xref: /freebsd/sys/contrib/openzfs/module/os/linux/zfs/zfs_acl.c (revision d4eeb02986980bf33dd56c41ceb9fc5f180c0d47)
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 https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013 by Delphix. All rights reserved.
24  * Copyright 2014 Nexenta Systems, Inc.  All rights reserved.
25  */
26 
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/sysmacros.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/fs/zfs.h>
41 #include <sys/policy.h>
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_fuid.h>
44 #include <sys/zfs_acl.h>
45 #include <sys/zfs_dir.h>
46 #include <sys/zfs_quota.h>
47 #include <sys/zfs_vfsops.h>
48 #include <sys/dmu.h>
49 #include <sys/dnode.h>
50 #include <sys/zap.h>
51 #include <sys/sa.h>
52 #include <sys/trace_acl.h>
53 #include <sys/zpl.h>
54 
55 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
56 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
57 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
58 #define	MIN_ACE_TYPE	ALLOW
59 
60 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
61 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
62     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
63 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
64     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
65 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
66     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
67 
68 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
69     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
70     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
71     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
72 
73 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
74 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
75     ACE_DELETE|ACE_DELETE_CHILD)
76 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
77 
78 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
79     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
80 
81 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
82     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
83 
84 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
85     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
86 
87 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
88 
89 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
90     ZFS_ACL_PROTECTED)
91 
92 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
93     ZFS_ACL_OBJ_ACE)
94 
95 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
96 
97 #define	IDMAP_WK_CREATOR_OWNER_UID	2147483648U
98 
99 static uint16_t
100 zfs_ace_v0_get_type(void *acep)
101 {
102 	return (((zfs_oldace_t *)acep)->z_type);
103 }
104 
105 static uint16_t
106 zfs_ace_v0_get_flags(void *acep)
107 {
108 	return (((zfs_oldace_t *)acep)->z_flags);
109 }
110 
111 static uint32_t
112 zfs_ace_v0_get_mask(void *acep)
113 {
114 	return (((zfs_oldace_t *)acep)->z_access_mask);
115 }
116 
117 static uint64_t
118 zfs_ace_v0_get_who(void *acep)
119 {
120 	return (((zfs_oldace_t *)acep)->z_fuid);
121 }
122 
123 static void
124 zfs_ace_v0_set_type(void *acep, uint16_t type)
125 {
126 	((zfs_oldace_t *)acep)->z_type = type;
127 }
128 
129 static void
130 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
131 {
132 	((zfs_oldace_t *)acep)->z_flags = flags;
133 }
134 
135 static void
136 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
137 {
138 	((zfs_oldace_t *)acep)->z_access_mask = mask;
139 }
140 
141 static void
142 zfs_ace_v0_set_who(void *acep, uint64_t who)
143 {
144 	((zfs_oldace_t *)acep)->z_fuid = who;
145 }
146 
147 static size_t
148 zfs_ace_v0_size(void *acep)
149 {
150 	(void) acep;
151 	return (sizeof (zfs_oldace_t));
152 }
153 
154 static size_t
155 zfs_ace_v0_abstract_size(void)
156 {
157 	return (sizeof (zfs_oldace_t));
158 }
159 
160 static int
161 zfs_ace_v0_mask_off(void)
162 {
163 	return (offsetof(zfs_oldace_t, z_access_mask));
164 }
165 
166 static int
167 zfs_ace_v0_data(void *acep, void **datap)
168 {
169 	(void) acep;
170 	*datap = NULL;
171 	return (0);
172 }
173 
174 static const acl_ops_t zfs_acl_v0_ops = {
175 	.ace_mask_get = zfs_ace_v0_get_mask,
176 	.ace_mask_set = zfs_ace_v0_set_mask,
177 	.ace_flags_get = zfs_ace_v0_get_flags,
178 	.ace_flags_set = zfs_ace_v0_set_flags,
179 	.ace_type_get = zfs_ace_v0_get_type,
180 	.ace_type_set = zfs_ace_v0_set_type,
181 	.ace_who_get = zfs_ace_v0_get_who,
182 	.ace_who_set = zfs_ace_v0_set_who,
183 	.ace_size = zfs_ace_v0_size,
184 	.ace_abstract_size = zfs_ace_v0_abstract_size,
185 	.ace_mask_off = zfs_ace_v0_mask_off,
186 	.ace_data = zfs_ace_v0_data
187 };
188 
189 static uint16_t
190 zfs_ace_fuid_get_type(void *acep)
191 {
192 	return (((zfs_ace_hdr_t *)acep)->z_type);
193 }
194 
195 static uint16_t
196 zfs_ace_fuid_get_flags(void *acep)
197 {
198 	return (((zfs_ace_hdr_t *)acep)->z_flags);
199 }
200 
201 static uint32_t
202 zfs_ace_fuid_get_mask(void *acep)
203 {
204 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
205 }
206 
207 static uint64_t
208 zfs_ace_fuid_get_who(void *args)
209 {
210 	uint16_t entry_type;
211 	zfs_ace_t *acep = args;
212 
213 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
214 
215 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
216 	    entry_type == ACE_EVERYONE)
217 		return (-1);
218 	return (((zfs_ace_t *)acep)->z_fuid);
219 }
220 
221 static void
222 zfs_ace_fuid_set_type(void *acep, uint16_t type)
223 {
224 	((zfs_ace_hdr_t *)acep)->z_type = type;
225 }
226 
227 static void
228 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
229 {
230 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
231 }
232 
233 static void
234 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
235 {
236 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
237 }
238 
239 static void
240 zfs_ace_fuid_set_who(void *arg, uint64_t who)
241 {
242 	zfs_ace_t *acep = arg;
243 
244 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
245 
246 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
247 	    entry_type == ACE_EVERYONE)
248 		return;
249 	acep->z_fuid = who;
250 }
251 
252 static size_t
253 zfs_ace_fuid_size(void *acep)
254 {
255 	zfs_ace_hdr_t *zacep = acep;
256 	uint16_t entry_type;
257 
258 	switch (zacep->z_type) {
259 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
260 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
261 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
262 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
263 		return (sizeof (zfs_object_ace_t));
264 	case ALLOW:
265 	case DENY:
266 		entry_type =
267 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
268 		if (entry_type == ACE_OWNER ||
269 		    entry_type == OWNING_GROUP ||
270 		    entry_type == ACE_EVERYONE)
271 			return (sizeof (zfs_ace_hdr_t));
272 		zfs_fallthrough;
273 	default:
274 		return (sizeof (zfs_ace_t));
275 	}
276 }
277 
278 static size_t
279 zfs_ace_fuid_abstract_size(void)
280 {
281 	return (sizeof (zfs_ace_hdr_t));
282 }
283 
284 static int
285 zfs_ace_fuid_mask_off(void)
286 {
287 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
288 }
289 
290 static int
291 zfs_ace_fuid_data(void *acep, void **datap)
292 {
293 	zfs_ace_t *zacep = acep;
294 	zfs_object_ace_t *zobjp;
295 
296 	switch (zacep->z_hdr.z_type) {
297 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
298 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
299 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
300 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
301 		zobjp = acep;
302 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
303 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
304 	default:
305 		*datap = NULL;
306 		return (0);
307 	}
308 }
309 
310 static const acl_ops_t zfs_acl_fuid_ops = {
311 	.ace_mask_get = zfs_ace_fuid_get_mask,
312 	.ace_mask_set = zfs_ace_fuid_set_mask,
313 	.ace_flags_get = zfs_ace_fuid_get_flags,
314 	.ace_flags_set = zfs_ace_fuid_set_flags,
315 	.ace_type_get = zfs_ace_fuid_get_type,
316 	.ace_type_set = zfs_ace_fuid_set_type,
317 	.ace_who_get = zfs_ace_fuid_get_who,
318 	.ace_who_set = zfs_ace_fuid_set_who,
319 	.ace_size = zfs_ace_fuid_size,
320 	.ace_abstract_size = zfs_ace_fuid_abstract_size,
321 	.ace_mask_off = zfs_ace_fuid_mask_off,
322 	.ace_data = zfs_ace_fuid_data
323 };
324 
325 /*
326  * The following three functions are provided for compatibility with
327  * older ZPL version in order to determine if the file use to have
328  * an external ACL and what version of ACL previously existed on the
329  * file.  Would really be nice to not need this, sigh.
330  */
331 uint64_t
332 zfs_external_acl(znode_t *zp)
333 {
334 	zfs_acl_phys_t acl_phys;
335 	int error;
336 
337 	if (zp->z_is_sa)
338 		return (0);
339 
340 	/*
341 	 * Need to deal with a potential
342 	 * race where zfs_sa_upgrade could cause
343 	 * z_isa_sa to change.
344 	 *
345 	 * If the lookup fails then the state of z_is_sa should have
346 	 * changed.
347 	 */
348 
349 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
350 	    &acl_phys, sizeof (acl_phys))) == 0)
351 		return (acl_phys.z_acl_extern_obj);
352 	else {
353 		/*
354 		 * after upgrade the SA_ZPL_ZNODE_ACL should have been
355 		 * removed
356 		 */
357 		VERIFY(zp->z_is_sa && error == ENOENT);
358 		return (0);
359 	}
360 }
361 
362 /*
363  * Determine size of ACL in bytes
364  *
365  * This is more complicated than it should be since we have to deal
366  * with old external ACLs.
367  */
368 static int
369 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
370     zfs_acl_phys_t *aclphys)
371 {
372 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
373 	uint64_t acl_count;
374 	int size;
375 	int error;
376 
377 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
378 	if (zp->z_is_sa) {
379 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
380 		    &size)) != 0)
381 			return (error);
382 		*aclsize = size;
383 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
384 		    &acl_count, sizeof (acl_count))) != 0)
385 			return (error);
386 		*aclcount = acl_count;
387 	} else {
388 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
389 		    aclphys, sizeof (*aclphys))) != 0)
390 			return (error);
391 
392 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
393 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
394 			*aclcount = aclphys->z_acl_size;
395 		} else {
396 			*aclsize = aclphys->z_acl_size;
397 			*aclcount = aclphys->z_acl_count;
398 		}
399 	}
400 	return (0);
401 }
402 
403 int
404 zfs_znode_acl_version(znode_t *zp)
405 {
406 	zfs_acl_phys_t acl_phys;
407 
408 	if (zp->z_is_sa)
409 		return (ZFS_ACL_VERSION_FUID);
410 	else {
411 		int error;
412 
413 		/*
414 		 * Need to deal with a potential
415 		 * race where zfs_sa_upgrade could cause
416 		 * z_isa_sa to change.
417 		 *
418 		 * If the lookup fails then the state of z_is_sa should have
419 		 * changed.
420 		 */
421 		if ((error = sa_lookup(zp->z_sa_hdl,
422 		    SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
423 		    &acl_phys, sizeof (acl_phys))) == 0)
424 			return (acl_phys.z_acl_version);
425 		else {
426 			/*
427 			 * After upgrade SA_ZPL_ZNODE_ACL should have
428 			 * been removed.
