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 2017 Nexenta Systems, Inc. All rights reserved.
27 #include <sys/types.h>
28 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
32 #include <sys/resource.h>
34 #include <sys/vnode.h>
38 #include <sys/cmn_err.h>
39 #include <sys/errno.h>
40 #include <sys/unistd.h>
42 #include <sys/fs/zfs.h>
43 #include <sys/policy.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_fuid.h>
46 #include <sys/zfs_acl.h>
47 #include <sys/zfs_dir.h>
48 #include <sys/zfs_quota.h>
49 #include <sys/zfs_vfsops.h>
51 #include <sys/dnode.h>
54 #include <acl/acl_common.h>
57 #define ALLOW ACE_ACCESS_ALLOWED_ACE_TYPE
58 #define DENY ACE_ACCESS_DENIED_ACE_TYPE
59 #define MAX_ACE_TYPE ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
60 #define MIN_ACE_TYPE ALLOW
62 #define OWNING_GROUP (ACE_GROUP|ACE_IDENTIFIER_GROUP)
63 #define EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
64 ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
65 #define EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
66 ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
67 #define OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
68 ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
70 #define ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
71 ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
72 ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
73 ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
75 #define WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
76 #define WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
77 ACE_DELETE|ACE_DELETE_CHILD)
78 #define WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
80 #define OGE_CLEAR (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
81 ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
83 #define OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
84 ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
86 #define ALL_INHERIT (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
87 ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
89 #define RESTRICTED_CLEAR (ACE_WRITE_ACL|ACE_WRITE_OWNER)
91 #define V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
94 #define ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
97 #define ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
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 = {
177 zfs_ace_v0_get_flags,
178 zfs_ace_v0_set_flags,
184 zfs_ace_v0_abstract_size,
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 zfs_ace_fuid_get_mask,
312 zfs_ace_fuid_set_mask,
313 zfs_ace_fuid_get_flags,
314 zfs_ace_fuid_set_flags,
315 zfs_ace_fuid_get_type,
316 zfs_ace_fuid_set_type,
317 zfs_ace_fuid_get_who,
318 zfs_ace_fuid_set_who,
320 zfs_ace_fuid_abstract_size,
321 zfs_ace_fuid_mask_off,
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(zp->z_zfsvfs),
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 = zp->z_zfsvfs;
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(zp->z_zfsvfs),
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(zp->z_zfsvfs->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(vtype_t obj_type, 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 (obj_type == VDIR &&
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, vtype_t obj_type, zfs_acl_t *aclp,
653 void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
654 zfs_fuid_info_t **fuidp, cred_t *cr)
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_type, 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(vtype_t obj_type, 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_type, 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(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
830 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(znode_t *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;
1064 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1065 if (zp->z_zfsvfs->z_replay == B_FALSE)
1066 ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1068 if (zp->z_acl_cached && !will_modify) {
1069 *aclpp = zp->z_acl_cached;
1073 version = zfs_znode_acl_version(zp);
1075 if ((error = zfs_acl_znode_info(zp, &aclsize,
1076 &acl_count, &znode_acl)) != 0) {
1080 aclp = zfs_acl_alloc(version);
1082 aclp->z_acl_count = acl_count;
1083 aclp->z_acl_bytes = aclsize;
1085 aclnode = zfs_acl_node_alloc(aclsize);
1086 aclnode->z_ace_count = aclp->z_acl_count;
1087 aclnode->z_size = aclsize;
1090 if (znode_acl.z_acl_extern_obj) {
1091 error = dmu_read(zp->z_zfsvfs->z_os,
1092 znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1093 aclnode->z_acldata, DMU_READ_PREFETCH);
1095 bcopy(znode_acl.z_ace_data, aclnode->z_acldata,
1099 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1100 aclnode->z_acldata, aclnode->z_size);
1105 zfs_acl_node_free(aclnode);
1106 /* convert checksum errors into IO errors */
1107 if (error == ECKSUM)
1108 error = SET_ERROR(EIO);
1112 list_insert_head(&aclp->z_acl, aclnode);
1116 zp->z_acl_cached = aclp;
1123 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1124 boolean_t start, void *userdata)
1126 zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1129 cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1131 cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1134 *dataptr = cb->cb_acl_node->z_acldata;
1135 *length = cb->cb_acl_node->z_size;
1139 zfs_acl_chown_setattr(znode_t *zp)
1144 if (zp->z_zfsvfs->z_replay == B_FALSE)
1145 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1146 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1147 ASSERT_VOP_IN_SEQC(ZTOV(zp));
1149 if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1150 zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1151 &zp->z_pflags, zp->z_uid, zp->z_gid);
1156 * common code for setting ACLs.
1158 * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1159 * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1160 * already checked the acl and knows whether to inherit.
