1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
25 */
26
27#include <sys/types.h>
28#include <sys/param.h>
29#include <sys/time.h>
30#include <sys/systm.h>
31#include <sys/sysmacros.h>
32#include <sys/resource.h>
33#include <sys/vfs.h>
34#include <sys/vnode.h>
35#include <sys/file.h>
36#include <sys/stat.h>
37#include <sys/kmem.h>
38#include <sys/cmn_err.h>
39#include <sys/errno.h>
40#include <sys/unistd.h>
41#include <sys/sdt.h>
42#include <sys/fs/zfs.h>
43#include <sys/policy.h>
44#include <sys/zfs_znode.h>
45#include <sys/zfs_fuid.h>
46#include <sys/zfs_acl.h>
47#include <sys/zfs_dir.h>
48#include <sys/zfs_quota.h>
49#include <sys/zfs_vfsops.h>
50#include <sys/dmu.h>
51#include <sys/dnode.h>
52#include <sys/zap.h>
53#include <sys/sa.h>
54#include <acl/acl_common.h>
55
56
57#define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
58#define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
59#define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
60#define	MIN_ACE_TYPE	ALLOW
61
62#define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
63#define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
64    ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
65#define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
66    ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
67#define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
68    ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
69
70#define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
71    ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
72    ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
73    ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
74
75#define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
76#define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
77    ACE_DELETE|ACE_DELETE_CHILD)
78#define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
79
80#define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
81    ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
82
83#define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
84    ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
85
86#define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
87    ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
88
89#define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
90
91#define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
92    ZFS_ACL_PROTECTED)
93
94#define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
95    ZFS_ACL_OBJ_ACE)
96
97#define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
98
99static uint16_t
100zfs_ace_v0_get_type(void *acep)
101{
102	return (((zfs_oldace_t *)acep)->z_type);
103}
104
105static uint16_t
106zfs_ace_v0_get_flags(void *acep)
107{
108	return (((zfs_oldace_t *)acep)->z_flags);
109}
110
111static uint32_t
112zfs_ace_v0_get_mask(void *acep)
113{
114	return (((zfs_oldace_t *)acep)->z_access_mask);
115}
116
117static uint64_t
118zfs_ace_v0_get_who(void *acep)
119{
120	return (((zfs_oldace_t *)acep)->z_fuid);
121}
122
123static void
124zfs_ace_v0_set_type(void *acep, uint16_t type)
125{
126	((zfs_oldace_t *)acep)->z_type = type;
127}
128
129static void
130zfs_ace_v0_set_flags(void *acep, uint16_t flags)
131{
132	((zfs_oldace_t *)acep)->z_flags = flags;
133}
134
135static void
136zfs_ace_v0_set_mask(void *acep, uint32_t mask)
137{
138	((zfs_oldace_t *)acep)->z_access_mask = mask;
139}
140
141static void
142zfs_ace_v0_set_who(void *acep, uint64_t who)
143{
144	((zfs_oldace_t *)acep)->z_fuid = who;
145}
146
147static size_t
148zfs_ace_v0_size(void *acep)
149{
150	(void) acep;
151	return (sizeof (zfs_oldace_t));
152}
153
154static size_t
155zfs_ace_v0_abstract_size(void)
156{
157	return (sizeof (zfs_oldace_t));
158}
159
160static int
161zfs_ace_v0_mask_off(void)
162{
163	return (offsetof(zfs_oldace_t, z_access_mask));
164}
165
166static int
167zfs_ace_v0_data(void *acep, void **datap)
168{
169	(void) acep;
170	*datap = NULL;
171	return (0);
172}
173
174static const acl_ops_t zfs_acl_v0_ops = {
175	zfs_ace_v0_get_mask,
176	zfs_ace_v0_set_mask,
177	zfs_ace_v0_get_flags,
178	zfs_ace_v0_set_flags,
179	zfs_ace_v0_get_type,
180	zfs_ace_v0_set_type,
181	zfs_ace_v0_get_who,
182	zfs_ace_v0_set_who,
183	zfs_ace_v0_size,
184	zfs_ace_v0_abstract_size,
185	zfs_ace_v0_mask_off,
186	zfs_ace_v0_data
187};
188
189static uint16_t
190zfs_ace_fuid_get_type(void *acep)
191{
192	return (((zfs_ace_hdr_t *)acep)->z_type);
193}
194
195static uint16_t
196zfs_ace_fuid_get_flags(void *acep)
197{
198	return (((zfs_ace_hdr_t *)acep)->z_flags);
199}
200
201static uint32_t
202zfs_ace_fuid_get_mask(void *acep)
203{
204	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
205}
206
207static uint64_t
208zfs_ace_fuid_get_who(void *args)
209{
210	uint16_t entry_type;
211	zfs_ace_t *acep = args;
212
213	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
214
215	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
216	    entry_type == ACE_EVERYONE)
217		return (-1);
218	return (((zfs_ace_t *)acep)->z_fuid);
219}
220
221static void
222zfs_ace_fuid_set_type(void *acep, uint16_t type)
223{
224	((zfs_ace_hdr_t *)acep)->z_type = type;
225}
226
227static void
228zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
229{
230	((zfs_ace_hdr_t *)acep)->z_flags = flags;
231}
232
233static void
234zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
235{
236	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
237}
238
239static void
240zfs_ace_fuid_set_who(void *arg, uint64_t who)
241{
242	zfs_ace_t *acep = arg;
243
244	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
245
246	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
247	    entry_type == ACE_EVERYONE)
248		return;
249	acep->z_fuid = who;
250}
251
252static size_t
253zfs_ace_fuid_size(void *acep)
254{
255	zfs_ace_hdr_t *zacep = acep;
256	uint16_t entry_type;
257
258	switch (zacep->z_type) {
259	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
260	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
261	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
262	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
263		return (sizeof (zfs_object_ace_t));
264	case ALLOW:
265	case DENY:
266		entry_type =
267		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
268		if (entry_type == ACE_OWNER ||
269		    entry_type == OWNING_GROUP ||
270		    entry_type == ACE_EVERYONE)
271			return (sizeof (zfs_ace_hdr_t));
272		zfs_fallthrough;
273	default:
274		return (sizeof (zfs_ace_t));
275	}
276}
277
278static size_t
279zfs_ace_fuid_abstract_size(void)
280{
281	return (sizeof (zfs_ace_hdr_t));
282}
283
284static int
285zfs_ace_fuid_mask_off(void)
286{
287	return (offsetof(zfs_ace_hdr_t, z_access_mask));
288}
289
290static int
291zfs_ace_fuid_data(void *acep, void **datap)
292{
293	zfs_ace_t *zacep = acep;
294	zfs_object_ace_t *zobjp;
295
296	switch (zacep->z_hdr.z_type) {
297	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
298	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
299	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
300	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
301		zobjp = acep;
302		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
303		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
304	default:
305		*datap = NULL;
306		return (0);
307	}
308}
309
310static const acl_ops_t zfs_acl_fuid_ops = {
311	zfs_ace_fuid_get_mask,
312	zfs_ace_fuid_set_mask,
313	zfs_ace_fuid_get_flags,
314	zfs_ace_fuid_set_flags,
315	zfs_ace_fuid_get_type,
316	zfs_ace_fuid_set_type,
317	zfs_ace_fuid_get_who,
318	zfs_ace_fuid_set_who,
319	zfs_ace_fuid_size,
320	zfs_ace_fuid_abstract_size,
321	zfs_ace_fuid_mask_off,
322	zfs_ace_fuid_data
323};
324
325/*
326 * The following three functions are provided for compatibility with
327 * older ZPL version in order to determine if the file use to have
328 * an external ACL and what version of ACL previously existed on the
329 * file.  Would really be nice to not need this, sigh.
330 */
331uint64_t
332zfs_external_acl(znode_t *zp)
333{
334	zfs_acl_phys_t acl_phys;
335	int error;
336
337	if (zp->z_is_sa)
338		return (0);
339
340	/*
341	 * Need to deal with a potential
342	 * race where zfs_sa_upgrade could cause
343	 * z_isa_sa to change.
344	 *
345	 * If the lookup fails then the state of z_is_sa should have
346	 * changed.
