xref: /titanic_44/usr/src/uts/common/fs/zfs/zfs_acl.c (revision 870ad75a2b67a92c3449d93b4fef8a0baa982b4a)
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(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
543     void *datap, zfs_ace_t *z_acl, int aclcnt, size_t *size,
544     zfs_fuid_info_t **fuidp, cred_t *cr)
545 {
546 	int i;
547 	uint16_t entry_type;
548 	zfs_ace_t *aceptr = z_acl;
549 	ace_t *acep = datap;
550 	zfs_object_ace_t *zobjacep;
551 	ace_object_t *aceobjp;
552 
553 	for (i = 0; i != aclcnt; i++) {
554 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
555 		aceptr->z_hdr.z_flags = acep->a_flags;
556 		aceptr->z_hdr.z_type = acep->a_type;
557 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
558 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
559 		    entry_type != ACE_EVERYONE) {
560 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
561 			    cr, (entry_type == 0) ?
562 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
563 		}
564 
565 		/*
566 		 * Make sure ACE is valid
567 		 */
568 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
569 		    aceptr->z_hdr.z_flags) != B_TRUE)
570 			return (EINVAL);
571 
572 		switch (acep->a_type) {
573 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
574 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
575 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
576 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
577 			zobjacep = (zfs_object_ace_t *)aceptr;
578 			aceobjp = (ace_object_t *)acep;
579 
580 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
581 			    sizeof (aceobjp->a_obj_type));
582 			bcopy(aceobjp->a_inherit_obj_type,
583 			    zobjacep->z_inherit_type,
584 			    sizeof (aceobjp->a_inherit_obj_type));
585 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
586 			break;
587 		default:
588 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
589 		}
590 
591 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
592 		    aclp->z_ops.ace_size(aceptr));
593 	}
594 
595 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
596 
597 	return (0);
598 }
599 
600 /*
601  * Copy ZFS ACEs to fixed size ace_t layout
602  */
603 static void
604 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
605     void *datap, int filter)
606 {
607 	uint64_t who;
608 	uint32_t access_mask;
609 	uint16_t iflags, type;
610 	zfs_ace_hdr_t *zacep = NULL;
611 	ace_t *acep = datap;
612 	ace_object_t *objacep;
613 	zfs_object_ace_t *zobjacep;
614 	size_t ace_size;
615 	uint16_t entry_type;
616 
617 	while (zacep = zfs_acl_next_ace(aclp, zacep,
618 	    &who, &access_mask, &iflags, &type)) {
619 
620 		switch (type) {
621 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
622 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
623 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
624 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
625 			if (filter) {
626 				continue;
627 			}
628 			zobjacep = (zfs_object_ace_t *)zacep;
629 			objacep = (ace_object_t *)acep;
630 			bcopy(zobjacep->z_object_type,
631 			    objacep->a_obj_type,
632 			    sizeof (zobjacep->z_object_type));
633 			bcopy(zobjacep->z_inherit_type,
634 			    objacep->a_inherit_obj_type,
635 			    sizeof (zobjacep->z_inherit_type));
636 			ace_size = sizeof (ace_object_t);
637 			break;
638 		default:
639 			ace_size = sizeof (ace_t);
640 			break;
641 		}
642 
643 		entry_type = (iflags & ACE_TYPE_FLAGS);
644 		if ((entry_type != ACE_OWNER &&
645 		    entry_type != OWNING_GROUP &&
646 		    entry_type != ACE_EVERYONE)) {
647 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
648 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
649 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
650 		} else {
651 			acep->a_who = (uid_t)(int64_t)who;
652 		}
653 		acep->a_access_mask = access_mask;
654 		acep->a_flags = iflags;
655 		acep->a_type = type;
656 		acep = (ace_t *)((caddr_t)acep + ace_size);
657 	}
658 }
659 
660 static int
661 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
662     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
663 {
664 	int i;
665 	zfs_oldace_t *aceptr = z_acl;
666 
667 	for (i = 0; i != aclcnt; i++, aceptr++) {
668 		aceptr->z_access_mask = acep[i].a_access_mask;
669 		aceptr->z_type = acep[i].a_type;
670 		aceptr->z_flags = acep[i].a_flags;
671 		aceptr->z_fuid = acep[i].a_who;
672 		/*
673 		 * Make sure ACE is valid
674 		 */
675 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
676 		    aceptr->z_flags) != B_TRUE)
677 			return (EINVAL);
678 	}
679 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
680 	return (0);
681 }
682 
683 /*
684  * convert old ACL format to new
685  */
686 void
687 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
688 {
689 	zfs_oldace_t *oldaclp;
690 	int i;
691 	uint16_t type, iflags;
692 	uint32_t access_mask;
693 	uint64_t who;
694 	void *cookie = NULL;
695 	zfs_acl_node_t *newaclnode;
696 
697 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
698 	/*
699 	 * First create the ACE in a contiguous piece of memory
700 	 * for zfs_copy_ace_2_fuid().
701 	 *
702 	 * We only convert an ACL once, so this won't happen
703 	 * everytime.
704 	 */
705 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
706 	    KM_SLEEP);
707 	i = 0;
708 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
709 	    &access_mask, &iflags, &type)) {
710 		oldaclp[i].z_flags = iflags;
711 		oldaclp[i].z_type = type;
712 		oldaclp[i].z_fuid = who;
713 		oldaclp[i++].z_access_mask = access_mask;
714 	}
715 
716 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
717 	    sizeof (zfs_object_ace_t));
718 	aclp->z_ops = zfs_acl_fuid_ops;
719 	VERIFY(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
720 	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
721 	    &newaclnode->z_size, NULL, cr) == 0);
722 	newaclnode->z_ace_count = aclp->z_acl_count;
723 	aclp->z_version = ZFS_ACL_VERSION;
724 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
725 
726 	/*
727 	 * Release all previous ACL nodes
728 	 */
729 
730 	zfs_acl_release_nodes(aclp);
731 
732 	list_insert_head(&aclp->z_acl, newaclnode);
733 
734 	aclp->z_acl_bytes = newaclnode->z_size;
735 	aclp->z_acl_count = newaclnode->z_ace_count;
736 
737 }
738 
739 /*
740  * Convert unix access mask to v4 access mask
741  */
742 static uint32_t
743 zfs_unix_to_v4(uint32_t access_mask)
744 {
745 	uint32_t new_mask = 0;
746 
747 	if (access_mask & S_IXOTH)
748 		new_mask |= ACE_EXECUTE;
749 	if (access_mask & S_IWOTH)
750 		new_mask |= ACE_WRITE_DATA;
751 	if (access_mask & S_IROTH)
752 		new_mask |= ACE_READ_DATA;
753 	return (new_mask);
754 }
755 
756 static void
757 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
758     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
759 {
760 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
761 
762 	aclp->z_ops.ace_mask_set(acep, access_mask);
763 	aclp->z_ops.ace_type_set(acep, access_type);
764 	aclp->z_ops.ace_flags_set(acep, entry_type);
765 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
766 	    type != ACE_EVERYONE))
767 		aclp->z_ops.ace_who_set(acep, fuid);
768 }
769 
770 /*
771  * Determine mode of file based on ACL.
772  * Also, create FUIDs for any User/Group ACEs
773  */
774 static uint64_t
775 zfs_mode_compute(znode_t *zp, zfs_acl_t *aclp)
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 	return (mode);
911 }
912 
913 static zfs_acl_t *
914 zfs_acl_node_read_internal(znode_t *zp, boolean_t will_modify)
915 {
916 	zfs_acl_t	*aclp;
917 	zfs_acl_node_t	*aclnode;
918 
919 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
920 
921 	/*
922 	 * Version 0 to 1 znode_acl_phys has the size/count fields swapped.
923 	 * Version 0 didn't have a size field, only a count.
924 	 */
925 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
926 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_size;
927 		aclp->z_acl_bytes = ZFS_ACL_SIZE(aclp->z_acl_count);
928 	} else {
929 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_count;
930 		aclp->z_acl_bytes = zp->z_phys->zp_acl.z_acl_size;
931 	}
932 
933 	aclnode = zfs_acl_node_alloc(will_modify ? aclp->z_acl_bytes : 0);
934 	aclnode->z_ace_count = aclp->z_acl_count;
935 	if (will_modify) {
936 		bcopy(zp->z_phys->zp_acl.z_ace_data, aclnode->z_acldata,
937 		    aclp->z_acl_bytes);
938 	} else {
939 		aclnode->z_size = aclp->z_acl_bytes;
940 		aclnode->z_acldata = &zp->z_phys->zp_acl.z_ace_data[0];
941 	}
942 
943 	list_insert_head(&aclp->z_acl, aclnode);
944 
945 	return (aclp);
946 }
947 
948 /*
949  * Read an external acl object.
