xref: /freebsd/sys/contrib/openzfs/module/os/freebsd/zfs/zfs_acl.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2013 by Delphix. All rights reserved.
25  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
26  */
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/vfs.h>
35 #include <sys/vnode.h>
36 #include <sys/file.h>
37 #include <sys/stat.h>
38 #include <sys/kmem.h>
39 #include <sys/cmn_err.h>
40 #include <sys/errno.h>
41 #include <sys/unistd.h>
42 #include <sys/sdt.h>
43 #include <sys/fs/zfs.h>
44 #include <sys/policy.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_fuid.h>
47 #include <sys/zfs_acl.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zfs_quota.h>
50 #include <sys/zfs_vfsops.h>
51 #include <sys/dmu.h>
52 #include <sys/dnode.h>
53 #include <sys/zap.h>
54 #include <sys/sa.h>
55 #include <acl/acl_common.h>
56 
57 
58 #define	ALLOW	ACE_ACCESS_ALLOWED_ACE_TYPE
59 #define	DENY	ACE_ACCESS_DENIED_ACE_TYPE
60 #define	MAX_ACE_TYPE	ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
61 #define	MIN_ACE_TYPE	ALLOW
62 
63 #define	OWNING_GROUP		(ACE_GROUP|ACE_IDENTIFIER_GROUP)
64 #define	EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
65     ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
66 #define	EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
67     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
68 #define	OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
69     ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
70 
71 #define	ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
72     ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
73     ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
74     ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
75 
76 #define	WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
77 #define	WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
78     ACE_DELETE|ACE_DELETE_CHILD)
79 #define	WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
80 
81 #define	OGE_CLEAR	(ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
82     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
83 
84 #define	OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
85     ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
86 
87 #define	ALL_INHERIT	(ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
88     ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
89 
90 #define	RESTRICTED_CLEAR	(ACE_WRITE_ACL|ACE_WRITE_OWNER)
91 
92 #define	V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
93     ZFS_ACL_PROTECTED)
94 
95 #define	ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
96     ZFS_ACL_OBJ_ACE)
97 
98 #define	ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
99 
100 static uint16_t
zfs_ace_v0_get_type(void * acep)101 zfs_ace_v0_get_type(void *acep)
102 {
103 	return (((zfs_oldace_t *)acep)->z_type);
104 }
105 
106 static uint16_t
zfs_ace_v0_get_flags(void * acep)107 zfs_ace_v0_get_flags(void *acep)
108 {
109 	return (((zfs_oldace_t *)acep)->z_flags);
110 }
111 
112 static uint32_t
zfs_ace_v0_get_mask(void * acep)113 zfs_ace_v0_get_mask(void *acep)
114 {
115 	return (((zfs_oldace_t *)acep)->z_access_mask);
116 }
117 
118 static uint64_t
zfs_ace_v0_get_who(void * acep)119 zfs_ace_v0_get_who(void *acep)
120 {
121 	return (((zfs_oldace_t *)acep)->z_fuid);
122 }
123 
124 static void
zfs_ace_v0_set_type(void * acep,uint16_t type)125 zfs_ace_v0_set_type(void *acep, uint16_t type)
126 {
127 	((zfs_oldace_t *)acep)->z_type = type;
128 }
129 
130 static void
zfs_ace_v0_set_flags(void * acep,uint16_t flags)131 zfs_ace_v0_set_flags(void *acep, uint16_t flags)
132 {
133 	((zfs_oldace_t *)acep)->z_flags = flags;
134 }
135 
136 static void
zfs_ace_v0_set_mask(void * acep,uint32_t mask)137 zfs_ace_v0_set_mask(void *acep, uint32_t mask)
138 {
139 	((zfs_oldace_t *)acep)->z_access_mask = mask;
140 }
141 
142 static void
zfs_ace_v0_set_who(void * acep,uint64_t who)143 zfs_ace_v0_set_who(void *acep, uint64_t who)
144 {
145 	((zfs_oldace_t *)acep)->z_fuid = who;
146 }
147 
148 static size_t
zfs_ace_v0_size(void * acep)149 zfs_ace_v0_size(void *acep)
150 {
151 	(void) acep;
152 	return (sizeof (zfs_oldace_t));
153 }
154 
155 static size_t
zfs_ace_v0_abstract_size(void)156 zfs_ace_v0_abstract_size(void)
157 {
158 	return (sizeof (zfs_oldace_t));
159 }
160 
161 static int
zfs_ace_v0_mask_off(void)162 zfs_ace_v0_mask_off(void)
163 {
164 	return (offsetof(zfs_oldace_t, z_access_mask));
165 }
166 
167 static int
zfs_ace_v0_data(void * acep,void ** datap)168 zfs_ace_v0_data(void *acep, void **datap)
169 {
170 	(void) acep;
171 	*datap = NULL;
172 	return (0);
173 }
174 
175 static const acl_ops_t zfs_acl_v0_ops = {
176 	zfs_ace_v0_get_mask,
177 	zfs_ace_v0_set_mask,
178 	zfs_ace_v0_get_flags,
179 	zfs_ace_v0_set_flags,
180 	zfs_ace_v0_get_type,
181 	zfs_ace_v0_set_type,
182 	zfs_ace_v0_get_who,
183 	zfs_ace_v0_set_who,
184 	zfs_ace_v0_size,
185 	zfs_ace_v0_abstract_size,
186 	zfs_ace_v0_mask_off,
187 	zfs_ace_v0_data
188 };
189 
190 static uint16_t
zfs_ace_fuid_get_type(void * acep)191 zfs_ace_fuid_get_type(void *acep)
192 {
193 	return (((zfs_ace_hdr_t *)acep)->z_type);
194 }
195 
196 static uint16_t
zfs_ace_fuid_get_flags(void * acep)197 zfs_ace_fuid_get_flags(void *acep)
198 {
199 	return (((zfs_ace_hdr_t *)acep)->z_flags);
200 }
201 
202 static uint32_t
zfs_ace_fuid_get_mask(void * acep)203 zfs_ace_fuid_get_mask(void *acep)
204 {
205 	return (((zfs_ace_hdr_t *)acep)->z_access_mask);
206 }
207 
208 static uint64_t
zfs_ace_fuid_get_who(void * args)209 zfs_ace_fuid_get_who(void *args)
210 {
211 	uint16_t entry_type;
212 	zfs_ace_t *acep = args;
213 
214 	entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
215 
216 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
217 	    entry_type == ACE_EVERYONE)
218 		return (-1);
219 	return (((zfs_ace_t *)acep)->z_fuid);
220 }
221 
222 static void
zfs_ace_fuid_set_type(void * acep,uint16_t type)223 zfs_ace_fuid_set_type(void *acep, uint16_t type)
224 {
225 	((zfs_ace_hdr_t *)acep)->z_type = type;
226 }
227 
228 static void
zfs_ace_fuid_set_flags(void * acep,uint16_t flags)229 zfs_ace_fuid_set_flags(void *acep, uint16_t flags)
230 {
231 	((zfs_ace_hdr_t *)acep)->z_flags = flags;
232 }
233 
234 static void
zfs_ace_fuid_set_mask(void * acep,uint32_t mask)235 zfs_ace_fuid_set_mask(void *acep, uint32_t mask)
236 {
237 	((zfs_ace_hdr_t *)acep)->z_access_mask = mask;
238 }
239 
240 static void
zfs_ace_fuid_set_who(void * arg,uint64_t who)241 zfs_ace_fuid_set_who(void *arg, uint64_t who)
242 {
243 	zfs_ace_t *acep = arg;
244 
245 	uint16_t entry_type = acep->z_hdr.z_flags & ACE_TYPE_FLAGS;
246 
247 	if (entry_type == ACE_OWNER || entry_type == OWNING_GROUP ||
248 	    entry_type == ACE_EVERYONE)
249 		return;
250 	acep->z_fuid = who;
251 }
252 
253 static size_t
zfs_ace_fuid_size(void * acep)254 zfs_ace_fuid_size(void *acep)
255 {
256 	zfs_ace_hdr_t *zacep = acep;
257 	uint16_t entry_type;
258 
259 	switch (zacep->z_type) {
260 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
261 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
262 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
263 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
264 		return (sizeof (zfs_object_ace_t));
265 	case ALLOW:
266 	case DENY:
267 		entry_type =
268 		    (((zfs_ace_hdr_t *)acep)->z_flags & ACE_TYPE_FLAGS);
269 		if (entry_type == ACE_OWNER ||
270 		    entry_type == OWNING_GROUP ||
271 		    entry_type == ACE_EVERYONE)
272 			return (sizeof (zfs_ace_hdr_t));
273 		zfs_fallthrough;
274 	default:
275 		return (sizeof (zfs_ace_t));
276 	}
277 }
278 
279 static size_t
zfs_ace_fuid_abstract_size(void)280 zfs_ace_fuid_abstract_size(void)
281 {
282 	return (sizeof (zfs_ace_hdr_t));
283 }
284 
285 static int
zfs_ace_fuid_mask_off(void)286 zfs_ace_fuid_mask_off(void)
287 {
288 	return (offsetof(zfs_ace_hdr_t, z_access_mask));
289 }
290 
291 static int
zfs_ace_fuid_data(void * acep,void ** datap)292 zfs_ace_fuid_data(void *acep, void **datap)
293 {
294 	zfs_ace_t *zacep = acep;
295 	zfs_object_ace_t *zobjp;
296 
297 	switch (zacep->z_hdr.z_type) {
298 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
299 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
300 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
301 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
302 		zobjp = acep;
303 		*datap = (caddr_t)zobjp + sizeof (zfs_ace_t);
304 		return (sizeof (zfs_object_ace_t) - sizeof (zfs_ace_t));
305 	default:
306 		*datap = NULL;
307 		return (0);
308 	}
309 }
310 
311 static const acl_ops_t zfs_acl_fuid_ops = {
312 	zfs_ace_fuid_get_mask,
313 	zfs_ace_fuid_set_mask,
314 	zfs_ace_fuid_get_flags,
315 	zfs_ace_fuid_set_flags,
316 	zfs_ace_fuid_get_type,
317 	zfs_ace_fuid_set_type,
318 	zfs_ace_fuid_get_who,
319 	zfs_ace_fuid_set_who,
320 	zfs_ace_fuid_size,
321 	zfs_ace_fuid_abstract_size,
322 	zfs_ace_fuid_mask_off,
323 	zfs_ace_fuid_data
324 };
325 
326 /*
327  * The following three functions are provided for compatibility with
328  * older ZPL version in order to determine if the file use to have
329  * an external ACL and what version of ACL previously existed on the
330  * file.  Would really be nice to not need this, sigh.
331  */
332 uint64_t
zfs_external_acl(znode_t * zp)333 zfs_external_acl(znode_t *zp)
334 {
335 	zfs_acl_phys_t acl_phys;
336 	int error;
337 
338 	if (zp->z_is_sa)
339 		return (0);
340 
341 	/*
342 	 * Need to deal with a potential
343 	 * race where zfs_sa_upgrade could cause
344 	 * z_isa_sa to change.
345 	 *
346 	 * If the lookup fails then the state of z_is_sa should have
347 	 * changed.
348 	 */
349 
350 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
351 	    &acl_phys, sizeof (acl_phys))) == 0)
352 		return (acl_phys.z_acl_extern_obj);
353 	else {
354 		/*
355 		 * after upgrade the SA_ZPL_ZNODE_ACL should have been
356 		 * removed
357 		 */
358 		VERIFY(zp->z_is_sa);
359 		VERIFY3S(error, ==, ENOENT);
360 		return (0);
361 	}
362 }
363 
364 /*
365  * Determine size of ACL in bytes
366  *
367  * This is more complicated than it should be since we have to deal
368  * with old external ACLs.
369  */
370 static int
zfs_acl_znode_info(znode_t * zp,int * aclsize,int * aclcount,zfs_acl_phys_t * aclphys)371 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
372     zfs_acl_phys_t *aclphys)
373 {
374 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
375 	uint64_t acl_count;
376 	int size;
377 	int error;
378 
379 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
380 	if (zp->z_is_sa) {
381 		if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
382 		    &size)) != 0)
383 			return (error);
384 		*aclsize = size;
385 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
386 		    &acl_count, sizeof (acl_count))) != 0)
387 			return (error);
388 		*aclcount = acl_count;
389 	} else {
390 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
391 		    aclphys, sizeof (*aclphys))) != 0)
392 			return (error);
393 
394 		if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
395 			*aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
396 			*aclcount = aclphys->z_acl_size;
397 		} else {
398 			*aclsize = aclphys->z_acl_size;
399 			*aclcount = aclphys->z_acl_count;
400 		}
401 	}
402 	return (0);
403 }
404 
405 int
zfs_znode_acl_version(znode_t * zp)406 zfs_znode_acl_version(znode_t *zp)
407 {
408 	zfs_acl_phys_t acl_phys;
409 
410 	if (zp->z_is_sa)
411 		return (ZFS_ACL_VERSION_FUID);
412 	else {
413 		int error;
414 
415 		/*
416 		 * Need to deal with a potential
417 		 * race where zfs_sa_upgrade could cause
418 		 * z_isa_sa to change.
419 		 *
420 		 * If the lookup fails then the state of z_is_sa should have
421 		 * changed.
422 		 */
423 		if ((error = sa_lookup(zp->z_sa_hdl,
424 		    SA_ZPL_ZNODE_ACL(zp->z_zfsvfs),
425 		    &acl_phys, sizeof (acl_phys))) == 0)
426 			return (acl_phys.z_acl_version);
427 		else {
428 			/*
429 			 * After upgrade SA_ZPL_ZNODE_ACL should have
430 			 * been removed.
431 			 */
432 			VERIFY(zp->z_is_sa);
433 			VERIFY3S(error, ==, ENOENT);
434 			return (ZFS_ACL_VERSION_FUID);
435 		}
436 	}
437 }
438 
439 static int
zfs_acl_version(int version)440 zfs_acl_version(int version)
441 {
442 	if (version < ZPL_VERSION_FUID)
443 		return (ZFS_ACL_VERSION_INITIAL);
444 	else
445 		return (ZFS_ACL_VERSION_FUID);
446 }
447 
448 static int
zfs_acl_version_zp(znode_t * zp)449 zfs_acl_version_zp(znode_t *zp)
450 {
451 	return (zfs_acl_version(zp->z_zfsvfs->z_version));
452 }
453 
454 zfs_acl_t *
zfs_acl_alloc(int vers)455 zfs_acl_alloc(int vers)
456 {
457 	zfs_acl_t *aclp;
458 
459 	aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
460 	list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
461 	    offsetof(zfs_acl_node_t, z_next));
462 	aclp->z_version = vers;
463 	if (vers == ZFS_ACL_VERSION_FUID)
464 		aclp->z_ops = &zfs_acl_fuid_ops;
465 	else
466 		aclp->z_ops = &zfs_acl_v0_ops;
467 	return (aclp);
468 }
469 
470 zfs_acl_node_t *
zfs_acl_node_alloc(size_t bytes)471 zfs_acl_node_alloc(size_t bytes)
472 {
473 	zfs_acl_node_t *aclnode;
474 
475 	aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
476 	if (bytes) {
477 		aclnode->z_acldata = kmem_zalloc(bytes, KM_SLEEP);
478 		aclnode->z_allocdata = aclnode->z_acldata;
479 		aclnode->z_allocsize = bytes;
480 		aclnode->z_size = bytes;
481 	}
482 
483 	return (aclnode);
484 }
485 
486 static void
zfs_acl_node_free(zfs_acl_node_t * aclnode)487 zfs_acl_node_free(zfs_acl_node_t *aclnode)
488 {
489 	if (aclnode->z_allocsize)
490 		kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
491 	kmem_free(aclnode, sizeof (zfs_acl_node_t));
492 }
493 
494 static void
zfs_acl_release_nodes(zfs_acl_t * aclp)495 zfs_acl_release_nodes(zfs_acl_t *aclp)
496 {
497 	zfs_acl_node_t *aclnode;
498 
499 	while ((aclnode = list_remove_head(&aclp->z_acl)))
500 		zfs_acl_node_free(aclnode);
501 	aclp->z_acl_count = 0;
502 	aclp->z_acl_bytes = 0;
503 }
504 
505 void
zfs_acl_free(zfs_acl_t * aclp)506 zfs_acl_free(zfs_acl_t *aclp)
507 {
508 	zfs_acl_release_nodes(aclp);
509 	list_destroy(&aclp->z_acl);
510 	kmem_free(aclp, sizeof (zfs_acl_t));
511 }
512 
513 static boolean_t
zfs_acl_valid_ace_type(uint_t type,uint_t flags)514 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
515 {
516 	uint16_t entry_type;
517 
518 	switch (type) {
519 	case ALLOW:
520 	case DENY:
521 	case ACE_SYSTEM_AUDIT_ACE_TYPE:
522 	case ACE_SYSTEM_ALARM_ACE_TYPE:
523 		entry_type = flags & ACE_TYPE_FLAGS;
524 		return (entry_type == ACE_OWNER ||
525 		    entry_type == OWNING_GROUP ||
526 		    entry_type == ACE_EVERYONE || entry_type == 0 ||
527 		    entry_type == ACE_IDENTIFIER_GROUP);
528 	default:
529 		if (type <= MAX_ACE_TYPE)
530 			return (B_TRUE);
531 	}
532 	return (B_FALSE);
533 }
534 
535 static boolean_t
zfs_ace_valid(vtype_t obj_type,zfs_acl_t * aclp,uint16_t type,uint16_t iflags)536 zfs_ace_valid(vtype_t obj_type, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
537 {
538 	/*
539 	 * first check type of entry
540 	 */
541 
542 	if (!zfs_acl_valid_ace_type(type, iflags))
543 		return (B_FALSE);
544 
545 	switch (type) {
546 	case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
547 	case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
548 	case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
549 	case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
550 		if (aclp->z_version < ZFS_ACL_VERSION_FUID)
551 			return (B_FALSE);
552 		aclp->z_hints |= ZFS_ACL_OBJ_ACE;
553 	}
554 
555 	/*
556 	 * next check inheritance level flags
557 	 */
558 
559 	if (obj_type == VDIR &&
560 	    (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
561 		aclp->z_hints |= ZFS_INHERIT_ACE;
562 
563 	if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
564 		if ((iflags & (ACE_FILE_INHERIT_ACE|
565 		    ACE_DIRECTORY_INHERIT_ACE)) == 0) {
566 			return (B_FALSE);
567 		}
568 	}
569 
570 	return (B_TRUE);
571 }
572 
573 static void *
zfs_acl_next_ace(zfs_acl_t * aclp,void * start,uint64_t * who,uint32_t * access_mask,uint16_t * iflags,uint16_t * type)574 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
575     uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
576 {
577 	zfs_acl_node_t *aclnode;
578 
579 	ASSERT3P(aclp, !=, NULL);
580 
581 	if (start == NULL) {
582 		aclnode = list_head(&aclp->z_acl);
583 		if (aclnode == NULL)
584 			return (NULL);
585 
586 		aclp->z_next_ace = aclnode->z_acldata;
587 		aclp->z_curr_node = aclnode;
588 		aclnode->z_ace_idx = 0;
589 	}
590 
591 	aclnode = aclp->z_curr_node;
592 
593 	if (aclnode == NULL)
594 		return (NULL);
595 
596 	if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
597 		aclnode = list_next(&aclp->z_acl, aclnode);
598 		if (aclnode == NULL)
599 			return (NULL);
600 		else {
601 			aclp->z_curr_node = aclnode;
602 			aclnode->z_ace_idx = 0;
603 			aclp->z_next_ace = aclnode->z_acldata;
604 		}
605 	}
606 
607 	if (aclnode->z_ace_idx < aclnode->z_ace_count) {
608 		void *acep = aclp->z_next_ace;
609 		size_t ace_size;
610 
611 		/*
612 		 * Make sure we don't overstep our bounds
613 		 */
614 		ace_size = aclp->z_ops->ace_size(acep);
615 
616 		if (((caddr_t)acep + ace_size) >
617 		    ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
618 			return (NULL);
619 		}
620 
621 		*iflags = aclp->z_ops->ace_flags_get(acep);
622 		*type = aclp->z_ops->ace_type_get(acep);
623 		*access_mask = aclp->z_ops->ace_mask_get(acep);
624 		*who = aclp->z_ops->ace_who_get(acep);
625 		aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
626 		aclnode->z_ace_idx++;
627 
628 		return ((void *)acep);
629 	}
630 	return (NULL);
631 }
632 
633 static uintptr_t
zfs_ace_walk(void * datap,uintptr_t cookie,int aclcnt,uint16_t * flags,uint16_t * type,uint32_t * mask)634 zfs_ace_walk(void *datap, uintptr_t cookie, int aclcnt,
635     uint16_t *flags, uint16_t *type, uint32_t *mask)
636 {
637 	(void) aclcnt;
638 	zfs_acl_t *aclp = datap;
639 	zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)(uintptr_t)cookie;
640 	uint64_t who;
641 
642 	acep = zfs_acl_next_ace(aclp, acep, &who, mask,
643 	    flags, type);
644 	return ((uintptr_t)acep);
645 }
646 
647 /*
648  * Copy ACE to internal ZFS format.
649  * While processing the ACL each ACE will be validated for correctness.
650  * ACE FUIDs will be created later.
651  */
652 static int
zfs_copy_ace_2_fuid(zfsvfs_t * zfsvfs,vtype_t obj_type,zfs_acl_t * aclp,void * datap,zfs_ace_t * z_acl,uint64_t aclcnt,size_t * size,zfs_fuid_info_t ** fuidp,cred_t * cr)653 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, vtype_t obj_type, zfs_acl_t *aclp,
654     void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
655     zfs_fuid_info_t **fuidp, cred_t *cr)
656 {
657 	int i;
658 	uint16_t entry_type;
659 	zfs_ace_t *aceptr = z_acl;
660 	ace_t *acep = datap;
661 	zfs_object_ace_t *zobjacep;
662 	ace_object_t *aceobjp;
663 
664 	for (i = 0; i != aclcnt; i++) {
665 		aceptr->z_hdr.z_access_mask = acep->a_access_mask;
666 		aceptr->z_hdr.z_flags = acep->a_flags;
667 		aceptr->z_hdr.z_type = acep->a_type;
668 		entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
669 		if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
670 		    entry_type != ACE_EVERYONE) {
671 			aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
672 			    cr, (entry_type == 0) ?
673 			    ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
674 		}
675 
676 		/*
677 		 * Make sure ACE is valid
678 		 */
679 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_hdr.z_type,
680 		    aceptr->z_hdr.z_flags) != B_TRUE)
681 			return (SET_ERROR(EINVAL));
682 
683 		switch (acep->a_type) {
684 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
685 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
686 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
687 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
688 			zobjacep = (zfs_object_ace_t *)aceptr;
689 			aceobjp = (ace_object_t *)acep;
690 
691 			memcpy(zobjacep->z_object_type, aceobjp->a_obj_type,
692 			    sizeof (aceobjp->a_obj_type));
693 			memcpy(zobjacep->z_inherit_type,
694 			    aceobjp->a_inherit_obj_type,
695 			    sizeof (aceobjp->a_inherit_obj_type));
696 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
697 			break;
698 		default:
699 			acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
700 		}
701 
702 		aceptr = (zfs_ace_t *)((caddr_t)aceptr +
703 		    aclp->z_ops->ace_size(aceptr));
704 	}
705 
706 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
707 
708 	return (0);
709 }
710 
711 /*
712  * Copy ZFS ACEs to fixed size ace_t layout
713  */
714 static void
zfs_copy_fuid_2_ace(zfsvfs_t * zfsvfs,zfs_acl_t * aclp,cred_t * cr,void * datap,int filter)715 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
716     void *datap, int filter)
717 {
718 	uint64_t who;
719 	uint32_t access_mask;
720 	uint16_t iflags, type;
721 	zfs_ace_hdr_t *zacep = NULL;
722 	ace_t *acep = datap;
723 	ace_object_t *objacep;
724 	zfs_object_ace_t *zobjacep;
725 	size_t ace_size;
726 	uint16_t entry_type;
727 
728 	while ((zacep = zfs_acl_next_ace(aclp, zacep,
729 	    &who, &access_mask, &iflags, &type))) {
730 
731 		switch (type) {
732 		case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
733 		case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
734 		case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
735 		case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
736 			if (filter) {
737 				continue;
738 			}
739 			zobjacep = (zfs_object_ace_t *)zacep;
740 			objacep = (ace_object_t *)acep;
741 			memcpy(objacep->a_obj_type,
742 			    zobjacep->z_object_type,
743 			    sizeof (zobjacep->z_object_type));
744 			memcpy(objacep->a_inherit_obj_type,
745 			    zobjacep->z_inherit_type,
746 			    sizeof (zobjacep->z_inherit_type));
747 			ace_size = sizeof (ace_object_t);
748 			break;
749 		default:
750 			ace_size = sizeof (ace_t);
751 			break;
752 		}
753 
754 		entry_type = (iflags & ACE_TYPE_FLAGS);
755 		if ((entry_type != ACE_OWNER &&
756 		    entry_type != OWNING_GROUP &&
757 		    entry_type != ACE_EVERYONE)) {
758 			acep->a_who = zfs_fuid_map_id(zfsvfs, who,
759 			    cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
760 			    ZFS_ACE_GROUP : ZFS_ACE_USER);
761 		} else {
762 			acep->a_who = (uid_t)(int64_t)who;
763 		}
764 		acep->a_access_mask = access_mask;
765 		acep->a_flags = iflags;
766 		acep->a_type = type;
767 		acep = (ace_t *)((caddr_t)acep + ace_size);
768 	}
769 }
770 
771 static int
zfs_copy_ace_2_oldace(vtype_t obj_type,zfs_acl_t * aclp,ace_t * acep,zfs_oldace_t * z_acl,int aclcnt,size_t * size)772 zfs_copy_ace_2_oldace(vtype_t obj_type, zfs_acl_t *aclp, ace_t *acep,
773     zfs_oldace_t *z_acl, int aclcnt, size_t *size)
774 {
775 	int i;
776 	zfs_oldace_t *aceptr = z_acl;
777 
778 	for (i = 0; i != aclcnt; i++, aceptr++) {
779 		aceptr->z_access_mask = acep[i].a_access_mask;
780 		aceptr->z_type = acep[i].a_type;
781 		aceptr->z_flags = acep[i].a_flags;
782 		aceptr->z_fuid = acep[i].a_who;
783 		/*
784 		 * Make sure ACE is valid
785 		 */
786 		if (zfs_ace_valid(obj_type, aclp, aceptr->z_type,
787 		    aceptr->z_flags) != B_TRUE)
788 			return (SET_ERROR(EINVAL));
789 	}
790 	*size = (caddr_t)aceptr - (caddr_t)z_acl;
791 	return (0);
792 }
793 
794 /*
795  * convert old ACL format to new
796  */
797 void
zfs_acl_xform(znode_t * zp,zfs_acl_t * aclp,cred_t * cr)798 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
799 {
800 	zfs_oldace_t *oldaclp;
801 	int i;
802 	uint16_t type, iflags;
803 	uint32_t access_mask;
804 	uint64_t who;
805 	void *cookie = NULL;
806 	zfs_acl_node_t *newaclnode;
807 
808 	ASSERT3U(aclp->z_version, ==, ZFS_ACL_VERSION_INITIAL);
809 	/*
810 	 * First create the ACE in a contiguous piece of memory
811 	 * for zfs_copy_ace_2_fuid().
812 	 *
813 	 * We only convert an ACL once, so this won't happen
814 	 * everytime.
815 	 */
816 	oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
817 	    KM_SLEEP);
818 	i = 0;
819 	while ((cookie = zfs_acl_next_ace(aclp, cookie, &who,
820 	    &access_mask, &iflags, &type))) {
821 		oldaclp[i].z_flags = iflags;
822 		oldaclp[i].z_type = type;
823 		oldaclp[i].z_fuid = who;
824 		oldaclp[i++].z_access_mask = access_mask;
825 	}
826 
827 	newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
828 	    sizeof (zfs_object_ace_t));
829 	aclp->z_ops = &zfs_acl_fuid_ops;
830 	VERIFY0(zfs_copy_ace_2_fuid(zp->z_zfsvfs, ZTOV(zp)->v_type, aclp,
831 	    oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
832 	    &newaclnode->z_size, NULL, cr));
833 	newaclnode->z_ace_count = aclp->z_acl_count;
834 	aclp->z_version = ZFS_ACL_VERSION;
835 	kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
836 
837 	/*
838 	 * Release all previous ACL nodes
839 	 */
840 
841 	zfs_acl_release_nodes(aclp);
842 
843 	list_insert_head(&aclp->z_acl, newaclnode);
844 
845 	aclp->z_acl_bytes = newaclnode->z_size;
846 	aclp->z_acl_count = newaclnode->z_ace_count;
847 
848 }
849 
850 /*
851  * Convert unix access mask to v4 access mask
852  */
853 static uint32_t
zfs_unix_to_v4(uint32_t access_mask)854 zfs_unix_to_v4(uint32_t access_mask)
855 {
856 	uint32_t new_mask = 0;
857 
858 	if (access_mask & S_IXOTH)
859 		new_mask |= ACE_EXECUTE;
860 	if (access_mask & S_IWOTH)
861 		new_mask |= ACE_WRITE_DATA;
862 	if (access_mask & S_IROTH)
863 		new_mask |= ACE_READ_DATA;
864 	return (new_mask);
865 }
866 
867 static void
zfs_set_ace(zfs_acl_t * aclp,void * acep,uint32_t access_mask,uint16_t access_type,uint64_t fuid,uint16_t entry_type)868 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
869     uint16_t access_type, uint64_t fuid, uint16_t entry_type)
870 {
871 	uint16_t type = entry_type & ACE_TYPE_FLAGS;
872 
873 	aclp->z_ops->ace_mask_set(acep, access_mask);
874 	aclp->z_ops->ace_type_set(acep, access_type);
875 	aclp->z_ops->ace_flags_set(acep, entry_type);
876 	if ((type != ACE_OWNER && type != OWNING_GROUP &&
877 	    type != ACE_EVERYONE))
878 		aclp->z_ops->ace_who_set(acep, fuid);
879 }
880 
881 /*
882  * Determine mode of file based on ACL.
883  */
884 uint64_t
zfs_mode_compute(uint64_t fmode,zfs_acl_t * aclp,uint64_t * pflags,uint64_t fuid,uint64_t fgid)885 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
886     uint64_t *pflags, uint64_t fuid, uint64_t fgid)
887 {
888 	int		entry_type;
889 	mode_t		mode;
890 	mode_t		seen = 0;
891 	zfs_ace_hdr_t 	*acep = NULL;
892 	uint64_t	who;
893 	uint16_t	iflags, type;
894 	uint32_t	access_mask;
895 	boolean_t	an_exec_denied = B_FALSE;
896 
897 	mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
898 
899 	while ((acep = zfs_acl_next_ace(aclp, acep, &who,
900 	    &access_mask, &iflags, &type))) {
901 
902 		if (!zfs_acl_valid_ace_type(type, iflags))
903 			continue;
904 
905 		entry_type = (iflags & ACE_TYPE_FLAGS);
906 
907 		/*
908 		 * Skip over any inherit_only ACEs
909 		 */
910 		if (iflags & ACE_INHERIT_ONLY_ACE)
911 			continue;
912 
913 		if (entry_type == ACE_OWNER || (entry_type == 0 &&
914 		    who == fuid)) {
915 			if ((access_mask & ACE_READ_DATA) &&
916 			    (!(seen & S_IRUSR))) {
917 				seen |= S_IRUSR;
918 				if (type == ALLOW) {
919 					mode |= S_IRUSR;
920 				}
921 			}
922 			if ((access_mask & ACE_WRITE_DATA) &&
923 			    (!(seen & S_IWUSR))) {
924 				seen |= S_IWUSR;
925 				if (type == ALLOW) {
926 					mode |= S_IWUSR;
927 				}
928 			}
929 			if ((access_mask & ACE_EXECUTE) &&
930 			    (!(seen & S_IXUSR))) {
931 				seen |= S_IXUSR;
932 				if (type == ALLOW) {
933 					mode |= S_IXUSR;
934 				}
935 			}
936 		} else if (entry_type == OWNING_GROUP ||
937 		    (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
938 			if ((access_mask & ACE_READ_DATA) &&
939 			    (!(seen & S_IRGRP))) {
940 				seen |= S_IRGRP;
941 				if (type == ALLOW) {
942 					mode |= S_IRGRP;
943 				}
944 			}
945 			if ((access_mask & ACE_WRITE_DATA) &&
946 			    (!(seen & S_IWGRP))) {
947 				seen |= S_IWGRP;
948 				if (type == ALLOW) {
949 					mode |= S_IWGRP;
950 				}
951 			}
952 			if ((access_mask & ACE_EXECUTE) &&
953 			    (!(seen & S_IXGRP))) {
954 				seen |= S_IXGRP;
955 				if (type == ALLOW) {
956 					mode |= S_IXGRP;
957 				}
958 			}
959 		} else if (entry_type == ACE_EVERYONE) {
960 			if ((access_mask & ACE_READ_DATA)) {
961 				if (!(seen & S_IRUSR)) {
962 					seen |= S_IRUSR;
963 					if (type == ALLOW) {
964 						mode |= S_IRUSR;
965 					}
966 				}
967 				if (!(seen & S_IRGRP)) {
968 					seen |= S_IRGRP;
969 					if (type == ALLOW) {
970 						mode |= S_IRGRP;
971 					}
972 				}
973 				if (!(seen & S_IROTH)) {
974 					seen |= S_IROTH;
975 					if (type == ALLOW) {
976 						mode |= S_IROTH;
977 					}
978 				}
979 			}
980 			if ((access_mask & ACE_WRITE_DATA)) {
981 				if (!(seen & S_IWUSR)) {
982 					seen |= S_IWUSR;
983 					if (type == ALLOW) {
984 						mode |= S_IWUSR;
985 					}
986 				}
987 				if (!(seen & S_IWGRP)) {
988 					seen |= S_IWGRP;
989 					if (type == ALLOW) {
990 						mode |= S_IWGRP;
991 					}
992 				}
993 				if (!(seen & S_IWOTH)) {
994 					seen |= S_IWOTH;
995 					if (type == ALLOW) {
996 						mode |= S_IWOTH;
997 					}
998 				}
999 			}
1000 			if ((access_mask & ACE_EXECUTE)) {
1001 				if (!(seen & S_IXUSR)) {
1002 					seen |= S_IXUSR;
1003 					if (type == ALLOW) {
1004 						mode |= S_IXUSR;
1005 					}
1006 				}
1007 				if (!(seen & S_IXGRP)) {
1008 					seen |= S_IXGRP;
1009 					if (type == ALLOW) {
1010 						mode |= S_IXGRP;
1011 					}
1012 				}
1013 				if (!(seen & S_IXOTH)) {
1014 					seen |= S_IXOTH;
1015 					if (type == ALLOW) {
1016 						mode |= S_IXOTH;
1017 					}
1018 				}
1019 			}
1020 		} else {
1021 			/*
1022 			 * Only care if this IDENTIFIER_GROUP or
1023 			 * USER ACE denies execute access to someone,
1024 			 * mode is not affected
1025 			 */
1026 			if ((access_mask & ACE_EXECUTE) && type == DENY)
1027 				an_exec_denied = B_TRUE;
1028 		}
1029 	}
1030 
1031 	/*
1032 	 * Failure to allow is effectively a deny, so execute permission
1033 	 * is denied if it was never mentioned or if we explicitly
1034 	 * weren't allowed it.
1035 	 */
1036 	if (!an_exec_denied &&
1037 	    ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1038 	    (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1039 		an_exec_denied = B_TRUE;
1040 
1041 	if (an_exec_denied)
1042 		*pflags &= ~ZFS_NO_EXECS_DENIED;
1043 	else
1044 		*pflags |= ZFS_NO_EXECS_DENIED;
1045 
1046 	return (mode);
1047 }
1048 
1049 /*
1050  * Read an external acl object.  If the intent is to modify, always
1051  * create a new acl and leave any cached acl in place.
1052  */
1053 int
zfs_acl_node_read(znode_t * zp,boolean_t have_lock,zfs_acl_t ** aclpp,boolean_t will_modify)1054 zfs_acl_node_read(znode_t *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1055     boolean_t will_modify)
1056 {
1057 	zfs_acl_t	*aclp;
1058 	int		aclsize;
1059 	int		acl_count;
1060 	zfs_acl_node_t	*aclnode;
1061 	zfs_acl_phys_t	znode_acl;
1062 	int		version;
1063 	int		error;
1064 
1065 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1066 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1067 		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
1068 
1069 	if (zp->z_acl_cached && !will_modify) {
1070 		*aclpp = zp->z_acl_cached;
1071 		return (0);
1072 	}
1073 
1074 	version = zfs_znode_acl_version(zp);
1075 
1076 	if ((error = zfs_acl_znode_info(zp, &aclsize,
1077 	    &acl_count, &znode_acl)) != 0) {
1078 		goto done;
1079 	}
1080 
1081 	aclp = zfs_acl_alloc(version);
1082 
1083 	aclp->z_acl_count = acl_count;
1084 	aclp->z_acl_bytes = aclsize;
1085 
1086 	aclnode = zfs_acl_node_alloc(aclsize);
1087 	aclnode->z_ace_count = aclp->z_acl_count;
1088 	aclnode->z_size = aclsize;
1089 
1090 	if (!zp->z_is_sa) {
1091 		if (znode_acl.z_acl_extern_obj) {
1092 			error = dmu_read(zp->z_zfsvfs->z_os,
1093 			    znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1094 			    aclnode->z_acldata, DMU_READ_PREFETCH);
1095 		} else {
1096 			memcpy(aclnode->z_acldata, znode_acl.z_ace_data,
1097 			    aclnode->z_size);
1098 		}
1099 	} else {
1100 		error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zp->z_zfsvfs),
1101 		    aclnode->z_acldata, aclnode->z_size);
1102 	}
1103 
1104 	if (error != 0) {
1105 		zfs_acl_free(aclp);
1106 		zfs_acl_node_free(aclnode);
1107 		/* convert checksum errors into IO errors */
1108 		if (error == ECKSUM)
1109 			error = SET_ERROR(EIO);
1110 		goto done;
1111 	}
1112 
1113 	list_insert_head(&aclp->z_acl, aclnode);
1114 
1115 	*aclpp = aclp;
1116 	if (!will_modify)
1117 		zp->z_acl_cached = aclp;
1118 done:
1119 	return (error);
1120 }
1121 
1122 void
zfs_acl_data_locator(void ** dataptr,uint32_t * length,uint32_t buflen,boolean_t start,void * userdata)1123 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1124     boolean_t start, void *userdata)
1125 {
1126 	(void) buflen;
1127 	zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1128 
1129 	if (start) {
1130 		cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1131 	} else {
1132 		cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1133 		    cb->cb_acl_node);
1134 	}
1135 	ASSERT3P(cb->cb_acl_node, !=, NULL);
1136 	*dataptr = cb->cb_acl_node->z_acldata;
1137 	*length = cb->cb_acl_node->z_size;
1138 }
1139 
1140 int
zfs_acl_chown_setattr(znode_t * zp)1141 zfs_acl_chown_setattr(znode_t *zp)
1142 {
1143 	int error;
1144 	zfs_acl_t *aclp;
1145 
1146 	if (zp->z_zfsvfs->z_replay == B_FALSE) {
1147 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1148 		ASSERT_VOP_IN_SEQC(ZTOV(zp));
1149 	}
1150 	ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1151 
1152 	if ((error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE)) == 0)
1153 		zp->z_mode = zfs_mode_compute(zp->z_mode, aclp,
1154 		    &zp->z_pflags, zp->z_uid, zp->z_gid);
1155 	return (error);
1156 }
1157 
1158 /*
1159  * common code for setting ACLs.
1160  *
1161  * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1162  * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1163  * already checked the acl and knows whether to inherit.
1164  */
1165 int
zfs_aclset_common(znode_t * zp,zfs_acl_t * aclp,cred_t * cr,dmu_tx_t * tx)1166 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1167 {
1168 	int			error;
1169 	zfsvfs_t		*zfsvfs = zp->z_zfsvfs;
1170 	dmu_object_type_t	otype;
1171 	zfs_acl_locator_cb_t	locate = { 0 };
1172 	uint64_t		mode;
1173 	sa_bulk_attr_t		bulk[5];
1174 	uint64_t		ctime[2];
1175 	int			count = 0;
1176 	zfs_acl_phys_t		acl_phys;
1177 
1178 	if (zp->z_zfsvfs->z_replay == B_FALSE) {
1179 		ASSERT_VOP_IN_SEQC(ZTOV(zp));
1180 	}
1181 
1182 	mode = zp->z_mode;
1183 
1184 	mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1185 	    zp->z_uid, zp->z_gid);
1186 
1187 	zp->z_mode = mode;
1188 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1189 	    &mode, sizeof (mode));
1190 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1191 	    &zp->z_pflags, sizeof (zp->z_pflags));
1192 	SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1193 	    &ctime, sizeof (ctime));
1194 
1195 	if (zp->z_acl_cached) {
1196 		zfs_acl_free(zp->z_acl_cached);
1197 		zp->z_acl_cached = NULL;
1198 	}
1199 
1200 	/*
1201 	 * Upgrade needed?
1202 	 */
1203 	if (!zfsvfs->z_use_fuids) {
1204 		otype = DMU_OT_OLDACL;
1205 	} else {
1206 		if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1207 		    (zfsvfs->z_version >= ZPL_VERSION_FUID))
1208 			zfs_acl_xform(zp, aclp, cr);
1209 		ASSERT3U(aclp->z_version, >=, ZFS_ACL_VERSION_FUID);
1210 		otype = DMU_OT_ACL;
1211 	}
1212 
1213 	/*
1214 	 * Arrgh, we have to handle old on disk format
1215 	 * as well as newer (preferred) SA format.
1216 	 */
1217 
1218 	if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1219 		locate.cb_aclp = aclp;
1220 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1221 		    zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1222 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1223 		    NULL, &aclp->z_acl_count, sizeof (uint64_t));
1224 	} else { /* Painful legacy way */
1225 		zfs_acl_node_t *aclnode;
1226 		uint64_t off = 0;
1227 		uint64_t aoid;
1228 
1229 		if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1230 		    &acl_phys, sizeof (acl_phys))) != 0)
1231 			return (error);
1232 
1233 		aoid = acl_phys.z_acl_extern_obj;
1234 
1235 		if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1236 			/*
1237 			 * If ACL was previously external and we are now
1238 			 * converting to new ACL format then release old
1239 			 * ACL object and create a new one.
1240 			 */
1241 			if (aoid &&
1242 			    aclp->z_version != acl_phys.z_acl_version) {
1243 				error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1244 				if (error)
1245 					return (error);
1246 				aoid = 0;
1247 			}
1248 			if (aoid == 0) {
1249 				aoid = dmu_object_alloc(zfsvfs->z_os,
1250 				    otype, aclp->z_acl_bytes,
1251 				    otype == DMU_OT_ACL ?
1252 				    DMU_OT_SYSACL : DMU_OT_NONE,
1253 				    otype == DMU_OT_ACL ?
1254 				    DN_OLD_MAX_BONUSLEN : 0, tx);
1255 			} else {
1256 				(void) dmu_object_set_blocksize(zfsvfs->z_os,
1257 				    aoid, aclp->z_acl_bytes, 0, tx);
1258 			}
1259 			acl_phys.z_acl_extern_obj = aoid;
1260 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1261 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1262 				if (aclnode->z_ace_count == 0)
1263 					continue;
1264 				dmu_write(zfsvfs->z_os, aoid, off,
1265 				    aclnode->z_size, aclnode->z_acldata, tx);
1266 				off += aclnode->z_size;
1267 			}
1268 		} else {
1269 			void *start = acl_phys.z_ace_data;
1270 			/*
1271 			 * Migrating back embedded?
1272 			 */
1273 			if (acl_phys.z_acl_extern_obj) {
1274 				error = dmu_object_free(zfsvfs->z_os,
1275 				    acl_phys.z_acl_extern_obj, tx);
1276 				if (error)
1277 					return (error);
1278 				acl_phys.z_acl_extern_obj = 0;
1279 			}
1280 
1281 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1282 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1283 				if (aclnode->z_ace_count == 0)
1284 					continue;
1285 				memcpy(start, aclnode->z_acldata,
1286 				    aclnode->z_size);
1287 				start = (caddr_t)start + aclnode->z_size;
1288 			}
1289 		}
1290 		/*
1291 		 * If Old version then swap count/bytes to match old
1292 		 * layout of znode_acl_phys_t.
1293 		 */
1294 		if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1295 			acl_phys.z_acl_size = aclp->z_acl_count;
1296 			acl_phys.z_acl_count = aclp->z_acl_bytes;
1297 		} else {
1298 			acl_phys.z_acl_size = aclp->z_acl_bytes;
1299 			acl_phys.z_acl_count = aclp->z_acl_count;
1300 		}
1301 		acl_phys.z_acl_version = aclp->z_version;
1302 
1303 		SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1304 		    &acl_phys, sizeof (acl_phys));
1305 	}
1306 
1307 	/*
1308 	 * Replace ACL wide bits, but first clear them.
1309 	 */
1310 	zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1311 
1312 	zp->z_pflags |= aclp->z_hints;
1313 
1314 	if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1315 		zp->z_pflags |= ZFS_ACL_TRIVIAL;
1316 
1317 	zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1318 	return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1319 }
1320 
1321 static void
zfs_acl_chmod(vtype_t vtype,uint64_t mode,boolean_t split,boolean_t trim,zfs_acl_t * aclp)1322 zfs_acl_chmod(vtype_t vtype, uint64_t mode, boolean_t split, boolean_t trim,
1323     zfs_acl_t *aclp)
1324 {
1325 	void		*acep = NULL;
1326 	uint64_t	who;
1327 	int		new_count, new_bytes;
1328 	int		ace_size;
1329 	int 		entry_type;
1330 	uint16_t	iflags, type;
1331 	uint32_t	access_mask;
1332 	zfs_acl_node_t	*newnode;
1333 	size_t 		abstract_size = aclp->z_ops->ace_abstract_size();
1334 	void 		*zacep;
1335 	boolean_t	isdir;
1336 	trivial_acl_t	masks;
1337 
1338 	new_count = new_bytes = 0;
1339 
1340 	isdir = (vtype == VDIR);
1341 
1342 	acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1343 
1344 	newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1345 
1346 	zacep = newnode->z_acldata;
1347 	if (masks.allow0) {
1348 		zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1349 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1350 		new_count++;
1351 		new_bytes += abstract_size;
1352 	}
1353 	if (masks.deny1) {
1354 		zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1355 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1356 		new_count++;
1357 		new_bytes += abstract_size;
1358 	}
1359 	if (masks.deny2) {
1360 		zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1361 		zacep = (void *)((uintptr_t)zacep + abstract_size);
1362 		new_count++;
1363 		new_bytes += abstract_size;
1364 	}
1365 
1366 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1367 	    &iflags, &type))) {
1368 		entry_type = (iflags & ACE_TYPE_FLAGS);
1369 		/*
1370 		 * ACEs used to represent the file mode may be divided
1371 		 * into an equivalent pair of inherit-only and regular
1372 		 * ACEs, if they are inheritable.
1373 		 * Skip regular ACEs, which are replaced by the new mode.
1374 		 */
1375 		if (split && (entry_type == ACE_OWNER ||
1376 		    entry_type == OWNING_GROUP ||
1377 		    entry_type == ACE_EVERYONE)) {
1378 			if (!isdir || !(iflags &
1379 			    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1380 				continue;
1381 			/*
1382 			 * We preserve owner@, group@, or @everyone
1383 			 * permissions, if they are inheritable, by
1384 			 * copying them to inherit_only ACEs. This
1385 			 * prevents inheritable permissions from being
1386 			 * altered along with the file mode.
1387 			 */
1388 			iflags |= ACE_INHERIT_ONLY_ACE;
1389 		}
1390 
1391 		/*
1392 		 * If this ACL has any inheritable ACEs, mark that in
1393 		 * the hints (which are later masked into the pflags)
1394 		 * so create knows to do inheritance.
1395 		 */
1396 		if (isdir && (iflags &
1397 		    (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1398 			aclp->z_hints |= ZFS_INHERIT_ACE;
1399 
1400 		if ((type != ALLOW && type != DENY) ||
1401 		    (iflags & ACE_INHERIT_ONLY_ACE)) {
1402 			switch (type) {
1403 			case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1404 			case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1405 			case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1406 			case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1407 				aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1408 				break;
1409 			}
1410 		} else {
1411 			/*
1412 			 * Limit permissions granted by ACEs to be no greater
1413 			 * than permissions of the requested group mode.
1414 			 * Applies when the "aclmode" property is set to
1415 			 * "groupmask".
1416 			 */
1417 			if ((type == ALLOW) && trim)
1418 				access_mask &= masks.group;
1419 		}
1420 		zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1421 		ace_size = aclp->z_ops->ace_size(acep);
1422 		zacep = (void *)((uintptr_t)zacep + ace_size);
1423 		new_count++;
1424 		new_bytes += ace_size;
1425 	}
1426 	zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1427 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1428 	zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1429 	zacep = (void *)((uintptr_t)zacep + abstract_size);
1430 	zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1431 
1432 	new_count += 3;
1433 	new_bytes += abstract_size * 3;
1434 	zfs_acl_release_nodes(aclp);
1435 	aclp->z_acl_count = new_count;
1436 	aclp->z_acl_bytes = new_bytes;
1437 	newnode->z_ace_count = new_count;
1438 	newnode->z_size = new_bytes;
1439 	list_insert_tail(&aclp->z_acl, newnode);
1440 }
1441 
1442 int
zfs_acl_chmod_setattr(znode_t * zp,zfs_acl_t ** aclp,uint64_t mode)1443 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1444 {
1445 	int error = 0;
1446 
1447 	mutex_enter(&zp->z_acl_lock);
1448 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1449 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1450 	if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_DISCARD)
1451 		*aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1452 	else
1453 		error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1454 
1455 	if (error == 0) {
1456 		(*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1457 		zfs_acl_chmod(ZTOV(zp)->v_type, mode, B_TRUE,
1458 		    (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1459 	}
1460 	mutex_exit(&zp->z_acl_lock);
1461 
1462 	return (error);
1463 }
1464 
1465 /*
1466  * Should ACE be inherited?
1467  */
1468 static int
zfs_ace_can_use(vtype_t vtype,uint16_t acep_flags)1469 zfs_ace_can_use(vtype_t vtype, uint16_t acep_flags)
1470 {
1471 	int	iflags = (acep_flags & 0xf);
1472 
1473 	if ((vtype == VDIR) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1474 		return (1);
1475 	else if (iflags & ACE_FILE_INHERIT_ACE)
1476 		return (!((vtype == VDIR) &&
1477 		    (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1478 	return (0);
1479 }
1480 
1481 /*
1482  * inherit inheritable ACEs from parent
1483  */
1484 static zfs_acl_t *
zfs_acl_inherit(zfsvfs_t * zfsvfs,vtype_t vtype,zfs_acl_t * paclp,uint64_t mode,boolean_t * need_chmod)1485 zfs_acl_inherit(zfsvfs_t *zfsvfs, vtype_t vtype, zfs_acl_t *paclp,
1486     uint64_t mode, boolean_t *need_chmod)
1487 {
1488 	void		*pacep = NULL;
1489 	void		*acep;
1490 	zfs_acl_node_t  *aclnode;
1491 	zfs_acl_t	*aclp = NULL;
1492 	uint64_t	who;
1493 	uint32_t	access_mask;
1494 	uint16_t	iflags, newflags, type;
1495 	size_t		ace_size;
1496 	void		*data1, *data2;
1497 	size_t		data1sz, data2sz;
1498 	uint_t		aclinherit;
1499 	boolean_t	isdir = (vtype == VDIR);
1500 	boolean_t	isreg = (vtype == VREG);
1501 
1502 	*need_chmod = B_TRUE;
1503 
1504 	aclp = zfs_acl_alloc(paclp->z_version);
1505 	aclinherit = zfsvfs->z_acl_inherit;
1506 	if (aclinherit == ZFS_ACL_DISCARD || vtype == VLNK)
1507 		return (aclp);
1508 
1509 	while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1510 	    &access_mask, &iflags, &type))) {
1511 
1512 		/*
1513 		 * don't inherit bogus ACEs
1514 		 */
1515 		if (!zfs_acl_valid_ace_type(type, iflags))
1516 			continue;
1517 
1518 		/*
1519 		 * Check if ACE is inheritable by this vnode
1520 		 */
1521 		if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1522 		    !zfs_ace_can_use(vtype, iflags))
1523 			continue;
1524 
1525 		/*
1526 		 * If owner@, group@, or everyone@ inheritable
1527 		 * then zfs_acl_chmod() isn't needed.
1528 		 */
1529 		if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1530 		    aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1531 		    ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1532 		    ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1533 		    (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1534 			*need_chmod = B_FALSE;
1535 
1536 		/*
1537 		 * Strip inherited execute permission from file if
1538 		 * not in mode
1539 		 */
1540 		if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1541 		    !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1542 			access_mask &= ~ACE_EXECUTE;
1543 		}
1544 
1545 		/*
1546 		 * Strip write_acl and write_owner from permissions
1547 		 * when inheriting an ACE
1548 		 */
1549 		if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1550 			access_mask &= ~RESTRICTED_CLEAR;
1551 		}
1552 
1553 		ace_size = aclp->z_ops->ace_size(pacep);
1554 		aclnode = zfs_acl_node_alloc(ace_size);
1555 		list_insert_tail(&aclp->z_acl, aclnode);
1556 		acep = aclnode->z_acldata;
1557 
1558 		zfs_set_ace(aclp, acep, access_mask, type,
1559 		    who, iflags|ACE_INHERITED_ACE);
1560 
1561 		/*
1562 		 * Copy special opaque data if any
1563 		 */
1564 		if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1565 			data2sz = aclp->z_ops->ace_data(acep, &data2);
1566 			VERIFY3U(data2sz, ==, data1sz);
1567 			memcpy(data2, data1, data2sz);
1568 		}
1569 
1570 		aclp->z_acl_count++;
1571 		aclnode->z_ace_count++;
1572 		aclp->z_acl_bytes += aclnode->z_size;
1573 		newflags = aclp->z_ops->ace_flags_get(acep);
1574 
1575 		/*
1576 		 * If ACE is not to be inherited further, or if the vnode is
1577 		 * not a directory, remove all inheritance flags
1578 		 */
1579 		if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1580 			newflags &= ~ALL_INHERIT;
1581 			aclp->z_ops->ace_flags_set(acep,
1582 			    newflags|ACE_INHERITED_ACE);
1583 			continue;
1584 		}
1585 
1586 		/*
1587 		 * This directory has an inheritable ACE
1588 		 */
1589 		aclp->z_hints |= ZFS_INHERIT_ACE;
1590 
1591 		/*
1592 		 * If only FILE_INHERIT is set then turn on
1593 		 * inherit_only
1594 		 */
1595 		if ((iflags & (ACE_FILE_INHERIT_ACE |
1596 		    ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1597 			newflags |= ACE_INHERIT_ONLY_ACE;
1598 			aclp->z_ops->ace_flags_set(acep,
1599 			    newflags|ACE_INHERITED_ACE);
1600 		} else {
1601 			newflags &= ~ACE_INHERIT_ONLY_ACE;
1602 			aclp->z_ops->ace_flags_set(acep,
1603 			    newflags|ACE_INHERITED_ACE);
1604 		}
1605 	}
1606 	if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1607 	    aclp->z_acl_count != 0) {
1608 		*need_chmod = B_FALSE;
1609 	}
1610 
1611 	return (aclp);
1612 }
1613 
1614 /*
1615  * Create file system object initial permissions
1616  * including inheritable ACEs.
1617  * Also, create FUIDs for owner and group.
1618  */
1619 int
zfs_acl_ids_create(znode_t * dzp,int flag,vattr_t * vap,cred_t * cr,vsecattr_t * vsecp,zfs_acl_ids_t * acl_ids,zidmap_t * mnt_ns)1620 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1621     vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids, zidmap_t *mnt_ns)
1622 {
1623 	int		error;
1624 	zfsvfs_t	*zfsvfs = dzp->z_zfsvfs;
1625 	zfs_acl_t	*paclp;
1626 	gid_t		gid;
1627 	boolean_t	need_chmod = B_TRUE;
1628 	boolean_t	trim = B_FALSE;
1629 	boolean_t	inherited = B_FALSE;
1630 
1631 	if ((flag & IS_ROOT_NODE) == 0) {
1632 		if (zfsvfs->z_replay == B_FALSE)
1633 			ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__);
1634 	} else
1635 		ASSERT3P(dzp->z_vnode, ==, NULL);
1636 	memset(acl_ids, 0, sizeof (zfs_acl_ids_t));
1637 	acl_ids->z_mode = MAKEIMODE(vap->va_type, vap->va_mode);
1638 
1639 	if (vsecp)
1640 		if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_type, vsecp, cr,
1641 		    &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1642 			return (error);
1643 	/*
1644 	 * Determine uid and gid.
1645 	 */
1646 	if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1647 	    ((flag & IS_XATTR) && (vap->va_type == VDIR))) {
1648 		acl_ids->z_fuid = zfs_fuid_create(zfsvfs,
1649 		    (uint64_t)vap->va_uid, cr,
1650 		    ZFS_OWNER, &acl_ids->z_fuidp);
1651 		acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1652 		    (uint64_t)vap->va_gid, cr,
1653 		    ZFS_GROUP, &acl_ids->z_fuidp);
1654 		gid = vap->va_gid;
1655 	} else {
1656 		uid_t id = crgetuid(cr);
1657 		if (IS_EPHEMERAL(id))
1658 			id = UID_NOBODY;
1659 		acl_ids->z_fuid = (uint64_t)id;
1660 		acl_ids->z_fgid = 0;
1661 		if (vap->va_mask & AT_GID)  {
1662 			acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1663 			    (uint64_t)vap->va_gid,
1664 			    cr, ZFS_GROUP, &acl_ids->z_fuidp);
1665 			gid = vap->va_gid;
1666 			if (acl_ids->z_fgid != dzp->z_gid &&
1667 			    !groupmember(vap->va_gid, cr) &&
1668 			    secpolicy_vnode_create_gid(cr) != 0)
1669 				acl_ids->z_fgid = 0;
1670 		}
1671 		if (acl_ids->z_fgid == 0) {
1672 			const char	*domain;
1673 			uint32_t	rid;
1674 
1675 			acl_ids->z_fgid = dzp->z_gid;
1676 			gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1677 			    cr, ZFS_GROUP);
1678 
1679 			if (zfsvfs->z_use_fuids &&
1680 			    IS_EPHEMERAL(acl_ids->z_fgid)) {
1681 				domain =
1682 				    zfs_fuid_idx_domain(&zfsvfs->z_fuid_idx,
1683 				    FUID_INDEX(acl_ids->z_fgid));
1684 				rid = FUID_RID(acl_ids->z_fgid);
1685 				zfs_fuid_node_add(&acl_ids->z_fuidp,
1686 				    domain, rid, FUID_INDEX(acl_ids->z_fgid),
1687 				    acl_ids->z_fgid, ZFS_GROUP);
1688 			}
1689 		}
1690 	}
1691 
1692 	/*
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.
1697 	 */
1698 
1699 	if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1700 	    (vap->va_type == VDIR)) {
1701 		acl_ids->z_mode |= S_ISGID;
1702 	} else {
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;
1706 	}
1707 
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,
1714 			    &paclp, B_FALSE));
1715 			acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1716 			    vap->va_type, paclp, acl_ids->z_mode, &need_chmod);
1717 			inherited = B_TRUE;
1718 		} else {
1719 			acl_ids->z_aclp =
1720 			    zfs_acl_alloc(zfs_acl_version_zp(dzp));
1721 			acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1722 		}
1723 		mutex_exit(&dzp->z_acl_lock);
1724 
1725 		if (need_chmod) {
1726 			if (vap->va_type == VDIR)
1727 				acl_ids->z_aclp->z_hints |=
1728 				    ZFS_ACL_AUTO_INHERIT;
1729 
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)
1733 				trim = B_TRUE;
1734 			zfs_acl_chmod(vap->va_type, acl_ids->z_mode, B_FALSE,
1735 			    trim, acl_ids->z_aclp);
1736 		}
1737 	}
1738 
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;
1745 	}
1746 
1747 	return (0);
1748 }
1749 
1750 /*
1751  * Free ACL and fuid_infop, but not the acl_ids structure
1752  */
1753 void
zfs_acl_ids_free(zfs_acl_ids_t * acl_ids)1754 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1755 {
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;
1762 }
1763 
1764 boolean_t
zfs_acl_ids_overquota(zfsvfs_t * zv,zfs_acl_ids_t * acl_ids,uint64_t projid)1765 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1766 {
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)));
1771 }
1772 
1773 /*
1774  * Retrieve a file's ACL
1775  */
1776 int
zfs_getacl(znode_t * zp,vsecattr_t * vsecp,boolean_t skipaclchk,cred_t * cr)1777 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1778 {
1779 	zfs_acl_t	*aclp;
1780 	ulong_t		mask;
1781 	int		error;
1782 	int 		count = 0;
1783 	int		largeace = 0;
1784 
1785 	mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1786 	    VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1787 
1788 	if (mask == 0)
1789 		return (SET_ERROR(ENOSYS));
1790 
1791 	if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr, NULL)))
1792 		return (error);
1793 
1794 	mutex_enter(&zp->z_acl_lock);
1795 
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);
1799 	if (error != 0) {
1800 		mutex_exit(&zp->z_acl_lock);
1801 		return (error);
1802 	}
1803 
1804 	/*
1805 	 * Scan ACL to determine number of ACEs
1806 	 */
1807 	if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1808 		void *zacep = NULL;
1809 		uint64_t who;
1810 		uint32_t access_mask;
1811 		uint16_t type, iflags;
1812 
1813 		while ((zacep = zfs_acl_next_ace(aclp, zacep,
1814 		    &who, &access_mask, &iflags, &type))) {
1815 			switch (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:
1820 				largeace++;
1821 				continue;
1822 			default:
1823 				count++;
1824 			}
1825 		}
1826 		vsecp->vsa_aclcnt = count;
1827 	} else
1828 		count = (int)aclp->z_acl_count;
1829 
1830 	if (mask & VSA_ACECNT) {
1831 		vsecp->vsa_aclcnt = count;
1832 	}
1833 
1834 	if (mask & VSA_ACE) {
1835 		size_t aclsz;
1836 
1837 		aclsz = count * sizeof (ace_t) +
1838 		    sizeof (ace_object_t) * largeace;
1839 
1840 		vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
1841 		vsecp->vsa_aclentsz = aclsz;
1842 
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));
1846 		else {
1847 			zfs_acl_node_t *aclnode;
1848 			void *start = vsecp->vsa_aclentp;
1849 
1850 			for (aclnode = list_head(&aclp->z_acl); aclnode;
1851 			    aclnode = list_next(&aclp->z_acl, aclnode)) {
1852 				memcpy(start, aclnode->z_acldata,
1853 				    aclnode->z_size);
1854 				start = (caddr_t)start + aclnode->z_size;
1855 			}
1856 			ASSERT3U((caddr_t)start - (caddr_t)vsecp->vsa_aclentp,
1857 			    ==, aclp->z_acl_bytes);
1858 		}
1859 	}
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;
1868 	}
1869 
1870 	mutex_exit(&zp->z_acl_lock);
1871 
1872 	return (0);
1873 }
1874 
1875 int
zfs_vsec_2_aclp(zfsvfs_t * zfsvfs,umode_t obj_type,vsecattr_t * vsecp,cred_t * cr,zfs_fuid_info_t ** fuidp,zfs_acl_t ** zaclp)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)
1878 {
1879 	zfs_acl_t *aclp;
1880 	zfs_acl_node_t *aclnode;
1881 	int aclcnt = vsecp->vsa_aclcnt;
1882 	int error;
1883 
1884 	if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
1885 		return (SET_ERROR(EINVAL));
1886 
1887 	aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
1888 
1889 	aclp->z_hints = 0;
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) {
1895 			zfs_acl_free(aclp);
1896 			zfs_acl_node_free(aclnode);
1897 			return (error);
1898 		}
1899 	} else {
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) {
1903 			zfs_acl_free(aclp);
1904 			zfs_acl_node_free(aclnode);
1905 			return (error);
1906 		}
1907 	}
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);
1912 
1913 	/*
1914 	 * If flags are being set then add them to z_hints
1915 	 */
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;
1923 	}
1924 
1925 	*zaclp = aclp;
1926 
1927 	return (0);
1928 }
1929 
1930 /*
1931  * Set a file's ACL
1932  */
1933 int
zfs_setacl(znode_t * zp,vsecattr_t * vsecp,boolean_t skipaclchk,cred_t * cr)1934 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1935 {
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);
1939 	dmu_tx_t	*tx;
1940 	int		error;
1941 	zfs_acl_t	*aclp;
1942 	zfs_fuid_info_t	*fuidp = NULL;
1943 	boolean_t	fuid_dirtied;
1944 	uint64_t	acl_obj;
1945 
1946 	if (zp->z_zfsvfs->z_replay == B_FALSE)
1947 		ASSERT_VOP_ELOCKED(ZTOV(zp), __func__);
1948 	if (mask == 0)
1949 		return (SET_ERROR(ENOSYS));
1950 
1951 	if (zp->z_pflags & ZFS_IMMUTABLE)
1952 		return (SET_ERROR(EPERM));
1953 
1954 	if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, NULL)))
1955 		return (error);
1956 
1957 	error = zfs_vsec_2_aclp(zfsvfs, ZTOV(zp)->v_type, vsecp, cr, &fuidp,
1958 	    &aclp);
1959 	if (error)
1960 		return (error);
1961 
1962 	/*
1963 	 * If ACL wide flags aren't being set then preserve any
1964 	 * existing flags.
1965 	 */
1966 	if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
1967 		aclp->z_hints |=
1968 		    (zp->z_pflags & V4_ACL_WIDE_FLAGS);
1969 	}
1970 top:
1971 	mutex_enter(&zp->z_acl_lock);
1972 
1973 	tx = dmu_tx_create(zfsvfs->z_os);
1974 
1975 	dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1976 
1977 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1978 	if (fuid_dirtied)
1979 		zfs_fuid_txhold(zfsvfs, tx);
1980 
1981 	/*
1982 	 * If old version and ACL won't fit in bonus and we aren't
1983 	 * upgrading then take out necessary DMU holds
1984 	 */
1985 
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,
1990 			    DMU_OBJECT_END);
1991 			dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
1992 			    aclp->z_acl_bytes);
1993 		} else {
1994 			dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
1995 		}
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);
1998 	}
1999 
2000 	zfs_sa_upgrade_txholds(tx, zp);
2001 	error = dmu_tx_assign(tx, DMU_TX_NOWAIT);
2002 	if (error) {
2003 		mutex_exit(&zp->z_acl_lock);
2004 
2005 		if (error == ERESTART) {
2006 			dmu_tx_wait(tx);
2007 			dmu_tx_abort(tx);
2008 			goto top;
2009 		}
2010 		dmu_tx_abort(tx);
2011 		zfs_acl_free(aclp);
2012 		return (error);
2013 	}
2014 
2015 	error = zfs_aclset_common(zp, aclp, cr, tx);
2016 	ASSERT0(error);
2017 	ASSERT3P(zp->z_acl_cached, ==, NULL);
2018 	zp->z_acl_cached = aclp;
2019 
2020 	if (fuid_dirtied)
2021 		zfs_fuid_sync(zfsvfs, tx);
2022 
2023 	zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2024 
2025 	if (fuidp)
2026 		zfs_fuid_info_free(fuidp);
2027 	dmu_tx_commit(tx);
2028 	mutex_exit(&zp->z_acl_lock);
2029 
2030 	return (error);
2031 }
2032 
2033 /*
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.
2037  */
2038 static int
zfs_zaccess_dataset_check(znode_t * zp,uint32_t v4_mode)2039 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2040 {
2041 	if ((v4_mode & WRITE_MASK) &&
2042 	    (zp->z_zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
2043 	    (!IS_DEVVP(ZTOV(zp)) || (v4_mode & WRITE_MASK_ATTRS))) {
2044 		return (SET_ERROR(EROFS));
2045 	}
2046 
2047 	/*
2048 	 * Intentionally allow ZFS_READONLY through here.
2049 	 * See zfs_zaccess_common().