347	 */
348
349	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
350	    &acl_phys, sizeof (acl_phys))) == 0)
351		return (acl_phys.z_acl_extern_obj);
352	else {
353		/*
354		 * after upgrade the SA_ZPL_ZNODE_ACL should have been
355		 * removed
356		 */
357		VERIFY(zp->z_is_sa);
358		VERIFY3S(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 */
369static int
370zfs_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
404int
405zfs_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);
432			VERIFY3S(error, ==, ENOENT);
433			return (ZFS_ACL_VERSION_FUID);
434		}
435	}
436}
437
438static int
439zfs_acl_version(int version)
440{
441	if (version < ZPL_VERSION_FUID)
442		return (ZFS_ACL_VERSION_INITIAL);
443	else
444		return (ZFS_ACL_VERSION_FUID);
445}
446
447static int
448zfs_acl_version_zp(znode_t *zp)
449{
450	return (zfs_acl_version(zp->z_zfsvfs->z_version));
451}
452
453zfs_acl_t *
454zfs_acl_alloc(int vers)
455{
456	zfs_acl_t *aclp;
457
458	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
459	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
460	    offsetof(zfs_acl_node_t, z_next));
461	aclp->z_version = vers;
462	if (vers == ZFS_ACL_VERSION_FUID)
463		aclp->z_ops = &zfs_acl_fuid_ops;
464	else
465		aclp->z_ops = &zfs_acl_v0_ops;
466	return (aclp);
467}
468
469zfs_acl_node_t *
470zfs_acl_node_alloc(size_t bytes)
471{
472	zfs_acl_node_t *aclnode;
473
474	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
475	if (bytes) {
476		aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
477		aclnode->z_allocdata = aclnode->z_acldata;
478		aclnode->z_allocsize = bytes;
479		aclnode->z_size = bytes;
480	}
481
482	return (aclnode);
483}
484
485static void
486zfs_acl_node_free(zfs_acl_node_t *aclnode)
487{
488	if (aclnode->z_allocsize)
489		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
490	kmem_free(aclnode, sizeof (zfs_acl_node_t));
491}
492
493static void
494zfs_acl_release_nodes(zfs_acl_t *aclp)
495{
496	zfs_acl_node_t *aclnode;
497
498	while ((aclnode = list_remove_head(&aclp->z_acl)))
499		zfs_acl_node_free(aclnode);
500	aclp->z_acl_count = 0;
501	aclp->z_acl_bytes = 0;
502}
503
504void
505zfs_acl_free(zfs_acl_t *aclp)
506{
507	zfs_acl_release_nodes(aclp);
508	list_destroy(&aclp->z_acl);
509	kmem_free(aclp, sizeof (zfs_acl_t));
510}
511
512static boolean_t
513zfs_acl_valid_ace_type(uint_t type, uint_t flags)
514{
515	uint16_t entry_type;
516
517	switch (type) {
518	case ALLOW:
519	case DENY:
520	case ACE_SYSTEM_AUDIT_ACE_TYPE:
521	case ACE_SYSTEM_ALARM_ACE_TYPE:
522		entry_type = flags & ACE_TYPE_FLAGS;
523		return (entry_type == ACE_OWNER ||
524		    entry_type == OWNING_GROUP ||
525		    entry_type == ACE_EVERYONE || entry_type == 0 ||
526		    entry_type == ACE_IDENTIFIER_GROUP);
527	default:
528		if (type <= MAX_ACE_TYPE)
529			return (B_TRUE);
530	}
531	return (B_FALSE);
532}
533
534static boolean_t
535zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
536{
537	/*
538	 * first check type of entry
539	 */
540
541	if (!zfs_acl_valid_ace_type(type, iflags))
542		return (B_FALSE);
543
544	switch (type) {
545	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
546	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
547	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
548	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
549		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
550			return (B_FALSE);
551		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
552	}
553
554	/*
555	 * next check inheritance level flags
556	 */
557
558	if (obj_type == VDIR &&
559	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
560		aclp->z_hints |= ZFS_INHERIT_ACE;
561
562	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
563		if ((iflags & (ACE_FILE_INHERIT_ACE|
564		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
565			return (B_FALSE);
566		}
567	}
568
569	return (B_TRUE);
570}
571
572static void *
573zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
574    uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
575{
576	zfs_acl_node_t *aclnode;
577
578	ASSERT3P(aclp, !=, NULL);
579
580	if (start == NULL) {
581		aclnode = list_head(&aclp->z_acl);
582		if (aclnode == NULL)
583			return (NULL);
584
585		aclp->z_next_ace = aclnode->z_acldata;
586		aclp->z_curr_node = aclnode;
587		aclnode->z_ace_idx = 0;
588	}
589
590	aclnode = aclp->z_curr_node;
591
592	if (aclnode == NULL)
593		return (NULL);
594
595	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
596		aclnode = list_next(&aclp->z_acl, aclnode);
597		if (aclnode == NULL)
598			return (NULL);
599		else {
600			aclp->z_curr_node = aclnode;
601			aclnode->z_ace_idx = 0;
602			aclp->z_next_ace = aclnode->z_acldata;
603		}
604	}
605
606	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
607		void *acep = aclp->z_next_ace;
608		size_t ace_size;
609
610		/*
611		 * Make sure we don't overstep our bounds
612		 */
613		ace_size = aclp->z_ops->ace_size(acep);
614
615		if (((caddr_t)acep + ace_size) >
616		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
617			return (NULL);
618		}
619
620		*iflags = aclp->z_ops->ace_flags_get(acep);
621		*type = aclp->z_ops->ace_type_get(acep);
622		*access_mask = aclp->z_ops->ace_mask_get(acep);
623		*who = aclp->z_ops->ace_who_get(acep);
624		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
625		aclnode->z_ace_idx++;
626
627		return ((void *)acep);
628	}
629	return (NULL);
630}
631
632static uintptr_t
633zfs_ace_walk(void *datap, uintptr_t cookie, int aclcnt,
634    uint16_t *flags, uint16_t *type, uint32_t *mask)
635{
636	(void) aclcnt;
637	zfs_acl_t *aclp = datap;
638	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
639	uint64_t who;
640
641	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
642	    flags, type);
643	return ((uintptr_t)acep);
644}
645
646/*
647 * Copy ACE to internal ZFS format.
648 * While processing the ACL each ACE will be validated for correctness.
649 * ACE FUIDs will be created later.
650 */
651static int
652zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
653    void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
654    zfs_fuid_info_t **fuidp, cred_t *cr)
655{
656	int i;
657	uint16_t entry_type;
658	zfs_ace_t *aceptr = z_acl;
659	ace_t *acep = datap;
660	zfs_object_ace_t *zobjacep;
661	ace_object_t *aceobjp;
662
663	for (i = 0; i != aclcnt; i++) {
664		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
665		aceptr->z_hdr.z_flags = acep->a_flags;
666		aceptr->z_hdr.z_type = acep->a_type;
667		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
668		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
669		    entry_type != ACE_EVERYONE) {
670			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
671			    cr, (entry_type == 0) ?
672			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
673		}
674
675		/*
676		 * Make sure ACE is valid
677		 */
678		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
679		    aceptr->z_hdr.z_flags) != B_TRUE)
680			return (SET_ERROR(EINVAL));
681
682		switch (acep->a_type) {
683		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
684		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
685		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
686		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
687			zobjacep = (zfs_object_ace_t *)aceptr;
688			aceobjp = (ace_object_t *)acep;
689
690			memcpy(zobjacep->z_object_type, aceobjp->a_obj_type,
691			    sizeof (aceobjp->a_obj_type));
692			memcpy(zobjacep->z_inherit_type,
693			    aceobjp->a_inherit_obj_type,
694			    sizeof (aceobjp->a_inherit_obj_type));
695			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
696			break;
697		default:
698			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
699		}
700
701		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
702		    aclp->z_ops->ace_size(aceptr));
703	}
704
705	*size = (caddr_t)aceptr - (caddr_t)z_acl;
706
707	return (0);
708}
709
710/*
711 * Copy ZFS ACEs to fixed size ace_t layout
712 */
713static void
714zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
715    void *datap, int filter)
716{
717	uint64_t who;
718	uint32_t access_mask;
719	uint16_t iflags, type;
720	zfs_ace_hdr_t *zacep = NULL;
721	ace_t *acep = datap;
722	ace_object_t *objacep;
723	zfs_object_ace_t *zobjacep;
724	size_t ace_size;
725	uint16_t entry_type;
726
727	while ((zacep = zfs_acl_next_ace(aclp, zacep,
728	    &who, &access_mask, &iflags, &type))) {
729
730		switch (type) {
731		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
732		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
733		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
734		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
735			if (filter) {
736				continue;
737			}
738			zobjacep = (zfs_object_ace_t *)zacep;
739			objacep = (ace_object_t *)acep;
740			memcpy(objacep->a_obj_type,
741			    zobjacep->z_object_type,
742			    sizeof (zobjacep->z_object_type));
743			memcpy(objacep->a_inherit_obj_type,
744			    zobjacep->z_inherit_type,
745			    sizeof (zobjacep->z_inherit_type));
746			ace_size = sizeof (ace_object_t);
747			break;
748		default:
749			ace_size = sizeof (ace_t);
750			break;
751		}
752
753		entry_type = (iflags & ACE_TYPE_FLAGS);
754		if ((entry_type != ACE_OWNER &&
755		    entry_type != OWNING_GROUP &&
756		    entry_type != ACE_EVERYONE)) {
757			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
758			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
759			    ZFS_ACE_GROUP : ZFS_ACE_USER);
760		} else {
761			acep->a_who = (uid_t)(int64_t)who;
762		}
763		acep->a_access_mask = access_mask;
764		acep->a_flags = iflags;
765		acep->a_type = type;
766		acep = (ace_t *)((caddr_t)acep + ace_size);
767	}
768}
769
770static int
771zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
772    zfs_oldace_t *z_acl, int aclcnt, size_t *size)
773{
774	int i;
775	zfs_oldace_t *aceptr = z_acl;
776
777	for (i = 0; i != aclcnt; i++, aceptr++) {
778		aceptr->z_access_mask = acep[i].a_access_mask;
779		aceptr->z_type = acep[i].a_type;
780		aceptr->z_flags = acep[i].a_flags;
781		aceptr->z_fuid = acep[i].a_who;
782		/*
783		 * Make sure ACE is valid
784		 */
785		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
786		    aceptr->z_flags) != B_TRUE)
787			return (SET_ERROR(EINVAL));
788	}
789	*size = (caddr_t)aceptr - (caddr_t)z_acl;
790	return (0);
791}
792
793/*
794 * convert old ACL format to new
795 */
796void
797zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
798{
799	zfs_oldace_t *oldaclp;
800	int i;
801	uint16_t type, iflags;
802	uint32_t access_mask;
803	uint64_t who;
804	void *cookie = NULL;
805	zfs_acl_node_t *newaclnode;
806
807	ASSERT3U(aclp->z_version, ==, ZFS_ACL_VERSION_INITIAL);
808	/*
809	 * First create the ACE in a contiguous piece of memory
810	 * for zfs_copy_ace_2_fuid().