950  */
951 static int
952 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
953 {
954 	uint64_t extacl = zp->z_phys->zp_acl.z_acl_extern_obj;
955 	zfs_acl_t	*aclp;
956 	size_t		aclsize;
957 	size_t		acl_count;
958 	zfs_acl_node_t	*aclnode;
959 	int error;
960 
961 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
962 
963 	if (zp->z_phys->zp_acl.z_acl_extern_obj == 0) {
964 		*aclpp = zfs_acl_node_read_internal(zp, will_modify);
965 		return (0);
966 	}
967 
968 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
969 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
970 		zfs_acl_phys_v0_t *zacl0 =
971 		    (zfs_acl_phys_v0_t *)&zp->z_phys->zp_acl;
972 
973 		aclsize = ZFS_ACL_SIZE(zacl0->z_acl_count);
974 		acl_count = zacl0->z_acl_count;
975 	} else {
976 		aclsize = zp->z_phys->zp_acl.z_acl_size;
977 		acl_count = zp->z_phys->zp_acl.z_acl_count;
978 		if (aclsize == 0)
979 			aclsize = acl_count * sizeof (zfs_ace_t);
980 	}
981 	aclnode = zfs_acl_node_alloc(aclsize);
982 	list_insert_head(&aclp->z_acl, aclnode);
983 	error = dmu_read(zp->z_zfsvfs->z_os, extacl, 0,
984 	    aclsize, aclnode->z_acldata, DMU_READ_PREFETCH);
985 	aclnode->z_ace_count = acl_count;
986 	aclp->z_acl_count = acl_count;
987 	aclp->z_acl_bytes = aclsize;
988 
989 	if (error != 0) {
990 		zfs_acl_free(aclp);
991 		/* convert checksum errors into IO errors */
992 		if (error == ECKSUM)
993 			error = EIO;
994 		return (error);
995 	}
996 
997 	*aclpp = aclp;
998 	return (0);
999 }
1000 
1001 /*
1002  * common code for setting ACLs.
1003  *
1004  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1005  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1006  * already checked the acl and knows whether to inherit.
1007  */
1008 int
1009 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1010 {
1011 	int		error;
1012 	znode_phys_t	*zphys = zp->z_phys;
1013 	zfs_acl_phys_t	*zacl = &zphys->zp_acl;
1014 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1015 	uint64_t	aoid = zphys->zp_acl.z_acl_extern_obj;
1016 	uint64_t	off = 0;
1017 	dmu_object_type_t otype;
1018 	zfs_acl_node_t	*aclnode;
1019 
1020 	dmu_buf_will_dirty(zp->z_dbuf, tx);
1021 
1022 	zphys->zp_mode = zfs_mode_compute(zp, aclp);
1023 
1024 	/*
1025 	 * Decide which opbject type to use.  If we are forced to
1026 	 * use old ACL format than transform ACL into zfs_oldace_t
1027 	 * layout.
1028 	 */
1029 	if (!zfsvfs->z_use_fuids) {
1030 		otype = DMU_OT_OLDACL;
1031 	} else {
1032 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1033 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1034 			zfs_acl_xform(zp, aclp, cr);
1035 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1036 		otype = DMU_OT_ACL;
1037 	}
1038 
1039 	if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1040 		/*
1041 		 * If ACL was previously external and we are now
1042 		 * converting to new ACL format then release old
1043 		 * ACL object and create a new one.
1044 		 */
1045 		if (aoid && aclp->z_version != zacl->z_acl_version) {
1046 			error = dmu_object_free(zfsvfs->z_os,
1047 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1048 			if (error)
1049 				return (error);
1050 			aoid = 0;
1051 		}
1052 		if (aoid == 0) {
1053 			aoid = dmu_object_alloc(zfsvfs->z_os,
1054 			    otype, aclp->z_acl_bytes,
1055 			    otype == DMU_OT_ACL ? DMU_OT_SYSACL : DMU_OT_NONE,
1056 			    otype == DMU_OT_ACL ? DN_MAX_BONUSLEN : 0, tx);
1057 		} else {
1058 			(void) dmu_object_set_blocksize(zfsvfs->z_os, aoid,
1059 			    aclp->z_acl_bytes, 0, tx);
1060 		}
1061 		zphys->zp_acl.z_acl_extern_obj = aoid;
1062 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1063 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1064 			if (aclnode->z_ace_count == 0)
1065 				continue;
1066 			dmu_write(zfsvfs->z_os, aoid, off,
1067 			    aclnode->z_size, aclnode->z_acldata, tx);
1068 			off += aclnode->z_size;
1069 		}
1070 	} else {
1071 		void *start = zacl->z_ace_data;
1072 		/*
1073 		 * Migrating back embedded?
1074 		 */
1075 		if (zphys->zp_acl.z_acl_extern_obj) {
1076 			error = dmu_object_free(zfsvfs->z_os,
1077 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1078 			if (error)
1079 				return (error);
1080 			zphys->zp_acl.z_acl_extern_obj = 0;
1081 		}
1082 
1083 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1084 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1085 			if (aclnode->z_ace_count == 0)
1086 				continue;
1087 			bcopy(aclnode->z_acldata, start, aclnode->z_size);
1088 			start = (caddr_t)start + aclnode->z_size;
1089 		}
1090 	}
1091 
1092 	/*
1093 	 * If Old version then swap count/bytes to match old
1094 	 * layout of znode_acl_phys_t.
1095 	 */
1096 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1097 		zphys->zp_acl.z_acl_size = aclp->z_acl_count;
1098 		zphys->zp_acl.z_acl_count = aclp->z_acl_bytes;
1099 	} else {
1100 		zphys->zp_acl.z_acl_size = aclp->z_acl_bytes;
1101 		zphys->zp_acl.z_acl_count = aclp->z_acl_count;
1102 	}
1103 
1104 	zphys->zp_acl.z_acl_version = aclp->z_version;
1105 
1106 	/*
1107 	 * Replace ACL wide bits, but first clear them.
1108 	 */
1109 	zp->z_phys->zp_flags &= ~ZFS_ACL_WIDE_FLAGS;
1110 
1111 	zp->z_phys->zp_flags |= aclp->z_hints;
1112 
1113 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1114 		zp->z_phys->zp_flags |= ZFS_ACL_TRIVIAL;
1115 
1116 	return (0);
1117 }
1118 
1119 /*
1120  * Update access mask for prepended ACE
1121  *
1122  * This applies the "groupmask" value for aclmode property.
1123  */
1124 static void
1125 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1126     mode_t mode, uint64_t owner)
1127 {
1128 	int	rmask, wmask, xmask;
1129 	int	user_ace;
1130 	uint16_t aceflags;
1131 	uint32_t origmask, acepmask;
1132 	uint64_t fuid;
1133 
1134 	aceflags = aclp->z_ops.ace_flags_get(acep);
1135 	fuid = aclp->z_ops.ace_who_get(acep);
1136 	origmask = aclp->z_ops.ace_mask_get(origacep);
1137 	acepmask = aclp->z_ops.ace_mask_get(acep);
1138 
1139 	user_ace = (!(aceflags &
1140 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1141 
1142 	if (user_ace && (fuid == owner)) {
1143 		rmask = S_IRUSR;
1144 		wmask = S_IWUSR;
1145 		xmask = S_IXUSR;
1146 	} else {
1147 		rmask = S_IRGRP;
1148 		wmask = S_IWGRP;
1149 		xmask = S_IXGRP;
1150 	}
1151 
1152 	if (origmask & ACE_READ_DATA) {
1153 		if (mode & rmask) {
1154 			acepmask &= ~ACE_READ_DATA;
1155 		} else {
1156 			acepmask |= ACE_READ_DATA;
1157 		}
1158 	}
1159 
1160 	if (origmask & ACE_WRITE_DATA) {
1161 		if (mode & wmask) {
1162 			acepmask &= ~ACE_WRITE_DATA;
1163 		} else {
1164 			acepmask |= ACE_WRITE_DATA;
1165 		}
1166 	}
1167 
1168 	if (origmask & ACE_APPEND_DATA) {
1169 		if (mode & wmask) {
1170 			acepmask &= ~ACE_APPEND_DATA;
1171 		} else {
1172 			acepmask |= ACE_APPEND_DATA;
1173 		}
1174 	}
1175 
1176 	if (origmask & ACE_EXECUTE) {
1177 		if (mode & xmask) {
1178 			acepmask &= ~ACE_EXECUTE;
1179 		} else {
1180 			acepmask |= ACE_EXECUTE;
1181 		}
1182 	}
1183 	aclp->z_ops.ace_mask_set(acep, acepmask);
1184 }
1185 
1186 /*
1187  * Apply mode to canonical six ACEs.