429 			 */
430 			VERIFY(zp->z_is_sa && error == ENOENT);
431 			return (ZFS_ACL_VERSION_FUID);
432 		}
433 	}
434 }
435 
436 static int
437 zfs_acl_version(int version)
438 {
439 	if (version < ZPL_VERSION_FUID)
440 		return (ZFS_ACL_VERSION_INITIAL);
441 	else
442 		return (ZFS_ACL_VERSION_FUID);
443 }
444 
445 static int
446 zfs_acl_version_zp(znode_t *zp)
447 {
448 	return (zfs_acl_version(ZTOZSB(zp)->z_version));
449 }
450 
451 zfs_acl_t *
452 zfs_acl_alloc(int vers)
453 {
454 	zfs_acl_t *aclp;
455 
456 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
457 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
458 	    offsetof(zfs_acl_node_t, z_next));
459 	aclp->z_version = vers;
460 	if (vers == ZFS_ACL_VERSION_FUID)
461 		aclp->z_ops = &zfs_acl_fuid_ops;
462 	else
463 		aclp->z_ops = &zfs_acl_v0_ops;
464 	return (aclp);
465 }
466 
467 zfs_acl_node_t *
468 zfs_acl_node_alloc(size_t bytes)
469 {
470 	zfs_acl_node_t *aclnode;
471 
472 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
473 	if (bytes) {
474 		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
475 		aclnode->z_allocdata = aclnode->z_acldata;
476 		aclnode->z_allocsize = bytes;
477 		aclnode->z_size = bytes;
478 	}
479 
480 	return (aclnode);
481 }
482 
483 static void
484 zfs_acl_node_free(zfs_acl_node_t *aclnode)
485 {
486 	if (aclnode->z_allocsize)
487 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
488 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
489 }
490 
491 static void
492 zfs_acl_release_nodes(zfs_acl_t *aclp)
493 {
494 	zfs_acl_node_t *aclnode;
495 
496 	while ((aclnode = list_head(&aclp->z_acl))) {
497 		list_remove(&aclp->z_acl, aclnode);
498 		zfs_acl_node_free(aclnode);
499 	}
500 	aclp->z_acl_count = 0;
501 	aclp->z_acl_bytes = 0;
502 }
503 
504 void
505 zfs_acl_free(zfs_acl_t *aclp)
506 {
507 	zfs_acl_release_nodes(aclp);
508 	list_destroy(&aclp->z_acl);
509 	kmem_free(aclp, sizeof (zfs_acl_t));
510 }
511 
512 static boolean_t
513 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
514 {
515 	uint16_t entry_type;
516 
517 	switch (type) {
518 	case ALLOW:
519 	case DENY:
520 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
521 	case ACE_SYSTEM_ALARM_ACE_TYPE:
522 		entry_type = flags & ACE_TYPE_FLAGS;
523 		return (entry_type == ACE_OWNER ||
524 		    entry_type == OWNING_GROUP ||
525 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
526 		    entry_type == ACE_IDENTIFIER_GROUP);
527 	default:
528 		if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
529 			return (B_TRUE);
530 	}
531 	return (B_FALSE);
532 }
533 
534 static boolean_t
535 zfs_ace_valid(umode_t obj_mode, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
536 {
537 	/*
538 	 * first check type of entry
539 	 */
540 
541 	if (!zfs_acl_valid_ace_type(type, iflags))
542 		return (B_FALSE);
543 
544 	switch (type) {
545 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
546 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
547 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
548 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
549 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
550 			return (B_FALSE);
551 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
552 	}
553 
554 	/*
555 	 * next check inheritance level flags
556 	 */
557 
558 	if (S_ISDIR(obj_mode) &&
559 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
560 		aclp->z_hints |= ZFS_INHERIT_ACE;
561 
562 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
563 		if ((iflags & (ACE_FILE_INHERIT_ACE|
564 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
565 			return (B_FALSE);
566 		}
567 	}
568 
569 	return (B_TRUE);
570 }
571 
572 static void *
573 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
574     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
575 {
576 	zfs_acl_node_t *aclnode;
577 
578 	ASSERT(aclp);
579 
580 	if (start == NULL) {
581 		aclnode = list_head(&aclp->z_acl);
582 		if (aclnode == NULL)
583 			return (NULL);
584 
585 		aclp->z_next_ace = aclnode->z_acldata;
586 		aclp->z_curr_node = aclnode;
587 		aclnode->z_ace_idx = 0;
588 	}
589 
590 	aclnode = aclp->z_curr_node;
591 
592 	if (aclnode == NULL)
593 		return (NULL);
594 
595 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
596 		aclnode = list_next(&aclp->z_acl, aclnode);
597 		if (aclnode == NULL)
598 			return (NULL);
599 		else {
600 			aclp->z_curr_node = aclnode;
601 			aclnode->z_ace_idx = 0;
602 			aclp->z_next_ace = aclnode->z_acldata;
603 		}
604 	}
605 
606 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
607 		void *acep = aclp->z_next_ace;
608 		size_t ace_size;
609 
610 		/*
611 		 * Make sure we don't overstep our bounds
612 		 */
613 		ace_size = aclp->z_ops->ace_size(acep);
614 
615 		if (((caddr_t)acep + ace_size) >
616 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
617 			return (NULL);
618 		}
619 
620 		*iflags = aclp->z_ops->ace_flags_get(acep);
621 		*type = aclp->z_ops->ace_type_get(acep);
622 		*access_mask = aclp->z_ops->ace_mask_get(acep);
623 		*who = aclp->z_ops->ace_who_get(acep);
624 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
625 		aclnode->z_ace_idx++;
626 
627 		return ((void *)acep);
628 	}
629 	return (NULL);
630 }
631 
632 static uint64_t
633 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
634     uint16_t *flags, uint16_t *type, uint32_t *mask)
635 {
636 	(void) aclcnt;
637 	zfs_acl_t *aclp = datap;
638 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
639 	uint64_t who;
640 
641 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
642 	    flags, type);
643 	return ((uint64_t)(uintptr_t)acep);
644 }
645 
646 /*
647  * Copy ACE to internal ZFS format.
648  * While processing the ACL each ACE will be validated for correctness.
649  * ACE FUIDs will be created later.
650  */
651 static int
652 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, umode_t obj_mode, zfs_acl_t *aclp,
653     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
654     zfs_fuid_info_t **fuidp, cred_t *cr)
655 {
656 	int i;
657 	uint16_t entry_type;
658 	zfs_ace_t *aceptr = z_acl;
659 	ace_t *acep = datap;
660 	zfs_object_ace_t *zobjacep;
661 	ace_object_t *aceobjp;
662 
663 	for (i = 0; i != aclcnt; i++) {
664 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
665 		aceptr->z_hdr.z_flags = acep->a_flags;
666 		aceptr->z_hdr.z_type = acep->a_type;
667 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
668 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
669 		    entry_type != ACE_EVERYONE) {
670 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
671 			    cr, (entry_type == 0) ?
672 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
673 		}
674 
675 		/*
676 		 * Make sure ACE is valid
677 		 */
678 		if (zfs_ace_valid(obj_mode, aclp, aceptr->z_hdr.z_type,
679 		    aceptr->z_hdr.z_flags) != B_TRUE)
680 			return (SET_ERROR(EINVAL));
681 
682 		switch (acep->a_type) {
683 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
684 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
685 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
686 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
687 			zobjacep = (zfs_object_ace_t *)aceptr;
688 			aceobjp = (ace_object_t *)acep;
689 
690 			memcpy(zobjacep->z_object_type, aceobjp->a_obj_type,
691 			    sizeof (aceobjp->a_obj_type));
692 			memcpy(zobjacep->z_inherit_type,
693 			    aceobjp->a_inherit_obj_type,
694 			    sizeof (aceobjp->a_inherit_obj_type));
695 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
696 			break;
697 		default:
698 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
699 		}
700 
701 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
702 		    aclp->z_ops->ace_size(aceptr));
703 	}
704 
705 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
706 
707 	return (0);
708 }
709 
710 /*
711  * Copy ZFS ACEs to fixed size ace_t layout
712  */
713 static void
714 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
715     void *datap, int filter)
716 {
717 	uint64_t who;
718 	uint32_t access_mask;
719 	uint16_t iflags, type;
720 	zfs_ace_hdr_t *zacep = NULL;
721 	ace_t *acep = datap;
722 	ace_object_t *objacep;
723 	zfs_object_ace_t *zobjacep;
724 	size_t ace_size;
725 	uint16_t entry_type;
726 
727 	while ((zacep = zfs_acl_next_ace(aclp, zacep,
728 	    &who, &access_mask, &iflags, &type))) {
729 
730 		switch (type) {
731 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
732 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
733 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
734 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
735 			if (filter) {
736 				continue;
737 			}
738 			zobjacep = (zfs_object_ace_t *)zacep;
739 			objacep = (ace_object_t *)acep;
740 			memcpy(objacep->a_obj_type,
741 			    zobjacep->z_object_type,
742 			    sizeof (zobjacep->z_object_type));
743 			memcpy(objacep->a_inherit_obj_type,
744 			    zobjacep->z_inherit_type,
745 			    sizeof (zobjacep->z_inherit_type));
746 			ace_size = sizeof (ace_object_t);
747 			break;
748 		default:
749 			ace_size = sizeof (ace_t);
750 			break;
751 		}
752 
753 		entry_type = (iflags & ACE_TYPE_FLAGS);
754 		if ((entry_type != ACE_OWNER &&
755 		    entry_type != OWNING_GROUP &&
756 		    entry_type != ACE_EVERYONE)) {
757 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
758 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
759 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
760 		} else {
761 			acep->a_who = (uid_t)(int64_t)who;
762 		}
763 		acep->a_access_mask = access_mask;
764 		acep->a_flags = iflags;
765 		acep->a_type = type;
766 		acep = (ace_t *)((caddr_t)acep + ace_size);
767 	}
768 }
769 
770 static int
771 zfs_copy_ace_2_oldace(umode_t obj_mode, zfs_acl_t *aclp, ace_t *acep,
772     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
773 {
774 	int i;
775 	zfs_oldace_t *aceptr = z_acl;
776 
777 	for (i = 0; i != aclcnt; i++, aceptr++) {
778 		aceptr->z_access_mask = acep[i].a_access_mask;
779 		aceptr->z_type = acep[i].a_type;
780 		aceptr->z_flags = acep[i].a_flags;
781 		aceptr->z_fuid = acep[i].a_who;
782 		/*
783 		 * Make sure ACE is valid
784 		 */
785 		if (zfs_ace_valid(obj_mode, aclp, aceptr->z_type,
786 		    aceptr->z_flags) != B_TRUE)
787 			return (SET_ERROR(EINVAL));
788 	}
789 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
790 	return (0);
791 }
792 
793 /*
794  * convert old ACL format to new
795  */
796 void
797 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
798 {
799 	zfs_oldace_t *oldaclp;
800 	int i;
801 	uint16_t type, iflags;
802 	uint32_t access_mask;
803 	uint64_t who;
804 	void *cookie = NULL;
805 	zfs_acl_node_t *newaclnode;
806 
807 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
808 	/*
809 	 * First create the ACE in a contiguous piece of memory
810 	 * for zfs_copy_ace_2_fuid().
811 	 *
812 	 * We only convert an ACL once, so this won't happen
813 	 * every time.
814 	 */
815 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
816 	    KM_SLEEP);
817 	i = 0;
818 	while ((cookie = zfs_acl_next_ace(aclp, cookie, &who,
819 	    &access_mask, &iflags, &type))) {
820 		oldaclp[i].z_flags = iflags;
821 		oldaclp[i].z_type = type;
822 		oldaclp[i].z_fuid = who;
823 		oldaclp[i++].z_access_mask = access_mask;
824 	}
825 
826 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
827 	    sizeof (zfs_object_ace_t));
828 	aclp->z_ops = &zfs_acl_fuid_ops;
829 	VERIFY(zfs_copy_ace_2_fuid(ZTOZSB(zp), ZTOI(zp)->i_mode,
830 	    aclp, oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
831 	    &newaclnode->z_size, NULL, cr) == 0);
832 	newaclnode->z_ace_count = aclp->z_acl_count;
833 	aclp->z_version = ZFS_ACL_VERSION;
834 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
835 
836 	/*
837 	 * Release all previous ACL nodes
838 	 */
839 
840 	zfs_acl_release_nodes(aclp);
841 
842 	list_insert_head(&aclp->z_acl, newaclnode);
843 
844 	aclp->z_acl_bytes = newaclnode->z_size;
845 	aclp->z_acl_count = newaclnode->z_ace_count;
846 
847 }
848 
849 /*
850  * Convert unix access mask to v4 access mask
851  */
852 static uint32_t
853 zfs_unix_to_v4(uint32_t access_mask)
854 {
855 	uint32_t new_mask = 0;
856 
857 	if (access_mask & S_IXOTH)
858 		new_mask |= ACE_EXECUTE;
859 	if (access_mask & S_IWOTH)
860 		new_mask |= ACE_WRITE_DATA;
861 	if (access_mask & S_IROTH)
862 		new_mask |= ACE_READ_DATA;
863 	return (new_mask);
864 }
865 
866 
867 static int
868 zfs_v4_to_unix(uint32_t access_mask, int *unmapped)
869 {
870 	int new_mask = 0;
871 
872 	*unmapped = access_mask &
873 	    (ACE_WRITE_OWNER | ACE_WRITE_ACL | ACE_DELETE);
874 
875 	if (access_mask & WRITE_MASK)
876 		new_mask |= S_IWOTH;
877 	if (access_mask & ACE_READ_DATA)
878 		new_mask |= S_IROTH;
879 	if (access_mask & ACE_EXECUTE)
880 		new_mask |= S_IXOTH;
881 
882 	return (new_mask);
883 }
884 
885 
886 static void
887 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
888     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
889 {
890 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
891 
892 	aclp->z_ops->ace_mask_set(acep, access_mask);
893 	aclp->z_ops->ace_type_set(acep, access_type);
894 	aclp->z_ops->ace_flags_set(acep, entry_type);
895 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
896 	    type != ACE_EVERYONE))
897 		aclp->z_ops->ace_who_set(acep, fuid);
898 }
899 
900 /*
901  * Determine mode of file based on ACL.