811	 *
812	 * We only convert an ACL once, so this won't happen
813	 * everytime.
814	 */
815	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
816	    KM_SLEEP);
817	i = 0;
818	while ((cookie = zfs_acl_next_ace(aclp, cookie, &who,
819	    &access_mask, &iflags, &type))) {
820		oldaclp[i].z_flags = iflags;
821		oldaclp[i].z_type = type;
822		oldaclp[i].z_fuid = who;
823		oldaclp[i++].z_access_mask = access_mask;
824	}
825
826	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
827	    sizeof (zfs_object_ace_t));
828	aclp->z_ops = &zfs_acl_fuid_ops;
829	VERIFY0(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
830	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
831	    &newaclnode->z_size, NULL, cr));
832	newaclnode->z_ace_count = aclp->z_acl_count;
833	aclp->z_version = ZFS_ACL_VERSION;
834	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
835
836	/*
837	 * Release all previous ACL nodes
838	 */
839
840	zfs_acl_release_nodes(aclp);
841
842	list_insert_head(&aclp->z_acl, newaclnode);
843
844	aclp->z_acl_bytes = newaclnode->z_size;
845	aclp->z_acl_count = newaclnode->z_ace_count;
846
847}
848
849/*
850 * Convert unix access mask to v4 access mask
851 */
852static uint32_t
853zfs_unix_to_v4(uint32_t access_mask)
854{
855	uint32_t new_mask = 0;
856
857	if (access_mask & S_IXOTH)
858		new_mask |= ACE_EXECUTE;
859	if (access_mask & S_IWOTH)
860		new_mask |= ACE_WRITE_DATA;
861	if (access_mask & S_IROTH)
862		new_mask |= ACE_READ_DATA;
863	return (new_mask);
864}
865
866static void
867zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
868    uint16_t access_type, uint64_t fuid, uint16_t entry_type)
869{
870	uint16_t type = entry_type & ACE_TYPE_FLAGS;
871
872	aclp->z_ops->ace_mask_set(acep, access_mask);
873	aclp->z_ops->ace_type_set(acep, access_type);
874	aclp->z_ops->ace_flags_set(acep, entry_type);
875	if ((type != ACE_OWNER && type != OWNING_GROUP &&
876	    type != ACE_EVERYONE))
877		aclp->z_ops->ace_who_set(acep, fuid);
878}
879
880/*
881 * Determine mode of file based on ACL.
882 */
883uint64_t
884zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
885    uint64_t *pflags, uint64_t fuid, uint64_t fgid)
886{
887	int		entry_type;
888	mode_t		mode;
889	mode_t		seen = 0;
890	zfs_ace_hdr_t 	*acep = NULL;
891	uint64_t	who;
892	uint16_t	iflags, type;
893	uint32_t	access_mask;
894	boolean_t	an_exec_denied = B_FALSE;
895
896	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
897
898	while ((acep = zfs_acl_next_ace(aclp, acep, &who,
899	    &access_mask, &iflags, &type))) {
900
901		if (!zfs_acl_valid_ace_type(type, iflags))
902			continue;
903
904		entry_type = (iflags & ACE_TYPE_FLAGS);
905
906		/*
907		 * Skip over any inherit_only ACEs
908		 */
909		if (iflags & ACE_INHERIT_ONLY_ACE)
910			continue;
911
912		if (entry_type == ACE_OWNER || (entry_type == 0 &&
913		    who == fuid)) {
914			if ((access_mask & ACE_READ_DATA) &&
915			    (!(seen & S_IRUSR))) {
916				seen |= S_IRUSR;
917				if (type == ALLOW) {
918					mode |= S_IRUSR;
919				}
920			}
921			if ((access_mask & ACE_WRITE_DATA) &&
922			    (!(seen & S_IWUSR))) {
923				seen |= S_IWUSR;
924				if (type == ALLOW) {
925					mode |= S_IWUSR;
926				}
927			}
928			if ((access_mask & ACE_EXECUTE) &&
929			    (!(seen & S_IXUSR))) {
930				seen |= S_IXUSR;
931				if (type == ALLOW) {
932					mode |= S_IXUSR;
933				}
934			}
935		} else if (entry_type == OWNING_GROUP ||
936		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
937			if ((access_mask & ACE_READ_DATA) &&
938			    (!(seen & S_IRGRP))) {
939				seen |= S_IRGRP;
940				if (type == ALLOW) {
941					mode |= S_IRGRP;
942				}
943			}
944			if ((access_mask & ACE_WRITE_DATA) &&
945			    (!(seen & S_IWGRP))) {
946				seen |= S_IWGRP;
947				if (type == ALLOW) {
948					mode |= S_IWGRP;
949				}
950			}
951			if ((access_mask & ACE_EXECUTE) &&
952			    (!(seen & S_IXGRP))) {
953				seen |= S_IXGRP;
954				if (type == ALLOW) {
955					mode |= S_IXGRP;
956				}
957			}
958		} else if (entry_type == ACE_EVERYONE) {
959			if ((access_mask & ACE_READ_DATA)) {
960				if (!(seen & S_IRUSR)) {
961					seen |= S_IRUSR;
962					if (type == ALLOW) {
963						mode |= S_IRUSR;
964					}
965				}
966				if (!(seen & S_IRGRP)) {
967					seen |= S_IRGRP;
968					if (type == ALLOW) {
969						mode |= S_IRGRP;
970					}
971				}
972				if (!(seen & S_IROTH)) {
973					seen |= S_IROTH;
974					if (type == ALLOW) {
975						mode |= S_IROTH;
976					}
977				}
978			}
979			if ((access_mask & ACE_WRITE_DATA)) {
980				if (!(seen & S_IWUSR)) {
981					seen |= S_IWUSR;
982					if (type == ALLOW) {
983						mode |= S_IWUSR;
984					}
985				}
986				if (!(seen & S_IWGRP)) {
987					seen |= S_IWGRP;
988					if (type == ALLOW) {
989						mode |= S_IWGRP;
990					}
991				}
992				if (!(seen & S_IWOTH)) {
993					seen |= S_IWOTH;
994					if (type == ALLOW) {
995						mode |= S_IWOTH;
996					}
997				}
998			}
999			if ((access_mask & ACE_EXECUTE)) {
1000				if (!(seen & S_IXUSR)) {
1001					seen |= S_IXUSR;
1002					if (type == ALLOW) {
1003						mode |= S_IXUSR;
1004					}
1005				}
1006				if (!(seen & S_IXGRP)) {
1007					seen |= S_IXGRP;
1008					if (type == ALLOW) {
1009						mode |= S_IXGRP;
1010					}
1011				}
1012				if (!(seen & S_IXOTH)) {
1013					seen |= S_IXOTH;
1014					if (type == ALLOW) {
1015						mode |= S_IXOTH;
1016					}
1017				}
1018			}
1019		} else {
1020			/*
1021			 * Only care if this IDENTIFIER_GROUP or
1022			 * USER ACE denies execute access to someone,
1023			 * mode is not affected
1024			 */
1025			if ((access_mask & ACE_EXECUTE) && type == DENY)
1026				an_exec_denied = B_TRUE;
1027		}
1028	}
1029
1030	/*
1031	 * Failure to allow is effectively a deny, so execute permission
1032	 * is denied if it was never mentioned or if we explicitly
1033	 * weren't allowed it.
1034	 */
1035	if (!an_exec_denied &&
1036	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1037	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1038		an_exec_denied = B_TRUE;
1039
1040	if (an_exec_denied)
1041		*pflags &= ~ZFS_NO_EXECS_DENIED;
1042	else
1043		*pflags |= ZFS_NO_EXECS_DENIED;
1044
1045	return (mode);
1046}
1047
1048/*
1049 * Read an external acl object.  If the intent is to modify, always
1050 * create a new acl and leave any cached acl in place.
1051 */
1052int
1053zfs_acl_node_read(znode_t *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1054    boolean_t will_modify)
1055{
1056	zfs_acl_t	*aclp;
1057	int		aclsize;
1058	int		acl_count;
1059	zfs_acl_node_t	*aclnode;
1060	zfs_acl_phys_t	znode_acl;
1061	int		version;
1062	int		error;
1063
1064	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1065	if (zp->z_zfsvfs->z_replay == B_FALSE)
1066		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1067
1068	if (zp->z_acl_cached && !will_modify) {
1069		*aclpp = zp->z_acl_cached;
1070		return (0);
1071	}
1072
1073	version = zfs_znode_acl_version(zp);
1074
1075	if ((error = zfs_acl_znode_info(zp, &aclsize,
1076	    &acl_count, &znode_acl)) != 0) {
1077		goto done;
1078	}
1079
1080	aclp = zfs_acl_alloc(version);
1081
1082	aclp->z_acl_count = acl_count;
1083	aclp->z_acl_bytes = aclsize;
1084
1085	aclnode = zfs_acl_node_alloc(aclsize);
1086	aclnode->z_ace_count = aclp->z_acl_count;
1087	aclnode->z_size = aclsize;
1088
1089	if (!zp->z_is_sa) {
1090		if (znode_acl.z_acl_extern_obj) {
1091			error = dmu_read(zp->z_zfsvfs->z_os,
1092			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1093			    aclnode->z_acldata, DMU_READ_PREFETCH);
1094		} else {
1095			memcpy(aclnode->z_acldata, znode_acl.z_ace_data,
1096			    aclnode->z_size);
1097		}
1098	} else {
1099		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1100		    aclnode->z_acldata, aclnode->z_size);
1101	}
1102
1103	if (error != 0) {
1104		zfs_acl_free(aclp);
1105		zfs_acl_node_free(aclnode);
1106		/* convert checksum errors into IO errors */
1107		if (error == ECKSUM)
1108			error = SET_ERROR(EIO);
1109		goto done;
1110	}
1111
1112	list_insert_head(&aclp->z_acl, aclnode);
1113
1114	*aclpp = aclp;
1115	if (!will_modify)
1116		zp->z_acl_cached = aclp;
1117done:
1118	return (error);
1119}
1120
1121void
1122zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1123    boolean_t start, void *userdata)
1124{
1125	(void) buflen;
1126	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1127
1128	if (start) {
1129		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1130	} else {
1131		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1132		    cb->cb_acl_node);
1133	}
1134	ASSERT3P(cb->cb_acl_node, !=, NULL);
1135	*dataptr = cb->cb_acl_node->z_acldata;
1136	*length = cb->cb_acl_node->z_size;
1137}
1138
1139int
1140zfs_acl_chown_setattr(znode_t *zp)
1141{
1142	int error;
1143	zfs_acl_t *aclp;
1144
1145	if (zp->z_zfsvfs->z_replay == B_FALSE) {
1146		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1147		ASSERT_VOP_IN_SEQC(ZTOV(zp));
1148	}
1149	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1150
1151	if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1152		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1153		    &zp->z_pflags, zp->z_uid, zp->z_gid);
1154	return (error);
1155}
1156
1157/*
1158 * common code for setting ACLs.
