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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
28 #include <sys/types.h>
29 #include <sys/param.h>
31 #include <sys/sysmacros.h>
33 #include <sys/vnode.h>
38 #include <sys/cmn_err.h>
39 #include <sys/errno.h>
40 #include <sys/fs/zfs.h>
41 #include <sys/policy.h>
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_fuid.h>
44 #include <sys/zfs_acl.h>
45 #include <sys/zfs_dir.h>
46 #include <sys/zfs_quota.h>
47 #include <sys/zfs_vfsops.h>
49 #include <sys/dnode.h>
52 #include <sys/trace_acl.h>
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
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)
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)
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)
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)
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)
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)
87 #define RESTRICTED_CLEAR (ACE_WRITE_ACL|ACE_WRITE_OWNER)
89 #define V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
92 #define ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
95 #define ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
97 #define IDMAP_WK_CREATOR_OWNER_UID 2147483648U
100 zfs_ace_v0_get_type(void *acep)
102 return (((zfs_oldace_t *)acep)->z_type);
106 zfs_ace_v0_get_flags(void *acep)
108 return (((zfs_oldace_t *)acep)->z_flags);
112 zfs_ace_v0_get_mask(void *acep)
114 return (((zfs_oldace_t *)acep)->z_access_mask);
118 zfs_ace_v0_get_who(void *acep)
120 return (((zfs_oldace_t *)acep)->z_fuid);
124 zfs_ace_v0_set_type(void *acep, uint16_t type)
126 ((zfs_oldace_t *)acep)->z_type = type;
130 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
132 ((zfs_oldace_t *)acep)->z_flags = flags;
136 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
138 ((zfs_oldace_t *)acep)->z_access_mask = mask;
142 zfs_ace_v0_set_who(void *acep, uint64_t who)
144 ((zfs_oldace_t *)acep)->z_fuid = who;
149 zfs_ace_v0_size(void *acep)
151 return (sizeof (zfs_oldace_t));
155 zfs_ace_v0_abstract_size(void)
157 return (sizeof (zfs_oldace_t));
161 zfs_ace_v0_mask_off(void)
163 return (offsetof(zfs_oldace_t, z_access_mask));
168 zfs_ace_v0_data(void *acep, void **datap)
174 static acl_ops_t zfs_acl_v0_ops = {
175 .ace_mask_get = zfs_ace_v0_get_mask,
176 .ace_mask_set = zfs_ace_v0_set_mask,
177 .ace_flags_get = zfs_ace_v0_get_flags,
178 .ace_flags_set = zfs_ace_v0_set_flags,
179 .ace_type_get = zfs_ace_v0_get_type,
180 .ace_type_set = zfs_ace_v0_set_type,
181 .ace_who_get = zfs_ace_v0_get_who,
182 .ace_who_set = zfs_ace_v0_set_who,
183 .ace_size = zfs_ace_v0_size,
184 .ace_abstract_size = zfs_ace_v0_abstract_size,
185 .ace_mask_off = zfs_ace_v0_mask_off,
186 .ace_data = zfs_ace_v0_data
190 zfs_ace_fuid_get_type(void *acep)
192 return (((zfs_ace_hdr_t *)acep)->z_type);
196 zfs_ace_fuid_get_flags(void *acep)
198 return (((zfs_ace_hdr_t *)acep)->z_flags);
202 zfs_ace_fuid_get_mask(void *acep)
204 return (((zfs_ace_hdr_t *)acep)->z_access_mask);
208 zfs_ace_fuid_get_who(void *args)
211 zfs_ace_t *acep = args;
213 entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
215 if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
216 entry_type == ACE_EVERYONE)
218 return (((zfs_ace_t *)acep)->z_fuid);
222 zfs_ace_fuid_set_type(void *acep, uint16_t type)
224 ((zfs_ace_hdr_t *)acep)->z_type = type;
228 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
230 ((zfs_ace_hdr_t *)acep)->z_flags = flags;
234 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
236 ((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
240 zfs_ace_fuid_set_who(void *arg, uint64_t who)
242 zfs_ace_t *acep = arg;
244 uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
246 if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
247 entry_type == ACE_EVERYONE)
253 zfs_ace_fuid_size(void *acep)
255 zfs_ace_hdr_t *zacep = acep;
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));
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));
274 return (sizeof (zfs_ace_t));
279 zfs_ace_fuid_abstract_size(void)
281 return (sizeof (zfs_ace_hdr_t));
285 zfs_ace_fuid_mask_off(void)
287 return (offsetof(zfs_ace_hdr_t, z_access_mask));
291 zfs_ace_fuid_data(void *acep, void **datap)
293 zfs_ace_t *zacep = acep;
294 zfs_object_ace_t *zobjp;
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:
302 *datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
303 return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
310 static acl_ops_t zfs_acl_fuid_ops = {
311 .ace_mask_get = zfs_ace_fuid_get_mask,
312 .ace_mask_set = zfs_ace_fuid_set_mask,
313 .ace_flags_get = zfs_ace_fuid_get_flags,
314 .ace_flags_set = zfs_ace_fuid_set_flags,
315 .ace_type_get = zfs_ace_fuid_get_type,
316 .ace_type_set = zfs_ace_fuid_set_type,
317 .ace_who_get = zfs_ace_fuid_get_who,
318 .ace_who_set = zfs_ace_fuid_set_who,
319 .ace_size = zfs_ace_fuid_size,
320 .ace_abstract_size = zfs_ace_fuid_abstract_size,
321 .ace_mask_off = zfs_ace_fuid_mask_off,
322 .ace_data = zfs_ace_fuid_data
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.
332 zfs_external_acl(znode_t *zp)
334 zfs_acl_phys_t acl_phys;
341 * Need to deal with a potential
342 * race where zfs_sa_upgrade could cause
343 * z_isa_sa to change.
345 * If the lookup fails then the state of z_is_sa should have
349 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
350 &acl_phys, sizeof (acl_phys))) == 0)
351 return (acl_phys.z_acl_extern_obj);
354 * after upgrade the SA_ZPL_ZNODE_ACL should have been
357 VERIFY(zp->z_is_sa && error == ENOENT);
363 * Determine size of ACL in bytes
365 * This is more complicated than it should be since we have to deal
366 * with old external ACLs.
369 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
370 zfs_acl_phys_t *aclphys)
372 zfsvfs_t *zfsvfs = ZTOZSB(zp);
377 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
379 if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
383 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
384 &acl_count, sizeof (acl_count))) != 0)
386 *aclcount = acl_count;
388 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
389 aclphys, sizeof (*aclphys))) != 0)
392 if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
393 *aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
394 *aclcount = aclphys->z_acl_size;
396 *aclsize = aclphys->z_acl_size;
397 *aclcount = aclphys->z_acl_count;
404 zfs_znode_acl_version(znode_t *zp)
406 zfs_acl_phys_t acl_phys;
409 return (ZFS_ACL_VERSION_FUID);
414 * Need to deal with a potential
415 * race where zfs_sa_upgrade could cause
416 * z_isa_sa to change.
