xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_acl.c (revision dcf1eb702aeeccc639446ab5c5e8d725ce20cd76)
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 	boolean_t	an_exec_denied = B_FALSE;
785 
786 	mode = (zp->z_phys->zp_mode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
787 
788 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
789 	    &access_mask, &iflags, &type)) {
790 
791 		if (!zfs_acl_valid_ace_type(type, iflags))
792 			continue;
793 
794 		entry_type = (iflags & ACE_TYPE_FLAGS);
795 
796 		/*
797 		 * Skip over owner@, group@ or everyone@ inherit only ACEs
798 		 */
799 		if ((iflags & ACE_INHERIT_ONLY_ACE) &&
800 		    (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
801 		    entry_type == OWNING_GROUP))
802 			continue;
803 
804 		if (entry_type == ACE_OWNER) {
805 			if ((access_mask & ACE_READ_DATA) &&
806 			    (!(seen & S_IRUSR))) {
807 				seen |= S_IRUSR;
808 				if (type == ALLOW) {
809 					mode |= S_IRUSR;
810 				}
811 			}
812 			if ((access_mask & ACE_WRITE_DATA) &&
813 			    (!(seen & S_IWUSR))) {
814 				seen |= S_IWUSR;
815 				if (type == ALLOW) {
816 					mode |= S_IWUSR;
817 				}
818 			}
819 			if ((access_mask & ACE_EXECUTE) &&
820 			    (!(seen & S_IXUSR))) {
821 				seen |= S_IXUSR;
822 				if (type == ALLOW) {
823 					mode |= S_IXUSR;
824 				}
825 			}
826 		} else if (entry_type == OWNING_GROUP) {
827 			if ((access_mask & ACE_READ_DATA) &&
828 			    (!(seen & S_IRGRP))) {
829 				seen |= S_IRGRP;
830 				if (type == ALLOW) {
831 					mode |= S_IRGRP;
832 				}
833 			}
834 			if ((access_mask & ACE_WRITE_DATA) &&
835 			    (!(seen & S_IWGRP))) {
836 				seen |= S_IWGRP;
837 				if (type == ALLOW) {
838 					mode |= S_IWGRP;
839 				}
840 			}
841 			if ((access_mask & ACE_EXECUTE) &&
842 			    (!(seen & S_IXGRP))) {
843 				seen |= S_IXGRP;
844 				if (type == ALLOW) {
845 					mode |= S_IXGRP;
846 				}
847 			}
848 		} else if (entry_type == ACE_EVERYONE) {
849 			if ((access_mask & ACE_READ_DATA)) {
850 				if (!(seen & S_IRUSR)) {
851 					seen |= S_IRUSR;
852 					if (type == ALLOW) {
853 						mode |= S_IRUSR;
854 					}
855 				}
856 				if (!(seen & S_IRGRP)) {
857 					seen |= S_IRGRP;
858 					if (type == ALLOW) {
859 						mode |= S_IRGRP;
860 					}
861 				}
862 				if (!(seen & S_IROTH)) {
863 					seen |= S_IROTH;
864 					if (type == ALLOW) {
865 						mode |= S_IROTH;
866 					}
867 				}
868 			}
869 			if ((access_mask & ACE_WRITE_DATA)) {
870 				if (!(seen & S_IWUSR)) {
871 					seen |= S_IWUSR;
872 					if (type == ALLOW) {
873 						mode |= S_IWUSR;
874 					}
875 				}
876 				if (!(seen & S_IWGRP)) {
877 					seen |= S_IWGRP;
878 					if (type == ALLOW) {
879 						mode |= S_IWGRP;
880 					}
881 				}
882 				if (!(seen & S_IWOTH)) {
883 					seen |= S_IWOTH;
884 					if (type == ALLOW) {
885 						mode |= S_IWOTH;
886 					}
887 				}
888 			}
889 			if ((access_mask & ACE_EXECUTE)) {
890 				if (!(seen & S_IXUSR)) {
891 					seen |= S_IXUSR;
892 					if (type == ALLOW) {
893 						mode |= S_IXUSR;
894 					}
895 				}
896 				if (!(seen & S_IXGRP)) {
897 					seen |= S_IXGRP;
898 					if (type == ALLOW) {
899 						mode |= S_IXGRP;
900 					}
901 				}
902 				if (!(seen & S_IXOTH)) {
903 					seen |= S_IXOTH;
904 					if (type == ALLOW) {
905 						mode |= S_IXOTH;
906 					}
907 				}
908 			}
909 		} else {
910 			/*
911 			 * Only care if this IDENTIFIER_GROUP or
912 			 * USER ACE denies execute access to someone,
913 			 * mode is not affected
914 			 */
915 			if ((access_mask & ACE_EXECUTE) && type == DENY)
916 				an_exec_denied = B_TRUE;
917 		}
918 	}
919 
920 	if (!an_exec_denied && !(seen & (S_IXUSR | S_IXGRP | S_IXOTH)) ||
921 	    !(mode & (S_IXUSR | S_IXGRP | S_IXOTH)))
922 		an_exec_denied = B_TRUE;
923 
924 	if (an_exec_denied)
925 		zp->z_phys->zp_flags &= ~ZFS_NO_EXECS_DENIED;
926 	else
927 		zp->z_phys->zp_flags |= ZFS_NO_EXECS_DENIED;
928 
929 	return (mode);
930 }
931 
932 static zfs_acl_t *
933 zfs_acl_node_read_internal(znode_t *zp, boolean_t will_modify)
934 {
935 	zfs_acl_t	*aclp;
936 	zfs_acl_node_t	*aclnode;
937 
938 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
939 
940 	/*
941 	 * Version 0 to 1 znode_acl_phys has the size/count fields swapped.
942 	 * Version 0 didn't have a size field, only a count.
943 	 */
944 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
945 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_size;
946 		aclp->z_acl_bytes = ZFS_ACL_SIZE(aclp->z_acl_count);
947 	} else {
948 		aclp->z_acl_count = zp->z_phys->zp_acl.z_acl_count;
949 		aclp->z_acl_bytes = zp->z_phys->zp_acl.z_acl_size;
950 	}
951 
952 	aclnode = zfs_acl_node_alloc(will_modify ? aclp->z_acl_bytes : 0);
953 	aclnode->z_ace_count = aclp->z_acl_count;
954 	if (will_modify) {
955 		bcopy(zp->z_phys->zp_acl.z_ace_data, aclnode->z_acldata,
956 		    aclp->z_acl_bytes);
957 	} else {
958 		aclnode->z_size = aclp->z_acl_bytes;
959 		aclnode->z_acldata = &zp->z_phys->zp_acl.z_ace_data[0];
960 	}
961 
962 	list_insert_head(&aclp->z_acl, aclnode);
963 
964 	return (aclp);
965 }
966 
967 /*
968  * Read an external acl object.
969  */
970 static int
971 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
972 {
973 	uint64_t extacl = zp->z_phys->zp_acl.z_acl_extern_obj;
974 	zfs_acl_t	*aclp;
975 	size_t		aclsize;
976 	size_t		acl_count;
977 	zfs_acl_node_t	*aclnode;
978 	int error;
979 
980 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
981 
982 	if (zp->z_acl_cached) {
983 		*aclpp = zp->z_acl_cached;
984 		return (0);
985 	}
986 
987 	if (zp->z_phys->zp_acl.z_acl_extern_obj == 0) {
988 		*aclpp = zfs_acl_node_read_internal(zp, will_modify);
989 		zp->z_acl_cached = *aclpp;
990 		return (0);
991 	}
992 
993 	aclp = zfs_acl_alloc(zp->z_phys->zp_acl.z_acl_version);
994 	if (zp->z_phys->zp_acl.z_acl_version == ZFS_ACL_VERSION_INITIAL) {
995 		zfs_acl_phys_v0_t *zacl0 =
996 		    (zfs_acl_phys_v0_t *)&zp->z_phys->zp_acl;
997 
998 		aclsize = ZFS_ACL_SIZE(zacl0->z_acl_count);
999 		acl_count = zacl0->z_acl_count;
1000 	} else {
1001 		aclsize = zp->z_phys->zp_acl.z_acl_size;
1002 		acl_count = zp->z_phys->zp_acl.z_acl_count;
1003 		if (aclsize == 0)
1004 			aclsize = acl_count * sizeof (zfs_ace_t);
1005 	}
1006 	aclnode = zfs_acl_node_alloc(aclsize);
1007 	list_insert_head(&aclp->z_acl, aclnode);
1008 	error = dmu_read(zp->z_zfsvfs->z_os, extacl, 0,
1009 	    aclsize, aclnode->z_acldata, DMU_READ_PREFETCH);
1010 	aclnode->z_ace_count = acl_count;
1011 	aclp->z_acl_count = acl_count;
1012 	aclp->z_acl_bytes = aclsize;
1013 
1014 	if (error != 0) {
1015 		zfs_acl_free(aclp);
1016 		/* convert checksum errors into IO errors */
1017 		if (error == ECKSUM)
1018 			error = EIO;
1019 		return (error);
1020 	}
1021 
1022 	zp->z_acl_cached = *aclpp = aclp;
1023 	return (0);
1024 }
1025 
1026 /*
1027  * common code for setting ACLs.
