1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2013 by Delphix. All rights reserved.
24  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
25  */
26 
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/time.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
32 #include <sys/resource.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/file.h>
36 #include <sys/stat.h>
37 #include <sys/kmem.h>
38 #include <sys/cmn_err.h>
39 #include <sys/errno.h>
40 #include <sys/unistd.h>
41 #include <sys/sdt.h>
42 #include <sys/fs/zfs.h>
43 #include <sys/policy.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_fuid.h>
46 #include <sys/zfs_acl.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_quota.h>
49 #include <sys/zfs_vfsops.h>
50 #include <sys/dmu.h>
51 #include <sys/dnode.h>
52 #include <sys/zap.h>
53 #include <sys/sa.h>
54 #include <acl/acl_common.h>
55 
56 
57 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
58 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
59 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
60 #define	MIN_ACE_TYPE	ALLOW
61 
62 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
63 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
64     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
65 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
66     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
67 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
68     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
69 
70 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
71     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
72     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
73     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
74 
75 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
76 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
77     ACE_DELETE|ACE_DELETE_CHILD)
78 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
79 
80 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
81     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
82 
83 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
84     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
85 
86 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
87     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
88 
89 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
90 
91 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
92     ZFS_ACL_PROTECTED)
93 
94 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
95     ZFS_ACL_OBJ_ACE)
96 
97 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
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 /*ARGSUSED*/
148 static size_t
149 zfs_ace_v0_size(void *acep)
150 {
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 /*ARGSUSED*/
167 static int
168 zfs_ace_v0_data(void *acep, void **datap)
169 {
170 	*datap = NULL;
171 	return (0);
172 }
173 
174 static acl_ops_t zfs_acl_v0_ops = {
175 	zfs_ace_v0_get_mask,
176 	zfs_ace_v0_set_mask,
177 	zfs_ace_v0_get_flags,
178 	zfs_ace_v0_set_flags,
179 	zfs_ace_v0_get_type,
180 	zfs_ace_v0_set_type,
181 	zfs_ace_v0_get_who,
182 	zfs_ace_v0_set_who,
183 	zfs_ace_v0_size,
184 	zfs_ace_v0_abstract_size,
185 	zfs_ace_v0_mask_off,
186 	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 		/*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 acl_ops_t zfs_acl_fuid_ops = {
311 	zfs_ace_fuid_get_mask,
312 	zfs_ace_fuid_set_mask,
313 	zfs_ace_fuid_get_flags,
314 	zfs_ace_fuid_set_flags,
315 	zfs_ace_fuid_get_type,
316 	zfs_ace_fuid_set_type,
317 	zfs_ace_fuid_get_who,
318 	zfs_ace_fuid_set_who,
319 	zfs_ace_fuid_size,
320 	zfs_ace_fuid_abstract_size,
321 	zfs_ace_fuid_mask_off,
322 	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(zp->z_zfsvfs),
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 = zp->z_zfsvfs;
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(zp->z_zfsvfs),
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(zp->z_zfsvfs->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(vtype_t obj_type, 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 (obj_type == VDIR &&
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 /*ARGSUSED*/
633 static uint64_t
634 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
635     uint16_t *flags, uint16_t *type, uint32_t *mask)
636 {
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, vtype_t obj_type, 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_type, 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 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
691 			    sizeof (aceobjp->a_obj_type));
692 			bcopy(aceobjp->a_inherit_obj_type,
693 			    zobjacep->z_inherit_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 			bcopy(zobjacep->z_object_type,
741 			    objacep->a_obj_type,
742 			    sizeof (zobjacep->z_object_type));
743 			bcopy(zobjacep->z_inherit_type,
744 			    objacep->a_inherit_obj_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(vtype_t obj_type, 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_type, 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 	 * everytime.
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(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
830 	    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 static void
867 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
868     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
869 {
870 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
871 
872 	aclp->z_ops->ace_mask_set(acep, access_mask);
873 	aclp->z_ops->ace_type_set(acep, access_type);
874 	aclp->z_ops->ace_flags_set(acep, entry_type);
875 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
876 	    type != ACE_EVERYONE))
877 		aclp->z_ops->ace_who_set(acep, fuid);
878 }
879 
880 /*
881  * Determine mode of file based on ACL.
882  */
883 uint64_t
884 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
885     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
886 {
887 	int		entry_type;
888 	mode_t		mode;
889 	mode_t		seen = 0;
890 	zfs_ace_hdr_t 	*acep = NULL;
891 	uint64_t	who;
892 	uint16_t	iflags, type;
893 	uint32_t	access_mask;
894 	boolean_t	an_exec_denied = B_FALSE;
895 
896 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
897 
898 	while ((acep = zfs_acl_next_ace(aclp, acep, &who,
899 	    &access_mask, &iflags, &type))) {
900 
901 		if (!zfs_acl_valid_ace_type(type, iflags))
902 			continue;
903 
904 		entry_type = (iflags & ACE_TYPE_FLAGS);
905 
906 		/*
907 		 * Skip over any inherit_only ACEs
908 		 */
909 		if (iflags & ACE_INHERIT_ONLY_ACE)
910 			continue;
911 
912 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
913 		    who == fuid)) {
914 			if ((access_mask & ACE_READ_DATA) &&
915 			    (!(seen & S_IRUSR))) {
916 				seen |= S_IRUSR;
917 				if (type == ALLOW) {
918 					mode |= S_IRUSR;
919 				}
920 			}
921 			if ((access_mask & ACE_WRITE_DATA) &&
922 			    (!(seen & S_IWUSR))) {
923 				seen |= S_IWUSR;
924 				if (type == ALLOW) {
925 					mode |= S_IWUSR;
926 				}
927 			}
928 			if ((access_mask & ACE_EXECUTE) &&
929 			    (!(seen & S_IXUSR))) {
930 				seen |= S_IXUSR;
931 				if (type == ALLOW) {
932 					mode |= S_IXUSR;
933 				}
934 			}
935 		} else if (entry_type == OWNING_GROUP ||
936 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
937 			if ((access_mask & ACE_READ_DATA) &&
938 			    (!(seen & S_IRGRP))) {
939 				seen |= S_IRGRP;
940 				if (type == ALLOW) {
941 					mode |= S_IRGRP;
942 				}
943 			}
944 			if ((access_mask & ACE_WRITE_DATA) &&
945 			    (!(seen & S_IWGRP))) {
946 				seen |= S_IWGRP;
947 				if (type == ALLOW) {
948 					mode |= S_IWGRP;
949 				}
950 			}
951 			if ((access_mask & ACE_EXECUTE) &&
952 			    (!(seen & S_IXGRP))) {
953 				seen |= S_IXGRP;
954 				if (type == ALLOW) {
955 					mode |= S_IXGRP;
956 				}
957 			}
958 		} else if (entry_type == ACE_EVERYONE) {
959 			if ((access_mask & ACE_READ_DATA)) {
960 				if (!(seen & S_IRUSR)) {
961 					seen |= S_IRUSR;
962 					if (type == ALLOW) {
963 						mode |= S_IRUSR;
964 					}
965 				}
966 				if (!(seen & S_IRGRP)) {
967 					seen |= S_IRGRP;
968 					if (type == ALLOW) {
969 						mode |= S_IRGRP;
970 					}
971 				}
972 				if (!(seen & S_IROTH)) {
973 					seen |= S_IROTH;
974 					if (type == ALLOW) {
975 						mode |= S_IROTH;
976 					}
977 				}
978 			}
979 			if ((access_mask & ACE_WRITE_DATA)) {
980 				if (!(seen & S_IWUSR)) {
981 					seen |= S_IWUSR;
982 					if (type == ALLOW) {
983 						mode |= S_IWUSR;
984 					}
985 				}
986 				if (!(seen & S_IWGRP)) {
987 					seen |= S_IWGRP;
988 					if (type == ALLOW) {
989 						mode |= S_IWGRP;
990 					}
991 				}
992 				if (!(seen & S_IWOTH)) {
993 					seen |= S_IWOTH;
994 					if (type == ALLOW) {
995 						mode |= S_IWOTH;
996 					}
997 				}
998 			}
999 			if ((access_mask & ACE_EXECUTE)) {
1000 				if (!(seen & S_IXUSR)) {
1001 					seen |= S_IXUSR;
1002 					if (type == ALLOW) {
1003 						mode |= S_IXUSR;
1004 					}
1005 				}
1006 				if (!(seen & S_IXGRP)) {
1007 					seen |= S_IXGRP;
1008 					if (type == ALLOW) {
1009 						mode |= S_IXGRP;
1010 					}
1011 				}
1012 				if (!(seen & S_IXOTH)) {
1013 					seen |= S_IXOTH;
1014 					if (type == ALLOW) {
1015 						mode |= S_IXOTH;
1016 					}
1017 				}
1018 			}
1019 		} else {
1020 			/*
1021 			 * Only care if this IDENTIFIER_GROUP or
1022 			 * USER ACE denies execute access to someone,
1023 			 * mode is not affected
1024 			 */
1025 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1026 				an_exec_denied = B_TRUE;
1027 		}
1028 	}
1029 
1030 	/*
1031 	 * Failure to allow is effectively a deny, so execute permission
1032 	 * is denied if it was never mentioned or if we explicitly
1033 	 * weren't allowed it.