1163 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1166 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1167 dmu_object_type_t otype;
1168 zfs_acl_locator_cb_t locate = { 0 };
1170 sa_bulk_attr_t bulk[5];
1173 zfs_acl_phys_t acl_phys;
1175 ASSERT_VOP_IN_SEQC(ZTOV(zp));
1179 mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1180 zp->z_uid, zp->z_gid);
1183 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1184 &mode, sizeof (mode));
1185 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1186 &zp->z_pflags, sizeof (zp->z_pflags));
1187 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1188 &ctime, sizeof (ctime));
1190 if (zp->z_acl_cached) {
1191 zfs_acl_free(zp->z_acl_cached);
1192 zp->z_acl_cached = NULL;
1198 if (!zfsvfs->z_use_fuids) {
1199 otype = DMU_OT_OLDACL;
1201 if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1202 (zfsvfs->z_version >= ZPL_VERSION_FUID))
1203 zfs_acl_xform(zp, aclp, cr);
1204 ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1209 * Arrgh, we have to handle old on disk format
1210 * as well as newer (preferred) SA format.
1213 if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1214 locate.cb_aclp = aclp;
1215 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1216 zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1217 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1218 NULL, &aclp->z_acl_count, sizeof (uint64_t));
1219 } else { /* Painful legacy way */
1220 zfs_acl_node_t *aclnode;
1224 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1225 &acl_phys, sizeof (acl_phys))) != 0)
1228 aoid = acl_phys.z_acl_extern_obj;
1230 if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1232 * If ACL was previously external and we are now
1233 * converting to new ACL format then release old
1234 * ACL object and create a new one.
1237 aclp->z_version != acl_phys.z_acl_version) {
1238 error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1244 aoid = dmu_object_alloc(zfsvfs->z_os,
1245 otype, aclp->z_acl_bytes,
1246 otype == DMU_OT_ACL ?
1247 DMU_OT_SYSACL : DMU_OT_NONE,
1248 otype == DMU_OT_ACL ?
1249 DN_OLD_MAX_BONUSLEN : 0, tx);
1251 (void) dmu_object_set_blocksize(zfsvfs->z_os,
1252 aoid, aclp->z_acl_bytes, 0, tx);
1254 acl_phys.z_acl_extern_obj = aoid;
1255 for (aclnode = list_head(&aclp->z_acl); aclnode;
1256 aclnode = list_next(&aclp->z_acl, aclnode)) {
1257 if (aclnode->z_ace_count == 0)
1259 dmu_write(zfsvfs->z_os, aoid, off,
1260 aclnode->z_size, aclnode->z_acldata, tx);
1261 off += aclnode->z_size;
1264 void *start = acl_phys.z_ace_data;
1266 * Migrating back embedded?
1268 if (acl_phys.z_acl_extern_obj) {
1269 error = dmu_object_free(zfsvfs->z_os,
1270 acl_phys.z_acl_extern_obj, tx);
1273 acl_phys.z_acl_extern_obj = 0;
1276 for (aclnode = list_head(&aclp->z_acl); aclnode;
1277 aclnode = list_next(&aclp->z_acl, aclnode)) {
1278 if (aclnode->z_ace_count == 0)
1280 bcopy(aclnode->z_acldata, start,
1282 start = (caddr_t)start + aclnode->z_size;
1286 * If Old version then swap count/bytes to match old
1287 * layout of znode_acl_phys_t.
1289 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1290 acl_phys.z_acl_size = aclp->z_acl_count;
1291 acl_phys.z_acl_count = aclp->z_acl_bytes;
1293 acl_phys.z_acl_size = aclp->z_acl_bytes;
1294 acl_phys.z_acl_count = aclp->z_acl_count;
1296 acl_phys.z_acl_version = aclp->z_version;
1298 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1299 &acl_phys, sizeof (acl_phys));
1303 * Replace ACL wide bits, but first clear them.
1305 zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1307 zp->z_pflags |= aclp->z_hints;
1309 if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1310 zp->z_pflags |= ZFS_ACL_TRIVIAL;
1312 zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1313 return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1317 zfs_acl_chmod(vtype_t vtype, uint64_t mode, boolean_t split, boolean_t trim,
1322 int new_count, new_bytes;
1325 uint16_t iflags, type;
1326 uint32_t access_mask;
1327 zfs_acl_node_t *newnode;
1328 size_t abstract_size = aclp->z_ops->ace_abstract_size();
1331 trivial_acl_t masks;
1333 new_count = new_bytes = 0;
1335 isdir = (vtype == VDIR);
1337 acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1339 newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1341 zacep = newnode->z_acldata;
1343 zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1344 zacep = (void *)((uintptr_t)zacep + abstract_size);
1346 new_bytes += abstract_size;
1349 zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1350 zacep = (void *)((uintptr_t)zacep + abstract_size);
1352 new_bytes += abstract_size;
1355 zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1356 zacep = (void *)((uintptr_t)zacep + abstract_size);
1358 new_bytes += abstract_size;
1361 while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1363 entry_type = (iflags & ACE_TYPE_FLAGS);
1365 * ACEs used to represent the file mode may be divided
1366 * into an equivalent pair of inherit-only and regular
1367 * ACEs, if they are inheritable.
1368 * Skip regular ACEs, which are replaced by the new mode.
1370 if (split && (entry_type == ACE_OWNER ||
1371 entry_type == OWNING_GROUP ||
1372 entry_type == ACE_EVERYONE)) {
1373 if (!isdir || !(iflags &
1374 (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1377 * We preserve owner@, group@, or @everyone
1378 * permissions, if they are inheritable, by
1379 * copying them to inherit_only ACEs. This
1380 * prevents inheritable permissions from being
1381 * altered along with the file mode.