1028  *
1029  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1030  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1031  * already checked the acl and knows whether to inherit.
1032  */
1033 int
1034 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1035 {
1036 	int		error;
1037 	znode_phys_t	*zphys = zp->z_phys;
1038 	zfs_acl_phys_t	*zacl = &zphys->zp_acl;
1039 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1040 	uint64_t	aoid = zphys->zp_acl.z_acl_extern_obj;
1041 	uint64_t	off = 0;
1042 	dmu_object_type_t otype;
1043 	zfs_acl_node_t	*aclnode;
1044 
1045 	dmu_buf_will_dirty(zp->z_dbuf, tx);
1046 
1047 	if (zp->z_acl_cached != aclp && zp->z_acl_cached) {
1048 		zfs_acl_free(zp->z_acl_cached);
1049 		zp->z_acl_cached = NULL;
1050 	}
1051 
1052 	zphys->zp_mode = zfs_mode_compute(zp, aclp);
1053 
1054 	/*
1055 	 * Decide which opbject type to use.  If we are forced to
1056 	 * use old ACL format than transform ACL into zfs_oldace_t
1057 	 * layout.
1058 	 */
1059 	if (!zfsvfs->z_use_fuids) {
1060 		otype = DMU_OT_OLDACL;
1061 	} else {
1062 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1063 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1064 			zfs_acl_xform(zp, aclp, cr);
1065 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1066 		otype = DMU_OT_ACL;
1067 	}
1068 
1069 	if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1070 		/*
1071 		 * If ACL was previously external and we are now
1072 		 * converting to new ACL format then release old
1073 		 * ACL object and create a new one.
1074 		 */
1075 		if (aoid && aclp->z_version != zacl->z_acl_version) {
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 			aoid = 0;
1081 		}
1082 		if (aoid == 0) {
1083 			aoid = dmu_object_alloc(zfsvfs->z_os,
1084 			    otype, aclp->z_acl_bytes,
1085 			    otype == DMU_OT_ACL ? DMU_OT_SYSACL : DMU_OT_NONE,
1086 			    otype == DMU_OT_ACL ? DN_MAX_BONUSLEN : 0, tx);
1087 		} else {
1088 			(void) dmu_object_set_blocksize(zfsvfs->z_os, aoid,
1089 			    aclp->z_acl_bytes, 0, tx);
1090 		}
1091 		zphys->zp_acl.z_acl_extern_obj = aoid;
1092 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1093 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1094 			if (aclnode->z_ace_count == 0)
1095 				continue;
1096 			dmu_write(zfsvfs->z_os, aoid, off,
1097 			    aclnode->z_size, aclnode->z_acldata, tx);
1098 			off += aclnode->z_size;
1099 		}
1100 	} else {
1101 		void *start = zacl->z_ace_data;
1102 		/*
1103 		 * Migrating back embedded?
1104 		 */
1105 		if (zphys->zp_acl.z_acl_extern_obj) {
1106 			error = dmu_object_free(zfsvfs->z_os,
1107 			    zp->z_phys->zp_acl.z_acl_extern_obj, tx);
1108 			if (error)
1109 				return (error);
1110 			zphys->zp_acl.z_acl_extern_obj = 0;
1111 		}
1112 
1113 		for (aclnode = list_head(&aclp->z_acl); aclnode;
1114 		    aclnode = list_next(&aclp->z_acl, aclnode)) {
1115 			if (aclnode->z_ace_count == 0)
1116 				continue;
1117 			bcopy(aclnode->z_acldata, start, aclnode->z_size);
1118 			start = (caddr_t)start + aclnode->z_size;
1119 		}
1120 	}
1121 
1122 	/*
1123 	 * If Old version then swap count/bytes to match old
1124 	 * layout of znode_acl_phys_t.
1125 	 */
1126 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1127 		zphys->zp_acl.z_acl_size = aclp->z_acl_count;
1128 		zphys->zp_acl.z_acl_count = aclp->z_acl_bytes;
1129 	} else {
1130 		zphys->zp_acl.z_acl_size = aclp->z_acl_bytes;
1131 		zphys->zp_acl.z_acl_count = aclp->z_acl_count;
1132 	}
1133 
1134 	zphys->zp_acl.z_acl_version = aclp->z_version;
1135 
1136 	/*
1137 	 * Replace ACL wide bits, but first clear them.
1138 	 */
1139 	zp->z_phys->zp_flags &= ~ZFS_ACL_WIDE_FLAGS;
1140 
1141 	zp->z_phys->zp_flags |= aclp->z_hints;
1142 
1143 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1144 		zp->z_phys->zp_flags |= ZFS_ACL_TRIVIAL;
1145 
1146 	return (0);
1147 }
1148 
1149 /*
1150  * Update access mask for prepended ACE
1151  *
1152  * This applies the "groupmask" value for aclmode property.
1153  */
1154 static void
1155 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1156     mode_t mode, uint64_t owner)
1157 {
1158 	int	rmask, wmask, xmask;
1159 	int	user_ace;
1160 	uint16_t aceflags;
1161 	uint32_t origmask, acepmask;
1162 	uint64_t fuid;
1163 
1164 	aceflags = aclp->z_ops.ace_flags_get(acep);
1165 	fuid = aclp->z_ops.ace_who_get(acep);
1166 	origmask = aclp->z_ops.ace_mask_get(origacep);
1167 	acepmask = aclp->z_ops.ace_mask_get(acep);
1168 
1169 	user_ace = (!(aceflags &
1170 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1171 
1172 	if (user_ace && (fuid == owner)) {
1173 		rmask = S_IRUSR;
1174 		wmask = S_IWUSR;
1175 		xmask = S_IXUSR;
1176 	} else {
1177 		rmask = S_IRGRP;
1178 		wmask = S_IWGRP;
1179 		xmask = S_IXGRP;
1180 	}
1181 
1182 	if (origmask & ACE_READ_DATA) {
1183 		if (mode & rmask) {
1184 			acepmask &= ~ACE_READ_DATA;
1185 		} else {
1186 			acepmask |= ACE_READ_DATA;
1187 		}
1188 	}
1189 
1190 	if (origmask & ACE_WRITE_DATA) {
1191 		if (mode & wmask) {
1192 			acepmask &= ~ACE_WRITE_DATA;
1193 		} else {
1194 			acepmask |= ACE_WRITE_DATA;
1195 		}
1196 	}
1197 
1198 	if (origmask & ACE_APPEND_DATA) {
1199 		if (mode & wmask) {
1200 			acepmask &= ~ACE_APPEND_DATA;
1201 		} else {
1202 			acepmask |= ACE_APPEND_DATA;
1203 		}
1204 	}
1205 
1206 	if (origmask & ACE_EXECUTE) {
1207 		if (mode & xmask) {
1208 			acepmask &= ~ACE_EXECUTE;
1209 		} else {
1210 			acepmask |= ACE_EXECUTE;
1211 		}
1212 	}
1213 	aclp->z_ops.ace_mask_set(acep, acepmask);
1214 }
1215 
1216 /*
1217  * Apply mode to canonical six ACEs.
1218  */
1219 static void
1220 zfs_acl_fixup_canonical_six(zfs_acl_t *aclp, mode_t mode)
1221 {
1222 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1223 	void	*acep;
1224 	int	maskoff = aclp->z_ops.ace_mask_off();
1225 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1226 
1227 	ASSERT(aclnode != NULL);
1228 
1229 	acep = (void *)((caddr_t)aclnode->z_acldata +
1230 	    aclnode->z_size - (abstract_size * 6));
1231 
1232 	/*
1233 	 * Fixup final ACEs to match the mode
1234 	 */
1235 
1236 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1237 	    (mode & 0700) >> 6);	/* owner@ */
1238 
1239 	acep = (caddr_t)acep + (abstract_size * 2);
1240 
1241 	adjust_ace_pair_common(acep, maskoff, abstract_size,
1242 	    (mode & 0070) >> 3);	/* group@ */
1243 
1244 	acep = (caddr_t)acep + (abstract_size * 2);
1245 	adjust_ace_pair_common(acep, maskoff,
1246 	    abstract_size, mode);	/* everyone@ */
1247 }
1248 
1249 
1250 static int
1251 zfs_acl_ace_match(zfs_acl_t *aclp, void *acep, int allow_deny,
1252     int entry_type, int accessmask)
1253 {
1254 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1255 	uint16_t type = aclp->z_ops.ace_type_get(acep);
1256 	uint16_t flags = aclp->z_ops.ace_flags_get(acep);
1257 
1258 	return (mask == accessmask && type == allow_deny &&
1259 	    ((flags & ACE_TYPE_FLAGS) == entry_type));
1260 }
1261 
1262 /*
1263  * Can prepended ACE be reused?