418 * If the lookup fails then the state of z_is_sa should have
421 if ((error = sa_lookup(zp->z_sa_hdl,
422 SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
423 &acl_phys, sizeof (acl_phys))) == 0)
424 return (acl_phys.z_acl_version);
427 * After upgrade SA_ZPL_ZNODE_ACL should have
430 VERIFY(zp->z_is_sa && error == ENOENT);
431 return (ZFS_ACL_VERSION_FUID);
437 zfs_acl_version(int version)
439 if (version < ZPL_VERSION_FUID)
440 return (ZFS_ACL_VERSION_INITIAL);
442 return (ZFS_ACL_VERSION_FUID);
446 zfs_acl_version_zp(znode_t *zp)
448 return (zfs_acl_version(ZTOZSB(zp)->z_version));
452 zfs_acl_alloc(int vers)
456 aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
457 list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
458 offsetof(zfs_acl_node_t, z_next));
459 aclp->z_version = vers;
460 if (vers == ZFS_ACL_VERSION_FUID)
461 aclp->z_ops = &zfs_acl_fuid_ops;
463 aclp->z_ops = &zfs_acl_v0_ops;
468 zfs_acl_node_alloc(size_t bytes)
470 zfs_acl_node_t *aclnode;
472 aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
474 aclnode->z_acldata = kmem_alloc(bytes, KM_SLEEP);
475 aclnode->z_allocdata = aclnode->z_acldata;
476 aclnode->z_allocsize = bytes;
477 aclnode->z_size = bytes;
484 zfs_acl_node_free(zfs_acl_node_t *aclnode)
486 if (aclnode->z_allocsize)
487 kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
488 kmem_free(aclnode, sizeof (zfs_acl_node_t));
492 zfs_acl_release_nodes(zfs_acl_t *aclp)
494 zfs_acl_node_t *aclnode;
496 while ((aclnode = list_head(&aclp->z_acl))) {
497 list_remove(&aclp->z_acl, aclnode);
498 zfs_acl_node_free(aclnode);
500 aclp->z_acl_count = 0;
501 aclp->z_acl_bytes = 0;
505 zfs_acl_free(zfs_acl_t *aclp)
507 zfs_acl_release_nodes(aclp);
508 list_destroy(&aclp->z_acl);
509 kmem_free(aclp, sizeof (zfs_acl_t));
513 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
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);
528 if (type >= MIN_ACE_TYPE && type <= MAX_ACE_TYPE)
535 zfs_ace_valid(umode_t obj_mode, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
538 * first check type of entry
541 if (!zfs_acl_valid_ace_type(type, iflags))
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)
551 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
555 * next check inheritance level flags
558 if (S_ISDIR(obj_mode) &&
559 (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
560 aclp->z_hints |= ZFS_INHERIT_ACE;
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) {
573 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
574 uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
576 zfs_acl_node_t *aclnode;
581 aclnode = list_head(&aclp->z_acl);
585 aclp->z_next_ace = aclnode->z_acldata;
586 aclp->z_curr_node = aclnode;
587 aclnode->z_ace_idx = 0;
590 aclnode = aclp->z_curr_node;
595 if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
596 aclnode = list_next(&aclp->z_acl, aclnode);
600 aclp->z_curr_node = aclnode;
601 aclnode->z_ace_idx = 0;
602 aclp->z_next_ace = aclnode->z_acldata;
606 if (aclnode->z_ace_idx < aclnode->z_ace_count) {
607 void *acep = aclp->z_next_ace;
611 * Make sure we don't overstep our bounds
613 ace_size = aclp->z_ops->ace_size(acep);
615 if (((caddr_t)acep + ace_size) >
616 ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
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++;
627 return ((void *)acep);
634 zfs_ace_walk(void *datap, uint64_t cookie, int aclcnt,
635 uint16_t *flags, uint16_t *type, uint32_t *mask)
637 zfs_acl_t *aclp = datap;
638 zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
641 acep = zfs_acl_next_ace(aclp, acep, &who, mask,
643 return ((uint64_t)(uintptr_t)acep);
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.
652 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, umode_t obj_mode, 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)
658 zfs_ace_t *aceptr = z_acl;
660 zfs_object_ace_t *zobjacep;
661 ace_object_t *aceobjp;
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);
676 * Make sure ACE is valid
678 if (zfs_ace_valid(obj_mode, aclp, aceptr->z_hdr.z_type,
679 aceptr->z_hdr.z_flags) != B_TRUE)
680 return (SET_ERROR(EINVAL));
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;
690 bcopy(aceobjp->a_obj_type, zobjacep->z_object_type,
691 sizeof (aceobjp->a_obj_type));
692 bcopy(aceobjp->a_inherit_obj_type,
693 zobjacep->z_inherit_type,
694 sizeof (aceobjp->a_inherit_obj_type));
695 acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
698 acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
701 aceptr = (zfs_ace_t *)((caddr_t)aceptr +
702 aclp->z_ops->ace_size(aceptr));
705 *size = (caddr_t)aceptr - (caddr_t)z_acl;
711 * Copy ZFS ACEs to fixed size ace_t layout
714 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
715 void *datap, int filter)
718 uint32_t access_mask;
719 uint16_t iflags, type;
720 zfs_ace_hdr_t *zacep = NULL;
722 ace_object_t *objacep;
723 zfs_object_ace_t *zobjacep;
727 while ((zacep = zfs_acl_next_ace(aclp, zacep,
728 &who, &access_mask, &iflags, &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:
738 zobjacep = (zfs_object_ace_t *)zacep;
739 objacep = (ace_object_t *)acep;
740 bcopy(zobjacep->z_object_type,
742 sizeof (zobjacep->z_object_type));
743 bcopy(zobjacep->z_inherit_type,
744 objacep->a_inherit_obj_type,
745 sizeof (zobjacep->z_inherit_type));
746 ace_size = sizeof (ace_object_t);
749 ace_size = sizeof (ace_t);
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);
761 acep->a_who = (uid_t)(int64_t)who;
763 acep->a_access_mask = access_mask;
764 acep->a_flags = iflags;
766 acep = (ace_t *)((caddr_t)acep + ace_size);
771 zfs_copy_ace_2_oldace(umode_t obj_mode, zfs_acl_t *aclp, ace_t *acep,
772 zfs_oldace_t *z_acl, int aclcnt, size_t *size)
775 zfs_oldace_t *aceptr = z_acl;
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;
783 * Make sure ACE is valid
785 if (zfs_ace_valid(obj_mode, aclp, aceptr->z_type,
786 aceptr->z_flags) != B_TRUE)
787 return (SET_ERROR(EINVAL));
789 *size = (caddr_t)aceptr - (caddr_t)z_acl;
794 * convert old ACL format to new
797 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
799 zfs_oldace_t *oldaclp;
801 uint16_t type, iflags;
802 uint32_t access_mask;
805 zfs_acl_node_t *newaclnode;
807 ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
809 * First create the ACE in a contiguous piece of memory
810 * for zfs_copy_ace_2_fuid().
812 * We only convert an ACL once, so this won't happen
815 oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
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;
826 newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
827 sizeof (zfs_object_ace_t));
828 aclp->z_ops = &zfs_acl_fuid_ops;
829 VERIFY(zfs_copy_ace_2_fuid(ZTOZSB(zp), ZTOI(zp)->i_mode,
830 aclp, oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
831 &newaclnode->z_size, NULL, cr) == 0);
832 newaclnode->z_ace_count = aclp->z_acl_count;
833 aclp->z_version = ZFS_ACL_VERSION;
834 kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
837 * Release all previous ACL nodes
840 zfs_acl_release_nodes(aclp);
842 list_insert_head(&aclp->z_acl, newaclnode);
844 aclp->z_acl_bytes = newaclnode->z_size;
845 aclp->z_acl_count = newaclnode->z_ace_count;
850 * Convert unix access mask to v4 access mask
853 zfs_unix_to_v4(uint32_t access_mask)
855 uint32_t new_mask = 0;
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;
867 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
868 uint16_t access_type, uint64_t fuid, uint16_t entry_type)
870 uint16_t type = entry_type & ACE_TYPE_FLAGS;
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);
881 * Determine mode of file based on ACL.
884 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
885 uint64_t *pflags, uint64_t fuid, uint64_t fgid)
890 zfs_ace_hdr_t *acep = NULL;
892 uint16_t iflags, type;
893 uint32_t access_mask;
894 boolean_t an_exec_denied = B_FALSE;
896 mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
898 while ((acep = zfs_acl_next_ace(aclp, acep, &who,
899 &access_mask, &iflags, &type))) {
901 if (!zfs_acl_valid_ace_type(type, iflags))
904 entry_type = (iflags & ACE_TYPE_FLAGS);
907 * Skip over any inherit_only ACEs
909 if (iflags & ACE_INHERIT_ONLY_ACE)
912 if (entry_type == ACE_OWNER || (entry_type == 0 &&
914 if ((access_mask & ACE_READ_DATA) &&
915 (!(seen & S_IRUSR))) {
921 if ((access_mask & ACE_WRITE_DATA) &&
922 (!(seen & S_IWUSR))) {
928 if ((access_mask & ACE_EXECUTE) &&
929 (!(seen & S_IXUSR))) {
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))) {
944 if ((access_mask & ACE_WRITE_DATA) &&
945 (!(seen & S_IWGRP))) {
951 if ((access_mask & ACE_EXECUTE) &&
952 (!(seen & S_IXGRP))) {
958 } else if (entry_type == ACE_EVERYONE) {
959 if ((access_mask & ACE_READ_DATA)) {
960 if (!(seen & S_IRUSR)) {
966 if (!(seen & S_IRGRP)) {
972 if (!(seen & S_IROTH)) {
979 if ((access_mask & ACE_WRITE_DATA)) {
980 if (!(seen & S_IWUSR)) {
986 if (!(seen & S_IWGRP)) {
992 if (!(seen & S_IWOTH)) {
999 if ((access_mask & ACE_EXECUTE)) {
1000 if (!(seen & S_IXUSR)) {
1002 if (type == ALLOW) {
1006 if (!(seen & S_IXGRP)) {
1008 if (type == ALLOW) {
1012 if (!(seen & S_IXOTH)) {
1014 if (type == ALLOW) {
1021 * Only care if this IDENTIFIER_GROUP or
1022 * USER ACE denies execute access to someone,
1023 * mode is not affected
1025 if ((access_mask & ACE_EXECUTE) && type == DENY)
1026 an_exec_denied = B_TRUE;
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.