1383 iflags |= ACE_INHERIT_ONLY_ACE;
1387 * If this ACL has any inheritable ACEs, mark that in
1388 * the hints (which are later masked into the pflags)
1389 * so create knows to do inheritance.
1391 if (isdir && (iflags &
1392 (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1393 aclp->z_hints |= ZFS_INHERIT_ACE;
1395 if ((type != ALLOW && type != DENY) ||
1396 (iflags & ACE_INHERIT_ONLY_ACE)) {
1398 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1399 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1400 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1401 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1402 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1407 * Limit permissions granted by ACEs to be no greater
1408 * than permissions of the requested group mode.
1409 * Applies when the "aclmode" property is set to
1412 if ((type == ALLOW) && trim)
1413 access_mask &= masks.group;
1415 zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1416 ace_size = aclp->z_ops->ace_size(acep);
1417 zacep = (void *)((uintptr_t)zacep + ace_size);
1419 new_bytes += ace_size;
1421 zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1422 zacep = (void *)((uintptr_t)zacep + abstract_size);
1423 zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1424 zacep = (void *)((uintptr_t)zacep + abstract_size);
1425 zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1428 new_bytes += abstract_size * 3;
1429 zfs_acl_release_nodes(aclp);
1430 aclp->z_acl_count = new_count;
1431 aclp->z_acl_bytes = new_bytes;
1432 newnode->z_ace_count = new_count;
1433 newnode->z_size = new_bytes;
1434 list_insert_tail(&aclp->z_acl, newnode);
1438 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1442 mutex_enter(&zp->z_acl_lock);
1443 if (zp->z_zfsvfs->z_replay == B_FALSE)
1444 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1445 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1446 *aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1448 error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1451 (*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1452 zfs_acl_chmod(ZTOV(zp)->v_type, mode, B_TRUE,
1453 (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1455 mutex_exit(&zp->z_acl_lock);
1461 * Should ACE be inherited?
1464 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1466 int iflags = (acep_flags & 0xf);
1468 if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1470 else if (iflags & ACE_FILE_INHERIT_ACE)
1471 return (!((vtype == VDIR) &&
1472 (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1477 * inherit inheritable ACEs from parent
1480 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1481 uint64_t mode, boolean_t *need_chmod)
1485 zfs_acl_node_t *aclnode;
1486 zfs_acl_t *aclp = NULL;
1488 uint32_t access_mask;
1489 uint16_t iflags, newflags, type;
1491 void *data1, *data2;
1492 size_t data1sz, data2sz;
1494 boolean_t isdir = (vtype == VDIR);
1495 boolean_t isreg = (vtype == VREG);
1497 *need_chmod = B_TRUE;
1499 aclp = zfs_acl_alloc(paclp->z_version);
1500 aclinherit = zfsvfs->z_acl_inherit;
1501 if (aclinherit == ZFS_ACL_DISCARD || vtype == VLNK)
1504 while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1505 &access_mask, &iflags, &type))) {
1508 * don't inherit bogus ACEs
1510 if (!zfs_acl_valid_ace_type(type, iflags))
1514 * Check if ACE is inheritable by this vnode
1516 if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1517 !zfs_ace_can_use(vtype, iflags))
1521 * If owner@, group@, or everyone@ inheritable
1522 * then zfs_acl_chmod() isn't needed.
1524 if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1525 aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1526 ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1527 ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1528 (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1529 *need_chmod = B_FALSE;
1532 * Strip inherited execute permission from file if
1535 if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1536 !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1537 access_mask &= ~ACE_EXECUTE;
1541 * Strip write_acl and write_owner from permissions
1542 * when inheriting an ACE
1544 if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1545 access_mask &= ~RESTRICTED_CLEAR;
1548 ace_size = aclp->z_ops->ace_size(pacep);
1549 aclnode = zfs_acl_node_alloc(ace_size);
1550 list_insert_tail(&aclp->z_acl, aclnode);
1551 acep = aclnode->z_acldata;
1553 zfs_set_ace(aclp, acep, access_mask, type,
1554 who, iflags|ACE_INHERITED_ACE);
1557 * Copy special opaque data if any
1559 if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1560 VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1561 &data2)) == data1sz);
1562 bcopy(data1, data2, data2sz);
1565 aclp->z_acl_count++;
1566 aclnode->z_ace_count++;
1567 aclp->z_acl_bytes += aclnode->z_size;
1568 newflags = aclp->z_ops->ace_flags_get(acep);
1571 * If ACE is not to be inherited further, or if the vnode is
1572 * not a directory, remove all inheritance flags
1574 if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1575 newflags &= ~ALL_INHERIT;
1576 aclp->z_ops->ace_flags_set(acep,
1577 newflags|ACE_INHERITED_ACE);
1582 * This directory has an inheritable ACE
1584 aclp->z_hints |= ZFS_INHERIT_ACE;
1587 * If only FILE_INHERIT is set then turn on
1590 if ((iflags & (ACE_FILE_INHERIT_ACE |
1591 ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1592 newflags |= ACE_INHERIT_ONLY_ACE;
1593 aclp->z_ops->ace_flags_set(acep,
1594 newflags|ACE_INHERITED_ACE);
1596 newflags &= ~ACE_INHERIT_ONLY_ACE;
1597 aclp->z_ops->ace_flags_set(acep,
1598 newflags|ACE_INHERITED_ACE);
1606 * Create file system object initial permissions
1607 * including inheritable ACEs.
