xref: /titanic_50/usr/src/uts/common/fs/zfs/zfs_acl.c (revision cb8a054b1ab30d5caa746e6c44f29d4c9d3071c1)
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  */
24 
25 #include <sys/types.h>
26 #include <sys/param.h>
27 #include <sys/time.h>
28 #include <sys/systm.h>
29 #include <sys/sysmacros.h>
30 #include <sys/resource.h>
31 #include <sys/vfs.h>
32 #include <sys/vnode.h>
33 #include <sys/sid.h>
34 #include <sys/file.h>
35 #include <sys/stat.h>
36 #include <sys/kmem.h>
37 #include <sys/cmn_err.h>
38 #include <sys/errno.h>
39 #include <sys/unistd.h>
40 #include <sys/sdt.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/mode.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_vfsops.h>
49 #include <sys/dmu.h>
50 #include <sys/dnode.h>
51 #include <sys/zap.h>
52 #include <sys/sa.h>
53 #include "fs/fs_subr.h"
54 #include <acl/acl_common.h>
55 
56 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
57 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
58 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
59 #define	MIN_ACE_TYPE	ALLOW
60 
61 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
62 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
63     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
64 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
65     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
66 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
67     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
68 
69 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
70     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
71     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
72     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
73 
74 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
75 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
76     ACE_DELETE|ACE_DELETE_CHILD)
77 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
78 
79 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
80     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
81 
82 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
83     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
84 
85 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
86     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
87 
88 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
89 
90 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
91     ZFS_ACL_PROTECTED)
92 
93 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
94     ZFS_ACL_OBJ_ACE)
95 
96 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
97 
98 static uint16_t
99 zfs_ace_v0_get_type(void *acep)
100 {
101 	return (((zfs_oldace_t *)acep)->z_type);
102 }
103 
104 static uint16_t
105 zfs_ace_v0_get_flags(void *acep)
106 {
107 	return (((zfs_oldace_t *)acep)->z_flags);
108 }
109 
110 static uint32_t
111 zfs_ace_v0_get_mask(void *acep)
112 {
113 	return (((zfs_oldace_t *)acep)->z_access_mask);
114 }
115 
116 static uint64_t
117 zfs_ace_v0_get_who(void *acep)
118 {
119 	return (((zfs_oldace_t *)acep)->z_fuid);
120 }
121 
122 static void
123 zfs_ace_v0_set_type(void *acep, uint16_t type)
124 {
125 	((zfs_oldace_t *)acep)->z_type = type;
126 }
127 
128 static void
129 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
130 {
131 	((zfs_oldace_t *)acep)->z_flags = flags;
132 }
133 
134 static void
135 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
136 {
137 	((zfs_oldace_t *)acep)->z_access_mask = mask;
138 }
139 
140 static void
141 zfs_ace_v0_set_who(void *acep, uint64_t who)
142 {
143 	((zfs_oldace_t *)acep)->z_fuid = who;
144 }
145 
146 /*ARGSUSED*/
147 static size_t
148 zfs_ace_v0_size(void *acep)
149 {
150 	return (sizeof (zfs_oldace_t));
151 }
152 
153 static size_t
154 zfs_ace_v0_abstract_size(void)
155 {
156 	return (sizeof (zfs_oldace_t));
157 }
158 
159 static int
160 zfs_ace_v0_mask_off(void)
161 {
162 	return (offsetof(zfs_oldace_t, z_access_mask));
163 }
164 
165 /*ARGSUSED*/
166 static int
167 zfs_ace_v0_data(void *acep, void **datap)
168 {
169 	*datap = NULL;
170 	return (0);
171 }
172 
173 static acl_ops_t zfs_acl_v0_ops = {
174 	zfs_ace_v0_get_mask,
175 	zfs_ace_v0_set_mask,
176 	zfs_ace_v0_get_flags,
177 	zfs_ace_v0_set_flags,
178 	zfs_ace_v0_get_type,
179 	zfs_ace_v0_set_type,
180 	zfs_ace_v0_get_who,
181 	zfs_ace_v0_set_who,
182 	zfs_ace_v0_size,
183 	zfs_ace_v0_abstract_size,
184 	zfs_ace_v0_mask_off,
185 	zfs_ace_v0_data
186 };
187 
188 static uint16_t
189 zfs_ace_fuid_get_type(void *acep)
190 {
191 	return (((zfs_ace_hdr_t *)acep)->z_type);
192 }
193 
194 static uint16_t
195 zfs_ace_fuid_get_flags(void *acep)
196 {
197 	return (((zfs_ace_hdr_t *)acep)->z_flags);
198 }
199 
200 static uint32_t
201 zfs_ace_fuid_get_mask(void *acep)
202 {
203 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
204 }
205 
206 static uint64_t
207 zfs_ace_fuid_get_who(void *args)
208 {
209 	uint16_t entry_type;
210 	zfs_ace_t *acep = args;
211 
212 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
213 
214 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
215 	    entry_type == ACE_EVERYONE)
216 		return (-1);
217 	return (((zfs_ace_t *)acep)->z_fuid);
218 }
219 
220 static void
221 zfs_ace_fuid_set_type(void *acep, uint16_t type)
222 {
223 	((zfs_ace_hdr_t *)acep)->z_type = type;
224 }
225 
226 static void
227 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
228 {
229 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
230 }
231 
232 static void
233 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
234 {
235 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
236 }
237 
238 static void
239 zfs_ace_fuid_set_who(void *arg, uint64_t who)
240 {
241 	zfs_ace_t *acep = arg;
242 
243 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
244 
245 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
246 	    entry_type == ACE_EVERYONE)
247 		return;
248 	acep->z_fuid = who;
249 }
250 
251 static size_t
252 zfs_ace_fuid_size(void *acep)
253 {
254 	zfs_ace_hdr_t *zacep = acep;
255 	uint16_t entry_type;
256 
257 	switch (zacep->z_type) {
258 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
259 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
260 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
261 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
262 		return (sizeof (zfs_object_ace_t));
263 	case ALLOW:
264 	case DENY:
265 		entry_type =
266 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
267 		if (entry_type == ACE_OWNER ||
268 		    entry_type == OWNING_GROUP ||
269 		    entry_type == ACE_EVERYONE)
270 			return (sizeof (zfs_ace_hdr_t));
271 		/*FALLTHROUGH*/
272 	default:
273 		return (sizeof (zfs_ace_t));
274 	}
275 }
276 
277 static size_t
278 zfs_ace_fuid_abstract_size(void)
279 {
280 	return (sizeof (zfs_ace_hdr_t));
281 }
282 
283 static int
284 zfs_ace_fuid_mask_off(void)
285 {
286 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
287 }
288 
289 static int
290 zfs_ace_fuid_data(void *acep, void **datap)
291 {
292 	zfs_ace_t *zacep = acep;
293 	zfs_object_ace_t *zobjp;
294 
295 	switch (zacep->z_hdr.z_type) {
296 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
297 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
298 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
299 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
300 		zobjp = acep;
301 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
302 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
303 	default:
304 		*datap = NULL;
305 		return (0);
306 	}
307 }
308 
309 static acl_ops_t zfs_acl_fuid_ops = {
310 	zfs_ace_fuid_get_mask,
311 	zfs_ace_fuid_set_mask,
312 	zfs_ace_fuid_get_flags,
313 	zfs_ace_fuid_set_flags,
314 	zfs_ace_fuid_get_type,
315 	zfs_ace_fuid_set_type,
316 	zfs_ace_fuid_get_who,
317 	zfs_ace_fuid_set_who,
318 	zfs_ace_fuid_size,
319 	zfs_ace_fuid_abstract_size,
320 	zfs_ace_fuid_mask_off,
321 	zfs_ace_fuid_data
322 };
323 
324 /*
325  * The following three functions are provided for compatibility with
326  * older ZPL version in order to determine if the file use to have
327  * an external ACL and what version of ACL previously existed on the
328  * file.  Would really be nice to not need this, sigh.
329  */
330 
331 uint64_t
332 zfs_external_acl(znode_t *zp)
333 {
334 	zfs_acl_phys_t acl_phys;
335 
336 	if (zp->z_is_sa)
337 		return (0);
338 
339 	VERIFY(0 == sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
340 	    &acl_phys, sizeof (acl_phys)));
341 
342 	return (acl_phys.z_acl_extern_obj);
343 }
344 
345 /*
346  * Determine size of ACL in bytes
347  *
348  * This is more complicated than it should be since we have to deal
349  * with old external ACLs.
350  */
351 static int
352 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
353     zfs_acl_phys_t *aclphys)
354 {
355 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
356 	uint64_t acl_count;
357 	int size;
358 	int error;
359 
360 	if (zp->z_is_sa) {
361 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
362 		    &size)) != 0)
363 			return (error);
364 		*aclsize = size;
365 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
366 		    &acl_count, sizeof (acl_count))) != 0)
367 			return (error);
368 		*aclcount = acl_count;
369 	} else {
370 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
371 		    aclphys, sizeof (*aclphys))) != 0)
372 			return (error);
373 
374 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
375 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
376 			*aclcount = aclphys->z_acl_size;
377 		} else {
378 			*aclsize = aclphys->z_acl_size;
379 			*aclcount = aclphys->z_acl_count;
380 		}
381 	}
382 	return (0);
383 }
384 
385 int
386 zfs_znode_acl_version(znode_t *zp)
387 {
388 	zfs_acl_phys_t acl_phys;
389 
390 	if (zp->z_is_sa) {
391 		return (ZFS_ACL_VERSION_FUID);
392 	} else {
393 		VERIFY(0 == sa_lookup(zp->z_sa_hdl,
394 		    SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
395 		    &acl_phys, sizeof (acl_phys)));
396 		return (acl_phys.z_acl_version);
397 	}
398 }
399 
400 static int
401 zfs_acl_version(int version)
402 {
403 	if (version < ZPL_VERSION_FUID)
404 		return (ZFS_ACL_VERSION_INITIAL);
405 	else
406 		return (ZFS_ACL_VERSION_FUID);
407 }
408 
409 static int
410 zfs_acl_version_zp(znode_t *zp)
411 {
412 	return (zfs_acl_version(zp->z_zfsvfs->z_version));
413 }
414 
415 zfs_acl_t *
416 zfs_acl_alloc(int vers)
417 {
418 	zfs_acl_t *aclp;
419 
420 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
421 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
422 	    offsetof(zfs_acl_node_t, z_next));
423 	aclp->z_version = vers;
424 	if (vers == ZFS_ACL_VERSION_FUID)
425 		aclp->z_ops = zfs_acl_fuid_ops;
426 	else
427 		aclp->z_ops = zfs_acl_v0_ops;
428 	return (aclp);
429 }
430 
431 zfs_acl_node_t *
432 zfs_acl_node_alloc(size_t bytes)
433 {
434 	zfs_acl_node_t *aclnode;
435 
436 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
437 	if (bytes) {
438 		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
439 		aclnode->z_allocdata = aclnode->z_acldata;
440 		aclnode->z_allocsize = bytes;
441 		aclnode->z_size = bytes;
442 	}
443 
444 	return (aclnode);
445 }
446 
447 static void
448 zfs_acl_node_free(zfs_acl_node_t *aclnode)
449 {
450 	if (aclnode->z_allocsize)
451 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
452 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
453 }
454 
455 static void
456 zfs_acl_release_nodes(zfs_acl_t *aclp)
457 {
458 	zfs_acl_node_t *aclnode;
459 
460 	while (aclnode = list_head(&aclp->z_acl)) {
461 		list_remove(&aclp->z_acl, aclnode);
462 		zfs_acl_node_free(aclnode);
463 	}
464 	aclp->z_acl_count = 0;
465 	aclp->z_acl_bytes = 0;
466 }
467 
468 void
469 zfs_acl_free(zfs_acl_t *aclp)
470 {
471 	zfs_acl_release_nodes(aclp);
472 	list_destroy(&aclp->z_acl);
473 	kmem_free(aclp, sizeof (zfs_acl_t));
474 }
475 
476 static boolean_t
477 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
478 {
479 	uint16_t entry_type;
480 
481 	switch (type) {
482 	case ALLOW:
483 	case DENY:
484 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
485 	case ACE_SYSTEM_ALARM_ACE_TYPE:
486 		entry_type = flags & ACE_TYPE_FLAGS;
487 		return (entry_type == ACE_OWNER ||
488 		    entry_type == OWNING_GROUP ||
489 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
490 		    entry_type == ACE_IDENTIFIER_GROUP);
491 	default:
492 		if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
493 			return (B_TRUE);
494 	}
495 	return (B_FALSE);
496 }
497 
498 static boolean_t
499 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
500 {
501 	/*
502 	 * first check type of entry
503 	 */
504 
505 	if (!zfs_acl_valid_ace_type(type, iflags))
506 		return (B_FALSE);
507 
508 	switch (type) {
509 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
510 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
511 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
512 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
513 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
514 			return (B_FALSE);
515 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
516 	}
517 
518 	/*
519 	 * next check inheritance level flags
520 	 */
521 
522 	if (obj_type == VDIR &&
523 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
524 		aclp->z_hints |= ZFS_INHERIT_ACE;
525 
526 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
527 		if ((iflags & (ACE_FILE_INHERIT_ACE|
528 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
529 			return (B_FALSE);
530 		}
531 	}
532 
533 	return (B_TRUE);
534 }
535 
536 static void *
537 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
538     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
539 {
540 	zfs_acl_node_t *aclnode;
541 
542 	ASSERT(aclp);
543 
544 	if (start == NULL) {
545 		aclnode = list_head(&aclp->z_acl);
546 		if (aclnode == NULL)
547 			return (NULL);
548 
549 		aclp->z_next_ace = aclnode->z_acldata;
550 		aclp->z_curr_node = aclnode;
551 		aclnode->z_ace_idx = 0;
552 	}
553 
554 	aclnode = aclp->z_curr_node;
555 
556 	if (aclnode == NULL)
557 		return (NULL);
558 
559 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
560 		aclnode = list_next(&aclp->z_acl, aclnode);
561 		if (aclnode == NULL)
562 			return (NULL);
563 		else {
564 			aclp->z_curr_node = aclnode;
565 			aclnode->z_ace_idx = 0;
566 			aclp->z_next_ace = aclnode->z_acldata;
567 		}
568 	}
569 
570 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
571 		void *acep = aclp->z_next_ace;
572 		size_t ace_size;
573 
574 		/*
575 		 * Make sure we don't overstep our bounds
576 		 */
577 		ace_size = aclp->z_ops.ace_size(acep);
578 
579 		if (((caddr_t)acep + ace_size) >
580 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
581 			return (NULL);
582 		}
583 
584 		*iflags = aclp->z_ops.ace_flags_get(acep);
585 		*type = aclp->z_ops.ace_type_get(acep);
586 		*access_mask = aclp->z_ops.ace_mask_get(acep);
587 		*who = aclp->z_ops.ace_who_get(acep);
588 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
589 		aclnode->z_ace_idx++;
590 
591 		return ((void *)acep);
592 	}
593 	return (NULL);
594 }
595 
596 /*ARGSUSED*/
597 static uint64_t
598 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
599     uint16_t *flags, uint16_t *type, uint32_t *mask)
600 {
601 	zfs_acl_t *aclp = datap;
602 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
603 	uint64_t who;
604 
605 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
606 	    flags, type);
607 	return ((uint64_t)(uintptr_t)acep);
608 }
609 
610 static zfs_acl_node_t *
611 zfs_acl_curr_node(zfs_acl_t *aclp)
612 {
613 	ASSERT(aclp->z_curr_node);
614 	return (aclp->z_curr_node);
615 }
616 
617 /*
618  * Copy ACE to internal ZFS format.
619  * While processing the ACL each ACE will be validated for correctness.
620  * ACE FUIDs will be created later.
621  */
622 int
623 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
624     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
625     zfs_fuid_info_t **fuidp, cred_t *cr)
626 {
627 	int i;
628 	uint16_t entry_type;
629 	zfs_ace_t *aceptr = z_acl;
630 	ace_t *acep = datap;
631 	zfs_object_ace_t *zobjacep;
632 	ace_object_t *aceobjp;
633 
634 	for (i = 0; i != aclcnt; i++) {
635 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
636 		aceptr->z_hdr.z_flags = acep->a_flags;
637 		aceptr->z_hdr.z_type = acep->a_type;
638 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
639 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
640 		    entry_type != ACE_EVERYONE) {
641 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
642 			    cr, (entry_type == 0) ?
643 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
644 		}
645 
646 		/*
647 		 * Make sure ACE is valid
648 		 */
649 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
650 		    aceptr->z_hdr.z_flags) != B_TRUE)
651 			return (EINVAL);
652 
653 		switch (acep->a_type) {
654 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
655 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
656 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
657 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
658 			zobjacep = (zfs_object_ace_t *)aceptr;
659 			aceobjp = (ace_object_t *)acep;
660 
661 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
662 			    sizeof (aceobjp->a_obj_type));
663 			bcopy(aceobjp->a_inherit_obj_type,
664 			    zobjacep->z_inherit_type,
665 			    sizeof (aceobjp->a_inherit_obj_type));
666 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
667 			break;
668 		default:
669 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
670 		}
671 
672 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
673 		    aclp->z_ops.ace_size(aceptr));
674 	}
675 
676 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
677 
678 	return (0);
679 }
680 
681 /*
682  * Copy ZFS ACEs to fixed size ace_t layout
683  */
684 static void
685 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
686     void *datap, int filter)
687 {
688 	uint64_t who;
689 	uint32_t access_mask;
690 	uint16_t iflags, type;
691 	zfs_ace_hdr_t *zacep = NULL;
692 	ace_t *acep = datap;
693 	ace_object_t *objacep;
694 	zfs_object_ace_t *zobjacep;
695 	size_t ace_size;
696 	uint16_t entry_type;
697 
698 	while (zacep = zfs_acl_next_ace(aclp, zacep,
699 	    &who, &access_mask, &iflags, &type)) {
700 
701 		switch (type) {
702 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
703 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
704 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
705 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
706 			if (filter) {
707 				continue;
708 			}
709 			zobjacep = (zfs_object_ace_t *)zacep;
710 			objacep = (ace_object_t *)acep;
711 			bcopy(zobjacep->z_object_type,
712 			    objacep->a_obj_type,
713 			    sizeof (zobjacep->z_object_type));
714 			bcopy(zobjacep->z_inherit_type,
715 			    objacep->a_inherit_obj_type,
716 			    sizeof (zobjacep->z_inherit_type));
717 			ace_size = sizeof (ace_object_t);
718 			break;
719 		default:
720 			ace_size = sizeof (ace_t);
721 			break;
722 		}
723 
724 		entry_type = (iflags & ACE_TYPE_FLAGS);
725 		if ((entry_type != ACE_OWNER &&
726 		    entry_type != OWNING_GROUP &&
727 		    entry_type != ACE_EVERYONE)) {
728 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
729 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
730 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
731 		} else {
732 			acep->a_who = (uid_t)(int64_t)who;
733 		}
734 		acep->a_access_mask = access_mask;
735 		acep->a_flags = iflags;
736 		acep->a_type = type;
737 		acep = (ace_t *)((caddr_t)acep + ace_size);
738 	}
739 }
740 
741 static int
742 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
743     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
744 {
745 	int i;
746 	zfs_oldace_t *aceptr = z_acl;
747 
748 	for (i = 0; i != aclcnt; i++, aceptr++) {
749 		aceptr->z_access_mask = acep[i].a_access_mask;
750 		aceptr->z_type = acep[i].a_type;
751 		aceptr->z_flags = acep[i].a_flags;
752 		aceptr->z_fuid = acep[i].a_who;
753 		/*
754 		 * Make sure ACE is valid
755 		 */
756 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
757 		    aceptr->z_flags) != B_TRUE)
758 			return (EINVAL);
759 	}
760 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
761 	return (0);
762 }
763 
764 /*
765  * convert old ACL format to new
766  */
767 void
768 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
769 {
770 	zfs_oldace_t *oldaclp;
771 	int i;
772 	uint16_t type, iflags;
773 	uint32_t access_mask;
774 	uint64_t who;
775 	void *cookie = NULL;
776 	zfs_acl_node_t *newaclnode;
777 
778 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
779 	/*
780 	 * First create the ACE in a contiguous piece of memory
781 	 * for zfs_copy_ace_2_fuid().
