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 2014 Nexenta Systems, Inc. All rights reserved.
26 */
27
28
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/time.h>
32 #include <sys/sysmacros.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/sid.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/fs/zfs.h>
42 #include <sys/policy.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_fuid.h>
45 #include <sys/zfs_acl.h>
46 #include <sys/zfs_dir.h>
47 #include <sys/zfs_quota.h>
48 #include <sys/zfs_vfsops.h>
49 #include <sys/dmu.h>
50 #include <sys/dnode.h>
51 #include <sys/zap.h>
52 #include <sys/sa.h>
53 #include <sys/trace_acl.h>
54 #include <sys/zpl.h>
55
56 #define ALLOW ACE_ACCESS_ALLOWED_ACE_TYPE
57 #define DENY ACE_ACCESS_DENIED_ACE_TYPE
58 #define MAX_ACE_TYPE ACE_SYSTEM_ALARM_CALLBACK_OBJECT_ACE_TYPE
59 #define MIN_ACE_TYPE ALLOW
60
61 #define OWNING_GROUP (ACE_GROUP|ACE_IDENTIFIER_GROUP)
62 #define EVERYONE_ALLOW_MASK (ACE_READ_ACL|ACE_READ_ATTRIBUTES | \
63 ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE)
64 #define EVERYONE_DENY_MASK (ACE_WRITE_ACL|ACE_WRITE_OWNER | \
65 ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
66 #define OWNER_ALLOW_MASK (ACE_WRITE_ACL | ACE_WRITE_OWNER | \
67 ACE_WRITE_ATTRIBUTES|ACE_WRITE_NAMED_ATTRS)
68
69 #define ZFS_CHECKED_MASKS (ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_DATA| \
70 ACE_READ_NAMED_ATTRS|ACE_WRITE_DATA|ACE_WRITE_ATTRIBUTES| \
71 ACE_WRITE_NAMED_ATTRS|ACE_APPEND_DATA|ACE_EXECUTE|ACE_WRITE_OWNER| \
72 ACE_WRITE_ACL|ACE_DELETE|ACE_DELETE_CHILD|ACE_SYNCHRONIZE)
73
74 #define WRITE_MASK_DATA (ACE_WRITE_DATA|ACE_APPEND_DATA|ACE_WRITE_NAMED_ATTRS)
75 #define WRITE_MASK_ATTRS (ACE_WRITE_ACL|ACE_WRITE_OWNER|ACE_WRITE_ATTRIBUTES| \
76 ACE_DELETE|ACE_DELETE_CHILD)
77 #define WRITE_MASK (WRITE_MASK_DATA|WRITE_MASK_ATTRS)
78
79 #define OGE_CLEAR (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
80 ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
81
82 #define OKAY_MASK_BITS (ACE_READ_DATA|ACE_LIST_DIRECTORY|ACE_WRITE_DATA| \
83 ACE_ADD_FILE|ACE_APPEND_DATA|ACE_ADD_SUBDIRECTORY|ACE_EXECUTE)
84
85 #define ALL_INHERIT (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE | \
86 ACE_NO_PROPAGATE_INHERIT_ACE|ACE_INHERIT_ONLY_ACE|ACE_INHERITED_ACE)
87
88 #define RESTRICTED_CLEAR (ACE_WRITE_ACL|ACE_WRITE_OWNER)
89
90 #define V4_ACL_WIDE_FLAGS (ZFS_ACL_AUTO_INHERIT|ZFS_ACL_DEFAULTED|\
91 ZFS_ACL_PROTECTED)
92
93 #define ZFS_ACL_WIDE_FLAGS (V4_ACL_WIDE_FLAGS|ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|\
94 ZFS_ACL_OBJ_ACE)
95
96 #define ALL_MODE_EXECS (S_IXUSR | S_IXGRP | S_IXOTH)
97
98 #define IDMAP_WK_CREATOR_OWNER_UID 2147483648U
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 .ace_mask_get = zfs_ace_v0_get_mask,
177 .ace_mask_set = zfs_ace_v0_set_mask,
178 .ace_flags_get = zfs_ace_v0_get_flags,
179 .ace_flags_set = zfs_ace_v0_set_flags,
180 .ace_type_get = zfs_ace_v0_get_type,
181 .ace_type_set = zfs_ace_v0_set_type,
182 .ace_who_get = zfs_ace_v0_get_who,
183 .ace_who_set = zfs_ace_v0_set_who,
184 .ace_size = zfs_ace_v0_size,
185 .ace_abstract_size = zfs_ace_v0_abstract_size,
186 .ace_mask_off = zfs_ace_v0_mask_off,
187 .ace_data = 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 .ace_mask_get = zfs_ace_fuid_get_mask,
313 .ace_mask_set = zfs_ace_fuid_set_mask,
314 .ace_flags_get = zfs_ace_fuid_get_flags,
315 .ace_flags_set = zfs_ace_fuid_set_flags,
316 .ace_type_get = zfs_ace_fuid_get_type,
317 .ace_type_set = zfs_ace_fuid_set_type,
318 .ace_who_get = zfs_ace_fuid_get_who,
319 .ace_who_set = zfs_ace_fuid_set_who,
320 .ace_size = zfs_ace_fuid_size,
321 .ace_abstract_size = zfs_ace_fuid_abstract_size,
322 .ace_mask_off = zfs_ace_fuid_mask_off,
323 .ace_data = 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(ZTOZSB(zp)),
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 && error == ENOENT);
359 return (0);
360 }
361 }
362
363 /*
364 * Determine size of ACL in bytes
365 *
366 * This is more complicated than it should be since we have to deal
367 * with old external ACLs.
368 */
369 static int
zfs_acl_znode_info(znode_t * zp,int * aclsize,int * aclcount,zfs_acl_phys_t * aclphys)370 zfs_acl_znode_info(znode_t *zp, int *aclsize, int *aclcount,
371 zfs_acl_phys_t *aclphys)
372 {
373 zfsvfs_t *zfsvfs = ZTOZSB(zp);
374 uint64_t acl_count;
375 int size;
376 int error;
377
378 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
379 if (zp->z_is_sa) {
380 if ((error = sa_size(zp->z_sa_hdl, SA_ZPL_DACL_ACES(zfsvfs),
381 &size)) != 0)
382 return (error);
383 *aclsize = size;
384 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_COUNT(zfsvfs),
385 &acl_count, sizeof (acl_count))) != 0)
386 return (error);
387 *aclcount = acl_count;
388 } else {
389 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
390 aclphys, sizeof (*aclphys))) != 0)
391 return (error);
392
393 if (aclphys->z_acl_version == ZFS_ACL_VERSION_INITIAL) {
394 *aclsize = ZFS_ACL_SIZE(aclphys->z_acl_size);
395 *aclcount = aclphys->z_acl_size;
396 } else {
397 *aclsize = aclphys->z_acl_size;
398 *aclcount = aclphys->z_acl_count;
399 }
400 }
401 return (0);
402 }
403
404 int
zfs_znode_acl_version(znode_t * zp)405 zfs_znode_acl_version(znode_t *zp)
406 {
407 zfs_acl_phys_t acl_phys;
408
409 if (zp->z_is_sa)
410 return (ZFS_ACL_VERSION_FUID);
411 else {
412 int error;
413
414 /*
415 * Need to deal with a potential
416 * race where zfs_sa_upgrade could cause
417 * z_isa_sa to change.
418 *
419 * If the lookup fails then the state of z_is_sa should have
420 * changed.
421 */
422 if ((error = sa_lookup(zp->z_sa_hdl,
423 SA_ZPL_ZNODE_ACL(ZTOZSB(zp)),
424 &acl_phys, sizeof (acl_phys))) == 0)
425 return (acl_phys.z_acl_version);
426 else {
427 /*
428 * After upgrade SA_ZPL_ZNODE_ACL should have
429 * been removed.
430 */
431 VERIFY(zp->z_is_sa && error == ENOENT);
432 return (ZFS_ACL_VERSION_FUID);
433 }
434 }
435 }
436
437 static int
zfs_acl_version(int version)438 zfs_acl_version(int version)
439 {
440 if (version < ZPL_VERSION_FUID)
441 return (ZFS_ACL_VERSION_INITIAL);
442 else
443 return (ZFS_ACL_VERSION_FUID);
444 }
445
446 static int
zfs_acl_version_zp(znode_t * zp)447 zfs_acl_version_zp(znode_t *zp)
448 {
449 return (zfs_acl_version(ZTOZSB(zp)->z_version));
450 }
451
452 zfs_acl_t *
zfs_acl_alloc(int vers)453 zfs_acl_alloc(int vers)
454 {
455 zfs_acl_t *aclp;
456
457 aclp = kmem_zalloc(sizeof (zfs_acl_t), KM_SLEEP);
458 list_create(&aclp->z_acl, sizeof (zfs_acl_node_t),
459 offsetof(zfs_acl_node_t, z_next));
460 aclp->z_version = vers;
461 if (vers == ZFS_ACL_VERSION_FUID)
462 aclp->z_ops = &zfs_acl_fuid_ops;
463 else
464 aclp->z_ops = &zfs_acl_v0_ops;
465 return (aclp);
466 }
467
468 zfs_acl_node_t *
zfs_acl_node_alloc(size_t bytes)469 zfs_acl_node_alloc(size_t bytes)
470 {
471 zfs_acl_node_t *aclnode;
472
473 aclnode = kmem_zalloc(sizeof (zfs_acl_node_t), KM_SLEEP);
474 if (bytes) {
475 aclnode->z_acldata = kmem_zalloc(bytes, KM_SLEEP);
476 aclnode->z_allocdata = aclnode->z_acldata;
477 aclnode->z_allocsize = bytes;
478 aclnode->z_size = bytes;
479 }
480
481 return (aclnode);
482 }
483
484 static void
zfs_acl_node_free(zfs_acl_node_t * aclnode)485 zfs_acl_node_free(zfs_acl_node_t *aclnode)
486 {
487 if (aclnode->z_allocsize)
488 kmem_free(aclnode->z_allocdata, aclnode->z_allocsize);
489 kmem_free(aclnode, sizeof (zfs_acl_node_t));
490 }
491
492 static void
zfs_acl_release_nodes(zfs_acl_t * aclp)493 zfs_acl_release_nodes(zfs_acl_t *aclp)
494 {
495 zfs_acl_node_t *aclnode;
496
497 while ((aclnode = list_remove_head(&aclp->z_acl)))
498 zfs_acl_node_free(aclnode);
499 aclp->z_acl_count = 0;
500 aclp->z_acl_bytes = 0;
501 }
502
503 void
zfs_acl_free(zfs_acl_t * aclp)504 zfs_acl_free(zfs_acl_t *aclp)
505 {
506 zfs_acl_release_nodes(aclp);
507 list_destroy(&aclp->z_acl);
508 kmem_free(aclp, sizeof (zfs_acl_t));
509 }
510
511 static boolean_t
zfs_acl_valid_ace_type(uint_t type,uint_t flags)512 zfs_acl_valid_ace_type(uint_t type, uint_t flags)
513 {
514 uint16_t entry_type;
515
516 switch (type) {
517 case ALLOW:
518 case DENY:
519 case ACE_SYSTEM_AUDIT_ACE_TYPE:
520 case ACE_SYSTEM_ALARM_ACE_TYPE:
521 entry_type = flags & ACE_TYPE_FLAGS;
522 return (entry_type == ACE_OWNER ||
523 entry_type == OWNING_GROUP ||
524 entry_type == ACE_EVERYONE || entry_type == 0 ||
525 entry_type == ACE_IDENTIFIER_GROUP);
526 default:
527 if (type <= MAX_ACE_TYPE)
528 return (B_TRUE);
529 }
530 return (B_FALSE);
531 }
532
533 static boolean_t
zfs_ace_valid(umode_t obj_mode,zfs_acl_t * aclp,uint16_t type,uint16_t iflags)534 zfs_ace_valid(umode_t obj_mode, zfs_acl_t *aclp, uint16_t type, uint16_t iflags)
535 {
536 /*
537 * first check type of entry
538 */
539
540 if (!zfs_acl_valid_ace_type(type, iflags))
541 return (B_FALSE);
542
543 switch (type) {
544 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
545 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
546 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
547 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
548 if (aclp->z_version < ZFS_ACL_VERSION_FUID)
549 return (B_FALSE);
550 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
551 }
552
553 /*
554 * next check inheritance level flags
555 */
556
557 if (S_ISDIR(obj_mode) &&
558 (iflags & (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
559 aclp->z_hints |= ZFS_INHERIT_ACE;
560
561 if (iflags & (ACE_INHERIT_ONLY_ACE|ACE_NO_PROPAGATE_INHERIT_ACE)) {
562 if ((iflags & (ACE_FILE_INHERIT_ACE|
563 ACE_DIRECTORY_INHERIT_ACE)) == 0) {
564 return (B_FALSE);
565 }
566 }
567
568 return (B_TRUE);
569 }
570
571 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)572 zfs_acl_next_ace(zfs_acl_t *aclp, void *start, uint64_t *who,
573 uint32_t *access_mask, uint16_t *iflags, uint16_t *type)
574 {
575 zfs_acl_node_t *aclnode;
576
577 ASSERT(aclp);
578
579 if (start == NULL) {
580 aclnode = list_head(&aclp->z_acl);
581 if (aclnode == NULL)
582 return (NULL);
583
584 aclp->z_next_ace = aclnode->z_acldata;
585 aclp->z_curr_node = aclnode;
586 aclnode->z_ace_idx = 0;
587 }
588
589 aclnode = aclp->z_curr_node;
590
591 if (aclnode == NULL)
592 return (NULL);
593
594 if (aclnode->z_ace_idx >= aclnode->z_ace_count) {
595 aclnode = list_next(&aclp->z_acl, aclnode);
596 if (aclnode == NULL)
597 return (NULL);
598 else {
599 aclp->z_curr_node = aclnode;
600 aclnode->z_ace_idx = 0;
601 aclp->z_next_ace = aclnode->z_acldata;
602 }
603 }
604
605 if (aclnode->z_ace_idx < aclnode->z_ace_count) {
606 void *acep = aclp->z_next_ace;
607 size_t ace_size;
608
609 /*
610 * Make sure we don't overstep our bounds
611 */
612 ace_size = aclp->z_ops->ace_size(acep);
613
614 if (((caddr_t)acep + ace_size) >
615 ((caddr_t)aclnode->z_acldata + aclnode->z_size)) {
616 return (NULL);
617 }
618
619 *iflags = aclp->z_ops->ace_flags_get(acep);
620 *type = aclp->z_ops->ace_type_get(acep);
621 *access_mask = aclp->z_ops->ace_mask_get(acep);
622 *who = aclp->z_ops->ace_who_get(acep);
623 aclp->z_next_ace = (caddr_t)aclp->z_next_ace + ace_size;
624 aclnode->z_ace_idx++;
625
626 return ((void *)acep);
627 }
628 return (NULL);
629 }
630
631 static uintptr_t
zfs_ace_walk(void * datap,uintptr_t cookie,int aclcnt,uint16_t * flags,uint16_t * type,uint32_t * mask)632 zfs_ace_walk(void *datap, uintptr_t cookie, int aclcnt,
633 uint16_t *flags, uint16_t *type, uint32_t *mask)
634 {
635 (void) aclcnt;
636 zfs_acl_t *aclp = datap;
637 zfs_ace_hdr_t *acep = (zfs_ace_hdr_t *)cookie;
638 uint64_t who;
639
640 acep = zfs_acl_next_ace(aclp, acep, &who, mask,
641 flags, type);
642 return ((uintptr_t)acep);
643 }
644
645 /*
646 * Copy ACE to internal ZFS format.
