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