902  */
903 uint64_t
904 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
905     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
906 {
907 	int		entry_type;
908 	mode_t		mode;
909 	mode_t		seen = 0;
910 	zfs_ace_hdr_t 	*acep = NULL;
911 	uint64_t	who;
912 	uint16_t	iflags, type;
913 	uint32_t	access_mask;
914 	boolean_t	an_exec_denied = B_FALSE;
915 
916 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
917 
918 	while ((acep = zfs_acl_next_ace(aclp, acep, &who,
919 	    &access_mask, &iflags, &type))) {
920 
921 		if (!zfs_acl_valid_ace_type(type, iflags))
922 			continue;
923 
924 		entry_type = (iflags & ACE_TYPE_FLAGS);
925 
926 		/*
927 		 * Skip over any inherit_only ACEs
928 		 */
929 		if (iflags & ACE_INHERIT_ONLY_ACE)
930 			continue;
931 
932 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
933 		    who == fuid)) {
934 			if ((access_mask & ACE_READ_DATA) &&
935 			    (!(seen & S_IRUSR))) {
936 				seen |= S_IRUSR;
937 				if (type == ALLOW) {
938 					mode |= S_IRUSR;
939 				}
940 			}
941 			if ((access_mask & ACE_WRITE_DATA) &&
942 			    (!(seen & S_IWUSR))) {
943 				seen |= S_IWUSR;
944 				if (type == ALLOW) {
945 					mode |= S_IWUSR;
946 				}
947 			}
948 			if ((access_mask & ACE_EXECUTE) &&
949 			    (!(seen & S_IXUSR))) {
950 				seen |= S_IXUSR;
951 				if (type == ALLOW) {
952 					mode |= S_IXUSR;
953 				}
954 			}
955 		} else if (entry_type == OWNING_GROUP ||
956 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
957 			if ((access_mask & ACE_READ_DATA) &&
958 			    (!(seen & S_IRGRP))) {
959 				seen |= S_IRGRP;
960 				if (type == ALLOW) {
961 					mode |= S_IRGRP;
962 				}
963 			}
964 			if ((access_mask & ACE_WRITE_DATA) &&
965 			    (!(seen & S_IWGRP))) {
966 				seen |= S_IWGRP;
967 				if (type == ALLOW) {
968 					mode |= S_IWGRP;
969 				}
970 			}
971 			if ((access_mask & ACE_EXECUTE) &&
972 			    (!(seen & S_IXGRP))) {
973 				seen |= S_IXGRP;
974 				if (type == ALLOW) {
975 					mode |= S_IXGRP;
976 				}
977 			}
978 		} else if (entry_type == ACE_EVERYONE) {
979 			if ((access_mask & ACE_READ_DATA)) {
980 				if (!(seen & S_IRUSR)) {
981 					seen |= S_IRUSR;
982 					if (type == ALLOW) {
983 						mode |= S_IRUSR;
984 					}
985 				}
986 				if (!(seen & S_IRGRP)) {
987 					seen |= S_IRGRP;
988 					if (type == ALLOW) {
989 						mode |= S_IRGRP;
990 					}
991 				}
992 				if (!(seen & S_IROTH)) {
993 					seen |= S_IROTH;
994 					if (type == ALLOW) {
995 						mode |= S_IROTH;
996 					}
997 				}
998 			}
999 			if ((access_mask & ACE_WRITE_DATA)) {
1000 				if (!(seen & S_IWUSR)) {
1001 					seen |= S_IWUSR;
1002 					if (type == ALLOW) {
1003 						mode |= S_IWUSR;
1004 					}
1005 				}
1006 				if (!(seen & S_IWGRP)) {
1007 					seen |= S_IWGRP;
1008 					if (type == ALLOW) {
1009 						mode |= S_IWGRP;
1010 					}
1011 				}
1012 				if (!(seen & S_IWOTH)) {
1013 					seen |= S_IWOTH;
1014 					if (type == ALLOW) {
1015 						mode |= S_IWOTH;
1016 					}
1017 				}
1018 			}
1019 			if ((access_mask & ACE_EXECUTE)) {
1020 				if (!(seen & S_IXUSR)) {
1021 					seen |= S_IXUSR;
1022 					if (type == ALLOW) {
1023 						mode |= S_IXUSR;
1024 					}
1025 				}
1026 				if (!(seen & S_IXGRP)) {
1027 					seen |= S_IXGRP;
1028 					if (type == ALLOW) {
1029 						mode |= S_IXGRP;
1030 					}
1031 				}
1032 				if (!(seen & S_IXOTH)) {
1033 					seen |= S_IXOTH;
1034 					if (type == ALLOW) {
1035 						mode |= S_IXOTH;
1036 					}
1037 				}
1038 			}
1039 		} else {
1040 			/*
1041 			 * Only care if this IDENTIFIER_GROUP or
1042 			 * USER ACE denies execute access to someone,
1043 			 * mode is not affected
1044 			 */
1045 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1046 				an_exec_denied = B_TRUE;
1047 		}
1048 	}
1049 
1050 	/*
1051 	 * Failure to allow is effectively a deny, so execute permission
1052 	 * is denied if it was never mentioned or if we explicitly
1053 	 * weren't allowed it.
1054 	 */
1055 	if (!an_exec_denied &&
1056 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1057 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1058 		an_exec_denied = B_TRUE;
1059 
1060 	if (an_exec_denied)
1061 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1062 	else
1063 		*pflags |= ZFS_NO_EXECS_DENIED;
1064 
1065 	return (mode);
1066 }
1067 
1068 /*
1069  * Read an external acl object.  If the intent is to modify, always
1070  * create a new acl and leave any cached acl in place.
1071  */
1072 int
1073 zfs_acl_node_read(struct znode *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1074     boolean_t will_modify)
1075 {
1076 	zfs_acl_t	*aclp;
1077 	int		aclsize = 0;
1078 	int		acl_count = 0;
1079 	zfs_acl_node_t	*aclnode;
1080 	zfs_acl_phys_t	znode_acl;
1081 	int		version;
1082 	int		error;
1083 	boolean_t	drop_lock = B_FALSE;
1084 
1085 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1086 
1087 	if (zp->z_acl_cached && !will_modify) {
1088 		*aclpp = zp->z_acl_cached;
1089 		return (0);
1090 	}
1091 
1092 	/*
1093 	 * close race where znode could be upgrade while trying to
1094 	 * read the znode attributes.
1095 	 *
1096 	 * But this could only happen if the file isn't already an SA
1097 	 * znode
1098 	 */
1099 	if (!zp->z_is_sa && !have_lock) {
1100 		mutex_enter(&zp->z_lock);
1101 		drop_lock = B_TRUE;
1102 	}
1103 	version = zfs_znode_acl_version(zp);
1104 
1105 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1106 	    &acl_count, &znode_acl)) != 0) {
1107 		goto done;
1108 	}
1109 
1110 	aclp = zfs_acl_alloc(version);
1111 
1112 	aclp->z_acl_count = acl_count;
1113 	aclp->z_acl_bytes = aclsize;
1114 
1115 	aclnode = zfs_acl_node_alloc(aclsize);
1116 	aclnode->z_ace_count = aclp->z_acl_count;
1117 	aclnode->z_size = aclsize;
1118 
1119 	if (!zp->z_is_sa) {
1120 		if (znode_acl.z_acl_extern_obj) {
1121 			error = dmu_read(ZTOZSB(zp)->z_os,
1122 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1123 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1124 		} else {
1125 			memcpy(aclnode->z_acldata, znode_acl.z_ace_data,
1126 			    aclnode->z_size);
1127 		}
1128 	} else {
1129 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(ZTOZSB(zp)),
1130 		    aclnode->z_acldata, aclnode->z_size);
1131 	}
1132 
1133 	if (error != 0) {
1134 		zfs_acl_free(aclp);
1135 		zfs_acl_node_free(aclnode);
1136 		/* convert checksum errors into IO errors */
1137 		if (error == ECKSUM)
1138 			error = SET_ERROR(EIO);
1139 		goto done;
1140 	}
1141 
1142 	list_insert_head(&aclp->z_acl, aclnode);
1143 
1144 	*aclpp = aclp;
1145 	if (!will_modify)
1146 		zp->z_acl_cached = aclp;
1147 done:
1148 	if (drop_lock)
1149 		mutex_exit(&zp->z_lock);
1150 	return (error);
1151 }
1152 
1153 void
1154 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1155     boolean_t start, void *userdata)
1156 {
1157 	(void) buflen;
1158 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1159 
1160 	if (start) {
1161 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1162 	} else {
1163 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1164 		    cb->cb_acl_node);
1165 	}
1166 	*dataptr = cb->cb_acl_node->z_acldata;
1167 	*length = cb->cb_acl_node->z_size;
1168 }
1169 
1170 int
1171 zfs_acl_chown_setattr(znode_t *zp)
1172 {
1173 	int error;
1174 	zfs_acl_t *aclp;
1175 
1176 	if (ZTOZSB(zp)->z_acl_type == ZFS_ACLTYPE_POSIX)
1177 		return (0);
1178 
1179 	ASSERT(MUTEX_HELD(&zp->z_lock));
1180 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1181 
1182 	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1183 	if (error == 0 && aclp->z_acl_count > 0)
1184 		zp->z_mode = ZTOI(zp)->i_mode =
1185 		    zfs_mode_compute(zp->z_mode, aclp,
1186 		    &zp->z_pflags, KUID_TO_SUID(ZTOI(zp)->i_uid),
1187 		    KGID_TO_SGID(ZTOI(zp)->i_gid));
1188 
1189 	/*
1190 	 * Some ZFS implementations (ZEVO) create neither a ZNODE_ACL
1191 	 * nor a DACL_ACES SA in which case ENOENT is returned from
1192 	 * zfs_acl_node_read() when the SA can't be located.
1193 	 * Allow chown/chgrp to succeed in these cases rather than
1194 	 * returning an error that makes no sense in the context of
1195 	 * the caller.