647 * While processing the ACL each ACE will be validated for correctness.
648 * ACE FUIDs will be created later.
649 */
650 static int
zfs_copy_ace_2_fuid(zfsvfs_t * zfsvfs,umode_t obj_mode,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)651 zfs_copy_ace_2_fuid(zfsvfs_t *zfsvfs, umode_t obj_mode, zfs_acl_t *aclp,
652 void *datap, zfs_ace_t *z_acl, uint64_t aclcnt, size_t *size,
653 zfs_fuid_info_t **fuidp, cred_t *cr)
654 {
655 int i;
656 uint16_t entry_type;
657 zfs_ace_t *aceptr = z_acl;
658 ace_t *acep = datap;
659 zfs_object_ace_t *zobjacep;
660 ace_object_t *aceobjp;
661
662 for (i = 0; i != aclcnt; i++) {
663 aceptr->z_hdr.z_access_mask = acep->a_access_mask;
664 aceptr->z_hdr.z_flags = acep->a_flags;
665 aceptr->z_hdr.z_type = acep->a_type;
666 entry_type = aceptr->z_hdr.z_flags & ACE_TYPE_FLAGS;
667 if (entry_type != ACE_OWNER && entry_type != OWNING_GROUP &&
668 entry_type != ACE_EVERYONE) {
669 aceptr->z_fuid = zfs_fuid_create(zfsvfs, acep->a_who,
670 cr, (entry_type == 0) ?
671 ZFS_ACE_USER : ZFS_ACE_GROUP, fuidp);
672 }
673
674 /*
675 * Make sure ACE is valid
676 */
677 if (zfs_ace_valid(obj_mode, aclp, aceptr->z_hdr.z_type,
678 aceptr->z_hdr.z_flags) != B_TRUE)
679 return (SET_ERROR(EINVAL));
680
681 switch (acep->a_type) {
682 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
683 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
684 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
685 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
686 zobjacep = (zfs_object_ace_t *)aceptr;
687 aceobjp = (ace_object_t *)acep;
688
689 memcpy(zobjacep->z_object_type, aceobjp->a_obj_type,
690 sizeof (aceobjp->a_obj_type));
691 memcpy(zobjacep->z_inherit_type,
692 aceobjp->a_inherit_obj_type,
693 sizeof (aceobjp->a_inherit_obj_type));
694 acep = (ace_t *)((caddr_t)acep + sizeof (ace_object_t));
695 break;
696 default:
697 acep = (ace_t *)((caddr_t)acep + sizeof (ace_t));
698 }
699
700 aceptr = (zfs_ace_t *)((caddr_t)aceptr +
701 aclp->z_ops->ace_size(aceptr));
702 }
703
704 *size = (caddr_t)aceptr - (caddr_t)z_acl;
705
706 return (0);
707 }
708
709 /*
710 * Copy ZFS ACEs to fixed size ace_t layout
711 */
712 static void
zfs_copy_fuid_2_ace(zfsvfs_t * zfsvfs,zfs_acl_t * aclp,cred_t * cr,void * datap,int filter)713 zfs_copy_fuid_2_ace(zfsvfs_t *zfsvfs, zfs_acl_t *aclp, cred_t *cr,
714 void *datap, int filter)
715 {
716 uint64_t who;
717 uint32_t access_mask;
718 uint16_t iflags, type;
719 zfs_ace_hdr_t *zacep = NULL;
720 ace_t *acep = datap;
721 ace_object_t *objacep;
722 zfs_object_ace_t *zobjacep;
723 size_t ace_size;
724 uint16_t entry_type;
725
726 while ((zacep = zfs_acl_next_ace(aclp, zacep,
727 &who, &access_mask, &iflags, &type))) {
728
729 switch (type) {
730 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
731 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
732 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
733 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
734 if (filter) {
735 continue;
736 }
737 zobjacep = (zfs_object_ace_t *)zacep;
738 objacep = (ace_object_t *)acep;
739 memcpy(objacep->a_obj_type,
740 zobjacep->z_object_type,
741 sizeof (zobjacep->z_object_type));
742 memcpy(objacep->a_inherit_obj_type,
743 zobjacep->z_inherit_type,
744 sizeof (zobjacep->z_inherit_type));
745 ace_size = sizeof (ace_object_t);
746 break;
747 default:
748 ace_size = sizeof (ace_t);
749 break;
750 }
751
752 entry_type = (iflags & ACE_TYPE_FLAGS);
753 if ((entry_type != ACE_OWNER &&
754 entry_type != OWNING_GROUP &&
755 entry_type != ACE_EVERYONE)) {
756 acep->a_who = zfs_fuid_map_id(zfsvfs, who,
757 cr, (entry_type & ACE_IDENTIFIER_GROUP) ?
758 ZFS_ACE_GROUP : ZFS_ACE_USER);
759 } else {
760 acep->a_who = (uid_t)(int64_t)who;
761 }
762 acep->a_access_mask = access_mask;
763 acep->a_flags = iflags;
764 acep->a_type = type;
765 acep = (ace_t *)((caddr_t)acep + ace_size);
766 }
767 }
768
769 static int
zfs_copy_ace_2_oldace(umode_t obj_mode,zfs_acl_t * aclp,ace_t * acep,zfs_oldace_t * z_acl,int aclcnt,size_t * size)770 zfs_copy_ace_2_oldace(umode_t obj_mode, zfs_acl_t *aclp, ace_t *acep,
771 zfs_oldace_t *z_acl, int aclcnt, size_t *size)
772 {
773 int i;
774 zfs_oldace_t *aceptr = z_acl;
775
776 for (i = 0; i != aclcnt; i++, aceptr++) {
777 aceptr->z_access_mask = acep[i].a_access_mask;
778 aceptr->z_type = acep[i].a_type;
779 aceptr->z_flags = acep[i].a_flags;
780 aceptr->z_fuid = acep[i].a_who;
781 /*
782 * Make sure ACE is valid
783 */
784 if (zfs_ace_valid(obj_mode, aclp, aceptr->z_type,
785 aceptr->z_flags) != B_TRUE)
786 return (SET_ERROR(EINVAL));
787 }
788 *size = (caddr_t)aceptr - (caddr_t)z_acl;
789 return (0);
790 }
791
792 /*
793 * convert old ACL format to new
794 */
795 void
zfs_acl_xform(znode_t * zp,zfs_acl_t * aclp,cred_t * cr)796 zfs_acl_xform(znode_t *zp, zfs_acl_t *aclp, cred_t *cr)
797 {
798 zfs_oldace_t *oldaclp;
799 int i;
800 uint16_t type, iflags;
801 uint32_t access_mask;
802 uint64_t who;
803 void *cookie = NULL;
804 zfs_acl_node_t *newaclnode;
805
806 ASSERT(aclp->z_version == ZFS_ACL_VERSION_INITIAL);
807 /*
808 * First create the ACE in a contiguous piece of memory
809 * for zfs_copy_ace_2_fuid().
810 *
811 * We only convert an ACL once, so this won't happen
812 * every time.
813 */
814 oldaclp = kmem_alloc(sizeof (zfs_oldace_t) * aclp->z_acl_count,
815 KM_SLEEP);
816 i = 0;
817 while ((cookie = zfs_acl_next_ace(aclp, cookie, &who,
818 &access_mask, &iflags, &type))) {
819 oldaclp[i].z_flags = iflags;
820 oldaclp[i].z_type = type;
821 oldaclp[i].z_fuid = who;
822 oldaclp[i++].z_access_mask = access_mask;
823 }
824
825 newaclnode = zfs_acl_node_alloc(aclp->z_acl_count *
826 sizeof (zfs_object_ace_t));
827 aclp->z_ops = &zfs_acl_fuid_ops;
828 VERIFY(zfs_copy_ace_2_fuid(ZTOZSB(zp), ZTOI(zp)->i_mode,
829 aclp, oldaclp, newaclnode->z_acldata, aclp->z_acl_count,
830 &newaclnode->z_size, NULL, cr) == 0);
831 newaclnode->z_ace_count = aclp->z_acl_count;
832 aclp->z_version = ZFS_ACL_VERSION;
833 kmem_free(oldaclp, aclp->z_acl_count * sizeof (zfs_oldace_t));
834
835 /*
836 * Release all previous ACL nodes
837 */
838
839 zfs_acl_release_nodes(aclp);
840
841 list_insert_head(&aclp->z_acl, newaclnode);
842
843 aclp->z_acl_bytes = newaclnode->z_size;
844 aclp->z_acl_count = newaclnode->z_ace_count;
845
846 }
847
848 /*
849 * Convert unix access mask to v4 access mask
850 */
851 static uint32_t
zfs_unix_to_v4(uint32_t access_mask)852 zfs_unix_to_v4(uint32_t access_mask)
853 {
854 uint32_t new_mask = 0;
855
856 if (access_mask & S_IXOTH)
857 new_mask |= ACE_EXECUTE;
858 if (access_mask & S_IWOTH)
859 new_mask |= ACE_WRITE_DATA;
860 if (access_mask & S_IROTH)
861 new_mask |= ACE_READ_DATA;
862 return (new_mask);
863 }
864
865
866 static int
zfs_v4_to_unix(uint32_t access_mask,int * unmapped)867 zfs_v4_to_unix(uint32_t access_mask, int *unmapped)
868 {
869 int new_mask = 0;
870
871 *unmapped = access_mask &
872 (ACE_WRITE_OWNER | ACE_WRITE_ACL | ACE_DELETE);
873
874 if (access_mask & WRITE_MASK)
875 new_mask |= S_IWOTH;
876 if (access_mask & ACE_READ_DATA)
877 new_mask |= S_IROTH;
878 if (access_mask & ACE_EXECUTE)
879 new_mask |= S_IXOTH;
880
881 return (new_mask);
882 }
883
884
885 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)886 zfs_set_ace(zfs_acl_t *aclp, void *acep, uint32_t access_mask,
887 uint16_t access_type, uint64_t fuid, uint16_t entry_type)
888 {
889 uint16_t type = entry_type & ACE_TYPE_FLAGS;
890
891 aclp->z_ops->ace_mask_set(acep, access_mask);
892 aclp->z_ops->ace_type_set(acep, access_type);
893 aclp->z_ops->ace_flags_set(acep, entry_type);
894 if ((type != ACE_OWNER && type != OWNING_GROUP &&
895 type != ACE_EVERYONE))
896 aclp->z_ops->ace_who_set(acep, fuid);
897 }
898
899 /*
900 * Determine mode of file based on ACL.
901 */
902 uint64_t
zfs_mode_compute(uint64_t fmode,zfs_acl_t * aclp,uint64_t * pflags,uint64_t fuid,uint64_t fgid)903 zfs_mode_compute(uint64_t fmode, zfs_acl_t *aclp,
904 uint64_t *pflags, uint64_t fuid, uint64_t fgid)
905 {
906 int entry_type;
907 mode_t mode;
908 mode_t seen = 0;
909 zfs_ace_hdr_t *acep = NULL;
910 uint64_t who;
911 uint16_t iflags, type;
912 uint32_t access_mask;
913 boolean_t an_exec_denied = B_FALSE;
914
915 mode = (fmode & (S_IFMT | S_ISUID | S_ISGID | S_ISVTX));
916
917 while ((acep = zfs_acl_next_ace(aclp, acep, &who,
918 &access_mask, &iflags, &type))) {
919
920 if (!zfs_acl_valid_ace_type(type, iflags))
921 continue;
922
923 entry_type = (iflags & ACE_TYPE_FLAGS);
924
925 /*
926 * Skip over any inherit_only ACEs
927 */
928 if (iflags & ACE_INHERIT_ONLY_ACE)
929 continue;
930
931 if (entry_type == ACE_OWNER || (entry_type == 0 &&
932 who == fuid)) {
933 if ((access_mask & ACE_READ_DATA) &&
934 (!(seen & S_IRUSR))) {
935 seen |= S_IRUSR;
936 if (type == ALLOW) {
937 mode |= S_IRUSR;
938 }
939 }
940 if ((access_mask & ACE_WRITE_DATA) &&
941 (!(seen & S_IWUSR))) {
942 seen |= S_IWUSR;
943 if (type == ALLOW) {
944 mode |= S_IWUSR;
945 }
946 }
947 if ((access_mask & ACE_EXECUTE) &&
948 (!(seen & S_IXUSR))) {
949 seen |= S_IXUSR;
950 if (type == ALLOW) {
951 mode |= S_IXUSR;
952 }
953 }
954 } else if (entry_type == OWNING_GROUP ||
955 (entry_type == ACE_IDENTIFIER_GROUP && who == fgid)) {
956 if ((access_mask & ACE_READ_DATA) &&
957 (!(seen & S_IRGRP))) {
958 seen |= S_IRGRP;
959 if (type == ALLOW) {
960 mode |= S_IRGRP;
961 }
962 }
963 if ((access_mask & ACE_WRITE_DATA) &&
964 (!(seen & S_IWGRP))) {
965 seen |= S_IWGRP;
966 if (type == ALLOW) {
967 mode |= S_IWGRP;
968 }
969 }
970 if ((access_mask & ACE_EXECUTE) &&
971 (!(seen & S_IXGRP))) {
972 seen |= S_IXGRP;
973 if (type == ALLOW) {
974 mode |= S_IXGRP;
975 }
976 }
977 } else if (entry_type == ACE_EVERYONE) {
978 if ((access_mask & ACE_READ_DATA)) {
979 if (!(seen & S_IRUSR)) {
980 seen |= S_IRUSR;
981 if (type == ALLOW) {
982 mode |= S_IRUSR;
983 }
984 }
985 if (!(seen & S_IRGRP)) {
986 seen |= S_IRGRP;
987 if (type == ALLOW) {
988 mode |= S_IRGRP;
989 }
990 }
991 if (!(seen & S_IROTH)) {
992 seen |= S_IROTH;
993 if (type == ALLOW) {
994 mode |= S_IROTH;
995 }
996 }
997 }
998 if ((access_mask & ACE_WRITE_DATA)) {
999 if (!(seen & S_IWUSR)) {
1000 seen |= S_IWUSR;
1001 if (type == ALLOW) {
1002 mode |= S_IWUSR;
1003 }
1004 }
1005 if (!(seen & S_IWGRP)) {
1006 seen |= S_IWGRP;
1007 if (type == ALLOW) {
1008 mode |= S_IWGRP;
1009 }
1010 }
1011 if (!(seen & S_IWOTH)) {
1012 seen |= S_IWOTH;
1013 if (type == ALLOW) {
1014 mode |= S_IWOTH;
1015 }
1016 }
1017 }
1018 if ((access_mask & ACE_EXECUTE)) {
1019 if (!(seen & S_IXUSR)) {
1020 seen |= S_IXUSR;
1021 if (type == ALLOW) {
1022 mode |= S_IXUSR;
1023 }
1024 }
1025 if (!(seen & S_IXGRP)) {
1026 seen |= S_IXGRP;
1027 if (type == ALLOW) {
1028 mode |= S_IXGRP;
1029 }
1030 }
1031 if (!(seen & S_IXOTH)) {
1032 seen |= S_IXOTH;
1033 if (type == ALLOW) {
1034 mode |= S_IXOTH;
1035 }
1036 }
1037 }
1038 } else {
1039 /*
1040 * Only care if this IDENTIFIER_GROUP or
1041 * USER ACE denies execute access to someone,
1042 * mode is not affected
1043 */
1044 if ((access_mask & ACE_EXECUTE) && type == DENY)
1045 an_exec_denied = B_TRUE;
1046 }
1047 }
1048
1049 /*
1050 * Failure to allow is effectively a deny, so execute permission
1051 * is denied if it was never mentioned or if we explicitly
1052 * weren't allowed it.