782 	 *
783 	 * We only convert an ACL once, so this won't happen
784 	 * everytime.
785 	 */
786 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
787 	    KM_SLEEP);
788 	i = 0;
789 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
790 	    &access_mask, &iflags, &type)) {
791 		oldaclp[i].z_flags = iflags;
792 		oldaclp[i].z_type = type;
793 		oldaclp[i].z_fuid = who;
794 		oldaclp[i++].z_access_mask = access_mask;
795 	}
796 
797 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
798 	    sizeof (zfs_object_ace_t));
799 	aclp->z_ops = zfs_acl_fuid_ops;
800 	VERIFY(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
801 	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
802 	    &newaclnode->z_size, NULL, cr) == 0);
803 	newaclnode->z_ace_count = aclp->z_acl_count;
804 	aclp->z_version = ZFS_ACL_VERSION;
805 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
806 
807 	/*
808 	 * Release all previous ACL nodes
809 	 */
810 
811 	zfs_acl_release_nodes(aclp);
812 
813 	list_insert_head(&aclp->z_acl, newaclnode);
814 
815 	aclp->z_acl_bytes = newaclnode->z_size;
816 	aclp->z_acl_count = newaclnode->z_ace_count;
817 
818 }
819 
820 /*
821  * Convert unix access mask to v4 access mask
822  */
823 static uint32_t
824 zfs_unix_to_v4(uint32_t access_mask)
825 {
826 	uint32_t new_mask = 0;
827 
828 	if (access_mask & S_IXOTH)
829 		new_mask |= ACE_EXECUTE;
830 	if (access_mask & S_IWOTH)
831 		new_mask |= ACE_WRITE_DATA;
832 	if (access_mask & S_IROTH)
833 		new_mask |= ACE_READ_DATA;
834 	return (new_mask);
835 }
836 
837 static void
838 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
839     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
840 {
841 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
842 
843 	aclp->z_ops.ace_mask_set(acep, access_mask);
844 	aclp->z_ops.ace_type_set(acep, access_type);
845 	aclp->z_ops.ace_flags_set(acep, entry_type);
846 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
847 	    type != ACE_EVERYONE))
848 		aclp->z_ops.ace_who_set(acep, fuid);
849 }
850 
851 /*
852  * Determine mode of file based on ACL.
853  * Also, create FUIDs for any User/Group ACEs
854  */
855 uint64_t
856 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
857     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
858 {
859 	int		entry_type;
860 	mode_t		mode;
861 	mode_t		seen = 0;
862 	zfs_ace_hdr_t 	*acep = NULL;
863 	uint64_t	who;
864 	uint16_t	iflags, type;
865 	uint32_t	access_mask;
866 	boolean_t	an_exec_denied = B_FALSE;
867 
868 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
869 
870 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
871 	    &access_mask, &iflags, &type)) {
872 
873 		if (!zfs_acl_valid_ace_type(type, iflags))
874 			continue;
875 
876 		entry_type = (iflags & ACE_TYPE_FLAGS);
877 
878 		/*
879 		 * Skip over owner@, group@ or everyone@ inherit only ACEs
880 		 */
881 		if ((iflags & ACE_INHERIT_ONLY_ACE) &&
882 		    (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
883 		    entry_type == OWNING_GROUP))
884 			continue;
885 
886 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
887 		    who == fuid)) {
888 			if ((access_mask & ACE_READ_DATA) &&
889 			    (!(seen & S_IRUSR))) {
890 				seen |= S_IRUSR;
891 				if (type == ALLOW) {
892 					mode |= S_IRUSR;
893 				}
894 			}
895 			if ((access_mask & ACE_WRITE_DATA) &&
896 			    (!(seen & S_IWUSR))) {
897 				seen |= S_IWUSR;
898 				if (type == ALLOW) {
899 					mode |= S_IWUSR;
900 				}
901 			}
902 			if ((access_mask & ACE_EXECUTE) &&
903 			    (!(seen & S_IXUSR))) {
904 				seen |= S_IXUSR;
905 				if (type == ALLOW) {
906 					mode |= S_IXUSR;
907 				}
908 			}
909 		} else if (entry_type == OWNING_GROUP ||
910 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
911 			if ((access_mask & ACE_READ_DATA) &&
912 			    (!(seen & S_IRGRP))) {
913 				seen |= S_IRGRP;
914 				if (type == ALLOW) {
915 					mode |= S_IRGRP;
916 				}
917 			}
918 			if ((access_mask & ACE_WRITE_DATA) &&
919 			    (!(seen & S_IWGRP))) {
920 				seen |= S_IWGRP;
921 				if (type == ALLOW) {
922 					mode |= S_IWGRP;
923 				}
924 			}
925 			if ((access_mask & ACE_EXECUTE) &&
926 			    (!(seen & S_IXGRP))) {
927 				seen |= S_IXGRP;
928 				if (type == ALLOW) {
929 					mode |= S_IXGRP;
930 				}
931 			}
932 		} else if (entry_type == ACE_EVERYONE) {
933 			if ((access_mask & ACE_READ_DATA)) {
934 				if (!(seen & S_IRUSR)) {
935 					seen |= S_IRUSR;
936 					if (type == ALLOW) {
937 						mode |= S_IRUSR;
938 					}
939 				}
940 				if (!(seen & S_IRGRP)) {
941 					seen |= S_IRGRP;
942 					if (type == ALLOW) {
943 						mode |= S_IRGRP;
944 					}
945 				}
946 				if (!(seen & S_IROTH)) {
947 					seen |= S_IROTH;
948 					if (type == ALLOW) {
949 						mode |= S_IROTH;
950 					}
951 				}
952 			}
953 			if ((access_mask & ACE_WRITE_DATA)) {
954 				if (!(seen & S_IWUSR)) {
955 					seen |= S_IWUSR;
956 					if (type == ALLOW) {
957 						mode |= S_IWUSR;
958 					}
959 				}
960 				if (!(seen & S_IWGRP)) {
961 					seen |= S_IWGRP;
962 					if (type == ALLOW) {
963 						mode |= S_IWGRP;
964 					}
965 				}
966 				if (!(seen & S_IWOTH)) {
967 					seen |= S_IWOTH;
968 					if (type == ALLOW) {
969 						mode |= S_IWOTH;
970 					}
971 				}
972 			}
973 			if ((access_mask & ACE_EXECUTE)) {
974 				if (!(seen & S_IXUSR)) {
975 					seen |= S_IXUSR;
976 					if (type == ALLOW) {
977 						mode |= S_IXUSR;
978 					}
979 				}
980 				if (!(seen & S_IXGRP)) {
981 					seen |= S_IXGRP;
982 					if (type == ALLOW) {
983 						mode |= S_IXGRP;
984 					}
985 				}
986 				if (!(seen & S_IXOTH)) {
987 					seen |= S_IXOTH;
988 					if (type == ALLOW) {
989 						mode |= S_IXOTH;
990 					}
991 				}
992 			}
993 		} else {
994 			/*
995 			 * Only care if this IDENTIFIER_GROUP or
996 			 * USER ACE denies execute access to someone,
997 			 * mode is not affected
998 			 */
999 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1000 				an_exec_denied = B_TRUE;
1001 		}
1002 	}
1003 
1004 	/*
1005 	 * Failure to allow is effectively a deny, so execute permission
1006 	 * is denied if it was never mentioned or if we explicitly
1007 	 * weren't allowed it.
1008 	 */
1009 	if (!an_exec_denied &&
1010 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1011 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1012 		an_exec_denied = B_TRUE;
1013 
1014 	if (an_exec_denied)
1015 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1016 	else
1017 		*pflags |= ZFS_NO_EXECS_DENIED;
1018 
1019 	return (mode);
1020 }
1021 
1022 /*
1023  * Read an external acl object.  If the intent is to modify, always
1024  * create a new acl and leave any cached acl in place.
1025  */
1026 static int
1027 zfs_acl_node_read(znode_t *zp, zfs_acl_t **aclpp, boolean_t will_modify)
1028 {
1029 	zfs_acl_t	*aclp;
1030 	int		aclsize;
1031 	int		acl_count;
1032 	zfs_acl_node_t	*aclnode;
1033 	zfs_acl_phys_t	znode_acl;
1034 	int		version;
1035 	int		error;
1036 
1037 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1038 
1039 	if (zp->z_acl_cached && !will_modify) {
1040 		*aclpp = zp->z_acl_cached;
1041 		return (0);
1042 	}
1043 
1044 	version = ZNODE_ACL_VERSION(zp);
1045 
1046 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1047 	    &acl_count, &znode_acl)) != 0)
1048 		return (error);
1049 
1050 	aclp = zfs_acl_alloc(version);
1051 
1052 	aclp->z_acl_count = acl_count;
1053 	aclp->z_acl_bytes = aclsize;
1054 
1055 	aclnode = zfs_acl_node_alloc(aclsize);
1056 	aclnode->z_ace_count = aclp->z_acl_count;
1057 	aclnode->z_size = aclsize;
1058 
1059 	if (!zp->z_is_sa) {
1060 		if (znode_acl.z_acl_extern_obj) {
1061 			error = dmu_read(zp->z_zfsvfs->z_os,
1062 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1063 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1064 		} else {
1065 			bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1066 			    aclnode->z_size);
1067 		}
1068 	} else {
1069 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1070 		    aclnode->z_acldata, aclnode->z_size);
1071 	}
1072 
1073 	if (error != 0) {
1074 		zfs_acl_free(aclp);
1075 		zfs_acl_node_free(aclnode);
1076 		/* convert checksum errors into IO errors */
1077 		if (error == ECKSUM)
1078 			error = EIO;
1079 		return (error);
1080 	}
1081 
1082 	list_insert_head(&aclp->z_acl, aclnode);
1083 
1084 	*aclpp = aclp;
1085 	if (!will_modify)
1086 		zp->z_acl_cached = aclp;
1087 	return (0);
1088 }
1089 
1090 /*ARGSUSED*/
1091 void
1092 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1093     boolean_t start, void *userdata)
1094 {
1095 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1096 
1097 	if (start) {
1098 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1099 	} else {
1100 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1101 		    cb->cb_acl_node);
1102 	}
1103 	*dataptr = cb->cb_acl_node->z_acldata;
1104 	*length = cb->cb_acl_node->z_size;
1105 }
1106 
1107 
1108 static int
1109 zfs_acl_get_owner_fuids(znode_t *zp, uint64_t *fuid, uint64_t *fgid)
1110 {
1111 	int count = 0;
1112 	sa_bulk_attr_t	bulk[2];
1113 	int error;
1114 
1115 	if (IS_EPHEMERAL(zp->z_uid) || IS_EPHEMERAL(zp->z_gid)) {
1116 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zp->z_zfsvfs), NULL,
1117 		    &fuid, sizeof (fuid));
1118 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zp->z_zfsvfs), NULL,
1119 		    &fgid, sizeof (fuid));
1120 		if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) {
1121 			return (error);
1122 		}
1123 	} else {
1124 		*fuid = zp->z_uid;
1125 		*fgid = zp->z_gid;
1126 	}
1127 	return (0);
1128 }
1129 
1130 int
1131 zfs_acl_chown_setattr(znode_t *zp)
1132 {
1133 	int error;
1134 	zfs_acl_t *aclp;
1135 	uint64_t fuid, fgid;
1136 
1137 	if ((error = zfs_acl_get_owner_fuids(zp, &fuid, &fgid)) != 0)
1138 		return (error);
1139 
1140 	mutex_enter(&zp->z_acl_lock);
1141 	if ((error = zfs_acl_node_read(zp, &aclp, B_FALSE)) == 0)
1142 		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1143 		    &zp->z_pflags, fuid, fgid);
1144 	mutex_exit(&zp->z_acl_lock);
1145 	return (error);
1146 }
1147 
1148 /*
1149  * common code for setting ACLs.
1150  *
1151  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1152  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1153  * already checked the acl and knows whether to inherit.
1154  */
1155 int
1156 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1157 {
1158 	int			error;
1159 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1160 	dmu_object_type_t	otype;
1161 	zfs_acl_locator_cb_t	locate = { 0 };
1162 	uint64_t		mode;
1163 	sa_bulk_attr_t		bulk[5];
1164 	uint64_t		ctime[2];
1165 	int			count = 0;
1166 	uint64_t		fuid, fgid;
1167 
1168 	mode = zp->z_mode;
1169 
1170 	if ((error = zfs_acl_get_owner_fuids(zp, &fuid, &fgid)) != 0)
1171 		return (error);
1172 
1173 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags, fuid, fgid);
1174 
1175 	zp->z_mode = mode;
1176 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1177 	    &mode, sizeof (mode));
1178 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1179 	    &zp->z_pflags, sizeof (zp->z_pflags));
1180 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1181 	    &ctime, sizeof (ctime));
1182 
1183 	if (zp->z_acl_cached) {
1184 		zfs_acl_free(zp->z_acl_cached);
1185 		zp->z_acl_cached = NULL;
1186 	}
1187 
1188 	/*
1189 	 * Upgrade needed?
1190 	 */
1191 	if (!zfsvfs->z_use_fuids) {
1192 		otype = DMU_OT_OLDACL;
1193 	} else {
1194 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1195 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1196 			zfs_acl_xform(zp, aclp, cr);
1197 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1198 		otype = DMU_OT_ACL;
1199 	}
1200 
1201 	/*
1202 	 * Arrgh, we have to handle old on disk format
1203 	 * as well as newer (preferred) SA format.
1204 	 */
1205 
1206 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1207 		locate.cb_aclp = aclp;
1208 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1209 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1210 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1211 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1212 	} else { /* Painful legacy way */
1213 		zfs_acl_node_t *aclnode;
1214 		uint64_t off = 0;
1215 		zfs_acl_phys_t acl_phys;
1216 		uint64_t aoid;
1217 
1218 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1219 		    &acl_phys, sizeof (acl_phys))) != 0)
1220 			return (error);
1221 
1222 		aoid = acl_phys.z_acl_extern_obj;
1223 
1224 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1225 			/*
1226 			 * If ACL was previously external and we are now
1227 			 * converting to new ACL format then release old
1228 			 * ACL object and create a new one.
1229 			 */
1230 			if (aoid &&
1231 			    aclp->z_version != acl_phys.z_acl_version) {
1232 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1233 				if (error)
1234 					return (error);
1235 				aoid = 0;
1236 			}
1237 			if (aoid == 0) {
1238 				aoid = dmu_object_alloc(zfsvfs->z_os,
1239 				    otype, aclp->z_acl_bytes,
1240 				    otype == DMU_OT_ACL ?
1241 				    DMU_OT_SYSACL : DMU_OT_NONE,
1242 				    otype == DMU_OT_ACL ?
1243 				    DN_MAX_BONUSLEN : 0, tx);
1244 			} else {
1245 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1246 				    aoid, aclp->z_acl_bytes, 0, tx);
1247 			}
1248 			acl_phys.z_acl_extern_obj = aoid;
1249 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1250 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1251 				if (aclnode->z_ace_count == 0)
1252 					continue;
1253 				dmu_write(zfsvfs->z_os, aoid, off,
1254 				    aclnode->z_size, aclnode->z_acldata, tx);
1255 				off += aclnode->z_size;
1256 			}
1257 		} else {
1258 			void *start = acl_phys.z_ace_data;
1259 			/*
1260 			 * Migrating back embedded?
1261 			 */
1262 			if (acl_phys.z_acl_extern_obj) {
1263 				error = dmu_object_free(zfsvfs->z_os,
1264 				    acl_phys.z_acl_extern_obj, tx);
1265 				if (error)
1266 					return (error);
1267 				acl_phys.z_acl_extern_obj = 0;
1268 			}
1269 
1270 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1271 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1272 				if (aclnode->z_ace_count == 0)
1273 					continue;
1274 				bcopy(aclnode->z_acldata, start,
1275 				    aclnode->z_size);
1276 				start = (caddr_t)start + aclnode->z_size;
1277 			}
1278 		}
1279 		/*
1280 		 * If Old version then swap count/bytes to match old
1281 		 * layout of znode_acl_phys_t.
1282 		 */
1283 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1284 			acl_phys.z_acl_size = aclp->z_acl_count;
1285 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1286 		} else {
1287 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1288 			acl_phys.z_acl_count = aclp->z_acl_count;
1289 		}
1290 		acl_phys.z_acl_version = aclp->z_version;
1291 
1292 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1293 		    &acl_phys, sizeof (acl_phys));
1294 	}
1295 
1296 	/*
1297 	 * Replace ACL wide bits, but first clear them.
1298 	 */
1299 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1300 
1301 	zp->z_pflags |= aclp->z_hints;
1302 
1303 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1304 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1305 
1306 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime, B_TRUE);
1307 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1308 }
1309 
1310 /*
1311  * Update access mask for prepended ACE
1312  *
1313  * This applies the "groupmask" value for aclmode property.
1314  */
1315 static void
1316 zfs_acl_prepend_fixup(zfs_acl_t *aclp, void  *acep, void  *origacep,
1317     mode_t mode, uint64_t owner)
1318 {
1319 	int	rmask, wmask, xmask;
1320 	int	user_ace;
1321 	uint16_t aceflags;
1322 	uint32_t origmask, acepmask;
1323 	uint64_t fuid;
1324 
1325 	aceflags = aclp->z_ops.ace_flags_get(acep);
1326 	fuid = aclp->z_ops.ace_who_get(acep);
1327 	origmask = aclp->z_ops.ace_mask_get(origacep);
1328 	acepmask = aclp->z_ops.ace_mask_get(acep);
1329 
1330 	user_ace = (!(aceflags &
1331 	    (ACE_OWNER|ACE_GROUP|ACE_IDENTIFIER_GROUP)));
1332 
1333 	if (user_ace && (fuid == owner)) {
1334 		rmask = S_IRUSR;
1335 		wmask = S_IWUSR;
1336 		xmask = S_IXUSR;
1337 	} else {
1338 		rmask = S_IRGRP;
1339 		wmask = S_IWGRP;
1340 		xmask = S_IXGRP;
1341 	}
1342 
1343 	if (origmask & ACE_READ_DATA) {
1344 		if (mode & rmask) {
1345 			acepmask &= ~ACE_READ_DATA;
1346 		} else {
1347 			acepmask |= ACE_READ_DATA;
1348 		}
1349 	}
1350 
1351 	if (origmask & ACE_WRITE_DATA) {
1352 		if (mode & wmask) {
1353 			acepmask &= ~ACE_WRITE_DATA;
1354 		} else {
1355 			acepmask |= ACE_WRITE_DATA;
1356 		}
1357 	}
1358 
1359 	if (origmask & ACE_APPEND_DATA) {
1360 		if (mode & wmask) {
1361 			acepmask &= ~ACE_APPEND_DATA;
1362 		} else {
1363 			acepmask |= ACE_APPEND_DATA;
1364 		}
1365 	}
1366 
1367 	if (origmask & ACE_EXECUTE) {
1368 		if (mode & xmask) {
1369 			acepmask &= ~ACE_EXECUTE;
1370 		} else {
1371 			acepmask |= ACE_EXECUTE;
1372 		}
1373 	}
1374 	aclp->z_ops.ace_mask_set(acep, acepmask);
1375 }
1376 
1377 static void
1378 zfs_acl_chmod(zfsvfs_t *zfsvfs, uint64_t mode, zfs_acl_t *aclp)
1379 {
1380 	void		*acep = NULL;
1381 	uint64_t	who;
1382 	int		new_count, new_bytes;
1383 	int		ace_size;
1384 	int 		entry_type;
1385 	uint16_t	iflags, type;
1386 	uint32_t	access_mask;
1387 	zfs_acl_node_t	*newnode;
1388 	size_t 		abstract_size = aclp->z_ops.ace_abstract_size();
1389 	void 		*zacep;
1390 	uint32_t 	owner, group, everyone;
1391 	uint32_t	deny1, deny2, allow0;
1392 
1393 	new_count = new_bytes = 0;
1394 
1395 	acl_trivial_access_masks((mode_t)mode, &allow0, &deny1, &deny2,
1396 	    &owner, &group, &everyone);
1397 
1398 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1399 
1400 	zacep = newnode->z_acldata;
1401 	if (allow0) {
1402 		zfs_set_ace(aclp, zacep, allow0, ALLOW, -1, ACE_OWNER);
1403 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1404 		new_count++;
1405 		new_bytes += abstract_size;
1406 	} if (deny1) {
1407 		zfs_set_ace(aclp, zacep, deny1, DENY, -1, ACE_OWNER);
1408 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1409 		new_count++;
1410 		new_bytes += abstract_size;
1411 	}
1412 	if (deny2) {
1413 		zfs_set_ace(aclp, zacep, deny2, DENY, -1, OWNING_GROUP);
1414 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1415 		new_count++;
1416 		new_bytes += abstract_size;
1417 	}
1418 
1419 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1420 	    &iflags, &type)) {
1421 		uint16_t inherit_flags;
1422 
1423 		entry_type = (iflags & ACE_TYPE_FLAGS);
1424 		inherit_flags = (iflags & ALL_INHERIT);
1425 
1426 		if ((entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1427 		    (entry_type == OWNING_GROUP)) &&
1428 		    ((inherit_flags & ACE_INHERIT_ONLY_ACE) == 0)) {
1429 			continue;
1430 		}
1431 
1432 		if ((type != ALLOW && type != DENY) ||
1433 		    (inherit_flags & ACE_INHERIT_ONLY_ACE)) {
1434 			if (inherit_flags)
1435 				aclp->z_hints |= ZFS_INHERIT_ACE;
1436 			switch (type) {
1437 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1438 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1439 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1440 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1441 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1442 				break;
1443 			}
1444 		} else {
1445 
1446 			/*