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;
1041 *pflags &= ~ZFS_NO_EXECS_DENIED;
1043 *pflags |= ZFS_NO_EXECS_DENIED;
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.
1053 zfs_acl_node_read(struct znode *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1054 boolean_t will_modify)
1059 zfs_acl_node_t *aclnode;
1060 zfs_acl_phys_t znode_acl;
1063 boolean_t drop_lock = B_FALSE;
1065 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1067 if (zp->z_acl_cached && !will_modify) {
1068 *aclpp = zp->z_acl_cached;
1073 * close race where znode could be upgrade while trying to
1074 * read the znode attributes.
1076 * But this could only happen if the file isn't already an SA
1079 if (!zp->z_is_sa && !have_lock) {
1080 mutex_enter(&zp->z_lock);
1083 version = zfs_znode_acl_version(zp);
1085 if ((error = zfs_acl_znode_info(zp, &aclsize,
1086 &acl_count, &znode_acl)) != 0) {
1090 aclp = zfs_acl_alloc(version);
1092 aclp->z_acl_count = acl_count;
1093 aclp->z_acl_bytes = aclsize;
1095 aclnode = zfs_acl_node_alloc(aclsize);
1096 aclnode->z_ace_count = aclp->z_acl_count;
1097 aclnode->z_size = aclsize;
1100 if (znode_acl.z_acl_extern_obj) {
1101 error = dmu_read(ZTOZSB(zp)->z_os,
1102 znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1103 aclnode->z_acldata, DMU_READ_PREFETCH);
1105 bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1109 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(ZTOZSB(zp)),
1110 aclnode->z_acldata, aclnode->z_size);
1115 zfs_acl_node_free(aclnode);
1116 /* convert checksum errors into IO errors */
1117 if (error == ECKSUM)
1118 error = SET_ERROR(EIO);
1122 list_insert_head(&aclp->z_acl, aclnode);
1126 zp->z_acl_cached = aclp;
1129 mutex_exit(&zp->z_lock);
1135 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1136 boolean_t start, void *userdata)
1138 zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1141 cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1143 cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1146 *dataptr = cb->cb_acl_node->z_acldata;
1147 *length = cb->cb_acl_node->z_size;
1151 zfs_acl_chown_setattr(znode_t *zp)
1156 if (ZTOZSB(zp)->z_acl_type == ZFS_ACLTYPE_POSIX)
1159 ASSERT(MUTEX_HELD(&zp->z_lock));
1160 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1162 error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1163 if (error == 0 && aclp->z_acl_count > 0)
1164 zp->z_mode = ZTOI(zp)->i_mode =
1165 zfs_mode_compute(zp->z_mode, aclp,
1166 &zp->z_pflags, KUID_TO_SUID(ZTOI(zp)->i_uid),
1167 KGID_TO_SGID(ZTOI(zp)->i_gid));
1170 * Some ZFS implementations (ZEVO) create neither a ZNODE_ACL
1171 * nor a DACL_ACES SA in which case ENOENT is returned from
1172 * zfs_acl_node_read() when the SA can't be located.
1173 * Allow chown/chgrp to succeed in these cases rather than
1174 * returning an error that makes no sense in the context of
1177 if (error == ENOENT)
1183 typedef struct trivial_acl {
1184 uint32_t allow0; /* allow mask for bits only in owner */
1185 uint32_t deny1; /* deny mask for bits not in owner */
1186 uint32_t deny2; /* deny mask for bits not in group */
1187 uint32_t owner; /* allow mask matching mode */
1188 uint32_t group; /* allow mask matching mode */
1189 uint32_t everyone; /* allow mask matching mode */
1193 acl_trivial_access_masks(mode_t mode, boolean_t isdir, trivial_acl_t *masks)
1195 uint32_t read_mask = ACE_READ_DATA;
1196 uint32_t write_mask = ACE_WRITE_DATA|ACE_APPEND_DATA;
1197 uint32_t execute_mask = ACE_EXECUTE;
1200 write_mask |= ACE_DELETE_CHILD;
1204 if (!(mode & S_IRUSR) && (mode & (S_IRGRP|S_IROTH)))
1205 masks->deny1 |= read_mask;
1206 if (!(mode & S_IWUSR) && (mode & (S_IWGRP|S_IWOTH)))
1207 masks->deny1 |= write_mask;
1208 if (!(mode & S_IXUSR) && (mode & (S_IXGRP|S_IXOTH)))
1209 masks->deny1 |= execute_mask;
1212 if (!(mode & S_IRGRP) && (mode & S_IROTH))
1213 masks->deny2 |= read_mask;
1214 if (!(mode & S_IWGRP) && (mode & S_IWOTH))
1215 masks->deny2 |= write_mask;
1216 if (!(mode & S_IXGRP) && (mode & S_IXOTH))
1217 masks->deny2 |= execute_mask;
1220 if ((mode & S_IRUSR) && (!(mode & S_IRGRP) && (mode & S_IROTH)))
1221 masks->allow0 |= read_mask;
1222 if ((mode & S_IWUSR) && (!(mode & S_IWGRP) && (mode & S_IWOTH)))
1223 masks->allow0 |= write_mask;
1224 if ((mode & S_IXUSR) && (!(mode & S_IXGRP) && (mode & S_IXOTH)))
1225 masks->allow0 |= execute_mask;
1227 masks->owner = ACE_WRITE_ATTRIBUTES|ACE_WRITE_OWNER|ACE_WRITE_ACL|
1228 ACE_WRITE_NAMED_ATTRS|ACE_READ_ACL|ACE_READ_ATTRIBUTES|
1229 ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE;
1231 masks->owner |= read_mask;
1233 masks->owner |= write_mask;
1235 masks->owner |= execute_mask;
1237 masks->group = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1240 masks->group |= read_mask;
1242 masks->group |= write_mask;
1244 masks->group |= execute_mask;
1246 masks->everyone = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1249 masks->everyone |= read_mask;
1251 masks->everyone |= write_mask;
1253 masks->everyone |= execute_mask;
1258 * determine whether an ace_t acl is trivial
1260 * Trivialness implies that the acl is composed of only
1261 * owner, group, everyone entries. ACL can't
1262 * have read_acl denied, and write_owner/write_acl/write_attributes
1263 * can only be owner@ entry.
1266 ace_trivial_common(void *acep, int aclcnt,
1267 uint64_t (*walk)(void *, uint64_t, int aclcnt,
1268 uint16_t *, uint16_t *, uint32_t *))
1273 uint64_t cookie = 0;
1275 while ((cookie = walk(acep, cookie, aclcnt, &flags, &type, &mask))) {
1276 switch (flags & ACE_TYPE_FLAGS) {
1278 case ACE_GROUP|ACE_IDENTIFIER_GROUP:
1285 if (flags & (ACE_FILE_INHERIT_ACE|
1286 ACE_DIRECTORY_INHERIT_ACE|ACE_NO_PROPAGATE_INHERIT_ACE|
1287 ACE_INHERIT_ONLY_ACE))
1291 * Special check for some special bits
1293 * Don't allow anybody to deny reading basic
1294 * attributes or a files ACL.