1608 * Also, create FUIDs for owner and group.
1611 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1612 vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids)
1615 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
1618 boolean_t need_chmod = B_TRUE;
1619 boolean_t trim = B_FALSE;
1620 boolean_t inherited = B_FALSE;
1622 if ((flag & IS_ROOT_NODE) == 0) {
1623 if (zfsvfs->z_replay == B_FALSE)
1624 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
1626 ASSERT(dzp->z_vnode == NULL);
1627 bzero(acl_ids, sizeof (zfs_acl_ids_t));
1628 acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1631 if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1632 &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1635 * Determine uid and gid.
1637 if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1638 ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1639 acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1640 (uint64_t)vap->va_uid, cr,
1641 ZFS_OWNER, &acl_ids->z_fuidp);
1642 acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1643 (uint64_t)vap->va_gid, cr,
1644 ZFS_GROUP, &acl_ids->z_fuidp);
1647 acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1648 cr, &acl_ids->z_fuidp);
1649 acl_ids->z_fgid = 0;
1650 if (vap->va_mask & AT_GID) {
1651 acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1652 (uint64_t)vap->va_gid,
1653 cr, ZFS_GROUP, &acl_ids->z_fuidp);
1655 if (acl_ids->z_fgid != dzp->z_gid &&
1656 !groupmember(vap->va_gid, cr) &&
1657 secpolicy_vnode_create_gid(cr) != 0)
1658 acl_ids->z_fgid = 0;
1660 if (acl_ids->z_fgid == 0) {
1661 if (dzp->z_mode & S_ISGID) {
1665 acl_ids->z_fgid = dzp->z_gid;
1666 gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1669 if (zfsvfs->z_use_fuids &&
1670 IS_EPHEMERAL(acl_ids->z_fgid)) {
1671 domain = zfs_fuid_idx_domain(
1672 &zfsvfs->z_fuid_idx,
1673 FUID_INDEX(acl_ids->z_fgid));
1674 rid = FUID_RID(acl_ids->z_fgid);
1675 zfs_fuid_node_add(&acl_ids->z_fuidp,
1677 FUID_INDEX(acl_ids->z_fgid),
1678 acl_ids->z_fgid, ZFS_GROUP);
1681 acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1682 ZFS_GROUP, cr, &acl_ids->z_fuidp);
1683 #ifdef __FreeBSD_kernel__
1684 gid = acl_ids->z_fgid = dzp->z_gid;
1693 * If we're creating a directory, and the parent directory has the
1694 * set-GID bit set, set in on the new directory.
1695 * Otherwise, if the user is neither privileged nor a member of the
1696 * file's new group, clear the file's set-GID bit.
1699 if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1700 (vap->va_type == VDIR)) {
1701 acl_ids->z_mode |= S_ISGID;
1703 if ((acl_ids->z_mode & S_ISGID) &&
1704 secpolicy_vnode_setids_setgids(ZTOV(dzp), cr, gid) != 0)
1705 acl_ids->z_mode &= ~S_ISGID;
1708 if (acl_ids->z_aclp == NULL) {
1709 mutex_enter(&dzp->z_acl_lock);
1710 if (!(flag & IS_ROOT_NODE) &&
1711 (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1712 !(dzp->z_pflags & ZFS_XATTR)) {
1713 VERIFY0(zfs_acl_node_read(dzp, B_TRUE,
1715 acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1716 vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1720 zfs_acl_alloc(zfs_acl_version_zp(dzp));
1721 acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1723 mutex_exit(&dzp->z_acl_lock);
1726 if (vap->va_type == VDIR)
1727 acl_ids->z_aclp->z_hints |=
1728 ZFS_ACL_AUTO_INHERIT;
1730 if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1731 zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1732 zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1734 zfs_acl_chmod(vap->va_type, acl_ids->z_mode, B_FALSE,
1735 trim, acl_ids->z_aclp);
1739 if (inherited || vsecp) {
1740 acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1741 acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1742 acl_ids->z_fuid, acl_ids->z_fgid);
1743 if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1744 acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1751 * Free ACL and fuid_infop, but not the acl_ids structure
1754 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1756 if (acl_ids->z_aclp)
1757 zfs_acl_free(acl_ids->z_aclp);
1758 if (acl_ids->z_fuidp)
1759 zfs_fuid_info_free(acl_ids->z_fuidp);
1760 acl_ids->z_aclp = NULL;
1761 acl_ids->z_fuidp = NULL;
1765 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1767 return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1768 zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1769 (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1770 zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1774 * Retrieve a file's ACL
1777 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1785 mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1786 VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1789 return (SET_ERROR(ENOSYS));
1791 if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr)))
1794 mutex_enter(&zp->z_acl_lock);
1796 if (zp->z_zfsvfs->z_replay == B_FALSE)
1797 ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1798 error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1800 mutex_exit(&zp->z_acl_lock);
1805 * Scan ACL to determine number of ACEs
1807 if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1810 uint32_t access_mask;