2050 	 */
2051 	if ((v4_mode & WRITE_MASK_DATA) &&
2052 	    (zp->z_pflags & ZFS_IMMUTABLE)) {
2053 		return (SET_ERROR(EPERM));
2054 	}
2055 
2056 	/*
2057 	 * In FreeBSD we allow to modify directory's content is ZFS_NOUNLINK
2058 	 * (sunlnk) is set. We just don't allow directory removal, which is
2059 	 * handled in zfs_zaccess_delete().
2060 	 */
2061 	if ((v4_mode & ACE_DELETE) &&
2062 	    (zp->z_pflags & ZFS_NOUNLINK)) {
2063 		return (EPERM);
2064 	}
2065 
2066 	if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2067 	    (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2068 		return (SET_ERROR(EACCES));
2069 	}
2070 
2071 	return (0);
2072 }
2073 
2074 /*
2075  * The primary usage of this function is to loop through all of the
2076  * ACEs in the znode, determining what accesses of interest (AoI) to
2077  * the caller are allowed or denied.  The AoI are expressed as bits in
2078  * the working_mode parameter.  As each ACE is processed, bits covered
2079  * by that ACE are removed from the working_mode.  This removal
2080  * facilitates two things.  The first is that when the working mode is
2081  * empty (= 0), we know we've looked at all the AoI. The second is
2082  * that the ACE interpretation rules don't allow a later ACE to undo
2083  * something granted or denied by an earlier ACE.  Removing the
2084  * discovered access or denial enforces this rule.  At the end of
2085  * processing the ACEs, all AoI that were found to be denied are
2086  * placed into the working_mode, giving the caller a mask of denied
2087  * accesses.  Returns:
2088  *	0		if all AoI granted
2089  *	EACCESS 	if the denied mask is non-zero
2090  *	other error	if abnormal failure (e.g., IO error)
2091  *
2092  * A secondary usage of the function is to determine if any of the
2093  * AoI are granted.  If an ACE grants any access in
2094  * the working_mode, we immediately short circuit out of the function.
2095  * This mode is chosen by setting anyaccess to B_TRUE.  The
2096  * working_mode is not a denied access mask upon exit if the function
2097  * is used in this manner.
2098  */
2099 static int
zfs_zaccess_aces_check(znode_t * zp,uint32_t * working_mode,boolean_t anyaccess,cred_t * cr)2100 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2101     boolean_t anyaccess, cred_t *cr)
2102 {
2103 	zfsvfs_t	*zfsvfs = zp->z_zfsvfs;
2104 	zfs_acl_t	*aclp;
2105 	int		error;
2106 	uid_t		uid = crgetuid(cr);
2107 	uint64_t 	who;
2108 	uint16_t	type, iflags;
2109 	uint16_t	entry_type;
2110 	uint32_t	access_mask;
2111 	uint32_t	deny_mask = 0;
2112 	zfs_ace_hdr_t	*acep = NULL;
2113 	boolean_t	checkit;
2114 	uid_t		gowner;
2115 	uid_t		fowner;
2116 
2117 	zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2118 
2119 	mutex_enter(&zp->z_acl_lock);
2120 
2121 	if (zp->z_zfsvfs->z_replay == B_FALSE)
2122 		ASSERT_VOP_LOCKED(ZTOV(zp), __func__);
2123 	error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
2124 	if (error != 0) {
2125 		mutex_exit(&zp->z_acl_lock);
2126 		return (error);
2127 	}
2128 
2129 	ASSERT3P(zp->z_acl_cached, !=, NULL);
2130 
2131 	while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2132 	    &iflags, &type))) {
2133 		uint32_t mask_matched;
2134 
2135 		if (!zfs_acl_valid_ace_type(type, iflags))
2136 			continue;
2137 
2138 		if (ZTOV(zp)->v_type == VDIR && (iflags & ACE_INHERIT_ONLY_ACE))
2139 			continue;
2140 
2141 		/* Skip ACE if it does not affect any AoI */
2142 		mask_matched = (access_mask & *working_mode);
2143 		if (!mask_matched)
2144 			continue;
2145 
2146 		entry_type = (iflags & ACE_TYPE_FLAGS);
2147 
2148 		checkit = B_FALSE;
2149 
2150 		switch (entry_type) {
2151 		case ACE_OWNER:
2152 			if (uid == fowner)
2153 				checkit = B_TRUE;
2154 			break;
2155 		case OWNING_GROUP:
2156 			who = gowner;
2157 			zfs_fallthrough;
2158 		case ACE_IDENTIFIER_GROUP:
2159 			checkit = zfs_groupmember(zfsvfs, who, cr);
2160 			break;
2161 		case ACE_EVERYONE:
2162 			checkit = B_TRUE;
2163 			break;
2164 
2165 		/* USER Entry */
2166 		default:
2167 			if (entry_type == 0) {
2168 				uid_t newid;
2169 
2170 				newid = zfs_fuid_map_id(zfsvfs, who, cr,
2171 				    ZFS_ACE_USER);
2172 				if (newid !=  UID_NOBODY &&
2173 				    uid == newid)
2174 					checkit = B_TRUE;
2175 				break;
2176 			} else {
2177 				mutex_exit(&zp->z_acl_lock);
2178 				return (SET_ERROR(EIO));
2179 			}
2180 		}
2181 
2182 		if (checkit) {
2183 			if (type == DENY) {
2184 				DTRACE_PROBE3(zfs__ace__denies,
2185 				    znode_t *, zp,
2186 				    zfs_ace_hdr_t *, acep,
2187 				    uint32_t, mask_matched);
2188 				deny_mask |= mask_matched;
2189 			} else {
2190 				DTRACE_PROBE3(zfs__ace__allows,
2191 				    znode_t *, zp,
2192 				    zfs_ace_hdr_t *, acep,
2193 				    uint32_t, mask_matched);
2194 				if (anyaccess) {
2195 					mutex_exit(&zp->z_acl_lock);
2196 					return (0);
2197 				}
2198 			}
2199 			*working_mode &= ~mask_matched;
2200 		}
2201 
2202 		/* Are we done? */
2203 		if (*working_mode == 0)
2204 			break;
2205 	}
2206 
2207 	mutex_exit(&zp->z_acl_lock);
2208 
2209 	/* Put the found 'denies' back on the working mode */
2210 	if (deny_mask) {
2211 		*working_mode |= deny_mask;
2212 		return (SET_ERROR(EACCES));
2213 	} else if (*working_mode) {
2214 		return (-1);
2215 	}
2216 
2217 	return (0);
2218 }
2219 
2220 /*
2221  * Return true if any access whatsoever granted, we don't actually
2222  * care what access is granted.
2223  */
2224 boolean_t
zfs_has_access(znode_t * zp,cred_t * cr)2225 zfs_has_access(znode_t *zp, cred_t *cr)
2226 {
2227 	uint32_t have = ACE_ALL_PERMS;
2228 
2229 	if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr) != 0) {
2230 		uid_t owner;
2231 
2232 		owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2233 		return (secpolicy_vnode_any_access(cr, ZTOV(zp), owner) == 0);
2234 	}
2235 	return (B_TRUE);
2236 }
2237 
2238 static int
zfs_zaccess_common(znode_t * zp,uint32_t v4_mode,uint32_t * working_mode,boolean_t * check_privs,boolean_t skipaclchk,cred_t * cr)2239 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2240     boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr)
2241 {
2242 	zfsvfs_t *zfsvfs = zp->z_zfsvfs;
2243 	int err;
2244 
2245 	*working_mode = v4_mode;
2246 	*check_privs = B_TRUE;
2247 
2248 	/*
2249 	 * Short circuit empty requests
2250 	 */
2251 	if (v4_mode == 0 || zfsvfs->z_replay) {
2252 		*working_mode = 0;
2253 		return (0);
2254 	}
2255 
2256 	if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2257 		*check_privs = B_FALSE;
2258 		return (err);
2259 	}
2260 
2261 	/*
2262 	 * The caller requested that the ACL check be skipped.  This
2263 	 * would only happen if the caller checked VOP_ACCESS() with a
2264 	 * 32 bit ACE mask and already had the appropriate permissions.
2265 	 */
2266 	if (skipaclchk) {
2267 		*working_mode = 0;
2268 		return (0);
2269 	}
2270 
2271 	/*
2272 	 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2273 	 * When that flag is set, we should behave as if write access
2274 	 * were not granted by anything in the ACL.  In particular:
2275 	 * We _must_ allow writes after opening the file r/w, then
2276 	 * setting the DOS R/O attribute, and writing some more.
2277 	 * (Similar to how you can write after fchmod(fd, 0444).)
2278 	 *
2279 	 * Therefore ZFS_READONLY is ignored in the dataset check
2280 	 * above, and checked here as if part of the ACL check.
2281 	 * Also note: DOS R/O is ignored for directories.
2282 	 */
2283 	if ((v4_mode & WRITE_MASK_DATA) &&
2284 	    (ZTOV(zp)->v_type != VDIR) &&
2285 	    (zp->z_pflags & ZFS_READONLY)) {
2286 		return (SET_ERROR(EPERM));
2287 	}
2288 
2289 	return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr));
2290 }
2291 
2292 static int
zfs_zaccess_append(znode_t * zp,uint32_t * working_mode,boolean_t * check_privs,cred_t * cr)2293 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2294     cred_t *cr)
2295 {
2296 	if (*working_mode != ACE_WRITE_DATA)
2297 		return (SET_ERROR(EACCES));
2298 
2299 	return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2300 	    check_privs, B_FALSE, cr));
2301 }
2302 
2303 /*
2304  * Check if VEXEC is allowed.
2305  *
2306  * This routine is based on zfs_fastaccesschk_execute which has slowpath
2307  * calling zfs_zaccess. This would be incorrect on FreeBSD (see
2308  * zfs_freebsd_access for the difference). Thus this variant let's the
2309  * caller handle the slowpath (if necessary).
2310  *
2311  * On top of that we perform a lockless check for ZFS_NO_EXECS_DENIED.
2312  *
2313  * Safe access to znode_t is provided by the vnode lock.
2314  */
2315 int
zfs_fastaccesschk_execute(znode_t * zdp,cred_t * cr)2316 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2317 {
2318 	boolean_t is_attr;
2319 
2320 	if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2321 		return (1);
2322 
2323 	is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2324 	    (ZTOV(zdp)->v_type == VDIR));
2325 	if (is_attr)
2326 		return (1);
2327 
2328 	if (zdp->z_pflags & ZFS_NO_EXECS_DENIED)
2329 		return (0);
2330 
2331 	return (1);
2332 }
2333 
2334 
2335 /*
2336  * Determine whether Access should be granted/denied.
2337  *
2338  * The least priv subsystem is always consulted as a basic privilege
2339  * can define any form of access.
2340  */
2341 int
zfs_zaccess(znode_t * zp,int mode,int flags,boolean_t skipaclchk,cred_t * cr,zidmap_t * mnt_ns)2342 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr,
2343     zidmap_t *mnt_ns)
2344 {
2345 	uint32_t	working_mode;
2346 	int		error;
2347 	int		is_attr;
2348 	boolean_t 	check_privs;
2349 	znode_t		*xzp = NULL;
2350 	znode_t 	*check_zp = zp;
2351 	mode_t		needed_bits;
2352 	uid_t		owner;
2353 
2354 	is_attr = ((zp->z_pflags & ZFS_XATTR) && (ZTOV(zp)->v_type == VDIR));
2355 
2356 	/*
2357 	 * In FreeBSD, we don't care about permissions of individual ADS.
2358 	 * Note that not checking them is not just an optimization - without
2359 	 * this shortcut, EA operations may bogusly fail with EACCES.
2360 	 */
2361 	if (zp->z_pflags & ZFS_XATTR)
2362 		return (0);
2363 
2364 	owner = zfs_fuid_map_id(zp->z_zfsvfs, zp->z_uid, cr, ZFS_OWNER);
2365 
2366 	/*
2367 	 * Map the bits required to the standard vnode flags VREAD|VWRITE|VEXEC
2368 	 * in needed_bits.  Map the bits mapped by working_mode (currently
2369 	 * missing) in missing_bits.
2370 	 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2371 	 * needed_bits.
2372 	 */
2373 	needed_bits = 0;
2374 
2375 	working_mode = mode;
2376 	if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2377 	    owner == crgetuid(cr))
2378 		working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2379 
2380 	if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2381 	    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2382 		needed_bits |= VREAD;
2383 	if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2384 	    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2385 		needed_bits |= VWRITE;
2386 	if (working_mode & ACE_EXECUTE)
2387 		needed_bits |= VEXEC;
2388 
2389 	if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2390 	    &check_privs, skipaclchk, cr)) == 0) {
2391 		if (is_attr)
2392 			VN_RELE(ZTOV(xzp));
2393 		return (secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2394 		    needed_bits, needed_bits));
2395 	}
2396 
2397 	if (error && !check_privs) {
2398 		if (is_attr)
2399 			VN_RELE(ZTOV(xzp));
2400 		return (error);
2401 	}
2402 
2403 	if (error && (flags & V_APPEND)) {
2404 		error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr);
2405 	}
2406 
2407 	if (error && check_privs) {
2408 		mode_t		checkmode = 0;
2409 		vnode_t *check_vp = ZTOV(check_zp);
2410 
2411 		/*
2412 		 * First check for implicit owner permission on
2413 		 * read_acl/read_attributes
2414 		 */
2415 
2416 		ASSERT3U(working_mode, !=, 0);
2417 
2418 		if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2419 		    owner == crgetuid(cr)))
2420 			working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2421 
2422 		if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2423 		    ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2424 			checkmode |= VREAD;
2425 		if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2426 		    ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2427 			checkmode |= VWRITE;
2428 		if (working_mode & ACE_EXECUTE)
2429 			checkmode |= VEXEC;
2430 
2431 		error = secpolicy_vnode_access2(cr, check_vp, owner,
2432 		    needed_bits & ~checkmode, needed_bits);
2433 
2434 		if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2435 			error = secpolicy_vnode_chown(check_vp, cr, owner);
2436 		if (error == 0 && (working_mode & ACE_WRITE_ACL))
2437 			error = secpolicy_vnode_setdac(check_vp, cr, owner);
2438 
2439 		if (error == 0 && (working_mode &
2440 		    (ACE_DELETE|ACE_DELETE_CHILD)))
2441 			error = secpolicy_vnode_remove(check_vp, cr);
2442 
2443 		if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2444 			error = secpolicy_vnode_chown(check_vp, cr, owner);
2445 		}
2446 		if (error == 0) {
2447 			/*
2448 			 * See if any bits other than those already checked
2449 			 * for are still present.  If so then return EACCES
2450 			 */
2451 			if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2452 				error = SET_ERROR(EACCES);
2453 			}
2454 		}
2455 	} else if (error == 0) {
2456 		error = secpolicy_vnode_access2(cr, ZTOV(zp), owner,
2457 		    needed_bits, needed_bits);
2458 	}
2459 
2460 
2461 	if (is_attr)
2462 		VN_RELE(ZTOV(xzp));
2463 
2464 	return (error);
2465 }
2466 
2467 /*
2468  * Translate traditional unix VREAD/VWRITE/VEXEC mode into
2469  * NFSv4-style ZFS ACL format and call zfs_zaccess()
2470  */
2471 int
zfs_zaccess_rwx(znode_t * zp,mode_t mode,int flags,cred_t * cr,zidmap_t * mnt_ns)2472 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr,
2473     zidmap_t *mnt_ns)
2474 {
2475 	return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr,
2476 	    mnt_ns));
2477 }
2478 
2479 /*
2480  * Access function for secpolicy_vnode_setattr
2481  */
2482 int
zfs_zaccess_unix(void * zp,int mode,cred_t * cr)2483 zfs_zaccess_unix(void *zp, int mode, cred_t *cr)
2484 {
2485 	int v4_mode = zfs_unix_to_v4(mode >> 6);
2486 
2487 	return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr, NULL));
2488 }
2489 
2490 static int
zfs_delete_final_check(znode_t * zp,znode_t * dzp,mode_t available_perms,cred_t * cr)2491 zfs_delete_final_check(znode_t *zp, znode_t *dzp,
2492     mode_t available_perms, cred_t *cr)
2493 {
2494 	int error;
2495 	uid_t downer;
2496 
2497 	downer = zfs_fuid_map_id(dzp->z_zfsvfs, dzp->z_uid, cr, ZFS_OWNER);
2498 
2499 	error = secpolicy_vnode_access2(cr, ZTOV(dzp),
2500 	    downer, available_perms, VWRITE|VEXEC);
2501 
2502 	if (error == 0)
2503 		error = zfs_sticky_remove_access(dzp, zp, cr);
2504 
2505 	return (error);
2506 }
2507 
2508 /*
2509  * Determine whether Access should be granted/deny, without
2510  * consulting least priv subsystem.
2511  *
2512  * The following chart is the recommended NFSv4 enforcement for
2513  * ability to delete an object.
2514  *
2515  *      -------------------------------------------------------
2516  *      |   Parent Dir  |           Target Object Permissions |
2517  *      |  permissions  |                                     |
2518  *      -------------------------------------------------------
2519  *      |               | ACL Allows | ACL Denies| Delete     |
2520  *      |               |  Delete    |  Delete   | unspecified|
2521  *      -------------------------------------------------------
2522  *      |  ACL Allows   | Permit     | Permit    | Permit     |
2523  *      |  DELETE_CHILD |                                     |
2524  *      -------------------------------------------------------
2525  *      |  ACL Denies   | Permit     | Deny      | Deny       |
2526  *      |  DELETE_CHILD |            |           |            |
2527  *      -------------------------------------------------------
2528  *      | ACL specifies |            |           |            |
2529  *      | only allow    | Permit     | Permit    | Permit     |
2530  *      | write and     |            |           |            |
2531  *      | execute       |            |           |            |
2532  *      -------------------------------------------------------
2533  *      | ACL denies    |            |           |            |
2534  *      | write and     | Permit     | Deny      | Deny       |
2535  *      | execute       |            |           |            |
2536  *      -------------------------------------------------------
2537  *         ^
2538  *         |
2539  *         No search privilege, can't even look up file?
2540  *
2541  */
2542 int
zfs_zaccess_delete(znode_t * dzp,znode_t * zp,cred_t * cr,zidmap_t * mnt_ns)2543 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr, zidmap_t *mnt_ns)
2544 {
2545 	uint32_t dzp_working_mode = 0;
2546 	uint32_t zp_working_mode = 0;
2547 	int dzp_error, zp_error;
2548 	mode_t available_perms;
2549 	boolean_t dzpcheck_privs = B_TRUE;
2550 	boolean_t zpcheck_privs = B_TRUE;
2551 
2552 	/*
2553 	 * We want specific DELETE permissions to
2554 	 * take precedence over WRITE/EXECUTE.  We don't
2555 	 * want an ACL such as this to mess us up.
2556 	 * user:joe:write_data:deny,user:joe:delete:allow
2557 	 *
2558 	 * However, deny permissions may ultimately be overridden
2559 	 * by secpolicy_vnode_access().
2560 	 *
2561 	 * We will ask for all of the necessary permissions and then
2562 	 * look at the working modes from the directory and target object
2563 	 * to determine what was found.
2564 	 */
2565 
2566 	if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2567 		return (SET_ERROR(EPERM));
2568 
2569 	/*
2570 	 * First row
2571 	 * If the directory permissions allow the delete, we are done.
2572 	 */
2573 	if ((dzp_error = zfs_zaccess_common(dzp, ACE_DELETE_CHILD,
2574 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr)) == 0)
2575 		return (0);
2576 
2577 	/*
2578 	 * If target object has delete permission then we are done
2579 	 */
2580 	if ((zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2581 	    &zpcheck_privs, B_FALSE, cr)) == 0)
2582 		return (0);
2583 
2584 	ASSERT(dzp_error);
2585 	ASSERT(zp_error);
2586 
2587 	if (!dzpcheck_privs)
2588 		return (dzp_error);
2589 	if (!zpcheck_privs)
2590 		return (zp_error);
2591 
2592 	/*
2593 	 * Second row
2594 	 *
2595 	 * If directory returns EACCES then delete_child was denied
2596 	 * due to deny delete_child.  In this case send the request through
2597 	 * secpolicy_vnode_remove().  We don't use zfs_delete_final_check()
2598 	 * since that *could* allow the delete based on write/execute permission
2599 	 * and we want delete permissions to override write/execute.
2600 	 */
2601 
2602 	if (dzp_error == EACCES) {
2603 		/* XXXPJD: s/dzp/zp/ ? */
2604 		return (secpolicy_vnode_remove(ZTOV(dzp), cr));
2605 	}
2606 	/*
2607 	 * Third Row
2608 	 * only need to see if we have write/execute on directory.
2609 	 */
2610 
2611 	dzp_error = zfs_zaccess_common(dzp, ACE_EXECUTE|ACE_WRITE_DATA,
2612 	    &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr);
2613 
2614 	if (dzp_error != 0 && !dzpcheck_privs)
2615 		return (dzp_error);
2616 
2617 	/*
2618 	 * Fourth row
2619 	 */
2620 
2621 	available_perms = (dzp_working_mode & ACE_WRITE_DATA) ? 0 : VWRITE;
2622 	available_perms |= (dzp_working_mode & ACE_EXECUTE) ? 0 : VEXEC;
2623 
2624 	return (zfs_delete_final_check(zp, dzp, available_perms, cr));
2625 
2626 }
2627 
2628 int
zfs_zaccess_rename(znode_t * sdzp,znode_t * szp,znode_t * tdzp,znode_t * tzp,cred_t * cr,zidmap_t * mnt_ns)2629 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2630     znode_t *tzp, cred_t *cr, zidmap_t *mnt_ns)
2631 {
2632 	int add_perm;
2633 	int error;
2634 
2635 	if (szp->z_pflags & ZFS_AV_QUARANTINED)
2636 		return (SET_ERROR(EACCES));
2637 
2638 	add_perm = (ZTOV(szp)->v_type == VDIR) ?
2639 	    ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
2640 
2641 	/*
2642 	 * Rename permissions are combination of delete permission +
2643 	 * add file/subdir permission.
2644 	 *
2645 	 * BSD operating systems also require write permission
2646 	 * on the directory being moved from one parent directory
2647 	 * to another.
2648 	 */
2649 	if (ZTOV(szp)->v_type == VDIR && ZTOV(sdzp) != ZTOV(tdzp)) {
2650 		if ((error = zfs_zaccess(szp, ACE_WRITE_DATA, 0, B_FALSE, cr,
2651 		    mnt_ns)))
2652 			return (error);
2653 	}
2654 
2655 	/*
2656 	 * first make sure we do the delete portion.
2657 	 *
2658 	 * If that succeeds then check for add_file/add_subdir permissions
2659 	 */
2660 
2661 	if ((error = zfs_zaccess_delete(sdzp, szp, cr, mnt_ns)))
2662 		return (error);
2663 
2664 	/*
2665 	 * If we have a tzp, see if we can delete it?
2666 	 */
2667 	if (tzp && (error = zfs_zaccess_delete(tdzp, tzp, cr, mnt_ns)))
2668 		return (error);
2669 
2670 	/*
2671 	 * Now check for add permissions
2672 	 */
2673 	error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr, mnt_ns);
2674 
2675 	return (error);
2676 }
2677