1159 *
1160 * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1161 * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1162 * already checked the acl and knows whether to inherit.
1163 */
1164int
1165zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1166{
1167	int			error;
1168	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1169	dmu_object_type_t	otype;
1170	zfs_acl_locator_cb_t	locate = { 0 };
1171	uint64_t		mode;
1172	sa_bulk_attr_t		bulk[5];
1173	uint64_t		ctime[2];
1174	int			count = 0;
1175	zfs_acl_phys_t		acl_phys;
1176
1177	if (zp->z_zfsvfs->z_replay == B_FALSE) {
1178		ASSERT_VOP_IN_SEQC(ZTOV(zp));
1179	}
1180
1181	mode = zp->z_mode;
1182
1183	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1184	    zp->z_uid, zp->z_gid);
1185
1186	zp->z_mode = mode;
1187	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1188	    &mode, sizeof (mode));
1189	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1190	    &zp->z_pflags, sizeof (zp->z_pflags));
1191	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1192	    &ctime, sizeof (ctime));
1193
1194	if (zp->z_acl_cached) {
1195		zfs_acl_free(zp->z_acl_cached);
1196		zp->z_acl_cached = NULL;
1197	}
1198
1199	/*
1200	 * Upgrade needed?
1201	 */
1202	if (!zfsvfs->z_use_fuids) {
1203		otype = DMU_OT_OLDACL;
1204	} else {
1205		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1206		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1207			zfs_acl_xform(zp, aclp, cr);
1208		ASSERT3U(aclp->z_version, >=, ZFS_ACL_VERSION_FUID);
1209		otype = DMU_OT_ACL;
1210	}
1211
1212	/*
1213	 * Arrgh, we have to handle old on disk format
1214	 * as well as newer (preferred) SA format.
1215	 */
1216
1217	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1218		locate.cb_aclp = aclp;
1219		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1220		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1221		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1222		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1223	} else { /* Painful legacy way */
1224		zfs_acl_node_t *aclnode;
1225		uint64_t off = 0;
1226		uint64_t aoid;
1227
1228		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1229		    &acl_phys, sizeof (acl_phys))) != 0)
1230			return (error);
1231
1232		aoid = acl_phys.z_acl_extern_obj;
1233
1234		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1235			/*
1236			 * If ACL was previously external and we are now
1237			 * converting to new ACL format then release old
1238			 * ACL object and create a new one.
1239			 */
1240			if (aoid &&
1241			    aclp->z_version != acl_phys.z_acl_version) {
1242				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1243				if (error)
1244					return (error);
1245				aoid = 0;
1246			}
1247			if (aoid == 0) {
1248				aoid = dmu_object_alloc(zfsvfs->z_os,
1249				    otype, aclp->z_acl_bytes,
1250				    otype == DMU_OT_ACL ?
1251				    DMU_OT_SYSACL : DMU_OT_NONE,
1252				    otype == DMU_OT_ACL ?
1253				    DN_OLD_MAX_BONUSLEN : 0, tx);
1254			} else {
1255				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1256				    aoid, aclp->z_acl_bytes, 0, tx);
1257			}
1258			acl_phys.z_acl_extern_obj = aoid;
1259			for (aclnode = list_head(&aclp->z_acl); aclnode;
1260			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1261				if (aclnode->z_ace_count == 0)
1262					continue;
1263				dmu_write(zfsvfs->z_os, aoid, off,
1264				    aclnode->z_size, aclnode->z_acldata, tx);
1265				off += aclnode->z_size;
1266			}
1267		} else {
1268			void *start = acl_phys.z_ace_data;
1269			/*
1270			 * Migrating back embedded?
1271			 */
1272			if (acl_phys.z_acl_extern_obj) {
1273				error = dmu_object_free(zfsvfs->z_os,
1274				    acl_phys.z_acl_extern_obj, tx);
1275				if (error)
1276					return (error);
1277				acl_phys.z_acl_extern_obj = 0;
1278			}
1279
1280			for (aclnode = list_head(&aclp->z_acl); aclnode;
1281			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1282				if (aclnode->z_ace_count == 0)
1283					continue;
1284				memcpy(start, aclnode->z_acldata,
1285				    aclnode->z_size);
1286				start = (caddr_t)start + aclnode->z_size;
1287			}
1288		}
1289		/*
1290		 * If Old version then swap count/bytes to match old
1291		 * layout of znode_acl_phys_t.
1292		 */
1293		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1294			acl_phys.z_acl_size = aclp->z_acl_count;
1295			acl_phys.z_acl_count = aclp->z_acl_bytes;
1296		} else {
1297			acl_phys.z_acl_size = aclp->z_acl_bytes;
1298			acl_phys.z_acl_count = aclp->z_acl_count;
1299		}
1300		acl_phys.z_acl_version = aclp->z_version;
1301
1302		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1303		    &acl_phys, sizeof (acl_phys));
1304	}
1305
1306	/*
1307	 * Replace ACL wide bits, but first clear them.
1308	 */
1309	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1310
1311	zp->z_pflags |= aclp->z_hints;
1312
1313	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1314		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1315
1316	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1317	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1318}
1319
1320static void
1321zfs_acl_chmod(vtype_t vtype, uint64_t mode, boolean_t split, boolean_t trim,
1322    zfs_acl_t *aclp)
1323{
1324	void		*acep = NULL;
1325	uint64_t	who;
1326	int		new_count, new_bytes;
1327	int		ace_size;
1328	int 		entry_type;
1329	uint16_t	iflags, type;
1330	uint32_t	access_mask;
1331	zfs_acl_node_t	*newnode;
1332	size_t 		abstract_size = aclp->z_ops->ace_abstract_size();
1333	void 		*zacep;
1334	boolean_t	isdir;
1335	trivial_acl_t	masks;
1336
1337	new_count = new_bytes = 0;
1338
1339	isdir = (vtype == VDIR);
1340
1341	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1342
1343	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1344
1345	zacep = newnode->z_acldata;
1346	if (masks.allow0) {
1347		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1348		zacep = (void *)((uintptr_t)zacep + abstract_size);
1349		new_count++;
1350		new_bytes += abstract_size;
1351	}
1352	if (masks.deny1) {
1353		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1354		zacep = (void *)((uintptr_t)zacep + abstract_size);
1355		new_count++;
1356		new_bytes += abstract_size;
1357	}
1358	if (masks.deny2) {
1359		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1360		zacep = (void *)((uintptr_t)zacep + abstract_size);
1361		new_count++;
1362		new_bytes += abstract_size;
1363	}
1364
1365	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1366	    &iflags, &type))) {
1367		entry_type = (iflags & ACE_TYPE_FLAGS);
1368		/*
1369		 * ACEs used to represent the file mode may be divided
1370		 * into an equivalent pair of inherit-only and regular
1371		 * ACEs, if they are inheritable.
1372		 * Skip regular ACEs, which are replaced by the new mode.
1373		 */
1374		if (split && (entry_type == ACE_OWNER ||
1375		    entry_type == OWNING_GROUP ||
1376		    entry_type == ACE_EVERYONE)) {
1377			if (!isdir || !(iflags &
1378			    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1379				continue;
1380			/*
1381			 * We preserve owner@, group@, or @everyone
1382			 * permissions, if they are inheritable, by
1383			 * copying them to inherit_only ACEs. This
1384			 * prevents inheritable permissions from being
1385			 * altered along with the file mode.
1386			 */
1387			iflags |= ACE_INHERIT_ONLY_ACE;
1388		}
1389
1390		/*
1391		 * If this ACL has any inheritable ACEs, mark that in
1392		 * the hints (which are later masked into the pflags)
1393		 * so create knows to do inheritance.
1394		 */
1395		if (isdir && (iflags &
1396		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1397			aclp->z_hints |= ZFS_INHERIT_ACE;
1398
1399		if ((type != ALLOW && type != DENY) ||
1400		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1401			switch (type) {
1402			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1403			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1404			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1405			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1406				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1407				break;
1408			}
1409		} else {
1410			/*
1411			 * Limit permissions granted by ACEs to be no greater
1412			 * than permissions of the requested group mode.
1413			 * Applies when the "aclmode" property is set to
1414			 * "groupmask".