1264  */
1265 static int
1266 zfs_reuse_deny(zfs_acl_t *aclp, void *acep, void *prevacep)
1267 {
1268 	int okay_masks;
1269 	uint16_t prevtype;
1270 	uint16_t prevflags;
1271 	uint16_t flags;
1272 	uint32_t mask, prevmask;
1273 
1274 	if (prevacep == NULL)
1275 		return (B_FALSE);
1276 
1277 	prevtype = aclp->z_ops.ace_type_get(prevacep);
1278 	prevflags = aclp->z_ops.ace_flags_get(prevacep);
1279 	flags = aclp->z_ops.ace_flags_get(acep);
1280 	mask = aclp->z_ops.ace_mask_get(acep);
1281 	prevmask = aclp->z_ops.ace_mask_get(prevacep);
1282 
1283 	if (prevtype != DENY)
1284 		return (B_FALSE);
1285 
1286 	if (prevflags != (flags & ACE_IDENTIFIER_GROUP))
1287 		return (B_FALSE);
1288 
1289 	okay_masks = (mask & OKAY_MASK_BITS);
1290 
1291 	if (prevmask & ~okay_masks)
1292 		return (B_FALSE);
1293 
1294 	return (B_TRUE);
1295 }
1296 
1297 
1298 /*
1299  * Insert new ACL node into chain of zfs_acl_node_t's
1300  *
1301  * This will result in two possible results.
1302  * 1. If the ACL is currently just a single zfs_acl_node and
1303  *    we are prepending the entry then current acl node will have
1304  *    a new node inserted above it.
1305  *
1306  * 2. If we are inserting in the middle of current acl node then
1307  *    the current node will be split in two and new node will be inserted
1308  *    in between the two split nodes.
1309  */
1310 static zfs_acl_node_t *
1311 zfs_acl_ace_insert(zfs_acl_t *aclp, void  *acep)
1312 {
1313 	zfs_acl_node_t 	*newnode;
1314 	zfs_acl_node_t 	*trailernode = NULL;
1315 	zfs_acl_node_t 	*currnode = zfs_acl_curr_node(aclp);
1316 	int		curr_idx = aclp->z_curr_node->z_ace_idx;
1317 	int		trailer_count;
1318 	size_t		oldsize;
1319 
1320 	newnode = zfs_acl_node_alloc(aclp->z_ops.ace_size(acep));
1321 	newnode->z_ace_count = 1;
1322 
1323 	oldsize = currnode->z_size;
1324 
1325 	if (curr_idx != 1) {
1326 		trailernode = zfs_acl_node_alloc(0);
1327 		trailernode->z_acldata = acep;
1328 
1329 		trailer_count = currnode->z_ace_count - curr_idx + 1;
1330 		currnode->z_ace_count = curr_idx - 1;
1331 		currnode->z_size = (caddr_t)acep - (caddr_t)currnode->z_acldata;
1332 		trailernode->z_size = oldsize - currnode->z_size;
1333 		trailernode->z_ace_count = trailer_count;
1334 	}
1335 
1336 	aclp->z_acl_count += 1;
1337 	aclp->z_acl_bytes += aclp->z_ops.ace_size(acep);
1338 
1339 	if (curr_idx == 1)
1340 		list_insert_before(&aclp->z_acl, currnode, newnode);
1341 	else
1342 		list_insert_after(&aclp->z_acl, currnode, newnode);
1343 	if (trailernode) {
1344 		list_insert_after(&aclp->z_acl, newnode, trailernode);
1345 		aclp->z_curr_node = trailernode;
1346 		trailernode->z_ace_idx = 1;
1347 	}
1348 
1349 	return (newnode);
1350 }
1351 
1352 /*
1353  * Prepend deny ACE
1354  */
1355 static void *
1356 zfs_acl_prepend_deny(uint64_t uid, zfs_acl_t *aclp, void *acep,
1357     mode_t mode)
1358 {
1359 	zfs_acl_node_t *aclnode;
1360 	void  *newacep;
1361 	uint64_t fuid;
1362 	uint16_t flags;
1363 
1364 	aclnode = zfs_acl_ace_insert(aclp, acep);
1365 	newacep = aclnode->z_acldata;
1366 	fuid = aclp->z_ops.ace_who_get(acep);
1367 	flags = aclp->z_ops.ace_flags_get(acep);
1368 	zfs_set_ace(aclp, newacep, 0, DENY, fuid, (flags & ACE_TYPE_FLAGS));
1369 	zfs_acl_prepend_fixup(aclp, newacep, acep, mode, uid);
1370 
1371 	return (newacep);
1372 }
1373 
1374 /*
1375  * Split an inherited ACE into inherit_only ACE
1376  * and original ACE with inheritance flags stripped off.
1377  */
1378 static void
1379 zfs_acl_split_ace(zfs_acl_t *aclp, zfs_ace_hdr_t *acep)
1380 {
1381 	zfs_acl_node_t *aclnode;
1382 	zfs_acl_node_t *currnode;
1383 	void  *newacep;
1384 	uint16_t type, flags;
1385 	uint32_t mask;
1386 	uint64_t fuid;
1387 
1388 	type = aclp->z_ops.ace_type_get(acep);
1389 	flags = aclp->z_ops.ace_flags_get(acep);
1390 	mask = aclp->z_ops.ace_mask_get(acep);
1391 	fuid = aclp->z_ops.ace_who_get(acep);
1392 
1393 	aclnode = zfs_acl_ace_insert(aclp, acep);
1394 	newacep = aclnode->z_acldata;
1395 
1396 	aclp->z_ops.ace_type_set(newacep, type);
1397 	aclp->z_ops.ace_flags_set(newacep, flags | ACE_INHERIT_ONLY_ACE);
1398 	aclp->z_ops.ace_mask_set(newacep, mask);
1399 	aclp->z_ops.ace_type_set(newacep, type);
1400 	aclp->z_ops.ace_who_set(newacep, fuid);
1401 	aclp->z_next_ace = acep;
1402 	flags &= ~ALL_INHERIT;
1403 	aclp->z_ops.ace_flags_set(acep, flags);
1404 	currnode = zfs_acl_curr_node(aclp);
1405 	ASSERT(currnode->z_ace_idx >= 1);
1406 	currnode->z_ace_idx -= 1;
1407 }
1408 
1409 /*
1410  * Are ACES started at index i, the canonical six ACES?
1411  */
1412 static int
1413 zfs_have_canonical_six(zfs_acl_t *aclp)
1414 {
1415 	void *acep;
1416 	zfs_acl_node_t *aclnode = list_tail(&aclp->z_acl);
1417 	int		i = 0;
1418 	size_t abstract_size = aclp->z_ops.ace_abstract_size();
1419 
1420 	ASSERT(aclnode != NULL);
1421 
1422 	if (aclnode->z_ace_count < 6)
1423 		return (0);
1424 
1425 	acep = (void *)((caddr_t)aclnode->z_acldata +
1426 	    aclnode->z_size - (aclp->z_ops.ace_abstract_size() * 6));
1427 
1428 	if ((zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1429 	    DENY, ACE_OWNER, 0) &&
1430 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1431 	    ALLOW, ACE_OWNER, OWNER_ALLOW_MASK) &&
1432 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++), DENY,
1433 	    OWNING_GROUP, 0) && zfs_acl_ace_match(aclp, (caddr_t)acep +
1434 	    (abstract_size * i++),
1435 	    ALLOW, OWNING_GROUP, 0) &&
1436 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1437 	    DENY, ACE_EVERYONE, EVERYONE_DENY_MASK) &&
1438 	    zfs_acl_ace_match(aclp, (caddr_t)acep + (abstract_size * i++),
1439 	    ALLOW, ACE_EVERYONE, EVERYONE_ALLOW_MASK))) {
1440 		return (1);
1441 	} else {
1442 		return (0);
1443 	}
1444 }
1445 
1446 
1447 /*
1448  * Apply step 1g, to group entries
1449  *
1450  * Need to deal with corner case where group may have
1451  * greater permissions than owner.  If so then limit
1452  * group permissions, based on what extra permissions
1453  * group has.