1053 */
1054 if (!an_exec_denied &&
1055 ((seen & ALL_MODE_EXECS) != ALL_MODE_EXECS ||
1056 (mode & ALL_MODE_EXECS) != ALL_MODE_EXECS))
1057 an_exec_denied = B_TRUE;
1058
1059 if (an_exec_denied)
1060 *pflags &= ~ZFS_NO_EXECS_DENIED;
1061 else
1062 *pflags |= ZFS_NO_EXECS_DENIED;
1063
1064 return (mode);
1065 }
1066
1067 /*
1068 * Read an external acl object. If the intent is to modify, always
1069 * create a new acl and leave any cached acl in place.
1070 */
1071 int
zfs_acl_node_read(struct znode * zp,boolean_t have_lock,zfs_acl_t ** aclpp,boolean_t will_modify)1072 zfs_acl_node_read(struct znode *zp, boolean_t have_lock, zfs_acl_t **aclpp,
1073 boolean_t will_modify)
1074 {
1075 zfs_acl_t *aclp;
1076 int aclsize = 0;
1077 int acl_count = 0;
1078 zfs_acl_node_t *aclnode;
1079 zfs_acl_phys_t znode_acl;
1080 int version;
1081 int error;
1082 boolean_t drop_lock = B_FALSE;
1083
1084 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1085
1086 if (zp->z_acl_cached && !will_modify) {
1087 *aclpp = zp->z_acl_cached;
1088 return (0);
1089 }
1090
1091 /*
1092 * close race where znode could be upgrade while trying to
1093 * read the znode attributes.
1094 *
1095 * But this could only happen if the file isn't already an SA
1096 * znode
1097 */
1098 if (!zp->z_is_sa && !have_lock) {
1099 mutex_enter(&zp->z_lock);
1100 drop_lock = B_TRUE;
1101 }
1102 version = zfs_znode_acl_version(zp);
1103
1104 if ((error = zfs_acl_znode_info(zp, &aclsize,
1105 &acl_count, &znode_acl)) != 0) {
1106 goto done;
1107 }
1108
1109 aclp = zfs_acl_alloc(version);
1110
1111 aclp->z_acl_count = acl_count;
1112 aclp->z_acl_bytes = aclsize;
1113
1114 aclnode = zfs_acl_node_alloc(aclsize);
1115 aclnode->z_ace_count = aclp->z_acl_count;
1116 aclnode->z_size = aclsize;
1117
1118 if (!zp->z_is_sa) {
1119 if (znode_acl.z_acl_extern_obj) {
1120 error = dmu_read(ZTOZSB(zp)->z_os,
1121 znode_acl.z_acl_extern_obj, 0, aclnode->z_size,
1122 aclnode->z_acldata, DMU_READ_PREFETCH);
1123 } else {
1124 memcpy(aclnode->z_acldata, znode_acl.z_ace_data,
1125 aclnode->z_size);
1126 }
1127 } else {
1128 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_DACL_ACES(ZTOZSB(zp)),
1129 aclnode->z_acldata, aclnode->z_size);
1130 }
1131
1132 if (error != 0) {
1133 zfs_acl_free(aclp);
1134 zfs_acl_node_free(aclnode);
1135 /* convert checksum errors into IO errors */
1136 if (error == ECKSUM)
1137 error = SET_ERROR(EIO);
1138 goto done;
1139 }
1140
1141 list_insert_head(&aclp->z_acl, aclnode);
1142
1143 *aclpp = aclp;
1144 if (!will_modify)
1145 zp->z_acl_cached = aclp;
1146 done:
1147 if (drop_lock)
1148 mutex_exit(&zp->z_lock);
1149 return (error);
1150 }
1151
1152 void
zfs_acl_data_locator(void ** dataptr,uint32_t * length,uint32_t buflen,boolean_t start,void * userdata)1153 zfs_acl_data_locator(void **dataptr, uint32_t *length, uint32_t buflen,
1154 boolean_t start, void *userdata)
1155 {
1156 (void) buflen;
1157 zfs_acl_locator_cb_t *cb = (zfs_acl_locator_cb_t *)userdata;
1158
1159 if (start) {
1160 cb->cb_acl_node = list_head(&cb->cb_aclp->z_acl);
1161 } else {
1162 cb->cb_acl_node = list_next(&cb->cb_aclp->z_acl,
1163 cb->cb_acl_node);
1164 }
1165 ASSERT3P(cb->cb_acl_node, !=, NULL);
1166 *dataptr = cb->cb_acl_node->z_acldata;
1167 *length = cb->cb_acl_node->z_size;
1168 }
1169
1170 int
zfs_acl_chown_setattr(znode_t * zp)1171 zfs_acl_chown_setattr(znode_t *zp)
1172 {
1173 int error;
1174 zfs_acl_t *aclp;
1175
1176 if (ZTOZSB(zp)->z_acl_type == ZFS_ACLTYPE_POSIX)
1177 return (0);
1178
1179 ASSERT(MUTEX_HELD(&zp->z_lock));
1180 ASSERT(MUTEX_HELD(&zp->z_acl_lock));
1181
1182 error = zfs_acl_node_read(zp, B_TRUE, &aclp, B_FALSE);
1183 if (error == 0 && aclp->z_acl_count > 0)
1184 zp->z_mode = ZTOI(zp)->i_mode =
1185 zfs_mode_compute(zp->z_mode, aclp,
1186 &zp->z_pflags, KUID_TO_SUID(ZTOI(zp)->i_uid),
1187 KGID_TO_SGID(ZTOI(zp)->i_gid));
1188
1189 /*
1190 * Some ZFS implementations (ZEVO) create neither a ZNODE_ACL
1191 * nor a DACL_ACES SA in which case ENOENT is returned from
1192 * zfs_acl_node_read() when the SA can't be located.
1193 * Allow chown/chgrp to succeed in these cases rather than
1194 * returning an error that makes no sense in the context of
1195 * the caller.
1196 */
1197 if (error == ENOENT)
1198 return (0);
1199
1200 return (error);
1201 }
1202
1203 typedef struct trivial_acl {
1204 uint32_t allow0; /* allow mask for bits only in owner */
1205 uint32_t deny1; /* deny mask for bits not in owner */
1206 uint32_t deny2; /* deny mask for bits not in group */
1207 uint32_t owner; /* allow mask matching mode */
1208 uint32_t group; /* allow mask matching mode */
1209 uint32_t everyone; /* allow mask matching mode */
1210 } trivial_acl_t;
1211
1212 static void
acl_trivial_access_masks(mode_t mode,boolean_t isdir,trivial_acl_t * masks)1213 acl_trivial_access_masks(mode_t mode, boolean_t isdir, trivial_acl_t *masks)
1214 {
1215 uint32_t read_mask = ACE_READ_DATA;
1216 uint32_t write_mask = ACE_WRITE_DATA|ACE_APPEND_DATA;
1217 uint32_t execute_mask = ACE_EXECUTE;
1218
1219 if (isdir)
1220 write_mask |= ACE_DELETE_CHILD;
1221
1222 masks->deny1 = 0;
1223
1224 if (!(mode & S_IRUSR) && (mode & (S_IRGRP|S_IROTH)))
1225 masks->deny1 |= read_mask;
1226 if (!(mode & S_IWUSR) && (mode & (S_IWGRP|S_IWOTH)))
1227 masks->deny1 |= write_mask;
1228 if (!(mode & S_IXUSR) && (mode & (S_IXGRP|S_IXOTH)))
1229 masks->deny1 |= execute_mask;
1230
1231 masks->deny2 = 0;
1232 if (!(mode & S_IRGRP) && (mode & S_IROTH))
1233 masks->deny2 |= read_mask;
1234 if (!(mode & S_IWGRP) && (mode & S_IWOTH))
1235 masks->deny2 |= write_mask;
1236 if (!(mode & S_IXGRP) && (mode & S_IXOTH))
1237 masks->deny2 |= execute_mask;
1238
1239 masks->allow0 = 0;
1240 if ((mode & S_IRUSR) && (!(mode & S_IRGRP) && (mode & S_IROTH)))
1241 masks->allow0 |= read_mask;
1242 if ((mode & S_IWUSR) && (!(mode & S_IWGRP) && (mode & S_IWOTH)))
1243 masks->allow0 |= write_mask;
1244 if ((mode & S_IXUSR) && (!(mode & S_IXGRP) && (mode & S_IXOTH)))
1245 masks->allow0 |= execute_mask;
1246
1247 masks->owner = ACE_WRITE_ATTRIBUTES|ACE_WRITE_OWNER|ACE_WRITE_ACL|
1248 ACE_WRITE_NAMED_ATTRS|ACE_READ_ACL|ACE_READ_ATTRIBUTES|
1249 ACE_READ_NAMED_ATTRS|ACE_SYNCHRONIZE;
1250 if (mode & S_IRUSR)
1251 masks->owner |= read_mask;
1252 if (mode & S_IWUSR)
1253 masks->owner |= write_mask;
1254 if (mode & S_IXUSR)
1255 masks->owner |= execute_mask;
1256
1257 masks->group = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1258 ACE_SYNCHRONIZE;
1259 if (mode & S_IRGRP)
1260 masks->group |= read_mask;
1261 if (mode & S_IWGRP)
1262 masks->group |= write_mask;
1263 if (mode & S_IXGRP)
1264 masks->group |= execute_mask;
1265
1266 masks->everyone = ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_READ_NAMED_ATTRS|
1267 ACE_SYNCHRONIZE;
1268 if (mode & S_IROTH)
1269 masks->everyone |= read_mask;
1270 if (mode & S_IWOTH)
1271 masks->everyone |= write_mask;
1272 if (mode & S_IXOTH)
1273 masks->everyone |= execute_mask;
1274 }
1275
1276 /*
1277 * ace_trivial:
1278 * determine whether an ace_t acl is trivial
1279 *
1280 * Trivialness implies that the acl is composed of only
1281 * owner, group, everyone entries. ACL can't
1282 * have read_acl denied, and write_owner/write_acl/write_attributes
1283 * can only be owner@ entry.
1284 */
1285 static int
ace_trivial_common(void * acep,int aclcnt,uintptr_t (* walk)(void *,uintptr_t,int,uint16_t *,uint16_t *,uint32_t *))1286 ace_trivial_common(void *acep, int aclcnt,
1287 uintptr_t (*walk)(void *, uintptr_t, int,
1288 uint16_t *, uint16_t *, uint32_t *))
1289 {
1290 uint16_t flags;
1291 uint32_t mask;
1292 uint16_t type;
1293 uint64_t cookie = 0;
1294
1295 while ((cookie = walk(acep, cookie, aclcnt, &flags, &type, &mask))) {
1296 switch (flags & ACE_TYPE_FLAGS) {
1297 case ACE_OWNER:
1298 case ACE_GROUP|ACE_IDENTIFIER_GROUP:
1299 case ACE_EVERYONE:
1300 break;
1301 default:
1302 return (1);
1303 }
1304
1305 if (flags & (ACE_FILE_INHERIT_ACE|
1306 ACE_DIRECTORY_INHERIT_ACE|ACE_NO_PROPAGATE_INHERIT_ACE|
1307 ACE_INHERIT_ONLY_ACE))
1308 return (1);
1309
1310 /*
1311 * Special check for some special bits
1312 *
1313 * Don't allow anybody to deny reading basic
1314 * attributes or a files ACL.
1315 */
1316 if ((mask & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
1317 (type == ACE_ACCESS_DENIED_ACE_TYPE))
1318 return (1);
1319
1320 /*
1321 * Delete permission is never set by default
1322 */
1323 if (mask & ACE_DELETE)
1324 return (1);
1325
1326 /*
1327 * Child delete permission should be accompanied by write
1328 */
1329 if ((mask & ACE_DELETE_CHILD) && !(mask & ACE_WRITE_DATA))
1330 return (1);
1331
1332 /*
1333 * only allow owner@ to have
1334 * write_acl/write_owner/write_attributes/write_xattr/
1335 */
1336 if (type == ACE_ACCESS_ALLOWED_ACE_TYPE &&
1337 (!(flags & ACE_OWNER) && (mask &
1338 (ACE_WRITE_OWNER|ACE_WRITE_ACL| ACE_WRITE_ATTRIBUTES|
1339 ACE_WRITE_NAMED_ATTRS))))
1340 return (1);
1341
1342 }
1343
1344 return (0);
1345 }
1346
1347 /*
1348 * common code for setting ACLs.