1296 if ((mask & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
1297 (type == ACE_ACCESS_DENIED_ACE_TYPE))
1301 * Delete permission is never set by default
1303 if (mask & ACE_DELETE)
1307 * Child delete permission should be accompanied by write
1309 if ((mask & ACE_DELETE_CHILD) && !(mask & ACE_WRITE_DATA))
1313 * only allow owner@ to have
1314 * write_acl/write_owner/write_attributes/write_xattr/
1316 if (type == ACE_ACCESS_ALLOWED_ACE_TYPE &&
1317 (!(flags & ACE_OWNER) && (mask &
1318 (ACE_WRITE_OWNER|ACE_WRITE_ACL| ACE_WRITE_ATTRIBUTES|
1319 ACE_WRITE_NAMED_ATTRS))))
1328 * common code for setting ACLs.
1330 * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1331 * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1332 * already checked the acl and knows whether to inherit.
1335 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1338 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1339 dmu_object_type_t otype;
1340 zfs_acl_locator_cb_t locate = { 0 };
1342 sa_bulk_attr_t bulk[5];
1345 zfs_acl_phys_t acl_phys;
1349 mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1350 KUID_TO_SUID(ZTOI(zp)->i_uid), KGID_TO_SGID(ZTOI(zp)->i_gid));
1352 zp->z_mode = ZTOI(zp)->i_mode = mode;
1353 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1354 &mode, sizeof (mode));
1355 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1356 &zp->z_pflags, sizeof (zp->z_pflags));
1357 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1358 &ctime, sizeof (ctime));
1360 if (zp->z_acl_cached) {
1361 zfs_acl_free(zp->z_acl_cached);
1362 zp->z_acl_cached = NULL;
1368 if (!zfsvfs->z_use_fuids) {
1369 otype = DMU_OT_OLDACL;
1371 if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1372 (zfsvfs->z_version >= ZPL_VERSION_FUID))
1373 zfs_acl_xform(zp, aclp, cr);
1374 ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1379 * Arrgh, we have to handle old on disk format
1380 * as well as newer (preferred) SA format.
1383 if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1384 locate.cb_aclp = aclp;
1385 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1386 zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1387 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1388 NULL, &aclp->z_acl_count, sizeof (uint64_t));
1389 } else { /* Painful legacy way */
1390 zfs_acl_node_t *aclnode;
1394 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1395 &acl_phys, sizeof (acl_phys))) != 0)
1398 aoid = acl_phys.z_acl_extern_obj;
1400 if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1402 * If ACL was previously external and we are now
1403 * converting to new ACL format then release old
1404 * ACL object and create a new one.
1407 aclp->z_version != acl_phys.z_acl_version) {
1408 error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1414 aoid = dmu_object_alloc(zfsvfs->z_os,
1415 otype, aclp->z_acl_bytes,
1416 otype == DMU_OT_ACL ?
1417 DMU_OT_SYSACL : DMU_OT_NONE,
1418 otype == DMU_OT_ACL ?
1419 DN_OLD_MAX_BONUSLEN : 0, tx);
1421 (void) dmu_object_set_blocksize(zfsvfs->z_os,
1422 aoid, aclp->z_acl_bytes, 0, tx);
1424 acl_phys.z_acl_extern_obj = aoid;
1425 for (aclnode = list_head(&aclp->z_acl); aclnode;
1426 aclnode = list_next(&aclp->z_acl, aclnode)) {
1427 if (aclnode->z_ace_count == 0)
1429 dmu_write(zfsvfs->z_os, aoid, off,
1430 aclnode->z_size, aclnode->z_acldata, tx);
1431 off += aclnode->z_size;
1434 void *start = acl_phys.z_ace_data;
1436 * Migrating back embedded?
1438 if (acl_phys.z_acl_extern_obj) {
1439 error = dmu_object_free(zfsvfs->z_os,
1440 acl_phys.z_acl_extern_obj, tx);
1443 acl_phys.z_acl_extern_obj = 0;
1446 for (aclnode = list_head(&aclp->z_acl); aclnode;
1447 aclnode = list_next(&aclp->z_acl, aclnode)) {
1448 if (aclnode->z_ace_count == 0)
1450 bcopy(aclnode->z_acldata, start,
1452 start = (caddr_t)start + aclnode->z_size;
1456 * If Old version then swap count/bytes to match old
1457 * layout of znode_acl_phys_t.
1459 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1460 acl_phys.z_acl_size = aclp->z_acl_count;
1461 acl_phys.z_acl_count = aclp->z_acl_bytes;
1463 acl_phys.z_acl_size = aclp->z_acl_bytes;
1464 acl_phys.z_acl_count = aclp->z_acl_count;
1466 acl_phys.z_acl_version = aclp->z_version;
1468 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1469 &acl_phys, sizeof (acl_phys));
1473 * Replace ACL wide bits, but first clear them.
1475 zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1477 zp->z_pflags |= aclp->z_hints;
1479 if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1480 zp->z_pflags |= ZFS_ACL_TRIVIAL;
1482 zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1483 return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1487 zfs_acl_chmod(boolean_t isdir, uint64_t mode, boolean_t split, boolean_t trim,
1492 int new_count, new_bytes;
1495 uint16_t iflags, type;
1496 uint32_t access_mask;
1497 zfs_acl_node_t *newnode;
1498 size_t abstract_size = aclp->z_ops->ace_abstract_size();
1500 trivial_acl_t masks;
1502 new_count = new_bytes = 0;
1504 acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1506 newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1508 zacep = newnode->z_acldata;
1510 zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1511 zacep = (void *)((uintptr_t)zacep + abstract_size);
1513 new_bytes += abstract_size;
1516 zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1517 zacep = (void *)((uintptr_t)zacep + abstract_size);
1519 new_bytes += abstract_size;
1522 zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1523 zacep = (void *)((uintptr_t)zacep + abstract_size);
1525 new_bytes += abstract_size;
1528 while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1530 entry_type = (iflags & ACE_TYPE_FLAGS);
1532 * ACEs used to represent the file mode may be divided
1533 * into an equivalent pair of inherit-only and regular
1534 * ACEs, if they are inheritable.
1535 * Skip regular ACEs, which are replaced by the new mode.
1537 if (split && (entry_type == ACE_OWNER ||
1538 entry_type == OWNING_GROUP ||
1539 entry_type == ACE_EVERYONE)) {
1540 if (!isdir || !(iflags &
1541 (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1544 * We preserve owner@, group@, or @everyone
1545 * permissions, if they are inheritable, by
1546 * copying them to inherit_only ACEs. This
1547 * prevents inheritable permissions from being
1548 * altered along with the file mode.
1550 iflags |= ACE_INHERIT_ONLY_ACE;
1554 * If this ACL has any inheritable ACEs, mark that in
1555 * the hints (which are later masked into the pflags)
1556 * so create knows to do inheritance.
1558 if (isdir && (iflags &
1559 (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1560 aclp->z_hints |= ZFS_INHERIT_ACE;
1562 if ((type != ALLOW && type != DENY) ||
1563 (iflags & ACE_INHERIT_ONLY_ACE)) {
1565 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1566 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1567 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1568 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1569 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1574 * Limit permissions to be no greater than
1575 * group permissions.
1576 * The "aclinherit" and "aclmode" properties
1577 * affect policy for create and chmod(2),
1580 if ((type == ALLOW) && trim)
1581 access_mask &= masks.group;
1583 zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1584 ace_size = aclp->z_ops->ace_size(acep);
1585 zacep = (void *)((uintptr_t)zacep + ace_size);
1587 new_bytes += ace_size;
1589 zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1590 zacep = (void *)((uintptr_t)zacep + abstract_size);
1591 zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1592 zacep = (void *)((uintptr_t)zacep + abstract_size);
1593 zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1596 new_bytes += abstract_size * 3;
1597 zfs_acl_release_nodes(aclp);
1598 aclp->z_acl_count = new_count;
1599 aclp->z_acl_bytes = new_bytes;
1600 newnode->z_ace_count = new_count;
1601 newnode->z_size = new_bytes;
1602 list_insert_tail(&aclp->z_acl, newnode);
1606 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1610 mutex_enter(&zp->z_acl_lock);
1611 mutex_enter(&zp->z_lock);
1612 if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_DISCARD)
1613 *aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1615 error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1618 (*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1619 zfs_acl_chmod(S_ISDIR(ZTOI(zp)->i_mode), mode, B_TRUE,
1620 (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1622 mutex_exit(&zp->z_lock);
1623 mutex_exit(&zp->z_acl_lock);
1629 * Should ACE be inherited?
1632 zfs_ace_can_use(umode_t obj_mode, uint16_t acep_flags)
1634 int iflags = (acep_flags & 0xf);
1636 if (S_ISDIR(obj_mode) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1638 else if (iflags & ACE_FILE_INHERIT_ACE)
1639 return (!(S_ISDIR(obj_mode) &&
1640 (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1645 * inherit inheritable ACEs from parent
1648 zfs_acl_inherit(zfsvfs_t *zfsvfs, umode_t va_mode, zfs_acl_t *paclp,
1649 uint64_t mode, boolean_t *need_chmod)
1653 zfs_acl_node_t *aclnode;
1654 zfs_acl_t *aclp = NULL;
1656 uint32_t access_mask;
1657 uint16_t iflags, newflags, type;
1659 void *data1, *data2;
1660 size_t data1sz, data2sz;
1662 boolean_t isdir = S_ISDIR(va_mode);
1663 boolean_t isreg = S_ISREG(va_mode);
1665 *need_chmod = B_TRUE;
1667 aclp = zfs_acl_alloc(paclp->z_version);
1668 aclinherit = zfsvfs->z_acl_inherit;
1669 if (aclinherit == ZFS_ACL_DISCARD || S_ISLNK(va_mode))
1672 while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1673 &access_mask, &iflags, &type))) {
1676 * don't inherit bogus ACEs
1678 if (!zfs_acl_valid_ace_type(type, iflags))
1682 * Check if ACE is inheritable by this vnode
1684 if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1685 !zfs_ace_can_use(va_mode, iflags))
1689 * If owner@, group@, or everyone@ inheritable
1690 * then zfs_acl_chmod() isn't needed.
1692 if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1693 aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1694 ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1695 ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1696 (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1697 *need_chmod = B_FALSE;
1700 * Strip inherited execute permission from file if
1703 if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1704 !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1705 access_mask &= ~ACE_EXECUTE;
1709 * Strip write_acl and write_owner from permissions
1710 * when inheriting an ACE
1712 if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1713 access_mask &= ~RESTRICTED_CLEAR;
1716 ace_size = aclp->z_ops->ace_size(pacep);
1717 aclnode = zfs_acl_node_alloc(ace_size);
1718 list_insert_tail(&aclp->z_acl, aclnode);
1719 acep = aclnode->z_acldata;
1721 zfs_set_ace(aclp, acep, access_mask, type,
1722 who, iflags|ACE_INHERITED_ACE);
1725 * Copy special opaque data if any
1727 if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1728 VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1729 &data2)) == data1sz);
1730 bcopy(data1, data2, data2sz);
1733 aclp->z_acl_count++;
1734 aclnode->z_ace_count++;
1735 aclp->z_acl_bytes += aclnode->z_size;
1736 newflags = aclp->z_ops->ace_flags_get(acep);
1739 * If ACE is not to be inherited further, or if the vnode is
1740 * not a directory, remove all inheritance flags
1742 if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1743 newflags &= ~ALL_INHERIT;
1744 aclp->z_ops->ace_flags_set(acep,
1745 newflags|ACE_INHERITED_ACE);
1750 * This directory has an inheritable ACE
1752 aclp->z_hints |= ZFS_INHERIT_ACE;
1755 * If only FILE_INHERIT is set then turn on
1758 if ((iflags & (ACE_FILE_INHERIT_ACE |
1759 ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1760 newflags |= ACE_INHERIT_ONLY_ACE;
1761 aclp->z_ops->ace_flags_set(acep,
1762 newflags|ACE_INHERITED_ACE);
1764 newflags &= ~ACE_INHERIT_ONLY_ACE;
1765 aclp->z_ops->ace_flags_set(acep,
1766 newflags|ACE_INHERITED_ACE);
1769 if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1770 aclp->z_acl_count != 0) {
1771 *need_chmod = B_FALSE;
1778 * Create file system object initial permissions
1779 * including inheritable ACEs.
1780 * Also, create FUIDs for owner and group.
1783 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1784 vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1787 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
1789 gid_t gid = vap->va_gid;
1790 boolean_t need_chmod = B_TRUE;
1791 boolean_t trim = B_FALSE;
1792 boolean_t inherited = B_FALSE;
1794 bzero(acl_ids, sizeof (zfs_acl_ids_t));
1795 acl_ids->z_mode = vap->va_mode;
1798 if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_mode, vsecp,
1799 cr, &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1802 acl_ids->z_fuid = vap->va_uid;
1803 acl_ids->z_fgid = vap->va_gid;
1806 * Determine uid and gid.
1808 if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1809 ((flag & IS_XATTR) && (S_ISDIR(vap->va_mode)))) {
1810 acl_ids->z_fuid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_uid,
1811 cr, ZFS_OWNER, &acl_ids->z_fuidp);
1812 acl_ids->z_fgid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
1813 cr, ZFS_GROUP, &acl_ids->z_fuidp);
1816 acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1817 cr, &acl_ids->z_fuidp);
1818 acl_ids->z_fgid = 0;
1819 if (vap->va_mask & AT_GID) {
1820 acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1821 (uint64_t)vap->va_gid,
1822 cr, ZFS_GROUP, &acl_ids->z_fuidp);
1824 if (acl_ids->z_fgid != KGID_TO_SGID(ZTOI(dzp)->i_gid) &&
1825 !groupmember(vap->va_gid, cr) &&
1826 secpolicy_vnode_create_gid(cr) != 0)
1827 acl_ids->z_fgid = 0;
1829 if (acl_ids->z_fgid == 0) {
1830 if (dzp->z_mode & S_ISGID) {
1834 acl_ids->z_fgid = KGID_TO_SGID(
1836 gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1839 if (zfsvfs->z_use_fuids &&
1840 IS_EPHEMERAL(acl_ids->z_fgid)) {
1841 domain = zfs_fuid_idx_domain(
1842 &zfsvfs->z_fuid_idx,
1843 FUID_INDEX(acl_ids->z_fgid));
1844 rid = FUID_RID(acl_ids->z_fgid);
1845 zfs_fuid_node_add(&acl_ids->z_fuidp,
1847 FUID_INDEX(acl_ids->z_fgid),
1848 acl_ids->z_fgid, ZFS_GROUP);
1851 acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1852 ZFS_GROUP, cr, &acl_ids->z_fuidp);
1857 #endif /* HAVE_KSID */
1860 * If we're creating a directory, and the parent directory has the
1861 * set-GID bit set, set in on the new directory.
1862 * Otherwise, if the user is neither privileged nor a member of the
1863 * file's new group, clear the file's set-GID bit.