1034 	 */
1035 	if (!an_exec_denied &&
1036 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1037 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1038 		an_exec_denied = B_TRUE;
1039 
1040 	if (an_exec_denied)
1041 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1042 	else
1043 		*pflags |= ZFS_NO_EXECS_DENIED;
1044 
1045 	return (mode);
1046 }
1047 
1048 /*
1049  * Read an external acl object.  If the intent is to modify, always
1050  * create a new acl and leave any cached acl in place.
1051  */
1052 int
1053 zfs_acl_node_read(znode_t *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1054     boolean_t will_modify)
1055 {
1056 	zfs_acl_t	*aclp;
1057 	int		aclsize;
1058 	int		acl_count;
1059 	zfs_acl_node_t	*aclnode;
1060 	zfs_acl_phys_t	znode_acl;
1061 	int		version;
1062 	int		error;
1063 
1064 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1065 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1066 		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1067 
1068 	if (zp->z_acl_cached && !will_modify) {
1069 		*aclpp = zp->z_acl_cached;
1070 		return (0);
1071 	}
1072 
1073 	version = zfs_znode_acl_version(zp);
1074 
1075 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1076 	    &acl_count, &znode_acl)) != 0) {
1077 		goto done;
1078 	}
1079 
1080 	aclp = zfs_acl_alloc(version);
1081 
1082 	aclp->z_acl_count = acl_count;
1083 	aclp->z_acl_bytes = aclsize;
1084 
1085 	aclnode = zfs_acl_node_alloc(aclsize);
1086 	aclnode->z_ace_count = aclp->z_acl_count;
1087 	aclnode->z_size = aclsize;
1088 
1089 	if (!zp->z_is_sa) {
1090 		if (znode_acl.z_acl_extern_obj) {
1091 			error = dmu_read(zp->z_zfsvfs->z_os,
1092 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1093 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1094 		} else {
1095 			bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1096 			    aclnode->z_size);
1097 		}
1098 	} else {
1099 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1100 		    aclnode->z_acldata, aclnode->z_size);
1101 	}
1102 
1103 	if (error != 0) {
1104 		zfs_acl_free(aclp);
1105 		zfs_acl_node_free(aclnode);
1106 		/* convert checksum errors into IO errors */
1107 		if (error == ECKSUM)
1108 			error = SET_ERROR(EIO);
1109 		goto done;
1110 	}
1111 
1112 	list_insert_head(&aclp->z_acl, aclnode);
1113 
1114 	*aclpp = aclp;
1115 	if (!will_modify)
1116 		zp->z_acl_cached = aclp;
1117 done:
1118 	return (error);
1119 }
1120 
1121 /*ARGSUSED*/
1122 void
1123 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1124     boolean_t start, void *userdata)
1125 {
1126 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1127 
1128 	if (start) {
1129 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1130 	} else {
1131 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1132 		    cb->cb_acl_node);
1133 	}
1134 	*dataptr = cb->cb_acl_node->z_acldata;
1135 	*length = cb->cb_acl_node->z_size;
1136 }
1137 
1138 int
1139 zfs_acl_chown_setattr(znode_t *zp)
1140 {
1141 	int error;
1142 	zfs_acl_t *aclp;
1143 
1144 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1145 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1146 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1147 	ASSERT_VOP_IN_SEQC(ZTOV(zp));
1148 
1149 	if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1150 		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1151 		    &zp->z_pflags, zp->z_uid, zp->z_gid);
1152 	return (error);
1153 }
1154 
1155 /*
1156  * common code for setting ACLs.
1157  *
1158  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1159  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1160  * already checked the acl and knows whether to inherit.
1161  */
1162 int
1163 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1164 {
1165 	int			error;
1166 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1167 	dmu_object_type_t	otype;
1168 	zfs_acl_locator_cb_t	locate = { 0 };
1169 	uint64_t		mode;
1170 	sa_bulk_attr_t		bulk[5];
1171 	uint64_t		ctime[2];
1172 	int			count = 0;
1173 	zfs_acl_phys_t		acl_phys;
1174 
1175 	ASSERT_VOP_IN_SEQC(ZTOV(zp));
1176 
1177 	mode = zp->z_mode;
1178 
1179 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1180 	    zp->z_uid, zp->z_gid);
1181 
1182 	zp->z_mode = mode;
1183 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1184 	    &mode, sizeof (mode));
1185 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1186 	    &zp->z_pflags, sizeof (zp->z_pflags));
1187 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1188 	    &ctime, sizeof (ctime));
1189 
1190 	if (zp->z_acl_cached) {
1191 		zfs_acl_free(zp->z_acl_cached);
1192 		zp->z_acl_cached = NULL;
1193 	}
1194 
1195 	/*
1196 	 * Upgrade needed?
1197 	 */
1198 	if (!zfsvfs->z_use_fuids) {
1199 		otype = DMU_OT_OLDACL;
1200 	} else {
1201 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1202 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1203 			zfs_acl_xform(zp, aclp, cr);
1204 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1205 		otype = DMU_OT_ACL;
1206 	}
1207 
1208 	/*
1209 	 * Arrgh, we have to handle old on disk format
1210 	 * as well as newer (preferred) SA format.
1211 	 */
1212 
1213 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1214 		locate.cb_aclp = aclp;
1215 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1216 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1217 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1218 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1219 	} else { /* Painful legacy way */
1220 		zfs_acl_node_t *aclnode;
1221 		uint64_t off = 0;
1222 		uint64_t aoid;
1223 
1224 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1225 		    &acl_phys, sizeof (acl_phys))) != 0)
1226 			return (error);
1227 
1228 		aoid = acl_phys.z_acl_extern_obj;
1229 
1230 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1231 			/*
1232 			 * If ACL was previously external and we are now
1233 			 * converting to new ACL format then release old
1234 			 * ACL object and create a new one.