1415			 */
1416			if ((type == ALLOW) && trim)
1417				access_mask &= masks.group;
1418		}
1419		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1420		ace_size = aclp->z_ops->ace_size(acep);
1421		zacep = (void *)((uintptr_t)zacep + ace_size);
1422		new_count++;
1423		new_bytes += ace_size;
1424	}
1425	zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1426	zacep = (void *)((uintptr_t)zacep + abstract_size);
1427	zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1428	zacep = (void *)((uintptr_t)zacep + abstract_size);
1429	zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1430
1431	new_count += 3;
1432	new_bytes += abstract_size * 3;
1433	zfs_acl_release_nodes(aclp);
1434	aclp->z_acl_count = new_count;
1435	aclp->z_acl_bytes = new_bytes;
1436	newnode->z_ace_count = new_count;
1437	newnode->z_size = new_bytes;
1438	list_insert_tail(&aclp->z_acl, newnode);
1439}
1440
1441int
1442zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1443{
1444	int error = 0;
1445
1446	mutex_enter(&zp->z_acl_lock);
1447	if (zp->z_zfsvfs->z_replay == B_FALSE)
1448		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1449	if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1450		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1451	else
1452		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1453
1454	if (error == 0) {
1455		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1456		zfs_acl_chmod(ZTOV(zp)->v_type, mode, B_TRUE,
1457		    (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1458	}
1459	mutex_exit(&zp->z_acl_lock);
1460
1461	return (error);
1462}
1463
1464/*
1465 * Should ACE be inherited?
1466 */
1467static int
1468zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1469{
1470	int	iflags = (acep_flags & 0xf);
1471
1472	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1473		return (1);
1474	else if (iflags & ACE_FILE_INHERIT_ACE)
1475		return (!((vtype == VDIR) &&
1476		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1477	return (0);
1478}
1479
1480/*
1481 * inherit inheritable ACEs from parent
1482 */
1483static zfs_acl_t *
1484zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1485    uint64_t mode, boolean_t *need_chmod)
1486{
1487	void		*pacep = NULL;
1488	void		*acep;
1489	zfs_acl_node_t  *aclnode;
1490	zfs_acl_t	*aclp = NULL;
1491	uint64_t	who;
1492	uint32_t	access_mask;
1493	uint16_t	iflags, newflags, type;
1494	size_t		ace_size;
1495	void		*data1, *data2;
1496	size_t		data1sz, data2sz;
1497	uint_t		aclinherit;
1498	boolean_t	isdir = (vtype == VDIR);
1499	boolean_t	isreg = (vtype == VREG);
1500
1501	*need_chmod = B_TRUE;
1502
1503	aclp = zfs_acl_alloc(paclp->z_version);
1504	aclinherit = zfsvfs->z_acl_inherit;
1505	if (aclinherit == ZFS_ACL_DISCARD || vtype == VLNK)
1506		return (aclp);
1507
1508	while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1509	    &access_mask, &iflags, &type))) {
1510
1511		/*
1512		 * don't inherit bogus ACEs
1513		 */
1514		if (!zfs_acl_valid_ace_type(type, iflags))
1515			continue;
1516
1517		/*
1518		 * Check if ACE is inheritable by this vnode
1519		 */
1520		if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1521		    !zfs_ace_can_use(vtype, iflags))
1522			continue;
1523
1524		/*
1525		 * If owner@, group@, or everyone@ inheritable
1526		 * then zfs_acl_chmod() isn't needed.
1527		 */
1528		if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1529		    aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1530		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1531		    ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1532		    (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1533			*need_chmod = B_FALSE;
1534
1535		/*
1536		 * Strip inherited execute permission from file if
1537		 * not in mode
1538		 */
1539		if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1540		    !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1541			access_mask &= ~ACE_EXECUTE;
1542		}
1543
1544		/*
1545		 * Strip write_acl and write_owner from permissions
1546		 * when inheriting an ACE
1547		 */
1548		if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1549			access_mask &= ~RESTRICTED_CLEAR;
1550		}
1551
1552		ace_size = aclp->z_ops->ace_size(pacep);
1553		aclnode = zfs_acl_node_alloc(ace_size);
1554		list_insert_tail(&aclp->z_acl, aclnode);
1555		acep = aclnode->z_acldata;
1556
1557		zfs_set_ace(aclp, acep, access_mask, type,
1558		    who, iflags|ACE_INHERITED_ACE);
1559
1560		/*
1561		 * Copy special opaque data if any
1562		 */
1563		if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1564			data2sz = aclp->z_ops->ace_data(acep, &data2);
1565			VERIFY3U(data2sz, ==, data1sz);
1566			memcpy(data2, data1, data2sz);
1567		}
1568
1569		aclp->z_acl_count++;
1570		aclnode->z_ace_count++;
1571		aclp->z_acl_bytes += aclnode->z_size;
1572		newflags = aclp->z_ops->ace_flags_get(acep);
1573
1574		/*
1575		 * If ACE is not to be inherited further, or if the vnode is
1576		 * not a directory, remove all inheritance flags
1577		 */
1578		if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1579			newflags &= ~ALL_INHERIT;
1580			aclp->z_ops->ace_flags_set(acep,
1581			    newflags|ACE_INHERITED_ACE);
1582			continue;
1583		}
1584
1585		/*
1586		 * This directory has an inheritable ACE
1587		 */
1588		aclp->z_hints |= ZFS_INHERIT_ACE;
1589
1590		/*
1591		 * If only FILE_INHERIT is set then turn on
1592		 * inherit_only
1593		 */
1594		if ((iflags & (ACE_FILE_INHERIT_ACE |
1595		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1596			newflags |= ACE_INHERIT_ONLY_ACE;
1597			aclp->z_ops->ace_flags_set(acep,
1598			    newflags|ACE_INHERITED_ACE);
1599		} else {
1600			newflags &= ~ACE_INHERIT_ONLY_ACE;
1601			aclp->z_ops->ace_flags_set(acep,
1602			    newflags|ACE_INHERITED_ACE);
1603		}
1604	}
1605	if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1606	    aclp->z_acl_count != 0) {
1607		*need_chmod = B_FALSE;
1608	}
1609
1610	return (aclp);
1611}
1612
1613/*
1614 * Create file system object initial permissions
1615 * including inheritable ACEs.
1616 * Also, create FUIDs for owner and group.
1617 */
1618int
1619zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1620    vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids, zidmap_t *mnt_ns)
1621{
1622	int		error;
1623	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1624	zfs_acl_t	*paclp;
1625	gid_t		gid;
1626	boolean_t	need_chmod = B_TRUE;
1627	boolean_t	trim = B_FALSE;
1628	boolean_t	inherited = B_FALSE;
1629
1630	if ((flag & IS_ROOT_NODE) == 0) {
1631		if (zfsvfs->z_replay == B_FALSE)
1632			ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
1633	} else
1634		ASSERT3P(dzp->z_vnode, ==, NULL);
1635	memset(acl_ids, 0, sizeof (zfs_acl_ids_t));
1636	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1637
1638	if (vsecp)
1639		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1640		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1641			return (error);
1642	/*
1643	 * Determine uid and gid.
1644	 */
1645	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1646	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1647		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1648		    (uint64_t)vap->va_uid, cr,
1649		    ZFS_OWNER, &acl_ids->z_fuidp);
1650		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1651		    (uint64_t)vap->va_gid, cr,
1652		    ZFS_GROUP, &acl_ids->z_fuidp);
1653		gid = vap->va_gid;
1654	} else {
1655		uid_t id = crgetuid(cr);
1656		if (IS_EPHEMERAL(id))
1657			id = UID_NOBODY;
1658		acl_ids->z_fuid = (uint64_t)id;
1659		acl_ids->z_fgid = 0;
1660		if (vap->va_mask & AT_GID)  {
1661			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1662			    (uint64_t)vap->va_gid,
1663			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1664			gid = vap->va_gid;
1665			if (acl_ids->z_fgid != dzp->z_gid &&
1666			    !groupmember(vap->va_gid, cr) &&
1667			    secpolicy_vnode_create_gid(cr) != 0)
1668				acl_ids->z_fgid = 0;
1669		}
1670		if (acl_ids->z_fgid == 0) {
1671			const char	*domain;
1672			uint32_t	rid;
1673
1674			acl_ids->z_fgid = dzp->z_gid;
1675			gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1676			    cr, ZFS_GROUP);
1677
1678			if (zfsvfs->z_use_fuids &&
1679			    IS_EPHEMERAL(acl_ids->z_fgid)) {
1680				domain =
1681				    zfs_fuid_idx_domain(&zfsvfs->z_fuid_idx,
1682				    FUID_INDEX(acl_ids->z_fgid));
1683				rid = FUID_RID(acl_ids->z_fgid);
1684				zfs_fuid_node_add(&acl_ids->z_fuidp,
1685				    domain, rid, FUID_INDEX(acl_ids->z_fgid),
1686				    acl_ids->z_fgid, ZFS_GROUP);
1687			}
1688		}
1689	}
1690
1691	/*
1692	 * If we're creating a directory, and the parent directory has the
1693	 * set-GID bit set, set in on the new directory.
1694	 * Otherwise, if the user is neither privileged nor a member of the
1695	 * file's new group, clear the file's set-GID bit.