1811 uint16_t type, iflags;
1813 while ((zacep = zfs_acl_next_ace(aclp, zacep,
1814 &who, &access_mask, &iflags, &type))) {
1816 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1817 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1818 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1819 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1826 vsecp->vsa_aclcnt = count;
1828 count = (int)aclp->z_acl_count;
1830 if (mask & VSA_ACECNT) {
1831 vsecp->vsa_aclcnt = count;
1834 if (mask & VSA_ACE) {
1837 aclsz = count * sizeof (ace_t) +
1838 sizeof (ace_object_t) * largeace;
1840 vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1841 vsecp->vsa_aclentsz = aclsz;
1843 if (aclp->z_version == ZFS_ACL_VERSION_FUID)
1844 zfs_copy_fuid_2_ace(zp->z_zfsvfs, aclp, cr,
1845 vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
1847 zfs_acl_node_t *aclnode;
1848 void *start = vsecp->vsa_aclentp;
1850 for (aclnode = list_head(&aclp->z_acl); aclnode;
1851 aclnode = list_next(&aclp->z_acl, aclnode)) {
1852 bcopy(aclnode->z_acldata, start,
1854 start = (caddr_t)start + aclnode->z_size;
1856 ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
1860 if (mask & VSA_ACE_ACLFLAGS) {
1861 vsecp->vsa_aclflags = 0;
1862 if (zp->z_pflags & ZFS_ACL_DEFAULTED)
1863 vsecp->vsa_aclflags |= ACL_DEFAULTED;
1864 if (zp->z_pflags & ZFS_ACL_PROTECTED)
1865 vsecp->vsa_aclflags |= ACL_PROTECTED;
1866 if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
1867 vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
1870 mutex_exit(&zp->z_acl_lock);
1876 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_type,
1877 vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
1880 zfs_acl_node_t *aclnode;
1881 int aclcnt = vsecp->vsa_aclcnt;
1884 if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1885 return (SET_ERROR(EINVAL));
1887 aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1890 aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
1891 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1892 if ((error = zfs_copy_ace_2_oldace(obj_type, aclp,
1893 (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
1894 aclcnt, &aclnode->z_size)) != 0) {
1896 zfs_acl_node_free(aclnode);
1900 if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_type, aclp,
1901 vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
1902 &aclnode->z_size, fuidp, cr)) != 0) {
1904 zfs_acl_node_free(aclnode);
1908 aclp->z_acl_bytes = aclnode->z_size;
1909 aclnode->z_ace_count = aclcnt;
1910 aclp->z_acl_count = aclcnt;
1911 list_insert_head(&aclp->z_acl, aclnode);
1914 * If flags are being set then add them to z_hints
1916 if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
1917 if (vsecp->vsa_aclflags & ACL_PROTECTED)
1918 aclp->z_hints |= ZFS_ACL_PROTECTED;
1919 if (vsecp->vsa_aclflags & ACL_DEFAULTED)
1920 aclp->z_hints |= ZFS_ACL_DEFAULTED;
1921 if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
1922 aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
1934 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1936 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1937 zilog_t *zilog = zfsvfs->z_log;
1938 ulong_t mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
1942 zfs_fuid_info_t *fuidp = NULL;
1943 boolean_t fuid_dirtied;
1946 if (zp->z_zfsvfs->z_replay == B_FALSE)
1947 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1949 return (SET_ERROR(ENOSYS));
1951 if (zp->z_pflags & ZFS_IMMUTABLE)
1952 return (SET_ERROR(EPERM));
1954 if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr)))
1957 error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1963 * If ACL wide flags aren't being set then preserve any
1966 if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1968 (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1971 mutex_enter(&zp->z_acl_lock);
1973 tx = dmu_tx_create(zfsvfs->z_os);
1975 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1977 fuid_dirtied = zfsvfs->z_fuid_dirty;
1979 zfs_fuid_txhold(zfsvfs, tx);
1982 * If old version and ACL won't fit in bonus and we aren't
1983 * upgrading then take out necessary DMU holds
1986 if ((acl_obj = zfs_external_acl(zp)) != 0) {
1987 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
1988 zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
1989 dmu_tx_hold_free(tx, acl_obj, 0,
1991 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1994 dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
1996 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1997 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2000 zfs_sa_upgrade_txholds(tx, zp);
2001 error = dmu_tx_assign(tx, TXG_NOWAIT);
2003 mutex_exit(&zp->z_acl_lock);
2005 if (error == ERESTART) {
2015 error = zfs_aclset_common(zp, aclp, cr, tx);
2017 ASSERT(zp->z_acl_cached == NULL);
2018 zp->z_acl_cached = aclp;
2021 zfs_fuid_sync(zfsvfs, tx);
2023 zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2026 zfs_fuid_info_free(fuidp);
2028 mutex_exit(&zp->z_acl_lock);
2034 * Check accesses of interest (AoI) against attributes of the dataset
2035 * such as read-only. Returns zero if no AoI conflict with dataset
2036 * attributes, otherwise an appropriate errno is returned.
