aclsz;
1994 
1995 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1996 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1997 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1998 		else {
1999 			bcopy(aclnode->z_acldata, vsecp->vsa_aclentp,
2000 			    count * sizeof (ace_t));
2001 		}
2002 	}
2003 	if (mask & VSA_ACE_ACLFLAGS) {
2004 		vsecp->vsa_aclflags = 0;
2005 		if (zp->z_phys->zp_flags & ZFS_ACL_DEFAULTED)
2006 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
2007 		if (zp->z_phys->zp_flags & ZFS_ACL_PROTECTED)
2008 			vsecp->vsa_aclflags |= ACL_PROTECTED;
2009 		if (zp->z_phys->zp_flags & ZFS_ACL_AUTO_INHERIT)
2010 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2011 	}
2012 
2013 	mutex_exit(&zp->z_acl_lock);
2014 
2015 	zfs_acl_free(aclp);
2016 
2017 	return (0);
2018 }
2019 
2020 int
2021 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
2022     vsecattr_t *vsecp, zfs_acl_t **zaclp)
2023 {
2024 	zfs_acl_t *aclp;
2025 	zfs_acl_node_t *aclnode;
2026 	int aclcnt = vsecp->vsa_aclcnt;
2027 	int error;
2028 
2029 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2030 		return (EINVAL);
2031 
2032 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2033 
2034 	aclp->z_hints = 0;
2035 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2036 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2037 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
2038 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2039 		    aclcnt, &aclnode->z_size)) != 0) {
2040 			zfs_acl_free(aclp);
2041 			zfs_acl_node_free(aclnode);
2042 			return (error);
2043 		}
2044 	} else {
2045 		if ((error = zfs_copy_ace_2_fuid(obj_type, aclp,
2046 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2047 		    &aclnode->z_size)) != 0) {
2048 			zfs_acl_free(aclp);
2049 			zfs_acl_node_free(aclnode);
2050 			return (error);
2051 		}
2052 	}
2053 	aclp->z_acl_bytes = aclnode->z_size;
2054 	aclnode->z_ace_count = aclcnt;
2055 	aclp->z_acl_count = aclcnt;
2056 	list_insert_head(&aclp->z_acl, aclnode);
2057 
2058 	/*
2059 	 * If flags are being set then add them to z_hints
2060 	 */
2061 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2062 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2063 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2064 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2065 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2066 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2067 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2068 	}
2069 
2070 	*zaclp = aclp;
2071 
2072 	return (0);
2073 }
2074 
2075 /*
2076  * Set a files ACL
2077  */
2078 int
2079 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2080 {
2081 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2082 	zilog_t		*zilog = zfsvfs->z_log;
2083 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2084 	dmu_tx_t	*tx;
2085 	int		error;
2086 	zfs_acl_t	*aclp;
2087 	zfs_fuid_info_t	*fuidp = NULL;
2088 
2089 	if (mask == 0)
2090 		return (ENOSYS);
2091 
2092 	if (zp->z_phys->zp_flags & ZFS_IMMUTABLE)
2093 		return (EPERM);
2094 
2095 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
2096 		return (error);
2097 
2098 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, &aclp);
2099 	if (error)
2100 		return (error);
2101 
2102 	/*
2103 	 * If ACL wide flags aren't being set then preserve any
2104 	 * existing flags.
