xref: /titanic_52/usr/src/uts/common/fs/zfs/zfs_acl.c (revision f7170741490edba9d1d9c697c177c887172bc741)
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 2011 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2013 by Delphix. 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/sid.h>
36 #include <sys/file.h>
37 #include <sys/stat.h>
38 #include <sys/kmem.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
41 #include <sys/unistd.h>
42 #include <sys/sdt.h>
43 #include <sys/fs/zfs.h>
44 #include <sys/mode.h>
45 #include <sys/policy.h>
46 #include <sys/zfs_znode.h>
47 #include <sys/zfs_fuid.h>
48 #include <sys/zfs_acl.h>
49 #include <sys/zfs_dir.h>
50 #include <sys/zfs_vfsops.h>
51 #include <sys/dmu.h>
52 #include <sys/dnode.h>
53 #include <sys/zap.h>
54 #include <sys/sa.h>
55 #include "fs/fs_subr.h"
56 #include <acl/acl_common.h>
57 
58 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
59 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
60 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
61 #define	MIN_ACE_TYPE	ALLOW
62 
63 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
64 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
65     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
66 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
67     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
68 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
69     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
70 
71 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
72     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
73     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
74     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
75 
76 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
77 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
78     ACE_DELETE|ACE_DELETE_CHILD)
79 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
80 
81 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
82     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
83 
84 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
85     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
86 
87 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
88     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
89 
90 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
91 
92 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
93     ZFS_ACL_PROTECTED)
94 
95 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
96     ZFS_ACL_OBJ_ACE)
97 
98 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
99 
100 static uint16_t
101 zfs_ace_v0_get_type(void *acep)
102 {
103 	return (((zfs_oldace_t *)acep)->z_type);
104 }
105 
106 static uint16_t
107 zfs_ace_v0_get_flags(void *acep)
108 {
109 	return (((zfs_oldace_t *)acep)->z_flags);
110 }
111 
112 static uint32_t
113 zfs_ace_v0_get_mask(void *acep)
114 {
115 	return (((zfs_oldace_t *)acep)->z_access_mask);
116 }
117 
118 static uint64_t
119 zfs_ace_v0_get_who(void *acep)
120 {
121 	return (((zfs_oldace_t *)acep)->z_fuid);
122 }
123 
124 static void
125 zfs_ace_v0_set_type(void *acep, uint16_t type)
126 {
127 	((zfs_oldace_t *)acep)->z_type = type;
128 }
129 
130 static void
131 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
132 {
133 	((zfs_oldace_t *)acep)->z_flags = flags;
134 }
135 
136 static void
137 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
138 {
139 	((zfs_oldace_t *)acep)->z_access_mask = mask;
140 }
141 
142 static void
143 zfs_ace_v0_set_who(void *acep, uint64_t who)
144 {
145 	((zfs_oldace_t *)acep)->z_fuid = who;
146 }
147 
148 /*ARGSUSED*/
149 static size_t
150 zfs_ace_v0_size(void *acep)
151 {
152 	return (sizeof (zfs_oldace_t));
153 }
154 
155 static size_t
156 zfs_ace_v0_abstract_size(void)
157 {
158 	return (sizeof (zfs_oldace_t));
159 }
160 
161 static int
162 zfs_ace_v0_mask_off(void)
163 {
164 	return (offsetof(zfs_oldace_t, z_access_mask));
165 }
166 
167 /*ARGSUSED*/
168 static int
169 zfs_ace_v0_data(void *acep, void **datap)
170 {
171 	*datap = NULL;
172 	return (0);
173 }
174 
175 static acl_ops_t zfs_acl_v0_ops = {
176 	zfs_ace_v0_get_mask,
177 	zfs_ace_v0_set_mask,
178 	zfs_ace_v0_get_flags,
179 	zfs_ace_v0_set_flags,
180 	zfs_ace_v0_get_type,
181 	zfs_ace_v0_set_type,
182 	zfs_ace_v0_get_who,
183 	zfs_ace_v0_set_who,
184 	zfs_ace_v0_size,
185 	zfs_ace_v0_abstract_size,
186 	zfs_ace_v0_mask_off,
187 	zfs_ace_v0_data
188 };
189 
190 static uint16_t
191 zfs_ace_fuid_get_type(void *acep)
192 {
193 	return (((zfs_ace_hdr_t *)acep)->z_type);
194 }
195 
196 static uint16_t
197 zfs_ace_fuid_get_flags(void *acep)
198 {
199 	return (((zfs_ace_hdr_t *)acep)->z_flags);
200 }
201 
202 static uint32_t
203 zfs_ace_fuid_get_mask(void *acep)
204 {
205 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
206 }
207 
208 static uint64_t
209 zfs_ace_fuid_get_who(void *args)
210 {
211 	uint16_t entry_type;
212 	zfs_ace_t *acep = args;
213 
214 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
215 
216 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
217 	    entry_type == ACE_EVERYONE)
218 		return (-1);
219 	return (((zfs_ace_t *)acep)->z_fuid);
220 }
221 
222 static void
223 zfs_ace_fuid_set_type(void *acep, uint16_t type)
224 {
225 	((zfs_ace_hdr_t *)acep)->z_type = type;
226 }
227 
228 static void
229 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
230 {
231 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
232 }
233 
234 static void
235 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
236 {
237 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
238 }
239 
240 static void
241 zfs_ace_fuid_set_who(void *arg, uint64_t who)
242 {
243 	zfs_ace_t *acep = arg;
244 
245 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
246 
247 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
248 	    entry_type == ACE_EVERYONE)
249 		return;
250 	acep->z_fuid = who;
251 }
252 
253 static size_t
254 zfs_ace_fuid_size(void *acep)
255 {
256 	zfs_ace_hdr_t *zacep = acep;
257 	uint16_t entry_type;
258 
259 	switch (zacep->z_type) {
260 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
261 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
262 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
263 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
264 		return (sizeof (zfs_object_ace_t));
265 	case ALLOW:
266 	case DENY:
267 		entry_type =
268 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
269 		if (entry_type == ACE_OWNER ||
270 		    entry_type == OWNING_GROUP ||
271 		    entry_type == ACE_EVERYONE)
272 			return (sizeof (zfs_ace_hdr_t));
273 		/*FALLTHROUGH*/
274 	default:
275 		return (sizeof (zfs_ace_t));
276 	}
277 }
278 
279 static size_t
280 zfs_ace_fuid_abstract_size(void)
281 {
282 	return (sizeof (zfs_ace_hdr_t));
283 }
284 
285 static int
286 zfs_ace_fuid_mask_off(void)
287 {
288 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
289 }
290 
291 static int
292 zfs_ace_fuid_data(void *acep, void **datap)
293 {
294 	zfs_ace_t *zacep = acep;
295 	zfs_object_ace_t *zobjp;
296 
297 	switch (zacep->z_hdr.z_type) {
298 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
299 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
300 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
301 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
302 		zobjp = acep;
303 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
304 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
305 	default:
306 		*datap = NULL;
307 		return (0);
308 	}
309 }
310 
311 static acl_ops_t zfs_acl_fuid_ops = {
312 	zfs_ace_fuid_get_mask,
313 	zfs_ace_fuid_set_mask,
314 	zfs_ace_fuid_get_flags,
315 	zfs_ace_fuid_set_flags,
316 	zfs_ace_fuid_get_type,
317 	zfs_ace_fuid_set_type,
318 	zfs_ace_fuid_get_who,
319 	zfs_ace_fuid_set_who,
320 	zfs_ace_fuid_size,
321 	zfs_ace_fuid_abstract_size,
322 	zfs_ace_fuid_mask_off,
323 	zfs_ace_fuid_data
324 };
325 
326 /*
327  * The following three functions are provided for compatibility with
328  * older ZPL version in order to determine if the file use to have
329  * an external ACL and what version of ACL previously existed on the
330  * file.  Would really be nice to not need this, sigh.
331  */
332 uint64_t
333 zfs_external_acl(znode_t *zp)
334 {
335 	zfs_acl_phys_t acl_phys;
336 	int error;
337 
338 	if (zp->z_is_sa)
339 		return (0);
340 
341 	/*
342 	 * Need to deal with a potential
343 	 * race where zfs_sa_upgrade could cause
344 	 * z_isa_sa to change.
345 	 *
346 	 * If the lookup fails then the state of z_is_sa should have
347 	 * changed.
348 	 */
349 
350 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
351 	    &acl_phys, sizeof (acl_phys))) == 0)
352 		return (acl_phys.z_acl_extern_obj);
353 	else {
354 		/*
355 		 * after upgrade the SA_ZPL_ZNODE_ACL should have been
356 		 * removed
357 		 */
358 		VERIFY(zp->z_is_sa && error == ENOENT);
359 		return (0);
360 	}
361 }
362 
363 /*
364  * Determine size of ACL in bytes
365  *
366  * This is more complicated than it should be since we have to deal
367  * with old external ACLs.
368  */
369 static int
370 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
371     zfs_acl_phys_t *aclphys)
372 {
373 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
374 	uint64_t acl_count;
375 	int size;
376 	int error;
377 
378 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
379 	if (zp->z_is_sa) {
380 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
381 		    &size)) != 0)
382 			return (error);
383 		*aclsize = size;
384 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
385 		    &acl_count, sizeof (acl_count))) != 0)
386 			return (error);
387 		*aclcount = acl_count;
388 	} else {
389 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
390 		    aclphys, sizeof (*aclphys))) != 0)
391 			return (error);
392 
393 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
394 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
395 			*aclcount = aclphys->z_acl_size;
396 		} else {
397 			*aclsize = aclphys->z_acl_size;
398 			*aclcount = aclphys->z_acl_count;
399 		}
400 	}
401 	return (0);
402 }
403 
404 int
405 zfs_znode_acl_version(znode_t *zp)
406 {
407 	zfs_acl_phys_t acl_phys;
408 
409 	if (zp->z_is_sa)
410 		return (ZFS_ACL_VERSION_FUID);
411 	else {
412 		int error;
413 
414 		/*
415 		 * Need to deal with a potential
416 		 * race where zfs_sa_upgrade could cause
417 		 * z_isa_sa to change.