1196 	 */
1197 	if (error == ENOENT)
1198 		return (0);
1199 
1200 	return (error);
1201 }
1202 
1203 typedef struct trivial_acl {
1204 	uint32_t	allow0;		/* allow mask for bits only in owner */
1205 	uint32_t	deny1;		/* deny mask for bits not in owner */
1206 	uint32_t	deny2;		/* deny mask for bits not in group */
1207 	uint32_t	owner;		/* allow mask matching mode */
1208 	uint32_t	group;		/* allow mask matching mode */
1209 	uint32_t	everyone;	/* allow mask matching mode */
1210 } trivial_acl_t;
1211 
1212 static void
1213 acl_trivial_access_masks(mode_t mode, boolean_t isdir, trivial_acl_t *masks)
1214 {
1215 	uint32_t read_mask = ACE_READ_DATA;
1216 	uint32_t write_mask = ACE_WRITE_DATA|ACE_APPEND_DATA;
1217 	uint32_t execute_mask = ACE_EXECUTE;
1218 
1219 	if (isdir)
1220 		write_mask |= ACE_DELETE_CHILD;
1221 
1222 	masks->deny1 = 0;
1223 
1224 	if (!(mode & S_IRUSR) && (mode & (S_IRGRP|S_IROTH)))
1225 		masks->deny1 |= read_mask;
1226 	if (!(mode & S_IWUSR) && (mode & (S_IWGRP|S_IWOTH)))
1227 		masks->deny1 |= write_mask;
1228 	if (!(mode & S_IXUSR) && (mode & (S_IXGRP|S_IXOTH)))
1229 		masks->deny1 |= execute_mask;
1230 
1231 	masks->deny2 = 0;
1232 	if (!(mode & S_IRGRP) && (mode & S_IROTH))
1233 		masks->deny2 |= read_mask;
1234 	if (!(mode & S_IWGRP) && (mode & S_IWOTH))
1235 		masks->deny2 |= write_mask;
1236 	if (!(mode & S_IXGRP) && (mode & S_IXOTH))
1237 		masks->deny2 |= execute_mask;
1238 
1239 	masks->allow0 = 0;
1240 	if ((mode & S_IRUSR) && (!(mode & S_IRGRP) && (mode & S_IROTH)))
1241 		masks->allow0 |= read_mask;
1242 	if ((mode & S_IWUSR) && (!(mode & S_IWGRP) && (mode & S_IWOTH)))
1243 		masks->allow0 |= write_mask;
1244 	if ((mode & S_IXUSR) && (!(mode & S_IXGRP) && (mode & S_IXOTH)))
1245 		masks->allow0 |= execute_mask;
1246 
1247 	masks->owner = ACE_WRITE_ATTRIBUTES|ACE_WRITE_OWNER|ACE_WRITE_ACL|
1248 	    ACE_WRITE_NAMED_ATTRS|ACE_READ_ACL|ACE_READ_ATTRIBUTES|
1249 	    ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE;
1250 	if (mode & S_IRUSR)
1251 		masks->owner |= read_mask;
1252 	if (mode & S_IWUSR)
1253 		masks->owner |= write_mask;
1254 	if (mode & S_IXUSR)
1255 		masks->owner |= execute_mask;
1256 
1257 	masks->group = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1258 	    ACE_SYNCHRONIZE;
1259 	if (mode & S_IRGRP)
1260 		masks->group |= read_mask;
1261 	if (mode & S_IWGRP)
1262 		masks->group |= write_mask;
1263 	if (mode & S_IXGRP)
1264 		masks->group |= execute_mask;
1265 
1266 	masks->everyone = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1267 	    ACE_SYNCHRONIZE;
1268 	if (mode & S_IROTH)
1269 		masks->everyone |= read_mask;
1270 	if (mode & S_IWOTH)
1271 		masks->everyone |= write_mask;
1272 	if (mode & S_IXOTH)
1273 		masks->everyone |= execute_mask;
1274 }
1275 
1276 /*
1277  * ace_trivial:
1278  * determine whether an ace_t acl is trivial
1279  *
1280  * Trivialness implies that the acl is composed of only
1281  * owner, group, everyone entries.  ACL can't
1282  * have read_acl denied, and write_owner/write_acl/write_attributes
1283  * can only be owner@ entry.
1284  */
1285 static int
1286 ace_trivial_common(void *acep, int aclcnt,
1287     uint64_t (*walk)(void *, uint64_t, int aclcnt,
1288     uint16_t *, uint16_t *, uint32_t *))
1289 {
1290 	uint16_t flags;
1291 	uint32_t mask;
1292 	uint16_t type;
1293 	uint64_t cookie = 0;
1294 
1295 	while ((cookie = walk(acep, cookie, aclcnt, &flags, &type, &mask))) {
1296 		switch (flags & ACE_TYPE_FLAGS) {
1297 		case ACE_OWNER:
1298 		case ACE_GROUP|ACE_IDENTIFIER_GROUP:
1299 		case ACE_EVERYONE:
1300 			break;
1301 		default:
1302 			return (1);
1303 		}
1304 
1305 		if (flags & (ACE_FILE_INHERIT_ACE|
1306 		    ACE_DIRECTORY_INHERIT_ACE|ACE_NO_PROPAGATE_INHERIT_ACE|
1307 		    ACE_INHERIT_ONLY_ACE))
1308 			return (1);
1309 
1310 		/*
1311 		 * Special check for some special bits
1312 		 *
1313 		 * Don't allow anybody to deny reading basic
1314 		 * attributes or a files ACL.
1315 		 */
1316 		if ((mask & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
1317 		    (type == ACE_ACCESS_DENIED_ACE_TYPE))
1318 			return (1);
1319 
1320 		/*
1321 		 * Delete permission is never set by default
1322 		 */
1323 		if (mask & ACE_DELETE)
1324 			return (1);
1325 
1326 		/*
1327 		 * Child delete permission should be accompanied by write
1328 		 */
1329 		if ((mask & ACE_DELETE_CHILD) && !(mask & ACE_WRITE_DATA))
1330 			return (1);
1331 
1332 		/*
1333 		 * only allow owner@ to have
1334 		 * write_acl/write_owner/write_attributes/write_xattr/
1335 		 */
1336 		if (type == ACE_ACCESS_ALLOWED_ACE_TYPE &&
1337 		    (!(flags & ACE_OWNER) && (mask &
1338 		    (ACE_WRITE_OWNER|ACE_WRITE_ACL| ACE_WRITE_ATTRIBUTES|
1339 		    ACE_WRITE_NAMED_ATTRS))))
1340 			return (1);
1341 
1342 	}
1343 
1344 	return (0);
1345 }
1346 
1347 /*
1348  * common code for setting ACLs.
1349  *
1350  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1351  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1352  * already checked the acl and knows whether to inherit.
1353  */
1354 int
1355 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1356 {
1357 	int			error;
1358 	zfsvfs_t		*zfsvfs = ZTOZSB(zp);
1359 	dmu_object_type_t	otype;
1360 	zfs_acl_locator_cb_t	locate = { 0 };
1361 	uint64_t		mode;
1362 	sa_bulk_attr_t		bulk[5];
1363 	uint64_t		ctime[2];
1364 	int			count = 0;
1365 	zfs_acl_phys_t		acl_phys;
1366 
1367 	mode = zp->z_mode;
1368 
1369 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1370 	    KUID_TO_SUID(ZTOI(zp)->i_uid), KGID_TO_SGID(ZTOI(zp)->i_gid));
1371 
1372 	zp->z_mode = ZTOI(zp)->i_mode = mode;
1373 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1374 	    &mode, sizeof (mode));
1375 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1376 	    &zp->z_pflags, sizeof (zp->z_pflags));
1377 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1378 	    &ctime, sizeof (ctime));
1379 
1380 	if (zp->z_acl_cached) {
1381 		zfs_acl_free(zp->z_acl_cached);
1382 		zp->z_acl_cached = NULL;
1383 	}
1384 
1385 	/*
1386 	 * Upgrade needed?
1387 	 */
1388 	if (!zfsvfs->z_use_fuids) {
1389 		otype = DMU_OT_OLDACL;
1390 	} else {
1391 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1392 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1393 			zfs_acl_xform(zp, aclp, cr);
1394 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1395 		otype = DMU_OT_ACL;
1396 	}
1397 
1398 	/*
1399 	 * Arrgh, we have to handle old on disk format
1400 	 * as well as newer (preferred) SA format.
1401 	 */
1402 
1403 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1404 		locate.cb_aclp = aclp;
1405 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1406 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1407 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1408 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1409 	} else { /* Painful legacy way */
1410 		zfs_acl_node_t *aclnode;
1411 		uint64_t off = 0;
1412 		uint64_t aoid;
1413 
1414 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1415 		    &acl_phys, sizeof (acl_phys))) != 0)
1416 			return (error);
1417 
1418 		aoid = acl_phys.z_acl_extern_obj;
1419 
1420 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1421 			/*
1422 			 * If ACL was previously external and we are now
1423 			 * converting to new ACL format then release old
1424 			 * ACL object and create a new one.
1425 			 */
1426 			if (aoid &&
1427 			    aclp->z_version != acl_phys.z_acl_version) {
1428 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1429 				if (error)
1430 					return (error);
1431 				aoid = 0;
1432 			}
1433 			if (aoid == 0) {
1434 				aoid = dmu_object_alloc(zfsvfs->z_os,
1435 				    otype, aclp->z_acl_bytes,
1436 				    otype == DMU_OT_ACL ?
1437 				    DMU_OT_SYSACL : DMU_OT_NONE,
1438 				    otype == DMU_OT_ACL ?
1439 				    DN_OLD_MAX_BONUSLEN : 0, tx);
1440 			} else {
1441 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1442 				    aoid, aclp->z_acl_bytes, 0, tx);
1443 			}
1444 			acl_phys.z_acl_extern_obj = aoid;
1445 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1446 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1447 				if (aclnode->z_ace_count == 0)
1448 					continue;
1449 				dmu_write(zfsvfs->z_os, aoid, off,
1450 				    aclnode->z_size, aclnode->z_acldata, tx);
1451 				off += aclnode->z_size;
1452 			}
1453 		} else {
1454 			void *start = acl_phys.z_ace_data;
1455 			/*
1456 			 * Migrating back embedded?
1457 			 */
1458 			if (acl_phys.z_acl_extern_obj) {
1459 				error = dmu_object_free(zfsvfs->z_os,
1460 				    acl_phys.z_acl_extern_obj, tx);
1461 				if (error)
1462 					return (error);
1463 				acl_phys.z_acl_extern_obj = 0;
1464 			}
1465 
1466 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1467 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1468 				if (aclnode->z_ace_count == 0)
1469 					continue;
1470 				memcpy(start, aclnode->z_acldata,
1471 				    aclnode->z_size);
1472 				start = (caddr_t)start + aclnode->z_size;
1473 			}
1474 		}
1475 		/*
1476 		 * If Old version then swap count/bytes to match old
1477 		 * layout of znode_acl_phys_t.
1478 		 */
1479 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1480 			acl_phys.z_acl_size = aclp->z_acl_count;
1481 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1482 		} else {
1483 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1484 			acl_phys.z_acl_count = aclp->z_acl_count;
1485 		}
1486 		acl_phys.z_acl_version = aclp->z_version;
1487 
1488 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1489 		    &acl_phys, sizeof (acl_phys));
1490 	}
1491 
1492 	/*
1493 	 * Replace ACL wide bits, but first clear them.
1494 	 */
1495 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1496 
1497 	zp->z_pflags |= aclp->z_hints;
1498 
1499 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1500 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1501 
1502 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1503 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1504 }
1505 
1506 static void
1507 zfs_acl_chmod(boolean_t isdir, uint64_t mode, boolean_t split, boolean_t trim,
1508     zfs_acl_t *aclp)
1509 {
1510 	void		*acep = NULL;
1511 	uint64_t	who;
1512 	int		new_count, new_bytes;
1513 	int		ace_size;
1514 	int		entry_type;
1515 	uint16_t	iflags, type;
1516 	uint32_t	access_mask;
1517 	zfs_acl_node_t	*newnode;
1518 	size_t		abstract_size = aclp->z_ops->ace_abstract_size();
1519 	void		*zacep;
1520 	trivial_acl_t	masks;
1521 
1522 	new_count = new_bytes = 0;
1523 
1524 	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1525 
1526 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1527 
1528 	zacep = newnode->z_acldata;
1529 	if (masks.allow0) {
1530 		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1531 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1532 		new_count++;
1533 		new_bytes += abstract_size;
1534 	}
1535 	if (masks.deny1) {
1536 		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1537 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1538 		new_count++;
1539 		new_bytes += abstract_size;
1540 	}
1541 	if (masks.deny2) {
1542 		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1543 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1544 		new_count++;
1545 		new_bytes += abstract_size;
1546 	}
1547 
1548 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1549 	    &iflags, &type))) {
1550 		entry_type = (iflags & ACE_TYPE_FLAGS);
1551 		/*
1552 		 * ACEs used to represent the file mode may be divided
1553 		 * into an equivalent pair of inherit-only and regular
1554 		 * ACEs, if they are inheritable.
1555 		 * Skip regular ACEs, which are replaced by the new mode.
1556 		 */
1557 		if (split && (entry_type == ACE_OWNER ||
1558 		    entry_type == OWNING_GROUP ||
1559 		    entry_type == ACE_EVERYONE)) {
1560 			if (!isdir || !(iflags &
1561 			    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1562 				continue;
1563 			/*
1564 			 * We preserve owner@, group@, or @everyone
1565 			 * permissions, if they are inheritable, by
1566 			 * copying them to inherit_only ACEs. This
1567 			 * prevents inheritable permissions from being
1568 			 * altered along with the file mode.
1569 			 */
1570 			iflags |= ACE_INHERIT_ONLY_ACE;
1571 		}
1572 
1573 		/*
1574 		 * If this ACL has any inheritable ACEs, mark that in
1575 		 * the hints (which are later masked into the pflags)
1576 		 * so create knows to do inheritance.