1696	 */
1697
1698	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1699	    (vap->va_type == VDIR)) {
1700		acl_ids->z_mode |= S_ISGID;
1701	} else {
1702		if ((acl_ids->z_mode & S_ISGID) &&
1703		    secpolicy_vnode_setids_setgids(ZTOV(dzp), cr, gid) != 0)
1704			acl_ids->z_mode &= ~S_ISGID;
1705	}
1706
1707	if (acl_ids->z_aclp == NULL) {
1708		mutex_enter(&dzp->z_acl_lock);
1709		if (!(flag & IS_ROOT_NODE) &&
1710		    (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1711		    !(dzp->z_pflags & ZFS_XATTR)) {
1712			VERIFY0(zfs_acl_node_read(dzp, B_TRUE,
1713			    &paclp, B_FALSE));
1714			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1715			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1716			inherited = B_TRUE;
1717		} else {
1718			acl_ids->z_aclp =
1719			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1720			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1721		}
1722		mutex_exit(&dzp->z_acl_lock);
1723
1724		if (need_chmod) {
1725			if (vap->va_type == VDIR)
1726				acl_ids->z_aclp->z_hints |=
1727				    ZFS_ACL_AUTO_INHERIT;
1728
1729			if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1730			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1731			    zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1732				trim = B_TRUE;
1733			zfs_acl_chmod(vap->va_type, acl_ids->z_mode, B_FALSE,
1734			    trim, acl_ids->z_aclp);
1735		}
1736	}
1737
1738	if (inherited || vsecp) {
1739		acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1740		    acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1741		    acl_ids->z_fuid, acl_ids->z_fgid);
1742		if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1743			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1744	}
1745
1746	return (0);
1747}
1748
1749/*
1750 * Free ACL and fuid_infop, but not the acl_ids structure
1751 */
1752void
1753zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1754{
1755	if (acl_ids->z_aclp)
1756		zfs_acl_free(acl_ids->z_aclp);
1757	if (acl_ids->z_fuidp)
1758		zfs_fuid_info_free(acl_ids->z_fuidp);
1759	acl_ids->z_aclp = NULL;
1760	acl_ids->z_fuidp = NULL;
1761}
1762
1763boolean_t
1764zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1765{
1766	return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1767	    zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1768	    (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1769	    zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1770}
1771
1772/*
1773 * Retrieve a file's ACL
1774 */
1775int
1776zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1777{
1778	zfs_acl_t	*aclp;
1779	ulong_t		mask;
1780	int		error;
1781	int 		count = 0;
1782	int		largeace = 0;
1783
1784	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1785	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1786
1787	if (mask == 0)
1788		return (SET_ERROR(ENOSYS));
1789
1790	if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr, NULL)))
1791		return (error);
1792
1793	mutex_enter(&zp->z_acl_lock);
1794
1795	if (zp->z_zfsvfs->z_replay == B_FALSE)
1796		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1797	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1798	if (error != 0) {
1799		mutex_exit(&zp->z_acl_lock);
1800		return (error);
1801	}
1802
1803	/*
1804	 * Scan ACL to determine number of ACEs
1805	 */
1806	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1807		void *zacep = NULL;
1808		uint64_t who;
1809		uint32_t access_mask;
1810		uint16_t type, iflags;
1811
1812		while ((zacep = zfs_acl_next_ace(aclp, zacep,
1813		    &who, &access_mask, &iflags, &type))) {
1814			switch (type) {
1815			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1816			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1817			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1818			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1819				largeace++;
1820				continue;
1821			default:
1822				count++;
1823			}
1824		}
1825		vsecp->vsa_aclcnt = count;
1826	} else
1827		count = (int)aclp->z_acl_count;
1828
1829	if (mask & VSA_ACECNT) {
1830		vsecp->vsa_aclcnt = count;
1831	}
1832
1833	if (mask & VSA_ACE) {
1834		size_t aclsz;
1835
1836		aclsz = count * sizeof (ace_t) +
1837		    sizeof (ace_object_t) * largeace;
1838
1839		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1840		vsecp->vsa_aclentsz = aclsz;
1841
1842		if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1843			zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1844			    vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1845		else {
1846			zfs_acl_node_t *aclnode;
1847			void *start = vsecp->vsa_aclentp;
1848
1849			for (aclnode = list_head(&aclp->z_acl); aclnode;
1850			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1851				memcpy(start, aclnode->z_acldata,
1852				    aclnode->z_size);
1853				start = (caddr_t)start + aclnode->z_size;
1854			}
1855			ASSERT3U((caddr_t)start - (caddr_t)vsecp->vsa_aclentp,
1856			    ==, aclp->z_acl_bytes);
1857		}
1858	}
1859	if (mask & VSA_ACE_ACLFLAGS) {
1860		vsecp->vsa_aclflags = 0;
1861		if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1862			vsecp->vsa_aclflags |= ACL_DEFAULTED;
1863		if (zp->z_pflags & ZFS_ACL_PROTECTED)
1864			vsecp->vsa_aclflags |= ACL_PROTECTED;
1865		if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1866			vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1867	}
1868
1869	mutex_exit(&zp->z_acl_lock);
1870
1871	return (0);
1872}
1873
1874int
1875zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_type,
1876    vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1877{
1878	zfs_acl_t *aclp;
1879	zfs_acl_node_t *aclnode;
1880	int aclcnt = vsecp->vsa_aclcnt;
1881	int error;
1882
1883	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1884		return (SET_ERROR(EINVAL));
1885
1886	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1887
1888	aclp->z_hints = 0;
1889	aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1890	if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1891		if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1892		    (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1893		    aclcnt, &aclnode->z_size)) != 0) {
1894			zfs_acl_free(aclp);
1895			zfs_acl_node_free(aclnode);
1896			return (error);
1897		}
1898	} else {
1899		if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1900		    vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1901		    &aclnode->z_size, fuidp, cr)) != 0) {
1902			zfs_acl_free(aclp);
1903			zfs_acl_node_free(aclnode);
1904			return (error);
1905		}
1906	}
1907	aclp->z_acl_bytes = aclnode->z_size;
1908	aclnode->z_ace_count = aclcnt;
1909	aclp->z_acl_count = aclcnt;
1910	list_insert_head(&aclp->z_acl, aclnode);
1911
1912	/*
1913	 * If flags are being set then add them to z_hints
1914	 */
1915	if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1916		if (vsecp->vsa_aclflags & ACL_PROTECTED)
1917			aclp->z_hints |= ZFS_ACL_PROTECTED;
1918		if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1919			aclp->z_hints |= ZFS_ACL_DEFAULTED;
1920		if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1921			aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1922	}
1923
1924	*zaclp = aclp;
1925
1926	return (0);
1927}
1928
1929/*
1930 * Set a file's ACL
1931 */
1932int
1933zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1934{
1935	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
1936	zilog_t		*zilog = zfsvfs->z_log;
1937	ulong_t		mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1938	dmu_tx_t	*tx;
1939	int		error;
1940	zfs_acl_t	*aclp;
1941	zfs_fuid_info_t	*fuidp = NULL;
1942	boolean_t	fuid_dirtied;
1943	uint64_t	acl_obj;
1944
1945	if (zp->z_zfsvfs->z_replay == B_FALSE)
1946		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1947	if (mask == 0)
1948		return (SET_ERROR(ENOSYS));
1949
1950	if (zp->z_pflags & ZFS_IMMUTABLE)
1951		return (SET_ERROR(EPERM));
1952
1953	if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, NULL)))
1954		return (error);
1955
1956	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1957	    &aclp);
1958	if (error)
1959		return (error);
1960
1961	/*
1962	 * If ACL wide flags aren't being set then preserve any
1963	 * existing flags.
1964	 */
1965	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1966		aclp->z_hints |=
1967		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1968	}
1969top:
1970	mutex_enter(&zp->z_acl_lock);
1971
1972	tx = dmu_tx_create(zfsvfs->z_os);
1973
1974	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1975
1976	fuid_dirtied = zfsvfs->z_fuid_dirty;
1977	if (fuid_dirtied)
1978		zfs_fuid_txhold(zfsvfs, tx);
1979
1980	/*
1981	 * If old version and ACL won't fit in bonus and we aren't
1982	 * upgrading then take out necessary DMU holds
1983	 */
1984
1985	if ((acl_obj = zfs_external_acl(zp)) != 0) {
1986		if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
1987		    zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
1988			dmu_tx_hold_free(tx, acl_obj, 0,
1989			    DMU_OBJECT_END);
1990			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1991			    aclp->z_acl_bytes);
1992		} else {
1993			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
1994		}
1995	} else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1996		dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
1997	}
1998
1999	zfs_sa_upgrade_txholds(tx, zp);
2000	error = dmu_tx_assign(tx, TXG_NOWAIT);
2001	if (error) {
2002		mutex_exit(&zp->z_acl_lock);
2003
2004		if (error == ERESTART) {
2005			dmu_tx_wait(tx);
2006			dmu_tx_abort(tx);
2007			goto top;
2008		}
2009		dmu_tx_abort(tx);
2010		zfs_acl_free(aclp);
2011		return (error);
2012	}
2013
2014	error = zfs_aclset_common(zp, aclp, cr, tx);
2015	ASSERT0(error);
2016	ASSERT3P(zp->z_acl_cached, ==, NULL);
2017	zp->z_acl_cached = aclp;
2018
2019	if (fuid_dirtied)
2020		zfs_fuid_sync(zfsvfs, tx);
2021
2022	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2023
2024	if (fuidp)
2025		zfs_fuid_info_free(fuidp);
2026	dmu_tx_commit(tx);
2027	mutex_exit(&zp->z_acl_lock);
2028
2029	return (error);
2030}
2031
2032/*
2033 * Check accesses of interest (AoI) against attributes of the dataset
2034 * such as read-only.  Returns zero if no AoI conflict with dataset
2035 * attributes, otherwise an appropriate errno is returned.