1866 if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1867 (S_ISDIR(vap->va_mode))) {
1868 acl_ids->z_mode |= S_ISGID;
1870 if ((acl_ids->z_mode & S_ISGID) &&
1871 secpolicy_vnode_setids_setgids(cr, gid) != 0)
1872 acl_ids->z_mode &= ~S_ISGID;
1875 if (acl_ids->z_aclp == NULL) {
1876 mutex_enter(&dzp->z_acl_lock);
1877 mutex_enter(&dzp->z_lock);
1878 if (!(flag & IS_ROOT_NODE) &&
1879 (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1880 !(dzp->z_pflags & ZFS_XATTR)) {
1881 VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1883 acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1884 vap->va_mode, paclp, acl_ids->z_mode, &need_chmod);
1888 zfs_acl_alloc(zfs_acl_version_zp(dzp));
1889 acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1891 mutex_exit(&dzp->z_lock);
1892 mutex_exit(&dzp->z_acl_lock);
1895 if (S_ISDIR(vap->va_mode))
1896 acl_ids->z_aclp->z_hints |=
1897 ZFS_ACL_AUTO_INHERIT;
1899 if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1900 zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1901 zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1903 zfs_acl_chmod(vap->va_mode, acl_ids->z_mode, B_FALSE,
1904 trim, acl_ids->z_aclp);
1908 if (inherited || vsecp) {
1909 acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1910 acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1911 acl_ids->z_fuid, acl_ids->z_fgid);
1912 if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1913 acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1920 * Free ACL and fuid_infop, but not the acl_ids structure
1923 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1925 if (acl_ids->z_aclp)
1926 zfs_acl_free(acl_ids->z_aclp);
1927 if (acl_ids->z_fuidp)
1928 zfs_fuid_info_free(acl_ids->z_fuidp);
1929 acl_ids->z_aclp = NULL;
1930 acl_ids->z_fuidp = NULL;
1934 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1936 return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1937 zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1938 (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1939 zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1943 * Retrieve a file's ACL
1946 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1954 mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1955 VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1958 return (SET_ERROR(ENOSYS));
1960 if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr)))
1963 mutex_enter(&zp->z_acl_lock);
1965 error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1967 mutex_exit(&zp->z_acl_lock);
1972 * Scan ACL to determine number of ACEs
1974 if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1977 uint32_t access_mask;
1978 uint16_t type, iflags;
1980 while ((zacep = zfs_acl_next_ace(aclp, zacep,
1981 &who, &access_mask, &iflags, &type))) {
1983 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1984 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1985 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1986 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1993 vsecp->vsa_aclcnt = count;
1995 count = (int)aclp->z_acl_count;
1997 if (mask & VSA_ACECNT) {
1998 vsecp->vsa_aclcnt = count;
2001 if (mask & VSA_ACE) {
2004 aclsz = count * sizeof (ace_t) +
2005 sizeof (ace_object_t) * largeace;
2007 vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2008 vsecp->vsa_aclentsz = aclsz;
2010 if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2011 zfs_copy_fuid_2_ace(ZTOZSB(zp), aclp, cr,
2012 vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2014 zfs_acl_node_t *aclnode;
2015 void *start = vsecp->vsa_aclentp;
2017 for (aclnode = list_head(&aclp->z_acl); aclnode;
2018 aclnode = list_next(&aclp->z_acl, aclnode)) {
2019 bcopy(aclnode->z_acldata, start,
2021 start = (caddr_t)start + aclnode->z_size;
2023 ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
2027 if (mask & VSA_ACE_ACLFLAGS) {
2028 vsecp->vsa_aclflags = 0;
2029 if (zp->z_pflags & ZFS_ACL_DEFAULTED)
2030 vsecp->vsa_aclflags |= ACL_DEFAULTED;
2031 if (zp->z_pflags & ZFS_ACL_PROTECTED)
2032 vsecp->vsa_aclflags |= ACL_PROTECTED;
2033 if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
2034 vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2037 mutex_exit(&zp->z_acl_lock);
2043 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_mode,
2044 vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2047 zfs_acl_node_t *aclnode;
2048 int aclcnt = vsecp->vsa_aclcnt;
2051 if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2052 return (SET_ERROR(EINVAL));
2054 aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2057 aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2058 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2059 if ((error = zfs_copy_ace_2_oldace(obj_mode, aclp,
2060 (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2061 aclcnt, &aclnode->z_size)) != 0) {
2063 zfs_acl_node_free(aclnode);
2067 if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_mode, aclp,
2068 vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2069 &aclnode->z_size, fuidp, cr)) != 0) {
2071 zfs_acl_node_free(aclnode);
2075 aclp->z_acl_bytes = aclnode->z_size;
2076 aclnode->z_ace_count = aclcnt;
2077 aclp->z_acl_count = aclcnt;
2078 list_insert_head(&aclp->z_acl, aclnode);
2081 * If flags are being set then add them to z_hints
2083 if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2084 if (vsecp->vsa_aclflags & ACL_PROTECTED)
2085 aclp->z_hints |= ZFS_ACL_PROTECTED;
2086 if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2087 aclp->z_hints |= ZFS_ACL_DEFAULTED;
2088 if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2089 aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2101 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2103 zfsvfs_t *zfsvfs = ZTOZSB(zp);
2104 zilog_t *zilog = zfsvfs->z_log;
2105 ulong_t mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2109 zfs_fuid_info_t *fuidp = NULL;
2110 boolean_t fuid_dirtied;
2114 return (SET_ERROR(ENOSYS));
2116 if (zp->z_pflags & ZFS_IMMUTABLE)
2117 return (SET_ERROR(EPERM));
2119 if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)))
2122 error = zfs_vsec_2_aclp(zfsvfs, ZTOI(zp)->i_mode, vsecp, cr, &fuidp,
2128 * If ACL wide flags aren't being set then preserve any
2131 if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2133 (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2136 mutex_enter(&zp->z_acl_lock);
2137 mutex_enter(&zp->z_lock);
2139 tx = dmu_tx_create(zfsvfs->z_os);
2141 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2143 fuid_dirtied = zfsvfs->z_fuid_dirty;
2145 zfs_fuid_txhold(zfsvfs, tx);
2148 * If old version and ACL won't fit in bonus and we aren't
2149 * upgrading then take out necessary DMU holds
2152 if ((acl_obj = zfs_external_acl(zp)) != 0) {
2153 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2154 zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2155 dmu_tx_hold_free(tx, acl_obj, 0,
2157 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2160 dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2162 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2163 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2166 zfs_sa_upgrade_txholds(tx, zp);
2167 error = dmu_tx_assign(tx, TXG_NOWAIT);
2169 mutex_exit(&zp->z_acl_lock);
2170 mutex_exit(&zp->z_lock);
2172 if (error == ERESTART) {
2182 error = zfs_aclset_common(zp, aclp, cr, tx);
2184 ASSERT(zp->z_acl_cached == NULL);
2185 zp->z_acl_cached = aclp;
2188 zfs_fuid_sync(zfsvfs, tx);
2190 zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2193 zfs_fuid_info_free(fuidp);
2196 mutex_exit(&zp->z_lock);
2197 mutex_exit(&zp->z_acl_lock);
2203 * Check accesses of interest (AoI) against attributes of the dataset
2204 * such as read-only. Returns zero if no AoI conflict with dataset
2205 * attributes, otherwise an appropriate errno is returned.
2208 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2210 if ((v4_mode & WRITE_MASK) && (zfs_is_readonly(ZTOZSB(zp))) &&
2211 (!Z_ISDEV(ZTOI(zp)->i_mode) ||
2212 (Z_ISDEV(ZTOI(zp)->i_mode) && (v4_mode & WRITE_MASK_ATTRS)))) {
2213 return (SET_ERROR(EROFS));
2217 * Only check for READONLY on non-directories.
2219 if ((v4_mode & WRITE_MASK_DATA) &&
2220 ((!S_ISDIR(ZTOI(zp)->i_mode) &&
2221 (zp->z_pflags & (ZFS_READONLY | ZFS_IMMUTABLE))) ||
2222 (S_ISDIR(ZTOI(zp)->i_mode) &&
2223 (zp->z_pflags & ZFS_IMMUTABLE)))) {
2224 return (SET_ERROR(EPERM));
2227 if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2228 (zp->z_pflags & ZFS_NOUNLINK)) {
2229 return (SET_ERROR(EPERM));
2232 if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2233 (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2234 return (SET_ERROR(EACCES));
2241 * The primary usage of this function is to loop through all of the
2242 * ACEs in the znode, determining what accesses of interest (AoI) to
2243 * the caller are allowed or denied. The AoI are expressed as bits in
2244 * the working_mode parameter. As each ACE is processed, bits covered
2245 * by that ACE are removed from the working_mode. This removal
2246 * facilitates two things. The first is that when the working mode is
2247 * empty (= 0), we know we've looked at all the AoI. The second is
2248 * that the ACE interpretation rules don't allow a later ACE to undo
2249 * something granted or denied by an earlier ACE. Removing the
2250 * discovered access or denial enforces this rule. At the end of
2251 * processing the ACEs, all AoI that were found to be denied are
2252 * placed into the working_mode, giving the caller a mask of denied
2253 * accesses. Returns:
2254 * 0 if all AoI granted
2255 * EACCES if the denied mask is non-zero
2256 * other error if abnormal failure (e.g., IO error)
2258 * A secondary usage of the function is to determine if any of the
2259 * AoI are granted. If an ACE grants any access in
2260 * the working_mode, we immediately short circuit out of the function.
2261 * This mode is chosen by setting anyaccess to B_TRUE. The
2262 * working_mode is not a denied access mask upon exit if the function
2263 * is used in this manner.
2266 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2267 boolean_t anyaccess, cred_t *cr)
2269 zfsvfs_t *zfsvfs = ZTOZSB(zp);
2272 uid_t uid = crgetuid(cr);
2274 uint16_t type, iflags;
2275 uint16_t entry_type;
2276 uint32_t access_mask;
2277 uint32_t deny_mask = 0;
2278 zfs_ace_hdr_t *acep = NULL;
2283 zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2285 mutex_enter(&zp->z_acl_lock);
2287 error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2289 mutex_exit(&zp->z_acl_lock);
2293 ASSERT(zp->z_acl_cached);
2295 while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2297 uint32_t mask_matched;
2299 if (!zfs_acl_valid_ace_type(type, iflags))
2302 if (S_ISDIR(ZTOI(zp)->i_mode) &&
2303 (iflags & ACE_INHERIT_ONLY_ACE))
2306 /* Skip ACE if it does not affect any AoI */
2307 mask_matched = (access_mask & *working_mode);
2311 entry_type = (iflags & ACE_TYPE_FLAGS);
2315 switch (entry_type) {
2323 case ACE_IDENTIFIER_GROUP:
2324 checkit = zfs_groupmember(zfsvfs, who, cr);
2332 if (entry_type == 0) {
2335 newid = zfs_fuid_map_id(zfsvfs, who, cr,
2337 if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2342 mutex_exit(&zp->z_acl_lock);
2343 return (SET_ERROR(EIO));
2349 DTRACE_PROBE3(zfs__ace__denies,
2351 zfs_ace_hdr_t *, acep,
2352 uint32_t, mask_matched);
2353 deny_mask |= mask_matched;
2355 DTRACE_PROBE3(zfs__ace__allows,
2357 zfs_ace_hdr_t *, acep,
2358 uint32_t, mask_matched);
2360 mutex_exit(&zp->z_acl_lock);
2364 *working_mode &= ~mask_matched;
2368 if (*working_mode == 0)
2372 mutex_exit(&zp->z_acl_lock);
2374 /* Put the found 'denies' back on the working mode */
2376 *working_mode |= deny_mask;
2377 return (SET_ERROR(EACCES));
2378 } else if (*working_mode) {
2386 * Return true if any access whatsoever granted, we don't actually
2387 * care what access is granted.
2390 zfs_has_access(znode_t *zp, cred_t *cr)
2392 uint32_t have = ACE_ALL_PERMS;
2394 if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2397 owner = zfs_fuid_map_id(ZTOZSB(zp),
2398 KUID_TO_SUID(ZTOI(zp)->i_uid), cr, ZFS_OWNER);
2399 return (secpolicy_vnode_any_access(cr, ZTOI(zp), owner) == 0);
2405 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2406 boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2408 zfsvfs_t *zfsvfs = ZTOZSB(zp);
2411 *working_mode = v4_mode;
2412 *check_privs = B_TRUE;
2415 * Short circuit empty requests
2417 if (v4_mode == 0 || zfsvfs->z_replay) {
2422 if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2423 *check_privs = B_FALSE;
2428 * The caller requested that the ACL check be skipped. This
2429 * would only happen if the caller checked VOP_ACCESS() with a
2430 * 32 bit ACE mask and already had the appropriate permissions.
2437 return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2441 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2444 if (*working_mode != ACE_WRITE_DATA)
2445 return (SET_ERROR(EACCES));
2447 return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2448 check_privs, B_FALSE, cr));
2452 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2454 boolean_t owner = B_FALSE;
2455 boolean_t groupmbr = B_FALSE;
2457 uid_t uid = crgetuid(cr);
2460 if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2461 return (SET_ERROR(EACCES));
2463 is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2464 (S_ISDIR(ZTOI(zdp)->i_mode)));
2469 mutex_enter(&zdp->z_acl_lock);
2471 if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2472 mutex_exit(&zdp->z_acl_lock);
2476 if (KUID_TO_SUID(ZTOI(zdp)->i_uid) != 0 ||
2477 KGID_TO_SGID(ZTOI(zdp)->i_gid) != 0) {
2478 mutex_exit(&zdp->z_acl_lock);
2482 if (uid == KUID_TO_SUID(ZTOI(zdp)->i_uid)) {
2484 if (zdp->z_mode & S_IXUSR) {
2485 mutex_exit(&zdp->z_acl_lock);
2488 mutex_exit(&zdp->z_acl_lock);
2492 if (groupmember(KGID_TO_SGID(ZTOI(zdp)->i_gid), cr)) {
2494 if (zdp->z_mode & S_IXGRP) {
2495 mutex_exit(&zdp->z_acl_lock);
2498 mutex_exit(&zdp->z_acl_lock);
2502 if (!owner && !groupmbr) {
2503 if (zdp->z_mode & S_IXOTH) {
2504 mutex_exit(&zdp->z_acl_lock);
2509 mutex_exit(&zdp->z_acl_lock);
2512 DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2513 ZFS_ENTER(ZTOZSB(zdp));
2514 error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr);
2515 ZFS_EXIT(ZTOZSB(zdp));
2520 * Determine whether Access should be granted/denied.
2522 * The least priv subsystem is always consulted as a basic privilege
2523 * can define any form of access.
2526 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2528 uint32_t working_mode;
2531 boolean_t check_privs;
2533 znode_t *check_zp = zp;
2537 is_attr = ((zp->z_pflags & ZFS_XATTR) && S_ISDIR(ZTOI(zp)->i_mode));
2540 * If attribute then validate against base file
2543 if ((error = zfs_zget(ZTOZSB(zp),
2544 zp->z_xattr_parent, &xzp)) != 0) {
2551 * fixup mode to map to xattr perms
2554 if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2555 mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2556 mode |= ACE_WRITE_NAMED_ATTRS;
2559 if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2560 mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2561 mode |= ACE_READ_NAMED_ATTRS;
2565 owner = zfs_fuid_map_id(ZTOZSB(zp), KUID_TO_SUID(ZTOI(zp)->i_uid),
2568 * Map the bits required to the standard inode flags
2569 * S_IRUSR|S_IWUSR|S_IXUSR in the needed_bits. Map the bits
2570 * mapped by working_mode (currently missing) in missing_bits.
2571 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2576 working_mode = mode;
2577 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2578 owner == crgetuid(cr))
2579 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2581 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2582 ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2583 needed_bits |= S_IRUSR;
2584 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2585 ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2586 needed_bits |= S_IWUSR;
2587 if (working_mode & ACE_EXECUTE)
2588 needed_bits |= S_IXUSR;
2590 if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2591 &check_privs, skipaclchk, cr)) == 0) {
2594 return (secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2595 needed_bits, needed_bits));
2598 if (error && !check_privs) {
2604 if (error && (flags & V_APPEND)) {
2605 error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2608 if (error && check_privs) {
2609 mode_t checkmode = 0;
2612 * First check for implicit owner permission on
2613 * read_acl/read_attributes
2617 ASSERT(working_mode != 0);
2619 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2620 owner == crgetuid(cr)))
2621 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2623 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2624 ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2625 checkmode |= S_IRUSR;
2626 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2627 ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2628 checkmode |= S_IWUSR;
2629 if (working_mode & ACE_EXECUTE)
2630 checkmode |= S_IXUSR;
2632 error = secpolicy_vnode_access2(cr, ZTOI(check_zp), owner,
2633 needed_bits & ~checkmode, needed_bits);
2635 if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2636 error = secpolicy_vnode_chown(cr, owner);
2637 if (error == 0 && (working_mode & ACE_WRITE_ACL))
2638 error = secpolicy_vnode_setdac(cr, owner);
2640 if (error == 0 && (working_mode &
2641 (ACE_DELETE|ACE_DELETE_CHILD)))
2642 error = secpolicy_vnode_remove(cr);
2644 if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2645 error = secpolicy_vnode_chown(cr, owner);
2649 * See if any bits other than those already checked
2650 * for are still present. If so then return EACCES
2652 if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2653 error = SET_ERROR(EACCES);
2656 } else if (error == 0) {
2657 error = secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2658 needed_bits, needed_bits);
2668 * Translate traditional unix S_IRUSR/S_IWUSR/S_IXUSR mode into
2669 * NFSv4-style ZFS ACL format and call zfs_zaccess()
2672 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2674 return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2678 * Access function for secpolicy_vnode_setattr
2681 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2683 int v4_mode = zfs_unix_to_v4(mode >> 6);
2685 return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2688 /* See zfs_zaccess_delete() */
2689 int zfs_write_implies_delete_child = 1;
2692 * Determine whether delete access should be granted.