1349 *
1350 * This function is called from zfs_mode_update, zfs_perm_init, and zfs_setacl.
1351 * zfs_setacl passes a non-NULL inherit pointer (ihp) to indicate that it's
1352 * already checked the acl and knows whether to inherit.
1353 */
1354 int
zfs_aclset_common(znode_t * zp,zfs_acl_t * aclp,cred_t * cr,dmu_tx_t * tx)1355 zfs_aclset_common(znode_t *zp, zfs_acl_t *aclp, cred_t *cr, dmu_tx_t *tx)
1356 {
1357 int error;
1358 zfsvfs_t *zfsvfs = ZTOZSB(zp);
1359 dmu_object_type_t otype;
1360 zfs_acl_locator_cb_t locate = { 0 };
1361 uint64_t mode;
1362 sa_bulk_attr_t bulk[5];
1363 uint64_t ctime[2];
1364 int count = 0;
1365 zfs_acl_phys_t acl_phys;
1366
1367 mode = zp->z_mode;
1368
1369 mode = zfs_mode_compute(mode, aclp, &zp->z_pflags,
1370 KUID_TO_SUID(ZTOI(zp)->i_uid), KGID_TO_SGID(ZTOI(zp)->i_gid));
1371
1372 zp->z_mode = ZTOI(zp)->i_mode = mode;
1373 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL,
1374 &mode, sizeof (mode));
1375 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
1376 &zp->z_pflags, sizeof (zp->z_pflags));
1377 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
1378 &ctime, sizeof (ctime));
1379
1380 if (zp->z_acl_cached) {
1381 zfs_acl_free(zp->z_acl_cached);
1382 zp->z_acl_cached = NULL;
1383 }
1384
1385 /*
1386 * Upgrade needed?
1387 */
1388 if (!zfsvfs->z_use_fuids) {
1389 otype = DMU_OT_OLDACL;
1390 } else {
1391 if ((aclp->z_version == ZFS_ACL_VERSION_INITIAL) &&
1392 (zfsvfs->z_version >= ZPL_VERSION_FUID))
1393 zfs_acl_xform(zp, aclp, cr);
1394 ASSERT(aclp->z_version >= ZFS_ACL_VERSION_FUID);
1395 otype = DMU_OT_ACL;
1396 }
1397
1398 /*
1399 * Arrgh, we have to handle old on disk format
1400 * as well as newer (preferred) SA format.
1401 */
1402
1403 if (zp->z_is_sa) { /* the easy case, just update the ACL attribute */
1404 locate.cb_aclp = aclp;
1405 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_ACES(zfsvfs),
1406 zfs_acl_data_locator, &locate, aclp->z_acl_bytes);
1407 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_DACL_COUNT(zfsvfs),
1408 NULL, &aclp->z_acl_count, sizeof (uint64_t));
1409 } else { /* Painful legacy way */
1410 zfs_acl_node_t *aclnode;
1411 uint64_t off = 0;
1412 uint64_t aoid;
1413
1414 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_ZNODE_ACL(zfsvfs),
1415 &acl_phys, sizeof (acl_phys))) != 0)
1416 return (error);
1417
1418 aoid = acl_phys.z_acl_extern_obj;
1419
1420 if (aclp->z_acl_bytes > ZFS_ACE_SPACE) {
1421 /*
1422 * If ACL was previously external and we are now
1423 * converting to new ACL format then release old
1424 * ACL object and create a new one.
1425 */
1426 if (aoid &&
1427 aclp->z_version != acl_phys.z_acl_version) {
1428 error = dmu_object_free(zfsvfs->z_os, aoid, tx);
1429 if (error)
1430 return (error);
1431 aoid = 0;
1432 }
1433 if (aoid == 0) {
1434 aoid = dmu_object_alloc(zfsvfs->z_os,
1435 otype, aclp->z_acl_bytes,
1436 otype == DMU_OT_ACL ?
1437 DMU_OT_SYSACL : DMU_OT_NONE,
1438 otype == DMU_OT_ACL ?
1439 DN_OLD_MAX_BONUSLEN : 0, tx);
1440 } else {
1441 (void) dmu_object_set_blocksize(zfsvfs->z_os,
1442 aoid, aclp->z_acl_bytes, 0, tx);
1443 }
1444 acl_phys.z_acl_extern_obj = aoid;
1445 for (aclnode = list_head(&aclp->z_acl); aclnode;
1446 aclnode = list_next(&aclp->z_acl, aclnode)) {
1447 if (aclnode->z_ace_count == 0)
1448 continue;
1449 dmu_write(zfsvfs->z_os, aoid, off,
1450 aclnode->z_size, aclnode->z_acldata, tx);
1451 off += aclnode->z_size;
1452 }
1453 } else {
1454 void *start = acl_phys.z_ace_data;
1455 /*
1456 * Migrating back embedded?
1457 */
1458 if (acl_phys.z_acl_extern_obj) {
1459 error = dmu_object_free(zfsvfs->z_os,
1460 acl_phys.z_acl_extern_obj, tx);
1461 if (error)
1462 return (error);
1463 acl_phys.z_acl_extern_obj = 0;
1464 }
1465
1466 for (aclnode = list_head(&aclp->z_acl); aclnode;
1467 aclnode = list_next(&aclp->z_acl, aclnode)) {
1468 if (aclnode->z_ace_count == 0)
1469 continue;
1470 memcpy(start, aclnode->z_acldata,
1471 aclnode->z_size);
1472 start = (caddr_t)start + aclnode->z_size;
1473 }
1474 }
1475 /*
1476 * If Old version then swap count/bytes to match old
1477 * layout of znode_acl_phys_t.
1478 */
1479 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
1480 acl_phys.z_acl_size = aclp->z_acl_count;
1481 acl_phys.z_acl_count = aclp->z_acl_bytes;
1482 } else {
1483 acl_phys.z_acl_size = aclp->z_acl_bytes;
1484 acl_phys.z_acl_count = aclp->z_acl_count;
1485 }
1486 acl_phys.z_acl_version = aclp->z_version;
1487
1488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ZNODE_ACL(zfsvfs), NULL,
1489 &acl_phys, sizeof (acl_phys));
1490 }
1491
1492 /*
1493 * Replace ACL wide bits, but first clear them.
1494 */
1495 zp->z_pflags &= ~ZFS_ACL_WIDE_FLAGS;
1496
1497 zp->z_pflags |= aclp->z_hints;
1498
1499 if (ace_trivial_common(aclp, 0, zfs_ace_walk) == 0)
1500 zp->z_pflags |= ZFS_ACL_TRIVIAL;
1501
1502 zfs_tstamp_update_setup(zp, STATE_CHANGED, NULL, ctime);
1503 return (sa_bulk_update(zp->z_sa_hdl, bulk, count, tx));
1504 }
1505
1506 static void
zfs_acl_chmod(boolean_t isdir,uint64_t mode,boolean_t split,boolean_t trim,zfs_acl_t * aclp)1507 zfs_acl_chmod(boolean_t isdir, uint64_t mode, boolean_t split, boolean_t trim,
1508 zfs_acl_t *aclp)
1509 {
1510 void *acep = NULL;
1511 uint64_t who;
1512 int new_count, new_bytes;
1513 int ace_size;
1514 int entry_type;
1515 uint16_t iflags, type;
1516 uint32_t access_mask;
1517 zfs_acl_node_t *newnode;
1518 size_t abstract_size = aclp->z_ops->ace_abstract_size();
1519 void *zacep;
1520 trivial_acl_t masks;
1521
1522 new_count = new_bytes = 0;
1523
1524 acl_trivial_access_masks((mode_t)mode, isdir, &masks);
1525
1526 newnode = zfs_acl_node_alloc((abstract_size * 6) + aclp->z_acl_bytes);
1527
1528 zacep = newnode->z_acldata;
1529 if (masks.allow0) {
1530 zfs_set_ace(aclp, zacep, masks.allow0, ALLOW, -1, ACE_OWNER);
1531 zacep = (void *)((uintptr_t)zacep + abstract_size);
1532 new_count++;
1533 new_bytes += abstract_size;
1534 }
1535 if (masks.deny1) {
1536 zfs_set_ace(aclp, zacep, masks.deny1, DENY, -1, ACE_OWNER);
1537 zacep = (void *)((uintptr_t)zacep + abstract_size);
1538 new_count++;
1539 new_bytes += abstract_size;
1540 }
1541 if (masks.deny2) {
1542 zfs_set_ace(aclp, zacep, masks.deny2, DENY, -1, OWNING_GROUP);
1543 zacep = (void *)((uintptr_t)zacep + abstract_size);
1544 new_count++;
1545 new_bytes += abstract_size;
1546 }
1547
1548 while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
1549 &iflags, &type))) {
1550 entry_type = (iflags & ACE_TYPE_FLAGS);
1551 /*
1552 * ACEs used to represent the file mode may be divided
1553 * into an equivalent pair of inherit-only and regular
1554 * ACEs, if they are inheritable.
1555 * Skip regular ACEs, which are replaced by the new mode.
1556 */
1557 if (split && (entry_type == ACE_OWNER ||
1558 entry_type == OWNING_GROUP ||
1559 entry_type == ACE_EVERYONE)) {
1560 if (!isdir || !(iflags &
1561 (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1562 continue;
1563 /*
1564 * We preserve owner@, group@, or @everyone
1565 * permissions, if they are inheritable, by
1566 * copying them to inherit_only ACEs. This
1567 * prevents inheritable permissions from being
1568 * altered along with the file mode.
1569 */
1570 iflags |= ACE_INHERIT_ONLY_ACE;
1571 }
1572
1573 /*
1574 * If this ACL has any inheritable ACEs, mark that in
1575 * the hints (which are later masked into the pflags)
1576 * so create knows to do inheritance.
1577 */
1578 if (isdir && (iflags &
1579 (ACE_FILE_INHERIT_ACE|ACE_DIRECTORY_INHERIT_ACE)))
1580 aclp->z_hints |= ZFS_INHERIT_ACE;
1581
1582 if ((type != ALLOW && type != DENY) ||
1583 (iflags & ACE_INHERIT_ONLY_ACE)) {
1584 switch (type) {
1585 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
1586 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
1587 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
1588 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
1589 aclp->z_hints |= ZFS_ACL_OBJ_ACE;
1590 break;
1591 }
1592 } else {
1593 /*
1594 * Limit permissions to be no greater than
1595 * group permissions.
1596 * The "aclinherit" and "aclmode" properties
1597 * affect policy for create and chmod(2),
1598 * respectively.
1599 */
1600 if ((type == ALLOW) && trim)
1601 access_mask &= masks.group;
1602 }
1603 zfs_set_ace(aclp, zacep, access_mask, type, who, iflags);
1604 ace_size = aclp->z_ops->ace_size(acep);
1605 zacep = (void *)((uintptr_t)zacep + ace_size);
1606 new_count++;
1607 new_bytes += ace_size;
1608 }
1609 zfs_set_ace(aclp, zacep, masks.owner, ALLOW, -1, ACE_OWNER);
1610 zacep = (void *)((uintptr_t)zacep + abstract_size);
1611 zfs_set_ace(aclp, zacep, masks.group, ALLOW, -1, OWNING_GROUP);
1612 zacep = (void *)((uintptr_t)zacep + abstract_size);
1613 zfs_set_ace(aclp, zacep, masks.everyone, ALLOW, -1, ACE_EVERYONE);
1614
1615 new_count += 3;
1616 new_bytes += abstract_size * 3;
1617 zfs_acl_release_nodes(aclp);
1618 aclp->z_acl_count = new_count;
1619 aclp->z_acl_bytes = new_bytes;
1620 newnode->z_ace_count = new_count;
1621 newnode->z_size = new_bytes;
1622 list_insert_tail(&aclp->z_acl, newnode);
1623 }
1624
1625 int
zfs_acl_chmod_setattr(znode_t * zp,zfs_acl_t ** aclp,uint64_t mode)1626 zfs_acl_chmod_setattr(znode_t *zp, zfs_acl_t **aclp, uint64_t mode)
1627 {
1628 int error = 0;
1629
1630 mutex_enter(&zp->z_acl_lock);
1631 mutex_enter(&zp->z_lock);
1632 if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_DISCARD)
1633 *aclp = zfs_acl_alloc(zfs_acl_version_zp(zp));
1634 else
1635 error = zfs_acl_node_read(zp, B_TRUE, aclp, B_TRUE);
1636
1637 if (error == 0) {
1638 (*aclp)->z_hints = zp->z_pflags & V4_ACL_WIDE_FLAGS;
1639 zfs_acl_chmod(S_ISDIR(ZTOI(zp)->i_mode), mode, B_TRUE,
1640 (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_GROUPMASK), *aclp);
1641 }
1642 mutex_exit(&zp->z_lock);
1643 mutex_exit(&zp->z_acl_lock);
1644
1645 return (error);
1646 }
1647
1648 /*
1649 * Should ACE be inherited?
1650 */
1651 static int
zfs_ace_can_use(umode_t obj_mode,uint16_t acep_flags)1652 zfs_ace_can_use(umode_t obj_mode, uint16_t acep_flags)
1653 {
1654 int iflags = (acep_flags & 0xf);
1655
1656 if (S_ISDIR(obj_mode) && (iflags & ACE_DIRECTORY_INHERIT_ACE))
1657 return (1);
1658 else if (iflags & ACE_FILE_INHERIT_ACE)
1659 return (!(S_ISDIR(obj_mode) &&
1660 (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)));
1661 return (0);
1662 }
1663
1664 /*
1665 * inherit inheritable ACEs from parent
1666 */
1667 static zfs_acl_t *
zfs_acl_inherit(zfsvfs_t * zfsvfs,umode_t va_mode,zfs_acl_t * paclp,uint64_t mode,boolean_t * need_chmod)1668 zfs_acl_inherit(zfsvfs_t *zfsvfs, umode_t va_mode, zfs_acl_t *paclp,
1669 uint64_t mode, boolean_t *need_chmod)
1670 {
1671 void *pacep = NULL;
1672 void *acep;
1673 zfs_acl_node_t *aclnode;
1674 zfs_acl_t *aclp = NULL;
1675 uint64_t who;
1676 uint32_t access_mask;
1677 uint16_t iflags, newflags, type;
1678 size_t ace_size;
1679 void *data1, *data2;
1680 size_t data1sz, data2sz;
1681 uint_t aclinherit;
1682 boolean_t isdir = S_ISDIR(va_mode);
1683 boolean_t isreg = S_ISREG(va_mode);
1684
1685 *need_chmod = B_TRUE;
1686
1687 aclp = zfs_acl_alloc(paclp->z_version);
1688 aclinherit = zfsvfs->z_acl_inherit;
1689 if (aclinherit == ZFS_ACL_DISCARD || S_ISLNK(va_mode))
1690 return (aclp);
1691
1692 while ((pacep = zfs_acl_next_ace(paclp, pacep, &who,
1693 &access_mask, &iflags, &type))) {
1694
1695 /*
1696 * don't inherit bogus ACEs
1697 */
1698 if (!zfs_acl_valid_ace_type(type, iflags))
1699 continue;
1700
1701 /*
1702 * Check if ACE is inheritable by this vnode
1703 */
1704 if ((aclinherit == ZFS_ACL_NOALLOW && type == ALLOW) ||
1705 !zfs_ace_can_use(va_mode, iflags))
1706 continue;
1707
1708 /*
1709 * If owner@, group@, or everyone@ inheritable
1710 * then zfs_acl_chmod() isn't needed.
1711 */
1712 if ((aclinherit == ZFS_ACL_PASSTHROUGH ||
1713 aclinherit == ZFS_ACL_PASSTHROUGH_X) &&
1714 ((iflags & (ACE_OWNER|ACE_EVERYONE)) ||
1715 ((iflags & OWNING_GROUP) == OWNING_GROUP)) &&
1716 (isreg || (isdir && (iflags & ACE_DIRECTORY_INHERIT_ACE))))
1717 *need_chmod = B_FALSE;
1718
1719 /*
1720 * Strip inherited execute permission from file if
1721 * not in mode
1722 */
1723 if (aclinherit == ZFS_ACL_PASSTHROUGH_X && type == ALLOW &&
1724 !isdir && ((mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)) {
1725 access_mask &= ~ACE_EXECUTE;
1726 }
1727
1728 /*
1729 * Strip write_acl and write_owner from permissions
1730 * when inheriting an ACE
1731 */
1732 if (aclinherit == ZFS_ACL_RESTRICTED && type == ALLOW) {
1733 access_mask &= ~RESTRICTED_CLEAR;
1734 }
1735
1736 ace_size = aclp->z_ops->ace_size(pacep);
1737 aclnode = zfs_acl_node_alloc(ace_size);
1738 list_insert_tail(&aclp->z_acl, aclnode);
1739 acep = aclnode->z_acldata;
1740
1741 zfs_set_ace(aclp, acep, access_mask, type,
1742 who, iflags|ACE_INHERITED_ACE);
1743
1744 /*
1745 * Copy special opaque data if any
1746 */
1747 if ((data1sz = paclp->z_ops->ace_data(pacep, &data1)) != 0) {
1748 VERIFY((data2sz = aclp->z_ops->ace_data(acep,
1749 &data2)) == data1sz);
1750 memcpy(data2, data1, data2sz);
1751 }
1752
1753 aclp->z_acl_count++;
1754 aclnode->z_ace_count++;
1755 aclp->z_acl_bytes += aclnode->z_size;
1756 newflags = aclp->z_ops->ace_flags_get(acep);
1757
1758 /*
1759 * If ACE is not to be inherited further, or if the vnode is
1760 * not a directory, remove all inheritance flags
1761 */
1762 if (!isdir || (iflags & ACE_NO_PROPAGATE_INHERIT_ACE)) {
1763 newflags &= ~ALL_INHERIT;
1764 aclp->z_ops->ace_flags_set(acep,
1765 newflags|ACE_INHERITED_ACE);
1766 continue;
1767 }
1768
1769 /*
1770 * This directory has an inheritable ACE
1771 */
1772 aclp->z_hints |= ZFS_INHERIT_ACE;
1773
1774 /*
1775 * If only FILE_INHERIT is set then turn on
1776 * inherit_only
1777 */
1778 if ((iflags & (ACE_FILE_INHERIT_ACE |
1779 ACE_DIRECTORY_INHERIT_ACE)) == ACE_FILE_INHERIT_ACE) {
1780 newflags |= ACE_INHERIT_ONLY_ACE;
1781 aclp->z_ops->ace_flags_set(acep,
1782 newflags|ACE_INHERITED_ACE);
1783 } else {
1784 newflags &= ~ACE_INHERIT_ONLY_ACE;
1785 aclp->z_ops->ace_flags_set(acep,
1786 newflags|ACE_INHERITED_ACE);
1787 }
1788 }
1789 if (zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED &&
1790 aclp->z_acl_count != 0) {
1791 *need_chmod = B_FALSE;
1792 }
1793
1794 return (aclp);
1795 }
1796
1797 /*
1798 * Create file system object initial permissions
1799 * including inheritable ACEs.
1800 * Also, create FUIDs for owner and group.
1801 */
1802 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)1803 zfs_acl_ids_create(znode_t *dzp, int flag, vattr_t *vap, cred_t *cr,
1804 vsecattr_t *vsecp, zfs_acl_ids_t *acl_ids, zidmap_t *mnt_ns)
1805 {
1806 int error;
1807 zfsvfs_t *zfsvfs = ZTOZSB(dzp);
1808 zfs_acl_t *paclp;
1809 gid_t gid = vap->va_gid;
1810 boolean_t need_chmod = B_TRUE;
1811 boolean_t trim = B_FALSE;
1812 boolean_t inherited = B_FALSE;
1813
1814 memset(acl_ids, 0, sizeof (zfs_acl_ids_t));
1815 acl_ids->z_mode = vap->va_mode;
1816
1817 if (vsecp)
1818 if ((error = zfs_vsec_2_aclp(zfsvfs, vap->va_mode, vsecp,
1819 cr, &acl_ids->z_fuidp, &acl_ids->z_aclp)) != 0)
1820 return (error);
1821
1822 acl_ids->z_fuid = vap->va_uid;
1823 acl_ids->z_fgid = vap->va_gid;
1824 #ifdef HAVE_KSID
1825 /*
1826 * Determine uid and gid.
1827 */
1828 if ((flag & IS_ROOT_NODE) || zfsvfs->z_replay ||
1829 ((flag & IS_XATTR) && (S_ISDIR(vap->va_mode)))) {
1830 acl_ids->z_fuid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_uid,
1831 cr, ZFS_OWNER, &acl_ids->z_fuidp);
1832 acl_ids->z_fgid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid,
1833 cr, ZFS_GROUP, &acl_ids->z_fuidp);
1834 gid = vap->va_gid;
1835 } else {
1836 acl_ids->z_fuid = zfs_fuid_create_cred(zfsvfs, ZFS_OWNER,
1837 cr, &acl_ids->z_fuidp);
1838 acl_ids->z_fgid = 0;
1839 if (vap->va_mask & AT_GID) {
1840 acl_ids->z_fgid = zfs_fuid_create(zfsvfs,
1841 (uint64_t)vap->va_gid,
1842 cr, ZFS_GROUP, &acl_ids->z_fuidp);
1843 gid = vap->va_gid;
1844 if (acl_ids->z_fgid != KGID_TO_SGID(ZTOI(dzp)->i_gid) &&
1845 !groupmember(vap->va_gid, cr) &&
1846 secpolicy_vnode_create_gid(cr) != 0)
1847 acl_ids->z_fgid = 0;
1848 }
1849 if (acl_ids->z_fgid == 0) {
1850 if (dzp->z_mode & S_ISGID) {
1851 char *domain;
1852 uint32_t rid;
1853
1854 acl_ids->z_fgid = KGID_TO_SGID(
1855 ZTOI(dzp)->i_gid);
1856 gid = zfs_fuid_map_id(zfsvfs, acl_ids->z_fgid,
1857 cr, ZFS_GROUP);
1858
1859 if (zfsvfs->z_use_fuids &&
1860 IS_EPHEMERAL(acl_ids->z_fgid)) {
1861 domain = zfs_fuid_idx_domain(
1862 &zfsvfs->z_fuid_idx,
1863 FUID_INDEX(acl_ids->z_fgid));
1864 rid = FUID_RID(acl_ids->z_fgid);
1865 zfs_fuid_node_add(&acl_ids->z_fuidp,
1866 domain, rid,
1867 FUID_INDEX(acl_ids->z_fgid),
1868 acl_ids->z_fgid, ZFS_GROUP);
1869 }
1870 } else {
1871 acl_ids->z_fgid = zfs_fuid_create_cred(zfsvfs,
1872 ZFS_GROUP, cr, &acl_ids->z_fuidp);
1873 gid = crgetgid(cr);
1874 }
1875 }
1876 }
1877 #endif /* HAVE_KSID */
1878
1879 /*
1880 * If we're creating a directory, and the parent directory has the
1881 * set-GID bit set, set in on the new directory.
1882 * Otherwise, if the user is neither privileged nor a member of the
1883 * file's new group, clear the file's set-GID bit.
1884 */
1885
1886 if (!(flag & IS_ROOT_NODE) && (dzp->z_mode & S_ISGID) &&
1887 (S_ISDIR(vap->va_mode))) {
1888 acl_ids->z_mode |= S_ISGID;
1889 } else {
1890 if ((acl_ids->z_mode & S_ISGID) &&
1891 secpolicy_vnode_setids_setgids(cr, gid, mnt_ns,
1892 zfs_i_user_ns(ZTOI(dzp))) != 0) {
1893 acl_ids->z_mode &= ~S_ISGID;
1894 }
1895 }
1896
1897 if (acl_ids->z_aclp == NULL) {
1898 mutex_enter(&dzp->z_acl_lock);
1899 mutex_enter(&dzp->z_lock);
1900 if (!(flag & IS_ROOT_NODE) &&
1901 (dzp->z_pflags & ZFS_INHERIT_ACE) &&
1902 !(dzp->z_pflags & ZFS_XATTR)) {
1903 VERIFY(0 == zfs_acl_node_read(dzp, B_TRUE,
1904 &paclp, B_FALSE));
1905 acl_ids->z_aclp = zfs_acl_inherit(zfsvfs,
1906 vap->va_mode, paclp, acl_ids->z_mode, &need_chmod);
1907 inherited = B_TRUE;
1908 } else {
1909 acl_ids->z_aclp =
1910 zfs_acl_alloc(zfs_acl_version_zp(dzp));
1911 acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1912 }
1913 mutex_exit(&dzp->z_lock);
1914 mutex_exit(&dzp->z_acl_lock);
1915
1916 if (need_chmod) {
1917 if (S_ISDIR(vap->va_mode))
1918 acl_ids->z_aclp->z_hints |=
1919 ZFS_ACL_AUTO_INHERIT;
1920
1921 if (zfsvfs->z_acl_mode == ZFS_ACL_GROUPMASK &&
1922 zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH &&
1923 zfsvfs->z_acl_inherit != ZFS_ACL_PASSTHROUGH_X)
1924 trim = B_TRUE;
1925 zfs_acl_chmod(S_ISDIR(vap->va_mode), acl_ids->z_mode,
1926 B_FALSE, trim, acl_ids->z_aclp);
1927 }
1928 }
1929
1930 if (inherited || vsecp) {
1931 acl_ids->z_mode = zfs_mode_compute(acl_ids->z_mode,
1932 acl_ids->z_aclp, &acl_ids->z_aclp->z_hints,
1933 acl_ids->z_fuid, acl_ids->z_fgid);
1934 if (ace_trivial_common(acl_ids->z_aclp, 0, zfs_ace_walk) == 0)
1935 acl_ids->z_aclp->z_hints |= ZFS_ACL_TRIVIAL;
1936 }
1937
1938 return (0);
1939 }
1940
1941 /*
1942 * Free ACL and fuid_infop, but not the acl_ids structure
1943 */
1944 void
zfs_acl_ids_free(zfs_acl_ids_t * acl_ids)1945 zfs_acl_ids_free(zfs_acl_ids_t *acl_ids)
1946 {
1947 if (acl_ids->z_aclp)
1948 zfs_acl_free(acl_ids->z_aclp);
1949 if (acl_ids->z_fuidp)
1950 zfs_fuid_info_free(acl_ids->z_fuidp);
1951 acl_ids->z_aclp = NULL;
1952 acl_ids->z_fuidp = NULL;
1953 }
1954
1955 boolean_t
zfs_acl_ids_overquota(zfsvfs_t * zv,zfs_acl_ids_t * acl_ids,uint64_t projid)1956 zfs_acl_ids_overquota(zfsvfs_t *zv, zfs_acl_ids_t *acl_ids, uint64_t projid)
1957 {
1958 return (zfs_id_overquota(zv, DMU_USERUSED_OBJECT, acl_ids->z_fuid) ||
1959 zfs_id_overquota(zv, DMU_GROUPUSED_OBJECT, acl_ids->z_fgid) ||
1960 (projid != ZFS_DEFAULT_PROJID && projid != ZFS_INVALID_PROJID &&
1961 zfs_id_overquota(zv, DMU_PROJECTUSED_OBJECT, projid)));
1962 }
1963
1964 /*
1965 * Retrieve a file's ACL
1966 */
1967 int
zfs_getacl(znode_t * zp,vsecattr_t * vsecp,boolean_t skipaclchk,cred_t * cr)1968 zfs_getacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
1969 {
1970 zfs_acl_t *aclp;
1971 ulong_t mask;
1972 int error;
1973 int count = 0;
1974 int largeace = 0;
1975
1976 mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT |
1977 VSA_ACE_ACLFLAGS | VSA_ACE_ALLTYPES);
1978
1979 if (mask == 0)
1980 return (SET_ERROR(ENOSYS));
1981
1982 if ((error = zfs_zaccess(zp, ACE_READ_ACL, 0, skipaclchk, cr,
1983 zfs_init_idmap)))
1984 return (error);
1985
1986 mutex_enter(&zp->z_acl_lock);
1987
1988 error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
1989 if (error != 0) {
1990 mutex_exit(&zp->z_acl_lock);
1991 return (error);
1992 }
1993
1994 /*
1995 * Scan ACL to determine number of ACEs
1996 */
1997 if ((zp->z_pflags & ZFS_ACL_OBJ_ACE) && !(mask & VSA_ACE_ALLTYPES)) {
1998 void *zacep = NULL;
1999 uint64_t who;
2000 uint32_t access_mask;
2001 uint16_t type, iflags;
2002
2003 while ((zacep = zfs_acl_next_ace(aclp, zacep,
2004 &who, &access_mask, &iflags, &type))) {
2005 switch (type) {
2006 case ACE_ACCESS_ALLOWED_OBJECT_ACE_TYPE:
2007 case ACE_ACCESS_DENIED_OBJECT_ACE_TYPE:
2008 case ACE_SYSTEM_AUDIT_OBJECT_ACE_TYPE:
2009 case ACE_SYSTEM_ALARM_OBJECT_ACE_TYPE:
2010 largeace++;
2011 continue;
2012 default:
2013 count++;
2014 }
2015 }
2016 vsecp->vsa_aclcnt = count;
2017 } else
2018 count = (int)aclp->z_acl_count;
2019
2020 if (mask & VSA_ACECNT) {
2021 vsecp->vsa_aclcnt = count;
2022 }
2023
2024 if (mask & VSA_ACE) {
2025 size_t aclsz;
2026
2027 aclsz = count * sizeof (ace_t) +
2028 sizeof (ace_object_t) * largeace;
2029
2030 vsecp->vsa_aclentp = kmem_alloc(aclsz, KM_SLEEP);
2031 vsecp->vsa_aclentsz = aclsz;
2032
2033 if (aclp->z_version == ZFS_ACL_VERSION_FUID)
2034 zfs_copy_fuid_2_ace(ZTOZSB(zp), aclp, cr,
2035 vsecp->vsa_aclentp, !(mask & VSA_ACE_ALLTYPES));
2036 else {
2037 zfs_acl_node_t *aclnode;
2038 void *start = vsecp->vsa_aclentp;
2039
2040 for (aclnode = list_head(&aclp->z_acl); aclnode;
2041 aclnode = list_next(&aclp->z_acl, aclnode)) {
2042 memcpy(start, aclnode->z_acldata,
2043 aclnode->z_size);
2044 start = (caddr_t)start + aclnode->z_size;
2045 }
2046 ASSERT((caddr_t)start - (caddr_t)vsecp->vsa_aclentp ==
2047 aclp->z_acl_bytes);
2048 }
2049 }
2050 if (mask & VSA_ACE_ACLFLAGS) {
2051 vsecp->vsa_aclflags = 0;
2052 if (zp->z_pflags & ZFS_ACL_DEFAULTED)
2053 vsecp->vsa_aclflags |= ACL_DEFAULTED;
2054 if (zp->z_pflags & ZFS_ACL_PROTECTED)
2055 vsecp->vsa_aclflags |= ACL_PROTECTED;
2056 if (zp->z_pflags & ZFS_ACL_AUTO_INHERIT)
2057 vsecp->vsa_aclflags |= ACL_AUTO_INHERIT;
2058 }
2059
2060 mutex_exit(&zp->z_acl_lock);
2061
2062 return (0);
2063 }
2064
2065 int
zfs_vsec_2_aclp(zfsvfs_t * zfsvfs,umode_t obj_mode,vsecattr_t * vsecp,cred_t * cr,zfs_fuid_info_t ** fuidp,zfs_acl_t ** zaclp)2066 zfs_vsec_2_aclp(zfsvfs_t *zfsvfs, umode_t obj_mode,
2067 vsecattr_t *vsecp, cred_t *cr, zfs_fuid_info_t **fuidp, zfs_acl_t **zaclp)
2068 {
2069 zfs_acl_t *aclp;
2070 zfs_acl_node_t *aclnode;
2071 int aclcnt = vsecp->vsa_aclcnt;
2072 int error;
2073
2074 if (vsecp->vsa_aclcnt > MAX_ACL_ENTRIES || vsecp->vsa_aclcnt <= 0)
2075 return (SET_ERROR(EINVAL));
2076
2077 aclp = zfs_acl_alloc(zfs_acl_version(zfsvfs->z_version));
2078
2079 aclp->z_hints = 0;
2080 aclnode = zfs_acl_node_alloc(aclcnt * sizeof (zfs_object_ace_t));
2081 if (aclp->z_version == ZFS_ACL_VERSION_INITIAL) {
2082 if ((error = zfs_copy_ace_2_oldace(obj_mode, aclp,
2083 (ace_t *)vsecp->vsa_aclentp, aclnode->z_acldata,
2084 aclcnt, &aclnode->z_size)) != 0) {
2085 zfs_acl_free(aclp);
2086 zfs_acl_node_free(aclnode);
2087 return (error);
2088 }
2089 } else {
2090 if ((error = zfs_copy_ace_2_fuid(zfsvfs, obj_mode, aclp,
2091 vsecp->vsa_aclentp, aclnode->z_acldata, aclcnt,
2092 &aclnode->z_size, fuidp, cr)) != 0) {
2093 zfs_acl_free(aclp);
2094 zfs_acl_node_free(aclnode);
2095 return (error);
2096 }
2097 }
2098 aclp->z_acl_bytes = aclnode->z_size;
2099 aclnode->z_ace_count = aclcnt;
2100 aclp->z_acl_count = aclcnt;
2101 list_insert_head(&aclp->z_acl, aclnode);
2102
2103 /*
2104 * If flags are being set then add them to z_hints
2105 */
2106 if (vsecp->vsa_mask & VSA_ACE_ACLFLAGS) {
2107 if (vsecp->vsa_aclflags & ACL_PROTECTED)
2108 aclp->z_hints |= ZFS_ACL_PROTECTED;
2109 if (vsecp->vsa_aclflags & ACL_DEFAULTED)
2110 aclp->z_hints |= ZFS_ACL_DEFAULTED;
2111 if (vsecp->vsa_aclflags & ACL_AUTO_INHERIT)
2112 aclp->z_hints |= ZFS_ACL_AUTO_INHERIT;
2113 }
2114
2115 *zaclp = aclp;
2116
2117 return (0);
2118 }
2119
2120 /*
2121 * Set a file's ACL
2122 */
2123 int
zfs_setacl(znode_t * zp,vsecattr_t * vsecp,boolean_t skipaclchk,cred_t * cr)2124 zfs_setacl(znode_t *zp, vsecattr_t *vsecp, boolean_t skipaclchk, cred_t *cr)
2125 {
2126 zfsvfs_t *zfsvfs = ZTOZSB(zp);
2127 zilog_t *zilog = zfsvfs->z_log;
2128 ulong_t mask = vsecp->vsa_mask & (VSA_ACE | VSA_ACECNT);
2129 dmu_tx_t *tx;
2130 int error;
2131 zfs_acl_t *aclp;
2132 zfs_fuid_info_t *fuidp = NULL;
2133 boolean_t fuid_dirtied;
2134 uint64_t acl_obj;
2135
2136 if (mask == 0)
2137 return (SET_ERROR(ENOSYS));
2138
2139 if (zp->z_pflags & ZFS_IMMUTABLE)
2140 return (SET_ERROR(EPERM));
2141
2142 if ((error = zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr,
2143 zfs_init_idmap)))
2144 return (error);
2145
2146 error = zfs_vsec_2_aclp(zfsvfs, ZTOI(zp)->i_mode, vsecp, cr, &fuidp,
2147 &aclp);
2148 if (error)
2149 return (error);
2150
2151 /*
2152 * If ACL wide flags aren't being set then preserve any
2153 * existing flags.
2154 */
2155 if (!(vsecp->vsa_mask & VSA_ACE_ACLFLAGS)) {
2156 aclp->z_hints |=
2157 (zp->z_pflags & V4_ACL_WIDE_FLAGS);
2158 }
2159 top:
2160 mutex_enter(&zp->z_acl_lock);
2161 mutex_enter(&zp->z_lock);
2162
2163 tx = dmu_tx_create(zfsvfs->z_os);
2164
2165 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
2166
2167 fuid_dirtied = zfsvfs->z_fuid_dirty;
2168 if (fuid_dirtied)
2169 zfs_fuid_txhold(zfsvfs, tx);
2170
2171 /*
2172 * If old version and ACL won't fit in bonus and we aren't
2173 * upgrading then take out necessary DMU holds
2174 */
2175
2176 if ((acl_obj = zfs_external_acl(zp)) != 0) {
2177 if (zfsvfs->z_version >= ZPL_VERSION_FUID &&
2178 zfs_znode_acl_version(zp) <= ZFS_ACL_VERSION_INITIAL) {
2179 dmu_tx_hold_free(tx, acl_obj, 0,
2180 DMU_OBJECT_END);
2181 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0,
2182 aclp->z_acl_bytes);
2183 } else {
2184 dmu_tx_hold_write(tx, acl_obj, 0, aclp->z_acl_bytes);
2185 }
2186 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) {
2187 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, aclp->z_acl_bytes);
2188 }
2189
2190 zfs_sa_upgrade_txholds(tx, zp);
2191 error = dmu_tx_assign(tx, DMU_TX_NOWAIT);
2192 if (error) {
2193 mutex_exit(&zp->z_acl_lock);
2194 mutex_exit(&zp->z_lock);
2195
2196 if (error == ERESTART) {
2197 dmu_tx_wait(tx);
2198 dmu_tx_abort(tx);
2199 goto top;
2200 }
2201 dmu_tx_abort(tx);
2202 zfs_acl_free(aclp);
2203 return (error);
2204 }
2205
2206 error = zfs_aclset_common(zp, aclp, cr, tx);
2207 ASSERT(error == 0);
2208 ASSERT(zp->z_acl_cached == NULL);
2209 zp->z_acl_cached = aclp;
2210
2211 if (fuid_dirtied)
2212 zfs_fuid_sync(zfsvfs, tx);
2213
2214 zfs_log_acl(zilog, tx, zp, vsecp, fuidp);
2215
2216 if (fuidp)
2217 zfs_fuid_info_free(fuidp);
2218 dmu_tx_commit(tx);
2219
2220 mutex_exit(&zp->z_lock);
2221 mutex_exit(&zp->z_acl_lock);
2222
2223 return (error);
2224 }
2225
2226 /*
2227 * Check accesses of interest (AoI) against attributes of the dataset
2228 * such as read-only. Returns zero if no AoI conflict with dataset
2229 * attributes, otherwise an appropriate errno is returned.
2230 */
2231 static int
zfs_zaccess_dataset_check(znode_t * zp,uint32_t v4_mode)2232 zfs_zaccess_dataset_check(znode_t *zp, uint32_t v4_mode)
2233 {
2234 if ((v4_mode & WRITE_MASK) && (zfs_is_readonly(ZTOZSB(zp))) &&
2235 (!Z_ISDEV(ZTOI(zp)->i_mode) || (v4_mode & WRITE_MASK_ATTRS))) {
2236 return (SET_ERROR(EROFS));
2237 }
2238
2239 /*
2240 * Intentionally allow ZFS_READONLY through here.
2241 * See zfs_zaccess_common().
2242 */
2243 if ((v4_mode & WRITE_MASK_DATA) &&
2244 (zp->z_pflags & ZFS_IMMUTABLE)) {
2245 return (SET_ERROR(EPERM));
2246 }
2247
2248 if ((v4_mode & (ACE_DELETE | ACE_DELETE_CHILD)) &&
2249 (zp->z_pflags & ZFS_NOUNLINK)) {
2250 return (SET_ERROR(EPERM));
2251 }
2252
2253 if (((v4_mode & (ACE_READ_DATA|ACE_EXECUTE)) &&
2254 (zp->z_pflags & ZFS_AV_QUARANTINED))) {
2255 return (SET_ERROR(EACCES));
2256 }
2257
2258 return (0);
2259 }
2260
2261 /*
2262 * The primary usage of this function is to loop through all of the
2263 * ACEs in the znode, determining what accesses of interest (AoI) to
2264 * the caller are allowed or denied. The AoI are expressed as bits in
2265 * the working_mode parameter. As each ACE is processed, bits covered
2266 * by that ACE are removed from the working_mode. This removal
2267 * facilitates two things. The first is that when the working mode is
2268 * empty (= 0), we know we've looked at all the AoI. The second is
2269 * that the ACE interpretation rules don't allow a later ACE to undo
2270 * something granted or denied by an earlier ACE. Removing the
2271 * discovered access or denial enforces this rule. At the end of
2272 * processing the ACEs, all AoI that were found to be denied are
2273 * placed into the working_mode, giving the caller a mask of denied
2274 * accesses. Returns:
2275 * 0 if all AoI granted
2276 * EACCES if the denied mask is non-zero
2277 * other error if abnormal failure (e.g., IO error)
2278 *
2279 * A secondary usage of the function is to determine if any of the
2280 * AoI are granted. If an ACE grants any access in
2281 * the working_mode, we immediately short circuit out of the function.
2282 * This mode is chosen by setting anyaccess to B_TRUE. The
2283 * working_mode is not a denied access mask upon exit if the function
2284 * is used in this manner.
2285 */
2286 static int
zfs_zaccess_aces_check(znode_t * zp,uint32_t * working_mode,boolean_t anyaccess,cred_t * cr,zidmap_t * mnt_ns)2287 zfs_zaccess_aces_check(znode_t *zp, uint32_t *working_mode,
2288 boolean_t anyaccess, cred_t *cr, zidmap_t *mnt_ns)
2289 {
2290 zfsvfs_t *zfsvfs = ZTOZSB(zp);
2291 zfs_acl_t *aclp;
2292 int error;
2293 uid_t uid = crgetuid(cr);
2294 uint64_t who;
2295 uint16_t type, iflags;
2296 uint16_t entry_type;
2297 uint32_t access_mask;
2298 uint32_t deny_mask = 0;
2299 zfs_ace_hdr_t *acep = NULL;
2300 boolean_t checkit;
2301 uid_t gowner;
2302 uid_t fowner;
2303
2304 if (mnt_ns) {
2305 fowner = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2306 KUID_TO_SUID(ZTOI(zp)->i_uid));
2307 gowner = zfs_gid_to_vfsgid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2308 KGID_TO_SGID(ZTOI(zp)->i_gid));
2309 } else
2310 zfs_fuid_map_ids(zp, cr, &fowner, &gowner);
2311
2312 mutex_enter(&zp->z_acl_lock);
2313
2314 error = zfs_acl_node_read(zp, B_FALSE, &aclp, B_FALSE);
2315 if (error != 0) {
2316 mutex_exit(&zp->z_acl_lock);
2317 return (error);
2318 }
2319
2320 ASSERT(zp->z_acl_cached);
2321
2322 while ((acep = zfs_acl_next_ace(aclp, acep, &who, &access_mask,
2323 &iflags, &type))) {
2324 uint32_t mask_matched;
2325
2326 if (!zfs_acl_valid_ace_type(type, iflags))
2327 continue;
2328
2329 if (S_ISDIR(ZTOI(zp)->i_mode) &&
2330 (iflags & ACE_INHERIT_ONLY_ACE))
2331 continue;
2332
2333 /* Skip ACE if it does not affect any AoI */
2334 mask_matched = (access_mask & *working_mode);
2335 if (!mask_matched)
2336 continue;
2337
2338 entry_type = (iflags & ACE_TYPE_FLAGS);
2339
2340 checkit = B_FALSE;
2341
2342 switch (entry_type) {
2343 case ACE_OWNER:
2344 if (uid == fowner)
2345 checkit = B_TRUE;
2346 break;
2347 case OWNING_GROUP:
2348 who = gowner;
2349 zfs_fallthrough;
2350 case ACE_IDENTIFIER_GROUP:
2351 checkit = zfs_groupmember(zfsvfs, who, cr);
2352 break;
2353 case ACE_EVERYONE:
2354 checkit = B_TRUE;
2355 break;
2356
2357 /* USER Entry */
2358 default:
2359 if (entry_type == 0) {
2360 uid_t newid;
2361
2362 newid = zfs_fuid_map_id(zfsvfs, who, cr,
2363 ZFS_ACE_USER);
2364 if (newid != IDMAP_WK_CREATOR_OWNER_UID &&
2365 uid == newid)
2366 checkit = B_TRUE;
2367 break;
2368 } else {
2369 mutex_exit(&zp->z_acl_lock);
2370 return (SET_ERROR(EIO));
2371 }
2372 }
2373
2374 if (checkit) {
2375 if (type == DENY) {
2376 DTRACE_PROBE3(zfs__ace__denies,
2377 znode_t *, zp,
2378 zfs_ace_hdr_t *, acep,
2379 uint32_t, mask_matched);
2380 deny_mask |= mask_matched;
2381 } else {
2382 DTRACE_PROBE3(zfs__ace__allows,
2383 znode_t *, zp,
2384 zfs_ace_hdr_t *, acep,
2385 uint32_t, mask_matched);
2386 if (anyaccess) {
2387 mutex_exit(&zp->z_acl_lock);
2388 return (0);
2389 }
2390 }
2391 *working_mode &= ~mask_matched;
2392 }
2393
2394 /* Are we done? */
2395 if (*working_mode == 0)
2396 break;
2397 }
2398
2399 mutex_exit(&zp->z_acl_lock);
2400
2401 /* Put the found 'denies' back on the working mode */
2402 if (deny_mask) {
2403 *working_mode |= deny_mask;
2404 return (SET_ERROR(EACCES));
2405 } else if (*working_mode) {
2406 return (-1);
2407 }
2408
2409 return (0);
2410 }
2411
2412 /*
2413 * Return true if any access whatsoever granted, we don't actually
2414 * care what access is granted.
2415 */
2416 boolean_t
zfs_has_access(znode_t * zp,cred_t * cr)2417 zfs_has_access(znode_t *zp, cred_t *cr)
2418 {
2419 uint32_t have = ACE_ALL_PERMS;
2420
2421 if (zfs_zaccess_aces_check(zp, &have, B_TRUE, cr,
2422 zfs_init_idmap) != 0) {
2423 uid_t owner;
2424
2425 owner = zfs_fuid_map_id(ZTOZSB(zp),
2426 KUID_TO_SUID(ZTOI(zp)->i_uid), cr, ZFS_OWNER);
2427 return (secpolicy_vnode_any_access(cr, ZTOI(zp), owner) == 0);
2428 }
2429 return (B_TRUE);
2430 }
2431
2432 /*
2433 * Simplified access check for case where ACL is known to not contain
2434 * information beyond what is defined in the mode. In this case, we
2435 * can pass along to the kernel / vfs generic_permission() check, which
2436 * evaluates the mode and POSIX ACL.
2437 *
2438 * NFSv4 ACLs allow granting permissions that are usually relegated only
2439 * to the file owner or superuser. Examples are ACE_WRITE_OWNER (chown),
2440 * ACE_WRITE_ACL(chmod), and ACE_DELETE. ACE_DELETE requests must fail
2441 * because with conventional posix permissions, right to delete file
2442 * is determined by write bit on the parent dir.
2443 *
2444 * If unmappable perms are requested, then we must return EPERM
2445 * and include those bits in the working_mode so that the caller of
2446 * zfs_zaccess_common() can decide whether to perform additional
2447 * policy / capability checks. EACCES is used in zfs_zaccess_aces_check()
2448 * to indicate access check failed due to explicit DENY entry, and so
2449 * we want to avoid that here.
2450 */
2451 static int
zfs_zaccess_trivial(znode_t * zp,uint32_t * working_mode,cred_t * cr,zidmap_t * mnt_ns)2452 zfs_zaccess_trivial(znode_t *zp, uint32_t *working_mode, cred_t *cr,
2453 zidmap_t *mnt_ns)
2454 {
2455 int err, mask;
2456 int unmapped = 0;
2457
2458 ASSERT(zp->z_pflags & ZFS_ACL_TRIVIAL);
2459
2460 mask = zfs_v4_to_unix(*working_mode, &unmapped);
2461 if (mask == 0 || unmapped) {
2462 *working_mode = unmapped;
2463 return (unmapped ? SET_ERROR(EPERM) : 0);
2464 }
2465
2466 #if (defined(HAVE_IOPS_PERMISSION_USERNS) || \
2467 defined(HAVE_IOPS_PERMISSION_IDMAP))
2468 err = generic_permission(mnt_ns, ZTOI(zp), mask);
2469 #else
2470 err = generic_permission(ZTOI(zp), mask);
2471 #endif
2472 if (err != 0) {
2473 return (SET_ERROR(EPERM));
2474 }
2475
2476 *working_mode = unmapped;
2477
2478 return (0);
2479 }
2480
2481 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,zidmap_t * mnt_ns)2482 zfs_zaccess_common(znode_t *zp, uint32_t v4_mode, uint32_t *working_mode,
2483 boolean_t *check_privs, boolean_t skipaclchk, cred_t *cr, zidmap_t *mnt_ns)
2484 {
2485 zfsvfs_t *zfsvfs = ZTOZSB(zp);
2486 int err;
2487
2488 *working_mode = v4_mode;
2489 *check_privs = B_TRUE;
2490
2491 /*
2492 * Short circuit empty requests
2493 */
2494 if (v4_mode == 0 || zfsvfs->z_replay) {
2495 *working_mode = 0;
2496 return (0);
2497 }
2498
2499 if ((err = zfs_zaccess_dataset_check(zp, v4_mode)) != 0) {
2500 *check_privs = B_FALSE;
2501 return (err);
2502 }
2503
2504 /*
2505 * The caller requested that the ACL check be skipped. This
2506 * would only happen if the caller checked VOP_ACCESS() with a
2507 * 32 bit ACE mask and already had the appropriate permissions.
2508 */
2509 if (skipaclchk) {
2510 *working_mode = 0;
2511 return (0);
2512 }
2513
2514 /*
2515 * Note: ZFS_READONLY represents the "DOS R/O" attribute.
2516 * When that flag is set, we should behave as if write access
2517 * were not granted by anything in the ACL. In particular:
2518 * We _must_ allow writes after opening the file r/w, then
2519 * setting the DOS R/O attribute, and writing some more.
2520 * (Similar to how you can write after fchmod(fd, 0444).)
2521 *
2522 * Therefore ZFS_READONLY is ignored in the dataset check
2523 * above, and checked here as if part of the ACL check.
2524 * Also note: DOS R/O is ignored for directories.
2525 */
2526 if ((v4_mode & WRITE_MASK_DATA) &&
2527 S_ISDIR(ZTOI(zp)->i_mode) &&
2528 (zp->z_pflags & ZFS_READONLY)) {
2529 return (SET_ERROR(EPERM));
2530 }
2531
2532 if (zp->z_pflags & ZFS_ACL_TRIVIAL)
2533 return (zfs_zaccess_trivial(zp, working_mode, cr, mnt_ns));
2534
2535 return (zfs_zaccess_aces_check(zp, working_mode, B_FALSE, cr, mnt_ns));
2536 }
2537
2538 static int
zfs_zaccess_append(znode_t * zp,uint32_t * working_mode,boolean_t * check_privs,cred_t * cr,zidmap_t * mnt_ns)2539 zfs_zaccess_append(znode_t *zp, uint32_t *working_mode, boolean_t *check_privs,
2540 cred_t *cr, zidmap_t *mnt_ns)
2541 {
2542 if (*working_mode != ACE_WRITE_DATA)
2543 return (SET_ERROR(EACCES));
2544
2545 return (zfs_zaccess_common(zp, ACE_APPEND_DATA, working_mode,
2546 check_privs, B_FALSE, cr, mnt_ns));
2547 }
2548
2549 int
zfs_fastaccesschk_execute(znode_t * zdp,cred_t * cr)2550 zfs_fastaccesschk_execute(znode_t *zdp, cred_t *cr)
2551 {
2552 boolean_t owner = B_FALSE;
2553 boolean_t groupmbr = B_FALSE;
2554 boolean_t is_attr;
2555 uid_t uid = crgetuid(cr);
2556 int error;
2557
2558 if (zdp->z_pflags & ZFS_AV_QUARANTINED)
2559 return (SET_ERROR(EACCES));
2560
2561 is_attr = ((zdp->z_pflags & ZFS_XATTR) &&
2562 (S_ISDIR(ZTOI(zdp)->i_mode)));
2563 if (is_attr)
2564 goto slow;
2565
2566
2567 mutex_enter(&zdp->z_acl_lock);
2568
2569 if (zdp->z_pflags & ZFS_NO_EXECS_DENIED) {
2570 mutex_exit(&zdp->z_acl_lock);
2571 return (0);
2572 }
2573
2574 if (KUID_TO_SUID(ZTOI(zdp)->i_uid) != 0 ||
2575 KGID_TO_SGID(ZTOI(zdp)->i_gid) != 0) {
2576 mutex_exit(&zdp->z_acl_lock);
2577 goto slow;
2578 }
2579
2580 if (uid == KUID_TO_SUID(ZTOI(zdp)->i_uid)) {
2581 if (zdp->z_mode & S_IXUSR) {
2582 mutex_exit(&zdp->z_acl_lock);
2583 return (0);
2584 } else {
2585 mutex_exit(&zdp->z_acl_lock);
2586 goto slow;
2587 }
2588 }
2589 if (groupmember(KGID_TO_SGID(ZTOI(zdp)->i_gid), cr)) {
2590 if (zdp->z_mode & S_IXGRP) {
2591 mutex_exit(&zdp->z_acl_lock);
2592 return (0);
2593 } else {
2594 mutex_exit(&zdp->z_acl_lock);
2595 goto slow;
2596 }
2597 }
2598 if (!owner && !groupmbr) {
2599 if (zdp->z_mode & S_IXOTH) {
2600 mutex_exit(&zdp->z_acl_lock);
2601 return (0);
2602 }
2603 }
2604
2605 mutex_exit(&zdp->z_acl_lock);
2606
2607 slow:
2608 DTRACE_PROBE(zfs__fastpath__execute__access__miss);
2609 if ((error = zfs_enter(ZTOZSB(zdp), FTAG)) != 0)
2610 return (error);
2611 error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr,
2612 zfs_init_idmap);
2613 zfs_exit(ZTOZSB(zdp), FTAG);
2614 return (error);
2615 }
2616
2617 /*
2618 * Determine whether Access should be granted/denied.
2619 *
2620 * The least priv subsystem is always consulted as a basic privilege
2621 * can define any form of access.
2622 */
2623 int
zfs_zaccess(znode_t * zp,int mode,int flags,boolean_t skipaclchk,cred_t * cr,zidmap_t * mnt_ns)2624 zfs_zaccess(znode_t *zp, int mode, int flags, boolean_t skipaclchk, cred_t *cr,
2625 zidmap_t *mnt_ns)
2626 {
2627 uint32_t working_mode;
2628 int error;
2629 int is_attr;
2630 boolean_t check_privs;
2631 znode_t *xzp;
2632 znode_t *check_zp = zp;
2633 mode_t needed_bits;
2634 uid_t owner;
2635
2636 is_attr = ((zp->z_pflags & ZFS_XATTR) && S_ISDIR(ZTOI(zp)->i_mode));
2637
2638 /*
2639 * If attribute then validate against base file
2640 */
2641 if (is_attr) {
2642 if ((error = zfs_zget(ZTOZSB(zp),
2643 zp->z_xattr_parent, &xzp)) != 0) {
2644 return (error);
2645 }
2646
2647 check_zp = xzp;
2648
2649 /*
2650 * fixup mode to map to xattr perms
2651 */
2652
2653 if (mode & (ACE_WRITE_DATA|ACE_APPEND_DATA)) {
2654 mode &= ~(ACE_WRITE_DATA|ACE_APPEND_DATA);
2655 mode |= ACE_WRITE_NAMED_ATTRS;
2656 }
2657
2658 if (mode & (ACE_READ_DATA|ACE_EXECUTE)) {
2659 mode &= ~(ACE_READ_DATA|ACE_EXECUTE);
2660 mode |= ACE_READ_NAMED_ATTRS;
2661 }
2662 }
2663
2664 owner = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ZTOI(zp)),
2665 KUID_TO_SUID(ZTOI(zp)->i_uid));
2666 owner = zfs_fuid_map_id(ZTOZSB(zp), owner, cr, ZFS_OWNER);
2667
2668 /*
2669 * Map the bits required to the standard inode flags
2670 * S_IRUSR|S_IWUSR|S_IXUSR in the needed_bits. Map the bits
2671 * mapped by working_mode (currently missing) in missing_bits.
2672 * Call secpolicy_vnode_access2() with (needed_bits & ~checkmode),
2673 * needed_bits.
2674 */
2675 needed_bits = 0;
2676
2677 working_mode = mode;
2678 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES)) &&
2679 owner == crgetuid(cr))
2680 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2681
2682 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2683 ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2684 needed_bits |= S_IRUSR;
2685 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2686 ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2687 needed_bits |= S_IWUSR;
2688 if (working_mode & ACE_EXECUTE)
2689 needed_bits |= S_IXUSR;
2690
2691 if ((error = zfs_zaccess_common(check_zp, mode, &working_mode,
2692 &check_privs, skipaclchk, cr, mnt_ns)) == 0) {
2693 if (is_attr)
2694 zrele(xzp);
2695 return (secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2696 needed_bits, needed_bits));
2697 }
2698
2699 if (error && !check_privs) {
2700 if (is_attr)
2701 zrele(xzp);
2702 return (error);
2703 }
2704
2705 if (error && (flags & V_APPEND)) {
2706 error = zfs_zaccess_append(zp, &working_mode, &check_privs, cr,
2707 mnt_ns);
2708 }
2709
2710 if (error && check_privs) {
2711 mode_t checkmode = 0;
2712
2713 /*
2714 * First check for implicit owner permission on
2715 * read_acl/read_attributes
2716 */
2717
2718 ASSERT(working_mode != 0);
2719
2720 if ((working_mode & (ACE_READ_ACL|ACE_READ_ATTRIBUTES) &&
2721 owner == crgetuid(cr)))
2722 working_mode &= ~(ACE_READ_ACL|ACE_READ_ATTRIBUTES);
2723
2724 if (working_mode & (ACE_READ_DATA|ACE_READ_NAMED_ATTRS|
2725 ACE_READ_ACL|ACE_READ_ATTRIBUTES|ACE_SYNCHRONIZE))
2726 checkmode |= S_IRUSR;
2727 if (working_mode & (ACE_WRITE_DATA|ACE_WRITE_NAMED_ATTRS|
2728 ACE_APPEND_DATA|ACE_WRITE_ATTRIBUTES|ACE_SYNCHRONIZE))
2729 checkmode |= S_IWUSR;
2730 if (working_mode & ACE_EXECUTE)
2731 checkmode |= S_IXUSR;
2732
2733 error = secpolicy_vnode_access2(cr, ZTOI(check_zp), owner,
2734 needed_bits & ~checkmode, needed_bits);
2735
2736 if (error == 0 && (working_mode & ACE_WRITE_OWNER))
2737 error = secpolicy_vnode_chown(cr, owner);
2738 if (error == 0 && (working_mode & ACE_WRITE_ACL))
2739 error = secpolicy_vnode_setdac(cr, owner);
2740
2741 if (error == 0 && (working_mode &
2742 (ACE_DELETE|ACE_DELETE_CHILD)))
2743 error = secpolicy_vnode_remove(cr);
2744
2745 if (error == 0 && (working_mode & ACE_SYNCHRONIZE)) {
2746 error = secpolicy_vnode_chown(cr, owner);
2747 }
2748 if (error == 0) {
2749 /*
2750 * See if any bits other than those already checked
2751 * for are still present. If so then return EACCES
2752 */
2753 if (working_mode & ~(ZFS_CHECKED_MASKS)) {
2754 error = SET_ERROR(EACCES);
2755 }
2756 }
2757 } else if (error == 0) {
2758 error = secpolicy_vnode_access2(cr, ZTOI(zp), owner,
2759 needed_bits, needed_bits);
2760 }
2761
2762 if (is_attr)
2763 zrele(xzp);
2764
2765 return (error);
2766 }
2767
2768 /*
2769 * Translate traditional unix S_IRUSR/S_IWUSR/S_IXUSR mode into
2770 * NFSv4-style ZFS ACL format and call zfs_zaccess()
2771 */
2772 int
zfs_zaccess_rwx(znode_t * zp,mode_t mode,int flags,cred_t * cr,zidmap_t * mnt_ns)2773 zfs_zaccess_rwx(znode_t *zp, mode_t mode, int flags, cred_t *cr,
2774 zidmap_t *mnt_ns)
2775 {
2776 return (zfs_zaccess(zp, zfs_unix_to_v4(mode >> 6), flags, B_FALSE, cr,
2777 mnt_ns));
2778 }
2779
2780 /*
2781 * Access function for secpolicy_vnode_setattr
2782 */
2783 int
zfs_zaccess_unix(void * zp,int mode,cred_t * cr)2784 zfs_zaccess_unix(void *zp, int mode, cred_t *cr)
2785 {
2786 int v4_mode = zfs_unix_to_v4(mode >> 6);
2787
2788 return (zfs_zaccess(zp, v4_mode, 0, B_FALSE, cr, zfs_init_idmap));
2789 }
2790
2791 /* See zfs_zaccess_delete() */
2792 static const boolean_t zfs_write_implies_delete_child = B_TRUE;
2793
2794 /*
2795 * Determine whether delete access should be granted.
2796 *
2797 * The following chart outlines how we handle delete permissions which is
2798 * how recent versions of windows (Windows 2008) handles it. The efficiency
2799 * comes from not having to check the parent ACL where the object itself grants
2800 * delete:
2801 *
2802 * -------------------------------------------------------
2803 * | Parent Dir | Target Object Permissions |
2804 * | permissions | |
2805 * -------------------------------------------------------
2806 * | | ACL Allows | ACL Denies| Delete |
2807 * | | Delete | Delete | unspecified|
2808 * -------------------------------------------------------
2809 * | ACL Allows | Permit | Deny * | Permit |
2810 * | DELETE_CHILD | | | |
2811 * -------------------------------------------------------
2812 * | ACL Denies | Permit | Deny | Deny |
2813 * | DELETE_CHILD | | | |
2814 * -------------------------------------------------------
2815 * | ACL specifies | | | |
2816 * | only allow | Permit | Deny * | Permit |
2817 * | write and | | | |
2818 * | execute | | | |
2819 * -------------------------------------------------------
2820 * | ACL denies | | | |
2821 * | write and | Permit | Deny | Deny |
2822 * | execute | | | |
2823 * -------------------------------------------------------
2824 * ^
2825 * |
2826 * Re. execute permission on the directory: if that's missing,
2827 * the vnode lookup of the target will fail before we get here.
2828 *
2829 * Re [*] in the table above: NFSv4 would normally Permit delete for
2830 * these two cells of the matrix.
2831 * See acl.h for notes on which ACE_... flags should be checked for which
2832 * operations. Specifically, the NFSv4 committee recommendation is in
2833 * conflict with the Windows interpretation of DENY ACEs, where DENY ACEs
2834 * should take precedence ahead of ALLOW ACEs.
2835 *
2836 * This implementation always consults the target object's ACL first.
2837 * If a DENY ACE is present on the target object that specifies ACE_DELETE,
2838 * delete access is denied. If an ALLOW ACE with ACE_DELETE is present on
2839 * the target object, access is allowed. If and only if no entries with
2840 * ACE_DELETE are present in the object's ACL, check the container's ACL
2841 * for entries with ACE_DELETE_CHILD.
2842 *
2843 * A summary of the logic implemented from the table above is as follows:
2844 *
2845 * First check for DENY ACEs that apply.
2846 * If either target or container has a deny, EACCES.
2847 *
2848 * Delete access can then be summarized as follows:
2849 * 1: The object to be deleted grants ACE_DELETE, or
2850 * 2: The containing directory grants ACE_DELETE_CHILD.
2851 * In a Windows system, that would be the end of the story.
2852 * In this system, (2) has some complications...
2853 * 2a: "sticky" bit on a directory adds restrictions, and
2854 * 2b: existing ACEs from previous versions of ZFS may
2855 * not carry ACE_DELETE_CHILD where they should, so we
2856 * also allow delete when ACE_WRITE_DATA is granted.
2857 *
2858 * Note: 2b is technically a work-around for a prior bug,
2859 * which hopefully can go away some day. For those who
2860 * no longer need the work around, and for testing, this
2861 * work-around is made conditional via the tunable:
2862 * zfs_write_implies_delete_child
2863 */
2864 int
zfs_zaccess_delete(znode_t * dzp,znode_t * zp,cred_t * cr,zidmap_t * mnt_ns)2865 zfs_zaccess_delete(znode_t *dzp, znode_t *zp, cred_t *cr, zidmap_t *mnt_ns)
2866 {
2867 uint32_t wanted_dirperms;
2868 uint32_t dzp_working_mode = 0;
2869 uint32_t zp_working_mode = 0;
2870 int dzp_error, zp_error;
2871 boolean_t dzpcheck_privs;
2872 boolean_t zpcheck_privs;
2873
2874 if (zp->z_pflags & (ZFS_IMMUTABLE | ZFS_NOUNLINK))
2875 return (SET_ERROR(EPERM));
2876
2877 /*
2878 * Case 1:
2879 * If target object grants ACE_DELETE then we are done. This is
2880 * indicated by a return value of 0. For this case we don't worry
2881 * about the sticky bit because sticky only applies to the parent
2882 * directory and this is the child access result.
2883 *
2884 * If we encounter a DENY ACE here, we're also done (EACCES).
2885 * Note that if we hit a DENY ACE here (on the target) it should
2886 * take precedence over a DENY ACE on the container, so that when
2887 * we have more complete auditing support we will be able to
2888 * report an access failure against the specific target.
2889 * (This is part of why we're checking the target first.)
2890 */
2891 zp_error = zfs_zaccess_common(zp, ACE_DELETE, &zp_working_mode,
2892 &zpcheck_privs, B_FALSE, cr, mnt_ns);
2893 if (zp_error == EACCES) {
2894 /* We hit a DENY ACE. */
2895 if (!zpcheck_privs)
2896 return (SET_ERROR(zp_error));
2897 return (secpolicy_vnode_remove(cr));
2898
2899 }
2900 if (zp_error == 0)
2901 return (0);
2902
2903 /*
2904 * Case 2:
2905 * If the containing directory grants ACE_DELETE_CHILD,
2906 * or we're in backward compatibility mode and the
2907 * containing directory has ACE_WRITE_DATA, allow.
2908 * Case 2b is handled with wanted_dirperms.
2909 */
2910 wanted_dirperms = ACE_DELETE_CHILD;
2911 if (zfs_write_implies_delete_child)
2912 wanted_dirperms |= ACE_WRITE_DATA;
2913 dzp_error = zfs_zaccess_common(dzp, wanted_dirperms,
2914 &dzp_working_mode, &dzpcheck_privs, B_FALSE, cr, mnt_ns);
2915 if (dzp_error == EACCES) {
2916 /* We hit a DENY ACE. */
2917 if (!dzpcheck_privs)
2918 return (SET_ERROR(dzp_error));
2919 return (secpolicy_vnode_remove(cr));
2920 }
2921
2922 /*
2923 * Cases 2a, 2b (continued)
2924 *
2925 * Note: dzp_working_mode now contains any permissions
2926 * that were NOT granted. Therefore, if any of the
2927 * wanted_dirperms WERE granted, we will have:
2928 * dzp_working_mode != wanted_dirperms
2929 * We're really asking if ANY of those permissions
2930 * were granted, and if so, grant delete access.
2931 */
2932 if (dzp_working_mode != wanted_dirperms)
2933 dzp_error = 0;
2934
2935 /*
2936 * dzp_error is 0 if the container granted us permissions to "modify".
2937 * If we do not have permission via one or more ACEs, our current
2938 * privileges may still permit us to modify the container.
2939 *
2940 * dzpcheck_privs is false when i.e. the FS is read-only.
2941 * Otherwise, do privilege checks for the container.
2942 */
2943 if (dzp_error != 0 && dzpcheck_privs) {
2944 uid_t owner;
2945
2946 /*
2947 * The secpolicy call needs the requested access and
2948 * the current access mode of the container, but it
2949 * only knows about Unix-style modes (VEXEC, VWRITE),
2950 * so this must condense the fine-grained ACE bits into
2951 * Unix modes.
2952 *
2953 * The VEXEC flag is easy, because we know that has
2954 * always been checked before we get here (during the
2955 * lookup of the target vnode). The container has not
2956 * granted us permissions to "modify", so we do not set
2957 * the VWRITE flag in the current access mode.
2958 */
2959 owner = zfs_fuid_map_id(ZTOZSB(dzp),
2960 KUID_TO_SUID(ZTOI(dzp)->i_uid), cr, ZFS_OWNER);
2961 dzp_error = secpolicy_vnode_access2(cr, ZTOI(dzp),
2962 owner, S_IXUSR, S_IWUSR|S_IXUSR);
2963 }
2964 if (dzp_error != 0) {
2965 /*
2966 * Note: We may have dzp_error = -1 here (from
2967 * zfs_zacess_common). Don't return that.
2968 */
2969 return (SET_ERROR(EACCES));
2970 }
2971
2972
2973 /*
2974 * At this point, we know that the directory permissions allow
2975 * us to modify, but we still need to check for the additional
2976 * restrictions that apply when the "sticky bit" is set.
2977 *
2978 * Yes, zfs_sticky_remove_access() also checks this bit, but
2979 * checking it here and skipping the call below is nice when
2980 * you're watching all of this with dtrace.
2981 */
2982 if ((dzp->z_mode & S_ISVTX) == 0)
2983 return (0);
2984
2985 /*
2986 * zfs_sticky_remove_access will succeed if:
2987 * 1. The sticky bit is absent.
2988 * 2. We pass the sticky bit restrictions.
2989 * 3. We have privileges that always allow file removal.
2990 */
2991 return (zfs_sticky_remove_access(dzp, zp, cr));
2992 }
2993
2994 int
zfs_zaccess_rename(znode_t * sdzp,znode_t * szp,znode_t * tdzp,znode_t * tzp,cred_t * cr,zidmap_t * mnt_ns)2995 zfs_zaccess_rename(znode_t *sdzp, znode_t *szp, znode_t *tdzp,
2996 znode_t *tzp, cred_t *cr, zidmap_t *mnt_ns)
2997 {
2998 int add_perm;
2999 int error;
3000
3001 if (szp->z_pflags & ZFS_AV_QUARANTINED)
3002 return (SET_ERROR(EACCES));
3003
3004 add_perm = S_ISDIR(ZTOI(szp)->i_mode) ?
3005 ACE_ADD_SUBDIRECTORY : ACE_ADD_FILE;
3006
3007 /*
3008 * Rename permissions are combination of delete permission +
3009 * add file/subdir permission.
3010 */
3011
3012 /*
3013 * first make sure we do the delete portion.
3014 *
3015 * If that succeeds then check for add_file/add_subdir permissions
3016 */
3017
3018 if ((error = zfs_zaccess_delete(sdzp, szp, cr, mnt_ns)))
3019 return (error);
3020
3021 /*
3022 * If we have a tzp, see if we can delete it?
3023 */
3024 if (tzp) {
3025 if ((error = zfs_zaccess_delete(tdzp, tzp, cr, mnt_ns)))
3026 return (error);
3027 }
3028
3029 /*
3030 * Now check for add permissions
3031 */
3032 error = zfs_zaccess(tdzp, add_perm, 0, B_FALSE, cr, mnt_ns);
3033
3034 return (error);
3035 }
3036