1188  */
1189 static void
1190 zfs_acl_fixup_canonical_six(zfs_acl_t *aclp, mode_t mode)
1191 {
1192 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1193 	void	*acep;
1194 	int	maskoff = aclp->z_ops.ace_mask_off();
1195 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1196 
1197 	ASSERT(aclnode != NULL);
1198 
1199 	acep = (void *)((caddr_t)aclnode->z_acldata +
1200 	    aclnode->z_size - (abstract_size * 6));
1201 
1202 	/*
1203 	 * Fixup final ACEs to match the mode
1204 	 */
1205 
1206 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1207 	    (mode & 0700) >> 6);	/* owner@ */
1208 
1209 	acep = (caddr_t)acep + (abstract_size * 2);
1210 
1211 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1212 	    (mode & 0070) >> 3);	/* group@ */
1213 
1214 	acep = (caddr_t)acep + (abstract_size * 2);
1215 	adjust_ace_pair_common(acep, maskoff,
1216 	    abstract_size, mode);	/* everyone@ */
1217 }
1218 
1219 
1220 static int
1221 zfs_acl_ace_match(zfs_acl_t *aclp, void *acep, int allow_deny,
1222     int entry_type, int accessmask)
1223 {
1224 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1225 	uint16_t type = aclp->z_ops.ace_type_get(acep);
1226 	uint16_t flags = aclp->z_ops.ace_flags_get(acep);
1227 
1228 	return (mask == accessmask && type == allow_deny &&
1229 	    ((flags & ACE_TYPE_FLAGS) == entry_type));
1230 }
1231 
1232 /*
1233  * Can prepended ACE be reused?
1234  */
1235 static int
1236 zfs_reuse_deny(zfs_acl_t *aclp, void *acep, void *prevacep)
1237 {
1238 	int okay_masks;
1239 	uint16_t prevtype;
1240 	uint16_t prevflags;
1241 	uint16_t flags;
1242 	uint32_t mask, prevmask;
1243 
1244 	if (prevacep == NULL)
1245 		return (B_FALSE);
1246 
1247 	prevtype = aclp->z_ops.ace_type_get(prevacep);
1248 	prevflags = aclp->z_ops.ace_flags_get(prevacep);
1249 	flags = aclp->z_ops.ace_flags_get(acep);
1250 	mask = aclp->z_ops.ace_mask_get(acep);
1251 	prevmask = aclp->z_ops.ace_mask_get(prevacep);
1252 
1253 	if (prevtype != DENY)
1254 		return (B_FALSE);
1255 
1256 	if (prevflags != (flags & ACE_IDENTIFIER_GROUP))
1257 		return (B_FALSE);
1258 
1259 	okay_masks = (mask & OKAY_MASK_BITS);
1260 
1261 	if (prevmask & ~okay_masks)
1262 		return (B_FALSE);
1263 
1264 	return (B_TRUE);
1265 }
1266 
1267 
1268 /*
1269  * Insert new ACL node into chain of zfs_acl_node_t's
1270  *
1271  * This will result in two possible results.
1272  * 1. If the ACL is currently just a single zfs_acl_node and
1273  *    we are prepending the entry then current acl node will have
1274  *    a new node inserted above it.
1275  *
1276  * 2. If we are inserting in the middle of current acl node then
1277  *    the current node will be split in two and new node will be inserted
1278  *    in between the two split nodes.
1279  */
1280 static zfs_acl_node_t *
1281 zfs_acl_ace_insert(zfs_acl_t *aclp, void  *acep)
1282 {
1283 	zfs_acl_node_t 	*newnode;
1284 	zfs_acl_node_t 	*trailernode = NULL;
1285 	zfs_acl_node_t 	*currnode = zfs_acl_curr_node(aclp);
1286 	int		curr_idx = aclp->z_curr_node->z_ace_idx;
1287 	int		trailer_count;
1288 	size_t		oldsize;
1289 
1290 	newnode = zfs_acl_node_alloc(aclp->z_ops.ace_size(acep));
1291 	newnode->z_ace_count = 1;
1292 
1293 	oldsize = currnode->z_size;
1294 
1295 	if (curr_idx != 1) {
1296 		trailernode = zfs_acl_node_alloc(0);
1297 		trailernode->z_acldata = acep;
1298 
1299 		trailer_count = currnode->z_ace_count - curr_idx + 1;
1300 		currnode->z_ace_count = curr_idx - 1;
1301 		currnode->z_size = (caddr_t)acep - (caddr_t)currnode->z_acldata;
1302 		trailernode->z_size = oldsize - currnode->z_size;
1303 		trailernode->z_ace_count = trailer_count;
1304 	}
1305 
1306 	aclp->z_acl_count += 1;
1307 	aclp->z_acl_bytes += aclp->z_ops.ace_size(acep);
1308 
1309 	if (curr_idx == 1)
1310 		list_insert_before(&aclp->z_acl, currnode, newnode);
1311 	else
1312 		list_insert_after(&aclp->z_acl, currnode, newnode);
1313 	if (trailernode) {
1314 		list_insert_after(&aclp->z_acl, newnode, trailernode);
1315 		aclp->z_curr_node = trailernode;
1316 		trailernode->z_ace_idx = 1;
1317 	}
1318 
1319 	return (newnode);
1320 }
1321 
1322 /*
1323  * Prepend deny ACE
1324  */
1325 static void *
1326 zfs_acl_prepend_deny(uint64_t uid, zfs_acl_t *aclp, void *acep,
1327     mode_t mode)
1328 {
1329 	zfs_acl_node_t *aclnode;
1330 	void  *newacep;
1331 	uint64_t fuid;
1332 	uint16_t flags;
1333 
1334 	aclnode = zfs_acl_ace_insert(aclp, acep);
1335 	newacep = aclnode->z_acldata;
1336 	fuid = aclp->z_ops.ace_who_get(acep);
1337 	flags = aclp->z_ops.ace_flags_get(acep);
1338 	zfs_set_ace(aclp, newacep, 0, DENY, fuid, (flags & ACE_TYPE_FLAGS));
1339 	zfs_acl_prepend_fixup(aclp, newacep, acep, mode, uid);
1340 
1341 	return (newacep);
1342 }
1343 
1344 /*
1345  * Split an inherited ACE into inherit_only ACE
1346  * and original ACE with inheritance flags stripped off.
1347  */
1348 static void
1349 zfs_acl_split_ace(zfs_acl_t *aclp, zfs_ace_hdr_t *acep)
1350 {
1351 	zfs_acl_node_t *aclnode;
1352 	zfs_acl_node_t *currnode;
1353 	void  *newacep;
1354 	uint16_t type, flags;
1355 	uint32_t mask;
1356 	uint64_t fuid;
1357 
1358 	type = aclp->z_ops.ace_type_get(acep);
1359 	flags = aclp->z_ops.ace_flags_get(acep);
1360 	mask = aclp->z_ops.ace_mask_get(acep);
1361 	fuid = aclp->z_ops.ace_who_get(acep);
1362 
1363 	aclnode = zfs_acl_ace_insert(aclp, acep);
1364 	newacep = aclnode->z_acldata;
1365 
1366 	aclp->z_ops.ace_type_set(newacep, type);
1367 	aclp->z_ops.ace_flags_set(newacep, flags | ACE_INHERIT_ONLY_ACE);
1368 	aclp->z_ops.ace_mask_set(newacep, mask);
1369 	aclp->z_ops.ace_type_set(newacep, type);
1370 	aclp->z_ops.ace_who_set(newacep, fuid);
1371 	aclp->z_next_ace = acep;
1372 	flags &= ~ALL_INHERIT;
1373 	aclp->z_ops.ace_flags_set(acep, flags);
1374 	currnode = zfs_acl_curr_node(aclp);
1375 	ASSERT(currnode->z_ace_idx >= 1);
1376 	currnode->z_ace_idx -= 1;
1377 }
1378 
1379 /*
1380  * Are ACES started at index i, the canonical six ACES?