418 		 *
419 		 * If the lookup fails then the state of z_is_sa should have
420 		 * changed.
421 		 */
422 		if ((error = sa_lookup(zp->z_sa_hdl,
423 		    SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
424 		    &acl_phys, sizeof (acl_phys))) == 0)
425 			return (acl_phys.z_acl_version);
426 		else {
427 			/*
428 			 * After upgrade SA_ZPL_ZNODE_ACL should have
429 			 * been removed.
430 			 */
431 			VERIFY(zp->z_is_sa && error == ENOENT);
432 			return (ZFS_ACL_VERSION_FUID);
433 		}
434 	}
435 }
436 
437 static int
438 zfs_acl_version(int version)
439 {
440 	if (version < ZPL_VERSION_FUID)
441 		return (ZFS_ACL_VERSION_INITIAL);
442 	else
443 		return (ZFS_ACL_VERSION_FUID);
444 }
445 
446 static int
447 zfs_acl_version_zp(znode_t *zp)
448 {
449 	return (zfs_acl_version(zp->z_zfsvfs->z_version));
450 }
451 
452 zfs_acl_t *
453 zfs_acl_alloc(int vers)
454 {
455 	zfs_acl_t *aclp;
456 
457 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
458 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
459 	    offsetof(zfs_acl_node_t, z_next));
460 	aclp->z_version = vers;
461 	if (vers == ZFS_ACL_VERSION_FUID)
462 		aclp->z_ops = zfs_acl_fuid_ops;
463 	else
464 		aclp->z_ops = zfs_acl_v0_ops;
465 	return (aclp);
466 }
467 
468 zfs_acl_node_t *
469 zfs_acl_node_alloc(size_t bytes)
470 {
471 	zfs_acl_node_t *aclnode;
472 
473 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
474 	if (bytes) {
475 		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
476 		aclnode->z_allocdata = aclnode->z_acldata;
477 		aclnode->z_allocsize = bytes;
478 		aclnode->z_size = bytes;
479 	}
480 
481 	return (aclnode);
482 }
483 
484 static void
485 zfs_acl_node_free(zfs_acl_node_t *aclnode)
486 {
487 	if (aclnode->z_allocsize)
488 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
489 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
490 }
491 
492 static void
493 zfs_acl_release_nodes(zfs_acl_t *aclp)
494 {
495 	zfs_acl_node_t *aclnode;
496 
497 	while (aclnode = list_head(&aclp->z_acl)) {
498 		list_remove(&aclp->z_acl, aclnode);
499 		zfs_acl_node_free(aclnode);
500 	}
501 	aclp->z_acl_count = 0;
502 	aclp->z_acl_bytes = 0;
503 }
504 
505 void
506 zfs_acl_free(zfs_acl_t *aclp)
507 {
508 	zfs_acl_release_nodes(aclp);
509 	list_destroy(&aclp->z_acl);
510 	kmem_free(aclp, sizeof (zfs_acl_t));
511 }
512 
513 static boolean_t
514 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
515 {
516 	uint16_t entry_type;
517 
518 	switch (type) {
519 	case ALLOW:
520 	case DENY:
521 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
522 	case ACE_SYSTEM_ALARM_ACE_TYPE:
523 		entry_type = flags & ACE_TYPE_FLAGS;
524 		return (entry_type == ACE_OWNER ||
525 		    entry_type == OWNING_GROUP ||
526 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
527 		    entry_type == ACE_IDENTIFIER_GROUP);
528 	default:
529 		if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
530 			return (B_TRUE);
531 	}
532 	return (B_FALSE);
533 }
534 
535 static boolean_t
536 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
537 {
538 	/*
539 	 * first check type of entry
540 	 */
541 
542 	if (!zfs_acl_valid_ace_type(type, iflags))
543 		return (B_FALSE);
544 
545 	switch (type) {
546 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
547 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
548 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
549 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
550 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
551 			return (B_FALSE);
552 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
553 	}
554 
555 	/*
556 	 * next check inheritance level flags
557 	 */
558 
559 	if (obj_type == VDIR &&
560 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
561 		aclp->z_hints |= ZFS_INHERIT_ACE;
562 
563 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
564 		if ((iflags & (ACE_FILE_INHERIT_ACE|
565 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
566 			return (B_FALSE);
567 		}
568 	}
569 
570 	return (B_TRUE);
571 }
572 
573 static void *
574 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
575     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
576 {
577 	zfs_acl_node_t *aclnode;
578 
579 	ASSERT(aclp);
580 
581 	if (start == NULL) {
582 		aclnode = list_head(&aclp->z_acl);
583 		if (aclnode == NULL)
584 			return (NULL);
585 
586 		aclp->z_next_ace = aclnode->z_acldata;
587 		aclp->z_curr_node = aclnode;
588 		aclnode->z_ace_idx = 0;
589 	}
590 
591 	aclnode = aclp->z_curr_node;
592 
593 	if (aclnode == NULL)
594 		return (NULL);
595 
596 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
597 		aclnode = list_next(&aclp->z_acl, aclnode);
598 		if (aclnode == NULL)
599 			return (NULL);
600 		else {
601 			aclp->z_curr_node = aclnode;
602 			aclnode->z_ace_idx = 0;
603 			aclp->z_next_ace = aclnode->z_acldata;
604 		}
605 	}
606 
607 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
608 		void *acep = aclp->z_next_ace;
609 		size_t ace_size;
610 
611 		/*
612 		 * Make sure we don't overstep our bounds
613 		 */
614 		ace_size = aclp->z_ops.ace_size(acep);
615 
616 		if (((caddr_t)acep + ace_size) >
617 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
618 			return (NULL);
619 		}
620 
621 		*iflags = aclp->z_ops.ace_flags_get(acep);
622 		*type = aclp->z_ops.ace_type_get(acep);
623 		*access_mask = aclp->z_ops.ace_mask_get(acep);
624 		*who = aclp->z_ops.ace_who_get(acep);
625 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
626 		aclnode->z_ace_idx++;
627 
628 		return ((void *)acep);
629 	}
630 	return (NULL);
631 }
632 
633 /*ARGSUSED*/
634 static uint64_t
635 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
636     uint16_t *flags, uint16_t *type, uint32_t *mask)
637 {
638 	zfs_acl_t *aclp = datap;
639 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
640 	uint64_t who;
641 
642 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
643 	    flags, type);
644 	return ((uint64_t)(uintptr_t)acep);
645 }
646 
647 static zfs_acl_node_t *
648 zfs_acl_curr_node(zfs_acl_t *aclp)
649 {
650 	ASSERT(aclp->z_curr_node);
651 	return (aclp->z_curr_node);
652 }
653 
654 /*
655  * Copy ACE to internal ZFS format.
656  * While processing the ACL each ACE will be validated for correctness.
657  * ACE FUIDs will be created later.
658  */
659 int
660 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
661     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
662     zfs_fuid_info_t **fuidp, cred_t *cr)
663 {
664 	int i;
665 	uint16_t entry_type;
666 	zfs_ace_t *aceptr = z_acl;
667 	ace_t *acep = datap;
668 	zfs_object_ace_t *zobjacep;
669 	ace_object_t *aceobjp;
670 
671 	for (i = 0; i != aclcnt; i++) {
672 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
673 		aceptr->z_hdr.z_flags = acep->a_flags;
674 		aceptr->z_hdr.z_type = acep->a_type;
675 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
676 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
677 		    entry_type != ACE_EVERYONE) {
678 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
679 			    cr, (entry_type == 0) ?
680 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
681 		}
682 
683 		/*
684 		 * Make sure ACE is valid
685 		 */
686 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
687 		    aceptr->z_hdr.z_flags) != B_TRUE)
688 			return (SET_ERROR(EINVAL));
689 
690 		switch (acep->a_type) {
691 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
692 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
693 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
694 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
695 			zobjacep = (zfs_object_ace_t *)aceptr;
696 			aceobjp = (ace_object_t *)acep;
697 
698 			bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
699 			    sizeof (aceobjp->a_obj_type));
700 			bcopy(aceobjp->a_inherit_obj_type,
701 			    zobjacep->z_inherit_type,
702 			    sizeof (aceobjp->a_inherit_obj_type));
703 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
704 			break;
705 		default:
706 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
707 		}
708 
709 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
710 		    aclp->z_ops.ace_size(aceptr));
711 	}
712 
713 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
714 
715 	return (0);
716 }
717 
718 /*
719  * Copy ZFS ACEs to fixed size ace_t layout
720  */
721 static void
722 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
723     void *datap, int filter)
724 {
725 	uint64_t who;
726 	uint32_t access_mask;
727 	uint16_t iflags, type;
728 	zfs_ace_hdr_t *zacep = NULL;
729 	ace_t *acep = datap;
730 	ace_object_t *objacep;
731 	zfs_object_ace_t *zobjacep;
732 	size_t ace_size;
733 	uint16_t entry_type;
734 
735 	while (zacep = zfs_acl_next_ace(aclp, zacep,
736 	    &who, &access_mask, &iflags, &type)) {
737 
738 		switch (type) {
739 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
740 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
741 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
742 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
743 			if (filter) {
744 				continue;
745 			}
746 			zobjacep = (zfs_object_ace_t *)zacep;
747 			objacep = (ace_object_t *)acep;
748 			bcopy(zobjacep->z_object_type,
749 			    objacep->a_obj_type,
750 			    sizeof (zobjacep->z_object_type));
751 			bcopy(zobjacep->z_inherit_type,
752 			    objacep->a_inherit_obj_type,
753 			    sizeof (zobjacep->z_inherit_type));
754 			ace_size = sizeof (ace_object_t);
755 			break;
756 		default:
757 			ace_size = sizeof (ace_t);
758 			break;
759 		}
760 
761 		entry_type = (iflags & ACE_TYPE_FLAGS);
762 		if ((entry_type != ACE_OWNER &&
763 		    entry_type != OWNING_GROUP &&
764 		    entry_type != ACE_EVERYONE)) {
765 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
766 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
767 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
768 		} else {
769 			acep->a_who = (uid_t)(int64_t)who;
770 		}
771 		acep->a_access_mask = access_mask;
772 		acep->a_flags = iflags;
773 		acep->a_type = type;
774 		acep = (ace_t *)((caddr_t)acep + ace_size);
775 	}
776 }
777 
778 static int
779 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
780     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
781 {
782 	int i;
783 	zfs_oldace_t *aceptr = z_acl;
784 
785 	for (i = 0; i != aclcnt; i++, aceptr++) {
786 		aceptr->z_access_mask = acep[i].a_access_mask;
787 		aceptr->z_type = acep[i].a_type;
788 		aceptr->z_flags = acep[i].a_flags;
789 		aceptr->z_fuid = acep[i].a_who;
790 		/*
791 		 * Make sure ACE is valid
792 		 */
793 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
794 		    aceptr->z_flags) != B_TRUE)
795 			return (SET_ERROR(EINVAL));
796 	}
797 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
798 	return (0);
799 }
800 
801 /*
802  * convert old ACL format to new
803  */
804 void
805 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
806 {
807 	zfs_oldace_t *oldaclp;
808 	int i;
809 	uint16_t type, iflags;
810 	uint32_t access_mask;
811 	uint64_t who;
812 	void *cookie = NULL;
813 	zfs_acl_node_t *newaclnode;
814 
815 	ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
816 	/*
817 	 * First create the ACE in a contiguous piece of memory
818 	 * for zfs_copy_ace_2_fuid().
819 	 *
820 	 * We only convert an ACL once, so this won't happen
821 	 * everytime.
822 	 */
823 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
824 	    KM_SLEEP);
825 	i = 0;
826 	while (cookie = zfs_acl_next_ace(aclp, cookie, &who,
827 	    &access_mask, &iflags, &type)) {
828 		oldaclp[i].z_flags = iflags;
829 		oldaclp[i].z_type = type;
830 		oldaclp[i].z_fuid = who;
831 		oldaclp[i++].z_access_mask = access_mask;
832 	}
833 
834 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
835 	    sizeof (zfs_object_ace_t));
836 	aclp->z_ops = zfs_acl_fuid_ops;
837 	VERIFY(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
838 	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
839 	    &newaclnode->z_size, NULL, cr) == 0);
840 	newaclnode->z_ace_count = aclp->z_acl_count;
841 	aclp->z_version = ZFS_ACL_VERSION;
842 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
843 
844 	/*
845 	 * Release all previous ACL nodes
846 	 */
847 
848 	zfs_acl_release_nodes(aclp);
849 
850 	list_insert_head(&aclp->z_acl, newaclnode);
851 
852 	aclp->z_acl_bytes = newaclnode->z_size;
853 	aclp->z_acl_count = newaclnode->z_ace_count;
854 
855 }
856 
857 /*
858  * Convert unix access mask to v4 access mask
859  */
860 static uint32_t
861 zfs_unix_to_v4(uint32_t access_mask)
862 {
863 	uint32_t new_mask = 0;
864 
865 	if (access_mask & S_IXOTH)
866 		new_mask |= ACE_EXECUTE;
867 	if (access_mask & S_IWOTH)
868 		new_mask |= ACE_WRITE_DATA;
869 	if (access_mask & S_IROTH)
870 		new_mask |= ACE_READ_DATA;
871 	return (new_mask);
872 }
873 
874 static void
875 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
876     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
877 {
878 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
879 
880 	aclp->z_ops.ace_mask_set(acep, access_mask);
881 	aclp->z_ops.ace_type_set(acep, access_type);
882 	aclp->z_ops.ace_flags_set(acep, entry_type);
883 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
884 	    type != ACE_EVERYONE))
885 		aclp->z_ops.ace_who_set(acep, fuid);
886 }
887 
888 /*
889  * Determine mode of file based on ACL.
890  * Also, create FUIDs for any User/Group ACEs
891  */
892 uint64_t
893 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
894     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
895 {
896 	int		entry_type;
897 	mode_t		mode;
898 	mode_t		seen = 0;
899 	zfs_ace_hdr_t 	*acep = NULL;
900 	uint64_t	who;
901 	uint16_t	iflags, type;
902 	uint32_t	access_mask;
903 	boolean_t	an_exec_denied = B_FALSE;
904 
905 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
906 
907 	while (acep = zfs_acl_next_ace(aclp, acep, &who,
908 	    &access_mask, &iflags, &type)) {
909 
910 		if (!zfs_acl_valid_ace_type(type, iflags))
911 			continue;
912 
913 		entry_type = (iflags & ACE_TYPE_FLAGS);
914 
915 		/*
916 		 * Skip over owner@, group@ or everyone@ inherit only ACEs
917 		 */
918 		if ((iflags & ACE_INHERIT_ONLY_ACE) &&
919 		    (entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
920 		    entry_type == OWNING_GROUP))
921 			continue;
922 
923 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
924 		    who == fuid)) {
925 			if ((access_mask & ACE_READ_DATA) &&
926 			    (!(seen & S_IRUSR))) {
927 				seen |= S_IRUSR;
928 				if (type == ALLOW) {
929 					mode |= S_IRUSR;
930 				}
931 			}
932 			if ((access_mask & ACE_WRITE_DATA) &&
933 			    (!(seen & S_IWUSR))) {
934 				seen |= S_IWUSR;
935 				if (type == ALLOW) {
936 					mode |= S_IWUSR;
937 				}
938 			}
939 			if ((access_mask & ACE_EXECUTE) &&
940 			    (!(seen & S_IXUSR))) {
941 				seen |= S_IXUSR;
942 				if (type == ALLOW) {
943 					mode |= S_IXUSR;
944 				}
945 			}
946 		} else if (entry_type == OWNING_GROUP ||
947 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
948 			if ((access_mask & ACE_READ_DATA) &&
949 			    (!(seen & S_IRGRP))) {
950 				seen |= S_IRGRP;
951 				if (type == ALLOW) {
952 					mode |= S_IRGRP;
953 				}
954 			}
955 			if ((access_mask & ACE_WRITE_DATA) &&
956 			    (!(seen & S_IWGRP))) {
957 				seen |= S_IWGRP;
958 				if (type == ALLOW) {
959 					mode |= S_IWGRP;
960 				}
961 			}
962 			if ((access_mask & ACE_EXECUTE) &&
963 			    (!(seen & S_IXGRP))) {
964 				seen |= S_IXGRP;
965 				if (type == ALLOW) {
966 					mode |= S_IXGRP;
967 				}
968 			}
969 		} else if (entry_type == ACE_EVERYONE) {
970 			if ((access_mask & ACE_READ_DATA)) {
971 				if (!(seen & S_IRUSR)) {
972 					seen |= S_IRUSR;
973 					if (type == ALLOW) {
974 						mode |= S_IRUSR;
975 					}
976 				}
977 				if (!(seen & S_IRGRP)) {
978 					seen |= S_IRGRP;
979 					if (type == ALLOW) {
980 						mode |= S_IRGRP;
981 					}
982 				}
983 				if (!(seen & S_IROTH)) {
984 					seen |= S_IROTH;
985 					if (type == ALLOW) {
986 						mode |= S_IROTH;
987 					}
988 				}
989 			}
990 			if ((access_mask & ACE_WRITE_DATA)) {
991 				if (!(seen & S_IWUSR)) {
992 					seen |= S_IWUSR;
993 					if (type == ALLOW) {
994 						mode |= S_IWUSR;
995 					}
996 				}
997 				if (!(seen & S_IWGRP)) {
998 					seen |= S_IWGRP;
999 					if (type == ALLOW) {
1000 						mode |= S_IWGRP;
1001 					}
1002 				}
1003 				if (!(seen & S_IWOTH)) {
1004 					seen |= S_IWOTH;
1005 					if (type == ALLOW) {
1006 						mode |= S_IWOTH;
1007 					}
1008 				}
1009 			}
1010 			if ((access_mask & ACE_EXECUTE)) {
1011 				if (!(seen & S_IXUSR)) {
1012 					seen |= S_IXUSR;
1013 					if (type == ALLOW) {
1014 						mode |= S_IXUSR;
1015 					}
1016 				}
1017 				if (!(seen & S_IXGRP)) {
1018 					seen |= S_IXGRP;
1019 					if (type == ALLOW) {
1020 						mode |= S_IXGRP;
1021 					}
1022 				}
1023 				if (!(seen & S_IXOTH)) {
1024 					seen |= S_IXOTH;
1025 					if (type == ALLOW) {
1026 						mode |= S_IXOTH;
1027 					}
1028 				}
1029 			}
1030 		} else {
1031 			/*
1032 			 * Only care if this IDENTIFIER_GROUP or
1033 			 * USER ACE denies execute access to someone,
1034 			 * mode is not affected
1035 			 */
1036 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1037 				an_exec_denied = B_TRUE;
1038 		}
1039 	}
1040 
1041 	/*
1042 	 * Failure to allow is effectively a deny, so execute permission
1043 	 * is denied if it was never mentioned or if we explicitly
1044 	 * weren't allowed it.
1045 	 */
1046 	if (!an_exec_denied &&
1047 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1048 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1049 		an_exec_denied = B_TRUE;
1050 
1051 	if (an_exec_denied)
1052 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1053 	else
1054 		*pflags |= ZFS_NO_EXECS_DENIED;
1055 
1056 	return (mode);
1057 }
1058 
1059 /*
1060  * Read an external acl object.  If the intent is to modify, always
1061  * create a new acl and leave any cached acl in place.
1062  */
1063 static int
1064 zfs_acl_node_read(znode_t *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1065     boolean_t will_modify)
1066 {
1067 	zfs_acl_t	*aclp;
1068 	int		aclsize;
1069 	int		acl_count;
1070 	zfs_acl_node_t	*aclnode;
1071 	zfs_acl_phys_t	znode_acl;
1072 	int		version;
1073 	int		error;
1074 	boolean_t	drop_lock = B_FALSE;
1075 
1076 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1077 
1078 	if (zp->z_acl_cached && !will_modify) {
1079 		*aclpp = zp->z_acl_cached;
1080 		return (0);
1081 	}
1082 
1083 	/*
1084 	 * close race where znode could be upgrade while trying to
1085 	 * read the znode attributes.
1086 	 *
1087 	 * But this could only happen if the file isn't already an SA
1088 	 * znode
1089 	 */
1090 	if (!zp->z_is_sa && !have_lock) {
1091 		mutex_enter(&zp->z_lock);
1092 		drop_lock = B_TRUE;
1093 	}
1094 	version = zfs_znode_acl_version(zp);
1095 
1096 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1097 	    &acl_count, &znode_acl)) != 0) {
1098 		goto done;
1099 	}
1100 
1101 	aclp = zfs_acl_alloc(version);
1102 
1103 	aclp->z_acl_count = acl_count;
1104 	aclp->z_acl_bytes = aclsize;
1105 
1106 	aclnode = zfs_acl_node_alloc(aclsize);
1107 	aclnode->z_ace_count = aclp->z_acl_count;
1108 	aclnode->z_size = aclsize;
1109 
1110 	if (!zp->z_is_sa) {
1111 		if (znode_acl.z_acl_extern_obj) {
1112 			error = dmu_read(zp->z_zfsvfs->z_os,
1113 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1114 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1115 		} else {
1116 			bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1117 			    aclnode->z_size);
1118 		}
1119 	} else {
1120 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1121 		    aclnode->z_acldata, aclnode->z_size);
1122 	}
1123 
1124 	if (error != 0) {
1125 		zfs_acl_free(aclp);
1126 		zfs_acl_node_free(aclnode);
1127 		/* convert checksum errors into IO errors */
1128 		if (error == ECKSUM)
1129 			error = SET_ERROR(EIO);
1130 		goto done;
1131 	}
1132 
1133 	list_insert_head(&aclp->z_acl, aclnode);
1134 
1135 	*aclpp = aclp;
1136 	if (!will_modify)
1137 		zp->z_acl_cached = aclp;
1138 done:
1139 	if (drop_lock)
1140 		mutex_exit(&zp->z_lock);
1141 	return (error);
1142 }
1143 
1144 /*ARGSUSED*/
1145 void
1146 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1147     boolean_t start, void *userdata)
1148 {
1149 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1150 
1151 	if (start) {
1152 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1153 	} else {
1154 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1155 		    cb->cb_acl_node);
1156 	}
1157 	*dataptr = cb->cb_acl_node->z_acldata;
1158 	*length = cb->cb_acl_node->z_size;
1159 }
1160 
1161 int
1162 zfs_acl_chown_setattr(znode_t *zp)
1163 {
1164 	int error;
1165 	zfs_acl_t *aclp;
1166 
1167 	ASSERT(MUTEX_HELD(&zp->z_lock));
1168 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1169 
1170 	if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1171 		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1172 		    &zp->z_pflags, zp->z_uid, zp->z_gid);
1173 	return (error);
1174 }
1175 
1176 /*
1177  * common code for setting ACLs.
1178  *
1179  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1180  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1181  * already checked the acl and knows whether to inherit.
1182  */
1183 int
1184 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1185 {
1186 	int			error;
1187 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1188 	dmu_object_type_t	otype;
1189 	zfs_acl_locator_cb_t	locate = { 0 };
1190 	uint64_t		mode;
1191 	sa_bulk_attr_t		bulk[5];
1192 	uint64_t		ctime[2];
1193 	int			count = 0;
1194 
1195 	mode = zp->z_mode;
1196 
1197 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1198 	    zp->z_uid, zp->z_gid);
1199 
1200 	zp->z_mode = mode;
1201 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1202 	    &mode, sizeof (mode));
1203 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1204 	    &zp->z_pflags, sizeof (zp->z_pflags));
1205 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1206 	    &ctime, sizeof (ctime));
1207 
1208 	if (zp->z_acl_cached) {
1209 		zfs_acl_free(zp->z_acl_cached);
1210 		zp->z_acl_cached = NULL;
1211 	}
1212 
1213 	/*
1214 	 * Upgrade needed?
1215 	 */
1216 	if (!zfsvfs->z_use_fuids) {
1217 		otype = DMU_OT_OLDACL;
1218 	} else {
1219 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1220 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1221 			zfs_acl_xform(zp, aclp, cr);
1222 		ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1223 		otype = DMU_OT_ACL;
1224 	}
1225 
1226 	/*
1227 	 * Arrgh, we have to handle old on disk format
1228 	 * as well as newer (preferred) SA format.
1229 	 */
1230 
1231 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1232 		locate.cb_aclp = aclp;
1233 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1234 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1235 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1236 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1237 	} else { /* Painful legacy way */
1238 		zfs_acl_node_t *aclnode;
1239 		uint64_t off = 0;
1240 		zfs_acl_phys_t acl_phys;
1241 		uint64_t aoid;
1242 
1243 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1244 		    &acl_phys, sizeof (acl_phys))) != 0)
1245 			return (error);
1246 
1247 		aoid = acl_phys.z_acl_extern_obj;
1248 
1249 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1250 			/*
1251 			 * If ACL was previously external and we are now
1252 			 * converting to new ACL format then release old
1253 			 * ACL object and create a new one.
1254 			 */
1255 			if (aoid &&
1256 			    aclp->z_version != acl_phys.z_acl_version) {
1257 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1258 				if (error)
1259 					return (error);
1260 				aoid = 0;
1261 			}
1262 			if (aoid == 0) {
1263 				aoid = dmu_object_alloc(zfsvfs->z_os,
1264 				    otype, aclp->z_acl_bytes,
1265 				    otype == DMU_OT_ACL ?
1266 				    DMU_OT_SYSACL : DMU_OT_NONE,
1267 				    otype == DMU_OT_ACL ?
1268 				    DN_MAX_BONUSLEN : 0, tx);
1269 			} else {
1270 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1271 				    aoid, aclp->z_acl_bytes, 0, tx);
1272 			}
1273 			acl_phys.z_acl_extern_obj = aoid;
1274 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1275 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1276 				if (aclnode->z_ace_count == 0)
1277 					continue;
1278 				dmu_write(zfsvfs->z_os, aoid, off,
1279 				    aclnode->z_size, aclnode->z_acldata, tx);
1280 				off += aclnode->z_size;
1281 			}
1282 		} else {
1283 			void *start = acl_phys.z_ace_data;
1284 			/*
1285 			 * Migrating back embedded?
1286 			 */
1287 			if (acl_phys.z_acl_extern_obj) {
1288 				error = dmu_object_free(zfsvfs->z_os,
1289 				    acl_phys.z_acl_extern_obj, tx);
1290 				if (error)
1291 					return (error);
1292 				acl_phys.z_acl_extern_obj = 0;
1293 			}
1294 
1295 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1296 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1297 				if (aclnode->z_ace_count == 0)
1298 					continue;
1299 				bcopy(aclnode->z_acldata, start,
1300 				    aclnode->z_size);
1301 				start = (caddr_t)start + aclnode->z_size;
1302 			}
1303 		}
1304 		/*
1305 		 * If Old version then swap count/bytes to match old
1306 		 * layout of znode_acl_phys_t.
1307 		 */
1308 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1309 			acl_phys.z_acl_size = aclp->z_acl_count;
1310 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1311 		} else {
1312 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1313 			acl_phys.z_acl_count = aclp->z_acl_count;
1314 		}
1315 		acl_phys.z_acl_version = aclp->z_version;
1316 
1317 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1318 		    &acl_phys, sizeof (acl_phys));
1319 	}
1320 
1321 	/*
1322 	 * Replace ACL wide bits, but first clear them.
1323 	 */
1324 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1325 
1326 	zp->z_pflags |= aclp->z_hints;
1327 
1328 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1329 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1330 
1331 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime, B_TRUE);
1332 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1333 }
1334 
1335 static void
1336 zfs_acl_chmod(vtype_t vtype, uint64_t mode, boolean_t trim, zfs_acl_t *aclp)
1337 {
1338 	void		*acep = NULL;
1339 	uint64_t	who;
1340 	int		new_count, new_bytes;
1341 	int		ace_size;
1342 	int 		entry_type;
1343 	uint16_t	iflags, type;
1344 	uint32_t	access_mask;
1345 	zfs_acl_node_t	*newnode;
1346 	size_t 		abstract_size = aclp->z_ops.ace_abstract_size();
1347 	void 		*zacep;
1348 	boolean_t	isdir;
1349 	trivial_acl_t	masks;
1350 
1351 	new_count = new_bytes = 0;
1352 
1353 	isdir = (vtype == VDIR);
1354 
1355 	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1356 
1357 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1358 
1359 	zacep = newnode->z_acldata;
1360 	if (masks.allow0) {
1361 		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1362 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1363 		new_count++;
1364 		new_bytes += abstract_size;
1365 	}
1366 	if (masks.deny1) {
1367 		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1368 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1369 		new_count++;
1370 		new_bytes += abstract_size;
1371 	}
1372 	if (masks.deny2) {
1373 		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1374 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1375 		new_count++;
1376 		new_bytes += abstract_size;
1377 	}
1378 
1379 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1380 	    &iflags, &type)) {
1381 		uint16_t inherit_flags;
1382 
1383 		entry_type = (iflags & ACE_TYPE_FLAGS);
1384 		inherit_flags = (iflags & ALL_INHERIT);
1385 
1386 		if ((entry_type == ACE_OWNER || entry_type == ACE_EVERYONE ||
1387 		    (entry_type == OWNING_GROUP)) &&
1388 		    ((inherit_flags & ACE_INHERIT_ONLY_ACE) == 0)) {
1389 			continue;
1390 		}
1391 
1392 		/*
1393 		 * If this ACL has any inheritable ACEs, mark that in
1394 		 * the hints (which are later masked into the pflags)
1395 		 * so create knows to do inheritance.
1396 		 */
1397 		if (isdir && (inherit_flags &
1398 		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1399 			aclp->z_hints |= ZFS_INHERIT_ACE;
1400 
1401 		if ((type != ALLOW && type != DENY) ||
1402 		    (inherit_flags & ACE_INHERIT_ONLY_ACE)) {
1403 			switch (type) {
1404 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1405 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1406 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1407 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1408 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1409 				break;
1410 			}
1411 		} else {
1412 
1413 			/*
1414 			 * Limit permissions to be no greater than
1415 			 * group permissions.
1416 			 * The "aclinherit" and "aclmode" properties
1417 			 * affect policy for create and chmod(2),
1418 			 * respectively.
1419 			 */
1420 			if ((type == ALLOW) && trim)
1421 				access_mask &= masks.group;
1422 		}
1423 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1424 		ace_size = aclp->z_ops.ace_size(acep);
1425 		zacep = (void *)((uintptr_t)zacep + ace_size);
1426 		new_count++;
1427 		new_bytes += ace_size;
1428 	}
1429 	zfs_set_ace(aclp, zacep, masks.owner, 0, -1, ACE_OWNER);
1430 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1431 	zfs_set_ace(aclp, zacep, masks.group, 0, -1, OWNING_GROUP);
1432 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1433 	zfs_set_ace(aclp, zacep, masks.everyone, 0, -1, ACE_EVERYONE);
1434 
1435 	new_count += 3;
1436 	new_bytes += abstract_size * 3;
1437 	zfs_acl_release_nodes(aclp);
1438 	aclp->z_acl_count = new_count;
1439 	aclp->z_acl_bytes = new_bytes;
1440 	newnode->z_ace_count = new_count;
1441 	newnode->z_size = new_bytes;
1442 	list_insert_tail(&aclp->z_acl, newnode);
1443 }
1444 
1445 int
1446 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1447 {
1448 	int error = 0;
1449 
1450 	mutex_enter(&zp->z_acl_lock);
1451 	mutex_enter(&zp->z_lock);
1452 	if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1453 		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1454 	else
1455 		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1456 
1457 	if (error == 0) {
1458 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1459 		zfs_acl_chmod(ZTOV(zp)->v_type, mode,
1460 		    (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1461 	}
1462 	mutex_exit(&zp->z_lock);
1463 	mutex_exit(&zp->z_acl_lock);
1464 
1465 	return (error);
1466 }
1467 
1468 /*
1469  * strip off write_owner and write_acl
1470  */
1471 static void
1472 zfs_restricted_update(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, void *acep)
1473 {
1474 	uint32_t mask = aclp->z_ops.ace_mask_get(acep);
1475 
1476 	if ((zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED) &&
1477 	    (aclp->z_ops.ace_type_get(acep) == ALLOW)) {
1478 		mask &= ~RESTRICTED_CLEAR;
1479 		aclp->z_ops.ace_mask_set(acep, mask);
1480 	}
1481 }
1482 
1483 /*
1484  * Should ACE be inherited?
1485  */
1486 static int
1487 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1488 {
1489 	int	iflags = (acep_flags & 0xf);
1490 
1491 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1492 		return (1);
1493 	else if (iflags & ACE_FILE_INHERIT_ACE)
1494 		return (!((vtype == VDIR) &&
1495 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1496 	return (0);
1497 }
1498 
1499 /*
1500  * inherit inheritable ACEs from parent
1501  */
1502 static zfs_acl_t *
1503 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1504     uint64_t mode, boolean_t *need_chmod)
1505 {
1506 	void		*pacep;
1507 	void		*acep;
1508 	zfs_acl_node_t  *aclnode;
1509 	zfs_acl_t	*aclp = NULL;
1510 	uint64_t	who;
1511 	uint32_t	access_mask;
1512 	uint16_t	iflags, newflags, type;
1513 	size_t		ace_size;
1514 	void		*data1, *data2;
1515 	size_t		data1sz, data2sz;
1516 	boolean_t	vdir = vtype == VDIR;
1517 	boolean_t	vreg = vtype == VREG;
1518 	boolean_t	passthrough, passthrough_x, noallow;
1519 
1520 	passthrough_x =
1521 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH_X;
1522 	passthrough = passthrough_x ||
1523 	    zfsvfs->z_acl_inherit == ZFS_ACL_PASSTHROUGH;
1524 	noallow =
1525 	    zfsvfs->z_acl_inherit == ZFS_ACL_NOALLOW;
1526 
1527 	*need_chmod = B_TRUE;
1528 	pacep = NULL;
1529 	aclp = zfs_acl_alloc(paclp->z_version);
1530 	if (zfsvfs->z_acl_inherit == ZFS_ACL_DISCARD || vtype == VLNK)
1531 		return (aclp);
1532 	while (pacep = zfs_acl_next_ace(paclp, pacep, &who,
1533 	    &access_mask, &iflags, &type)) {
1534 
1535 		/*
1536 		 * don't inherit bogus ACEs
1537 		 */
1538 		if (!zfs_acl_valid_ace_type(type, iflags))
1539 			continue;
1540 
1541 		if (noallow && type == ALLOW)
1542 			continue;
1543 
1544 		ace_size = aclp->z_ops.ace_size(pacep);
1545 
1546 		if (!zfs_ace_can_use(vtype, iflags))
1547 			continue;
1548 
1549 		/*
1550 		 * If owner@, group@, or everyone@ inheritable
1551 		 * then zfs_acl_chmod() isn't needed.
1552 		 */
1553 		if (passthrough &&
1554 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1555 		    ((iflags & OWNING_GROUP) ==
1556 		    OWNING_GROUP)) && (vreg || (vdir && (iflags &
1557 		    ACE_DIRECTORY_INHERIT_ACE)))) {
1558 			*need_chmod = B_FALSE;
1559 		}
1560 
1561 		if (!vdir && passthrough_x &&
1562 		    ((mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
1563 			access_mask &= ~ACE_EXECUTE;
1564 		}
1565 
1566 		aclnode = zfs_acl_node_alloc(ace_size);
1567 		list_insert_tail(&aclp->z_acl, aclnode);
1568 		acep = aclnode->z_acldata;
1569 
1570 		zfs_set_ace(aclp, acep, access_mask, type,
1571 		    who, iflags|ACE_INHERITED_ACE);
1572 
1573 		/*
1574 		 * Copy special opaque data if any
1575 		 */
1576 		if ((data1sz = paclp->z_ops.ace_data(pacep, &data1)) != 0) {
1577 			VERIFY((data2sz = aclp->z_ops.ace_data(acep,
1578 			    &data2)) == data1sz);
1579 			bcopy(data1, data2, data2sz);
1580 		}
1581 
1582 		aclp->z_acl_count++;
1583 		aclnode->z_ace_count++;
1584 		aclp->z_acl_bytes += aclnode->z_size;
1585 		newflags = aclp->z_ops.ace_flags_get(acep);
1586 
1587 		if (vdir)
1588 			aclp->z_hints |= ZFS_INHERIT_ACE;
1589 
1590 		if ((iflags & ACE_NO_PROPAGATE_INHERIT_ACE) || !vdir) {
1591 			newflags &= ~ALL_INHERIT;
1592 			aclp->z_ops.ace_flags_set(acep,
1593 			    newflags|ACE_INHERITED_ACE);
1594 			zfs_restricted_update(zfsvfs, aclp, acep);
1595 			continue;
1596 		}
1597 
1598 		ASSERT(vdir);
1599 
1600 		/*
1601 		 * If only FILE_INHERIT is set then turn on
1602 		 * inherit_only
1603 		 */
1604 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1605 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1606 			newflags |= ACE_INHERIT_ONLY_ACE;
1607 			aclp->z_ops.ace_flags_set(acep,
1608 			    newflags|ACE_INHERITED_ACE);
1609 		} else {
1610 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1611 			aclp->z_ops.ace_flags_set(acep,
1612 			    newflags|ACE_INHERITED_ACE);
1613 		}
1614 	}
1615 	return (aclp);
1616 }
1617 
1618 /*
1619  * Create file system object initial permissions
1620  * including inheritable ACEs.
1621  */
1622 int
1623 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1624     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1625 {
1626 	int		error;
1627 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1628 	zfs_acl_t	*paclp;
1629 	gid_t		gid;
1630 	boolean_t	need_chmod = B_TRUE;
1631 	boolean_t	inherited = B_FALSE;
1632 
1633 	bzero(acl_ids, sizeof (zfs_acl_ids_t));
1634 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1635 
1636 	if (vsecp)
1637 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1638 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1639 			return (error);
1640 	/*
1641 	 * Determine uid and gid.
1642 	 */
1643 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1644 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1645 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1646 		    (uint64_t)vap->va_uid, cr,
1647 		    ZFS_OWNER, &acl_ids->z_fuidp);
1648 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1649 		    (uint64_t)vap->va_gid, cr,
1650 		    ZFS_GROUP, &acl_ids->z_fuidp);
1651 		gid = vap->va_gid;
1652 	} else {
1653 		acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1654 		    cr, &acl_ids->z_fuidp);
1655 		acl_ids->z_fgid = 0;
1656 		if (vap->va_mask & AT_GID)  {
1657 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1658 			    (uint64_t)vap->va_gid,
1659 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1660 			gid = vap->va_gid;
1661 			if (acl_ids->z_fgid != dzp->z_gid &&
1662 			    !groupmember(vap->va_gid, cr) &&
1663 			    secpolicy_vnode_create_gid(cr) != 0)
1664 				acl_ids->z_fgid = 0;
1665 		}
1666 		if (acl_ids->z_fgid == 0) {
1667 			if (dzp->z_mode & S_ISGID) {
1668 				char		*domain;
1669 				uint32_t	rid;
1670 
1671 				acl_ids->z_fgid = dzp->z_gid;
1672 				gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1673 				    cr, ZFS_GROUP);
1674 
1675 				if (zfsvfs->z_use_fuids &&
1676 				    IS_EPHEMERAL(acl_ids->z_fgid)) {
1677 					domain = zfs_fuid_idx_domain(
1678 					    &zfsvfs->z_fuid_idx,
1679 					    FUID_INDEX(acl_ids->z_fgid));
1680 					rid = FUID_RID(acl_ids->z_fgid);
1681 					zfs_fuid_node_add(&acl_ids->z_fuidp,
1682 					    domain, rid,
1683 					    FUID_INDEX(acl_ids->z_fgid),
1684 					    acl_ids->z_fgid, ZFS_GROUP);
1685 				}
1686 			} else {
1687 				acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1688 				    ZFS_GROUP, cr, &acl_ids->z_fuidp);
1689 				gid = crgetgid(cr);
1690 			}
1691 		}
1692 	}
1693 
1694 	/*
1695 	 * If we're creating a directory, and the parent directory has the
1696 	 * set-GID bit set, set in on the new directory.
1697 	 * Otherwise, if the user is neither privileged nor a member of the
1698 	 * file's new group, clear the file's set-GID bit.
1699 	 */
1700 
1701 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1702 	    (vap->va_type == VDIR)) {
1703 		acl_ids->z_mode |= S_ISGID;
1704 	} else {
1705 		if ((acl_ids->z_mode & S_ISGID) &&
1706 		    secpolicy_vnode_setids_setgids(cr, gid) != 0)
1707 			acl_ids->z_mode &= ~S_ISGID;
1708 	}
1709 
1710 	if (acl_ids->z_aclp == NULL) {
1711 		mutex_enter(&dzp->z_acl_lock);
1712 		mutex_enter(&dzp->z_lock);
1713 		if (!(flag & IS_ROOT_NODE) &&
1714 		    (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1715 		    !(dzp->z_pflags & ZFS_XATTR)) {
1716 			VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1717 			    &paclp, B_FALSE));
1718 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1719 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1720 			inherited = B_TRUE;
1721 		} else {
1722 			acl_ids->z_aclp =
1723 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1724 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1725 		}
1726 		mutex_exit(&dzp->z_lock);
1727 		mutex_exit(&dzp->z_acl_lock);
1728 		if (need_chmod) {
1729 			acl_ids->z_aclp->z_hints |= (vap->va_type == VDIR) ?
1730 			    ZFS_ACL_AUTO_INHERIT : 0;
1731 			zfs_acl_chmod(vap->va_type, acl_ids->z_mode,
1732 			    (zfsvfs->z_acl_inherit == ZFS_ACL_RESTRICTED),
1733 			    acl_ids->z_aclp);
1734 		}
1735 	}
1736 
1737 	if (inherited || vsecp) {
1738 		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1739 		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1740 		    acl_ids->z_fuid, acl_ids->z_fgid);
1741 		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1742 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1743 	}
1744 
1745 	return (0);
1746 }
1747 
1748 /*
1749  * Free ACL and fuid_infop, but not the acl_ids structure
1750  */
1751 void
1752 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1753 {
1754 	if (acl_ids->z_aclp)
1755 		zfs_acl_free(acl_ids->z_aclp);
1756 	if (acl_ids->z_fuidp)
1757 		zfs_fuid_info_free(acl_ids->z_fuidp);
1758 	acl_ids->z_aclp = NULL;
1759 	acl_ids->z_fuidp = NULL;
1760 }
1761 
1762 boolean_t
1763 zfs_acl_ids_overquota(zfsvfs_t *zfsvfs, zfs_acl_ids_t *acl_ids)
1764 {
1765 	return (zfs_fuid_overquota(zfsvfs, B_FALSE, acl_ids->z_fuid) ||
1766 	    zfs_fuid_overquota(zfsvfs, B_TRUE, acl_ids->z_fgid));
1767 }
1768 
1769 /*
1770  * Retrieve a file's ACL
1771  */
1772 int
1773 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1774 {
1775 	zfs_acl_t	*aclp;
1776 	ulong_t		mask;
1777 	int		error;
1778 	int 		count = 0;
1779 	int		largeace = 0;
1780 
1781 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1782 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1783 
1784 	if (mask == 0)
1785 		return (SET_ERROR(ENOSYS));
1786 
1787 	if (error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr))
1788 		return (error);
1789 
1790 	mutex_enter(&zp->z_acl_lock);
1791 
1792 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1793 	if (error != 0) {
1794 		mutex_exit(&zp->z_acl_lock);
1795 		return (error);
1796 	}
1797 
1798 	/*
1799 	 * Scan ACL to determine number of ACEs
1800 	 */
1801 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1802 		void *zacep = NULL;
1803 		uint64_t who;
1804 		uint32_t access_mask;
1805 		uint16_t type, iflags;
1806 
1807 		while (zacep = zfs_acl_next_ace(aclp, zacep,
1808 		    &who, &access_mask, &iflags, &type)) {
1809 			switch (type) {
1810 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1811 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1812 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1813 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1814 				largeace++;
1815 				continue;
1816 			default:
1817 				count++;
1818 			}
1819 		}
1820 		vsecp->vsa_aclcnt = count;
1821 	} else
1822 		count = (int)aclp->z_acl_count;
1823 
1824 	if (mask & VSA_ACECNT) {
1825 		vsecp->vsa_aclcnt = count;
1826 	}
1827 
1828 	if (mask & VSA_ACE) {
1829 		size_t aclsz;
1830 
1831 		aclsz = count * sizeof (ace_t) +
1832 		    sizeof (ace_object_t) * largeace;
1833 
1834 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1835 		vsecp->vsa_aclentsz = aclsz;
1836 
1837 		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1838 			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1839 			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1840 		else {
1841 			zfs_acl_node_t *aclnode;
1842 			void *start = vsecp->vsa_aclentp;
1843 
1844 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1845 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1846 				bcopy(aclnode->z_acldata, start,
1847 				    aclnode->z_size);
1848 				start = (caddr_t)start + aclnode->z_size;
1849 			}
1850 			ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1851 			    aclp->z_acl_bytes);
1852 		}
1853 	}
1854 	if (mask & VSA_ACE_ACLFLAGS) {
1855 		vsecp->vsa_aclflags = 0;
1856 		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1857 			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1858 		if (zp->z_pflags & ZFS_ACL_PROTECTED)
1859 			vsecp->vsa_aclflags |= ACL_PROTECTED;
1860 		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1861 			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1862 	}
1863 
1864 	mutex_exit(&zp->z_acl_lock);
1865 
1866 	return (0);
1867 }
1868 
1869 int
1870 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, vtype_t obj_type,
1871     vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1872 {
1873 	zfs_acl_t *aclp;
1874 	zfs_acl_node_t *aclnode;
1875 	int aclcnt = vsecp->vsa_aclcnt;
1876 	int error;
1877 
1878 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1879 		return (SET_ERROR(EINVAL));
1880 
1881 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1882 
1883 	aclp->z_hints = 0;
1884 	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1885 	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1886 		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1887 		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1888 		    aclcnt, &aclnode->z_size)) != 0) {
1889 			zfs_acl_free(aclp);
1890 			zfs_acl_node_free(aclnode);
1891 			return (error);
1892 		}
1893 	} else {
1894 		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1895 		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1896 		    &aclnode->z_size, fuidp, cr)) != 0) {
1897 			zfs_acl_free(aclp);
1898 			zfs_acl_node_free(aclnode);
1899 			return (error);
1900 		}
1901 	}
1902 	aclp->z_acl_bytes = aclnode->z_size;
1903 	aclnode->z_ace_count = aclcnt;
1904 	aclp->z_acl_count = aclcnt;
1905 	list_insert_head(&aclp->z_acl, aclnode);
1906 
1907 	/*
1908 	 * If flags are being set then add them to z_hints
1909 	 */
1910 	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1911 		if (vsecp->vsa_aclflags & ACL_PROTECTED)
1912 			aclp->z_hints |= ZFS_ACL_PROTECTED;
1913 		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1914 			aclp->z_hints |= ZFS_ACL_DEFAULTED;
1915 		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1916 			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1917 	}
1918 
1919 	*zaclp = aclp;
1920 
1921 	return (0);
1922 }
1923 
1924 /*
1925  * Set a file's ACL
1926  */
1927 int
1928 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1929 {
1930 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1931 	zilog_t		*zilog = zfsvfs->z_log;
1932 	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1933 	dmu_tx_t	*tx;
1934 	int		error;
1935 	zfs_acl_t	*aclp;
1936 	zfs_fuid_info_t	*fuidp = NULL;
1937 	boolean_t	fuid_dirtied;
1938 	uint64_t	acl_obj;
1939 
1940 	if (mask == 0)
1941 		return (SET_ERROR(ENOSYS));
1942 
1943 	if (zp->z_pflags & ZFS_IMMUTABLE)
1944 		return (SET_ERROR(EPERM));
1945 
1946 	if (error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr))
1947 		return (error);
1948 
1949 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1950 	    &aclp);
1951 	if (error)
1952 		return (error);
1953 
1954 	/*
1955 	 * If ACL wide flags aren't being set then preserve any
1956 	 * existing flags.
1957 	 */
1958 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1959 		aclp->z_hints |=
1960 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1961 	}
1962 top:
1963 	mutex_enter(&zp->z_acl_lock);
1964 	mutex_enter(&zp->z_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 		mutex_exit(&zp->z_lock);
1998 
1999 		if (error == ERESTART) {
2000 			dmu_tx_wait(tx);
2001 			dmu_tx_abort(tx);
2002 			goto top;
2003 		}
2004 		dmu_tx_abort(tx);
2005 		zfs_acl_free(aclp);
2006 		return (error);
2007 	}
2008 
2009 	error = zfs_aclset_common(zp, aclp, cr, tx);
2010 	ASSERT(error == 0);
2011 	ASSERT(zp->z_acl_cached == NULL);
2012 	zp->z_acl_cached = aclp;
2013 
2014 	if (fuid_dirtied)
2015 		zfs_fuid_sync(zfsvfs, tx);
2016 
2017 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2018 
2019 	if (fuidp)
2020 		zfs_fuid_info_free(fuidp);
2021 	dmu_tx_commit(tx);
2022 done:
2023 	mutex_exit(&zp->z_lock);
2024 	mutex_exit(&zp->z_acl_lock);
2025 
2026 	return (error);
2027 }
2028 
2029 /*
2030  * Check accesses of interest (AoI) against attributes of the dataset
2031  * such as read-only.  Returns zero if no AoI conflict with dataset
2032  * attributes, otherwise an appropriate errno is returned.
2033  */
2034 static int
2035 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2036 {
2037 	if ((v4_mode & WRITE_MASK) &&
2038 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2039 	    (!IS_DEVVP(ZTOV(zp)) ||
2040 	    (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2041 		return (SET_ERROR(EROFS));
2042 	}
2043 
2044 	/*
2045 	 * Only check for READONLY on non-directories.
2046 	 */
2047 	if ((v4_mode & WRITE_MASK_DATA) &&
2048 	    (((ZTOV(zp)->v_type != VDIR) &&
2049 	    (zp->z_pflags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2050 	    (ZTOV(zp)->v_type == VDIR &&
2051 	    (zp->z_pflags & ZFS_IMMUTABLE)))) {
2052 		return (SET_ERROR(EPERM));
2053 	}
2054 
2055 	if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2056 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2057 		return (SET_ERROR(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 	error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2116 	if (error != 0) {
2117 		mutex_exit(&zp->z_acl_lock);
2118 		return (error);
2119 	}
2120 
2121 	ASSERT(zp->z_acl_cached);
2122 
2123 	while (acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2124 	    &iflags, &type)) {
2125 		uint32_t mask_matched;
2126 
2127 		if (!zfs_acl_valid_ace_type(type, iflags))
2128 			continue;
2129 
2130 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2131 			continue;
2132 
2133 		/* Skip ACE if it does not affect any AoI */
2134 		mask_matched = (access_mask & *working_mode);
2135 		if (!mask_matched)
2136 			continue;
2137 
2138 		entry_type = (iflags & ACE_TYPE_FLAGS);
2139 
2140 		checkit = B_FALSE;
2141 
2142 		switch (entry_type) {
2143 		case ACE_OWNER:
2144 			if (uid == fowner)
2145 				checkit = B_TRUE;
2146 			break;
2147 		case OWNING_GROUP:
2148 			who = gowner;
2149 			/*FALLTHROUGH*/
2150 		case ACE_IDENTIFIER_GROUP:
2151 			checkit = zfs_groupmember(zfsvfs, who, cr);
2152 			break;
2153 		case ACE_EVERYONE:
2154 			checkit = B_TRUE;
2155 			break;
2156 
2157 		/* USER Entry */
2158 		default:
2159 			if (entry_type == 0) {
2160 				uid_t newid;
2161 
2162 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2163 				    ZFS_ACE_USER);
2164 				if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2165 				    uid == newid)
2166 					checkit = B_TRUE;
2167 				break;
2168 			} else {
2169 				mutex_exit(&zp->z_acl_lock);
2170 				return (SET_ERROR(EIO));
2171 			}
2172 		}
2173 
2174 		if (checkit) {
2175 			if (type == DENY) {
2176 				DTRACE_PROBE3(zfs__ace__denies,
2177 				    znode_t *, zp,
2178 				    zfs_ace_hdr_t *, acep,
2179 				    uint32_t, mask_matched);
2180 				deny_mask |= mask_matched;
2181 			} else {
2182 				DTRACE_PROBE3(zfs__ace__allows,
2183 				    znode_t *, zp,
2184 				    zfs_ace_hdr_t *, acep,
2185 				    uint32_t, mask_matched);
2186 				if (anyaccess) {
2187 					mutex_exit(&zp->z_acl_lock);
2188 					return (0);
2189 				}
2190 			}
2191 			*working_mode &= ~mask_matched;
2192 		}
2193 
2194 		/* Are we done? */
2195 		if (*working_mode == 0)
2196 			break;
2197 	}
2198 
2199 	mutex_exit(&zp->z_acl_lock);
2200 
2201 	/* Put the found 'denies' back on the working mode */
2202 	if (deny_mask) {
2203 		*working_mode |= deny_mask;
2204 		return (SET_ERROR(EACCES));
2205 	} else if (*working_mode) {
2206 		return (-1);
2207 	}
2208 
2209 	return (0);
2210 }
2211 
2212 /*
2213  * Return true if any access whatsoever granted, we don't actually
2214  * care what access is granted.
2215  */
2216 boolean_t
2217 zfs_has_access(znode_t *zp, cred_t *cr)
2218 {
2219 	uint32_t have = ACE_ALL_PERMS;
2220 
2221 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2222 		uid_t owner;
2223 
2224 		owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2225 		return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2226 	}
2227 	return (B_TRUE);
2228 }
2229 
2230 static int
2231 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2232     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2233 {
2234 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2235 	int err;
2236 
2237 	*working_mode = v4_mode;
2238 	*check_privs = B_TRUE;
2239 
2240 	/*
2241 	 * Short circuit empty requests
2242 	 */
2243 	if (v4_mode == 0 || zfsvfs->z_replay) {
2244 		*working_mode = 0;
2245 		return (0);
2246 	}
2247 
2248 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2249 		*check_privs = B_FALSE;
2250 		return (err);
2251 	}
2252 
2253 	/*
2254 	 * The caller requested that the ACL check be skipped.  This
2255 	 * would only happen if the caller checked VOP_ACCESS() with a
2256 	 * 32 bit ACE mask and already had the appropriate permissions.
2257 	 */
2258 	if (skipaclchk) {
2259 		*working_mode = 0;
2260 		return (0);
2261 	}
2262 
2263 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2264 }
2265 
2266 static int
2267 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2268     cred_t *cr)
2269 {
2270 	if (*working_mode != ACE_WRITE_DATA)
2271 		return (SET_ERROR(EACCES));
2272 
2273 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2274 	    check_privs, B_FALSE, cr));
2275 }
2276 
2277 int
2278 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2279 {
2280 	boolean_t owner = B_FALSE;
2281 	boolean_t groupmbr = B_FALSE;
2282 	boolean_t is_attr;
2283 	uid_t uid = crgetuid(cr);
2284 	int error;
2285 
2286 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2287 		return (SET_ERROR(EACCES));
2288 
2289 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2290 	    (ZTOV(zdp)->v_type == VDIR));
2291 	if (is_attr)
2292 		goto slow;
2293 
2294 
2295 	mutex_enter(&zdp->z_acl_lock);
2296 
2297 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2298 		mutex_exit(&zdp->z_acl_lock);
2299 		return (0);
2300 	}
2301 
2302 	if (FUID_INDEX(zdp->z_uid) != 0 || FUID_INDEX(zdp->z_gid) != 0) {
2303 		mutex_exit(&zdp->z_acl_lock);
2304 		goto slow;
2305 	}
2306 
2307 	if (uid == zdp->z_uid) {
2308 		owner = B_TRUE;
2309 		if (zdp->z_mode & S_IXUSR) {
2310 			mutex_exit(&zdp->z_acl_lock);
2311 			return (0);
2312 		} else {
2313 			mutex_exit(&zdp->z_acl_lock);
2314 			goto slow;
2315 		}
2316 	}
2317 	if (groupmember(zdp->z_gid, cr)) {
2318 		groupmbr = B_TRUE;
2319 		if (zdp->z_mode & S_IXGRP) {
2320 			mutex_exit(&zdp->z_acl_lock);
2321 			return (0);
2322 		} else {
2323 			mutex_exit(&zdp->z_acl_lock);
2324 			goto slow;
2325 		}
2326 	}
2327 	if (!owner && !groupmbr) {
2328 		if (zdp->z_mode & S_IXOTH) {
2329 			mutex_exit(&zdp->z_acl_lock);
2330 			return (0);
2331 		}
2332 	}
2333 
2334 	mutex_exit(&zdp->z_acl_lock);
2335 
2336 slow:
2337 	DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2338 	ZFS_ENTER(zdp->z_zfsvfs);
2339 	error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2340 	ZFS_EXIT(zdp->z_zfsvfs);
2341 	return (error);
2342 }
2343 
2344 /*
2345  * Determine whether Access should be granted/denied.
2346  *
2347  * The least priv subsytem is always consulted as a basic privilege
2348  * can define any form of access.
2349  */
2350 int
2351 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2352 {
2353 	uint32_t	working_mode;
2354 	int		error;
2355 	int		is_attr;
2356 	boolean_t 	check_privs;
2357 	znode_t		*xzp;
2358 	znode_t 	*check_zp = zp;
2359 	mode_t		needed_bits;
2360 	uid_t		owner;
2361 
2362 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2363 
2364 	/*
2365 	 * If attribute then validate against base file
2366 	 */
2367 	if (is_attr) {
2368 		uint64_t	parent;
2369 
2370 		if ((error = sa_lookup(zp->z_sa_hdl,
2371 		    SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2372 		    sizeof (parent))) != 0)
2373 			return (error);
2374 
2375 		if ((error = zfs_zget(zp->z_zfsvfs,
2376 		    parent, &xzp)) != 0)	{
2377 			return (error);
2378 		}
2379 
2380 		check_zp = xzp;
2381 
2382 		/*
2383 		 * fixup mode to map to xattr perms
2384 		 */
2385 
2386 		if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2387 			mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2388 			mode |= ACE_WRITE_NAMED_ATTRS;
2389 		}
2390 
2391 		if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2392 			mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2393 			mode |= ACE_READ_NAMED_ATTRS;
2394 		}
2395 	}
2396 
2397 	owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2398 	/*
2399 	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2400 	 * in needed_bits.  Map the bits mapped by working_mode (currently
2401 	 * missing) in missing_bits.
2402 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2403 	 * needed_bits.
2404 	 */
2405 	needed_bits = 0;
2406 
2407 	working_mode = mode;
2408 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2409 	    owner == crgetuid(cr))
2410 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2411 
2412 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2413 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2414 		needed_bits |= VREAD;
2415 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2416 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2417 		needed_bits |= VWRITE;
2418 	if (working_mode & ACE_EXECUTE)
2419 		needed_bits |= VEXEC;
2420 
2421 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2422 	    &check_privs, skipaclchk, cr)) == 0) {
2423 		if (is_attr)
2424 			VN_RELE(ZTOV(xzp));
2425 		return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2426 		    needed_bits, needed_bits));
2427 	}
2428 
2429 	if (error && !check_privs) {
2430 		if (is_attr)
2431 			VN_RELE(ZTOV(xzp));
2432 		return (error);
2433 	}
2434 
2435 	if (error && (flags & V_APPEND)) {
2436 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2437 	}
2438 
2439 	if (error && check_privs) {
2440 		mode_t		checkmode = 0;
2441 
2442 		/*
2443 		 * First check for implicit owner permission on
2444 		 * read_acl/read_attributes
2445 		 */
2446 
2447 		error = 0;
2448 		ASSERT(working_mode != 0);
2449 
2450 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2451 		    owner == crgetuid(cr)))
2452 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2453 
2454 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2455 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2456 			checkmode |= VREAD;
2457 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2458 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2459 			checkmode |= VWRITE;
2460 		if (working_mode & ACE_EXECUTE)
2461 			checkmode |= VEXEC;
2462 
2463 		error = secpolicy_vnode_access2(cr, ZTOV(check_zp), owner,
2464 		    needed_bits & ~checkmode, needed_bits);
2465 
2466 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2467 			error = secpolicy_vnode_chown(cr, owner);
2468 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2469 			error = secpolicy_vnode_setdac(cr, owner);
2470 
2471 		if (error == 0 && (working_mode &
2472 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2473 			error = secpolicy_vnode_remove(cr);
2474 
2475 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2476 			error = secpolicy_vnode_chown(cr, owner);
2477 		}
2478 		if (error == 0) {
2479 			/*
2480 			 * See if any bits other than those already checked
2481 			 * for are still present.  If so then return EACCES
2482 			 */
2483 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2484 				error = SET_ERROR(EACCES);
2485 			}
2486 		}
2487 	} else if (error == 0) {
2488 		error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2489 		    needed_bits, needed_bits);
2490 	}
2491 
2492 
2493 	if (is_attr)
2494 		VN_RELE(ZTOV(xzp));
2495 
2496 	return (error);
2497 }
2498 
2499 /*
2500  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2501  * native ACL format and call zfs_zaccess()
2502  */
2503 int
2504 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2505 {
2506 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2507 }
2508 
2509 /*
2510  * Access function for secpolicy_vnode_setattr
2511  */
2512 int
2513 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2514 {
2515 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2516 
2517 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2518 }
2519 
2520 static int
2521 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2522     mode_t available_perms, cred_t *cr)
2523 {
2524 	int error;
2525 	uid_t downer;
2526 
2527 	downer = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr, ZFS_OWNER);
2528 
2529 	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2530 	    downer, available_perms, VWRITE|VEXEC);
2531 
2532 	if (error == 0)
2533 		error = zfs_sticky_remove_access(dzp, zp, cr);
2534 
2535 	return (error);
2536 }
2537 
2538 /*
2539  * Determine whether Access should be granted/deny, without
2540  * consulting least priv subsystem.
2541  *
2542  * The following chart is the recommended NFSv4 enforcement for
2543  * ability to delete an object.
2544  *
2545  *      -------------------------------------------------------
2546  *      |   Parent Dir  |           Target Object Permissions |
2547  *      |  permissions  |                                     |
2548  *      -------------------------------------------------------
2549  *      |               | ACL Allows | ACL Denies| Delete     |
2550  *      |               |  Delete    |  Delete   | unspecified|
2551  *      -------------------------------------------------------
2552  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2553  *      |  DELETE_CHILD |                                     |
2554  *      -------------------------------------------------------
2555  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2556  *      |  DELETE_CHILD |            |           |            |
2557  *      -------------------------------------------------------
2558  *      | ACL specifies |            |           |            |
2559  *      | only allow    | Permit     | Permit    | Permit     |
2560  *      | write and     |            |           |            |
2561  *      | execute       |            |           |            |
2562  *      -------------------------------------------------------
2563  *      | ACL denies    |            |           |            |
2564  *      | write and     | Permit     | Deny      | Deny       |
2565  *      | execute       |            |           |            |
2566  *      -------------------------------------------------------
2567  *         ^
2568  *         |
2569  *         No search privilege, can't even look up file?
2570  *
2571  */
2572 int
2573 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2574 {
2575 	uint32_t dzp_working_mode = 0;
2576 	uint32_t zp_working_mode = 0;
2577 	int dzp_error, zp_error;
2578 	mode_t available_perms;
2579 	boolean_t dzpcheck_privs = B_TRUE;
2580 	boolean_t zpcheck_privs = B_TRUE;
2581 
2582 	/*
2583 	 * We want specific DELETE permissions to
2584 	 * take precedence over WRITE/EXECUTE.  We don't
2585 	 * want an ACL such as this to mess us up.
2586 	 * user:joe:write_data:deny,user:joe:delete:allow
2587 	 *
2588 	 * However, deny permissions may ultimately be overridden
2589 	 * by secpolicy_vnode_access().
2590 	 *
2591 	 * We will ask for all of the necessary permissions and then
2592 	 * look at the working modes from the directory and target object
2593 	 * to determine what was found.
2594 	 */
2595 
2596 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2597 		return (SET_ERROR(EPERM));
2598 
2599 	/*
2600 	 * First row
2601 	 * If the directory permissions allow the delete, we are done.
2602 	 */
2603 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2604 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2605 		return (0);
2606 
2607 	/*
2608 	 * If target object has delete permission then we are done
2609 	 */
2610 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2611 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2612 		return (0);
2613 
2614 	ASSERT(dzp_error && zp_error);
2615 
2616 	if (!dzpcheck_privs)
2617 		return (dzp_error);
2618 	if (!zpcheck_privs)
2619 		return (zp_error);
2620 
2621 	/*
2622 	 * Second row
2623 	 *
2624 	 * If directory returns EACCES then delete_child was denied
2625 	 * due to deny delete_child.  In this case send the request through
2626 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2627 	 * since that *could* allow the delete based on write/execute permission
2628 	 * and we want delete permissions to override write/execute.
2629 	 */
2630 
2631 	if (dzp_error == EACCES)
2632 		return (secpolicy_vnode_remove(cr));
2633 
2634 	/*
2635 	 * Third Row
2636 	 * only need to see if we have write/execute on directory.
2637 	 */
2638 
2639 	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2640 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2641 
2642 	if (dzp_error != 0 && !dzpcheck_privs)
2643 		return (dzp_error);
2644 
2645 	/*
2646 	 * Fourth row
2647 	 */
2648 
2649 	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2650 	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2651 
2652 	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2653 
2654 }
2655 
2656 int
2657 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2658     znode_t *tzp, cred_t *cr)
2659 {
2660 	int add_perm;
2661 	int error;
2662 
2663 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2664 		return (SET_ERROR(EACCES));
2665 
2666 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2667 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2668 
2669 	/*
2670 	 * Rename permissions are combination of delete permission +
2671 	 * add file/subdir permission.
2672 	 */
2673 
2674 	/*
2675 	 * first make sure we do the delete portion.
2676 	 *
2677 	 * If that succeeds then check for add_file/add_subdir permissions
2678 	 */
2679 
2680 	if (error = zfs_zaccess_delete(sdzp, szp, cr))
2681 		return (error);
2682 
2683 	/*
2684 	 * If we have a tzp, see if we can delete it?
2685 	 */
2686 	if (tzp) {
2687 		if (error = zfs_zaccess_delete(tdzp, tzp, cr))
2688 			return (error);
2689 	}
2690 
2691 	/*
2692 	 * Now check for add permissions
2693 	 */
2694 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);
2695 
2696 	return (error);
2697 }
2698