2036 */
2037static int
2038zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2039{
2040	if ((v4_mode & WRITE_MASK) &&
2041	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2042	    (!IS_DEVVP(ZTOV(zp)) || (v4_mode & WRITE_MASK_ATTRS))) {
2043		return (SET_ERROR(EROFS));
2044	}
2045
2046	/*
2047	 * Intentionally allow ZFS_READONLY through here.
2048	 * See zfs_zaccess_common().
2049	 */
2050	if ((v4_mode & WRITE_MASK_DATA) &&
2051	    (zp->z_pflags & ZFS_IMMUTABLE)) {
2052		return (SET_ERROR(EPERM));
2053	}
2054
2055	/*
2056	 * In FreeBSD we allow to modify directory's content is ZFS_NOUNLINK
2057	 * (sunlnk) is set. We just don't allow directory removal, which is
2058	 * handled in zfs_zaccess_delete().
2059	 */
2060	if ((v4_mode & ACE_DELETE) &&
2061	    (zp->z_pflags & ZFS_NOUNLINK)) {
2062		return (EPERM);
2063	}
2064
2065	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2066	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2067		return (SET_ERROR(EACCES));
2068	}
2069
2070	return (0);
2071}
2072
2073/*
2074 * The primary usage of this function is to loop through all of the
2075 * ACEs in the znode, determining what accesses of interest (AoI) to
2076 * the caller are allowed or denied.  The AoI are expressed as bits in
2077 * the working_mode parameter.  As each ACE is processed, bits covered
2078 * by that ACE are removed from the working_mode.  This removal
2079 * facilitates two things.  The first is that when the working mode is
2080 * empty (= 0), we know we've looked at all the AoI. The second is
2081 * that the ACE interpretation rules don't allow a later ACE to undo
2082 * something granted or denied by an earlier ACE.  Removing the
2083 * discovered access or denial enforces this rule.  At the end of
2084 * processing the ACEs, all AoI that were found to be denied are
2085 * placed into the working_mode, giving the caller a mask of denied
2086 * accesses.  Returns:
2087 *	0		if all AoI granted
2088 *	EACCESS 	if the denied mask is non-zero
2089 *	other error	if abnormal failure (e.g., IO error)
2090 *
2091 * A secondary usage of the function is to determine if any of the
2092 * AoI are granted.  If an ACE grants any access in
2093 * the working_mode, we immediately short circuit out of the function.
2094 * This mode is chosen by setting anyaccess to B_TRUE.  The
2095 * working_mode is not a denied access mask upon exit if the function
2096 * is used in this manner.
2097 */
2098static int
2099zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2100    boolean_t anyaccess, cred_t *cr)
2101{
2102	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2103	zfs_acl_t	*aclp;
2104	int		error;
2105	uid_t		uid = crgetuid(cr);
2106	uint64_t 	who;
2107	uint16_t	type, iflags;
2108	uint16_t	entry_type;
2109	uint32_t	access_mask;
2110	uint32_t	deny_mask = 0;
2111	zfs_ace_hdr_t	*acep = NULL;
2112	boolean_t	checkit;
2113	uid_t		gowner;
2114	uid_t		fowner;
2115
2116	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2117
2118	mutex_enter(&zp->z_acl_lock);
2119
2120	if (zp->z_zfsvfs->z_replay == B_FALSE)
2121		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
2122	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
2123	if (error != 0) {
2124		mutex_exit(&zp->z_acl_lock);
2125		return (error);
2126	}
2127
2128	ASSERT3P(zp->z_acl_cached, !=, NULL);
2129
2130	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2131	    &iflags, &type))) {
2132		uint32_t mask_matched;
2133
2134		if (!zfs_acl_valid_ace_type(type, iflags))
2135			continue;
2136
2137		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2138			continue;
2139
2140		/* Skip ACE if it does not affect any AoI */
2141		mask_matched = (access_mask & *working_mode);
2142		if (!mask_matched)
2143			continue;
2144
2145		entry_type = (iflags & ACE_TYPE_FLAGS);
2146
2147		checkit = B_FALSE;
2148
2149		switch (entry_type) {
2150		case ACE_OWNER:
2151			if (uid == fowner)
2152				checkit = B_TRUE;
2153			break;
2154		case OWNING_GROUP:
2155			who = gowner;
2156			zfs_fallthrough;
2157		case ACE_IDENTIFIER_GROUP:
2158			checkit = zfs_groupmember(zfsvfs, who, cr);
2159			break;
2160		case ACE_EVERYONE:
2161			checkit = B_TRUE;
2162			break;
2163
2164		/* USER Entry */
2165		default:
2166			if (entry_type == 0) {
2167				uid_t newid;
2168
2169				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2170				    ZFS_ACE_USER);
2171				if (newid !=  UID_NOBODY &&
2172				    uid == newid)
2173					checkit = B_TRUE;
2174				break;
2175			} else {
2176				mutex_exit(&zp->z_acl_lock);
2177				return (SET_ERROR(EIO));
2178			}
2179		}
2180
2181		if (checkit) {
2182			if (type == DENY) {
2183				DTRACE_PROBE3(zfs__ace__denies,
2184				    znode_t *, zp,
2185				    zfs_ace_hdr_t *, acep,
2186				    uint32_t, mask_matched);
2187				deny_mask |= mask_matched;
2188			} else {
2189				DTRACE_PROBE3(zfs__ace__allows,
2190				    znode_t *, zp,
2191				    zfs_ace_hdr_t *, acep,
2192				    uint32_t, mask_matched);
2193				if (anyaccess) {
2194					mutex_exit(&zp->z_acl_lock);
2195					return (0);
2196				}
2197			}
2198			*working_mode &= ~mask_matched;
2199		}
2200
2201		/* Are we done? */
2202		if (*working_mode == 0)
2203			break;
2204	}
2205
2206	mutex_exit(&zp->z_acl_lock);
2207
2208	/* Put the found 'denies' back on the working mode */
2209	if (deny_mask) {
2210		*working_mode |= deny_mask;
2211		return (SET_ERROR(EACCES));
2212	} else if (*working_mode) {
2213		return (-1);
2214	}
2215
2216	return (0);
2217}
2218
2219/*
2220 * Return true if any access whatsoever granted, we don't actually
2221 * care what access is granted.
2222 */
2223boolean_t
2224zfs_has_access(znode_t *zp, cred_t *cr)
2225{
2226	uint32_t have = ACE_ALL_PERMS;
2227
2228	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2229		uid_t owner;
2230
2231		owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2232		return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2233	}
2234	return (B_TRUE);
2235}
2236
2237static int
2238zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2239    boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2240{
2241	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2242	int err;
2243
2244	*working_mode = v4_mode;
2245	*check_privs = B_TRUE;
2246
2247	/*
2248	 * Short circuit empty requests
2249	 */
2250	if (v4_mode == 0 || zfsvfs->z_replay) {
2251		*working_mode = 0;
2252		return (0);
2253	}
2254
2255	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2256		*check_privs = B_FALSE;
2257		return (err);
2258	}
2259
2260	/*
2261	 * The caller requested that the ACL check be skipped.  This
2262	 * would only happen if the caller checked VOP_ACCESS() with a
2263	 * 32 bit ACE mask and already had the appropriate permissions.
2264	 */
2265	if (skipaclchk) {
2266		*working_mode = 0;
2267		return (0);
2268	}
2269
2270	/*
2271	 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2272	 * When that flag is set, we should behave as if write access
2273	 * were not granted by anything in the ACL.  In particular:
2274	 * We _must_ allow writes after opening the file r/w, then
2275	 * setting the DOS R/O attribute, and writing some more.
2276	 * (Similar to how you can write after fchmod(fd, 0444).)
2277	 *
2278	 * Therefore ZFS_READONLY is ignored in the dataset check
2279	 * above, and checked here as if part of the ACL check.
2280	 * Also note: DOS R/O is ignored for directories.
2281	 */
2282	if ((v4_mode & WRITE_MASK_DATA) &&
2283	    (ZTOV(zp)->v_type != VDIR) &&
2284	    (zp->z_pflags & ZFS_READONLY)) {
2285		return (SET_ERROR(EPERM));
2286	}
2287
2288	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2289}
2290
2291static int
2292zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2293    cred_t *cr)
2294{
2295	if (*working_mode != ACE_WRITE_DATA)
2296		return (SET_ERROR(EACCES));
2297
2298	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2299	    check_privs, B_FALSE, cr));
2300}
2301
2302/*
2303 * Check if VEXEC is allowed.
2304 *
2305 * This routine is based on zfs_fastaccesschk_execute which has slowpath
2306 * calling zfs_zaccess. This would be incorrect on FreeBSD (see
2307 * zfs_freebsd_access for the difference). Thus this variant let's the
2308 * caller handle the slowpath (if necessary).
2309 *
2310 * On top of that we perform a lockless check for ZFS_NO_EXECS_DENIED.
2311 *
2312 * Safe access to znode_t is provided by the vnode lock.
2313 */
2314int
2315zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2316{
2317	boolean_t is_attr;
2318
2319	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2320		return (1);
2321
2322	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2323	    (ZTOV(zdp)->v_type == VDIR));
2324	if (is_attr)
2325		return (1);
2326
2327	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED)
2328		return (0);
2329
2330	return (1);
2331}
2332
2333
2334/*
2335 * Determine whether Access should be granted/denied.
2336 *
2337 * The least priv subsystem is always consulted as a basic privilege
2338 * can define any form of access.