2694 * The following chart outlines how we handle delete permissions which is
2695 * how recent versions of windows (Windows 2008) handles it. The efficiency
2696 * comes from not having to check the parent ACL where the object itself grants
2699 * -------------------------------------------------------
2700 * | Parent Dir | Target Object Permissions |
2702 * -------------------------------------------------------
2703 * | | ACL Allows | ACL Denies| Delete |
2704 * | | Delete | Delete | unspecified|
2705 * -------------------------------------------------------
2706 * | ACL Allows | Permit | Deny * | Permit |
2707 * | DELETE_CHILD | | | |
2708 * -------------------------------------------------------
2709 * | ACL Denies | Permit | Deny | Deny |
2710 * | DELETE_CHILD | | | |
2711 * -------------------------------------------------------
2712 * | ACL specifies | | | |
2713 * | only allow | Permit | Deny * | Permit |
2714 * | write and | | | |
2716 * -------------------------------------------------------
2717 * | ACL denies | | | |
2718 * | write and | Permit | Deny | Deny |
2720 * -------------------------------------------------------
2723 * Re. execute permission on the directory: if that's missing,
2724 * the vnode lookup of the target will fail before we get here.
2726 * Re [*] in the table above: NFSv4 would normally Permit delete for
2727 * these two cells of the matrix.
2728 * See acl.h for notes on which ACE_... flags should be checked for which
2729 * operations. Specifically, the NFSv4 committee recommendation is in
2730 * conflict with the Windows interpretation of DENY ACEs, where DENY ACEs
2731 * should take precedence ahead of ALLOW ACEs.
2733 * This implementation always consults the target object's ACL first.
2734 * If a DENY ACE is present on the target object that specifies ACE_DELETE,
2735 * delete access is denied. If an ALLOW ACE with ACE_DELETE is present on
2736 * the target object, access is allowed. If and only if no entries with
2737 * ACE_DELETE are present in the object's ACL, check the container's ACL
2738 * for entries with ACE_DELETE_CHILD.
2740 * A summary of the logic implemented from the table above is as follows:
2742 * First check for DENY ACEs that apply.
2743 * If either target or container has a deny, EACCES.
2745 * Delete access can then be summarized as follows:
2746 * 1: The object to be deleted grants ACE_DELETE, or
2747 * 2: The containing directory grants ACE_DELETE_CHILD.
2748 * In a Windows system, that would be the end of the story.
2749 * In this system, (2) has some complications...
2750 * 2a: "sticky" bit on a directory adds restrictions, and
2751 * 2b: existing ACEs from previous versions of ZFS may
2752 * not carry ACE_DELETE_CHILD where they should, so we
2753 * also allow delete when ACE_WRITE_DATA is granted.
2755 * Note: 2b is technically a work-around for a prior bug,
2756 * which hopefully can go away some day. For those who
2757 * no longer need the work around, and for testing, this
2758 * work-around is made conditional via the tunable:
2759 * zfs_write_implies_delete_child
2762 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2764 uint32_t wanted_dirperms;
2765 uint32_t dzp_working_mode = 0;
2766 uint32_t zp_working_mode = 0;
2767 int dzp_error, zp_error;
2768 boolean_t dzpcheck_privs;
2769 boolean_t zpcheck_privs;
2771 if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2772 return (SET_ERROR(EPERM));
2776 * If target object grants ACE_DELETE then we are done. This is
2777 * indicated by a return value of 0. For this case we don't worry
2778 * about the sticky bit because sticky only applies to the parent
2779 * directory and this is the child access result.
2781 * If we encounter a DENY ACE here, we're also done (EACCES).
2782 * Note that if we hit a DENY ACE here (on the target) it should
2783 * take precedence over a DENY ACE on the container, so that when
2784 * we have more complete auditing support we will be able to
2785 * report an access failure against the specific target.
2786 * (This is part of why we're checking the target first.)
2788 zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2789 &zpcheck_privs, B_FALSE, cr);
2790 if (zp_error == EACCES) {
2791 /* We hit a DENY ACE. */
2793 return (SET_ERROR(zp_error));
2794 return (secpolicy_vnode_remove(cr));
2802 * If the containing directory grants ACE_DELETE_CHILD,
2803 * or we're in backward compatibility mode and the
2804 * containing directory has ACE_WRITE_DATA, allow.
2805 * Case 2b is handled with wanted_dirperms.
2807 wanted_dirperms = ACE_DELETE_CHILD;
2808 if (zfs_write_implies_delete_child)
2809 wanted_dirperms |= ACE_WRITE_DATA;
2810 dzp_error = zfs_zaccess_common(dzp, wanted_dirperms,
2811 &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2812 if (dzp_error == EACCES) {
2813 /* We hit a DENY ACE. */
2814 if (!dzpcheck_privs)
2815 return (SET_ERROR(dzp_error));
2816 return (secpolicy_vnode_remove(cr));
2820 * Cases 2a, 2b (continued)
2822 * Note: dzp_working_mode now contains any permissions
2823 * that were NOT granted. Therefore, if any of the
2824 * wanted_dirperms WERE granted, we will have:
2825 * dzp_working_mode != wanted_dirperms
2826 * We're really asking if ANY of those permissions
2827 * were granted, and if so, grant delete access.
2829 if (dzp_working_mode != wanted_dirperms)
2833 * dzp_error is 0 if the container granted us permissions to "modify".
2834 * If we do not have permission via one or more ACEs, our current
2835 * privileges may still permit us to modify the container.
2837 * dzpcheck_privs is false when i.e. the FS is read-only.
2838 * Otherwise, do privilege checks for the container.
2840 if (dzp_error != 0 && dzpcheck_privs) {
2844 * The secpolicy call needs the requested access and
2845 * the current access mode of the container, but it
2846 * only knows about Unix-style modes (VEXEC, VWRITE),
2847 * so this must condense the fine-grained ACE bits into
2850 * The VEXEC flag is easy, because we know that has
2851 * always been checked before we get here (during the
2852 * lookup of the target vnode). The container has not
2853 * granted us permissions to "modify", so we do not set
2854 * the VWRITE flag in the current access mode.
2856 owner = zfs_fuid_map_id(ZTOZSB(dzp),
2857 KUID_TO_SUID(ZTOI(dzp)->i_uid), cr, ZFS_OWNER);
2858 dzp_error = secpolicy_vnode_access2(cr, ZTOI(dzp),
2859 owner, S_IXUSR, S_IWUSR|S_IXUSR);
2861 if (dzp_error != 0) {
2863 * Note: We may have dzp_error = -1 here (from
2864 * zfs_zacess_common). Don't return that.
2866 return (SET_ERROR(EACCES));
2871 * At this point, we know that the directory permissions allow
2872 * us to modify, but we still need to check for the additional
2873 * restrictions that apply when the "sticky bit" is set.
2875 * Yes, zfs_sticky_remove_access() also checks this bit, but
2876 * checking it here and skipping the call below is nice when
2877 * you're watching all of this with dtrace.
2879 if ((dzp->z_mode & S_ISVTX) == 0)
2883 * zfs_sticky_remove_access will succeed if:
2884 * 1. The sticky bit is absent.
2885 * 2. We pass the sticky bit restrictions.
2886 * 3. We have privileges that always allow file removal.
2888 return (zfs_sticky_remove_access(dzp, zp, cr));
2892 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2893 znode_t *tzp, cred_t *cr)
2898 if (szp->z_pflags & ZFS_AV_QUARANTINED)
2899 return (SET_ERROR(EACCES));
2901 add_perm = S_ISDIR(ZTOI(szp)->i_mode) ?
2902 ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2905 * Rename permissions are combination of delete permission +
2906 * add file/subdir permission.
2910 * first make sure we do the delete portion.
2912 * If that succeeds then check for add_file/add_subdir permissions
2915 if ((error = zfs_zaccess_delete(sdzp, szp, cr)))
2919 * If we have a tzp, see if we can delete it?
2922 if ((error = zfs_zaccess_delete(tdzp, tzp, cr)))
2927 * Now check for add permissions
2929 error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);