1235 			 */
1236 			if (aoid &&
1237 			    aclp->z_version != acl_phys.z_acl_version) {
1238 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1239 				if (error)
1240 					return (error);
1241 				aoid = 0;
1242 			}
1243 			if (aoid == 0) {
1244 				aoid = dmu_object_alloc(zfsvfs->z_os,
1245 				    otype, aclp->z_acl_bytes,
1246 				    otype == DMU_OT_ACL ?
1247 				    DMU_OT_SYSACL : DMU_OT_NONE,
1248 				    otype == DMU_OT_ACL ?
1249 				    DN_OLD_MAX_BONUSLEN : 0, tx);
1250 			} else {
1251 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1252 				    aoid, aclp->z_acl_bytes, 0, tx);
1253 			}
1254 			acl_phys.z_acl_extern_obj = aoid;
1255 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1256 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1257 				if (aclnode->z_ace_count == 0)
1258 					continue;
1259 				dmu_write(zfsvfs->z_os, aoid, off,
1260 				    aclnode->z_size, aclnode->z_acldata, tx);
1261 				off += aclnode->z_size;
1262 			}
1263 		} else {
1264 			void *start = acl_phys.z_ace_data;
1265 			/*
1266 			 * Migrating back embedded?
1267 			 */
1268 			if (acl_phys.z_acl_extern_obj) {
1269 				error = dmu_object_free(zfsvfs->z_os,
1270 				    acl_phys.z_acl_extern_obj, tx);
1271 				if (error)
1272 					return (error);
1273 				acl_phys.z_acl_extern_obj = 0;
1274 			}
1275 
1276 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1277 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1278 				if (aclnode->z_ace_count == 0)
1279 					continue;
1280 				bcopy(aclnode->z_acldata, start,
1281 				    aclnode->z_size);
1282 				start = (caddr_t)start + aclnode->z_size;
1283 			}
1284 		}
1285 		/*
1286 		 * If Old version then swap count/bytes to match old
1287 		 * layout of znode_acl_phys_t.
1288 		 */
1289 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1290 			acl_phys.z_acl_size = aclp->z_acl_count;
1291 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1292 		} else {
1293 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1294 			acl_phys.z_acl_count = aclp->z_acl_count;
1295 		}
1296 		acl_phys.z_acl_version = aclp->z_version;
1297 
1298 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1299 		    &acl_phys, sizeof (acl_phys));
1300 	}
1301 
1302 	/*
1303 	 * Replace ACL wide bits, but first clear them.
1304 	 */
1305 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1306 
1307 	zp->z_pflags |= aclp->z_hints;
1308 
1309 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1310 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1311 
1312 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1313 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1314 }
1315 
1316 static void
1317 zfs_acl_chmod(vtype_t vtype, uint64_t mode, boolean_t split, boolean_t trim,
1318     zfs_acl_t *aclp)
1319 {
1320 	void		*acep = NULL;
1321 	uint64_t	who;
1322 	int		new_count, new_bytes;
1323 	int		ace_size;
1324 	int 		entry_type;
1325 	uint16_t	iflags, type;
1326 	uint32_t	access_mask;
1327 	zfs_acl_node_t	*newnode;
1328 	size_t 		abstract_size = aclp->z_ops->ace_abstract_size();
1329 	void 		*zacep;
1330 	boolean_t	isdir;
1331 	trivial_acl_t	masks;
1332 
1333 	new_count = new_bytes = 0;
1334 
1335 	isdir = (vtype == VDIR);
1336 
1337 	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1338 
1339 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1340 
1341 	zacep = newnode->z_acldata;
1342 	if (masks.allow0) {
1343 		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1344 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1345 		new_count++;
1346 		new_bytes += abstract_size;
1347 	}
1348 	if (masks.deny1) {
1349 		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1350 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1351 		new_count++;
1352 		new_bytes += abstract_size;
1353 	}
1354 	if (masks.deny2) {
1355 		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1356 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1357 		new_count++;
1358 		new_bytes += abstract_size;
1359 	}
1360 
1361 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1362 	    &iflags, &type))) {
1363 		entry_type = (iflags & ACE_TYPE_FLAGS);
1364 		/*
1365 		 * ACEs used to represent the file mode may be divided
1366 		 * into an equivalent pair of inherit-only and regular
1367 		 * ACEs, if they are inheritable.
1368 		 * Skip regular ACEs, which are replaced by the new mode.
1369 		 */
1370 		if (split && (entry_type == ACE_OWNER ||
1371 		    entry_type == OWNING_GROUP ||
1372 		    entry_type == ACE_EVERYONE)) {
1373 			if (!isdir || !(iflags &
1374 			    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1375 				continue;
1376 			/*
1377 			 * We preserve owner@, group@, or @everyone
1378 			 * permissions, if they are inheritable, by
1379 			 * copying them to inherit_only ACEs. This
1380 			 * prevents inheritable permissions from being
1381 			 * altered along with the file mode.
1382 			 */
1383 			iflags |= ACE_INHERIT_ONLY_ACE;
1384 		}
1385 
1386 		/*
1387 		 * If this ACL has any inheritable ACEs, mark that in
1388 		 * the hints (which are later masked into the pflags)
1389 		 * so create knows to do inheritance.
1390 		 */
1391 		if (isdir && (iflags &
1392 		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1393 			aclp->z_hints |= ZFS_INHERIT_ACE;
1394 
1395 		if ((type != ALLOW && type != DENY) ||
1396 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1397 			switch (type) {
1398 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1399 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1400 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1401 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1402 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1403 				break;
1404 			}
1405 		} else {
1406 			/*
1407 			 * Limit permissions granted by ACEs to be no greater
1408 			 * than permissions of the requested group mode.
1409 			 * Applies when the "aclmode" property is set to
1410 			 * "groupmask".
1411 			 */
1412 			if ((type == ALLOW) && trim)
1413 				access_mask &= masks.group;
1414 		}
1415 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1416 		ace_size = aclp->z_ops->ace_size(acep);
1417 		zacep = (void *)((uintptr_t)zacep + ace_size);
1418 		new_count++;
1419 		new_bytes += ace_size;
1420 	}
1421 	zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1422 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1423 	zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1424 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1425 	zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1426 
1427 	new_count += 3;
1428 	new_bytes += abstract_size * 3;
1429 	zfs_acl_release_nodes(aclp);
1430 	aclp->z_acl_count = new_count;
1431 	aclp->z_acl_bytes = new_bytes;
1432 	newnode->z_ace_count = new_count;
1433 	newnode->z_size = new_bytes;
1434 	list_insert_tail(&aclp->z_acl, newnode);
1435 }
1436 
1437 int
1438 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1439 {
1440 	int error = 0;
1441 
1442 	mutex_enter(&zp->z_acl_lock);
1443 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1444 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1445 	if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1446 		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1447 	else
1448 		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1449 
1450 	if (error == 0) {
1451 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1452 		zfs_acl_chmod(ZTOV(zp)->v_type, mode, B_TRUE,
1453 		    (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1454 	}
1455 	mutex_exit(&zp->z_acl_lock);
1456 
1457 	return (error);
1458 }
1459 
1460 /*
1461  * Should ACE be inherited?
1462  */
1463 static int
1464 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1465 {
1466 	int	iflags = (acep_flags & 0xf);
1467 
1468 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1469 		return (1);
1470 	else if (iflags & ACE_FILE_INHERIT_ACE)
1471 		return (!((vtype == VDIR) &&
1472 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1473 	return (0);
1474 }
1475 
1476 /*
1477  * inherit inheritable ACEs from parent
1478  */
1479 static zfs_acl_t *
1480 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1481     uint64_t mode, boolean_t *need_chmod)
1482 {
1483 	void		*pacep = NULL;
1484 	void		*acep;
1485 	zfs_acl_node_t  *aclnode;
1486 	zfs_acl_t	*aclp = NULL;
1487 	uint64_t	who;
1488 	uint32_t	access_mask;
1489 	uint16_t	iflags, newflags, type;
1490 	size_t		ace_size;
1491 	void		*data1, *data2;
1492 	size_t		data1sz, data2sz;
1493 	uint_t		aclinherit;
1494 	boolean_t	isdir = (vtype == VDIR);
1495 	boolean_t	isreg = (vtype == VREG);
1496 
1497 	*need_chmod = B_TRUE;
1498 
1499 	aclp = zfs_acl_alloc(paclp->z_version);
1500 	aclinherit = zfsvfs->z_acl_inherit;
1501 	if (aclinherit == ZFS_ACL_DISCARD || vtype == VLNK)
1502 		return (aclp);
1503 
1504 	while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1505 	    &access_mask, &iflags, &type))) {
1506 
1507 		/*
1508 		 * don't inherit bogus ACEs
1509 		 */
1510 		if (!zfs_acl_valid_ace_type(type, iflags))
1511 			continue;
1512 
1513 		/*
1514 		 * Check if ACE is inheritable by this vnode
1515 		 */
1516 		if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1517 		    !zfs_ace_can_use(vtype, iflags))
1518 			continue;
1519 
1520 		/*
1521 		 * If owner@, group@, or everyone@ inheritable
1522 		 * then zfs_acl_chmod() isn't needed.
1523 		 */
1524 		if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1525 		    aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1526 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1527 		    ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1528 		    (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1529 			*need_chmod = B_FALSE;
1530 
1531 		/*
1532 		 * Strip inherited execute permission from file if
1533 		 * not in mode
1534 		 */
1535 		if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1536 		    !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1537 			access_mask &= ~ACE_EXECUTE;
1538 		}
1539 
1540 		/*
1541 		 * Strip write_acl and write_owner from permissions
1542 		 * when inheriting an ACE
1543 		 */
1544 		if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1545 			access_mask &= ~RESTRICTED_CLEAR;
1546 		}
1547 
1548 		ace_size = aclp->z_ops->ace_size(pacep);
1549 		aclnode = zfs_acl_node_alloc(ace_size);
1550 		list_insert_tail(&aclp->z_acl, aclnode);
1551 		acep = aclnode->z_acldata;
1552 
1553 		zfs_set_ace(aclp, acep, access_mask, type,
1554 		    who, iflags|ACE_INHERITED_ACE);
1555 
1556 		/*
1557 		 * Copy special opaque data if any
1558 		 */
1559 		if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1560 			VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1561 			    &data2)) == data1sz);
1562 			bcopy(data1, data2, data2sz);
1563 		}
1564 
1565 		aclp->z_acl_count++;
1566 		aclnode->z_ace_count++;
1567 		aclp->z_acl_bytes += aclnode->z_size;
1568 		newflags = aclp->z_ops->ace_flags_get(acep);
1569 
1570 		/*
1571 		 * If ACE is not to be inherited further, or if the vnode is
1572 		 * not a directory, remove all inheritance flags
1573 		 */
1574 		if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1575 			newflags &= ~ALL_INHERIT;
1576 			aclp->z_ops->ace_flags_set(acep,
1577 			    newflags|ACE_INHERITED_ACE);
1578 			continue;
1579 		}
1580 
1581 		/*
1582 		 * This directory has an inheritable ACE
1583 		 */
1584 		aclp->z_hints |= ZFS_INHERIT_ACE;
1585 
1586 		/*
1587 		 * If only FILE_INHERIT is set then turn on
1588 		 * inherit_only
1589 		 */
1590 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1591 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1592 			newflags |= ACE_INHERIT_ONLY_ACE;
1593 			aclp->z_ops->ace_flags_set(acep,
1594 			    newflags|ACE_INHERITED_ACE);
1595 		} else {
1596 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1597 			aclp->z_ops->ace_flags_set(acep,
1598 			    newflags|ACE_INHERITED_ACE);
1599 		}
1600 	}
1601 	if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1602 	    aclp->z_acl_count != 0) {
1603 		*need_chmod = B_FALSE;
1604 	}
1605 
1606 	return (aclp);
1607 }
1608 
1609 /*
1610  * Create file system object initial permissions
1611  * including inheritable ACEs.
1612  * Also, create FUIDs for owner and group.
1613  */
1614 int
1615 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1616     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1617 {
1618 	int		error;
1619 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1620 	zfs_acl_t	*paclp;
1621 	gid_t		gid;
1622 	boolean_t	need_chmod = B_TRUE;
1623 	boolean_t	trim = B_FALSE;
1624 	boolean_t	inherited = B_FALSE;
1625 
1626 	if ((flag & IS_ROOT_NODE) == 0) {
1627 		if (zfsvfs->z_replay == B_FALSE)
1628 			ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
1629 	} else
1630 		ASSERT(dzp->z_vnode == NULL);
1631 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1632 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1633 
1634 	if (vsecp)
1635 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1636 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1637 			return (error);
1638 	/*
1639 	 * Determine uid and gid.
1640 	 */
1641 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1642 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1643 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1644 		    (uint64_t)vap->va_uid, cr,
1645 		    ZFS_OWNER, &acl_ids->z_fuidp);
1646 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1647 		    (uint64_t)vap->va_gid, cr,
1648 		    ZFS_GROUP, &acl_ids->z_fuidp);
1649 		gid = vap->va_gid;
1650 	} else {
1651 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1652 		    cr, &acl_ids->z_fuidp);
1653 		acl_ids->z_fgid = 0;
1654 		if (vap->va_mask & AT_GID)  {
1655 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1656 			    (uint64_t)vap->va_gid,
1657 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1658 			gid = vap->va_gid;
1659 			if (acl_ids->z_fgid != dzp->z_gid &&
1660 			    !groupmember(vap->va_gid, cr) &&
1661 			    secpolicy_vnode_create_gid(cr) != 0)
1662 				acl_ids->z_fgid = 0;
1663 		}
1664 		if (acl_ids->z_fgid == 0) {
1665 			char		*domain;
1666 			uint32_t	rid;
1667 
1668 			acl_ids->z_fgid = dzp->z_gid;
1669 			gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1670 			    cr, ZFS_GROUP);
1671 
1672 			if (zfsvfs->z_use_fuids &&
1673 			    IS_EPHEMERAL(acl_ids->z_fgid)) {
1674 				domain =
1675 				    zfs_fuid_idx_domain(&zfsvfs->z_fuid_idx,
1676 				    FUID_INDEX(acl_ids->z_fgid));
1677 				rid = FUID_RID(acl_ids->z_fgid);
1678 				zfs_fuid_node_add(&acl_ids->z_fuidp,
1679 				    domain, rid, FUID_INDEX(acl_ids->z_fgid),
1680 				    acl_ids->z_fgid, ZFS_GROUP);
1681 			}
1682 		}
1683 	}
1684 
1685 	/*
1686 	 * If we're creating a directory, and the parent directory has the
1687 	 * set-GID bit set, set in on the new directory.
1688 	 * Otherwise, if the user is neither privileged nor a member of the
1689 	 * file's new group, clear the file's set-GID bit.
1690 	 */
1691 
1692 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1693 	    (vap->va_type == VDIR)) {
1694 		acl_ids->z_mode |= S_ISGID;
1695 	} else {
1696 		if ((acl_ids->z_mode & S_ISGID) &&
1697 		    secpolicy_vnode_setids_setgids(ZTOV(dzp), cr, gid) != 0)
1698 			acl_ids->z_mode &= ~S_ISGID;
1699 	}
1700 
1701 	if (acl_ids->z_aclp == NULL) {
1702 		mutex_enter(&dzp->z_acl_lock);
1703 		if (!(flag & IS_ROOT_NODE) &&
1704 		    (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1705 		    !(dzp->z_pflags & ZFS_XATTR)) {
1706 			VERIFY0(zfs_acl_node_read(dzp, B_TRUE,
1707 			    &paclp, B_FALSE));
1708 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1709 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1710 			inherited = B_TRUE;
1711 		} else {
1712 			acl_ids->z_aclp =
1713 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1714 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1715 		}
1716 		mutex_exit(&dzp->z_acl_lock);
1717 
1718 		if (need_chmod) {
1719 			if (vap->va_type == VDIR)
1720 				acl_ids->z_aclp->z_hints |=
1721 				    ZFS_ACL_AUTO_INHERIT;
1722 
1723 			if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1724 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1725 			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1726 				trim = B_TRUE;
1727 			zfs_acl_chmod(vap->va_type, acl_ids->z_mode, B_FALSE,
1728 			    trim, acl_ids->z_aclp);
1729 		}
1730 	}
1731 
1732 	if (inherited || vsecp) {
1733 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1734 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1735 		    acl_ids->z_fuid, acl_ids->z_fgid);
1736 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1737 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1738 	}
1739 
1740 	return (0);
1741 }
1742 
1743 /*
1744  * Free ACL and fuid_infop, but not the acl_ids structure
1745  */
1746 void
1747 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1748 {
1749 	if (acl_ids->z_aclp)
1750 		zfs_acl_free(acl_ids->z_aclp);
1751 	if (acl_ids->z_fuidp)
1752 		zfs_fuid_info_free(acl_ids->z_fuidp);
1753 	acl_ids->z_aclp = NULL;
1754 	acl_ids->z_fuidp = NULL;
1755 }
1756 
1757 boolean_t
1758 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1759 {
1760 	return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1761 	    zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1762 	    (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1763 	    zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1764 }
1765 
1766 /*
1767  * Retrieve a file's ACL
1768  */
1769 int
1770 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1771 {
1772 	zfs_acl_t	*aclp;
1773 	ulong_t		mask;
1774 	int		error;
1775 	int 		count = 0;
1776 	int		largeace = 0;
1777 
1778 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1779 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1780 
1781 	if (mask == 0)
1782 		return (SET_ERROR(ENOSYS));
1783 
1784 	if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr)))
1785 		return (error);
1786 
1787 	mutex_enter(&zp->z_acl_lock);
1788 
1789 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1790 		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1791 	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1792 	if (error != 0) {
1793 		mutex_exit(&zp->z_acl_lock);
1794 		return (error);
1795 	}
1796 
1797 	/*
1798 	 * Scan ACL to determine number of ACEs
1799 	 */
1800 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1801 		void *zacep = NULL;
1802 		uint64_t who;
1803 		uint32_t access_mask;
1804 		uint16_t type, iflags;
1805 
1806 		while ((zacep = zfs_acl_next_ace(aclp, zacep,
1807 		    &who, &access_mask, &iflags, &type))) {
1808 			switch (type) {
1809 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1810 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1811 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1812 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1813 				largeace++;
1814 				continue;
1815 			default:
1816 				count++;
1817 			}
1818 		}
1819 		vsecp->vsa_aclcnt = count;
1820 	} else
1821 		count = (int)aclp->z_acl_count;
1822 
1823 	if (mask & VSA_ACECNT) {
1824 		vsecp->vsa_aclcnt = count;
1825 	}
1826 
1827 	if (mask & VSA_ACE) {
1828 		size_t aclsz;
1829 
1830 		aclsz = count * sizeof (ace_t) +
1831 		    sizeof (ace_object_t) * largeace;
1832 
1833 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1834 		vsecp->vsa_aclentsz = aclsz;
1835 
1836 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1837 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1838 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1839 		else {
1840 			zfs_acl_node_t *aclnode;
1841 			void *start = vsecp->vsa_aclentp;
1842 
1843 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1844 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1845 				bcopy(aclnode->z_acldata, start,
1846 				    aclnode->z_size);
1847 				start = (caddr_t)start + aclnode->z_size;
1848 			}
1849 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1850 			    aclp->z_acl_bytes);
1851 		}
1852 	}
1853 	if (mask & VSA_ACE_ACLFLAGS) {
1854 		vsecp->vsa_aclflags = 0;
1855 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1856 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1857 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
1858 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1859 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1860 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1861 	}
1862 
1863 	mutex_exit(&zp->z_acl_lock);
1864 
1865 	return (0);
1866 }
1867 
1868 int
1869 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_type,
1870     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1871 {
1872 	zfs_acl_t *aclp;
1873 	zfs_acl_node_t *aclnode;
1874 	int aclcnt = vsecp->vsa_aclcnt;
1875 	int error;
1876 
1877 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1878 		return (SET_ERROR(EINVAL));
1879 
1880 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1881 
1882 	aclp->z_hints = 0;
1883 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1884 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1885 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1886 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1887 		    aclcnt, &aclnode->z_size)) != 0) {
1888 			zfs_acl_free(aclp);
1889 			zfs_acl_node_free(aclnode);
1890 			return (error);
1891 		}
1892 	} else {
1893 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1894 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1895 		    &aclnode->z_size, fuidp, cr)) != 0) {
1896 			zfs_acl_free(aclp);
1897 			zfs_acl_node_free(aclnode);
1898 			return (error);
1899 		}
1900 	}
1901 	aclp->z_acl_bytes = aclnode->z_size;
1902 	aclnode->z_ace_count = aclcnt;
1903 	aclp->z_acl_count = aclcnt;
1904 	list_insert_head(&aclp->z_acl, aclnode);
1905 
1906 	/*
1907 	 * If flags are being set then add them to z_hints
1908 	 */
1909 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1910 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
1911 			aclp->z_hints |= ZFS_ACL_PROTECTED;
1912 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1913 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
1914 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1915 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1916 	}
1917 
1918 	*zaclp = aclp;
1919 
1920 	return (0);
1921 }
1922 
1923 /*
1924  * Set a file's ACL
1925  */
1926 int
1927 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1928 {
1929 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1930 	zilog_t		*zilog = zfsvfs->z_log;
1931 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1932 	dmu_tx_t	*tx;
1933 	int		error;
1934 	zfs_acl_t	*aclp;
1935 	zfs_fuid_info_t	*fuidp = NULL;
1936 	boolean_t	fuid_dirtied;
1937 	uint64_t	acl_obj;
1938 
1939 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1940 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1941 	if (mask == 0)
1942 		return (SET_ERROR(ENOSYS));
1943 
1944 	if (zp->z_pflags & ZFS_IMMUTABLE)
1945 		return (SET_ERROR(EPERM));
1946 
1947 	if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)))
1948 		return (error);
1949 
1950 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1951 	    &aclp);
1952 	if (error)
1953 		return (error);
1954 
1955 	/*
1956 	 * If ACL wide flags aren't being set then preserve any
1957 	 * existing flags.
1958 	 */
1959 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1960 		aclp->z_hints |=
1961 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1962 	}
1963 top:
1964 	mutex_enter(&zp->z_acl_lock);
1965 
1966 	tx = dmu_tx_create(zfsvfs->z_os);
1967 
1968 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1969 
1970 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1971 	if (fuid_dirtied)
1972 		zfs_fuid_txhold(zfsvfs, tx);
1973 
1974 	/*
1975 	 * If old version and ACL won't fit in bonus and we aren't
1976 	 * upgrading then take out necessary DMU holds
1977 	 */
1978 
1979 	if ((acl_obj = zfs_external_acl(zp)) != 0) {
1980 		if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
1981 		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
1982 			dmu_tx_hold_free(tx, acl_obj, 0,
1983 			    DMU_OBJECT_END);
1984 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1985 			    aclp->z_acl_bytes);
1986 		} else {
1987 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
1988 		}
1989 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1990 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
1991 	}
1992 
1993 	zfs_sa_upgrade_txholds(tx, zp);
1994 	error = dmu_tx_assign(tx, TXG_NOWAIT);
1995 	if (error) {
1996 		mutex_exit(&zp->z_acl_lock);
1997 
1998 		if (error == ERESTART) {
1999 			dmu_tx_wait(tx);
2000 			dmu_tx_abort(tx);
2001 			goto top;
2002 		}
2003 		dmu_tx_abort(tx);
2004 		zfs_acl_free(aclp);
2005 		return (error);
2006 	}
2007 
2008 	error = zfs_aclset_common(zp, aclp, cr, tx);
2009 	ASSERT(error == 0);
2010 	ASSERT(zp->z_acl_cached == NULL);
2011 	zp->z_acl_cached = aclp;
2012 
2013 	if (fuid_dirtied)
2014 		zfs_fuid_sync(zfsvfs, tx);
2015 
2016 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2017 
2018 	if (fuidp)
2019 		zfs_fuid_info_free(fuidp);
2020 	dmu_tx_commit(tx);
2021 	mutex_exit(&zp->z_acl_lock);
2022 
2023 	return (error);
2024 }
2025 
2026 /*
2027  * Check accesses of interest (AoI) against attributes of the dataset
2028  * such as read-only.  Returns zero if no AoI conflict with dataset
2029  * attributes, otherwise an appropriate errno is returned.
2030  */
2031 static int
2032 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2033 {
2034 	if ((v4_mode & WRITE_MASK) &&
2035 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2036 	    (!IS_DEVVP(ZTOV(zp)) ||
2037 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2038 		return (SET_ERROR(EROFS));
2039 	}
2040 
2041 	/*
2042 	 * Intentionally allow ZFS_READONLY through here.
2043 	 * See zfs_zaccess_common().
2044 	 */
2045 	if ((v4_mode & WRITE_MASK_DATA) &&
2046 	    (zp->z_pflags & ZFS_IMMUTABLE)) {
2047 		return (SET_ERROR(EPERM));
2048 	}
2049 
2050 	/*
2051 	 * In FreeBSD we allow to modify directory's content is ZFS_NOUNLINK
2052 	 * (sunlnk) is set. We just don't allow directory removal, which is
2053 	 * handled in zfs_zaccess_delete().
2054 	 */
2055 	if ((v4_mode & ACE_DELETE) &&
2056 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2057 		return (EPERM);
2058 	}
2059 
2060 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2061 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2062 		return (SET_ERROR(EACCES));
2063 	}
2064 
2065 	return (0);
2066 }
2067 
2068 /*
2069  * The primary usage of this function is to loop through all of the
2070  * ACEs in the znode, determining what accesses of interest (AoI) to
2071  * the caller are allowed or denied.  The AoI are expressed as bits in
2072  * the working_mode parameter.  As each ACE is processed, bits covered
2073  * by that ACE are removed from the working_mode.  This removal
2074  * facilitates two things.  The first is that when the working mode is
2075  * empty (= 0), we know we've looked at all the AoI. The second is
2076  * that the ACE interpretation rules don't allow a later ACE to undo
2077  * something granted or denied by an earlier ACE.  Removing the
2078  * discovered access or denial enforces this rule.  At the end of
2079  * processing the ACEs, all AoI that were found to be denied are
2080  * placed into the working_mode, giving the caller a mask of denied
2081  * accesses.  Returns:
2082  *	0		if all AoI granted
2083  *	EACCESS 	if the denied mask is non-zero
2084  *	other error	if abnormal failure (e.g., IO error)
2085  *
2086  * A secondary usage of the function is to determine if any of the
2087  * AoI are granted.  If an ACE grants any access in
2088  * the working_mode, we immediately short circuit out of the function.
2089  * This mode is chosen by setting anyaccess to B_TRUE.  The
2090  * working_mode is not a denied access mask upon exit if the function
2091  * is used in this manner.
2092  */
2093 static int
2094 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2095     boolean_t anyaccess, cred_t *cr)
2096 {
2097 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2098 	zfs_acl_t	*aclp;
2099 	int		error;
2100 	uid_t		uid = crgetuid(cr);
2101 	uint64_t 	who;
2102 	uint16_t	type, iflags;
2103 	uint16_t	entry_type;
2104 	uint32_t	access_mask;
2105 	uint32_t	deny_mask = 0;
2106 	zfs_ace_hdr_t	*acep = NULL;
2107 	boolean_t	checkit;
2108 	uid_t		gowner;
2109 	uid_t		fowner;
2110 
2111 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2112 
2113 	mutex_enter(&zp->z_acl_lock);
2114 
2115 	if (zp->z_zfsvfs->z_replay == B_FALSE)
2116 		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
2117 	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
2118 	if (error != 0) {
2119 		mutex_exit(&zp->z_acl_lock);
2120 		return (error);
2121 	}
2122 
2123 	ASSERT(zp->z_acl_cached);
2124 
2125 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2126 	    &iflags, &type))) {
2127 		uint32_t mask_matched;
2128 
2129 		if (!zfs_acl_valid_ace_type(type, iflags))
2130 			continue;
2131 
2132 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2133 			continue;
2134 
2135 		/* Skip ACE if it does not affect any AoI */
2136 		mask_matched = (access_mask & *working_mode);
2137 		if (!mask_matched)
2138 			continue;
2139 
2140 		entry_type = (iflags & ACE_TYPE_FLAGS);
2141 
2142 		checkit = B_FALSE;
2143 
2144 		switch (entry_type) {
2145 		case ACE_OWNER:
2146 			if (uid == fowner)
2147 				checkit = B_TRUE;
2148 			break;
2149 		case OWNING_GROUP:
2150 			who = gowner;
2151 			/*FALLTHROUGH*/
2152 		case ACE_IDENTIFIER_GROUP:
2153 			checkit = zfs_groupmember(zfsvfs, who, cr);
2154 			break;
2155 		case ACE_EVERYONE:
2156 			checkit = B_TRUE;
2157 			break;
2158 
2159 		/* USER Entry */
2160 		default:
2161 			if (entry_type == 0) {
2162 				uid_t newid;
2163 
2164 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2165 				    ZFS_ACE_USER);
2166 				if (newid !=  UID_NOBODY &&
2167 				    uid == newid)
2168 					checkit = B_TRUE;
2169 				break;
2170 			} else {
2171 				mutex_exit(&zp->z_acl_lock);
2172 				return (SET_ERROR(EIO));
2173 			}
2174 		}
2175 
2176 		if (checkit) {
2177 			if (type == DENY) {
2178 				DTRACE_PROBE3(zfs__ace__denies,
2179 				    znode_t *, zp,
2180 				    zfs_ace_hdr_t *, acep,
2181 				    uint32_t, mask_matched);
2182 				deny_mask |= mask_matched;
2183 			} else {
2184 				DTRACE_PROBE3(zfs__ace__allows,
2185 				    znode_t *, zp,
2186 				    zfs_ace_hdr_t *, acep,
2187 				    uint32_t, mask_matched);
2188 				if (anyaccess) {
2189 					mutex_exit(&zp->z_acl_lock);
2190 					return (0);
2191 				}
2192 			}
2193 			*working_mode &= ~mask_matched;
2194 		}
2195 
2196 		/* Are we done? */
2197 		if (*working_mode == 0)
2198 			break;
2199 	}
2200 
2201 	mutex_exit(&zp->z_acl_lock);
2202 
2203 	/* Put the found 'denies' back on the working mode */
2204 	if (deny_mask) {
2205 		*working_mode |= deny_mask;
2206 		return (SET_ERROR(EACCES));
2207 	} else if (*working_mode) {
2208 		return (-1);
2209 	}
2210 
2211 	return (0);
2212 }
2213 
2214 /*
2215  * Return true if any access whatsoever granted, we don't actually
2216  * care what access is granted.
2217  */
2218 boolean_t
2219 zfs_has_access(znode_t *zp, cred_t *cr)
2220 {
2221 	uint32_t have = ACE_ALL_PERMS;
2222 
2223 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2224 		uid_t owner;
2225 
2226 		owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2227 		return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2228 	}
2229 	return (B_TRUE);
2230 }
2231 
2232 static int
2233 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2234     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2235 {
2236 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2237 	int err;
2238 
2239 	*working_mode = v4_mode;
2240 	*check_privs = B_TRUE;
2241 
2242 	/*
2243 	 * Short circuit empty requests
2244 	 */
2245 	if (v4_mode == 0 || zfsvfs->z_replay) {
2246 		*working_mode = 0;
2247 		return (0);
2248 	}
2249 
2250 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2251 		*check_privs = B_FALSE;
2252 		return (err);
2253 	}
2254 
2255 	/*
2256 	 * The caller requested that the ACL check be skipped.  This
2257 	 * would only happen if the caller checked VOP_ACCESS() with a
2258 	 * 32 bit ACE mask and already had the appropriate permissions.
2259 	 */
2260 	if (skipaclchk) {
2261 		*working_mode = 0;
2262 		return (0);
2263 	}
2264 
2265 	/*
2266 	 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2267 	 * When that flag is set, we should behave as if write access
2268 	 * were not granted by anything in the ACL.  In particular:
2269 	 * We _must_ allow writes after opening the file r/w, then
2270 	 * setting the DOS R/O attribute, and writing some more.
2271 	 * (Similar to how you can write after fchmod(fd, 0444).)
2272 	 *
2273 	 * Therefore ZFS_READONLY is ignored in the dataset check
2274 	 * above, and checked here as if part of the ACL check.
2275 	 * Also note: DOS R/O is ignored for directories.
2276 	 */
2277 	if ((v4_mode & WRITE_MASK_DATA) &&
2278 	    (ZTOV(zp)->v_type != VDIR) &&
2279 	    (zp->z_pflags & ZFS_READONLY)) {
2280 		return (SET_ERROR(EPERM));
2281 	}
2282 
2283 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2284 }
2285 
2286 static int
2287 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2288     cred_t *cr)
2289 {
2290 	if (*working_mode != ACE_WRITE_DATA)
2291 		return (SET_ERROR(EACCES));
2292 
2293 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2294 	    check_privs, B_FALSE, cr));
2295 }
2296 
2297 /*
2298  * Check if VEXEC is allowed.
2299  *
2300  * This routine is based on zfs_fastaccesschk_execute which has slowpath
2301  * calling zfs_zaccess. This would be incorrect on FreeBSD (see
2302  * zfs_freebsd_access for the difference). Thus this variant let's the
2303  * caller handle the slowpath (if necessary).
2304  *
2305  * On top of that we perform a lockless check for ZFS_NO_EXECS_DENIED.
2306  *
2307  * Safe access to znode_t is provided by the vnode lock.
2308  */
2309 int
2310 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2311 {
2312 	boolean_t is_attr;
2313 
2314 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2315 		return (1);
2316 
2317 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2318 	    (ZTOV(zdp)->v_type == VDIR));
2319 	if (is_attr)
2320 		return (1);
2321 
2322 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED)
2323 		return (0);
2324 
2325 	return (1);
2326 }
2327 
2328 
2329 /*
2330  * Determine whether Access should be granted/denied.
2331  *
2332  * The least priv subsystem is always consulted as a basic privilege
2333  * can define any form of access.
2334  */
2335 int
2336 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2337 {
2338 	uint32_t	working_mode;
2339 	int		error;
2340 	int		is_attr;
2341 	boolean_t 	check_privs;
2342 	znode_t		*xzp = NULL;
2343 	znode_t 	*check_zp = zp;
2344 	mode_t		needed_bits;
2345 	uid_t		owner;
2346 
2347 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2348 
2349 #ifdef __FreeBSD_kernel__
2350 	/*
2351 	 * In FreeBSD, we don't care about permissions of individual ADS.
2352 	 * Note that not checking them is not just an optimization - without
2353 	 * this shortcut, EA operations may bogusly fail with EACCES.
2354 	 */
2355 	if (zp->z_pflags & ZFS_XATTR)
2356 		return (0);
2357 #else
2358 	/*
2359 	 * If attribute then validate against base file
2360 	 */
2361 	if (is_attr) {
2362 		uint64_t	parent;
2363 
2364 		if ((error = sa_lookup(zp->z_sa_hdl,
2365 		    SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2366 		    sizeof (parent))) != 0)
2367 			return (error);
2368 
2369 		if ((error = zfs_zget(zp->z_zfsvfs,
2370 		    parent, &xzp)) != 0)	{
2371 			return (error);
2372 		}
2373 
2374 		check_zp = xzp;
2375 
2376 		/*
2377 		 * fixup mode to map to xattr perms
2378 		 */
2379 
2380 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2381 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2382 			mode |= ACE_WRITE_NAMED_ATTRS;
2383 		}
2384 
2385 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2386 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2387 			mode |= ACE_READ_NAMED_ATTRS;
2388 		}
2389 	}
2390 #endif
2391 
2392 	owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2393 	/*
2394 	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2395 	 * in needed_bits.  Map the bits mapped by working_mode (currently
2396 	 * missing) in missing_bits.
2397 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2398 	 * needed_bits.
2399 	 */
2400 	needed_bits = 0;
2401 
2402 	working_mode = mode;
2403 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2404 	    owner == crgetuid(cr))
2405 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2406 
2407 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2408 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2409 		needed_bits |= VREAD;
2410 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2411 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2412 		needed_bits |= VWRITE;
2413 	if (working_mode & ACE_EXECUTE)
2414 		needed_bits |= VEXEC;
2415 
2416 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2417 	    &check_privs, skipaclchk, cr)) == 0) {
2418 		if (is_attr)
2419 			VN_RELE(ZTOV(xzp));
2420 		return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2421 		    needed_bits, needed_bits));
2422 	}
2423 
2424 	if (error && !check_privs) {
2425 		if (is_attr)
2426 			VN_RELE(ZTOV(xzp));
2427 		return (error);
2428 	}
2429 
2430 	if (error && (flags & V_APPEND)) {
2431 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2432 	}
2433 
2434 	if (error && check_privs) {
2435 		mode_t		checkmode = 0;
2436 		vnode_t *check_vp = ZTOV(check_zp);
2437 
2438 		/*
2439 		 * First check for implicit owner permission on
2440 		 * read_acl/read_attributes
2441 		 */
2442 
2443 		error = 0;
2444 		ASSERT(working_mode != 0);
2445 
2446 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2447 		    owner == crgetuid(cr)))
2448 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2449 
2450 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2451 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2452 			checkmode |= VREAD;
2453 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2454 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2455 			checkmode |= VWRITE;
2456 		if (working_mode & ACE_EXECUTE)
2457 			checkmode |= VEXEC;
2458 
2459 		error = secpolicy_vnode_access2(cr, check_vp, owner,
2460 		    needed_bits & ~checkmode, needed_bits);
2461 
2462 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2463 			error = secpolicy_vnode_chown(check_vp, cr, owner);
2464 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2465 			error = secpolicy_vnode_setdac(check_vp, cr, owner);
2466 
2467 		if (error == 0 && (working_mode &
2468 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2469 			error = secpolicy_vnode_remove(check_vp, cr);
2470 
2471 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2472 			error = secpolicy_vnode_chown(check_vp, cr, owner);
2473 		}
2474 		if (error == 0) {
2475 			/*
2476 			 * See if any bits other than those already checked
2477 			 * for are still present.  If so then return EACCES
2478 			 */
2479 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2480 				error = SET_ERROR(EACCES);
2481 			}
2482 		}
2483 	} else if (error == 0) {
2484 		error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2485 		    needed_bits, needed_bits);
2486 	}
2487 
2488 
2489 	if (is_attr)
2490 		VN_RELE(ZTOV(xzp));
2491 
2492 	return (error);
2493 }
2494 
2495 /*
2496  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2497  * NFSv4-style ZFS ACL format and call zfs_zaccess()
2498  */
2499 int
2500 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2501 {
2502 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2503 }
2504 
2505 /*
2506  * Access function for secpolicy_vnode_setattr
2507  */
2508 int
2509 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2510 {
2511 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2512 
2513 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2514 }
2515 
2516 static int
2517 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2518     mode_t available_perms, cred_t *cr)
2519 {
2520 	int error;
2521 	uid_t downer;
2522 
2523 	downer = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr, ZFS_OWNER);
2524 
2525 	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2526 	    downer, available_perms, VWRITE|VEXEC);
2527 
2528 	if (error == 0)
2529 		error = zfs_sticky_remove_access(dzp, zp, cr);
2530 
2531 	return (error);
2532 }
2533 
2534 /*
2535  * Determine whether Access should be granted/deny, without
2536  * consulting least priv subsystem.
2537  *
2538  * The following chart is the recommended NFSv4 enforcement for
2539  * ability to delete an object.
2540  *
2541  *      -------------------------------------------------------
2542  *      |   Parent Dir  |           Target Object Permissions |
2543  *      |  permissions  |                                     |
2544  *      -------------------------------------------------------
2545  *      |               | ACL Allows | ACL Denies| Delete     |
2546  *      |               |  Delete    |  Delete   | unspecified|
2547  *      -------------------------------------------------------
2548  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2549  *      |  DELETE_CHILD |                                     |
2550  *      -------------------------------------------------------
2551  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2552  *      |  DELETE_CHILD |            |           |            |
2553  *      -------------------------------------------------------
2554  *      | ACL specifies |            |           |            |
2555  *      | only allow    | Permit     | Permit    | Permit     |
2556  *      | write and     |            |           |            |
2557  *      | execute       |            |           |            |
2558  *      -------------------------------------------------------
2559  *      | ACL denies    |            |           |            |
2560  *      | write and     | Permit     | Deny      | Deny       |
2561  *      | execute       |            |           |            |
2562  *      -------------------------------------------------------
2563  *         ^
2564  *         |
2565  *         No search privilege, can't even look up file?
2566  *
2567  */
2568 int
2569 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2570 {
2571 	uint32_t dzp_working_mode = 0;
2572 	uint32_t zp_working_mode = 0;
2573 	int dzp_error, zp_error;
2574 	mode_t available_perms;
2575 	boolean_t dzpcheck_privs = B_TRUE;
2576 	boolean_t zpcheck_privs = B_TRUE;
2577 
2578 	/*
2579 	 * We want specific DELETE permissions to
2580 	 * take precedence over WRITE/EXECUTE.  We don't
2581 	 * want an ACL such as this to mess us up.
2582 	 * user:joe:write_data:deny,user:joe:delete:allow
2583 	 *
2584 	 * However, deny permissions may ultimately be overridden
2585 	 * by secpolicy_vnode_access().
2586 	 *
2587 	 * We will ask for all of the necessary permissions and then
2588 	 * look at the working modes from the directory and target object
2589 	 * to determine what was found.
2590 	 */
2591 
2592 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2593 		return (SET_ERROR(EPERM));
2594 
2595 	/*
2596 	 * First row
2597 	 * If the directory permissions allow the delete, we are done.
2598 	 */
2599 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2600 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2601 		return (0);
2602 
2603 	/*
2604 	 * If target object has delete permission then we are done
2605 	 */
2606 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2607 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2608 		return (0);
2609 
2610 	ASSERT(dzp_error && zp_error);
2611 
2612 	if (!dzpcheck_privs)
2613 		return (dzp_error);
2614 	if (!zpcheck_privs)
2615 		return (zp_error);
2616 
2617 	/*
2618 	 * Second row
2619 	 *
2620 	 * If directory returns EACCES then delete_child was denied
2621 	 * due to deny delete_child.  In this case send the request through
2622 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2623 	 * since that *could* allow the delete based on write/execute permission
2624 	 * and we want delete permissions to override write/execute.
2625 	 */
2626 
2627 	if (dzp_error == EACCES) {
2628 		/* XXXPJD: s/dzp/zp/ ? */
2629 		return (secpolicy_vnode_remove(ZTOV(dzp), cr));
2630 	}
2631 	/*
2632 	 * Third Row
2633 	 * only need to see if we have write/execute on directory.
2634 	 */
2635 
2636 	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2637 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2638 
2639 	if (dzp_error != 0 && !dzpcheck_privs)
2640 		return (dzp_error);
2641 
2642 	/*
2643 	 * Fourth row
2644 	 */
2645 
2646 	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2647 	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2648 
2649 	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2650 
2651 }
2652 
2653 int
2654 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2655     znode_t *tzp, cred_t *cr)
2656 {
2657 	int add_perm;
2658 	int error;
2659 
2660 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2661 		return (SET_ERROR(EACCES));
2662 
2663 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2664 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2665 
2666 	/*
2667 	 * Rename permissions are combination of delete permission +
2668 	 * add file/subdir permission.
2669 	 *
2670 	 * BSD operating systems also require write permission
2671 	 * on the directory being moved from one parent directory
2672 	 * to another.
2673 	 */
2674 	if (ZTOV(szp)->v_type == VDIR && ZTOV(sdzp) != ZTOV(tdzp)) {
2675 		if ((error = zfs_zaccess(szp, ACE_WRITE_DATA, 0, B_FALSE, cr)))
2676 			return (error);
2677 	}
2678 
2679 	/*
2680 	 * first make sure we do the delete portion.
2681 	 *
2682 	 * If that succeeds then check for add_file/add_subdir permissions
2683 	 */
2684 
2685 	if ((error = zfs_zaccess_delete(sdzp, szp, cr)))
2686 		return (error);
2687 
2688 	/*
2689 	 * If we have a tzp, see if we can delete it?
2690 	 */
2691 	if (tzp && (error = zfs_zaccess_delete(tdzp, tzp, cr)))
2692 		return (error);
2693 
2694 	/*
2695 	 * Now check for add permissions
2696 	 */
2697 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2698 
2699 	return (error);
2700 }
2701