2105 	 */
2106 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2107 		aclp->z_hints |= (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
2108 	}
2109 top:
2110 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)) {
2111 		zfs_acl_free(aclp);
2112 		return (error);
2113 	}
2114 
2115 	mutex_enter(&zp->z_lock);
2116 	mutex_enter(&zp->z_acl_lock);
2117 
2118 	tx = dmu_tx_create(zfsvfs->z_os);
2119 	dmu_tx_hold_bonus(tx, zp->z_id);
2120 
2121 	if (zp->z_phys->zp_acl.z_acl_extern_obj) {
2122 		/* Are we upgrading ACL? */
2123 		if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2124 		    zp->z_phys->zp_acl.z_acl_version ==
2125 		    ZFS_ACL_VERSION_INITIAL) {
2126 			dmu_tx_hold_free(tx,
2127 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2128 			    0, DMU_OBJECT_END);
2129 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2130 			    0, aclp->z_acl_bytes);
2131 		} else {
2132 			dmu_tx_hold_write(tx,
2133 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2134 			    0, aclp->z_acl_bytes);
2135 		}
2136 	} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2137 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2138 	}
2139 	if (aclp->z_has_fuids) {
2140 		if (zfsvfs->z_fuid_obj == 0) {
2141 			dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2142 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2143 			    FUID_SIZE_ESTIMATE(zfsvfs));
2144 			dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL);
2145 		} else {
2146 			dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj);
2147 			dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0,
2148 			    FUID_SIZE_ESTIMATE(zfsvfs));
2149 		}
2150 	}
2151 
2152 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2153 	if (error) {
2154 		mutex_exit(&zp->z_acl_lock);
2155 		mutex_exit(&zp->z_lock);
2156 
2157 		if (error == ERESTART) {
2158 			dmu_tx_wait(tx);
2159 			dmu_tx_abort(tx);
2160 			goto top;
2161 		}
2162 		dmu_tx_abort(tx);
2163 		zfs_acl_free(aclp);
2164 		return (error);
2165 	}
2166 
2167 	error = zfs_aclset_common(zp, aclp, cr, &fuidp, tx);
2168 	ASSERT(error == 0);
2169 
2170 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2171 
2172 	if (fuidp)
2173 		zfs_fuid_info_free(fuidp);
2174 	zfs_acl_free(aclp);
2175 	dmu_tx_commit(tx);
2176 done:
2177 	mutex_exit(&zp->z_acl_lock);
2178 	mutex_exit(&zp->z_lock);
2179 
2180 	return (error);
2181 }
2182 
2183 /*
2184  * working_mode returns the permissions that were not granted
2185  */
2186 static int
2187 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2188     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2189 {
2190 	zfs_acl_t	*aclp;
2191 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2192 	int		error;
2193 	uid_t		uid = crgetuid(cr);
2194 	uint64_t 	who;
2195 	uint16_t	type, iflags;
2196 	uint16_t	entry_type;
2197 	uint32_t	access_mask;
2198 	uint32_t	deny_mask = 0;
2199 	zfs_ace_hdr_t	*acep = NULL;
2200 	boolean_t	checkit;
2201 	uid_t		fowner;
2202 	uid_t		gowner;
2203 
2204 	/*
2205 	 * Short circuit empty requests
2206 	 */
2207 	if (v4_mode == 0)
2208 		return (0);
2209 
2210 	*check_privs = B_TRUE;
2211 
2212 	if (zfsvfs->z_replay) {
2213 		*working_mode = 0;
2214 		return (0);
2215 	}
2216 
2217 	*working_mode = v4_mode;
2218 
2219 	if ((v4_mode & WRITE_MASK) &&
2220 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2221 	    (!IS_DEVVP(ZTOV(zp)) ||
2222 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2223 		*check_privs = B_FALSE;
2224 		return (EROFS);
2225 	}
2226 
2227 	/*
2228 	 * Only check for READONLY on non-directories.
2229 	 */
2230 	if ((v4_mode & WRITE_MASK_DATA) &&
2231 	    (((ZTOV(zp)->v_type != VDIR) &&
2232 	    (zp->z_phys->zp_flags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2233 	    (ZTOV(zp)->v_type == VDIR &&
2234 	    (zp->z_phys->zp_flags & ZFS_IMMUTABLE)))) {
2235 		*check_privs = B_FALSE;
2236 		return (EPERM);
2237 	}
2238 
2239 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2240 	    (zp->z_phys->zp_flags & ZFS_NOUNLINK)) {
2241 		*check_privs = B_FALSE;
2242 		return (EPERM);
2243 	}
2244 
2245 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2246 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED))) {
2247 		*check_privs = B_FALSE;
2248 		return (EACCES);
2249 	}
2250 
2251 	/*
2252 	 * The caller requested that the ACL check be skipped.  This
2253 	 * would only happen if the caller checked VOP_ACCESS() with a
2254 	 * 32 bit ACE mask and already had the appropriate permissions.
2255 	 */
2256 	if (skipaclchk) {
2257 		*working_mode = 0;
2258 		return (0);
2259 	}
2260 
2261 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2262 
2263 	mutex_enter(&zp->z_acl_lock);
2264 
2265 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2266 	if (error != 0) {
2267 		mutex_exit(&zp->z_acl_lock);
2268 		return (error);
2269 	}
2270 
2271 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2272 	    &iflags, &type)) {
2273 
2274 		if (!zfs_acl_valid_ace_type(type, iflags))
2275 			continue;
2276 
2277 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2278 			continue;
2279 
2280 		entry_type = (iflags & ACE_TYPE_FLAGS);
2281 
2282 		checkit = B_FALSE;
2283 
2284 		switch (entry_type) {
2285 		case ACE_OWNER:
2286 			if (uid == fowner)
2287 				checkit = B_TRUE;
2288 			break;
2289 		case OWNING_GROUP:
2290 			who = gowner;
2291 			/*FALLTHROUGH*/
2292 		case ACE_IDENTIFIER_GROUP:
2293 			checkit = zfs_groupmember(zfsvfs, who, cr);
2294 			break;
2295 		case ACE_EVERYONE:
2296 			checkit = B_TRUE;
2297 			break;
2298 
2299 		/* USER Entry */
2300 		default:
2301 			if (entry_type == 0) {
2302 				uid_t newid;
2303 
2304 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2305 				    ZFS_ACE_USER);
2306 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2307 				    uid == newid)
2308 					checkit = B_TRUE;
2309 				break;
2310 			} else {
2311 				zfs_acl_free(aclp);
2312 				mutex_exit(&zp->z_acl_lock);
2313 				return (EIO);
2314 			}
2315 		}
2316 
2317 		if (checkit) {
2318 			uint32_t mask_matched = (access_mask & *working_mode);
2319 
2320 			if (mask_matched) {
2321 				if (type == DENY)
2322 					deny_mask |= mask_matched;
2323 
2324 				*working_mode &= ~mask_matched;
2325 			}
2326 		}
2327 
2328 		/* Are we done? */
2329 		if (*working_mode == 0)
2330 			break;
2331 	}
2332 
2333 	mutex_exit(&zp->z_acl_lock);
2334 	zfs_acl_free(aclp);
2335 
2336 	/* Put the found 'denies' back on the working mode */
2337 	if (deny_mask) {
2338 		*working_mode |= deny_mask;
2339 		return (EACCES);
2340 	} else if (*working_mode) {
2341 		return (-1);
2342 	}
2343 
2344 	return (0);
2345 }
2346 
2347 static int
2348 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2349     cred_t *cr)
2350 {
2351 	if (*working_mode != ACE_WRITE_DATA)
2352 		return (EACCES);
2353 
2354 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2355 	    check_privs, B_FALSE, cr));
2356 }
2357 
2358 /*
2359  * Determine whether Access should be granted/denied, invoking least
2360  * priv subsytem when a deny is determined.
2361  */
2362 int
2363 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2364 {
2365 	uint32_t	working_mode;
2366 	int		error;
2367 	int		is_attr;
2368 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2369 	boolean_t 	check_privs;
2370 	znode_t		*xzp;
2371 	znode_t 	*check_zp = zp;
2372 
2373 	is_attr = ((zp->z_phys->zp_flags & ZFS_XATTR) &&
2374 	    (ZTOV(zp)->v_type == VDIR));
2375 
2376 	/*
2377 	 * If attribute then validate against base file
2378 	 */
2379 	if (is_attr) {
2380 		if ((error = zfs_zget(zp->z_zfsvfs,
2381 		    zp->z_phys->zp_parent, &xzp)) != 0)	{
2382 			return (error);
2383 		}
2384 
2385 		check_zp = xzp;
2386 
2387 		/*
2388 		 * fixup mode to map to xattr perms
2389 		 */
2390 
2391 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2392 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2393 			mode |= ACE_WRITE_NAMED_ATTRS;
2394 		}
2395 
2396 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2397 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2398 			mode |= ACE_READ_NAMED_ATTRS;
2399 		}
2400 	}
2401 
2402 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2403 	    &check_privs, skipaclchk, cr)) == 0) {
2404 		if (is_attr)
2405 			VN_RELE(ZTOV(xzp));
2406 		return (0);
2407 	}
2408 
2409 	if (error && !check_privs) {
2410 		if (is_attr)
2411 			VN_RELE(ZTOV(xzp));
2412 		return (error);
2413 	}
2414 
2415 	if (error && (flags & V_APPEND)) {
2416 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2417 	}
2418 
2419 	if (error && check_privs) {
2420 		uid_t		owner;
2421 		mode_t		checkmode = 0;
2422 
2423 		owner = zfs_fuid_map_id(zfsvfs, check_zp->z_phys->zp_uid, cr,
2424 		    ZFS_OWNER);
2425 
2426 		/*
2427 		 * First check for implicit owner permission on
2428 		 * read_acl/read_attributes
2429 		 */
2430 
2431 		error = 0;
2432 		ASSERT(working_mode != 0);
2433 
2434 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2435 		    owner == crgetuid(cr)))
2436 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2437 
2438 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2439 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2440 			checkmode |= VREAD;
2441 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2442 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2443 			checkmode |= VWRITE;
2444 		if (working_mode & ACE_EXECUTE)
2445 			checkmode |= VEXEC;
2446 
2447 		if (checkmode)
2448 			error = secpolicy_vnode_access(cr, ZTOV(check_zp),
2449 			    owner, checkmode);
2450 
2451 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2452 			error = secpolicy_vnode_chown(cr, B_TRUE);
2453 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2454 			error = secpolicy_vnode_setdac(cr, owner);
2455 
2456 		if (error == 0 && (working_mode &
2457 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2458 			error = secpolicy_vnode_remove(cr);
2459 
2460 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2461 			error = secpolicy_vnode_chown(cr, B_FALSE);
2462 		}
2463 		if (error == 0) {
2464 			/*
2465 			 * See if any bits other than those already checked
2466 			 * for are still present.  If so then return EACCES
2467 			 */
2468 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2469 				error = EACCES;
2470 			}
2471 		}
2472 	}
2473 
2474 	if (is_attr)
2475 		VN_RELE(ZTOV(xzp));
2476 
2477 	return (error);
2478 }
2479 
2480 /*
2481  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2482  * native ACL format and call zfs_zaccess()
2483  */
2484 int
2485 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2486 {
2487 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2488 }
2489 
2490 /*
2491  * Access function for secpolicy_vnode_setattr
2492  */
2493 int
2494 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2495 {
2496 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2497 
2498 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2499 }
2500 
2501 static int
2502 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2503     mode_t missing_perms, cred_t *cr)
2504 {
2505 	int error;
2506 	uid_t downer;
2507 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2508 
2509 	downer = zfs_fuid_map_id(zfsvfs, dzp->z_phys->zp_uid, cr, ZFS_OWNER);
2510 
2511 	error = secpolicy_vnode_access(cr, ZTOV(dzp), downer, missing_perms);
2512 
2513 	if (error == 0)
2514 		error = zfs_sticky_remove_access(dzp, zp, cr);
2515 
2516 	return (error);
2517 }
2518 
2519 /*
2520  * Determine whether Access should be granted/deny, without
2521  * consulting least priv subsystem.
2522  *
2523  *
2524  * The following chart is the recommended NFSv4 enforcement for
2525  * ability to delete an object.
2526  *
2527  *      -------------------------------------------------------
2528  *      |   Parent Dir  |           Target Object Permissions |
2529  *      |  permissions  |                                     |
2530  *      -------------------------------------------------------
2531  *      |               | ACL Allows | ACL Denies| Delete     |
2532  *      |               |  Delete    |  Delete   | unspecified|
2533  *      -------------------------------------------------------
2534  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2535  *      |  DELETE_CHILD |                                     |
2536  *      -------------------------------------------------------
2537  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2538  *      |  DELETE_CHILD |            |           |            |
2539  *      -------------------------------------------------------
2540  *      | ACL specifies |            |           |            |
2541  *      | only allow    | Permit     | Permit    | Permit     |
2542  *      | write and     |            |           |            |
2543  *      | execute       |            |           |            |
2544  *      -------------------------------------------------------
2545  *      | ACL denies    |            |           |            |
2546  *      | write and     | Permit     | Deny      | Deny       |
2547  *      | execute       |            |           |            |
2548  *      -------------------------------------------------------
2549  *         ^
2550  *         |
2551  *         No search privilege, can't even look up file?
2552  *
2553  */
2554 int
2555 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2556 {
2557 	uint32_t dzp_working_mode = 0;
2558 	uint32_t zp_working_mode = 0;
2559 	int dzp_error, zp_error;
2560 	mode_t missing_perms;
2561 	boolean_t dzpcheck_privs = B_TRUE;
2562 	boolean_t zpcheck_privs = B_TRUE;
2563 
2564 	/*
2565 	 * We want specific DELETE permissions to
2566 	 * take precedence over WRITE/EXECUTE.  We don't
2567 	 * want an ACL such as this to mess us up.
2568 	 * user:joe:write_data:deny,user:joe:delete:allow
2569 	 *
2570 	 * However, deny permissions may ultimately be overridden
2571 	 * by secpolicy_vnode_access().
2572 	 *
2573 	 * We will ask for all of the necessary permissions and then
2574 	 * look at the working modes from the directory and target object
2575 	 * to determine what was found.
2576 	 */
2577 
2578 	if (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2579 		return (EPERM);
2580 
2581 	/*
2582 	 * First row
2583 	 * If the directory permissions allow the delete, we are done.
2584 	 */
2585 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2586 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2587 		return (0);
2588 
2589 	/*
2590 	 * If target object has delete permission then we are done
2591 	 */
2592 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2593 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2594 		return (0);
2595 
2596 	ASSERT(dzp_error && zp_error);
2597 
2598 	if (!dzpcheck_privs)
2599 		return (dzp_error);
2600 	if (!zpcheck_privs)
2601 		return (zp_error);
2602 
2603 	/*
2604 	 * Second row
2605 	 *
2606 	 * If directory returns EACCES then delete_child was denied
2607 	 * due to deny delete_child.  In this case send the request through
2608 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2609 	 * since that *could* allow the delete based on write/execute permission
2610 	 * and we want delete permissions to override write/execute.
2611 	 */
2612 
2613 	if (dzp_error == EACCES)
2614 		return (secpolicy_vnode_remove(cr));
2615 
2616 	/*
2617 	 * Third Row
2618 	 * only need to see if we have write/execute on directory.
2619 	 */
2620 
2621 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2622 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2623 		return (zfs_sticky_remove_access(dzp, zp, cr));
2624 
2625 	if (!dzpcheck_privs)
2626 		return (dzp_error);
2627 
2628 	/*
2629 	 * Fourth row
2630 	 */
2631 
2632 	missing_perms = (dzp_working_mode & ACE_WRITE_DATA) ? VWRITE : 0;
2633 	missing_perms |= (dzp_working_mode & ACE_EXECUTE) ? VEXEC : 0;
2634 
2635 	ASSERT(missing_perms);
2636 
2637 	return (zfs_delete_final_check(zp, dzp, missing_perms, cr));
2638 
2639 }
2640 
2641 int
2642 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2643     znode_t *tzp, cred_t *cr)
2644 {
2645 	int add_perm;
2646 	int error;
2647 
2648 	if (szp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
2649 		return (EACCES);
2650 
2651 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2652 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2653 
2654 	/*
2655 	 * Rename permissions are combination of delete permission +
2656 	 * add file/subdir permission.
2657 	 */
2658 
2659 	/*
2660 	 * first make sure we do the delete portion.
2661 	 *
2662 	 * If that succeeds then check for add_file/add_subdir permissions
2663 	 */
2664 
2665 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2666 		return (error);
2667 
2668 	/*
2669 	 * If we have a tzp, see if we can delete it?
2670 	 */
2671 	if (tzp) {
2672 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2673 			return (error);
2674 	}
2675 
2676 	/*
2677 	 * Now check for add permissions
2678 	 */
2679 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2680 
2681 	return (error);
2682 }
2683