2339 */
2340int
2341zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr,
2342    zidmap_t *mnt_ns)
2343{
2344	uint32_t	working_mode;
2345	int		error;
2346	int		is_attr;
2347	boolean_t 	check_privs;
2348	znode_t		*xzp = NULL;
2349	znode_t 	*check_zp = zp;
2350	mode_t		needed_bits;
2351	uid_t		owner;
2352
2353	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2354
2355	/*
2356	 * In FreeBSD, we don't care about permissions of individual ADS.
2357	 * Note that not checking them is not just an optimization - without
2358	 * this shortcut, EA operations may bogusly fail with EACCES.
2359	 */
2360	if (zp->z_pflags & ZFS_XATTR)
2361		return (0);
2362
2363	owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2364
2365	/*
2366	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2367	 * in needed_bits.  Map the bits mapped by working_mode (currently
2368	 * missing) in missing_bits.
2369	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2370	 * needed_bits.
2371	 */
2372	needed_bits = 0;
2373
2374	working_mode = mode;
2375	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2376	    owner == crgetuid(cr))
2377		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2378
2379	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2380	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2381		needed_bits |= VREAD;
2382	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2383	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2384		needed_bits |= VWRITE;
2385	if (working_mode & ACE_EXECUTE)
2386		needed_bits |= VEXEC;
2387
2388	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2389	    &check_privs, skipaclchk, cr)) == 0) {
2390		if (is_attr)
2391			VN_RELE(ZTOV(xzp));
2392		return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2393		    needed_bits, needed_bits));
2394	}
2395
2396	if (error && !check_privs) {
2397		if (is_attr)
2398			VN_RELE(ZTOV(xzp));
2399		return (error);
2400	}
2401
2402	if (error && (flags & V_APPEND)) {
2403		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2404	}
2405
2406	if (error && check_privs) {
2407		mode_t		checkmode = 0;
2408		vnode_t *check_vp = ZTOV(check_zp);
2409
2410		/*
2411		 * First check for implicit owner permission on
2412		 * read_acl/read_attributes
2413		 */
2414
2415		ASSERT3U(working_mode, !=, 0);
2416
2417		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2418		    owner == crgetuid(cr)))
2419			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2420
2421		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2422		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2423			checkmode |= VREAD;
2424		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2425		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2426			checkmode |= VWRITE;
2427		if (working_mode & ACE_EXECUTE)
2428			checkmode |= VEXEC;
2429
2430		error = secpolicy_vnode_access2(cr, check_vp, owner,
2431		    needed_bits & ~checkmode, needed_bits);
2432
2433		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2434			error = secpolicy_vnode_chown(check_vp, cr, owner);
2435		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2436			error = secpolicy_vnode_setdac(check_vp, cr, owner);
2437
2438		if (error == 0 && (working_mode &
2439		    (ACE_DELETE|ACE_DELETE_CHILD)))
2440			error = secpolicy_vnode_remove(check_vp, cr);
2441
2442		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2443			error = secpolicy_vnode_chown(check_vp, cr, owner);
2444		}
2445		if (error == 0) {
2446			/*
2447			 * See if any bits other than those already checked
2448			 * for are still present.  If so then return EACCES
2449			 */
2450			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2451				error = SET_ERROR(EACCES);
2452			}
2453		}
2454	} else if (error == 0) {
2455		error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2456		    needed_bits, needed_bits);
2457	}
2458
2459
2460	if (is_attr)
2461		VN_RELE(ZTOV(xzp));
2462
2463	return (error);
2464}
2465
2466/*
2467 * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2468 * NFSv4-style ZFS ACL format and call zfs_zaccess()
2469 */
2470int
2471zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr,
2472    zidmap_t *mnt_ns)
2473{
2474	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr,
2475	    mnt_ns));
2476}
2477
2478/*
2479 * Access function for secpolicy_vnode_setattr
2480 */
2481int
2482zfs_zaccess_unix(void *zp, int mode, cred_t *cr)
2483{
2484	int v4_mode = zfs_unix_to_v4(mode >> 6);
2485
2486	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr, NULL));
2487}
2488
2489static int
2490zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2491    mode_t available_perms, cred_t *cr)
2492{
2493	int error;
2494	uid_t downer;
2495
2496	downer = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr, ZFS_OWNER);
2497
2498	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2499	    downer, available_perms, VWRITE|VEXEC);
2500
2501	if (error == 0)
2502		error = zfs_sticky_remove_access(dzp, zp, cr);
2503
2504	return (error);
2505}
2506
2507/*
2508 * Determine whether Access should be granted/deny, without
2509 * consulting least priv subsystem.
2510 *
2511 * The following chart is the recommended NFSv4 enforcement for
2512 * ability to delete an object.
2513 *
2514 *      -------------------------------------------------------
2515 *      |   Parent Dir  |           Target Object Permissions |
2516 *      |  permissions  |                                     |
2517 *      -------------------------------------------------------
2518 *      |               | ACL Allows | ACL Denies| Delete     |
2519 *      |               |  Delete    |  Delete   | unspecified|
2520 *      -------------------------------------------------------
2521 *      |  ACL Allows   | Permit     | Permit    | Permit     |
2522 *      |  DELETE_CHILD |                                     |
2523 *      -------------------------------------------------------
2524 *      |  ACL Denies   | Permit     | Deny      | Deny       |
2525 *      |  DELETE_CHILD |            |           |            |
2526 *      -------------------------------------------------------
2527 *      | ACL specifies |            |           |            |
2528 *      | only allow    | Permit     | Permit    | Permit     |
2529 *      | write and     |            |           |            |
2530 *      | execute       |            |           |            |
2531 *      -------------------------------------------------------
2532 *      | ACL denies    |            |           |            |
2533 *      | write and     | Permit     | Deny      | Deny       |
2534 *      | execute       |            |           |            |
2535 *      -------------------------------------------------------
2536 *         ^
2537 *         |
2538 *         No search privilege, can't even look up file?
2539 *
2540 */
2541int
2542zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr, zidmap_t *mnt_ns)
2543{
2544	uint32_t dzp_working_mode = 0;
2545	uint32_t zp_working_mode = 0;
2546	int dzp_error, zp_error;
2547	mode_t available_perms;
2548	boolean_t dzpcheck_privs = B_TRUE;
2549	boolean_t zpcheck_privs = B_TRUE;
2550
2551	/*
2552	 * We want specific DELETE permissions to
2553	 * take precedence over WRITE/EXECUTE.  We don't
2554	 * want an ACL such as this to mess us up.
2555	 * user:joe:write_data:deny,user:joe:delete:allow
2556	 *
2557	 * However, deny permissions may ultimately be overridden
2558	 * by secpolicy_vnode_access().
2559	 *
2560	 * We will ask for all of the necessary permissions and then
2561	 * look at the working modes from the directory and target object
2562	 * to determine what was found.
2563	 */
2564
2565	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2566		return (SET_ERROR(EPERM));
2567
2568	/*
2569	 * First row
2570	 * If the directory permissions allow the delete, we are done.
2571	 */
2572	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2573	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2574		return (0);
2575
2576	/*
2577	 * If target object has delete permission then we are done
2578	 */
2579	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2580	    &zpcheck_privs, B_FALSE, cr)) == 0)
2581		return (0);
2582
2583	ASSERT(dzp_error);
2584	ASSERT(zp_error);
2585
2586	if (!dzpcheck_privs)
2587		return (dzp_error);
2588	if (!zpcheck_privs)
2589		return (zp_error);
2590
2591	/*
2592	 * Second row
2593	 *
2594	 * If directory returns EACCES then delete_child was denied
2595	 * due to deny delete_child.  In this case send the request through
2596	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2597	 * since that *could* allow the delete based on write/execute permission
2598	 * and we want delete permissions to override write/execute.
2599	 */
2600
2601	if (dzp_error == EACCES) {
2602		/* XXXPJD: s/dzp/zp/ ? */
2603		return (secpolicy_vnode_remove(ZTOV(dzp), cr));
2604	}
2605	/*
2606	 * Third Row
2607	 * only need to see if we have write/execute on directory.
2608	 */
2609
2610	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2611	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2612
2613	if (dzp_error != 0 && !dzpcheck_privs)
2614		return (dzp_error);
2615
2616	/*
2617	 * Fourth row
2618	 */
2619
2620	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2621	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2622
2623	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2624
2625}
2626
2627int
2628zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2629    znode_t *tzp, cred_t *cr, zidmap_t *mnt_ns)
2630{
2631	int add_perm;
2632	int error;
2633
2634	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2635		return (SET_ERROR(EACCES));
2636
2637	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2638	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2639
2640	/*
2641	 * Rename permissions are combination of delete permission +
2642	 * add file/subdir permission.
2643	 *
2644	 * BSD operating systems also require write permission
2645	 * on the directory being moved from one parent directory
2646	 * to another.
2647	 */
2648	if (ZTOV(szp)->v_type == VDIR && ZTOV(sdzp) != ZTOV(tdzp)) {
2649		if ((error = zfs_zaccess(szp, ACE_WRITE_DATA, 0, B_FALSE, cr,
2650		    mnt_ns)))
2651			return (error);
2652	}
2653
2654	/*
2655	 * first make sure we do the delete portion.
2656	 *
2657	 * If that succeeds then check for add_file/add_subdir permissions
2658	 */
2659
2660	if ((error = zfs_zaccess_delete(sdzp, szp, cr, mnt_ns)))
2661		return (error);
2662
2663	/*
2664	 * If we have a tzp, see if we can delete it?
2665	 */
2666	if (tzp && (error = zfs_zaccess_delete(tdzp, tzp, cr, mnt_ns)))
2667		return (error);
2668
2669	/*
2670	 * Now check for add permissions
2671	 */
2672	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr, mnt_ns);
2673
2674	return (error);
2675}
2676