1381  */
1382 static int
1383 zfs_have_canonical_six(zfs_acl_t *aclp)
1384 {
1385 	void *acep;
1386 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1387 	int		i = 0;
1388 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1389 
1390 	ASSERT(aclnode != NULL);
1391 
1392 	if (aclnode->z_ace_count < 6)
1393 		return (0);
1394 
1395 	acep = (void *)((caddr_t)aclnode->z_acldata +
1396 	    aclnode->z_size - (aclp->z_ops.ace_abstract_size() * 6));
1397 
1398 	if ((zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1399 	    DENY, ACE_OWNER, 0) &&
1400 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1401 	    ALLOW, ACE_OWNER, OWNER_ALLOW_MASK) &&
1402 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++), DENY,
1403 	    OWNING_GROUP, 0) && zfs_acl_ace_match(aclp, (caddr_t)acep +
1404 	    (abstract_size * i++),
1405 	    ALLOW, OWNING_GROUP, 0) &&
1406 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1407 	    DENY, ACE_EVERYONE, EVERYONE_DENY_MASK) &&
1408 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1409 	    ALLOW, ACE_EVERYONE, EVERYONE_ALLOW_MASK))) {
1410 		return (1);
1411 	} else {
1412 		return (0);
1413 	}
1414 }
1415 
1416 
1417 /*
1418  * Apply step 1g, to group entries
1419  *
1420  * Need to deal with corner case where group may have
1421  * greater permissions than owner.  If so then limit
1422  * group permissions, based on what extra permissions
1423  * group has.
1424  */
1425 static void
1426 zfs_fixup_group_entries(zfs_acl_t *aclp, void *acep, void *prevacep,
1427     mode_t mode)
1428 {
1429 	uint32_t prevmask = aclp->z_ops.ace_mask_get(prevacep);
1430 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1431 	uint16_t prevflags = aclp->z_ops.ace_flags_get(prevacep);
1432 	mode_t extramode = (mode >> 3) & 07;
1433 	mode_t ownermode = (mode >> 6);
1434 
1435 	if (prevflags & ACE_IDENTIFIER_GROUP) {
1436 
1437 		extramode &= ~ownermode;
1438 
1439 		if (extramode) {
1440 			if (extramode & S_IROTH) {
1441 				prevmask &= ~ACE_READ_DATA;
1442 				mask &= ~ACE_READ_DATA;
1443 			}
1444 			if (extramode & S_IWOTH) {
1445 				prevmask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1446 				mask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1447 			}
1448 			if (extramode & S_IXOTH) {
1449 				prevmask  &= ~ACE_EXECUTE;
1450 				mask &= ~ACE_EXECUTE;
1451 			}
1452 		}
1453 	}
1454 	aclp->z_ops.ace_mask_set(acep, mask);
1455 	aclp->z_ops.ace_mask_set(prevacep, prevmask);
1456 }
1457 
1458 /*
1459  * Apply the chmod algorithm as described
1460  * in PSARC/2002/240
1461  */
1462 static void
1463 zfs_acl_chmod(zfsvfs_t *zfsvfs, uint64_t uid,
1464     uint64_t mode, zfs_acl_t *aclp)
1465 {
1466 	void		*acep = NULL, *prevacep = NULL;
1467 	uint64_t	who;
1468 	int 		i;
1469 	int 		entry_type;
1470 	int 		reuse_deny;
1471 	int 		need_canonical_six = 1;
1472 	uint16_t	iflags, type;
1473 	uint32_t	access_mask;
1474 
1475 	/*
1476 	 * If discard then just discard all ACL nodes which
1477 	 * represent the ACEs.
1478 	 *
1479 	 * New owner@/group@/everone@ ACEs will be added
1480 	 * later.
1481 	 */
1482 	if (zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1483 		zfs_acl_release_nodes(aclp);
1484 
1485 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1486 	    &iflags, &type)) {
1487 
1488 		entry_type = (iflags & ACE_TYPE_FLAGS);
1489 		iflags = (iflags & ALL_INHERIT);
1490 
1491 		if ((type != ALLOW && type != DENY) ||
1492 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1493 			if (iflags)
1494 				aclp->z_hints |= ZFS_INHERIT_ACE;
1495 			switch (type) {
1496 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1497 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1498 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1499 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1500 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1501 				break;
1502 			}
1503 			goto nextace;
1504 		}
1505 
1506 		/*
1507 		 * Need to split ace into two?
1508 		 */
1509 		if ((iflags & (ACE_FILE_INHERIT_ACE|
1510 		    ACE_DIRECTORY_INHERIT_ACE)) &&
1511 		    (!(iflags & ACE_INHERIT_ONLY_ACE))) {
1512 			zfs_acl_split_ace(aclp, acep);
1513 			aclp->z_hints |= ZFS_INHERIT_ACE;
1514 			goto nextace;
1515 		}
1516 
1517 		if (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1518 		    (entry_type == OWNING_GROUP)) {
1519 			access_mask &= ~OGE_CLEAR;
1520 			aclp->z_ops.ace_mask_set(acep, access_mask);
1521 			goto nextace;
1522 		} else {
1523 			reuse_deny = B_TRUE;
1524 			if (type == ALLOW) {
1525 
1526 				/*
1527 				 * Check preceding ACE if any, to see
1528 				 * if we need to prepend a DENY ACE.
1529 				 * This is only applicable when the acl_mode
1530 				 * property == groupmask.
1531 				 */
1532 				if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK) {
1533 
1534 					reuse_deny = zfs_reuse_deny(aclp, acep,
1535 					    prevacep);
1536 
1537 					if (!reuse_deny) {
1538 						prevacep =
1539 						    zfs_acl_prepend_deny(uid,
1540 						    aclp, acep, mode);
1541 					} else {
1542 						zfs_acl_prepend_fixup(
1543 						    aclp, prevacep,
1544 						    acep, mode, uid);
1545 					}
1546 					zfs_fixup_group_entries(aclp, acep,
1547 					    prevacep, mode);
1548 				}
1549 			}
1550 		}
1551 nextace:
1552 		prevacep = acep;
1553 	}
1554 
1555 	/*
1556 	 * Check out last six aces, if we have six.
1557 	 */
1558 
1559 	if (aclp->z_acl_count >= 6) {
1560 		if (zfs_have_canonical_six(aclp)) {
1561 			need_canonical_six = 0;
1562 		}
1563 	}
1564 
1565 	if (need_canonical_six) {
1566 		size_t abstract_size = aclp->z_ops.ace_abstract_size();
1567 		void *zacep;
1568 		zfs_acl_node_t *aclnode =
1569 		    zfs_acl_node_alloc(abstract_size * 6);
1570 
1571 		aclnode->z_size = abstract_size * 6;
1572 		aclnode->z_ace_count = 6;
1573 		aclp->z_acl_bytes += aclnode->z_size;
1574 		list_insert_tail(&aclp->z_acl, aclnode);
1575 
1576 		zacep = aclnode->z_acldata;
1577 
1578 		i = 0;
1579 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1580 		    0, DENY, -1, ACE_OWNER);
1581 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1582 		    OWNER_ALLOW_MASK, ALLOW, -1, ACE_OWNER);
1583 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1584 		    DENY, -1, OWNING_GROUP);
1585 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1586 		    ALLOW, -1, OWNING_GROUP);
1587 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1588 		    EVERYONE_DENY_MASK, DENY, -1, ACE_EVERYONE);
1589 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1590 		    EVERYONE_ALLOW_MASK, ALLOW, -1, ACE_EVERYONE);
1591 		aclp->z_acl_count += 6;
1592 	}
1593 
1594 	zfs_acl_fixup_canonical_six(aclp, mode);
1595 }
1596 
1597 int
1598 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1599 {
1600 	int error;
1601 
1602 	mutex_enter(&zp->z_lock);
1603 	mutex_enter(&zp->z_acl_lock);
1604 	*aclp = NULL;
1605 	error = zfs_acl_node_read(zp, aclp, B_TRUE);
1606 	if (error == 0) {
1607 		(*aclp)->z_hints = zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS;
1608 		zfs_acl_chmod(zp->z_zfsvfs, zp->z_phys->zp_uid, mode, *aclp);
1609 	}
1610 	mutex_exit(&zp->z_acl_lock);
1611 	mutex_exit(&zp->z_lock);
1612 	return (error);
1613 }
1614 
1615 /*
1616  * strip off write_owner and write_acl
1617  */
1618 static void
1619 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1620 {
1621 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1622 
1623 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1624 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1625 		mask &= ~RESTRICTED_CLEAR;
1626 		aclp->z_ops.ace_mask_set(acep, mask);
1627 	}
1628 }
1629 
1630 /*
1631  * Should ACE be inherited?
1632  */
1633 static int
1634 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1635 {
1636 	int	iflags = (acep_flags & 0xf);
1637 
1638 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1639 		return (1);
1640 	else if (iflags & ACE_FILE_INHERIT_ACE)
1641 		return (!((vtype == VDIR) &&
1642 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1643 	return (0);
1644 }
1645 
1646 /*
1647  * inherit inheritable ACEs from parent
1648  */
1649 static zfs_acl_t *
1650 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1651     uint64_t mode, boolean_t *need_chmod)
1652 {
1653 	void		*pacep;
1654 	void		*acep, *acep2;
1655 	zfs_acl_node_t  *aclnode, *aclnode2;
1656 	zfs_acl_t	*aclp = NULL;
1657 	uint64_t	who;
1658 	uint32_t	access_mask;
1659 	uint16_t	iflags, newflags, type;
1660 	size_t		ace_size;
1661 	void		*data1, *data2;
1662 	size_t		data1sz, data2sz;
1663 	boolean_t	vdir = vtype == VDIR;
1664 	boolean_t	vreg = vtype == VREG;
1665 	boolean_t	passthrough, passthrough_x, noallow;
1666 
1667 	passthrough_x =
1668 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1669 	passthrough = passthrough_x ||
1670 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1671 	noallow =
1672 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1673 
1674 	*need_chmod = B_TRUE;
1675 	pacep = NULL;
1676 	aclp = zfs_acl_alloc(paclp->z_version);
1677 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD)
1678 		return (aclp);
1679 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1680 	    &access_mask, &iflags, &type)) {
1681 
1682 		/*
1683 		 * don't inherit bogus ACEs
1684 		 */
1685 		if (!zfs_acl_valid_ace_type(type, iflags))
1686 			continue;
1687 
1688 		if (noallow && type == ALLOW)
1689 			continue;
1690 
1691 		ace_size = aclp->z_ops.ace_size(pacep);
1692 
1693 		if (!zfs_ace_can_use(vtype, iflags))
1694 			continue;
1695 
1696 		/*
1697 		 * If owner@, group@, or everyone@ inheritable
1698 		 * then zfs_acl_chmod() isn't needed.
1699 		 */
1700 		if (passthrough &&
1701 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1702 		    ((iflags & OWNING_GROUP) ==
1703 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1704 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1705 			*need_chmod = B_FALSE;
1706 
1707 			if (!vdir && passthrough_x &&
1708 			    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1709 				access_mask &= ~ACE_EXECUTE;
1710 			}
1711 		}
1712 
1713 		aclnode = zfs_acl_node_alloc(ace_size);
1714 		list_insert_tail(&aclp->z_acl, aclnode);
1715 		acep = aclnode->z_acldata;
1716 
1717 		zfs_set_ace(aclp, acep, access_mask, type,
1718 		    who, iflags|ACE_INHERITED_ACE);
1719 
1720 		/*
1721 		 * Copy special opaque data if any
1722 		 */
1723 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1724 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1725 			    &data2)) == data1sz);
1726 			bcopy(data1, data2, data2sz);
1727 		}
1728 		aclp->z_acl_count++;
1729 		aclnode->z_ace_count++;
1730 		aclp->z_acl_bytes += aclnode->z_size;
1731 		newflags = aclp->z_ops.ace_flags_get(acep);
1732 
1733 		if (vdir)
1734 			aclp->z_hints |= ZFS_INHERIT_ACE;
1735 
1736 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1737 			newflags &= ~ALL_INHERIT;
1738 			aclp->z_ops.ace_flags_set(acep,
1739 			    newflags|ACE_INHERITED_ACE);
1740 			zfs_restricted_update(zfsvfs, aclp, acep);
1741 			continue;
1742 		}
1743 
1744 		ASSERT(vdir);
1745 
1746 		newflags = aclp->z_ops.ace_flags_get(acep);
1747 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1748 		    ACE_DIRECTORY_INHERIT_ACE)) !=
1749 		    ACE_FILE_INHERIT_ACE) {
1750 			aclnode2 = zfs_acl_node_alloc(ace_size);
1751 			list_insert_tail(&aclp->z_acl, aclnode2);
1752 			acep2 = aclnode2->z_acldata;
1753 			zfs_set_ace(aclp, acep2,
1754 			    access_mask, type, who,
1755 			    iflags|ACE_INHERITED_ACE);
1756 			newflags |= ACE_INHERIT_ONLY_ACE;
1757 			aclp->z_ops.ace_flags_set(acep, newflags);
1758 			newflags &= ~ALL_INHERIT;
1759 			aclp->z_ops.ace_flags_set(acep2,
1760 			    newflags|ACE_INHERITED_ACE);
1761 
1762 			/*
1763 			 * Copy special opaque data if any
1764 			 */
1765 			if ((data1sz = aclp->z_ops.ace_data(acep,
1766 			    &data1)) != 0) {
1767 				VERIFY((data2sz =
1768 				    aclp->z_ops.ace_data(acep2,
1769 				    &data2)) == data1sz);
1770 				bcopy(data1, data2, data1sz);
1771 			}
1772 			aclp->z_acl_count++;
1773 			aclnode2->z_ace_count++;
1774 			aclp->z_acl_bytes += aclnode->z_size;
1775 			zfs_restricted_update(zfsvfs, aclp, acep2);
1776 		} else {
1777 			newflags |= ACE_INHERIT_ONLY_ACE;
1778 			aclp->z_ops.ace_flags_set(acep,
1779 			    newflags|ACE_INHERITED_ACE);
1780 		}
1781 	}
1782 	return (aclp);
1783 }
1784 
1785 /*
1786  * Create file system object initial permissions
1787  * including inheritable ACEs.
1788  */
1789 int
1790 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1791     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1792 {
1793 	int		error;
1794 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1795 	zfs_acl_t	*paclp;
1796 	gid_t		gid;
1797 	boolean_t	need_chmod = B_TRUE;
1798 
1799 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1800 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1801 
1802 	if (vsecp)
1803 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1804 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1805 			return (error);
1806 
1807 	/*
1808 	 * Determine uid and gid.
1809 	 */
1810 	if ((flag & (IS_ROOT_NODE | IS_REPLAY)) ||
1811 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1812 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1813 		    (uint64_t)vap->va_uid, cr,
1814 		    ZFS_OWNER, &acl_ids->z_fuidp);
1815 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1816 		    (uint64_t)vap->va_gid, cr,
1817 		    ZFS_GROUP, &acl_ids->z_fuidp);
1818 		gid = vap->va_gid;
1819 	} else {
1820 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1821 		    cr, &acl_ids->z_fuidp);
1822 		acl_ids->z_fgid = 0;
1823 		if (vap->va_mask & AT_GID)  {
1824 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1825 			    (uint64_t)vap->va_gid,
1826 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1827 			gid = vap->va_gid;
1828 			if (acl_ids->z_fgid != dzp->z_phys->zp_gid &&
1829 			    !groupmember(vap->va_gid, cr) &&
1830 			    secpolicy_vnode_create_gid(cr) != 0)
1831 				acl_ids->z_fgid = 0;
1832 		}
1833 		if (acl_ids->z_fgid == 0) {
1834 			if (dzp->z_phys->zp_mode & S_ISGID) {
1835 				acl_ids->z_fgid = dzp->z_phys->zp_gid;
1836 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1837 				    cr, ZFS_GROUP);
1838 			} else {
1839 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1840 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1841 				gid = crgetgid(cr);
1842 			}
1843 		}
1844 	}
1845 
1846 	/*
1847 	 * If we're creating a directory, and the parent directory has the
1848 	 * set-GID bit set, set in on the new directory.
1849 	 * Otherwise, if the user is neither privileged nor a member of the
1850 	 * file's new group, clear the file's set-GID bit.
1851 	 */
1852 
1853 	if (!(flag & IS_ROOT_NODE) && (dzp->z_phys->zp_mode & S_ISGID) &&
1854 	    (vap->va_type == VDIR)) {
1855 		acl_ids->z_mode |= S_ISGID;
1856 	} else {
1857 		if ((acl_ids->z_mode & S_ISGID) &&
1858 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1859 			acl_ids->z_mode &= ~S_ISGID;
1860 	}
1861 
1862 	if (acl_ids->z_aclp == NULL) {
1863 		mutex_enter(&dzp->z_lock);
1864 		if (!(flag & IS_ROOT_NODE) && (ZTOV(dzp)->v_type == VDIR &&
1865 		    (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)) &&
1866 		    !(dzp->z_phys->zp_flags & ZFS_XATTR)) {
1867 			mutex_enter(&dzp->z_acl_lock);
1868 			VERIFY(0 == zfs_acl_node_read(dzp, &paclp, B_FALSE));
1869 			mutex_exit(&dzp->z_acl_lock);
1870 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1871 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1872 			zfs_acl_free(paclp);
1873 		} else {
1874 			acl_ids->z_aclp =
1875 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1876 		}
1877 		mutex_exit(&dzp->z_lock);
1878 		if (need_chmod) {
1879 			acl_ids->z_aclp->z_hints = (vap->va_type == VDIR) ?
1880 			    ZFS_ACL_AUTO_INHERIT : 0;
1881 			zfs_acl_chmod(zfsvfs, acl_ids->z_fuid,
1882 			    acl_ids->z_mode, acl_ids->z_aclp);
1883 		}
1884 	}
1885 
1886 	return (0);
1887 }
1888 
1889 /*
1890  * Free ACL and fuid_infop, but not the acl_ids structure
1891  */
1892 void
1893 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1894 {
1895 	if (acl_ids->z_aclp)
1896 		zfs_acl_free(acl_ids->z_aclp);
1897 	if (acl_ids->z_fuidp)
1898 		zfs_fuid_info_free(acl_ids->z_fuidp);
1899 	acl_ids->z_aclp = NULL;
1900 	acl_ids->z_fuidp = NULL;
1901 }
1902 
1903 boolean_t
1904 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1905 {
1906 	return (zfs_usergroup_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1907 	    zfs_usergroup_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1908 }
1909 
1910 /*
1911  * Retrieve a files ACL
1912  */
1913 int
1914 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1915 {
1916 	zfs_acl_t	*aclp;
1917 	ulong_t		mask;
1918 	int		error;
1919 	int 		count = 0;
1920 	int		largeace = 0;
1921 
1922 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1923 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1924 
1925 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1926 		return (error);
1927 
1928 	if (mask == 0)
1929 		return (ENOSYS);
1930 
1931 	mutex_enter(&zp->z_acl_lock);
1932 
1933 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
1934 	if (error != 0) {
1935 		mutex_exit(&zp->z_acl_lock);
1936 		return (error);
1937 	}
1938 
1939 	/*
1940 	 * Scan ACL to determine number of ACEs
1941 	 */
1942 	if ((zp->z_phys->zp_flags & ZFS_ACL_OBJ_ACE) &&
1943 	    !(mask & VSA_ACE_ALLTYPES)) {
1944 		void *zacep = NULL;
1945 		uint64_t who;
1946 		uint32_t access_mask;
1947 		uint16_t type, iflags;
1948 
1949 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1950 		    &who, &access_mask, &iflags, &type)) {
1951 			switch (type) {
1952 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1953 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1954 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1955 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1956 				largeace++;
1957 				continue;
1958 			default:
1959 				count++;
1960 			}
1961 		}
1962 		vsecp->vsa_aclcnt = count;
1963 	} else
1964 		count = aclp->z_acl_count;
1965 
1966 	if (mask & VSA_ACECNT) {
1967 		vsecp->vsa_aclcnt = count;
1968 	}
1969 
1970 	if (mask & VSA_ACE) {
1971 		size_t aclsz;
1972 
1973 		zfs_acl_node_t *aclnode = list_head(&aclp->z_acl);
1974 
1975 		aclsz = count * sizeof (ace_t) +
1976 		    sizeof (ace_object_t) * largeace;
1977 
1978 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1979 		vsecp->vsa_aclentsz = aclsz;
1980 
1981 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1982 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1983 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1984 		else {
1985 			bcopy(aclnode->z_acldata, vsecp->vsa_aclentp,
1986 			    count * sizeof (ace_t));
1987 		}
1988 	}
1989 	if (mask & VSA_ACE_ACLFLAGS) {
1990 		vsecp->vsa_aclflags = 0;
1991 		if (zp->z_phys->zp_flags & ZFS_ACL_DEFAULTED)
1992 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1993 		if (zp->z_phys->zp_flags & ZFS_ACL_PROTECTED)
1994 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1995 		if (zp->z_phys->zp_flags & ZFS_ACL_AUTO_INHERIT)
1996 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1997 	}
1998 
1999 	mutex_exit(&zp->z_acl_lock);
2000 
2001 	zfs_acl_free(aclp);
2002 
2003 	return (0);
2004 }
2005 
2006 int
2007 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
2008     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2009 {
2010 	zfs_acl_t *aclp;
2011 	zfs_acl_node_t *aclnode;
2012 	int aclcnt = vsecp->vsa_aclcnt;
2013 	int error;
2014 
2015 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2016 		return (EINVAL);
2017 
2018 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2019 
2020 	aclp->z_hints = 0;
2021 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2022 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2023 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
2024 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2025 		    aclcnt, &aclnode->z_size)) != 0) {
2026 			zfs_acl_free(aclp);
2027 			zfs_acl_node_free(aclnode);
2028 			return (error);
2029 		}
2030 	} else {
2031 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
2032 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2033 		    &aclnode->z_size, fuidp, cr)) != 0) {
2034 			zfs_acl_free(aclp);
2035 			zfs_acl_node_free(aclnode);
2036 			return (error);
2037 		}
2038 	}
2039 	aclp->z_acl_bytes = aclnode->z_size;
2040 	aclnode->z_ace_count = aclcnt;
2041 	aclp->z_acl_count = aclcnt;
2042 	list_insert_head(&aclp->z_acl, aclnode);
2043 
2044 	/*
2045 	 * If flags are being set then add them to z_hints
2046 	 */
2047 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2048 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2049 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2050 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2051 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2052 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2053 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2054 	}
2055 
2056 	*zaclp = aclp;
2057 
2058 	return (0);
2059 }
2060 
2061 /*
2062  * Set a files ACL
2063  */
2064 int
2065 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2066 {
2067 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2068 	zilog_t		*zilog = zfsvfs->z_log;
2069 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2070 	dmu_tx_t	*tx;
2071 	int		error;
2072 	zfs_acl_t	*aclp;
2073 	zfs_fuid_info_t	*fuidp = NULL;
2074 	boolean_t	fuid_dirtied;
2075 
2076 	if (mask == 0)
2077 		return (ENOSYS);
2078 
2079 	if (zp->z_phys->zp_flags & ZFS_IMMUTABLE)
2080 		return (EPERM);
2081 
2082 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
2083 		return (error);
2084 
2085 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
2086 	    &aclp);
2087 	if (error)
2088 		return (error);
2089 
2090 	/*
2091 	 * If ACL wide flags aren't being set then preserve any
2092 	 * existing flags.
2093 	 */
2094 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2095 		aclp->z_hints |= (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
2096 	}
2097 top:
2098 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)) {
2099 		zfs_acl_free(aclp);
2100 		return (error);
2101 	}
2102 
2103 	mutex_enter(&zp->z_lock);
2104 	mutex_enter(&zp->z_acl_lock);
2105 
2106 	tx = dmu_tx_create(zfsvfs->z_os);
2107 	dmu_tx_hold_bonus(tx, zp->z_id);
2108 
2109 	if (zp->z_phys->zp_acl.z_acl_extern_obj) {
2110 		/* Are we upgrading ACL? */
2111 		if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2112 		    zp->z_phys->zp_acl.z_acl_version ==
2113 		    ZFS_ACL_VERSION_INITIAL) {
2114 			dmu_tx_hold_free(tx,
2115 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2116 			    0, DMU_OBJECT_END);
2117 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2118 			    0, aclp->z_acl_bytes);
2119 		} else {
2120 			dmu_tx_hold_write(tx,
2121 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2122 			    0, aclp->z_acl_bytes);
2123 		}
2124 	} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2125 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2126 	}
2127 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2128 	if (fuid_dirtied)
2129 		zfs_fuid_txhold(zfsvfs, tx);
2130 
2131 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2132 	if (error) {
2133 		mutex_exit(&zp->z_acl_lock);
2134 		mutex_exit(&zp->z_lock);
2135 
2136 		if (error == ERESTART) {
2137 			dmu_tx_wait(tx);
2138 			dmu_tx_abort(tx);
2139 			goto top;
2140 		}
2141 		dmu_tx_abort(tx);
2142 		zfs_acl_free(aclp);
2143 		return (error);
2144 	}
2145 
2146 	error = zfs_aclset_common(zp, aclp, cr, tx);
2147 	ASSERT(error == 0);
2148 
2149 	if (fuid_dirtied)
2150 		zfs_fuid_sync(zfsvfs, tx);
2151 
2152 	zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2153 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2154 
2155 	if (fuidp)
2156 		zfs_fuid_info_free(fuidp);
2157 	zfs_acl_free(aclp);
2158 	dmu_tx_commit(tx);
2159 done:
2160 	mutex_exit(&zp->z_acl_lock);
2161 	mutex_exit(&zp->z_lock);
2162 
2163 	return (error);
2164 }
2165 
2166 /*
2167  * Check accesses of interest (AoI) against attributes of the dataset
2168  * such as read-only.  Returns zero if no AoI conflict with dataset
2169  * attributes, otherwise an appropriate errno is returned.
2170  */
2171 static int
2172 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2173 {
2174 	if ((v4_mode & WRITE_MASK) &&
2175 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2176 	    (!IS_DEVVP(ZTOV(zp)) ||
2177 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2178 		return (EROFS);
2179 	}
2180 
2181 	/*
2182 	 * Only check for READONLY on non-directories.
2183 	 */
2184 	if ((v4_mode & WRITE_MASK_DATA) &&
2185 	    (((ZTOV(zp)->v_type != VDIR) &&
2186 	    (zp->z_phys->zp_flags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2187 	    (ZTOV(zp)->v_type == VDIR &&
2188 	    (zp->z_phys->zp_flags & ZFS_IMMUTABLE)))) {
2189 		return (EPERM);
2190 	}
2191 
2192 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2193 	    (zp->z_phys->zp_flags & ZFS_NOUNLINK)) {
2194 		return (EPERM);
2195 	}
2196 
2197 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2198 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED))) {
2199 		return (EACCES);
2200 	}
2201 
2202 	return (0);
2203 }
2204 
2205 /*
2206  * The primary usage of this function is to loop through all of the
2207  * ACEs in the znode, determining what accesses of interest (AoI) to
2208  * the caller are allowed or denied.  The AoI are expressed as bits in
2209  * the working_mode parameter.  As each ACE is processed, bits covered
2210  * by that ACE are removed from the working_mode.  This removal
2211  * facilitates two things.  The first is that when the working mode is
2212  * empty (= 0), we know we've looked at all the AoI. The second is
2213  * that the ACE interpretation rules don't allow a later ACE to undo
2214  * something granted or denied by an earlier ACE.  Removing the
2215  * discovered access or denial enforces this rule.  At the end of
2216  * processing the ACEs, all AoI that were found to be denied are
2217  * placed into the working_mode, giving the caller a mask of denied
2218  * accesses.  Returns:
2219  *	0		if all AoI granted
2220  *	EACCESS 	if the denied mask is non-zero
2221  *	other error	if abnormal failure (e.g., IO error)
2222  *
2223  * A secondary usage of the function is to determine if any of the
2224  * AoI are granted.  If an ACE grants any access in
2225  * the working_mode, we immediately short circuit out of the function.
2226  * This mode is chosen by setting anyaccess to B_TRUE.  The
2227  * working_mode is not a denied access mask upon exit if the function
2228  * is used in this manner.
2229  */
2230 static int
2231 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2232     boolean_t anyaccess, cred_t *cr)
2233 {
2234 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2235 	zfs_acl_t	*aclp;
2236 	int		error;
2237 	uid_t		uid = crgetuid(cr);
2238 	uint64_t 	who;
2239 	uint16_t	type, iflags;
2240 	uint16_t	entry_type;
2241 	uint32_t	access_mask;
2242 	uint32_t	deny_mask = 0;
2243 	zfs_ace_hdr_t	*acep = NULL;
2244 	boolean_t	checkit;
2245 	uid_t		fowner;
2246 	uid_t		gowner;
2247 
2248 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2249 
2250 	mutex_enter(&zp->z_acl_lock);
2251 
2252 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2253 	if (error != 0) {
2254 		mutex_exit(&zp->z_acl_lock);
2255 		return (error);
2256 	}
2257 
2258 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2259 	    &iflags, &type)) {
2260 		uint32_t mask_matched;
2261 
2262 		if (!zfs_acl_valid_ace_type(type, iflags))
2263 			continue;
2264 
2265 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2266 			continue;
2267 
2268 		/* Skip ACE if it does not affect any AoI */
2269 		mask_matched = (access_mask & *working_mode);
2270 		if (!mask_matched)
2271 			continue;
2272 
2273 		entry_type = (iflags & ACE_TYPE_FLAGS);
2274 
2275 		checkit = B_FALSE;
2276 
2277 		switch (entry_type) {
2278 		case ACE_OWNER:
2279 			if (uid == fowner)
2280 				checkit = B_TRUE;
2281 			break;
2282 		case OWNING_GROUP:
2283 			who = gowner;
2284 			/*FALLTHROUGH*/
2285 		case ACE_IDENTIFIER_GROUP:
2286 			checkit = zfs_groupmember(zfsvfs, who, cr);
2287 			break;
2288 		case ACE_EVERYONE:
2289 			checkit = B_TRUE;
2290 			break;
2291 
2292 		/* USER Entry */
2293 		default:
2294 			if (entry_type == 0) {
2295 				uid_t newid;
2296 
2297 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2298 				    ZFS_ACE_USER);
2299 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2300 				    uid == newid)
2301 					checkit = B_TRUE;
2302 				break;
2303 			} else {
2304 				zfs_acl_free(aclp);
2305 				mutex_exit(&zp->z_acl_lock);
2306 				return (EIO);
2307 			}
2308 		}
2309 
2310 		if (checkit) {
2311 			if (type == DENY) {
2312 				DTRACE_PROBE3(zfs__ace__denies,
2313 				    znode_t *, zp,
2314 				    zfs_ace_hdr_t *, acep,
2315 				    uint32_t, mask_matched);
2316 				deny_mask |= mask_matched;
2317 			} else {
2318 				DTRACE_PROBE3(zfs__ace__allows,
2319 				    znode_t *, zp,
2320 				    zfs_ace_hdr_t *, acep,
2321 				    uint32_t, mask_matched);
2322 				if (anyaccess) {
2323 					mutex_exit(&zp->z_acl_lock);
2324 					zfs_acl_free(aclp);
2325 					return (0);
2326 				}
2327 			}
2328 			*working_mode &= ~mask_matched;
2329 		}
2330 
2331 		/* Are we done? */
2332 		if (*working_mode == 0)
2333 			break;
2334 	}
2335 
2336 	mutex_exit(&zp->z_acl_lock);
2337 	zfs_acl_free(aclp);
2338 
2339 	/* Put the found 'denies' back on the working mode */
2340 	if (deny_mask) {
2341 		*working_mode |= deny_mask;
2342 		return (EACCES);
2343 	} else if (*working_mode) {
2344 		return (-1);
2345 	}
2346 
2347 	return (0);
2348 }
2349 
2350 /*
2351  * Return true if any access whatsoever granted, we don't actually
2352  * care what access is granted.
2353  */
2354 boolean_t
2355 zfs_has_access(znode_t *zp, cred_t *cr)
2356 {
2357 	uint32_t have = ACE_ALL_PERMS;
2358 
2359 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2360 		uid_t		owner;
2361 
2362 		owner = zfs_fuid_map_id(zp->z_zfsvfs,
2363 		    zp->z_phys->zp_uid, cr, ZFS_OWNER);
2364 
2365 		return (
2366 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VREAD) == 0 ||
2367 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VWRITE) == 0 ||
2368 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VEXEC) == 0 ||
2369 		    secpolicy_vnode_chown(cr, B_TRUE) == 0 ||
2370 		    secpolicy_vnode_chown(cr, B_FALSE) == 0 ||
2371 		    secpolicy_vnode_setdac(cr, owner) == 0 ||
2372 		    secpolicy_vnode_remove(cr) == 0);
2373 	}
2374 	return (B_TRUE);
2375 }
2376 
2377 static int
2378 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2379     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2380 {
2381 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2382 	int err;
2383 
2384 	*working_mode = v4_mode;
2385 	*check_privs = B_TRUE;
2386 
2387 	/*
2388 	 * Short circuit empty requests
2389 	 */
2390 	if (v4_mode == 0 || zfsvfs->z_replay) {
2391 		*working_mode = 0;
2392 		return (0);
2393 	}
2394 
2395 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2396 		*check_privs = B_FALSE;
2397 		return (err);
2398 	}
2399 
2400 	/*
2401 	 * The caller requested that the ACL check be skipped.  This
2402 	 * would only happen if the caller checked VOP_ACCESS() with a
2403 	 * 32 bit ACE mask and already had the appropriate permissions.
2404 	 */
2405 	if (skipaclchk) {
2406 		*working_mode = 0;
2407 		return (0);
2408 	}
2409 
2410 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2411 }
2412 
2413 static int
2414 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2415     cred_t *cr)
2416 {
2417 	if (*working_mode != ACE_WRITE_DATA)
2418 		return (EACCES);
2419 
2420 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2421 	    check_privs, B_FALSE, cr));
2422 }
2423 
2424 /*
2425  * Determine whether Access should be granted/denied, invoking least
2426  * priv subsytem when a deny is determined.
2427  */
2428 int
2429 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2430 {
2431 	uint32_t	working_mode;
2432 	int		error;
2433 	int		is_attr;
2434 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2435 	boolean_t 	check_privs;
2436 	znode_t		*xzp;
2437 	znode_t 	*check_zp = zp;
2438 
2439 	is_attr = ((zp->z_phys->zp_flags & ZFS_XATTR) &&
2440 	    (ZTOV(zp)->v_type == VDIR));
2441 
2442 	/*
2443 	 * If attribute then validate against base file
2444 	 */
2445 	if (is_attr) {
2446 		if ((error = zfs_zget(zp->z_zfsvfs,
2447 		    zp->z_phys->zp_parent, &xzp)) != 0)	{
2448 			return (error);
2449 		}
2450 
2451 		check_zp = xzp;
2452 
2453 		/*
2454 		 * fixup mode to map to xattr perms
2455 		 */
2456 
2457 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2458 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2459 			mode |= ACE_WRITE_NAMED_ATTRS;
2460 		}
2461 
2462 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2463 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2464 			mode |= ACE_READ_NAMED_ATTRS;
2465 		}
2466 	}
2467 
2468 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2469 	    &check_privs, skipaclchk, cr)) == 0) {
2470 		if (is_attr)
2471 			VN_RELE(ZTOV(xzp));
2472 		return (0);
2473 	}
2474 
2475 	if (error && !check_privs) {
2476 		if (is_attr)
2477 			VN_RELE(ZTOV(xzp));
2478 		return (error);
2479 	}
2480 
2481 	if (error && (flags & V_APPEND)) {
2482 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2483 	}
2484 
2485 	if (error && check_privs) {
2486 		uid_t		owner;
2487 		mode_t		checkmode = 0;
2488 
2489 		owner = zfs_fuid_map_id(zfsvfs, check_zp->z_phys->zp_uid, cr,
2490 		    ZFS_OWNER);
2491 
2492 		/*
2493 		 * First check for implicit owner permission on
2494 		 * read_acl/read_attributes
2495 		 */
2496 
2497 		error = 0;
2498 		ASSERT(working_mode != 0);
2499 
2500 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2501 		    owner == crgetuid(cr)))
2502 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2503 
2504 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2505 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2506 			checkmode |= VREAD;
2507 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2508 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2509 			checkmode |= VWRITE;
2510 		if (working_mode & ACE_EXECUTE)
2511 			checkmode |= VEXEC;
2512 
2513 		if (checkmode)
2514 			error = secpolicy_vnode_access(cr, ZTOV(check_zp),
2515 			    owner, checkmode);
2516 
2517 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2518 			error = secpolicy_vnode_chown(cr, B_TRUE);
2519 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2520 			error = secpolicy_vnode_setdac(cr, owner);
2521 
2522 		if (error == 0 && (working_mode &
2523 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2524 			error = secpolicy_vnode_remove(cr);
2525 
2526 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2527 			error = secpolicy_vnode_chown(cr, B_FALSE);
2528 		}
2529 		if (error == 0) {
2530 			/*
2531 			 * See if any bits other than those already checked
2532 			 * for are still present.  If so then return EACCES
2533 			 */
2534 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2535 				error = EACCES;
2536 			}
2537 		}
2538 	}
2539 
2540 	if (is_attr)
2541 		VN_RELE(ZTOV(xzp));
2542 
2543 	return (error);
2544 }
2545 
2546 /*
2547  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2548  * native ACL format and call zfs_zaccess()
2549  */
2550 int
2551 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2552 {
2553 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2554 }
2555 
2556 /*
2557  * Access function for secpolicy_vnode_setattr
2558  */
2559 int
2560 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2561 {
2562 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2563 
2564 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2565 }
2566 
2567 static int
2568 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2569     mode_t missing_perms, cred_t *cr)
2570 {
2571 	int error;
2572 	uid_t downer;
2573 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2574 
2575 	downer = zfs_fuid_map_id(zfsvfs, dzp->z_phys->zp_uid, cr, ZFS_OWNER);
2576 
2577 	error = secpolicy_vnode_access(cr, ZTOV(dzp), downer, missing_perms);
2578 
2579 	if (error == 0)
2580 		error = zfs_sticky_remove_access(dzp, zp, cr);
2581 
2582 	return (error);
2583 }
2584 
2585 /*
2586  * Determine whether Access should be granted/deny, without
2587  * consulting least priv subsystem.
2588  *
2589  *
2590  * The following chart is the recommended NFSv4 enforcement for
2591  * ability to delete an object.
2592  *
2593  *      -------------------------------------------------------
2594  *      |   Parent Dir  |           Target Object Permissions |
2595  *      |  permissions  |                                     |
2596  *      -------------------------------------------------------
2597  *      |               | ACL Allows | ACL Denies| Delete     |
2598  *      |               |  Delete    |  Delete   | unspecified|
2599  *      -------------------------------------------------------
2600  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2601  *      |  DELETE_CHILD |                                     |
2602  *      -------------------------------------------------------
2603  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2604  *      |  DELETE_CHILD |            |           |            |
2605  *      -------------------------------------------------------
2606  *      | ACL specifies |            |           |            |
2607  *      | only allow    | Permit     | Permit    | Permit     |
2608  *      | write and     |            |           |            |
2609  *      | execute       |            |           |            |
2610  *      -------------------------------------------------------
2611  *      | ACL denies    |            |           |            |
2612  *      | write and     | Permit     | Deny      | Deny       |
2613  *      | execute       |            |           |            |
2614  *      -------------------------------------------------------
2615  *         ^
2616  *         |
2617  *         No search privilege, can't even look up file?
2618  *
2619  */
2620 int
2621 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2622 {
2623 	uint32_t dzp_working_mode = 0;
2624 	uint32_t zp_working_mode = 0;
2625 	int dzp_error, zp_error;
2626 	mode_t missing_perms;
2627 	boolean_t dzpcheck_privs = B_TRUE;
2628 	boolean_t zpcheck_privs = B_TRUE;
2629 
2630 	/*
2631 	 * We want specific DELETE permissions to
2632 	 * take precedence over WRITE/EXECUTE.  We don't
2633 	 * want an ACL such as this to mess us up.
2634 	 * user:joe:write_data:deny,user:joe:delete:allow
2635 	 *
2636 	 * However, deny permissions may ultimately be overridden
2637 	 * by secpolicy_vnode_access().
2638 	 *
2639 	 * We will ask for all of the necessary permissions and then
2640 	 * look at the working modes from the directory and target object
2641 	 * to determine what was found.
2642 	 */
2643 
2644 	if (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2645 		return (EPERM);
2646 
2647 	/*
2648 	 * First row
2649 	 * If the directory permissions allow the delete, we are done.
2650 	 */
2651 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2652 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2653 		return (0);
2654 
2655 	/*
2656 	 * If target object has delete permission then we are done
2657 	 */
2658 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2659 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2660 		return (0);
2661 
2662 	ASSERT(dzp_error && zp_error);
2663 
2664 	if (!dzpcheck_privs)
2665 		return (dzp_error);
2666 	if (!zpcheck_privs)
2667 		return (zp_error);
2668 
2669 	/*
2670 	 * Second row
2671 	 *
2672 	 * If directory returns EACCES then delete_child was denied
2673 	 * due to deny delete_child.  In this case send the request through
2674 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2675 	 * since that *could* allow the delete based on write/execute permission
2676 	 * and we want delete permissions to override write/execute.
2677 	 */
2678 
2679 	if (dzp_error == EACCES)
2680 		return (secpolicy_vnode_remove(cr));
2681 
2682 	/*
2683 	 * Third Row
2684 	 * only need to see if we have write/execute on directory.
2685 	 */
2686 
2687 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2688 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2689 		return (zfs_sticky_remove_access(dzp, zp, cr));
2690 
2691 	if (!dzpcheck_privs)
2692 		return (dzp_error);
2693 
2694 	/*
2695 	 * Fourth row
2696 	 */
2697 
2698 	missing_perms = (dzp_working_mode & ACE_WRITE_DATA) ? VWRITE : 0;
2699 	missing_perms |= (dzp_working_mode & ACE_EXECUTE) ? VEXEC : 0;
2700 
2701 	ASSERT(missing_perms);
2702 
2703 	return (zfs_delete_final_check(zp, dzp, missing_perms, cr));
2704 
2705 }
2706 
2707 int
2708 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2709     znode_t *tzp, cred_t *cr)
2710 {
2711 	int add_perm;
2712 	int error;
2713 
2714 	if (szp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
2715 		return (EACCES);
2716 
2717 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2718 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2719 
2720 	/*
2721 	 * Rename permissions are combination of delete permission +
2722 	 * add file/subdir permission.
2723 	 */
2724 
2725 	/*
2726 	 * first make sure we do the delete portion.
2727 	 *
2728 	 * If that succeeds then check for add_file/add_subdir permissions
2729 	 */
2730 
2731 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2732 		return (error);
2733 
2734 	/*
2735 	 * If we have a tzp, see if we can delete it?
2736 	 */
2737 	if (tzp) {
2738 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2739 			return (error);
2740 	}
2741 
2742 	/*
2743 	 * Now check for add permissions
2744 	 */
2745 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2746 
2747 	return (error);
2748 }
2749