1577 		 */
1578 		if (isdir && (iflags &
1579 		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1580 			aclp->z_hints |= ZFS_INHERIT_ACE;
1581 
1582 		if ((type != ALLOW && type != DENY) ||
1583 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1584 			switch (type) {
1585 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1586 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1587 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1588 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1589 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1590 				break;
1591 			}
1592 		} else {
1593 			/*
1594 			 * Limit permissions to be no greater than
1595 			 * group permissions.
1596 			 * The "aclinherit" and "aclmode" properties
1597 			 * affect policy for create and chmod(2),
1598 			 * respectively.
1599 			 */
1600 			if ((type == ALLOW) && trim)
1601 				access_mask &= masks.group;
1602 		}
1603 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1604 		ace_size = aclp->z_ops->ace_size(acep);
1605 		zacep = (void *)((uintptr_t)zacep + ace_size);
1606 		new_count++;
1607 		new_bytes += ace_size;
1608 	}
1609 	zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1610 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1611 	zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1612 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1613 	zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1614 
1615 	new_count += 3;
1616 	new_bytes += abstract_size * 3;
1617 	zfs_acl_release_nodes(aclp);
1618 	aclp->z_acl_count = new_count;
1619 	aclp->z_acl_bytes = new_bytes;
1620 	newnode->z_ace_count = new_count;
1621 	newnode->z_size = new_bytes;
1622 	list_insert_tail(&aclp->z_acl, newnode);
1623 }
1624 
1625 int
1626 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1627 {
1628 	int error = 0;
1629 
1630 	mutex_enter(&zp->z_acl_lock);
1631 	mutex_enter(&zp->z_lock);
1632 	if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_DISCARD)
1633 		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1634 	else
1635 		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1636 
1637 	if (error == 0) {
1638 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1639 		zfs_acl_chmod(S_ISDIR(ZTOI(zp)->i_mode), mode, B_TRUE,
1640 		    (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1641 	}
1642 	mutex_exit(&zp->z_lock);
1643 	mutex_exit(&zp->z_acl_lock);
1644 
1645 	return (error);
1646 }
1647 
1648 /*
1649  * Should ACE be inherited?
1650  */
1651 static int
1652 zfs_ace_can_use(umode_t obj_mode, uint16_t acep_flags)
1653 {
1654 	int	iflags = (acep_flags & 0xf);
1655 
1656 	if (S_ISDIR(obj_mode) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1657 		return (1);
1658 	else if (iflags & ACE_FILE_INHERIT_ACE)
1659 		return (!(S_ISDIR(obj_mode) &&
1660 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1661 	return (0);
1662 }
1663 
1664 /*
1665  * inherit inheritable ACEs from parent
1666  */
1667 static zfs_acl_t *
1668 zfs_acl_inherit(zfsvfs_t *zfsvfs, umode_t va_mode, zfs_acl_t *paclp,
1669     uint64_t mode, boolean_t *need_chmod)
1670 {
1671 	void		*pacep = NULL;
1672 	void		*acep;
1673 	zfs_acl_node_t  *aclnode;
1674 	zfs_acl_t	*aclp = NULL;
1675 	uint64_t	who;
1676 	uint32_t	access_mask;
1677 	uint16_t	iflags, newflags, type;
1678 	size_t		ace_size;
1679 	void		*data1, *data2;
1680 	size_t		data1sz, data2sz;
1681 	uint_t		aclinherit;
1682 	boolean_t	isdir = S_ISDIR(va_mode);
1683 	boolean_t	isreg = S_ISREG(va_mode);
1684 
1685 	*need_chmod = B_TRUE;
1686 
1687 	aclp = zfs_acl_alloc(paclp->z_version);
1688 	aclinherit = zfsvfs->z_acl_inherit;
1689 	if (aclinherit == ZFS_ACL_DISCARD || S_ISLNK(va_mode))
1690 		return (aclp);
1691 
1692 	while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1693 	    &access_mask, &iflags, &type))) {
1694 
1695 		/*
1696 		 * don't inherit bogus ACEs
1697 		 */
1698 		if (!zfs_acl_valid_ace_type(type, iflags))
1699 			continue;
1700 
1701 		/*
1702 		 * Check if ACE is inheritable by this vnode
1703 		 */
1704 		if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1705 		    !zfs_ace_can_use(va_mode, iflags))
1706 			continue;
1707 
1708 		/*
1709 		 * If owner@, group@, or everyone@ inheritable
1710 		 * then zfs_acl_chmod() isn't needed.
1711 		 */
1712 		if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1713 		    aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1714 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1715 		    ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1716 		    (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1717 			*need_chmod = B_FALSE;
1718 
1719 		/*
1720 		 * Strip inherited execute permission from file if
1721 		 * not in mode
1722 		 */
1723 		if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1724 		    !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1725 			access_mask &= ~ACE_EXECUTE;
1726 		}
1727 
1728 		/*
1729 		 * Strip write_acl and write_owner from permissions
1730 		 * when inheriting an ACE
1731 		 */
1732 		if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1733 			access_mask &= ~RESTRICTED_CLEAR;
1734 		}
1735 
1736 		ace_size = aclp->z_ops->ace_size(pacep);
1737 		aclnode = zfs_acl_node_alloc(ace_size);
1738 		list_insert_tail(&aclp->z_acl, aclnode);
1739 		acep = aclnode->z_acldata;
1740 
1741 		zfs_set_ace(aclp, acep, access_mask, type,
1742 		    who, iflags|ACE_INHERITED_ACE);
1743 
1744 		/*
1745 		 * Copy special opaque data if any
1746 		 */
1747 		if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1748 			VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1749 			    &data2)) == data1sz);
1750 			memcpy(data2, data1, data2sz);
1751 		}
1752 
1753 		aclp->z_acl_count++;
1754 		aclnode->z_ace_count++;
1755 		aclp->z_acl_bytes += aclnode->z_size;
1756 		newflags = aclp->z_ops->ace_flags_get(acep);
1757 
1758 		/*
1759 		 * If ACE is not to be inherited further, or if the vnode is
1760 		 * not a directory, remove all inheritance flags
1761 		 */
1762 		if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1763 			newflags &= ~ALL_INHERIT;
1764 			aclp->z_ops->ace_flags_set(acep,
1765 			    newflags|ACE_INHERITED_ACE);
1766 			continue;
1767 		}
1768 
1769 		/*
1770 		 * This directory has an inheritable ACE
1771 		 */
1772 		aclp->z_hints |= ZFS_INHERIT_ACE;
1773 
1774 		/*
1775 		 * If only FILE_INHERIT is set then turn on
1776 		 * inherit_only
1777 		 */
1778 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1779 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1780 			newflags |= ACE_INHERIT_ONLY_ACE;
1781 			aclp->z_ops->ace_flags_set(acep,
1782 			    newflags|ACE_INHERITED_ACE);
1783 		} else {
1784 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1785 			aclp->z_ops->ace_flags_set(acep,
1786 			    newflags|ACE_INHERITED_ACE);
1787 		}
1788 	}
1789 	if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1790 	    aclp->z_acl_count != 0) {
1791 		*need_chmod = B_FALSE;
1792 	}
1793 
1794 	return (aclp);
1795 }
1796 
1797 /*
1798  * Create file system object initial permissions
1799  * including inheritable ACEs.
1800  * Also, create FUIDs for owner and group.
1801  */
1802 int
1803 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1804     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1805 {
1806 	int		error;
1807 	zfsvfs_t	*zfsvfs = ZTOZSB(dzp);
1808 	zfs_acl_t	*paclp;
1809 	gid_t		gid = vap->va_gid;
1810 	boolean_t	need_chmod = B_TRUE;
1811 	boolean_t	trim = B_FALSE;
1812 	boolean_t	inherited = B_FALSE;
1813 
1814 	memset(acl_ids, 0, sizeof (zfs_acl_ids_t));
1815 	acl_ids->z_mode = vap->va_mode;
1816 
1817 	if (vsecp)
1818 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_mode, vsecp,
1819 		    cr, &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1820 			return (error);
1821 
1822 	acl_ids->z_fuid = vap->va_uid;
1823 	acl_ids->z_fgid = vap->va_gid;
1824 #ifdef HAVE_KSID
1825 	/*
1826 	 * Determine uid and gid.
1827 	 */
1828 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1829 	    ((flag & IS_XATTR) && (S_ISDIR(vap->va_mode)))) {
1830 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_uid,
1831 		    cr, ZFS_OWNER, &acl_ids->z_fuidp);
1832 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
1833 		    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1834 		gid = vap->va_gid;
1835 	} else {
1836 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1837 		    cr, &acl_ids->z_fuidp);
1838 		acl_ids->z_fgid = 0;
1839 		if (vap->va_mask & AT_GID)  {
1840 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1841 			    (uint64_t)vap->va_gid,
1842 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1843 			gid = vap->va_gid;
1844 			if (acl_ids->z_fgid != KGID_TO_SGID(ZTOI(dzp)->i_gid) &&
1845 			    !groupmember(vap->va_gid, cr) &&
1846 			    secpolicy_vnode_create_gid(cr) != 0)
1847 				acl_ids->z_fgid = 0;
1848 		}
1849 		if (acl_ids->z_fgid == 0) {
1850 			if (dzp->z_mode & S_ISGID) {
1851 				char		*domain;
1852 				uint32_t	rid;
1853 
1854 				acl_ids->z_fgid = KGID_TO_SGID(
1855 				    ZTOI(dzp)->i_gid);
1856 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1857 				    cr, ZFS_GROUP);
1858 
1859 				if (zfsvfs->z_use_fuids &&
1860 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1861 					domain = zfs_fuid_idx_domain(
1862 					    &zfsvfs->z_fuid_idx,
1863 					    FUID_INDEX(acl_ids->z_fgid));
1864 					rid = FUID_RID(acl_ids->z_fgid);
1865 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1866 					    domain, rid,
1867 					    FUID_INDEX(acl_ids->z_fgid),
1868 					    acl_ids->z_fgid, ZFS_GROUP);
1869 				}
1870 			} else {
1871 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1872 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1873 				gid = crgetgid(cr);
1874 			}
1875 		}
1876 	}
1877 #endif /* HAVE_KSID */
1878 
1879 	/*
1880 	 * If we're creating a directory, and the parent directory has the
1881 	 * set-GID bit set, set in on the new directory.
1882 	 * Otherwise, if the user is neither privileged nor a member of the
1883 	 * file's new group, clear the file's set-GID bit.