1454  */
1455 static void
1456 zfs_fixup_group_entries(zfs_acl_t *aclp, void *acep, void *prevacep,
1457     mode_t mode)
1458 {
1459 	uint32_t prevmask = aclp->z_ops.ace_mask_get(prevacep);
1460 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1461 	uint16_t prevflags = aclp->z_ops.ace_flags_get(prevacep);
1462 	mode_t extramode = (mode >> 3) & 07;
1463 	mode_t ownermode = (mode >> 6);
1464 
1465 	if (prevflags & ACE_IDENTIFIER_GROUP) {
1466 
1467 		extramode &= ~ownermode;
1468 
1469 		if (extramode) {
1470 			if (extramode & S_IROTH) {
1471 				prevmask &= ~ACE_READ_DATA;
1472 				mask &= ~ACE_READ_DATA;
1473 			}
1474 			if (extramode & S_IWOTH) {
1475 				prevmask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1476 				mask &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1477 			}
1478 			if (extramode & S_IXOTH) {
1479 				prevmask  &= ~ACE_EXECUTE;
1480 				mask &= ~ACE_EXECUTE;
1481 			}
1482 		}
1483 	}
1484 	aclp->z_ops.ace_mask_set(acep, mask);
1485 	aclp->z_ops.ace_mask_set(prevacep, prevmask);
1486 }
1487 
1488 /*
1489  * Apply the chmod algorithm as described
1490  * in PSARC/2002/240
1491  */
1492 static void
1493 zfs_acl_chmod(zfsvfs_t *zfsvfs, uint64_t uid,
1494     uint64_t mode, zfs_acl_t *aclp)
1495 {
1496 	void		*acep = NULL, *prevacep = NULL;
1497 	uint64_t	who;
1498 	int 		i;
1499 	int 		entry_type;
1500 	int 		reuse_deny;
1501 	int 		need_canonical_six = 1;
1502 	uint16_t	iflags, type;
1503 	uint32_t	access_mask;
1504 
1505 	/*
1506 	 * If discard then just discard all ACL nodes which
1507 	 * represent the ACEs.
1508 	 *
1509 	 * New owner@/group@/everone@ ACEs will be added
1510 	 * later.
1511 	 */
1512 	if (zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1513 		zfs_acl_release_nodes(aclp);
1514 
1515 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1516 	    &iflags, &type)) {
1517 
1518 		entry_type = (iflags & ACE_TYPE_FLAGS);
1519 		iflags = (iflags & ALL_INHERIT);
1520 
1521 		if ((type != ALLOW && type != DENY) ||
1522 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1523 			if (iflags)
1524 				aclp->z_hints |= ZFS_INHERIT_ACE;
1525 			switch (type) {
1526 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1527 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1528 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1529 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1530 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1531 				break;
1532 			}
1533 			goto nextace;
1534 		}
1535 
1536 		/*
1537 		 * Need to split ace into two?
1538 		 */
1539 		if ((iflags & (ACE_FILE_INHERIT_ACE|
1540 		    ACE_DIRECTORY_INHERIT_ACE)) &&
1541 		    (!(iflags & ACE_INHERIT_ONLY_ACE))) {
1542 			zfs_acl_split_ace(aclp, acep);
1543 			aclp->z_hints |= ZFS_INHERIT_ACE;
1544 			goto nextace;
1545 		}
1546 
1547 		if (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1548 		    (entry_type == OWNING_GROUP)) {
1549 			access_mask &= ~OGE_CLEAR;
1550 			aclp->z_ops.ace_mask_set(acep, access_mask);
1551 			goto nextace;
1552 		} else {
1553 			reuse_deny = B_TRUE;
1554 			if (type == ALLOW) {
1555 
1556 				/*
1557 				 * Check preceding ACE if any, to see
1558 				 * if we need to prepend a DENY ACE.
1559 				 * This is only applicable when the acl_mode
1560 				 * property == groupmask.
1561 				 */
1562 				if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK) {
1563 
1564 					reuse_deny = zfs_reuse_deny(aclp, acep,
1565 					    prevacep);
1566 
1567 					if (!reuse_deny) {
1568 						prevacep =
1569 						    zfs_acl_prepend_deny(uid,
1570 						    aclp, acep, mode);
1571 					} else {
1572 						zfs_acl_prepend_fixup(
1573 						    aclp, prevacep,
1574 						    acep, mode, uid);
1575 					}
1576 					zfs_fixup_group_entries(aclp, acep,
1577 					    prevacep, mode);
1578 				}
1579 			}
1580 		}
1581 nextace:
1582 		prevacep = acep;
1583 	}
1584 
1585 	/*
1586 	 * Check out last six aces, if we have six.
1587 	 */
1588 
1589 	if (aclp->z_acl_count >= 6) {
1590 		if (zfs_have_canonical_six(aclp)) {
1591 			need_canonical_six = 0;
1592 		}
1593 	}
1594 
1595 	if (need_canonical_six) {
1596 		size_t abstract_size = aclp->z_ops.ace_abstract_size();
1597 		void *zacep;
1598 		zfs_acl_node_t *aclnode =
1599 		    zfs_acl_node_alloc(abstract_size * 6);
1600 
1601 		aclnode->z_size = abstract_size * 6;
1602 		aclnode->z_ace_count = 6;
1603 		aclp->z_acl_bytes += aclnode->z_size;
1604 		list_insert_tail(&aclp->z_acl, aclnode);
1605 
1606 		zacep = aclnode->z_acldata;
1607 
1608 		i = 0;
1609 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1610 		    0, DENY, -1, ACE_OWNER);
1611 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1612 		    OWNER_ALLOW_MASK, ALLOW, -1, ACE_OWNER);
1613 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1614 		    DENY, -1, OWNING_GROUP);
1615 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++), 0,
1616 		    ALLOW, -1, OWNING_GROUP);
1617 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1618 		    EVERYONE_DENY_MASK, DENY, -1, ACE_EVERYONE);
1619 		zfs_set_ace(aclp, (caddr_t)zacep + (abstract_size * i++),
1620 		    EVERYONE_ALLOW_MASK, ALLOW, -1, ACE_EVERYONE);
1621 		aclp->z_acl_count += 6;
1622 	}
1623 
1624 	zfs_acl_fixup_canonical_six(aclp, mode);
1625 }
1626 
1627 int
1628 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1629 {
1630 	int error;
1631 
1632 	mutex_enter(&zp->z_lock);
1633 	mutex_enter(&zp->z_acl_lock);
1634 	*aclp = NULL;
1635 	error = zfs_acl_node_read(zp, aclp, B_TRUE);
1636 	if (error == 0) {
1637 		(*aclp)->z_hints = zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS;
1638 		zfs_acl_chmod(zp->z_zfsvfs, zp->z_phys->zp_uid, mode, *aclp);
1639 		zp->z_acl_cached = *aclp;
1640 	}
1641 	mutex_exit(&zp->z_acl_lock);
1642 	mutex_exit(&zp->z_lock);
1643 	return (error);
1644 }
1645 
1646 /*
1647  * strip off write_owner and write_acl
1648  */
1649 static void
1650 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1651 {
1652 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1653 
1654 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1655 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1656 		mask &= ~RESTRICTED_CLEAR;
1657 		aclp->z_ops.ace_mask_set(acep, mask);
1658 	}
1659 }
1660 
1661 /*
1662  * Should ACE be inherited?
1663  */
1664 static int
1665 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1666 {
1667 	int	iflags = (acep_flags & 0xf);
1668 
1669 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1670 		return (1);
1671 	else if (iflags & ACE_FILE_INHERIT_ACE)
1672 		return (!((vtype == VDIR) &&
1673 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1674 	return (0);
1675 }
1676 
1677 /*
1678  * inherit inheritable ACEs from parent
1679  */
1680 static zfs_acl_t *
1681 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1682     uint64_t mode, boolean_t *need_chmod)
1683 {
1684 	void		*pacep;
1685 	void		*acep, *acep2;
1686 	zfs_acl_node_t  *aclnode, *aclnode2;
1687 	zfs_acl_t	*aclp = NULL;
1688 	uint64_t	who;
1689 	uint32_t	access_mask;
1690 	uint16_t	iflags, newflags, type;
1691 	size_t		ace_size;
1692 	void		*data1, *data2;
1693 	size_t		data1sz, data2sz;
1694 	boolean_t	vdir = vtype == VDIR;
1695 	boolean_t	vreg = vtype == VREG;
1696 	boolean_t	passthrough, passthrough_x, noallow;
1697 
1698 	passthrough_x =
1699 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1700 	passthrough = passthrough_x ||
1701 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1702 	noallow =
1703 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1704 
1705 	*need_chmod = B_TRUE;
1706 	pacep = NULL;
1707 	aclp = zfs_acl_alloc(paclp->z_version);
1708 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD)
1709 		return (aclp);
1710 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1711 	    &access_mask, &iflags, &type)) {
1712 
1713 		/*
1714 		 * don't inherit bogus ACEs
1715 		 */
1716 		if (!zfs_acl_valid_ace_type(type, iflags))
1717 			continue;
1718 
1719 		if (noallow && type == ALLOW)
1720 			continue;
1721 
1722 		ace_size = aclp->z_ops.ace_size(pacep);
1723 
1724 		if (!zfs_ace_can_use(vtype, iflags))
1725 			continue;
1726 
1727 		/*
1728 		 * If owner@, group@, or everyone@ inheritable
1729 		 * then zfs_acl_chmod() isn't needed.