2039 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2041 if ((v4_mode & WRITE_MASK) &&
2042 (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2043 (!IS_DEVVP(ZTOV(zp)) ||
2044 (IS_DEVVP(ZTOV(zp)) && (v4_mode & WRITE_MASK_ATTRS)))) {
2045 return (SET_ERROR(EROFS));
2049 * Intentionally allow ZFS_READONLY through here.
2050 * See zfs_zaccess_common().
2052 if ((v4_mode & WRITE_MASK_DATA) &&
2053 (zp->z_pflags & ZFS_IMMUTABLE)) {
2054 return (SET_ERROR(EPERM));
2058 * In FreeBSD we allow to modify directory's content is ZFS_NOUNLINK
2059 * (sunlnk) is set. We just don't allow directory removal, which is
2060 * handled in zfs_zaccess_delete().
2062 if ((v4_mode & ACE_DELETE) &&
2063 (zp->z_pflags & ZFS_NOUNLINK)) {
2067 if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2068 (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2069 return (SET_ERROR(EACCES));
2076 * The primary usage of this function is to loop through all of the
2077 * ACEs in the znode, determining what accesses of interest (AoI) to
2078 * the caller are allowed or denied. The AoI are expressed as bits in
2079 * the working_mode parameter. As each ACE is processed, bits covered
2080 * by that ACE are removed from the working_mode. This removal
2081 * facilitates two things. The first is that when the working mode is
2082 * empty (= 0), we know we've looked at all the AoI. The second is
2083 * that the ACE interpretation rules don't allow a later ACE to undo
2084 * something granted or denied by an earlier ACE. Removing the
2085 * discovered access or denial enforces this rule. At the end of
2086 * processing the ACEs, all AoI that were found to be denied are
2087 * placed into the working_mode, giving the caller a mask of denied
2088 * accesses. Returns:
2089 * 0 if all AoI granted
2090 * EACCESS if the denied mask is non-zero
2091 * other error if abnormal failure (e.g., IO error)
2093 * A secondary usage of the function is to determine if any of the
2094 * AoI are granted. If an ACE grants any access in
2095 * the working_mode, we immediately short circuit out of the function.
2096 * This mode is chosen by setting anyaccess to B_TRUE. The
2097 * working_mode is not a denied access mask upon exit if the function
2098 * is used in this manner.
2101 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2102 boolean_t anyaccess, cred_t *cr)
2104 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2107 uid_t uid = crgetuid(cr);
2109 uint16_t type, iflags;
2110 uint16_t entry_type;
2111 uint32_t access_mask;
2112 uint32_t deny_mask = 0;
2113 zfs_ace_hdr_t *acep = NULL;
2118 zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2120 mutex_enter(&zp->z_acl_lock);
2122 if (zp->z_zfsvfs->z_replay == B_FALSE)
2123 ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
2124 error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
2126 mutex_exit(&zp->z_acl_lock);
2130 ASSERT(zp->z_acl_cached);
2132 while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2134 uint32_t mask_matched;
2136 if (!zfs_acl_valid_ace_type(type, iflags))
2139 if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2142 /* Skip ACE if it does not affect any AoI */
2143 mask_matched = (access_mask & *working_mode);
2147 entry_type = (iflags & ACE_TYPE_FLAGS);
2151 switch (entry_type) {
2159 case ACE_IDENTIFIER_GROUP:
2160 checkit = zfs_groupmember(zfsvfs, who, cr);
2168 if (entry_type == 0) {
2171 newid = zfs_fuid_map_id(zfsvfs, who, cr,
2173 if (newid != UID_NOBODY &&
2178 mutex_exit(&zp->z_acl_lock);
2179 return (SET_ERROR(EIO));
2185 DTRACE_PROBE3(zfs__ace__denies,
2187 zfs_ace_hdr_t *, acep,
2188 uint32_t, mask_matched);
2189 deny_mask |= mask_matched;
2191 DTRACE_PROBE3(zfs__ace__allows,
2193 zfs_ace_hdr_t *, acep,
2194 uint32_t, mask_matched);
2196 mutex_exit(&zp->z_acl_lock);
2200 *working_mode &= ~mask_matched;
2204 if (*working_mode == 0)
2208 mutex_exit(&zp->z_acl_lock);
2210 /* Put the found 'denies' back on the working mode */
2212 *working_mode |= deny_mask;
2213 return (SET_ERROR(EACCES));
2214 } else if (*working_mode) {
2222 * Return true if any access whatsoever granted, we don't actually
2223 * care what access is granted.
2226 zfs_has_access(znode_t *zp, cred_t *cr)
2228 uint32_t have = ACE_ALL_PERMS;
2230 if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2233 owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2234 return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2240 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2241 boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2243 zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2246 *working_mode = v4_mode;
2247 *check_privs = B_TRUE;
2250 * Short circuit empty requests
2252 if (v4_mode == 0 || zfsvfs->z_replay) {
2257 if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2258 *check_privs = B_FALSE;
2263 * The caller requested that the ACL check be skipped. This
2264 * would only happen if the caller checked VOP_ACCESS() with a
2265 * 32 bit ACE mask and already had the appropriate permissions.