1884 	 */
1885 
1886 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1887 	    (S_ISDIR(vap->va_mode))) {
1888 		acl_ids->z_mode |= S_ISGID;
1889 	} else {
1890 		if ((acl_ids->z_mode & S_ISGID) &&
1891 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1892 			acl_ids->z_mode &= ~S_ISGID;
1893 	}
1894 
1895 	if (acl_ids->z_aclp == NULL) {
1896 		mutex_enter(&dzp->z_acl_lock);
1897 		mutex_enter(&dzp->z_lock);
1898 		if (!(flag & IS_ROOT_NODE) &&
1899 		    (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1900 		    !(dzp->z_pflags & ZFS_XATTR)) {
1901 			VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1902 			    &paclp, B_FALSE));
1903 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1904 			    vap->va_mode, paclp, acl_ids->z_mode, &need_chmod);
1905 			inherited = B_TRUE;
1906 		} else {
1907 			acl_ids->z_aclp =
1908 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1909 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1910 		}
1911 		mutex_exit(&dzp->z_lock);
1912 		mutex_exit(&dzp->z_acl_lock);
1913 
1914 		if (need_chmod) {
1915 			if (S_ISDIR(vap->va_mode))
1916 				acl_ids->z_aclp->z_hints |=
1917 				    ZFS_ACL_AUTO_INHERIT;
1918 
1919 			if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1920 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1921 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1922 				trim = B_TRUE;
1923 			zfs_acl_chmod(vap->va_mode, acl_ids->z_mode, B_FALSE,
1924 			    trim, acl_ids->z_aclp);
1925 		}
1926 	}
1927 
1928 	if (inherited || vsecp) {
1929 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1930 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1931 		    acl_ids->z_fuid, acl_ids->z_fgid);
1932 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1933 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1934 	}
1935 
1936 	return (0);
1937 }
1938 
1939 /*
1940  * Free ACL and fuid_infop, but not the acl_ids structure
1941  */
1942 void
1943 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1944 {
1945 	if (acl_ids->z_aclp)
1946 		zfs_acl_free(acl_ids->z_aclp);
1947 	if (acl_ids->z_fuidp)
1948 		zfs_fuid_info_free(acl_ids->z_fuidp);
1949 	acl_ids->z_aclp = NULL;
1950 	acl_ids->z_fuidp = NULL;
1951 }
1952 
1953 boolean_t
1954 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1955 {
1956 	return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1957 	    zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1958 	    (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1959 	    zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1960 }
1961 
1962 /*
1963  * Retrieve a file's ACL
1964  */
1965 int
1966 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1967 {
1968 	zfs_acl_t	*aclp;
1969 	ulong_t		mask;
1970 	int		error;
1971 	int 		count = 0;
1972 	int		largeace = 0;
1973 
1974 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1975 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1976 
1977 	if (mask == 0)
1978 		return (SET_ERROR(ENOSYS));
1979 
1980 	if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr)))
1981 		return (error);
1982 
1983 	mutex_enter(&zp->z_acl_lock);
1984 
1985 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1986 	if (error != 0) {
1987 		mutex_exit(&zp->z_acl_lock);
1988 		return (error);
1989 	}
1990 
1991 	/*
1992 	 * Scan ACL to determine number of ACEs
1993 	 */
1994 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1995 		void *zacep = NULL;
1996 		uint64_t who;
1997 		uint32_t access_mask;
1998 		uint16_t type, iflags;
1999 
2000 		while ((zacep = zfs_acl_next_ace(aclp, zacep,
2001 		    &who, &access_mask, &iflags, &type))) {
2002 			switch (type) {
2003 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
2004 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
2005 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
2006 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
2007 				largeace++;
2008 				continue;
2009 			default:
2010 				count++;
2011 			}
2012 		}
2013 		vsecp->vsa_aclcnt = count;
2014 	} else
2015 		count = (int)aclp->z_acl_count;
2016 
2017 	if (mask & VSA_ACECNT) {
2018 		vsecp->vsa_aclcnt = count;
2019 	}
2020 
2021 	if (mask & VSA_ACE) {
2022 		size_t aclsz;
2023 
2024 		aclsz = count * sizeof (ace_t) +
2025 		    sizeof (ace_object_t) * largeace;
2026 
2027 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2028 		vsecp->vsa_aclentsz = aclsz;
2029 
2030 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2031 			zfs_copy_fuid_2_ace(ZTOZSB(zp), aclp, cr,
2032 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2033 		else {
2034 			zfs_acl_node_t *aclnode;
2035 			void *start = vsecp->vsa_aclentp;
2036 
2037 			for (aclnode = list_head(&aclp->z_acl); aclnode;
2038 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
2039 				memcpy(start, aclnode->z_acldata,
2040 				    aclnode->z_size);
2041 				start = (caddr_t)start + aclnode->z_size;
2042 			}
2043 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
2044 			    aclp->z_acl_bytes);
2045 		}
2046 	}
2047 	if (mask & VSA_ACE_ACLFLAGS) {
2048 		vsecp->vsa_aclflags = 0;
2049 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
2050 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
2051 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
2052 			vsecp->vsa_aclflags |= ACL_PROTECTED;
2053 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
2054 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2055 	}
2056 
2057 	mutex_exit(&zp->z_acl_lock);
2058 
2059 	return (0);
2060 }
2061 
2062 int
2063 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_mode,
2064     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2065 {
2066 	zfs_acl_t *aclp;
2067 	zfs_acl_node_t *aclnode;
2068 	int aclcnt = vsecp->vsa_aclcnt;
2069 	int error;
2070 
2071 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2072 		return (SET_ERROR(EINVAL));
2073 
2074 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2075 
2076 	aclp->z_hints = 0;
2077 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2078 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2079 		if ((error = zfs_copy_ace_2_oldace(obj_mode, aclp,
2080 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2081 		    aclcnt, &aclnode->z_size)) != 0) {
2082 			zfs_acl_free(aclp);
2083 			zfs_acl_node_free(aclnode);
2084 			return (error);
2085 		}
2086 	} else {
2087 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_mode, aclp,
2088 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2089 		    &aclnode->z_size, fuidp, cr)) != 0) {
2090 			zfs_acl_free(aclp);
2091 			zfs_acl_node_free(aclnode);
2092 			return (error);
2093 		}
2094 	}
2095 	aclp->z_acl_bytes = aclnode->z_size;
2096 	aclnode->z_ace_count = aclcnt;
2097 	aclp->z_acl_count = aclcnt;
2098 	list_insert_head(&aclp->z_acl, aclnode);
2099 
2100 	/*
2101 	 * If flags are being set then add them to z_hints
2102 	 */
2103 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2104 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
2105 			aclp->z_hints |= ZFS_ACL_PROTECTED;
2106 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2107 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
2108 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2109 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2110 	}
2111 
2112 	*zaclp = aclp;
2113 
2114 	return (0);
2115 }
2116 
2117 /*
2118  * Set a file's ACL
2119  */
2120 int
2121 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2122 {
2123 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
2124 	zilog_t		*zilog = zfsvfs->z_log;
2125 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2126 	dmu_tx_t	*tx;
2127 	int		error;
2128 	zfs_acl_t	*aclp;
2129 	zfs_fuid_info_t	*fuidp = NULL;
2130 	boolean_t	fuid_dirtied;
2131 	uint64_t	acl_obj;
2132 
2133 	if (mask == 0)
2134 		return (SET_ERROR(ENOSYS));
2135 
2136 	if (zp->z_pflags & ZFS_IMMUTABLE)
2137 		return (SET_ERROR(EPERM));
2138 
2139 	if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)))
2140 		return (error);
2141 
2142 	error = zfs_vsec_2_aclp(zfsvfs, ZTOI(zp)->i_mode, vsecp, cr, &fuidp,
2143 	    &aclp);
2144 	if (error)
2145 		return (error);
2146 
2147 	/*
2148 	 * If ACL wide flags aren't being set then preserve any
2149 	 * existing flags.
2150 	 */
2151 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2152 		aclp->z_hints |=
2153 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2154 	}
2155 top:
2156 	mutex_enter(&zp->z_acl_lock);
2157 	mutex_enter(&zp->z_lock);
2158 
2159 	tx = dmu_tx_create(zfsvfs->z_os);
2160 
2161 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2162 
2163 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2164 	if (fuid_dirtied)
2165 		zfs_fuid_txhold(zfsvfs, tx);
2166 
2167 	/*
2168 	 * If old version and ACL won't fit in bonus and we aren't
2169 	 * upgrading then take out necessary DMU holds
2170 	 */
2171 
2172 	if ((acl_obj = zfs_external_acl(zp)) != 0) {
2173 		if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2174 		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2175 			dmu_tx_hold_free(tx, acl_obj, 0,
2176 			    DMU_OBJECT_END);
2177 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2178 			    aclp->z_acl_bytes);
2179 		} else {
2180 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2181 		}
2182 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2183 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2184 	}
2185 
2186 	zfs_sa_upgrade_txholds(tx, zp);
2187 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2188 	if (error) {
2189 		mutex_exit(&zp->z_acl_lock);
2190 		mutex_exit(&zp->z_lock);
2191 
2192 		if (error == ERESTART) {
2193 			dmu_tx_wait(tx);
2194 			dmu_tx_abort(tx);
2195 			goto top;
2196 		}
2197 		dmu_tx_abort(tx);
2198 		zfs_acl_free(aclp);
2199 		return (error);
2200 	}
2201 
2202 	error = zfs_aclset_common(zp, aclp, cr, tx);
2203 	ASSERT(error == 0);
2204 	ASSERT(zp->z_acl_cached == NULL);
2205 	zp->z_acl_cached = aclp;
2206 
2207 	if (fuid_dirtied)
2208 		zfs_fuid_sync(zfsvfs, tx);
2209 
2210 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2211 
2212 	if (fuidp)
2213 		zfs_fuid_info_free(fuidp);
2214 	dmu_tx_commit(tx);
2215 
2216 	mutex_exit(&zp->z_lock);
2217 	mutex_exit(&zp->z_acl_lock);
2218 
2219 	return (error);
2220 }
2221 
2222 /*
2223  * Check accesses of interest (AoI) against attributes of the dataset
2224  * such as read-only.  Returns zero if no AoI conflict with dataset
2225  * attributes, otherwise an appropriate errno is returned.
2226  */
2227 static int
2228 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2229 {
2230 	if ((v4_mode & WRITE_MASK) && (zfs_is_readonly(ZTOZSB(zp))) &&
2231 	    (!Z_ISDEV(ZTOI(zp)->i_mode) ||
2232 	    (Z_ISDEV(ZTOI(zp)->i_mode) && (v4_mode & WRITE_MASK_ATTRS)))) {
2233 		return (SET_ERROR(EROFS));
2234 	}
2235 
2236 	/*
2237 	 * Intentionally allow ZFS_READONLY through here.
2238 	 * See zfs_zaccess_common().
2239 	 */
2240 	if ((v4_mode & WRITE_MASK_DATA) &&
2241 	    (zp->z_pflags & ZFS_IMMUTABLE)) {
2242 		return (SET_ERROR(EPERM));
2243 	}
2244 
2245 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2246 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2247 		return (SET_ERROR(EPERM));
2248 	}
2249 
2250 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2251 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2252 		return (SET_ERROR(EACCES));
2253 	}
2254 
2255 	return (0);
2256 }
2257 
2258 /*
2259  * The primary usage of this function is to loop through all of the
2260  * ACEs in the znode, determining what accesses of interest (AoI) to
2261  * the caller are allowed or denied.  The AoI are expressed as bits in
2262  * the working_mode parameter.  As each ACE is processed, bits covered
2263  * by that ACE are removed from the working_mode.  This removal
2264  * facilitates two things.  The first is that when the working mode is
2265  * empty (= 0), we know we've looked at all the AoI. The second is
2266  * that the ACE interpretation rules don't allow a later ACE to undo
2267  * something granted or denied by an earlier ACE.  Removing the
2268  * discovered access or denial enforces this rule.  At the end of
2269  * processing the ACEs, all AoI that were found to be denied are
2270  * placed into the working_mode, giving the caller a mask of denied
2271  * accesses.  Returns:
2272  *	0		if all AoI granted
2273  *	EACCES 		if the denied mask is non-zero
2274  *	other error	if abnormal failure (e.g., IO error)
2275  *
2276  * A secondary usage of the function is to determine if any of the
2277  * AoI are granted.  If an ACE grants any access in
2278  * the working_mode, we immediately short circuit out of the function.
2279  * This mode is chosen by setting anyaccess to B_TRUE.  The
2280  * working_mode is not a denied access mask upon exit if the function
2281  * is used in this manner.
2282  */
2283 static int
2284 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2285     boolean_t anyaccess, cred_t *cr)
2286 {
2287 	zfsvfs_t	*zfsvfs = ZTOZSB(zp);
2288 	zfs_acl_t	*aclp;
2289 	int		error;
2290 	uid_t		uid = crgetuid(cr);
2291 	uint64_t	who;
2292 	uint16_t	type, iflags;
2293 	uint16_t	entry_type;
2294 	uint32_t	access_mask;
2295 	uint32_t	deny_mask = 0;
2296 	zfs_ace_hdr_t	*acep = NULL;
2297 	boolean_t	checkit;
2298 	uid_t		gowner;
2299 	uid_t		fowner;
2300 
2301 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2302 
2303 	mutex_enter(&zp->z_acl_lock);
2304 
2305 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2306 	if (error != 0) {
2307 		mutex_exit(&zp->z_acl_lock);
2308 		return (error);
2309 	}
2310 
2311 	ASSERT(zp->z_acl_cached);
2312 
2313 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2314 	    &iflags, &type))) {
2315 		uint32_t mask_matched;
2316 
2317 		if (!zfs_acl_valid_ace_type(type, iflags))
2318 			continue;
2319 
2320 		if (S_ISDIR(ZTOI(zp)->i_mode) &&
2321 		    (iflags & ACE_INHERIT_ONLY_ACE))
2322 			continue;
2323 
2324 		/* Skip ACE if it does not affect any AoI */
2325 		mask_matched = (access_mask & *working_mode);
2326 		if (!mask_matched)
2327 			continue;
2328 
2329 		entry_type = (iflags & ACE_TYPE_FLAGS);
2330 
2331 		checkit = B_FALSE;
2332 
2333 		switch (entry_type) {
2334 		case ACE_OWNER:
2335 			if (uid == fowner)
2336 				checkit = B_TRUE;
2337 			break;
2338 		case OWNING_GROUP:
2339 			who = gowner;
2340 			zfs_fallthrough;
2341 		case ACE_IDENTIFIER_GROUP:
2342 			checkit = zfs_groupmember(zfsvfs, who, cr);
2343 			break;
2344 		case ACE_EVERYONE:
2345 			checkit = B_TRUE;
2346 			break;
2347 
2348 		/* USER Entry */
2349 		default:
2350 			if (entry_type == 0) {
2351 				uid_t newid;
2352 
2353 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2354 				    ZFS_ACE_USER);
2355 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2356 				    uid == newid)
2357 					checkit = B_TRUE;
2358 				break;
2359 			} else {
2360 				mutex_exit(&zp->z_acl_lock);
2361 				return (SET_ERROR(EIO));
2362 			}
2363 		}
2364 
2365 		if (checkit) {
2366 			if (type == DENY) {
2367 				DTRACE_PROBE3(zfs__ace__denies,
2368 				    znode_t *, zp,
2369 				    zfs_ace_hdr_t *, acep,
2370 				    uint32_t, mask_matched);
2371 				deny_mask |= mask_matched;
2372 			} else {
2373 				DTRACE_PROBE3(zfs__ace__allows,
2374 				    znode_t *, zp,
2375 				    zfs_ace_hdr_t *, acep,
2376 				    uint32_t, mask_matched);
2377 				if (anyaccess) {
2378 					mutex_exit(&zp->z_acl_lock);
2379 					return (0);
2380 				}
2381 			}
2382 			*working_mode &= ~mask_matched;
2383 		}
2384 
2385 		/* Are we done? */
2386 		if (*working_mode == 0)
2387 			break;
2388 	}
2389 
2390 	mutex_exit(&zp->z_acl_lock);
2391 
2392 	/* Put the found 'denies' back on the working mode */
2393 	if (deny_mask) {
2394 		*working_mode |= deny_mask;
2395 		return (SET_ERROR(EACCES));
2396 	} else if (*working_mode) {
2397 		return (-1);
2398 	}
2399 
2400 	return (0);
2401 }
2402 
2403 /*
2404  * Return true if any access whatsoever granted, we don't actually
2405  * care what access is granted.
2406  */
2407 boolean_t
2408 zfs_has_access(znode_t *zp, cred_t *cr)
2409 {
2410 	uint32_t have = ACE_ALL_PERMS;
2411 
2412 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2413 		uid_t owner;
2414 
2415 		owner = zfs_fuid_map_id(ZTOZSB(zp),
2416 		    KUID_TO_SUID(ZTOI(zp)->i_uid), cr, ZFS_OWNER);
2417 		return (secpolicy_vnode_any_access(cr, ZTOI(zp), owner) == 0);
2418 	}
2419 	return (B_TRUE);
2420 }
2421 
2422 /*
2423  * Simplified access check for case where ACL is known to not contain
2424  * information beyond what is defined in the mode. In this case, we
2425  * can pass along to the kernel / vfs generic_permission() check, which
2426  * evaluates the mode and POSIX ACL.
2427  *
2428  * NFSv4 ACLs allow granting permissions that are usually relegated only
2429  * to the file owner or superuser. Examples are ACE_WRITE_OWNER (chown),
2430  * ACE_WRITE_ACL(chmod), and ACE_DELETE. ACE_DELETE requests must fail
2431  * because with conventional posix permissions, right to delete file
2432  * is determined by write bit on the parent dir.
2433  *
2434  * If unmappable perms are requested, then we must return EPERM
2435  * and include those bits in the working_mode so that the caller of
2436  * zfs_zaccess_common() can decide whether to perform additional
2437  * policy / capability checks. EACCES is used in zfs_zaccess_aces_check()
2438  * to indicate access check failed due to explicit DENY entry, and so
2439  * we want to avoid that here.
2440  */
2441 static int
2442 zfs_zaccess_trivial(znode_t *zp, uint32_t *working_mode, cred_t *cr)
2443 {
2444 	int err, mask;
2445 	int unmapped = 0;
2446 
2447 	ASSERT(zp->z_pflags & ZFS_ACL_TRIVIAL);
2448 
2449 	mask = zfs_v4_to_unix(*working_mode, &unmapped);
2450 	if (mask == 0 || unmapped) {
2451 		*working_mode = unmapped;
2452 		return (unmapped ? SET_ERROR(EPERM) : 0);
2453 	}
2454 
2455 #if defined(HAVE_IOPS_PERMISSION_USERNS)
2456 	err = generic_permission(cr->user_ns, ZTOI(zp), mask);
2457 #else
2458 	err = generic_permission(ZTOI(zp), mask);
2459 #endif
2460 	if (err != 0) {
2461 		return (SET_ERROR(EPERM));
2462 	}
2463 
2464 	*working_mode = unmapped;
2465 
2466 	return (0);
2467 }
2468 
2469 static int
2470 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2471     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2472 {
2473 	zfsvfs_t *zfsvfs = ZTOZSB(zp);
2474 	int err;
2475 
2476 	*working_mode = v4_mode;
2477 	*check_privs = B_TRUE;
2478 
2479 	/*
2480 	 * Short circuit empty requests
2481 	 */
2482 	if (v4_mode == 0 || zfsvfs->z_replay) {
2483 		*working_mode = 0;
2484 		return (0);
2485 	}
2486 
2487 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2488 		*check_privs = B_FALSE;
2489 		return (err);
2490 	}
2491 
2492 	/*
2493 	 * The caller requested that the ACL check be skipped.  This
2494 	 * would only happen if the caller checked VOP_ACCESS() with a
2495 	 * 32 bit ACE mask and already had the appropriate permissions.
2496 	 */
2497 	if (skipaclchk) {
2498 		*working_mode = 0;
2499 		return (0);
2500 	}
2501 
2502 	/*
2503 	 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2504 	 * When that flag is set, we should behave as if write access
2505 	 * were not granted by anything in the ACL.  In particular:
2506 	 * We _must_ allow writes after opening the file r/w, then
2507 	 * setting the DOS R/O attribute, and writing some more.
2508 	 * (Similar to how you can write after fchmod(fd, 0444).)
2509 	 *
2510 	 * Therefore ZFS_READONLY is ignored in the dataset check
2511 	 * above, and checked here as if part of the ACL check.
2512 	 * Also note: DOS R/O is ignored for directories.
2513 	 */
2514 	if ((v4_mode & WRITE_MASK_DATA) &&
2515 	    S_ISDIR(ZTOI(zp)->i_mode) &&
2516 	    (zp->z_pflags & ZFS_READONLY)) {
2517 		return (SET_ERROR(EPERM));
2518 	}
2519 
2520 	if (zp->z_pflags & ZFS_ACL_TRIVIAL)
2521 		return (zfs_zaccess_trivial(zp, working_mode, cr));
2522 
2523 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2524 }
2525 
2526 static int
2527 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2528     cred_t *cr)
2529 {
2530 	if (*working_mode != ACE_WRITE_DATA)
2531 		return (SET_ERROR(EACCES));
2532 
2533 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2534 	    check_privs, B_FALSE, cr));
2535 }
2536 
2537 int
2538 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2539 {
2540 	boolean_t owner = B_FALSE;
2541 	boolean_t groupmbr = B_FALSE;
2542 	boolean_t is_attr;
2543 	uid_t uid = crgetuid(cr);
2544 	int error;
2545 
2546 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2547 		return (SET_ERROR(EACCES));
2548 
2549 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2550 	    (S_ISDIR(ZTOI(zdp)->i_mode)));
2551 	if (is_attr)
2552 		goto slow;
2553 
2554 
2555 	mutex_enter(&zdp->z_acl_lock);
2556 
2557 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2558 		mutex_exit(&zdp->z_acl_lock);
2559 		return (0);
2560 	}
2561 
2562 	if (KUID_TO_SUID(ZTOI(zdp)->i_uid) != 0 ||
2563 	    KGID_TO_SGID(ZTOI(zdp)->i_gid) != 0) {
2564 		mutex_exit(&zdp->z_acl_lock);
2565 		goto slow;
2566 	}
2567 
2568 	if (uid == KUID_TO_SUID(ZTOI(zdp)->i_uid)) {
2569 		owner = B_TRUE;
2570 		if (zdp->z_mode & S_IXUSR) {
2571 			mutex_exit(&zdp->z_acl_lock);
2572 			return (0);
2573 		} else {
2574 			mutex_exit(&zdp->z_acl_lock);
2575 			goto slow;
2576 		}
2577 	}
2578 	if (groupmember(KGID_TO_SGID(ZTOI(zdp)->i_gid), cr)) {
2579 		groupmbr = B_TRUE;
2580 		if (zdp->z_mode & S_IXGRP) {
2581 			mutex_exit(&zdp->z_acl_lock);
2582 			return (0);
2583 		} else {
2584 			mutex_exit(&zdp->z_acl_lock);
2585 			goto slow;
2586 		}
2587 	}
2588 	if (!owner && !groupmbr) {
2589 		if (zdp->z_mode & S_IXOTH) {
2590 			mutex_exit(&zdp->z_acl_lock);
2591 			return (0);
2592 		}
2593 	}
2594 
2595 	mutex_exit(&zdp->z_acl_lock);
2596 
2597 slow:
2598 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2599 	ZFS_ENTER(ZTOZSB(zdp));
2600 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2601 	ZFS_EXIT(ZTOZSB(zdp));
2602 	return (error);
2603 }
2604 
2605 /*
2606  * Determine whether Access should be granted/denied.
2607  *
2608  * The least priv subsystem is always consulted as a basic privilege
2609  * can define any form of access.
2610  */
2611 int
2612 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2613 {
2614 	uint32_t	working_mode;
2615 	int		error;
2616 	int		is_attr;
2617 	boolean_t 	check_privs;
2618 	znode_t		*xzp;
2619 	znode_t 	*check_zp = zp;
2620 	mode_t		needed_bits;
2621 	uid_t		owner;
2622 
2623 	is_attr = ((zp->z_pflags & ZFS_XATTR) && S_ISDIR(ZTOI(zp)->i_mode));
2624 
2625 	/*
2626 	 * If attribute then validate against base file
2627 	 */
2628 	if (is_attr) {
2629 		if ((error = zfs_zget(ZTOZSB(zp),
2630 		    zp->z_xattr_parent, &xzp)) != 0) {
2631 			return (error);
2632 		}
2633 
2634 		check_zp = xzp;
2635 
2636 		/*
2637 		 * fixup mode to map to xattr perms
2638 		 */
2639 
2640 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2641 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2642 			mode |= ACE_WRITE_NAMED_ATTRS;
2643 		}
2644 
2645 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2646 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2647 			mode |= ACE_READ_NAMED_ATTRS;
2648 		}
2649 	}
2650 
2651 	owner = zfs_fuid_map_id(ZTOZSB(zp), KUID_TO_SUID(ZTOI(zp)->i_uid),
2652 	    cr, ZFS_OWNER);
2653 	/*
2654 	 * Map the bits required to the standard inode flags
2655 	 * S_IRUSR|S_IWUSR|S_IXUSR in the needed_bits.  Map the bits
2656 	 * mapped by working_mode (currently missing) in missing_bits.
2657 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2658 	 * needed_bits.