1730 		 */
1731 		if (passthrough &&
1732 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1733 		    ((iflags & OWNING_GROUP) ==
1734 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1735 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1736 			*need_chmod = B_FALSE;
1737 
1738 			if (!vdir && passthrough_x &&
1739 			    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1740 				access_mask &= ~ACE_EXECUTE;
1741 			}
1742 		}
1743 
1744 		aclnode = zfs_acl_node_alloc(ace_size);
1745 		list_insert_tail(&aclp->z_acl, aclnode);
1746 		acep = aclnode->z_acldata;
1747 
1748 		zfs_set_ace(aclp, acep, access_mask, type,
1749 		    who, iflags|ACE_INHERITED_ACE);
1750 
1751 		/*
1752 		 * Copy special opaque data if any
1753 		 */
1754 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1755 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1756 			    &data2)) == data1sz);
1757 			bcopy(data1, data2, data2sz);
1758 		}
1759 		aclp->z_acl_count++;
1760 		aclnode->z_ace_count++;
1761 		aclp->z_acl_bytes += aclnode->z_size;
1762 		newflags = aclp->z_ops.ace_flags_get(acep);
1763 
1764 		if (vdir)
1765 			aclp->z_hints |= ZFS_INHERIT_ACE;
1766 
1767 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1768 			newflags &= ~ALL_INHERIT;
1769 			aclp->z_ops.ace_flags_set(acep,
1770 			    newflags|ACE_INHERITED_ACE);
1771 			zfs_restricted_update(zfsvfs, aclp, acep);
1772 			continue;
1773 		}
1774 
1775 		ASSERT(vdir);
1776 
1777 		newflags = aclp->z_ops.ace_flags_get(acep);
1778 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1779 		    ACE_DIRECTORY_INHERIT_ACE)) !=
1780 		    ACE_FILE_INHERIT_ACE) {
1781 			aclnode2 = zfs_acl_node_alloc(ace_size);
1782 			list_insert_tail(&aclp->z_acl, aclnode2);
1783 			acep2 = aclnode2->z_acldata;
1784 			zfs_set_ace(aclp, acep2,
1785 			    access_mask, type, who,
1786 			    iflags|ACE_INHERITED_ACE);
1787 			newflags |= ACE_INHERIT_ONLY_ACE;
1788 			aclp->z_ops.ace_flags_set(acep, newflags);
1789 			newflags &= ~ALL_INHERIT;
1790 			aclp->z_ops.ace_flags_set(acep2,
1791 			    newflags|ACE_INHERITED_ACE);
1792 
1793 			/*
1794 			 * Copy special opaque data if any
1795 			 */
1796 			if ((data1sz = aclp->z_ops.ace_data(acep,
1797 			    &data1)) != 0) {
1798 				VERIFY((data2sz =
1799 				    aclp->z_ops.ace_data(acep2,
1800 				    &data2)) == data1sz);
1801 				bcopy(data1, data2, data1sz);
1802 			}
1803 			aclp->z_acl_count++;
1804 			aclnode2->z_ace_count++;
1805 			aclp->z_acl_bytes += aclnode->z_size;
1806 			zfs_restricted_update(zfsvfs, aclp, acep2);
1807 		} else {
1808 			newflags |= ACE_INHERIT_ONLY_ACE;
1809 			aclp->z_ops.ace_flags_set(acep,
1810 			    newflags|ACE_INHERITED_ACE);
1811 		}
1812 	}
1813 	return (aclp);
1814 }
1815 
1816 /*
1817  * Create file system object initial permissions
1818  * including inheritable ACEs.
1819  */
1820 int
1821 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1822     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1823 {
1824 	int		error;
1825 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1826 	zfs_acl_t	*paclp;
1827 	gid_t		gid;
1828 	boolean_t	need_chmod = B_TRUE;
1829 
1830 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1831 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1832 
1833 	if (vsecp)
1834 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1835 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1836 			return (error);
1837 
1838 	/*
1839 	 * Determine uid and gid.
1840 	 */
1841 	if ((flag & (IS_ROOT_NODE | IS_REPLAY)) ||
1842 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1843 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1844 		    (uint64_t)vap->va_uid, cr,
1845 		    ZFS_OWNER, &acl_ids->z_fuidp);
1846 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1847 		    (uint64_t)vap->va_gid, cr,
1848 		    ZFS_GROUP, &acl_ids->z_fuidp);
1849 		gid = vap->va_gid;
1850 	} else {
1851 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1852 		    cr, &acl_ids->z_fuidp);
1853 		acl_ids->z_fgid = 0;
1854 		if (vap->va_mask & AT_GID)  {
1855 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1856 			    (uint64_t)vap->va_gid,
1857 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1858 			gid = vap->va_gid;
1859 			if (acl_ids->z_fgid != dzp->z_phys->zp_gid &&
1860 			    !groupmember(vap->va_gid, cr) &&
1861 			    secpolicy_vnode_create_gid(cr) != 0)
1862 				acl_ids->z_fgid = 0;
1863 		}
1864 		if (acl_ids->z_fgid == 0) {
1865 			if (dzp->z_phys->zp_mode & S_ISGID) {
1866 				acl_ids->z_fgid = dzp->z_phys->zp_gid;
1867 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1868 				    cr, ZFS_GROUP);
1869 			} else {
1870 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1871 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1872 				gid = crgetgid(cr);
1873 			}
1874 		}
1875 	}
1876 
1877 	/*
1878 	 * If we're creating a directory, and the parent directory has the
1879 	 * set-GID bit set, set in on the new directory.
1880 	 * Otherwise, if the user is neither privileged nor a member of the
1881 	 * file's new group, clear the file's set-GID bit.
1882 	 */
1883 
1884 	if (!(flag & IS_ROOT_NODE) && (dzp->z_phys->zp_mode & S_ISGID) &&
1885 	    (vap->va_type == VDIR)) {
1886 		acl_ids->z_mode |= S_ISGID;
1887 	} else {
1888 		if ((acl_ids->z_mode & S_ISGID) &&
1889 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1890 			acl_ids->z_mode &= ~S_ISGID;
1891 	}
1892 
1893 	if (acl_ids->z_aclp == NULL) {
1894 		mutex_enter(&dzp->z_lock);
1895 		if (!(flag & IS_ROOT_NODE) && (ZTOV(dzp)->v_type == VDIR &&
1896 		    (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE)) &&
1897 		    !(dzp->z_phys->zp_flags & ZFS_XATTR)) {
1898 			mutex_enter(&dzp->z_acl_lock);
1899 			VERIFY(0 == zfs_acl_node_read(dzp, &paclp, B_FALSE));
1900 			mutex_exit(&dzp->z_acl_lock);
1901 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1902 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1903 		} else {
1904 			acl_ids->z_aclp =
1905 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1906 		}
1907 		mutex_exit(&dzp->z_lock);
1908 		if (need_chmod) {
1909 			acl_ids->z_aclp->z_hints = (vap->va_type == VDIR) ?
1910 			    ZFS_ACL_AUTO_INHERIT : 0;
1911 			zfs_acl_chmod(zfsvfs, acl_ids->z_fuid,
1912 			    acl_ids->z_mode, acl_ids->z_aclp);
1913 		}
1914 	}
1915 
1916 	return (0);
1917 }
1918 
1919 /*
1920  * Free ACL and fuid_infop, but not the acl_ids structure
1921  */
1922 void
1923 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1924 {
1925 	if (acl_ids->z_aclp)
1926 		zfs_acl_free(acl_ids->z_aclp);
1927 	if (acl_ids->z_fuidp)
1928 		zfs_fuid_info_free(acl_ids->z_fuidp);
1929 	acl_ids->z_aclp = NULL;
1930 	acl_ids->z_fuidp = NULL;
1931 }
1932 
1933 boolean_t
1934 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1935 {
1936 	return (zfs_usergroup_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1937 	    zfs_usergroup_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1938 }
1939 
1940 /*
1941  * Retrieve a files ACL
1942  */
1943 int
1944 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1945 {
1946 	zfs_acl_t	*aclp;
1947 	ulong_t		mask;
1948 	int		error;
1949 	int 		count = 0;
1950 	int		largeace = 0;
1951 
1952 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1953 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1954 
1955 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1956 		return (error);
1957 
1958 	if (mask == 0)
1959 		return (ENOSYS);
1960 
1961 	mutex_enter(&zp->z_acl_lock);
1962 
1963 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
1964 	if (error != 0) {
1965 		mutex_exit(&zp->z_acl_lock);
1966 		return (error);
1967 	}
1968 
1969 	/*
1970 	 * Scan ACL to determine number of ACEs
1971 	 */
1972 	if ((zp->z_phys->zp_flags & ZFS_ACL_OBJ_ACE) &&
1973 	    !(mask & VSA_ACE_ALLTYPES)) {
1974 		void *zacep = NULL;
1975 		uint64_t who;
1976 		uint32_t access_mask;
1977 		uint16_t type, iflags;
1978 
1979 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1980 		    &who, &access_mask, &iflags, &type)) {
1981 			switch (type) {
1982 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1983 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1984 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1985 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1986 				largeace++;
1987 				continue;
1988 			default:
1989 				count++;
1990 			}
1991 		}
1992 		vsecp->vsa_aclcnt = count;
1993 	} else
1994 		count = aclp->z_acl_count;
1995 
1996 	if (mask & VSA_ACECNT) {
1997 		vsecp->vsa_aclcnt = count;
1998 	}
1999 
2000 	if (mask & VSA_ACE) {
2001 		size_t aclsz;
2002 
2003 		zfs_acl_node_t *aclnode = list_head(&aclp->z_acl);
2004 
2005 		aclsz = count * sizeof (ace_t) +
2006 		    sizeof (ace_object_t) * largeace;
2007 
2008 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2009 		vsecp->vsa_aclentsz = aclsz;
2010 
2011 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2012 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
2013 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2014 		else {
2015 			bcopy(aclnode->z_acldata, vsecp->vsa_aclentp,
2016 			    count * sizeof (ace_t));
2017 		}
2018 	}
2019 	if (mask & VSA_ACE_ACLFLAGS) {
2020 		vsecp->vsa_aclflags = 0;
2021 		if (zp->z_phys->zp_flags & ZFS_ACL_DEFAULTED)
2022 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
2023 		if (zp->z_phys->zp_flags & ZFS_ACL_PROTECTED)
2024 			vsecp->vsa_aclflags |= ACL_PROTECTED;
2025 		if (zp->z_phys->zp_flags & ZFS_ACL_AUTO_INHERIT)
2026 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2027 	}
2028 
2029 	mutex_exit(&zp->z_acl_lock);
2030 
2031 	return (0);
2032 }
2033 
2034 int
2035 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
2036     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2037 {
2038 	zfs_acl_t *aclp;
2039 	zfs_acl_node_t *aclnode;
2040 	int aclcnt = vsecp->vsa_aclcnt;
2041 	int error;
2042 
2043 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2044 		return (EINVAL);
2045 
2046 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2047 
2048 	aclp->z_hints = 0;
2049 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2050 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2051 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
2052 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2053 		    aclcnt, &aclnode->z_size)) != 0) {
2054 			zfs_acl_free(aclp);
2055 			zfs_acl_node_free(aclnode);
2056 			return (error);
2057 		}
2058 	} else {
2059 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
2060 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2061 		    &aclnode->z_size, fuidp, cr)) != 0) {
2062 			zfs_acl_free(aclp);
2063 			zfs_acl_node_free(aclnode);
2064 			return (error);
2065 		}
2066 	}
2067 	aclp->z_acl_bytes = aclnode->z_size;
2068 	aclnode->z_ace_count = aclcnt;
2069 	aclp->z_acl_count = aclcnt;
2070 	list_insert_head(&aclp->z_acl, aclnode);
2071 
2072 	/*
2073 	 * If flags are being set then add them to z_hints
2074 	 */
2075 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2076 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2077 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2078 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2079 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2080 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2081 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2082 	}
2083 
2084 	*zaclp = aclp;
2085 
2086 	return (0);
2087 }
2088 
2089 /*
2090  * Set a files ACL
2091  */
2092 int
2093 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2094 {
2095 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2096 	zilog_t		*zilog = zfsvfs->z_log;
2097 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2098 	dmu_tx_t	*tx;
2099 	int		error;
2100 	zfs_acl_t	*aclp;
2101 	zfs_fuid_info_t	*fuidp = NULL;
2102 	boolean_t	fuid_dirtied;
2103 
2104 	if (mask == 0)
2105 		return (ENOSYS);
2106 
2107 	if (zp->z_phys->zp_flags & ZFS_IMMUTABLE)
2108 		return (EPERM);
2109 
2110 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
2111 		return (error);
2112 
2113 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
2114 	    &aclp);
2115 	if (error)
2116 		return (error);
2117 
2118 	/*
2119 	 * If ACL wide flags aren't being set then preserve any
2120 	 * existing flags.
2121 	 */
2122 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2123 		aclp->z_hints |= (zp->z_phys->zp_flags & V4_ACL_WIDE_FLAGS);
2124 	}
2125 top:
2126 	mutex_enter(&zp->z_lock);
2127 	mutex_enter(&zp->z_acl_lock);
2128 
2129 	tx = dmu_tx_create(zfsvfs->z_os);
2130 	dmu_tx_hold_bonus(tx, zp->z_id);
2131 
2132 	if (zp->z_phys->zp_acl.z_acl_extern_obj) {
2133 		/* Are we upgrading ACL? */
2134 		if (zfsvfs->z_version <= ZPL_VERSION_FUID &&
2135 		    zp->z_phys->zp_acl.z_acl_version ==
2136 		    ZFS_ACL_VERSION_INITIAL) {
2137 			dmu_tx_hold_free(tx,
2138 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2139 			    0, DMU_OBJECT_END);
2140 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT,
2141 			    0, aclp->z_acl_bytes);
2142 		} else {
2143 			dmu_tx_hold_write(tx,
2144 			    zp->z_phys->zp_acl.z_acl_extern_obj,
2145 			    0, aclp->z_acl_bytes);
2146 		}
2147 	} else if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2148 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2149 	}
2150 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2151 	if (fuid_dirtied)
2152 		zfs_fuid_txhold(zfsvfs, tx);
2153 
2154 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2155 	if (error) {
2156 		mutex_exit(&zp->z_acl_lock);
2157 		mutex_exit(&zp->z_lock);
2158 
2159 		if (error == ERESTART) {
2160 			dmu_tx_wait(tx);
2161 			dmu_tx_abort(tx);
2162 			goto top;
2163 		}
2164 		dmu_tx_abort(tx);
2165 		zfs_acl_free(aclp);
2166 		return (error);
2167 	}
2168 
2169 	error = zfs_aclset_common(zp, aclp, cr, tx);
2170 	ASSERT(error == 0);
2171 
2172 	if (fuid_dirtied)
2173 		zfs_fuid_sync(zfsvfs, tx);
2174 
2175 	zfs_time_stamper_locked(zp, STATE_CHANGED, tx);
2176 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2177 
2178 	if (fuidp)
2179 		zfs_fuid_info_free(fuidp);
2180 	zp->z_acl_cached = aclp;
2181 	dmu_tx_commit(tx);
2182 done:
2183 	mutex_exit(&zp->z_acl_lock);
2184 	mutex_exit(&zp->z_lock);
2185 
2186 	return (error);
2187 }
2188 
2189 /*
2190  * Check accesses of interest (AoI) against attributes of the dataset
2191  * such as read-only.  Returns zero if no AoI conflict with dataset
2192  * attributes, otherwise an appropriate errno is returned.
2193  */
2194 static int
2195 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2196 {
2197 	if ((v4_mode & WRITE_MASK) &&
2198 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2199 	    (!IS_DEVVP(ZTOV(zp)) ||
2200 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2201 		return (EROFS);
2202 	}
2203 
2204 	/*
2205 	 * Only check for READONLY on non-directories.
2206 	 */
2207 	if ((v4_mode & WRITE_MASK_DATA) &&
2208 	    (((ZTOV(zp)->v_type != VDIR) &&
2209 	    (zp->z_phys->zp_flags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2210 	    (ZTOV(zp)->v_type == VDIR &&
2211 	    (zp->z_phys->zp_flags & ZFS_IMMUTABLE)))) {
2212 		return (EPERM);
2213 	}
2214 
2215 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2216 	    (zp->z_phys->zp_flags & ZFS_NOUNLINK)) {
2217 		return (EPERM);
2218 	}
2219 
2220 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2221 	    (zp->z_phys->zp_flags & ZFS_AV_QUARANTINED))) {
2222 		return (EACCES);
2223 	}
2224 
2225 	return (0);
2226 }
2227 
2228 /*
2229  * The primary usage of this function is to loop through all of the
2230  * ACEs in the znode, determining what accesses of interest (AoI) to
2231  * the caller are allowed or denied.  The AoI are expressed as bits in
2232  * the working_mode parameter.  As each ACE is processed, bits covered
2233  * by that ACE are removed from the working_mode.  This removal
2234  * facilitates two things.  The first is that when the working mode is
2235  * empty (= 0), we know we've looked at all the AoI. The second is
2236  * that the ACE interpretation rules don't allow a later ACE to undo
2237  * something granted or denied by an earlier ACE.  Removing the
2238  * discovered access or denial enforces this rule.  At the end of
2239  * processing the ACEs, all AoI that were found to be denied are
2240  * placed into the working_mode, giving the caller a mask of denied
2241  * accesses.  Returns:
2242  *	0		if all AoI granted
2243  *	EACCESS 	if the denied mask is non-zero
2244  *	other error	if abnormal failure (e.g., IO error)
2245  *
2246  * A secondary usage of the function is to determine if any of the
2247  * AoI are granted.  If an ACE grants any access in
2248  * the working_mode, we immediately short circuit out of the function.
2249  * This mode is chosen by setting anyaccess to B_TRUE.  The
2250  * working_mode is not a denied access mask upon exit if the function
2251  * is used in this manner.
2252  */
2253 static int
2254 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2255     boolean_t anyaccess, cred_t *cr)
2256 {
2257 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2258 	zfs_acl_t	*aclp;
2259 	int		error;
2260 	uid_t		uid = crgetuid(cr);
2261 	uint64_t 	who;
2262 	uint16_t	type, iflags;
2263 	uint16_t	entry_type;
2264 	uint32_t	access_mask;
2265 	uint32_t	deny_mask = 0;
2266 	zfs_ace_hdr_t	*acep = NULL;
2267 	boolean_t	checkit;
2268 	uid_t		fowner;
2269 	uid_t		gowner;
2270 
2271 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2272 
2273 	mutex_enter(&zp->z_acl_lock);
2274 
2275 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2276 	if (error != 0) {
2277 		mutex_exit(&zp->z_acl_lock);
2278 		return (error);
2279 	}
2280 
2281 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2282 	    &iflags, &type)) {
2283 		uint32_t mask_matched;
2284 
2285 		if (!zfs_acl_valid_ace_type(type, iflags))
2286 			continue;
2287 
2288 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2289 			continue;
2290 
2291 		/* Skip ACE if it does not affect any AoI */
2292 		mask_matched = (access_mask & *working_mode);
2293 		if (!mask_matched)
2294 			continue;
2295 
2296 		entry_type = (iflags & ACE_TYPE_FLAGS);
2297 
2298 		checkit = B_FALSE;
2299 
2300 		switch (entry_type) {
2301 		case ACE_OWNER:
2302 			if (uid == fowner)
2303 				checkit = B_TRUE;
2304 			break;
2305 		case OWNING_GROUP:
2306 			who = gowner;
2307 			/*FALLTHROUGH*/
2308 		case ACE_IDENTIFIER_GROUP:
2309 			checkit = zfs_groupmember(zfsvfs, who, cr);
2310 			break;
2311 		case ACE_EVERYONE:
2312 			checkit = B_TRUE;
2313 			break;
2314 
2315 		/* USER Entry */
2316 		default:
2317 			if (entry_type == 0) {
2318 				uid_t newid;
2319 
2320 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2321 				    ZFS_ACE_USER);
2322 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2323 				    uid == newid)
2324 					checkit = B_TRUE;
2325 				break;
2326 			} else {
2327 				mutex_exit(&zp->z_acl_lock);
2328 				return (EIO);
2329 			}
2330 		}
2331 
2332 		if (checkit) {
2333 			if (type == DENY) {
2334 				DTRACE_PROBE3(zfs__ace__denies,
2335 				    znode_t *, zp,
2336 				    zfs_ace_hdr_t *, acep,
2337 				    uint32_t, mask_matched);
2338 				deny_mask |= mask_matched;
2339 			} else {
2340 				DTRACE_PROBE3(zfs__ace__allows,
2341 				    znode_t *, zp,
2342 				    zfs_ace_hdr_t *, acep,
2343 				    uint32_t, mask_matched);
2344 				if (anyaccess) {
2345 					mutex_exit(&zp->z_acl_lock);
2346 					zfs_acl_free(aclp);
2347 					return (0);
2348 				}
2349 			}
2350 			*working_mode &= ~mask_matched;
2351 		}
2352 
2353 		/* Are we done? */
2354 		if (*working_mode == 0)
2355 			break;
2356 	}
2357 
2358 	mutex_exit(&zp->z_acl_lock);
2359 
2360 	/* Put the found 'denies' back on the working mode */
2361 	if (deny_mask) {
2362 		*working_mode |= deny_mask;
2363 		return (EACCES);
2364 	} else if (*working_mode) {
2365 		return (-1);
2366 	}
2367 
2368 	return (0);
2369 }
2370 
2371 /*
2372  * Return true if any access whatsoever granted, we don't actually
2373  * care what access is granted.
2374  */
2375 boolean_t
2376 zfs_has_access(znode_t *zp, cred_t *cr)
2377 {
2378 	uint32_t have = ACE_ALL_PERMS;
2379 
2380 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2381 		uid_t		owner;
2382 
2383 		owner = zfs_fuid_map_id(zp->z_zfsvfs,
2384 		    zp->z_phys->zp_uid, cr, ZFS_OWNER);
2385 
2386 		return (
2387 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VREAD) == 0 ||
2388 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VWRITE) == 0 ||
2389 		    secpolicy_vnode_access(cr, ZTOV(zp), owner, VEXEC) == 0 ||
2390 		    secpolicy_vnode_chown(cr, owner) == 0 ||
2391 		    secpolicy_vnode_setdac(cr, owner) == 0 ||
2392 		    secpolicy_vnode_remove(cr) == 0);
2393 	}
2394 	return (B_TRUE);
2395 }
2396 
2397 static int
2398 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2399     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2400 {
2401 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2402 	int err;
2403 
2404 	*working_mode = v4_mode;
2405 	*check_privs = B_TRUE;
2406 
2407 	/*
2408 	 * Short circuit empty requests
2409 	 */
2410 	if (v4_mode == 0 || zfsvfs->z_replay) {
2411 		*working_mode = 0;
2412 		return (0);
2413 	}
2414 
2415 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2416 		*check_privs = B_FALSE;
2417 		return (err);
2418 	}
2419 
2420 	/*
2421 	 * The caller requested that the ACL check be skipped.  This
2422 	 * would only happen if the caller checked VOP_ACCESS() with a
2423 	 * 32 bit ACE mask and already had the appropriate permissions.
2424 	 */
2425 	if (skipaclchk) {
2426 		*working_mode = 0;
2427 		return (0);
2428 	}
2429 
2430 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2431 }
2432 
2433 static int
2434 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2435     cred_t *cr)
2436 {
2437 	if (*working_mode != ACE_WRITE_DATA)
2438 		return (EACCES);
2439 
2440 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2441 	    check_privs, B_FALSE, cr));
2442 }
2443 
2444 int
2445 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2446 {
2447 	boolean_t owner = B_FALSE;
2448 	boolean_t groupmbr = B_FALSE;
2449 	boolean_t is_attr;
2450 	uid_t fowner;
2451 	uid_t gowner;
2452 	uid_t uid = crgetuid(cr);
2453 	int error;
2454 
2455 	if (zdp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
2456 		return (EACCES);
2457 
2458 	is_attr = ((zdp->z_phys->zp_flags & ZFS_XATTR) &&
2459 	    (ZTOV(zdp)->v_type == VDIR));
2460 	if (is_attr)
2461 		goto slow;
2462 
2463 	mutex_enter(&zdp->z_acl_lock);
2464 
2465 	if (zdp->z_phys->zp_flags & ZFS_NO_EXECS_DENIED) {
2466 		mutex_exit(&zdp->z_acl_lock);
2467 		return (0);
2468 	}
2469 
2470 	if (FUID_INDEX(zdp->z_phys->zp_uid) != 0 ||
2471 	    FUID_INDEX(zdp->z_phys->zp_gid) != 0) {
2472 		mutex_exit(&zdp->z_acl_lock);
2473 		goto slow;
2474 	}
2475 
2476 	fowner = (uid_t)zdp->z_phys->zp_uid;
2477 	gowner = (uid_t)zdp->z_phys->zp_gid;
2478 
2479 	if (uid == fowner) {
2480 		owner = B_TRUE;
2481 		if (zdp->z_phys->zp_mode & S_IXUSR) {
2482 			mutex_exit(&zdp->z_acl_lock);
2483 			return (0);
2484 		}
2485 	}
2486 	if (groupmember(gowner, cr)) {
2487 		groupmbr = B_TRUE;
2488 		if (zdp->z_phys->zp_mode & S_IXGRP) {
2489 			mutex_exit(&zdp->z_acl_lock);
2490 			return (0);
2491 		}
2492 	}
2493 	if (!owner && !groupmbr) {
2494 		if (zdp->z_phys->zp_mode & S_IXOTH) {
2495 			mutex_exit(&zdp->z_acl_lock);
2496 			return (0);
2497 		}
2498 	}
2499 
2500 	mutex_exit(&zdp->z_acl_lock);
2501 
2502 slow:
2503 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2504 	ZFS_ENTER(zdp->z_zfsvfs);
2505 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2506 	ZFS_EXIT(zdp->z_zfsvfs);
2507 	return (error);
2508 }
2509 
2510 /*
2511  * Determine whether Access should be granted/denied, invoking least
2512  * priv subsytem when a deny is determined.
2513  */
2514 int
2515 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2516 {
2517 	uint32_t	working_mode;
2518 	int		error;
2519 	int		is_attr;
2520 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2521 	boolean_t 	check_privs;
2522 	znode_t		*xzp;
2523 	znode_t 	*check_zp = zp;
2524 
2525 	is_attr = ((zp->z_phys->zp_flags & ZFS_XATTR) &&
2526 	    (ZTOV(zp)->v_type == VDIR));
2527 
2528 	/*
2529 	 * If attribute then validate against base file
2530 	 */
2531 	if (is_attr) {
2532 		if ((error = zfs_zget(zp->z_zfsvfs,
2533 		    zp->z_phys->zp_parent, &xzp)) != 0)	{
2534 			return (error);
2535 		}
2536 
2537 		check_zp = xzp;
2538 
2539 		/*
2540 		 * fixup mode to map to xattr perms
2541 		 */
2542 
2543 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2544 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2545 			mode |= ACE_WRITE_NAMED_ATTRS;
2546 		}
2547 
2548 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2549 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2550 			mode |= ACE_READ_NAMED_ATTRS;
2551 		}
2552 	}
2553 
2554 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2555 	    &check_privs, skipaclchk, cr)) == 0) {
2556 		if (is_attr)
2557 			VN_RELE(ZTOV(xzp));
2558 		return (0);
2559 	}
2560 
2561 	if (error && !check_privs) {
2562 		if (is_attr)
2563 			VN_RELE(ZTOV(xzp));
2564 		return (error);
2565 	}
2566 
2567 	if (error && (flags & V_APPEND)) {
2568 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2569 	}
2570 
2571 	if (error && check_privs) {
2572 		uid_t		owner;
2573 		mode_t		checkmode = 0;
2574 
2575 		owner = zfs_fuid_map_id(zfsvfs, check_zp->z_phys->zp_uid, cr,
2576 		    ZFS_OWNER);
2577 
2578 		/*
2579 		 * First check for implicit owner permission on
2580 		 * read_acl/read_attributes
2581 		 */
2582 
2583 		error = 0;
2584 		ASSERT(working_mode != 0);
2585 
2586 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2587 		    owner == crgetuid(cr)))
2588 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2589 
2590 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2591 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2592 			checkmode |= VREAD;
2593 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2594 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2595 			checkmode |= VWRITE;
2596 		if (working_mode & ACE_EXECUTE)
2597 			checkmode |= VEXEC;
2598 
2599 		if (checkmode)
2600 			error = secpolicy_vnode_access(cr, ZTOV(check_zp),
2601 			    owner, checkmode);
2602 
2603 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2604 			error = secpolicy_vnode_chown(cr, owner);
2605 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2606 			error = secpolicy_vnode_setdac(cr, owner);
2607 
2608 		if (error == 0 && (working_mode &
2609 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2610 			error = secpolicy_vnode_remove(cr);
2611 
2612 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2613 			error = secpolicy_vnode_chown(cr, owner);
2614 		}
2615 		if (error == 0) {
2616 			/*
2617 			 * See if any bits other than those already checked
2618 			 * for are still present.  If so then return EACCES
2619 			 */
2620 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2621 				error = EACCES;
2622 			}
2623 		}
2624 	}
2625 
2626 	if (is_attr)
2627 		VN_RELE(ZTOV(xzp));
2628 
2629 	return (error);
2630 }
2631 
2632 /*
2633  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2634  * native ACL format and call zfs_zaccess()
2635  */
2636 int
2637 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2638 {
2639 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2640 }
2641 
2642 /*
2643  * Access function for secpolicy_vnode_setattr
2644  */
2645 int
2646 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2647 {
2648 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2649 
2650 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2651 }
2652 
2653 static int
2654 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2655     mode_t missing_perms, cred_t *cr)
2656 {
2657 	int error;
2658 	uid_t downer;
2659 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2660 
2661 	downer = zfs_fuid_map_id(zfsvfs, dzp->z_phys->zp_uid, cr, ZFS_OWNER);
2662 
2663 	error = secpolicy_vnode_access(cr, ZTOV(dzp), downer, missing_perms);
2664 
2665 	if (error == 0)
2666 		error = zfs_sticky_remove_access(dzp, zp, cr);
2667 
2668 	return (error);
2669 }
2670 
2671 /*
2672  * Determine whether Access should be granted/deny, without
2673  * consulting least priv subsystem.
2674  *
2675  *
2676  * The following chart is the recommended NFSv4 enforcement for
2677  * ability to delete an object.
2678  *
2679  *      -------------------------------------------------------
2680  *      |   Parent Dir  |           Target Object Permissions |
2681  *      |  permissions  |                                     |
2682  *      -------------------------------------------------------
2683  *      |               | ACL Allows | ACL Denies| Delete     |
2684  *      |               |  Delete    |  Delete   | unspecified|
2685  *      -------------------------------------------------------
2686  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2687  *      |  DELETE_CHILD |                                     |
2688  *      -------------------------------------------------------
2689  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2690  *      |  DELETE_CHILD |            |           |            |
2691  *      -------------------------------------------------------
2692  *      | ACL specifies |            |           |            |
2693  *      | only allow    | Permit     | Permit    | Permit     |
2694  *      | write and     |            |           |            |
2695  *      | execute       |            |           |            |
2696  *      -------------------------------------------------------
2697  *      | ACL denies    |            |           |            |
2698  *      | write and     | Permit     | Deny      | Deny       |
2699  *      | execute       |            |           |            |
2700  *      -------------------------------------------------------
2701  *         ^
2702  *         |
2703  *         No search privilege, can't even look up file?
2704  *
2705  */
2706 int
2707 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2708 {
2709 	uint32_t dzp_working_mode = 0;
2710 	uint32_t zp_working_mode = 0;
2711 	int dzp_error, zp_error;
2712 	mode_t missing_perms;
2713 	boolean_t dzpcheck_privs = B_TRUE;
2714 	boolean_t zpcheck_privs = B_TRUE;
2715 
2716 	/*
2717 	 * We want specific DELETE permissions to
2718 	 * take precedence over WRITE/EXECUTE.  We don't
2719 	 * want an ACL such as this to mess us up.
2720 	 * user:joe:write_data:deny,user:joe:delete:allow
2721 	 *
2722 	 * However, deny permissions may ultimately be overridden
2723 	 * by secpolicy_vnode_access().
2724 	 *
2725 	 * We will ask for all of the necessary permissions and then
2726 	 * look at the working modes from the directory and target object
2727 	 * to determine what was found.
2728 	 */
2729 
2730 	if (zp->z_phys->zp_flags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2731 		return (EPERM);
2732 
2733 	/*
2734 	 * First row
2735 	 * If the directory permissions allow the delete, we are done.
2736 	 */
2737 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2738 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2739 		return (0);
2740 
2741 	/*
2742 	 * If target object has delete permission then we are done
2743 	 */
2744 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2745 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2746 		return (0);
2747 
2748 	ASSERT(dzp_error && zp_error);
2749 
2750 	if (!dzpcheck_privs)
2751 		return (dzp_error);
2752 	if (!zpcheck_privs)
2753 		return (zp_error);
2754 
2755 	/*
2756 	 * Second row
2757 	 *
2758 	 * If directory returns EACCES then delete_child was denied
2759 	 * due to deny delete_child.  In this case send the request through
2760 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2761 	 * since that *could* allow the delete based on write/execute permission
2762 	 * and we want delete permissions to override write/execute.
2763 	 */
2764 
2765 	if (dzp_error == EACCES)
2766 		return (secpolicy_vnode_remove(cr));
2767 
2768 	/*
2769 	 * Third Row
2770 	 * only need to see if we have write/execute on directory.
2771 	 */
2772 
2773 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2774 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2775 		return (zfs_sticky_remove_access(dzp, zp, cr));
2776 
2777 	if (!dzpcheck_privs)
2778 		return (dzp_error);
2779 
2780 	/*
2781 	 * Fourth row
2782 	 */
2783 
2784 	missing_perms = (dzp_working_mode & ACE_WRITE_DATA) ? VWRITE : 0;
2785 	missing_perms |= (dzp_working_mode & ACE_EXECUTE) ? VEXEC : 0;
2786 
2787 	ASSERT(missing_perms);
2788 
2789 	return (zfs_delete_final_check(zp, dzp, missing_perms, cr));
2790 
2791 }
2792 
2793 int
2794 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2795     znode_t *tzp, cred_t *cr)
2796 {
2797 	int add_perm;
2798 	int error;
2799 
2800 	if (szp->z_phys->zp_flags & ZFS_AV_QUARANTINED)
2801 		return (EACCES);
2802 
2803 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2804 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2805 
2806 	/*
2807 	 * Rename permissions are combination of delete permission +
2808 	 * add file/subdir permission.
2809 	 */
2810 
2811 	/*
2812 	 * first make sure we do the delete portion.
2813 	 *
2814 	 * If that succeeds then check for add_file/add_subdir permissions
2815 	 */
2816 
2817 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2818 		return (error);
2819 
2820 	/*
2821 	 * If we have a tzp, see if we can delete it?
2822 	 */
2823 	if (tzp) {
2824 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2825 			return (error);
2826 	}
2827 
2828 	/*
2829 	 * Now check for add permissions
2830 	 */
2831 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2832 
2833 	return (error);
2834 }
2835