2273 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2274 * When that flag is set, we should behave as if write access
2275 * were not granted by anything in the ACL. In particular:
2276 * We _must_ allow writes after opening the file r/w, then
2277 * setting the DOS R/O attribute, and writing some more.
2278 * (Similar to how you can write after fchmod(fd, 0444).)
2280 * Therefore ZFS_READONLY is ignored in the dataset check
2281 * above, and checked here as if part of the ACL check.
2282 * Also note: DOS R/O is ignored for directories.
2284 if ((v4_mode & WRITE_MASK_DATA) &&
2285 (ZTOV(zp)->v_type != VDIR) &&
2286 (zp->z_pflags & ZFS_READONLY)) {
2287 return (SET_ERROR(EPERM));
2290 return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2294 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2297 if (*working_mode != ACE_WRITE_DATA)
2298 return (SET_ERROR(EACCES));
2300 return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2301 check_privs, B_FALSE, cr));
2305 * Check if VEXEC is allowed.
2307 * This routine is based on zfs_fastaccesschk_execute which has slowpath
2308 * calling zfs_zaccess. This would be incorrect on FreeBSD (see
2309 * zfs_freebsd_access for the difference). Thus this variant let's the
2310 * caller handle the slowpath (if necessary).
2312 * On top of that we perform a lockless check for ZFS_NO_EXECS_DENIED.
2314 * Safe access to znode_t is provided by the vnode lock.
2317 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2319 boolean_t owner = B_FALSE;
2320 boolean_t groupmbr = B_FALSE;
2322 uid_t uid = crgetuid(cr);
2324 if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2327 is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2328 (ZTOV(zdp)->v_type == VDIR));
2332 if (zdp->z_pflags & ZFS_NO_EXECS_DENIED)
2335 mutex_enter(&zdp->z_acl_lock);
2336 if (FUID_INDEX(zdp->z_uid) != 0 || FUID_INDEX(zdp->z_gid) != 0) {
2340 if (uid == zdp->z_uid) {
2342 if (zdp->z_mode & S_IXUSR) {
2348 if (groupmember(zdp->z_gid, cr)) {
2350 if (zdp->z_mode & S_IXGRP) {
2356 if (!owner && !groupmbr) {
2357 if (zdp->z_mode & S_IXOTH) {
2362 mutex_exit(&zdp->z_acl_lock);
2365 mutex_exit(&zdp->z_acl_lock);
2371 * Determine whether Access should be granted/denied.
2373 * The least priv subsystem is always consulted as a basic privilege
2374 * can define any form of access.
2377 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr)
2379 uint32_t working_mode;
2382 boolean_t check_privs;
2383 znode_t *xzp = NULL;
2384 znode_t *check_zp = zp;
2388 is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2390 #ifdef __FreeBSD_kernel__
2392 * In FreeBSD, we don't care about permissions of individual ADS.
2393 * Note that not checking them is not just an optimization - without
2394 * this shortcut, EA operations may bogusly fail with EACCES.
2396 if (zp->z_pflags & ZFS_XATTR)
2400 * If attribute then validate against base file
2405 if ((error = sa_lookup(zp->z_sa_hdl,
2406 SA_ZPL_PARENT(zp->z_zfsvfs), &parent,
2407 sizeof (parent))) != 0)
2410 if ((error = zfs_zget(zp->z_zfsvfs,
2411 parent, &xzp)) != 0) {
2418 * fixup mode to map to xattr perms
2421 if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2422 mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2423 mode |= ACE_WRITE_NAMED_ATTRS;
2426 if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2427 mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2428 mode |= ACE_READ_NAMED_ATTRS;
2433 owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2435 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2436 * in needed_bits. Map the bits mapped by working_mode (currently
2437 * missing) in missing_bits.
2438 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2443 working_mode = mode;
2444 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2445 owner == crgetuid(cr))
2446 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2448 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2449 ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2450 needed_bits |= VREAD;
2451 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2452 ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2453 needed_bits |= VWRITE;
2454 if (working_mode & ACE_EXECUTE)
2455 needed_bits |= VEXEC;
2457 if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2458 &check_privs, skipaclchk, cr)) == 0) {
2461 return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2462 needed_bits, needed_bits));
2465 if (error && !check_privs) {
2471 if (error && (flags & V_APPEND)) {
2472 error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2475 if (error && check_privs) {
2476 mode_t checkmode = 0;
2477 vnode_t *check_vp = ZTOV(check_zp);
2480 * First check for implicit owner permission on
2481 * read_acl/read_attributes
2485 ASSERT(working_mode != 0);
2487 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2488 owner == crgetuid(cr)))
2489 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2491 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2492 ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2494 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2495 ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2496 checkmode |= VWRITE;
2497 if (working_mode & ACE_EXECUTE)
2500 error = secpolicy_vnode_access2(cr, check_vp, owner,
2501 needed_bits & ~checkmode, needed_bits);
2503 if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2504 error = secpolicy_vnode_chown(check_vp, cr, owner);
2505 if (error == 0 && (working_mode & ACE_WRITE_ACL))
2506 error = secpolicy_vnode_setdac(check_vp, cr, owner);
2508 if (error == 0 && (working_mode &
2509 (ACE_DELETE|ACE_DELETE_CHILD)))
2510 error = secpolicy_vnode_remove(check_vp, cr);
2512 if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2513 error = secpolicy_vnode_chown(check_vp, cr, owner);
2517 * See if any bits other than those already checked
2518 * for are still present. If so then return EACCES
2520 if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2521 error = SET_ERROR(EACCES);
2524 } else if (error == 0) {
2525 error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2526 needed_bits, needed_bits);
2537 * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2538 * native ACL format and call zfs_zaccess()
2541 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr)
2543 return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr));
2547 * Access function for secpolicy_vnode_setattr
2550 zfs_zaccess_unix(znode_t *zp, mode_t mode, cred_t *cr)
2552 int v4_mode = zfs_unix_to_v4(mode >> 6);
2554 return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr));
2558 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2559 mode_t available_perms, cred_t *cr)
2564 downer = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr, ZFS_OWNER);
2566 error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2567 downer, available_perms, VWRITE|VEXEC);
2570 error = zfs_sticky_remove_access(dzp, zp, cr);
2576 * Determine whether Access should be granted/deny, without
2577 * consulting least priv subsystem.