1447 			 * Limit permissions to be no greater than
1448 			 * group permissions
1449 			 */
1450 			if (zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) {
1451 				if (!(mode & S_IRGRP))
1452 					access_mask &= ~ACE_READ_DATA;
1453 				if (!(mode & S_IWGRP))
1454 					access_mask &=
1455 					    ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
1456 				if (!(mode & S_IXGRP))
1457 					access_mask &= ~ACE_EXECUTE;
1458 				access_mask &=
1459 				    ~(ACE_WRITE_OWNER|ACE_WRITE_ACL|
1460 				    ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS);
1461 			}
1462 		}
1463 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1464 		ace_size = aclp->z_ops.ace_size(acep);
1465 		zacep = (void *)((uintptr_t)zacep + ace_size);
1466 		new_count++;
1467 		new_bytes += ace_size;
1468 	}
1469 	zfs_set_ace(aclp, zacep, owner, 0, -1, ACE_OWNER);
1470 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1471 	zfs_set_ace(aclp, zacep, group, 0, -1, OWNING_GROUP);
1472 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1473 	zfs_set_ace(aclp, zacep, everyone, 0, -1, ACE_EVERYONE);
1474 
1475 	new_count += 3;
1476 	new_bytes += abstract_size * 3;
1477 	zfs_acl_release_nodes(aclp);
1478 	aclp->z_acl_count = new_count;
1479 	aclp->z_acl_bytes = new_bytes;
1480 	newnode->z_ace_count = new_count;
1481 	newnode->z_size = new_bytes;
1482 	list_insert_tail(&aclp->z_acl, newnode);
1483 }
1484 
1485 int
1486 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1487 {
1488 	mutex_enter(&zp->z_lock);
1489 	mutex_enter(&zp->z_acl_lock);
1490 	*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1491 	(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1492 	zfs_acl_chmod(zp->z_zfsvfs, mode, *aclp);
1493 	mutex_exit(&zp->z_acl_lock);
1494 	mutex_exit(&zp->z_lock);
1495 	ASSERT(*aclp);
1496 	return (0);
1497 }
1498 
1499 /*
1500  * strip off write_owner and write_acl
1501  */
1502 static void
1503 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1504 {
1505 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1506 
1507 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1508 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1509 		mask &= ~RESTRICTED_CLEAR;
1510 		aclp->z_ops.ace_mask_set(acep, mask);
1511 	}
1512 }
1513 
1514 /*
1515  * Should ACE be inherited?
1516  */
1517 static int
1518 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1519 {
1520 	int	iflags = (acep_flags & 0xf);
1521 
1522 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1523 		return (1);
1524 	else if (iflags & ACE_FILE_INHERIT_ACE)
1525 		return (!((vtype == VDIR) &&
1526 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1527 	return (0);
1528 }
1529 
1530 /*
1531  * inherit inheritable ACEs from parent
1532  */
1533 static zfs_acl_t *
1534 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1535     uint64_t mode, boolean_t *need_chmod)
1536 {
1537 	void		*pacep;
1538 	void		*acep;
1539 	zfs_acl_node_t  *aclnode;
1540 	zfs_acl_t	*aclp = NULL;
1541 	uint64_t	who;
1542 	uint32_t	access_mask;
1543 	uint16_t	iflags, newflags, type;
1544 	size_t		ace_size;
1545 	void		*data1, *data2;
1546 	size_t		data1sz, data2sz;
1547 	boolean_t	vdir = vtype == VDIR;
1548 	boolean_t	vreg = vtype == VREG;
1549 	boolean_t	passthrough, passthrough_x, noallow;
1550 
1551 	passthrough_x =
1552 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1553 	passthrough = passthrough_x ||
1554 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1555 	noallow =
1556 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1557 
1558 	*need_chmod = B_TRUE;
1559 	pacep = NULL;
1560 	aclp = zfs_acl_alloc(paclp->z_version);
1561 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD || vtype == VLNK)
1562 		return (aclp);
1563 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1564 	    &access_mask, &iflags, &type)) {
1565 
1566 		/*
1567 		 * don't inherit bogus ACEs
1568 		 */
1569 		if (!zfs_acl_valid_ace_type(type, iflags))
1570 			continue;
1571 
1572 		if (noallow && type == ALLOW)
1573 			continue;
1574 
1575 		ace_size = aclp->z_ops.ace_size(pacep);
1576 
1577 		if (!zfs_ace_can_use(vtype, iflags))
1578 			continue;
1579 
1580 		/*
1581 		 * If owner@, group@, or everyone@ inheritable
1582 		 * then zfs_acl_chmod() isn't needed.
1583 		 */
1584 		if (passthrough &&
1585 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1586 		    ((iflags & OWNING_GROUP) ==
1587 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1588 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1589 			*need_chmod = B_FALSE;
1590 		}
1591 
1592 		if (!vdir && passthrough_x &&
1593 		    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1594 			access_mask &= ~ACE_EXECUTE;
1595 		}
1596 
1597 		aclnode = zfs_acl_node_alloc(ace_size);
1598 		list_insert_tail(&aclp->z_acl, aclnode);
1599 		acep = aclnode->z_acldata;
1600 
1601 		zfs_set_ace(aclp, acep, access_mask, type,
1602 		    who, iflags|ACE_INHERITED_ACE);
1603 
1604 		/*
1605 		 * Copy special opaque data if any
1606 		 */
1607 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1608 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1609 			    &data2)) == data1sz);
1610 			bcopy(data1, data2, data2sz);
1611 		}
1612 
1613 		aclp->z_acl_count++;
1614 		aclnode->z_ace_count++;
1615 		aclp->z_acl_bytes += aclnode->z_size;
1616 		newflags = aclp->z_ops.ace_flags_get(acep);
1617 
1618 		if (vdir)
1619 			aclp->z_hints |= ZFS_INHERIT_ACE;
1620 
1621 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1622 			newflags &= ~ALL_INHERIT;
1623 			aclp->z_ops.ace_flags_set(acep,
1624 			    newflags|ACE_INHERITED_ACE);
1625 			zfs_restricted_update(zfsvfs, aclp, acep);
1626 			continue;
1627 		}
1628 
1629 		ASSERT(vdir);
1630 
1631 		/*
1632 		 * If only FILE_INHERIT is set then turn on
1633 		 * inherit_only
1634 		 */
1635 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1636 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1637 			newflags |= ACE_INHERIT_ONLY_ACE;
1638 			aclp->z_ops.ace_flags_set(acep,
1639 			    newflags|ACE_INHERITED_ACE);
1640 		} else {
1641 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1642 			aclp->z_ops.ace_flags_set(acep,
1643 			    newflags|ACE_INHERITED_ACE);
1644 		}
1645 	}
1646 	return (aclp);
1647 }
1648 
1649 /*
1650  * Create file system object initial permissions
1651  * including inheritable ACEs.
1652  */
1653 int
1654 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1655     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1656 {
1657 	int		error;
1658 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1659 	zfs_acl_t	*paclp;
1660 	gid_t		gid;
1661 	boolean_t	need_chmod = B_TRUE;
1662 	boolean_t	inherited = B_FALSE;
1663 	uint64_t	parentgid;
1664 
1665 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1666 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1667 
1668 	if (vsecp)
1669 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1670 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1671 			return (error);
1672 	/*
1673 	 * Determine uid and gid.
1674 	 */
1675 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1676 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1677 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1678 		    (uint64_t)vap->va_uid, cr,
1679 		    ZFS_OWNER, &acl_ids->z_fuidp);
1680 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1681 		    (uint64_t)vap->va_gid, cr,
1682 		    ZFS_GROUP, &acl_ids->z_fuidp);
1683 		gid = vap->va_gid;
1684 	} else {
1685 		if (IS_EPHEMERAL(dzp->z_gid))
1686 			VERIFY(0 == sa_lookup(dzp->z_sa_hdl, SA_ZPL_GID(zfsvfs),
1687 			    &parentgid, sizeof (parentgid)));
1688 		else
1689 			parentgid = (uint64_t)dzp->z_gid;
1690 
1691 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1692 		    cr, &acl_ids->z_fuidp);
1693 		acl_ids->z_fgid = 0;
1694 		if (vap->va_mask & AT_GID)  {
1695 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1696 			    (uint64_t)vap->va_gid,
1697 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1698 			gid = vap->va_gid;
1699 			if (acl_ids->z_fgid != parentgid &&
1700 			    !groupmember(vap->va_gid, cr) &&
1701 			    secpolicy_vnode_create_gid(cr) != 0)
1702 				acl_ids->z_fgid = 0;
1703 		}
1704 		if (acl_ids->z_fgid == 0) {
1705 			if (dzp->z_mode & S_ISGID) {
1706 				char		*domain;
1707 				uint32_t	rid;
1708 
1709 				acl_ids->z_fgid = parentgid;
1710 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1711 				    cr, ZFS_GROUP);
1712 
1713 				if (zfsvfs->z_use_fuids &&
1714 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1715 					domain = zfs_fuid_idx_domain(
1716 					    &zfsvfs->z_fuid_idx,
1717 					    FUID_INDEX(acl_ids->z_fgid));
1718 					rid = FUID_RID(acl_ids->z_fgid);
1719 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1720 					    domain, rid,
1721 					    FUID_INDEX(acl_ids->z_fgid),
1722 					    acl_ids->z_fgid, ZFS_GROUP);
1723 				}
1724 			} else {
1725 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1726 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1727 				gid = crgetgid(cr);
1728 			}
1729 		}
1730 	}
1731 
1732 	/*
1733 	 * If we're creating a directory, and the parent directory has the
1734 	 * set-GID bit set, set in on the new directory.
1735 	 * Otherwise, if the user is neither privileged nor a member of the
1736 	 * file's new group, clear the file's set-GID bit.
1737 	 */
1738 
1739 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1740 	    (vap->va_type == VDIR)) {
1741 		acl_ids->z_mode |= S_ISGID;
1742 	} else {
1743 		if ((acl_ids->z_mode & S_ISGID) &&
1744 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1745 			acl_ids->z_mode &= ~S_ISGID;
1746 	}
1747 
1748 	if (acl_ids->z_aclp == NULL) {
1749 		mutex_enter(&dzp->z_lock);
1750 		if (!(flag & IS_ROOT_NODE) && (ZTOV(dzp)->v_type == VDIR &&
1751 		    (dzp->z_pflags & ZFS_INHERIT_ACE)) &&
1752 		    !(dzp->z_pflags & ZFS_XATTR)) {
1753 			mutex_enter(&dzp->z_acl_lock);
1754 			VERIFY(0 == zfs_acl_node_read(dzp, &paclp, B_FALSE));
1755 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1756 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1757 			mutex_exit(&dzp->z_acl_lock);
1758 			inherited = B_TRUE;
1759 		} else {
1760 			acl_ids->z_aclp =
1761 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1762 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1763 		}
1764 		mutex_exit(&dzp->z_lock);
1765 		if (need_chmod) {
1766 			acl_ids->z_aclp->z_hints |= (vap->va_type == VDIR) ?
1767 			    ZFS_ACL_AUTO_INHERIT : 0;
1768 			zfs_acl_chmod(zfsvfs, acl_ids->z_mode, acl_ids->z_aclp);
1769 		}
1770 	}
1771 
1772 	if (inherited || vsecp) {
1773 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1774 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1775 		    acl_ids->z_fuid, acl_ids->z_fgid);
1776 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1777 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1778 	}
1779 
1780 	return (0);
1781 }
1782 
1783 /*
1784  * Free ACL and fuid_infop, but not the acl_ids structure
1785  */
1786 void
1787 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1788 {
1789 	if (acl_ids->z_aclp)
1790 		zfs_acl_free(acl_ids->z_aclp);
1791 	if (acl_ids->z_fuidp)
1792 		zfs_fuid_info_free(acl_ids->z_fuidp);
1793 	acl_ids->z_aclp = NULL;
1794 	acl_ids->z_fuidp = NULL;
1795 }
1796 
1797 boolean_t
1798 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1799 {
1800 	return (zfs_fuid_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1801 	    zfs_fuid_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1802 }
1803 
1804 /*
1805  * Retrieve a files ACL
1806  */
1807 int
1808 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1809 {
1810 	zfs_acl_t	*aclp;
1811 	ulong_t		mask;
1812 	int		error;
1813 	int 		count = 0;
1814 	int		largeace = 0;
1815 
1816 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1817 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1818 
1819 	if (mask == 0)
1820 		return (ENOSYS);
1821 
1822 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1823 		return (error);
1824 
1825 	mutex_enter(&zp->z_acl_lock);
1826 
1827 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
1828 	if (error != 0) {
1829 		mutex_exit(&zp->z_acl_lock);
1830 		return (error);
1831 	}
1832 
1833 	/*
1834 	 * Scan ACL to determine number of ACEs
1835 	 */
1836 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1837 		void *zacep = NULL;
1838 		uint64_t who;
1839 		uint32_t access_mask;
1840 		uint16_t type, iflags;
1841 
1842 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1843 		    &who, &access_mask, &iflags, &type)) {
1844 			switch (type) {
1845 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1846 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1847 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1848 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1849 				largeace++;
1850 				continue;
1851 			default:
1852 				count++;
1853 			}
1854 		}
1855 		vsecp->vsa_aclcnt = count;
1856 	} else
1857 		count = (int)aclp->z_acl_count;
1858 
1859 	if (mask & VSA_ACECNT) {
1860 		vsecp->vsa_aclcnt = count;
1861 	}
1862 
1863 	if (mask & VSA_ACE) {
1864 		size_t aclsz;
1865 
1866 		aclsz = count * sizeof (ace_t) +
1867 		    sizeof (ace_object_t) * largeace;
1868 
1869 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1870 		vsecp->vsa_aclentsz = aclsz;
1871 
1872 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1873 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1874 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1875 		else {
1876 			zfs_acl_node_t *aclnode;
1877 			void *start = vsecp->vsa_aclentp;
1878 
1879 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1880 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1881 				bcopy(aclnode->z_acldata, start,
1882 				    aclnode->z_size);
1883 				start = (caddr_t)start + aclnode->z_size;
1884 			}
1885 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1886 			    aclp->z_acl_bytes);
1887 		}
1888 	}
1889 	if (mask & VSA_ACE_ACLFLAGS) {
1890 		vsecp->vsa_aclflags = 0;
1891 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1892 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1893 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
1894 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1895 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1896 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1897 	}
1898 
1899 	mutex_exit(&zp->z_acl_lock);
1900 
1901 	return (0);
1902 }
1903 
1904 int
1905 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
1906     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1907 {
1908 	zfs_acl_t *aclp;
1909 	zfs_acl_node_t *aclnode;
1910 	int aclcnt = vsecp->vsa_aclcnt;
1911 	int error;
1912 
1913 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1914 		return (EINVAL);
1915 
1916 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1917 
1918 	aclp->z_hints = 0;
1919 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1920 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1921 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1922 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1923 		    aclcnt, &aclnode->z_size)) != 0) {
1924 			zfs_acl_free(aclp);
1925 			zfs_acl_node_free(aclnode);
1926 			return (error);
1927 		}
1928 	} else {
1929 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1930 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1931 		    &aclnode->z_size, fuidp, cr)) != 0) {
1932 			zfs_acl_free(aclp);
1933 			zfs_acl_node_free(aclnode);
1934 			return (error);
1935 		}
1936 	}
1937 	aclp->z_acl_bytes = aclnode->z_size;
1938 	aclnode->z_ace_count = aclcnt;
1939 	aclp->z_acl_count = aclcnt;
1940 	list_insert_head(&aclp->z_acl, aclnode);
1941 
1942 	/*
1943 	 * If flags are being set then add them to z_hints
1944 	 */
1945 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1946 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
1947 			aclp->z_hints |= ZFS_ACL_PROTECTED;
1948 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1949 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
1950 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1951 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1952 	}
1953 
1954 	*zaclp = aclp;
1955 
1956 	return (0);
1957 }
1958 
1959 /*
1960  * Set a files ACL
1961  */
1962 int
1963 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1964 {
1965 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1966 	zilog_t		*zilog = zfsvfs->z_log;
1967 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1968 	dmu_tx_t	*tx;
1969 	int		error;
1970 	zfs_acl_t	*aclp;
1971 	zfs_fuid_info_t	*fuidp = NULL;
1972 	boolean_t	fuid_dirtied;
1973 
1974 	if (mask == 0)
1975 		return (ENOSYS);
1976 
1977 	if (zp->z_pflags & ZFS_IMMUTABLE)
1978 		return (EPERM);
1979 
1980 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
1981 		return (error);
1982 
1983 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1984 	    &aclp);
1985 	if (error)
1986 		return (error);
1987 
1988 	/*
1989 	 * If ACL wide flags aren't being set then preserve any
1990 	 * existing flags.
1991 	 */
1992 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1993 		aclp->z_hints |=
1994 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1995 	}
1996 top:
1997 	mutex_enter(&zp->z_lock);
1998 	mutex_enter(&zp->z_acl_lock);
1999 
2000 	tx = dmu_tx_create(zfsvfs->z_os);
2001 
2002 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2003 
2004 	fuid_dirtied = zfsvfs->z_fuid_dirty;
2005 	if (fuid_dirtied)
2006 		zfs_fuid_txhold(zfsvfs, tx);
2007 
2008 	/*
2009 	 * If old version and ACL won't fit in bonus and we aren't
2010 	 * upgrading then take out necessary DMU holds
2011 	 */
2012 
2013 	if (ZFS_EXTERNAL_ACL(zp)) {
2014 		if (zfsvfs->z_version <= ZPL_VERSION_SA &&
2015 		    ZNODE_ACL_VERSION(zp) <= ZFS_ACL_VERSION_INITIAL) {
2016 			dmu_tx_hold_free(tx, ZFS_EXTERNAL_ACL(zp), 0,
2017 			    DMU_OBJECT_END);
2018 		} else {
2019 			dmu_tx_hold_write(tx, ZFS_EXTERNAL_ACL(zp),
2020 			    0, aclp->z_acl_bytes);
2021 		}
2022 	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2023 		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2024 	}
2025 
2026 	zfs_sa_upgrade_txholds(tx, zp);
2027 	error = dmu_tx_assign(tx, TXG_NOWAIT);
2028 	if (error) {
2029 		mutex_exit(&zp->z_acl_lock);
2030 		mutex_exit(&zp->z_lock);
2031 
2032 		if (error == ERESTART) {
2033 			dmu_tx_wait(tx);
2034 			dmu_tx_abort(tx);
2035 			goto top;
2036 		}
2037 		dmu_tx_abort(tx);
2038 		zfs_acl_free(aclp);
2039 		return (error);
2040 	}
2041 
2042 	error = zfs_aclset_common(zp, aclp, cr, tx);
2043 	ASSERT(error == 0);
2044 	zp->z_acl_cached = aclp;
2045 
2046 	if (fuid_dirtied)
2047 		zfs_fuid_sync(zfsvfs, tx);
2048 
2049 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2050 
2051 	if (fuidp)
2052 		zfs_fuid_info_free(fuidp);
2053 	dmu_tx_commit(tx);
2054 done:
2055 	mutex_exit(&zp->z_acl_lock);
2056 	mutex_exit(&zp->z_lock);
2057 
2058 	return (error);
2059 }
2060 
2061 /*
2062  * Check accesses of interest (AoI) against attributes of the dataset
2063  * such as read-only.  Returns zero if no AoI conflict with dataset
2064  * attributes, otherwise an appropriate errno is returned.
2065  */
2066 static int
2067 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2068 {
2069 	if ((v4_mode & WRITE_MASK) &&
2070 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2071 	    (!IS_DEVVP(ZTOV(zp)) ||
2072 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2073 		return (EROFS);
2074 	}
2075 
2076 	/*
2077 	 * Only check for READONLY on non-directories.
2078 	 */
2079 	if ((v4_mode & WRITE_MASK_DATA) &&
2080 	    (((ZTOV(zp)->v_type != VDIR) &&
2081 	    (zp->z_pflags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2082 	    (ZTOV(zp)->v_type == VDIR &&
2083 	    (zp->z_pflags & ZFS_IMMUTABLE)))) {
2084 		return (EPERM);
2085 	}
2086 
2087 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2088 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2089 		return (EPERM);
2090 	}
2091 
2092 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2093 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2094 		return (EACCES);
2095 	}
2096 
2097 	return (0);
2098 }
2099 
2100 /*
2101  * The primary usage of this function is to loop through all of the
2102  * ACEs in the znode, determining what accesses of interest (AoI) to
2103  * the caller are allowed or denied.  The AoI are expressed as bits in
2104  * the working_mode parameter.  As each ACE is processed, bits covered
2105  * by that ACE are removed from the working_mode.  This removal
2106  * facilitates two things.  The first is that when the working mode is
2107  * empty (= 0), we know we've looked at all the AoI. The second is
2108  * that the ACE interpretation rules don't allow a later ACE to undo
2109  * something granted or denied by an earlier ACE.  Removing the
2110  * discovered access or denial enforces this rule.  At the end of
2111  * processing the ACEs, all AoI that were found to be denied are
2112  * placed into the working_mode, giving the caller a mask of denied
2113  * accesses.  Returns:
2114  *	0		if all AoI granted
2115  *	EACCESS 	if the denied mask is non-zero
2116  *	other error	if abnormal failure (e.g., IO error)
2117  *
2118  * A secondary usage of the function is to determine if any of the
2119  * AoI are granted.  If an ACE grants any access in
2120  * the working_mode, we immediately short circuit out of the function.
2121  * This mode is chosen by setting anyaccess to B_TRUE.  The
2122  * working_mode is not a denied access mask upon exit if the function
2123  * is used in this manner.
2124  */
2125 static int
2126 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2127     boolean_t anyaccess, cred_t *cr)
2128 {
2129 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2130 	zfs_acl_t	*aclp;
2131 	int		error;
2132 	uid_t		uid = crgetuid(cr);
2133 	uint64_t 	who;
2134 	uint16_t	type, iflags;
2135 	uint16_t	entry_type;
2136 	uint32_t	access_mask;
2137 	uint32_t	deny_mask = 0;
2138 	zfs_ace_hdr_t	*acep = NULL;
2139 	boolean_t	checkit;
2140 	uint64_t	gowner;
2141 
2142 	mutex_enter(&zp->z_acl_lock);
2143 
2144 	error = zfs_acl_node_read(zp, &aclp, B_FALSE);
2145 	if (error != 0) {
2146 		mutex_exit(&zp->z_acl_lock);
2147 		return (error);
2148 	}
2149 
2150 	ASSERT(zp->z_acl_cached);
2151 
2152 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GID(zfsvfs),
2153 	    &gowner, sizeof (gowner))) != 0) {
2154 		mutex_exit(&zp->z_acl_lock);
2155 		return (error);
2156 	}
2157 
2158 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2159 	    &iflags, &type)) {
2160 		uint32_t mask_matched;
2161 
2162 		if (!zfs_acl_valid_ace_type(type, iflags))
2163 			continue;
2164 
2165 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2166 			continue;
2167 
2168 		/* Skip ACE if it does not affect any AoI */
2169 		mask_matched = (access_mask & *working_mode);
2170 		if (!mask_matched)
2171 			continue;
2172 
2173 		entry_type = (iflags & ACE_TYPE_FLAGS);
2174 
2175 		checkit = B_FALSE;
2176 
2177 		switch (entry_type) {
2178 		case ACE_OWNER:
2179 			if (uid == zp->z_uid)
2180 				checkit = B_TRUE;
2181 			break;
2182 		case OWNING_GROUP:
2183 			who = gowner;
2184 			/*FALLTHROUGH*/
2185 		case ACE_IDENTIFIER_GROUP:
2186 			checkit = zfs_groupmember(zfsvfs, who, cr);
2187 			break;
2188 		case ACE_EVERYONE:
2189 			checkit = B_TRUE;
2190 			break;
2191 
2192 		/* USER Entry */
2193 		default:
2194 			if (entry_type == 0) {
2195 				uid_t newid;
2196 
2197 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2198 				    ZFS_ACE_USER);
2199 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2200 				    uid == newid)
2201 					checkit = B_TRUE;
2202 				break;
2203 			} else {
2204 				mutex_exit(&zp->z_acl_lock);
2205 				return (EIO);
2206 			}
2207 		}
2208 
2209 		if (checkit) {
2210 			if (type == DENY) {
2211 				DTRACE_PROBE3(zfs__ace__denies,
2212 				    znode_t *, zp,
2213 				    zfs_ace_hdr_t *, acep,
2214 				    uint32_t, mask_matched);
2215 				deny_mask |= mask_matched;
2216 			} else {
2217 				DTRACE_PROBE3(zfs__ace__allows,
2218 				    znode_t *, zp,
2219 				    zfs_ace_hdr_t *, acep,
2220 				    uint32_t, mask_matched);
2221 				if (anyaccess) {
2222 					mutex_exit(&zp->z_acl_lock);
2223 					return (0);
2224 				}
2225 			}
2226 			*working_mode &= ~mask_matched;
2227 		}
2228 
2229 		/* Are we done? */
2230 		if (*working_mode == 0)
2231 			break;
2232 	}
2233 
2234 	mutex_exit(&zp->z_acl_lock);
2235 
2236 	/* Put the found 'denies' back on the working mode */
2237 	if (deny_mask) {
2238 		*working_mode |= deny_mask;
2239 		return (EACCES);
2240 	} else if (*working_mode) {
2241 		return (-1);
2242 	}
2243 
2244 	return (0);
2245 }
2246 
2247 /*
2248  * Return true if any access whatsoever granted, we don't actually
2249  * care what access is granted.
2250  */
2251 boolean_t
2252 zfs_has_access(znode_t *zp, cred_t *cr)
2253 {
2254 	uint32_t have = ACE_ALL_PERMS;
2255 
2256 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2257 		return (secpolicy_vnode_any_access(cr, ZTOV(zp),
2258 		    zp->z_uid) == 0);
2259 	}
2260 	return (B_TRUE);
2261 }
2262 
2263 static int
2264 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2265     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2266 {
2267 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2268 	int err;
2269 
2270 	*working_mode = v4_mode;
2271 	*check_privs = B_TRUE;
2272 
2273 	/*
2274 	 * Short circuit empty requests
2275 	 */
2276 	if (v4_mode == 0 || zfsvfs->z_replay) {
2277 		*working_mode = 0;
2278 		return (0);
2279 	}
2280 
2281 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2282 		*check_privs = B_FALSE;
2283 		return (err);
2284 	}
2285 
2286 	/*
2287 	 * The caller requested that the ACL check be skipped.  This
2288 	 * would only happen if the caller checked VOP_ACCESS() with a
2289 	 * 32 bit ACE mask and already had the appropriate permissions.
2290 	 */
2291 	if (skipaclchk) {
2292 		*working_mode = 0;
2293 		return (0);
2294 	}
2295 
2296 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2297 }
2298 
2299 static int
2300 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2301     cred_t *cr)
2302 {
2303 	if (*working_mode != ACE_WRITE_DATA)
2304 		return (EACCES);
2305 
2306 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2307 	    check_privs, B_FALSE, cr));
2308 }
2309 
2310 int
2311 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2312 {
2313 	boolean_t owner = B_FALSE;
2314 	boolean_t groupmbr = B_FALSE;
2315 	boolean_t is_attr;
2316 	uid_t uid = crgetuid(cr);
2317 	int error;
2318 
2319 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2320 		return (EACCES);
2321 
2322 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2323 	    (ZTOV(zdp)->v_type == VDIR));
2324 	if (is_attr)
2325 		goto slow;
2326 
2327 
2328 	mutex_enter(&zdp->z_acl_lock);
2329 
2330 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2331 		mutex_exit(&zdp->z_acl_lock);
2332 		return (0);
2333 	}
2334 
2335 	if (IS_EPHEMERAL(zdp->z_uid) != 0 || IS_EPHEMERAL(zdp->z_gid) != 0) {
2336 		mutex_exit(&zdp->z_acl_lock);
2337 		goto slow;
2338 	}
2339 
2340 	if (uid == zdp->z_uid) {
2341 		owner = B_TRUE;
2342 		if (zdp->z_mode & S_IXUSR) {
2343 			mutex_exit(&zdp->z_acl_lock);
2344 			return (0);
2345 		} else {
2346 			mutex_exit(&zdp->z_acl_lock);
2347 			goto slow;
2348 		}
2349 	}
2350 	if (groupmember(zdp->z_gid, cr)) {
2351 		groupmbr = B_TRUE;
2352 		if (zdp->z_mode & S_IXGRP) {
2353 			mutex_exit(&zdp->z_acl_lock);
2354 			return (0);
2355 		} else {
2356 			mutex_exit(&zdp->z_acl_lock);
2357 			goto slow;
2358 		}
2359 	}
2360 	if (!owner && !groupmbr) {
2361 		if (zdp->z_mode & S_IXOTH) {
2362 			mutex_exit(&zdp->z_acl_lock);
2363 			return (0);
2364 		}
2365 	}
2366 
2367 	mutex_exit(&zdp->z_acl_lock);
2368 
2369 slow:
2370 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2371 	ZFS_ENTER(zdp->z_zfsvfs);
2372 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2373 	ZFS_EXIT(zdp->z_zfsvfs);
2374 	return (error);
2375 }
2376 
2377 /*
2378  * Determine whether Access should be granted/denied.
2379  * The least priv subsytem is always consulted as a basic privilege
2380  * can define any form of access.
2381  */
2382 int
2383 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2384 {
2385 	uint32_t	working_mode;
2386 	int		error;
2387 	int		is_attr;
2388 	boolean_t 	check_privs;
2389 	znode_t		*xzp;
2390 	znode_t 	*check_zp = zp;
2391 	mode_t		needed_bits;
2392 
2393 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2394 
2395 	/*
2396 	 * If attribute then validate against base file
2397 	 */
2398 	if (is_attr) {
2399 		uint64_t	parent;
2400 
2401 		if ((error = sa_lookup(zp->z_sa_hdl,
2402 		    SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2403 		    sizeof (parent))) != 0)
2404 			return (error);
2405 
2406 		if ((error = zfs_zget(zp->z_zfsvfs,
2407 		    parent, &xzp)) != 0)	{
2408 			return (error);
2409 		}
2410 
2411 		check_zp = xzp;
2412 
2413 		/*
2414 		 * fixup mode to map to xattr perms
2415 		 */
2416 
2417 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2418 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2419 			mode |= ACE_WRITE_NAMED_ATTRS;
2420 		}
2421 
2422 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2423 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2424 			mode |= ACE_READ_NAMED_ATTRS;
2425 		}
2426 	}
2427 
2428 	/*
2429 	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2430 	 * in needed_bits.  Map the bits mapped by working_mode (currently
2431 	 * missing) in missing_bits.
2432 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2433 	 * needed_bits.
2434 	 */
2435 	needed_bits = 0;
2436 
2437 	working_mode = mode;
2438 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2439 	    zp->z_uid == crgetuid(cr))
2440 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2441 
2442 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2443 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2444 		needed_bits |= VREAD;
2445 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2446 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2447 		needed_bits |= VWRITE;
2448 	if (working_mode & ACE_EXECUTE)
2449 		needed_bits |= VEXEC;
2450 
2451 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2452 	    &check_privs, skipaclchk, cr)) == 0) {
2453 		if (is_attr)
2454 			VN_RELE(ZTOV(xzp));
2455 		return (secpolicy_vnode_access2(cr, ZTOV(zp), zp->z_uid,
2456 		    needed_bits, needed_bits));
2457 	}
2458 
2459 	if (error && !check_privs) {
2460 		if (is_attr)
2461 			VN_RELE(ZTOV(xzp));
2462 		return (error);
2463 	}
2464 
2465 	if (error && (flags & V_APPEND)) {
2466 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2467 	}
2468 
2469 	if (error && check_privs) {
2470 		mode_t		checkmode = 0;
2471 
2472 		/*
2473 		 * First check for implicit owner permission on
2474 		 * read_acl/read_attributes
2475 		 */
2476 
2477 		error = 0;
2478 		ASSERT(working_mode != 0);
2479 
2480 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2481 		    zp->z_uid == crgetuid(cr)))
2482 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2483 
2484 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2485 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2486 			checkmode |= VREAD;
2487 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2488 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2489 			checkmode |= VWRITE;
2490 		if (working_mode & ACE_EXECUTE)
2491 			checkmode |= VEXEC;
2492 
2493 		error = secpolicy_vnode_access2(cr, ZTOV(check_zp), zp->z_uid,
2494 		    needed_bits & ~checkmode, needed_bits);
2495 
2496 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2497 			error = secpolicy_vnode_chown(cr, zp->z_uid);
2498 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2499 			error = secpolicy_vnode_setdac(cr, zp->z_uid);
2500 
2501 		if (error == 0 && (working_mode &
2502 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2503 			error = secpolicy_vnode_remove(cr);
2504 
2505 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2506 			error = secpolicy_vnode_chown(cr, zp->z_uid);
2507 		}
2508 		if (error == 0) {
2509 			/*
2510 			 * See if any bits other than those already checked
2511 			 * for are still present.  If so then return EACCES
2512 			 */
2513 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2514 				error = EACCES;
2515 			}
2516 		}
2517 	} else if (error == 0) {
2518 		error = secpolicy_vnode_access2(cr, ZTOV(zp), zp->z_uid,
2519 		    needed_bits, needed_bits);
2520 	}
2521 
2522 
2523 	if (is_attr)
2524 		VN_RELE(ZTOV(xzp));
2525 
2526 	return (error);
2527 }
2528 
2529 /*
2530  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2531  * native ACL format and call zfs_zaccess()
2532  */
2533 int
2534 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2535 {
2536 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2537 }
2538 
2539 /*
2540  * Access function for secpolicy_vnode_setattr
2541  */
2542 int
2543 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2544 {
2545 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2546 
2547 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2548 }
2549 
2550 static int
2551 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2552     mode_t available_perms, cred_t *cr)
2553 {
2554 	int error;
2555 
2556 	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2557 	    dzp->z_uid, available_perms, VWRITE|VEXEC);
2558 
2559 	if (error == 0)
2560 		error = zfs_sticky_remove_access(dzp, zp, cr);
2561 
2562 	return (error);
2563 }
2564 
2565 /*
2566  * Determine whether Access should be granted/deny, without
2567  * consulting least priv subsystem.
2568  *
2569  *
2570  * The following chart is the recommended NFSv4 enforcement for
2571  * ability to delete an object.
2572  *
2573  *      -------------------------------------------------------
2574  *      |   Parent Dir  |           Target Object Permissions |
2575  *      |  permissions  |                                     |
2576  *      -------------------------------------------------------
2577  *      |               | ACL Allows | ACL Denies| Delete     |
2578  *      |               |  Delete    |  Delete   | unspecified|
2579  *      -------------------------------------------------------
2580  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2581  *      |  DELETE_CHILD |                                     |
2582  *      -------------------------------------------------------
2583  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2584  *      |  DELETE_CHILD |            |           |            |
2585  *      -------------------------------------------------------
2586  *      | ACL specifies |            |           |            |
2587  *      | only allow    | Permit     | Permit    | Permit     |
2588  *      | write and     |            |           |            |
2589  *      | execute       |            |           |            |
2590  *      -------------------------------------------------------
2591  *      | ACL denies    |            |           |            |
2592  *      | write and     | Permit     | Deny      | Deny       |
2593  *      | execute       |            |           |            |
2594  *      -------------------------------------------------------
2595  *         ^
2596  *         |
2597  *         No search privilege, can't even look up file?
2598  *
2599  */
2600 int
2601 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2602 {
2603 	uint32_t dzp_working_mode = 0;
2604 	uint32_t zp_working_mode = 0;
2605 	int dzp_error, zp_error;
2606 	mode_t available_perms;
2607 	boolean_t dzpcheck_privs = B_TRUE;
2608 	boolean_t zpcheck_privs = B_TRUE;
2609 
2610 	/*
2611 	 * We want specific DELETE permissions to
2612 	 * take precedence over WRITE/EXECUTE.  We don't
2613 	 * want an ACL such as this to mess us up.
2614 	 * user:joe:write_data:deny,user:joe:delete:allow
2615 	 *
2616 	 * However, deny permissions may ultimately be overridden
2617 	 * by secpolicy_vnode_access().
2618 	 *
2619 	 * We will ask for all of the necessary permissions and then
2620 	 * look at the working modes from the directory and target object
2621 	 * to determine what was found.
2622 	 */
2623 
2624 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2625 		return (EPERM);
2626 
2627 	/*
2628 	 * First row
2629 	 * If the directory permissions allow the delete, we are done.
2630 	 */
2631 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2632 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2633 		return (0);
2634 
2635 	/*
2636 	 * If target object has delete permission then we are done
2637 	 */
2638 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2639 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2640 		return (0);
2641 
2642 	ASSERT(dzp_error && zp_error);
2643 
2644 	if (!dzpcheck_privs)
2645 		return (dzp_error);
2646 	if (!zpcheck_privs)
2647 		return (zp_error);
2648 
2649 	/*
2650 	 * Second row
2651 	 *
2652 	 * If directory returns EACCES then delete_child was denied
2653 	 * due to deny delete_child.  In this case send the request through
2654 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2655 	 * since that *could* allow the delete based on write/execute permission
2656 	 * and we want delete permissions to override write/execute.
2657 	 */
2658 
2659 	if (dzp_error == EACCES)
2660 		return (secpolicy_vnode_remove(cr));
2661 
2662 	/*
2663 	 * Third Row
2664 	 * only need to see if we have write/execute on directory.
2665 	 */
2666 
2667 	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2668 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2669 
2670 	if (dzp_error != 0 && !dzpcheck_privs)
2671 		return (dzp_error);
2672 
2673 	/*
2674 	 * Fourth row
2675 	 */
2676 
2677 	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2678 	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2679 
2680 	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2681 
2682 }
2683 
2684 int
2685 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2686     znode_t *tzp, cred_t *cr)
2687 {
2688 	int add_perm;
2689 	int error;
2690 
2691 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2692 		return (EACCES);
2693 
2694 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2695 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2696 
2697 	/*
2698 	 * Rename permissions are combination of delete permission +
2699 	 * add file/subdir permission.
2700 	 */
2701 
2702 	/*
2703 	 * first make sure we do the delete portion.
2704 	 *
2705 	 * If that succeeds then check for add_file/add_subdir permissions
2706 	 */
2707 
2708 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2709 		return (error);
2710 
2711 	/*
2712 	 * If we have a tzp, see if we can delete it?
2713 	 */
2714 	if (tzp) {
2715 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2716 			return (error);
2717 	}
2718 
2719 	/*
2720 	 * Now check for add permissions
2721 	 */
2722 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2723 
2724 	return (error);
2725 }
2726