2659 	 */
2660 	needed_bits = 0;
2661 
2662 	working_mode = mode;
2663 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2664 	    owner == crgetuid(cr))
2665 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2666 
2667 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2668 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2669 		needed_bits |= S_IRUSR;
2670 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2671 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2672 		needed_bits |= S_IWUSR;
2673 	if (working_mode & ACE_EXECUTE)
2674 		needed_bits |= S_IXUSR;
2675 
2676 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2677 	    &check_privs, skipaclchk, cr)) == 0) {
2678 		if (is_attr)
2679 			zrele(xzp);
2680 		return (secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2681 		    needed_bits, needed_bits));
2682 	}
2683 
2684 	if (error && !check_privs) {
2685 		if (is_attr)
2686 			zrele(xzp);
2687 		return (error);
2688 	}
2689 
2690 	if (error && (flags & V_APPEND)) {
2691 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2692 	}
2693 
2694 	if (error && check_privs) {
2695 		mode_t		checkmode = 0;
2696 
2697 		/*
2698 		 * First check for implicit owner permission on
2699 		 * read_acl/read_attributes
2700 		 */
2701 
2702 		error = 0;
2703 		ASSERT(working_mode != 0);
2704 
2705 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2706 		    owner == crgetuid(cr)))
2707 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2708 
2709 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2710 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2711 			checkmode |= S_IRUSR;
2712 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2713 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2714 			checkmode |= S_IWUSR;
2715 		if (working_mode & ACE_EXECUTE)
2716 			checkmode |= S_IXUSR;
2717 
2718 		error = secpolicy_vnode_access2(cr, ZTOI(check_zp), owner,
2719 		    needed_bits & ~checkmode, needed_bits);
2720 
2721 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2722 			error = secpolicy_vnode_chown(cr, owner);
2723 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2724 			error = secpolicy_vnode_setdac(cr, owner);
2725 
2726 		if (error == 0 && (working_mode &
2727 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2728 			error = secpolicy_vnode_remove(cr);
2729 
2730 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2731 			error = secpolicy_vnode_chown(cr, owner);
2732 		}
2733 		if (error == 0) {
2734 			/*
2735 			 * See if any bits other than those already checked
2736 			 * for are still present.  If so then return EACCES
2737 			 */
2738 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2739 				error = SET_ERROR(EACCES);
2740 			}
2741 		}
2742 	} else if (error == 0) {
2743 		error = secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2744 		    needed_bits, needed_bits);
2745 	}
2746 
2747 	if (is_attr)
2748 		zrele(xzp);
2749 
2750 	return (error);
2751 }
2752 
2753 /*
2754  * Translate traditional unix S_IRUSR/S_IWUSR/S_IXUSR mode into
2755  * NFSv4-style ZFS ACL format and call zfs_zaccess()
2756  */
2757 int
2758 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2759 {
2760 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2761 }
2762 
2763 /*
2764  * Access function for secpolicy_vnode_setattr
2765  */
2766 int
2767 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2768 {
2769 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2770 
2771 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2772 }
2773 
2774 /* See zfs_zaccess_delete() */
2775 static const boolean_t zfs_write_implies_delete_child = B_TRUE;
2776 
2777 /*
2778  * Determine whether delete access should be granted.
2779  *
2780  * The following chart outlines how we handle delete permissions which is
2781  * how recent versions of windows (Windows 2008) handles it.  The efficiency
2782  * comes from not having to check the parent ACL where the object itself grants
2783  * delete:
2784  *
2785  *      -------------------------------------------------------
2786  *      |   Parent Dir  |      Target Object Permissions      |
2787  *      |  permissions  |                                     |
2788  *      -------------------------------------------------------
2789  *      |               | ACL Allows | ACL Denies| Delete     |
2790  *      |               |  Delete    |  Delete   | unspecified|
2791  *      -------------------------------------------------------
2792  *      | ACL Allows    | Permit     | Deny *    | Permit     |
2793  *      | DELETE_CHILD  |            |           |            |
2794  *      -------------------------------------------------------
2795  *      | ACL Denies    | Permit     | Deny      | Deny       |
2796  *      | DELETE_CHILD  |            |           |            |
2797  *      -------------------------------------------------------
2798  *      | ACL specifies |            |           |            |
2799  *      | only allow    | Permit     | Deny *    | Permit     |
2800  *      | write and     |            |           |            |
2801  *      | execute       |            |           |            |
2802  *      -------------------------------------------------------
2803  *      | ACL denies    |            |           |            |
2804  *      | write and     | Permit     | Deny      | Deny       |
2805  *      | execute       |            |           |            |
2806  *      -------------------------------------------------------
2807  *         ^
2808  *         |
2809  *         Re. execute permission on the directory:  if that's missing,
2810  *	   the vnode lookup of the target will fail before we get here.
2811  *
2812  * Re [*] in the table above:  NFSv4 would normally Permit delete for
2813  * these two cells of the matrix.
2814  * See acl.h for notes on which ACE_... flags should be checked for which
2815  * operations.  Specifically, the NFSv4 committee recommendation is in
2816  * conflict with the Windows interpretation of DENY ACEs, where DENY ACEs
2817  * should take precedence ahead of ALLOW ACEs.
2818  *
2819  * This implementation always consults the target object's ACL first.
2820  * If a DENY ACE is present on the target object that specifies ACE_DELETE,
2821  * delete access is denied.  If an ALLOW ACE with ACE_DELETE is present on
2822  * the target object, access is allowed.  If and only if no entries with
2823  * ACE_DELETE are present in the object's ACL, check the container's ACL
2824  * for entries with ACE_DELETE_CHILD.
2825  *
2826  * A summary of the logic implemented from the table above is as follows:
2827  *
2828  * First check for DENY ACEs that apply.
2829  * If either target or container has a deny, EACCES.
2830  *
2831  * Delete access can then be summarized as follows:
2832  * 1: The object to be deleted grants ACE_DELETE, or
2833  * 2: The containing directory grants ACE_DELETE_CHILD.
2834  * In a Windows system, that would be the end of the story.
2835  * In this system, (2) has some complications...
2836  * 2a: "sticky" bit on a directory adds restrictions, and
2837  * 2b: existing ACEs from previous versions of ZFS may
2838  * not carry ACE_DELETE_CHILD where they should, so we
2839  * also allow delete when ACE_WRITE_DATA is granted.
2840  *
2841  * Note: 2b is technically a work-around for a prior bug,
2842  * which hopefully can go away some day.  For those who
2843  * no longer need the work around, and for testing, this
2844  * work-around is made conditional via the tunable:
2845  * zfs_write_implies_delete_child
2846  */
2847 int
2848 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2849 {
2850 	uint32_t wanted_dirperms;
2851 	uint32_t dzp_working_mode = 0;
2852 	uint32_t zp_working_mode = 0;
2853 	int dzp_error, zp_error;
2854 	boolean_t dzpcheck_privs;
2855 	boolean_t zpcheck_privs;
2856 
2857 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2858 		return (SET_ERROR(EPERM));
2859 
2860 	/*
2861 	 * Case 1:
2862 	 * If target object grants ACE_DELETE then we are done.  This is
2863 	 * indicated by a return value of 0.  For this case we don't worry
2864 	 * about the sticky bit because sticky only applies to the parent
2865 	 * directory and this is the child access result.
2866 	 *
2867 	 * If we encounter a DENY ACE here, we're also done (EACCES).
2868 	 * Note that if we hit a DENY ACE here (on the target) it should
2869 	 * take precedence over a DENY ACE on the container, so that when
2870 	 * we have more complete auditing support we will be able to
2871 	 * report an access failure against the specific target.
2872 	 * (This is part of why we're checking the target first.)
2873 	 */
2874 	zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2875 	    &zpcheck_privs, B_FALSE, cr);
2876 	if (zp_error == EACCES) {
2877 		/* We hit a DENY ACE. */
2878 		if (!zpcheck_privs)
2879 			return (SET_ERROR(zp_error));
2880 		return (secpolicy_vnode_remove(cr));
2881 
2882 	}
2883 	if (zp_error == 0)
2884 		return (0);
2885 
2886 	/*
2887 	 * Case 2:
2888 	 * If the containing directory grants ACE_DELETE_CHILD,
2889 	 * or we're in backward compatibility mode and the
2890 	 * containing directory has ACE_WRITE_DATA, allow.
2891 	 * Case 2b is handled with wanted_dirperms.
2892 	 */
2893 	wanted_dirperms = ACE_DELETE_CHILD;
2894 	if (zfs_write_implies_delete_child)
2895 		wanted_dirperms |= ACE_WRITE_DATA;
2896 	dzp_error = zfs_zaccess_common(dzp, wanted_dirperms,
2897 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2898 	if (dzp_error == EACCES) {
2899 		/* We hit a DENY ACE. */
2900 		if (!dzpcheck_privs)
2901 			return (SET_ERROR(dzp_error));
2902 		return (secpolicy_vnode_remove(cr));
2903 	}
2904 
2905 	/*
2906 	 * Cases 2a, 2b (continued)
2907 	 *
2908 	 * Note: dzp_working_mode now contains any permissions
2909 	 * that were NOT granted.  Therefore, if any of the
2910 	 * wanted_dirperms WERE granted, we will have:
2911 	 *   dzp_working_mode != wanted_dirperms
2912 	 * We're really asking if ANY of those permissions
2913 	 * were granted, and if so, grant delete access.
2914 	 */
2915 	if (dzp_working_mode != wanted_dirperms)
2916 		dzp_error = 0;
2917 
2918 	/*
2919 	 * dzp_error is 0 if the container granted us permissions to "modify".
2920 	 * If we do not have permission via one or more ACEs, our current
2921 	 * privileges may still permit us to modify the container.
2922 	 *
2923 	 * dzpcheck_privs is false when i.e. the FS is read-only.
2924 	 * Otherwise, do privilege checks for the container.
2925 	 */
2926 	if (dzp_error != 0 && dzpcheck_privs) {
2927 		uid_t owner;
2928 
2929 		/*
2930 		 * The secpolicy call needs the requested access and
2931 		 * the current access mode of the container, but it
2932 		 * only knows about Unix-style modes (VEXEC, VWRITE),
2933 		 * so this must condense the fine-grained ACE bits into
2934 		 * Unix modes.
2935 		 *
2936 		 * The VEXEC flag is easy, because we know that has
2937 		 * always been checked before we get here (during the
2938 		 * lookup of the target vnode).  The container has not
2939 		 * granted us permissions to "modify", so we do not set
2940 		 * the VWRITE flag in the current access mode.
2941 		 */
2942 		owner = zfs_fuid_map_id(ZTOZSB(dzp),
2943 		    KUID_TO_SUID(ZTOI(dzp)->i_uid), cr, ZFS_OWNER);
2944 		dzp_error = secpolicy_vnode_access2(cr, ZTOI(dzp),
2945 		    owner, S_IXUSR, S_IWUSR|S_IXUSR);
2946 	}
2947 	if (dzp_error != 0) {
2948 		/*
2949 		 * Note: We may have dzp_error = -1 here (from
2950 		 * zfs_zacess_common).  Don't return that.
2951 		 */
2952 		return (SET_ERROR(EACCES));
2953 	}
2954 
2955 
2956 	/*
2957 	 * At this point, we know that the directory permissions allow
2958 	 * us to modify, but we still need to check for the additional
2959 	 * restrictions that apply when the "sticky bit" is set.
2960 	 *
2961 	 * Yes, zfs_sticky_remove_access() also checks this bit, but
2962 	 * checking it here and skipping the call below is nice when
2963 	 * you're watching all of this with dtrace.
2964 	 */
2965 	if ((dzp->z_mode & S_ISVTX) == 0)
2966 		return (0);
2967 
2968 	/*
2969 	 * zfs_sticky_remove_access will succeed if:
2970 	 * 1. The sticky bit is absent.
2971 	 * 2. We pass the sticky bit restrictions.
2972 	 * 3. We have privileges that always allow file removal.
2973 	 */
2974 	return (zfs_sticky_remove_access(dzp, zp, cr));
2975 }
2976 
2977 int
2978 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2979     znode_t *tzp, cred_t *cr)
2980 {
2981 	int add_perm;
2982 	int error;
2983 
2984 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2985 		return (SET_ERROR(EACCES));
2986 
2987 	add_perm = S_ISDIR(ZTOI(szp)->i_mode) ?
2988 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2989 
2990 	/*
2991 	 * Rename permissions are combination of delete permission +
2992 	 * add file/subdir permission.
2993 	 */
2994 
2995 	/*
2996 	 * first make sure we do the delete portion.
2997 	 *
2998 	 * If that succeeds then check for add_file/add_subdir permissions
2999 	 */
3000 
3001 	if ((error = zfs_zaccess_delete(sdzp, szp, cr)))
3002 		return (error);
3003 
3004 	/*
3005 	 * If we have a tzp, see if we can delete it?
3006 	 */
3007 	if (tzp) {
3008 		if ((error = zfs_zaccess_delete(tdzp, tzp, cr)))
3009 			return (error);
3010 	}
3011 
3012 	/*
3013 	 * Now check for add permissions
3014 	 */
3015 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
3016 
3017 	return (error);
3018 }
3019