2579 * The following chart is the recommended NFSv4 enforcement for
2580 * ability to delete an object.
2582 * -------------------------------------------------------
2583 * | Parent Dir | Target Object Permissions |
2585 * -------------------------------------------------------
2586 * | | ACL Allows | ACL Denies| Delete |
2587 * | | Delete | Delete | unspecified|
2588 * -------------------------------------------------------
2589 * | ACL Allows | Permit | Permit | Permit |
2590 * | DELETE_CHILD | |
2591 * -------------------------------------------------------
2592 * | ACL Denies | Permit | Deny | Deny |
2593 * | DELETE_CHILD | | | |
2594 * -------------------------------------------------------
2595 * | ACL specifies | | | |
2596 * | only allow | Permit | Permit | Permit |
2597 * | write and | | | |
2599 * -------------------------------------------------------
2600 * | ACL denies | | | |
2601 * | write and | Permit | Deny | Deny |
2603 * -------------------------------------------------------
2606 * No search privilege, can't even look up file?
2610 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr)
2612 uint32_t dzp_working_mode = 0;
2613 uint32_t zp_working_mode = 0;
2614 int dzp_error, zp_error;
2615 mode_t available_perms;
2616 boolean_t dzpcheck_privs = B_TRUE;
2617 boolean_t zpcheck_privs = B_TRUE;
2620 * We want specific DELETE permissions to
2621 * take precedence over WRITE/EXECUTE. We don't
2622 * want an ACL such as this to mess us up.
2623 * user:joe:write_data:deny,user:joe:delete:allow
2625 * However, deny permissions may ultimately be overridden
2626 * by secpolicy_vnode_access().
2628 * We will ask for all of the necessary permissions and then
2629 * look at the working modes from the directory and target object
2630 * to determine what was found.
2633 if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2634 return (SET_ERROR(EPERM));
2638 * If the directory permissions allow the delete, we are done.
2640 if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2641 &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2645 * If target object has delete permission then we are done
2647 if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2648 &zpcheck_privs, B_FALSE, cr)) == 0)
2651 ASSERT(dzp_error && zp_error);
2653 if (!dzpcheck_privs)
2661 * If directory returns EACCES then delete_child was denied
2662 * due to deny delete_child. In this case send the request through
2663 * secpolicy_vnode_remove(). We don't use zfs_delete_final_check()
2664 * since that *could* allow the delete based on write/execute permission
2665 * and we want delete permissions to override write/execute.
2668 if (dzp_error == EACCES) {
2669 /* XXXPJD: s/dzp/zp/ ? */
2670 return (secpolicy_vnode_remove(ZTOV(dzp), cr));
2674 * only need to see if we have write/execute on directory.
2677 dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2678 &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2680 if (dzp_error != 0 && !dzpcheck_privs)
2687 available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2688 available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2690 return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2695 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2696 znode_t *tzp, cred_t *cr)
2701 if (szp->z_pflags & ZFS_AV_QUARANTINED)
2702 return (SET_ERROR(EACCES));
2704 add_perm = (ZTOV(szp)->v_type == VDIR) ?
2705 ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2708 * Rename permissions are combination of delete permission +
2709 * add file/subdir permission.
2711 * BSD operating systems also require write permission
2712 * on the directory being moved from one parent directory
2715 if (ZTOV(szp)->v_type == VDIR && ZTOV(sdzp) != ZTOV(tdzp)) {
2716 if ((error = zfs_zaccess(szp, ACE_WRITE_DATA, 0, B_FALSE, cr)))
2721 * first make sure we do the delete portion.
2723 * If that succeeds then check for add_file/add_subdir permissions
2726 if ((error = zfs_zaccess_delete(sdzp, szp, cr)))
2730 * If we have a tzp, see if we can delete it?
2732 if (tzp && (error = zfs_zaccess_delete(tdzp, tzp, cr)))
2736 * Now check for add permissions
2738 error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr);