xref: /freebsd/sys/kern/subr_acl_nfs4.c (revision f0cfa1b168014f56c02b83e5f28412cc5f78d117)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2008-2010 Edward Tomasz Napierała <trasz@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 /*
30  * ACL support routines specific to NFSv4 access control lists.  These are
31  * utility routines for code common across file systems implementing NFSv4
32  * ACLs.
33  */
34 
35 #ifdef _KERNEL
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/param.h>
40 #include <sys/kernel.h>
41 #include <sys/module.h>
42 #include <sys/systm.h>
43 #include <sys/mount.h>
44 #include <sys/priv.h>
45 #include <sys/vnode.h>
46 #include <sys/errno.h>
47 #include <sys/stat.h>
48 #include <sys/sysctl.h>
49 #include <sys/acl.h>
50 #else
51 #include <errno.h>
52 #include <assert.h>
53 #include <sys/acl.h>
54 #include <sys/stat.h>
55 #define KASSERT(a, b) assert(a)
56 #define CTASSERT(a)
57 
58 #endif /* !_KERNEL */
59 
60 #ifdef _KERNEL
61 
62 static void	acl_nfs4_trivial_from_mode(struct acl *aclp, mode_t mode);
63 
64 static int	acl_nfs4_old_semantics = 0;
65 
66 SYSCTL_INT(_vfs, OID_AUTO, acl_nfs4_old_semantics, CTLFLAG_RW,
67     &acl_nfs4_old_semantics, 0, "Use pre-PSARC/2010/029 NFSv4 ACL semantics");
68 
69 static struct {
70 	accmode_t accmode;
71 	int mask;
72 } accmode2mask[] = {{VREAD, ACL_READ_DATA},
73 		    {VWRITE, ACL_WRITE_DATA},
74 		    {VAPPEND, ACL_APPEND_DATA},
75 		    {VEXEC, ACL_EXECUTE},
76 		    {VREAD_NAMED_ATTRS, ACL_READ_NAMED_ATTRS},
77 		    {VWRITE_NAMED_ATTRS, ACL_WRITE_NAMED_ATTRS},
78 		    {VDELETE_CHILD, ACL_DELETE_CHILD},
79 		    {VREAD_ATTRIBUTES, ACL_READ_ATTRIBUTES},
80 		    {VWRITE_ATTRIBUTES, ACL_WRITE_ATTRIBUTES},
81 		    {VDELETE, ACL_DELETE},
82 		    {VREAD_ACL, ACL_READ_ACL},
83 		    {VWRITE_ACL, ACL_WRITE_ACL},
84 		    {VWRITE_OWNER, ACL_WRITE_OWNER},
85 		    {VSYNCHRONIZE, ACL_SYNCHRONIZE},
86 		    {0, 0}};
87 
88 static int
89 _access_mask_from_accmode(accmode_t accmode)
90 {
91 	int access_mask = 0, i;
92 
93 	for (i = 0; accmode2mask[i].accmode != 0; i++) {
94 		if (accmode & accmode2mask[i].accmode)
95 			access_mask |= accmode2mask[i].mask;
96 	}
97 
98 	/*
99 	 * VAPPEND is just a modifier for VWRITE; if the caller asked
100 	 * for 'VAPPEND | VWRITE', we want to check for ACL_APPEND_DATA only.
101 	 */
102 	if (access_mask & ACL_APPEND_DATA)
103 		access_mask &= ~ACL_WRITE_DATA;
104 
105 	return (access_mask);
106 }
107 
108 /*
109  * Return 0, iff access is allowed, 1 otherwise.
110  */
111 static int
112 _acl_denies(const struct acl *aclp, int access_mask, struct ucred *cred,
113     int file_uid, int file_gid, int *denied_explicitly)
114 {
115 	int i;
116 	const struct acl_entry *entry;
117 
118 	if (denied_explicitly != NULL)
119 		*denied_explicitly = 0;
120 
121 	KASSERT(aclp->acl_cnt <= ACL_MAX_ENTRIES,
122 	    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
123 
124 	for (i = 0; i < aclp->acl_cnt; i++) {
125 		entry = &(aclp->acl_entry[i]);
126 
127 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
128 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
129 			continue;
130 		if (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY)
131 			continue;
132 		switch (entry->ae_tag) {
133 		case ACL_USER_OBJ:
134 			if (file_uid != cred->cr_uid)
135 				continue;
136 			break;
137 		case ACL_USER:
138 			if (entry->ae_id != cred->cr_uid)
139 				continue;
140 			break;
141 		case ACL_GROUP_OBJ:
142 			if (!groupmember(file_gid, cred))
143 				continue;
144 			break;
145 		case ACL_GROUP:
146 			if (!groupmember(entry->ae_id, cred))
147 				continue;
148 			break;
149 		default:
150 			KASSERT(entry->ae_tag == ACL_EVERYONE,
151 			    ("entry->ae_tag == ACL_EVERYONE"));
152 		}
153 
154 		if (entry->ae_entry_type == ACL_ENTRY_TYPE_DENY) {
155 			if (entry->ae_perm & access_mask) {
156 				if (denied_explicitly != NULL)
157 					*denied_explicitly = 1;
158 				return (1);
159 			}
160 		}
161 
162 		access_mask &= ~(entry->ae_perm);
163 		if (access_mask == 0)
164 			return (0);
165 	}
166 
167 	if (access_mask == 0)
168 		return (0);
169 
170 	return (1);
171 }
172 
173 int
174 vaccess_acl_nfs4(enum vtype type, uid_t file_uid, gid_t file_gid,
175     struct acl *aclp, accmode_t accmode, struct ucred *cred, int *privused)
176 {
177 	accmode_t priv_granted = 0;
178 	int denied, explicitly_denied, access_mask, is_directory,
179 	    must_be_owner = 0;
180 	mode_t file_mode = 0;
181 
182 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND |
183 	    VEXPLICIT_DENY | VREAD_NAMED_ATTRS | VWRITE_NAMED_ATTRS |
184 	    VDELETE_CHILD | VREAD_ATTRIBUTES | VWRITE_ATTRIBUTES | VDELETE |
185 	    VREAD_ACL | VWRITE_ACL | VWRITE_OWNER | VSYNCHRONIZE)) == 0,
186 	    ("invalid bit in accmode"));
187 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
188 	    	("VAPPEND without VWRITE"));
189 
190 	if (privused != NULL)
191 		*privused = 0;
192 
193 	if (accmode & VADMIN)
194 		must_be_owner = 1;
195 
196 	/*
197 	 * Ignore VSYNCHRONIZE permission.
198 	 */
199 	accmode &= ~VSYNCHRONIZE;
200 
201 	access_mask = _access_mask_from_accmode(accmode);
202 
203 	if (type == VDIR)
204 		is_directory = 1;
205 	else
206 		is_directory = 0;
207 
208 	/*
209 	 * File owner is always allowed to read and write the ACL
210 	 * and basic attributes.  This is to prevent a situation
211 	 * where user would change ACL in a way that prevents him
212 	 * from undoing the change.
213 	 */
214 	if (file_uid == cred->cr_uid)
215 		access_mask &= ~(ACL_READ_ACL | ACL_WRITE_ACL |
216 		    ACL_READ_ATTRIBUTES | ACL_WRITE_ATTRIBUTES);
217 
218 	/*
219 	 * Ignore append permission for regular files; use write
220 	 * permission instead.
221 	 */
222 	if (!is_directory && (access_mask & ACL_APPEND_DATA)) {
223 		access_mask &= ~ACL_APPEND_DATA;
224 		access_mask |= ACL_WRITE_DATA;
225 	}
226 
227 	denied = _acl_denies(aclp, access_mask, cred, file_uid, file_gid,
228 	    &explicitly_denied);
229 
230 	if (must_be_owner) {
231 		if (file_uid != cred->cr_uid)
232 			denied = EPERM;
233 	}
234 
235 	/*
236 	 * For VEXEC, ensure that at least one execute bit is set for
237 	 * non-directories. We have to check the mode here to stay
238 	 * consistent with execve(2). See the test in
239 	 * exec_check_permissions().
240 	 */
241 	acl_nfs4_sync_mode_from_acl(&file_mode, aclp);
242 	if (!denied && !is_directory && (accmode & VEXEC) &&
243 	    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)
244 		denied = EACCES;
245 
246 	if (!denied)
247 		return (0);
248 
249 	/*
250 	 * Access failed.  Iff it was not denied explicitly and
251 	 * VEXPLICIT_DENY flag was specified, allow access.
252 	 */
253 	if ((accmode & VEXPLICIT_DENY) && explicitly_denied == 0)
254 		return (0);
255 
256 	accmode &= ~VEXPLICIT_DENY;
257 
258 	/*
259 	 * No match.  Try to use privileges, if there are any.
260 	 */
261 	if (is_directory) {
262 		if ((accmode & VEXEC) && !priv_check_cred(cred,
263 		    PRIV_VFS_LOOKUP, 0))
264 			priv_granted |= VEXEC;
265 	} else {
266 		/*
267 		 * Ensure that at least one execute bit is on. Otherwise,
268 		 * a privileged user will always succeed, and we don't want
269 		 * this to happen unless the file really is executable.
270 		 */
271 		if ((accmode & VEXEC) && (file_mode &
272 		    (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
273 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
274 			priv_granted |= VEXEC;
275 	}
276 
277 	if ((accmode & VREAD) && !priv_check_cred(cred, PRIV_VFS_READ, 0))
278 		priv_granted |= VREAD;
279 
280 	if ((accmode & (VWRITE | VAPPEND | VDELETE_CHILD)) &&
281 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
282 		priv_granted |= (VWRITE | VAPPEND | VDELETE_CHILD);
283 
284 	if ((accmode & VADMIN_PERMS) &&
285 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
286 		priv_granted |= VADMIN_PERMS;
287 
288 	if ((accmode & VSTAT_PERMS) &&
289 	    !priv_check_cred(cred, PRIV_VFS_STAT, 0))
290 		priv_granted |= VSTAT_PERMS;
291 
292 	if ((accmode & priv_granted) == accmode) {
293 		if (privused != NULL)
294 			*privused = 1;
295 
296 		return (0);
297 	}
298 
299 	if (accmode & (VADMIN_PERMS | VDELETE_CHILD | VDELETE))
300 		denied = EPERM;
301 	else
302 		denied = EACCES;
303 
304 	return (denied);
305 }
306 #endif /* _KERNEL */
307 
308 static int
309 _acl_entry_matches(struct acl_entry *entry, acl_tag_t tag, acl_perm_t perm,
310     acl_entry_type_t entry_type)
311 {
312 	if (entry->ae_tag != tag)
313 		return (0);
314 
315 	if (entry->ae_id != ACL_UNDEFINED_ID)
316 		return (0);
317 
318 	if (entry->ae_perm != perm)
319 		return (0);
320 
321 	if (entry->ae_entry_type != entry_type)
322 		return (0);
323 
324 	if (entry->ae_flags != 0)
325 		return (0);
326 
327 	return (1);
328 }
329 
330 static struct acl_entry *
331 _acl_append(struct acl *aclp, acl_tag_t tag, acl_perm_t perm,
332     acl_entry_type_t entry_type)
333 {
334 	struct acl_entry *entry;
335 
336 	KASSERT(aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
337 	    ("aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
338 
339 	entry = &(aclp->acl_entry[aclp->acl_cnt]);
340 	aclp->acl_cnt++;
341 
342 	entry->ae_tag = tag;
343 	entry->ae_id = ACL_UNDEFINED_ID;
344 	entry->ae_perm = perm;
345 	entry->ae_entry_type = entry_type;
346 	entry->ae_flags = 0;
347 
348 	return (entry);
349 }
350 
351 static struct acl_entry *
352 _acl_duplicate_entry(struct acl *aclp, int entry_index)
353 {
354 	int i;
355 
356 	KASSERT(aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
357 	    ("aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
358 
359 	for (i = aclp->acl_cnt; i > entry_index; i--)
360 		aclp->acl_entry[i] = aclp->acl_entry[i - 1];
361 
362 	aclp->acl_cnt++;
363 
364 	return (&(aclp->acl_entry[entry_index + 1]));
365 }
366 
367 static void
368 acl_nfs4_sync_acl_from_mode_draft(struct acl *aclp, mode_t mode,
369     int file_owner_id)
370 {
371 	int i, meets, must_append;
372 	struct acl_entry *entry, *copy, *previous,
373 	    *a1, *a2, *a3, *a4, *a5, *a6;
374 	mode_t amode;
375 	const int READ = 04;
376 	const int WRITE = 02;
377 	const int EXEC = 01;
378 
379 	KASSERT(aclp->acl_cnt <= ACL_MAX_ENTRIES,
380 	    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
381 
382 	/*
383 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
384 	 *
385 	 * 3.16.6.3. Applying a Mode to an Existing ACL
386 	 */
387 
388 	/*
389 	 * 1. For each ACE:
390 	 */
391 	for (i = 0; i < aclp->acl_cnt; i++) {
392 		entry = &(aclp->acl_entry[i]);
393 
394 		/*
395 		 * 1.1. If the type is neither ALLOW or DENY - skip.
396 		 */
397 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
398 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
399 			continue;
400 
401 		/*
402 		 * 1.2. If ACL_ENTRY_INHERIT_ONLY is set - skip.
403 		 */
404 		if (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY)
405 			continue;
406 
407 		/*
408 		 * 1.3. If ACL_ENTRY_FILE_INHERIT or ACL_ENTRY_DIRECTORY_INHERIT
409 		 *      are set:
410 		 */
411 		if (entry->ae_flags &
412 		    (ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT)) {
413 			/*
414 			 * 1.3.1. A copy of the current ACE is made, and placed
415 			 *        in the ACL immediately following the current
416 			 *        ACE.
417 			 */
418 			copy = _acl_duplicate_entry(aclp, i);
419 
420 			/*
421 			 * 1.3.2. In the first ACE, the flag
422 			 *        ACL_ENTRY_INHERIT_ONLY is set.
423 			 */
424 			entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
425 
426 			/*
427 			 * 1.3.3. In the second ACE, the following flags
428 			 *        are cleared:
429 			 *        ACL_ENTRY_FILE_INHERIT,
430 			 *        ACL_ENTRY_DIRECTORY_INHERIT,
431 			 *        ACL_ENTRY_NO_PROPAGATE_INHERIT.
432 			 */
433 			copy->ae_flags &= ~(ACL_ENTRY_FILE_INHERIT |
434 			    ACL_ENTRY_DIRECTORY_INHERIT |
435 			    ACL_ENTRY_NO_PROPAGATE_INHERIT);
436 
437 			/*
438 			 * The algorithm continues on with the second ACE.
439 			 */
440 			i++;
441 			entry = copy;
442 		}
443 
444 		/*
445 		 * 1.4. If it's owner@, group@ or everyone@ entry, clear
446 		 *      ACL_READ_DATA, ACL_WRITE_DATA, ACL_APPEND_DATA
447 		 *      and ACL_EXECUTE.  Continue to the next entry.
448 		 */
449 		if (entry->ae_tag == ACL_USER_OBJ ||
450 		    entry->ae_tag == ACL_GROUP_OBJ ||
451 		    entry->ae_tag == ACL_EVERYONE) {
452 			entry->ae_perm &= ~(ACL_READ_DATA | ACL_WRITE_DATA |
453 			    ACL_APPEND_DATA | ACL_EXECUTE);
454 			continue;
455 		}
456 
457 		/*
458 		 * 1.5. Otherwise, if the "who" field did not match one
459 		 *      of OWNER@, GROUP@, EVERYONE@:
460 		 *
461 		 * 1.5.1. If the type is ALLOW, check the preceding ACE.
462 		 *        If it does not meet all of the following criteria:
463 		 */
464 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW)
465 			continue;
466 
467 		meets = 0;
468 		if (i > 0) {
469 			meets = 1;
470 			previous = &(aclp->acl_entry[i - 1]);
471 
472 			/*
473 			 * 1.5.1.1. The type field is DENY,
474 			 */
475 			if (previous->ae_entry_type != ACL_ENTRY_TYPE_DENY)
476 				meets = 0;
477 
478 			/*
479 			 * 1.5.1.2. The "who" field is the same as the current
480 			 *          ACE,
481 			 *
482 			 * 1.5.1.3. The flag bit ACE4_IDENTIFIER_GROUP
483 			 *          is the same as it is in the current ACE,
484 			 *          and no other flag bits are set,
485 			 */
486 			if (previous->ae_id != entry->ae_id ||
487 			    previous->ae_tag != entry->ae_tag)
488 				meets = 0;
489 
490 			if (previous->ae_flags)
491 				meets = 0;
492 
493 			/*
494 			 * 1.5.1.4. The mask bits are a subset of the mask bits
495 			 *          of the current ACE, and are also subset of
496 			 *          the following: ACL_READ_DATA,
497 			 *          ACL_WRITE_DATA, ACL_APPEND_DATA, ACL_EXECUTE
498 			 */
499 			if (previous->ae_perm & ~(entry->ae_perm))
500 				meets = 0;
501 
502 			if (previous->ae_perm & ~(ACL_READ_DATA |
503 			    ACL_WRITE_DATA | ACL_APPEND_DATA | ACL_EXECUTE))
504 				meets = 0;
505 		}
506 
507 		if (!meets) {
508 			/*
509 		 	 * Then the ACE of type DENY, with a who equal
510 			 * to the current ACE, flag bits equal to
511 			 * (<current ACE flags> & <ACE_IDENTIFIER_GROUP>)
512 			 * and no mask bits, is prepended.
513 			 */
514 			previous = entry;
515 			entry = _acl_duplicate_entry(aclp, i);
516 
517 			/* Adjust counter, as we've just added an entry. */
518 			i++;
519 
520 			previous->ae_tag = entry->ae_tag;
521 			previous->ae_id = entry->ae_id;
522 			previous->ae_flags = entry->ae_flags;
523 			previous->ae_perm = 0;
524 			previous->ae_entry_type = ACL_ENTRY_TYPE_DENY;
525 		}
526 
527 		/*
528 		 * 1.5.2. The following modifications are made to the prepended
529 		 *        ACE.  The intent is to mask the following ACE
530 		 *        to disallow ACL_READ_DATA, ACL_WRITE_DATA,
531 		 *        ACL_APPEND_DATA, or ACL_EXECUTE, based upon the group
532 		 *        permissions of the new mode.  As a special case,
533 		 *        if the ACE matches the current owner of the file,
534 		 *        the owner bits are used, rather than the group bits.
535 		 *        This is reflected in the algorithm below.
536 		 */
537 		amode = mode >> 3;
538 
539 		/*
540 		 * If ACE4_IDENTIFIER_GROUP is not set, and the "who" field
541 		 * in ACE matches the owner of the file, we shift amode three
542 		 * more bits, in order to have the owner permission bits
543 		 * placed in the three low order bits of amode.
544 		 */
545 		if (entry->ae_tag == ACL_USER && entry->ae_id == file_owner_id)
546 			amode = amode >> 3;
547 
548 		if (entry->ae_perm & ACL_READ_DATA) {
549 			if (amode & READ)
550 				previous->ae_perm &= ~ACL_READ_DATA;
551 			else
552 				previous->ae_perm |= ACL_READ_DATA;
553 		}
554 
555 		if (entry->ae_perm & ACL_WRITE_DATA) {
556 			if (amode & WRITE)
557 				previous->ae_perm &= ~ACL_WRITE_DATA;
558 			else
559 				previous->ae_perm |= ACL_WRITE_DATA;
560 		}
561 
562 		if (entry->ae_perm & ACL_APPEND_DATA) {
563 			if (amode & WRITE)
564 				previous->ae_perm &= ~ACL_APPEND_DATA;
565 			else
566 				previous->ae_perm |= ACL_APPEND_DATA;
567 		}
568 
569 		if (entry->ae_perm & ACL_EXECUTE) {
570 			if (amode & EXEC)
571 				previous->ae_perm &= ~ACL_EXECUTE;
572 			else
573 				previous->ae_perm |= ACL_EXECUTE;
574 		}
575 
576 		/*
577 		 * 1.5.3. If ACE4_IDENTIFIER_GROUP is set in the flags
578 		 *        of the ALLOW ace:
579 		 *
580 		 * XXX: This point is not there in the Falkner's draft.
581 		 */
582 		if (entry->ae_tag == ACL_GROUP &&
583 		    entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW) {
584 			mode_t extramode, ownermode;
585 			extramode = (mode >> 3) & 07;
586 			ownermode = mode >> 6;
587 			extramode &= ~ownermode;
588 
589 			if (extramode) {
590 				if (extramode & READ) {
591 					entry->ae_perm &= ~ACL_READ_DATA;
592 					previous->ae_perm &= ~ACL_READ_DATA;
593 				}
594 
595 				if (extramode & WRITE) {
596 					entry->ae_perm &=
597 					    ~(ACL_WRITE_DATA | ACL_APPEND_DATA);
598 					previous->ae_perm &=
599 					    ~(ACL_WRITE_DATA | ACL_APPEND_DATA);
600 				}
601 
602 				if (extramode & EXEC) {
603 					entry->ae_perm &= ~ACL_EXECUTE;
604 					previous->ae_perm &= ~ACL_EXECUTE;
605 				}
606 			}
607 		}
608 	}
609 
610 	/*
611 	 * 2. If there at least six ACEs, the final six ACEs are examined.
612 	 *    If they are not equal to what we want, append six ACEs.
613 	 */
614 	must_append = 0;
615 	if (aclp->acl_cnt < 6) {
616 		must_append = 1;
617 	} else {
618 		a6 = &(aclp->acl_entry[aclp->acl_cnt - 1]);
619 		a5 = &(aclp->acl_entry[aclp->acl_cnt - 2]);
620 		a4 = &(aclp->acl_entry[aclp->acl_cnt - 3]);
621 		a3 = &(aclp->acl_entry[aclp->acl_cnt - 4]);
622 		a2 = &(aclp->acl_entry[aclp->acl_cnt - 5]);
623 		a1 = &(aclp->acl_entry[aclp->acl_cnt - 6]);
624 
625 		if (!_acl_entry_matches(a1, ACL_USER_OBJ, 0,
626 		    ACL_ENTRY_TYPE_DENY))
627 			must_append = 1;
628 		if (!_acl_entry_matches(a2, ACL_USER_OBJ, ACL_WRITE_ACL |
629 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
630 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_ALLOW))
631 			must_append = 1;
632 		if (!_acl_entry_matches(a3, ACL_GROUP_OBJ, 0,
633 		    ACL_ENTRY_TYPE_DENY))
634 			must_append = 1;
635 		if (!_acl_entry_matches(a4, ACL_GROUP_OBJ, 0,
636 		    ACL_ENTRY_TYPE_ALLOW))
637 			must_append = 1;
638 		if (!_acl_entry_matches(a5, ACL_EVERYONE, ACL_WRITE_ACL |
639 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
640 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_DENY))
641 			must_append = 1;
642 		if (!_acl_entry_matches(a6, ACL_EVERYONE, ACL_READ_ACL |
643 		    ACL_READ_ATTRIBUTES | ACL_READ_NAMED_ATTRS |
644 		    ACL_SYNCHRONIZE, ACL_ENTRY_TYPE_ALLOW))
645 			must_append = 1;
646 	}
647 
648 	if (must_append) {
649 		KASSERT(aclp->acl_cnt + 6 <= ACL_MAX_ENTRIES,
650 		    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
651 
652 		a1 = _acl_append(aclp, ACL_USER_OBJ, 0, ACL_ENTRY_TYPE_DENY);
653 		a2 = _acl_append(aclp, ACL_USER_OBJ, ACL_WRITE_ACL |
654 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
655 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_ALLOW);
656 		a3 = _acl_append(aclp, ACL_GROUP_OBJ, 0, ACL_ENTRY_TYPE_DENY);
657 		a4 = _acl_append(aclp, ACL_GROUP_OBJ, 0, ACL_ENTRY_TYPE_ALLOW);
658 		a5 = _acl_append(aclp, ACL_EVERYONE, ACL_WRITE_ACL |
659 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
660 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_DENY);
661 		a6 = _acl_append(aclp, ACL_EVERYONE, ACL_READ_ACL |
662 		    ACL_READ_ATTRIBUTES | ACL_READ_NAMED_ATTRS |
663 		    ACL_SYNCHRONIZE, ACL_ENTRY_TYPE_ALLOW);
664 
665 		KASSERT(a1 != NULL && a2 != NULL && a3 != NULL && a4 != NULL &&
666 		    a5 != NULL && a6 != NULL, ("couldn't append to ACL."));
667 	}
668 
669 	/*
670 	 * 3. The final six ACEs are adjusted according to the incoming mode.
671 	 */
672 	if (mode & S_IRUSR)
673 		a2->ae_perm |= ACL_READ_DATA;
674 	else
675 		a1->ae_perm |= ACL_READ_DATA;
676 	if (mode & S_IWUSR)
677 		a2->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
678 	else
679 		a1->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
680 	if (mode & S_IXUSR)
681 		a2->ae_perm |= ACL_EXECUTE;
682 	else
683 		a1->ae_perm |= ACL_EXECUTE;
684 
685 	if (mode & S_IRGRP)
686 		a4->ae_perm |= ACL_READ_DATA;
687 	else
688 		a3->ae_perm |= ACL_READ_DATA;
689 	if (mode & S_IWGRP)
690 		a4->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
691 	else
692 		a3->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
693 	if (mode & S_IXGRP)
694 		a4->ae_perm |= ACL_EXECUTE;
695 	else
696 		a3->ae_perm |= ACL_EXECUTE;
697 
698 	if (mode & S_IROTH)
699 		a6->ae_perm |= ACL_READ_DATA;
700 	else
701 		a5->ae_perm |= ACL_READ_DATA;
702 	if (mode & S_IWOTH)
703 		a6->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
704 	else
705 		a5->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
706 	if (mode & S_IXOTH)
707 		a6->ae_perm |= ACL_EXECUTE;
708 	else
709 		a5->ae_perm |= ACL_EXECUTE;
710 }
711 
712 #ifdef _KERNEL
713 void
714 acl_nfs4_sync_acl_from_mode(struct acl *aclp, mode_t mode,
715     int file_owner_id)
716 {
717 
718 	if (acl_nfs4_old_semantics)
719 		acl_nfs4_sync_acl_from_mode_draft(aclp, mode, file_owner_id);
720 	else
721 		acl_nfs4_trivial_from_mode(aclp, mode);
722 }
723 #endif /* _KERNEL */
724 
725 void
726 acl_nfs4_sync_mode_from_acl(mode_t *_mode, const struct acl *aclp)
727 {
728 	int i;
729 	mode_t old_mode = *_mode, mode = 0, seen = 0;
730 	const struct acl_entry *entry;
731 
732 	KASSERT(aclp->acl_cnt <= ACL_MAX_ENTRIES,
733 	    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
734 
735 	/*
736 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
737 	 *
738 	 * 3.16.6.1. Recomputing mode upon SETATTR of ACL
739 	 */
740 
741 	for (i = 0; i < aclp->acl_cnt; i++) {
742 		entry = &(aclp->acl_entry[i]);
743 
744 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
745 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
746 			continue;
747 
748 		if (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY)
749 			continue;
750 
751 		if (entry->ae_tag == ACL_USER_OBJ) {
752 			if ((entry->ae_perm & ACL_READ_DATA) &&
753 			    ((seen & S_IRUSR) == 0)) {
754 				seen |= S_IRUSR;
755 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
756 					mode |= S_IRUSR;
757 			}
758 			if ((entry->ae_perm & ACL_WRITE_DATA) &&
759 			     ((seen & S_IWUSR) == 0)) {
760 				seen |= S_IWUSR;
761 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
762 					mode |= S_IWUSR;
763 			}
764 			if ((entry->ae_perm & ACL_EXECUTE) &&
765 			    ((seen & S_IXUSR) == 0)) {
766 				seen |= S_IXUSR;
767 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
768 					mode |= S_IXUSR;
769 			}
770 		} else if (entry->ae_tag == ACL_GROUP_OBJ) {
771 			if ((entry->ae_perm & ACL_READ_DATA) &&
772 			    ((seen & S_IRGRP) == 0)) {
773 				seen |= S_IRGRP;
774 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
775 					mode |= S_IRGRP;
776 			}
777 			if ((entry->ae_perm & ACL_WRITE_DATA) &&
778 			    ((seen & S_IWGRP) == 0)) {
779 				seen |= S_IWGRP;
780 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
781 					mode |= S_IWGRP;
782 			}
783 			if ((entry->ae_perm & ACL_EXECUTE) &&
784 			    ((seen & S_IXGRP) == 0)) {
785 				seen |= S_IXGRP;
786 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
787 					mode |= S_IXGRP;
788 			}
789 		} else if (entry->ae_tag == ACL_EVERYONE) {
790 			if (entry->ae_perm & ACL_READ_DATA) {
791 				if ((seen & S_IRUSR) == 0) {
792 					seen |= S_IRUSR;
793 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
794 						mode |= S_IRUSR;
795 				}
796 				if ((seen & S_IRGRP) == 0) {
797 					seen |= S_IRGRP;
798 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
799 						mode |= S_IRGRP;
800 				}
801 				if ((seen & S_IROTH) == 0) {
802 					seen |= S_IROTH;
803 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
804 						mode |= S_IROTH;
805 				}
806 			}
807 			if (entry->ae_perm & ACL_WRITE_DATA) {
808 				if ((seen & S_IWUSR) == 0) {
809 					seen |= S_IWUSR;
810 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
811 						mode |= S_IWUSR;
812 				}
813 				if ((seen & S_IWGRP) == 0) {
814 					seen |= S_IWGRP;
815 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
816 						mode |= S_IWGRP;
817 				}
818 				if ((seen & S_IWOTH) == 0) {
819 					seen |= S_IWOTH;
820 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
821 						mode |= S_IWOTH;
822 				}
823 			}
824 			if (entry->ae_perm & ACL_EXECUTE) {
825 				if ((seen & S_IXUSR) == 0) {
826 					seen |= S_IXUSR;
827 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
828 						mode |= S_IXUSR;
829 				}
830 				if ((seen & S_IXGRP) == 0) {
831 					seen |= S_IXGRP;
832 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
833 						mode |= S_IXGRP;
834 				}
835 				if ((seen & S_IXOTH) == 0) {
836 					seen |= S_IXOTH;
837 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
838 						mode |= S_IXOTH;
839 				}
840 			}
841 		}
842 	}
843 
844 	*_mode = mode | (old_mode & ACL_PRESERVE_MASK);
845 }
846 
847 #ifdef _KERNEL
848 /*
849  * Calculate inherited ACL in a manner compatible with NFSv4 Minor Version 1,
850  * draft-ietf-nfsv4-minorversion1-03.txt.
851  */
852 static void
853 acl_nfs4_compute_inherited_acl_draft(const struct acl *parent_aclp,
854     struct acl *child_aclp, mode_t mode, int file_owner_id,
855     int is_directory)
856 {
857 	int i, flags;
858 	const struct acl_entry *parent_entry;
859 	struct acl_entry *entry, *copy;
860 
861 	KASSERT(child_aclp->acl_cnt == 0, ("child_aclp->acl_cnt == 0"));
862 	KASSERT(parent_aclp->acl_cnt <= ACL_MAX_ENTRIES,
863 	    ("parent_aclp->acl_cnt <= ACL_MAX_ENTRIES"));
864 
865 	/*
866 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
867 	 *
868 	 * 3.16.6.2. Applying the mode given to CREATE or OPEN
869 	 *           to an inherited ACL
870 	 */
871 
872 	/*
873 	 * 1. Form an ACL that is the concatenation of all inheritable ACEs.
874 	 */
875 	for (i = 0; i < parent_aclp->acl_cnt; i++) {
876 		parent_entry = &(parent_aclp->acl_entry[i]);
877 		flags = parent_entry->ae_flags;
878 
879 		/*
880 		 * Entry is not inheritable at all.
881 		 */
882 		if ((flags & (ACL_ENTRY_DIRECTORY_INHERIT |
883 		    ACL_ENTRY_FILE_INHERIT)) == 0)
884 			continue;
885 
886 		/*
887 		 * We're creating a file, but entry is not inheritable
888 		 * by files.
889 		 */
890 		if (!is_directory && (flags & ACL_ENTRY_FILE_INHERIT) == 0)
891 			continue;
892 
893 		/*
894 		 * Entry is inheritable only by files, but has NO_PROPAGATE
895 		 * flag set, and we're creating a directory, so it wouldn't
896 		 * propagate to any file in that directory anyway.
897 		 */
898 		if (is_directory &&
899 		    (flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0 &&
900 		    (flags & ACL_ENTRY_NO_PROPAGATE_INHERIT))
901 			continue;
902 
903 		KASSERT(child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
904 		    ("child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
905 		child_aclp->acl_entry[child_aclp->acl_cnt] = *parent_entry;
906 		child_aclp->acl_cnt++;
907 	}
908 
909 	/*
910 	 * 2. For each entry in the new ACL, adjust its flags, possibly
911 	 *    creating two entries in place of one.
912 	 */
913 	for (i = 0; i < child_aclp->acl_cnt; i++) {
914 		entry = &(child_aclp->acl_entry[i]);
915 
916 		/*
917 		 * This is not in the specification, but SunOS
918 		 * apparently does that.
919 		 */
920 		if (((entry->ae_flags & ACL_ENTRY_NO_PROPAGATE_INHERIT) ||
921 		    !is_directory) &&
922 		    entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
923 			entry->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER);
924 
925 		/*
926 		 * 2.A. If the ACL_ENTRY_NO_PROPAGATE_INHERIT is set, or if the object
927 		 *      being created is not a directory, then clear the
928 		 *      following flags: ACL_ENTRY_NO_PROPAGATE_INHERIT,
929 		 *      ACL_ENTRY_FILE_INHERIT, ACL_ENTRY_DIRECTORY_INHERIT,
930 		 *      ACL_ENTRY_INHERIT_ONLY.
931 		 */
932 		if (entry->ae_flags & ACL_ENTRY_NO_PROPAGATE_INHERIT ||
933 		    !is_directory) {
934 			entry->ae_flags &= ~(ACL_ENTRY_NO_PROPAGATE_INHERIT |
935 			ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT |
936 			ACL_ENTRY_INHERIT_ONLY);
937 
938 			/*
939 			 * Continue on to the next ACE.
940 			 */
941 			continue;
942 		}
943 
944 		/*
945 		 * 2.B. If the object is a directory and ACL_ENTRY_FILE_INHERIT
946 		 *      is set, but ACL_ENTRY_NO_PROPAGATE_INHERIT is not set, ensure
947 		 *      that ACL_ENTRY_INHERIT_ONLY is set.  Continue to the
948 		 *      next ACE.  Otherwise...
949 		 */
950 		/*
951 		 * XXX: Read it again and make sure what does the "otherwise"
952 		 *      apply to.
953 		 */
954 		if (is_directory &&
955 		    (entry->ae_flags & ACL_ENTRY_FILE_INHERIT) &&
956 		    ((entry->ae_flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0)) {
957 			entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
958 			continue;
959 		}
960 
961 		/*
962 		 * 2.C. If the type of the ACE is neither ALLOW nor deny,
963 		 *      then continue.
964 		 */
965 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
966 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
967 			continue;
968 
969 		/*
970 		 * 2.D. Copy the original ACE into a second, adjacent ACE.
971 		 */
972 		copy = _acl_duplicate_entry(child_aclp, i);
973 
974 		/*
975 		 * 2.E. On the first ACE, ensure that ACL_ENTRY_INHERIT_ONLY
976 		 *      is set.
977 		 */
978 		entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
979 
980 		/*
981 		 * 2.F. On the second ACE, clear the following flags:
982 		 *      ACL_ENTRY_NO_PROPAGATE_INHERIT, ACL_ENTRY_FILE_INHERIT,
983 		 *      ACL_ENTRY_DIRECTORY_INHERIT, ACL_ENTRY_INHERIT_ONLY.
984 		 */
985 		copy->ae_flags &= ~(ACL_ENTRY_NO_PROPAGATE_INHERIT |
986 		    ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT |
987 		    ACL_ENTRY_INHERIT_ONLY);
988 
989 		/*
990 		 * 2.G. On the second ACE, if the type is ALLOW,
991 		 *      an implementation MAY clear the following
992 		 *      mask bits: ACL_WRITE_ACL, ACL_WRITE_OWNER.
993 		 */
994 		if (copy->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
995 			copy->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER);
996 
997 		/*
998 		 * Increment the counter to skip the copied entry.
999 		 */
1000 		i++;
1001 	}
1002 
1003 	/*
1004 	 * 3. To ensure that the mode is honored, apply the algorithm describe
1005 	 *    in Section 2.16.6.3, using the mode that is to be used for file
1006 	 *    creation.
1007 	 */
1008 	acl_nfs4_sync_acl_from_mode(child_aclp, mode, file_owner_id);
1009 }
1010 #endif /* _KERNEL */
1011 
1012 /*
1013  * Populate the ACL with entries inherited from parent_aclp.
1014  */
1015 static void
1016 acl_nfs4_inherit_entries(const struct acl *parent_aclp,
1017     struct acl *child_aclp, mode_t mode, int file_owner_id,
1018     int is_directory)
1019 {
1020 	int i, flags, tag;
1021 	const struct acl_entry *parent_entry;
1022 	struct acl_entry *entry;
1023 
1024 	KASSERT(parent_aclp->acl_cnt <= ACL_MAX_ENTRIES,
1025 	    ("parent_aclp->acl_cnt <= ACL_MAX_ENTRIES"));
1026 
1027 	for (i = 0; i < parent_aclp->acl_cnt; i++) {
1028 		parent_entry = &(parent_aclp->acl_entry[i]);
1029 		flags = parent_entry->ae_flags;
1030 		tag = parent_entry->ae_tag;
1031 
1032 		/*
1033 		 * Don't inherit owner@, group@, or everyone@ entries.
1034 		 */
1035 		if (tag == ACL_USER_OBJ || tag == ACL_GROUP_OBJ ||
1036 		    tag == ACL_EVERYONE)
1037 			continue;
1038 
1039 		/*
1040 		 * Entry is not inheritable at all.
1041 		 */
1042 		if ((flags & (ACL_ENTRY_DIRECTORY_INHERIT |
1043 		    ACL_ENTRY_FILE_INHERIT)) == 0)
1044 			continue;
1045 
1046 		/*
1047 		 * We're creating a file, but entry is not inheritable
1048 		 * by files.
1049 		 */
1050 		if (!is_directory && (flags & ACL_ENTRY_FILE_INHERIT) == 0)
1051 			continue;
1052 
1053 		/*
1054 		 * Entry is inheritable only by files, but has NO_PROPAGATE
1055 		 * flag set, and we're creating a directory, so it wouldn't
1056 		 * propagate to any file in that directory anyway.
1057 		 */
1058 		if (is_directory &&
1059 		    (flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0 &&
1060 		    (flags & ACL_ENTRY_NO_PROPAGATE_INHERIT))
1061 			continue;
1062 
1063 		/*
1064 		 * Entry qualifies for being inherited.
1065 		 */
1066 		KASSERT(child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
1067 		    ("child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
1068 		entry = &(child_aclp->acl_entry[child_aclp->acl_cnt]);
1069 		*entry = *parent_entry;
1070 		child_aclp->acl_cnt++;
1071 
1072 		entry->ae_flags &= ~ACL_ENTRY_INHERIT_ONLY;
1073 		entry->ae_flags |= ACL_ENTRY_INHERITED;
1074 
1075 		/*
1076 		 * If the type of the ACE is neither ALLOW nor DENY,
1077 		 * then leave it as it is and proceed to the next one.
1078 		 */
1079 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
1080 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
1081 			continue;
1082 
1083 		/*
1084 		 * If the ACL_ENTRY_NO_PROPAGATE_INHERIT is set, or if
1085 		 * the object being created is not a directory, then clear
1086 		 * the following flags: ACL_ENTRY_NO_PROPAGATE_INHERIT,
1087 		 * ACL_ENTRY_FILE_INHERIT, ACL_ENTRY_DIRECTORY_INHERIT,
1088 		 * ACL_ENTRY_INHERIT_ONLY.
1089 		 */
1090 		if (entry->ae_flags & ACL_ENTRY_NO_PROPAGATE_INHERIT ||
1091 		    !is_directory) {
1092 			entry->ae_flags &= ~(ACL_ENTRY_NO_PROPAGATE_INHERIT |
1093 			ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT |
1094 			ACL_ENTRY_INHERIT_ONLY);
1095 		}
1096 
1097 		/*
1098 		 * If the object is a directory and ACL_ENTRY_FILE_INHERIT
1099 		 * is set, but ACL_ENTRY_DIRECTORY_INHERIT is not set, ensure
1100 		 * that ACL_ENTRY_INHERIT_ONLY is set.
1101 		 */
1102 		if (is_directory &&
1103 		    (entry->ae_flags & ACL_ENTRY_FILE_INHERIT) &&
1104 		    ((entry->ae_flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0)) {
1105 			entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
1106 		}
1107 
1108 		if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW &&
1109 		    (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY) == 0) {
1110 			/*
1111 			 * Some permissions must never be inherited.
1112 			 */
1113 			entry->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER |
1114 			    ACL_WRITE_NAMED_ATTRS | ACL_WRITE_ATTRIBUTES);
1115 
1116 			/*
1117 			 * Others must be masked according to the file mode.
1118 			 */
1119 			if ((mode & S_IRGRP) == 0)
1120 				entry->ae_perm &= ~ACL_READ_DATA;
1121 			if ((mode & S_IWGRP) == 0)
1122 				entry->ae_perm &=
1123 				    ~(ACL_WRITE_DATA | ACL_APPEND_DATA);
1124 			if ((mode & S_IXGRP) == 0)
1125 				entry->ae_perm &= ~ACL_EXECUTE;
1126 		}
1127 	}
1128 }
1129 
1130 /*
1131  * Calculate inherited ACL in a manner compatible with PSARC/2010/029.
1132  * It's also being used to calculate a trivial ACL, by inheriting from
1133  * a NULL ACL.
1134  */
1135 static void
1136 acl_nfs4_compute_inherited_acl_psarc(const struct acl *parent_aclp,
1137     struct acl *aclp, mode_t mode, int file_owner_id, int is_directory)
1138 {
1139 	acl_perm_t user_allow_first = 0, user_deny = 0, group_deny = 0;
1140 	acl_perm_t user_allow, group_allow, everyone_allow;
1141 
1142 	KASSERT(aclp->acl_cnt == 0, ("aclp->acl_cnt == 0"));
1143 
1144 	user_allow = group_allow = everyone_allow = ACL_READ_ACL |
1145 	    ACL_READ_ATTRIBUTES | ACL_READ_NAMED_ATTRS | ACL_SYNCHRONIZE;
1146 	user_allow |= ACL_WRITE_ACL | ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
1147 	    ACL_WRITE_NAMED_ATTRS;
1148 
1149 	if (mode & S_IRUSR)
1150 		user_allow |= ACL_READ_DATA;
1151 	if (mode & S_IWUSR)
1152 		user_allow |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
1153 	if (mode & S_IXUSR)
1154 		user_allow |= ACL_EXECUTE;
1155 
1156 	if (mode & S_IRGRP)
1157 		group_allow |= ACL_READ_DATA;
1158 	if (mode & S_IWGRP)
1159 		group_allow |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
1160 	if (mode & S_IXGRP)
1161 		group_allow |= ACL_EXECUTE;
1162 
1163 	if (mode & S_IROTH)
1164 		everyone_allow |= ACL_READ_DATA;
1165 	if (mode & S_IWOTH)
1166 		everyone_allow |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
1167 	if (mode & S_IXOTH)
1168 		everyone_allow |= ACL_EXECUTE;
1169 
1170 	user_deny = ((group_allow | everyone_allow) & ~user_allow);
1171 	group_deny = everyone_allow & ~group_allow;
1172 	user_allow_first = group_deny & ~user_deny;
1173 
1174 	if (user_allow_first != 0)
1175 		_acl_append(aclp, ACL_USER_OBJ, user_allow_first,
1176 		    ACL_ENTRY_TYPE_ALLOW);
1177 	if (user_deny != 0)
1178 		_acl_append(aclp, ACL_USER_OBJ, user_deny,
1179 		    ACL_ENTRY_TYPE_DENY);
1180 	if (group_deny != 0)
1181 		_acl_append(aclp, ACL_GROUP_OBJ, group_deny,
1182 		    ACL_ENTRY_TYPE_DENY);
1183 
1184 	if (parent_aclp != NULL)
1185 		acl_nfs4_inherit_entries(parent_aclp, aclp, mode,
1186 		    file_owner_id, is_directory);
1187 
1188 	_acl_append(aclp, ACL_USER_OBJ, user_allow, ACL_ENTRY_TYPE_ALLOW);
1189 	_acl_append(aclp, ACL_GROUP_OBJ, group_allow, ACL_ENTRY_TYPE_ALLOW);
1190 	_acl_append(aclp, ACL_EVERYONE, everyone_allow, ACL_ENTRY_TYPE_ALLOW);
1191 }
1192 
1193 #ifdef _KERNEL
1194 void
1195 acl_nfs4_compute_inherited_acl(const struct acl *parent_aclp,
1196     struct acl *child_aclp, mode_t mode, int file_owner_id,
1197     int is_directory)
1198 {
1199 
1200 	if (acl_nfs4_old_semantics)
1201 		acl_nfs4_compute_inherited_acl_draft(parent_aclp, child_aclp,
1202 		    mode, file_owner_id, is_directory);
1203 	else
1204 		acl_nfs4_compute_inherited_acl_psarc(parent_aclp, child_aclp,
1205 		    mode, file_owner_id, is_directory);
1206 }
1207 #endif /* _KERNEL */
1208 
1209 /*
1210  * Calculate trivial ACL in a manner compatible with PSARC/2010/029.
1211  * Note that this results in an ACL different from (but semantically
1212  * equal to) the "canonical six" trivial ACL computed using algorithm
1213  * described in draft-ietf-nfsv4-minorversion1-03.txt, 3.16.6.2.
1214  */
1215 static void
1216 acl_nfs4_trivial_from_mode(struct acl *aclp, mode_t mode)
1217 {
1218 
1219 	aclp->acl_cnt = 0;
1220 	acl_nfs4_compute_inherited_acl_psarc(NULL, aclp, mode, -1, -1);
1221 }
1222 
1223 #ifndef _KERNEL
1224 /*
1225  * This routine is used by libc to implement acl_strip_np(3)
1226  * and acl_is_trivial_np(3).
1227  */
1228 void
1229 acl_nfs4_trivial_from_mode_libc(struct acl *aclp, int mode, int canonical_six)
1230 {
1231 
1232 	aclp->acl_cnt = 0;
1233 	if (canonical_six)
1234 		acl_nfs4_sync_acl_from_mode_draft(aclp, mode, -1);
1235 	else
1236 		acl_nfs4_trivial_from_mode(aclp, mode);
1237 }
1238 #endif /* !_KERNEL */
1239 
1240 #ifdef _KERNEL
1241 static int
1242 _acls_are_equal(const struct acl *a, const struct acl *b)
1243 {
1244 	int i;
1245 	const struct acl_entry *entrya, *entryb;
1246 
1247 	if (a->acl_cnt != b->acl_cnt)
1248 		return (0);
1249 
1250 	for (i = 0; i < b->acl_cnt; i++) {
1251 		entrya = &(a->acl_entry[i]);
1252 		entryb = &(b->acl_entry[i]);
1253 
1254 		if (entrya->ae_tag != entryb->ae_tag ||
1255 		    entrya->ae_id != entryb->ae_id ||
1256 		    entrya->ae_perm != entryb->ae_perm ||
1257 		    entrya->ae_entry_type != entryb->ae_entry_type ||
1258 		    entrya->ae_flags != entryb->ae_flags)
1259 			return (0);
1260 	}
1261 
1262 	return (1);
1263 }
1264 
1265 /*
1266  * This routine is used to determine whether to remove extended attribute
1267  * that stores ACL contents.
1268  */
1269 int
1270 acl_nfs4_is_trivial(const struct acl *aclp, int file_owner_id)
1271 {
1272 	int trivial;
1273 	mode_t tmpmode = 0;
1274 	struct acl *tmpaclp;
1275 
1276 	if (aclp->acl_cnt > 6)
1277 		return (0);
1278 
1279 	/*
1280 	 * Compute the mode from the ACL, then compute new ACL from that mode.
1281 	 * If the ACLs are identical, then the ACL is trivial.
1282 	 *
1283 	 * XXX: I guess there is a faster way to do this.  However, even
1284 	 *      this slow implementation significantly speeds things up
1285 	 *      for files that don't have non-trivial ACLs - it's critical
1286 	 *      for performance to not use EA when they are not needed.
1287 	 *
1288 	 * First try the PSARC/2010/029 semantics.
1289 	 */
1290 	tmpaclp = acl_alloc(M_WAITOK | M_ZERO);
1291 	acl_nfs4_sync_mode_from_acl(&tmpmode, aclp);
1292 	acl_nfs4_trivial_from_mode(tmpaclp, tmpmode);
1293 	trivial = _acls_are_equal(aclp, tmpaclp);
1294 	if (trivial) {
1295 		acl_free(tmpaclp);
1296 		return (trivial);
1297 	}
1298 
1299 	/*
1300 	 * Check if it's a draft-ietf-nfsv4-minorversion1-03.txt trivial ACL.
1301 	 */
1302 	tmpaclp->acl_cnt = 0;
1303 	acl_nfs4_sync_acl_from_mode_draft(tmpaclp, tmpmode, file_owner_id);
1304 	trivial = _acls_are_equal(aclp, tmpaclp);
1305 	acl_free(tmpaclp);
1306 
1307 	return (trivial);
1308 }
1309 #endif /* _KERNEL */
1310 
1311 int
1312 acl_nfs4_check(const struct acl *aclp, int is_directory)
1313 {
1314 	int i;
1315 	const struct acl_entry *entry;
1316 
1317 	/*
1318 	 * The spec doesn't seem to say anything about ACL validity.
1319 	 * It seems there is not much to do here.  There is even no need
1320 	 * to count "owner@" or "everyone@" (ACL_USER_OBJ and ACL_EVERYONE)
1321 	 * entries, as there can be several of them and that's perfectly
1322 	 * valid.  There can be none of them too.  Really.
1323 	 */
1324 
1325 	if (aclp->acl_cnt > ACL_MAX_ENTRIES || aclp->acl_cnt <= 0)
1326 		return (EINVAL);
1327 
1328 	for (i = 0; i < aclp->acl_cnt; i++) {
1329 		entry = &(aclp->acl_entry[i]);
1330 
1331 		switch (entry->ae_tag) {
1332 		case ACL_USER_OBJ:
1333 		case ACL_GROUP_OBJ:
1334 		case ACL_EVERYONE:
1335 			if (entry->ae_id != ACL_UNDEFINED_ID)
1336 				return (EINVAL);
1337 			break;
1338 
1339 		case ACL_USER:
1340 		case ACL_GROUP:
1341 			if (entry->ae_id == ACL_UNDEFINED_ID)
1342 				return (EINVAL);
1343 			break;
1344 
1345 		default:
1346 			return (EINVAL);
1347 		}
1348 
1349 		if ((entry->ae_perm | ACL_NFS4_PERM_BITS) != ACL_NFS4_PERM_BITS)
1350 			return (EINVAL);
1351 
1352 		/*
1353 		 * Disallow ACL_ENTRY_TYPE_AUDIT and ACL_ENTRY_TYPE_ALARM for now.
1354 		 */
1355 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
1356 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
1357 			return (EINVAL);
1358 
1359 		if ((entry->ae_flags | ACL_FLAGS_BITS) != ACL_FLAGS_BITS)
1360 			return (EINVAL);
1361 
1362 		/* Disallow unimplemented flags. */
1363 		if (entry->ae_flags & (ACL_ENTRY_SUCCESSFUL_ACCESS |
1364 		    ACL_ENTRY_FAILED_ACCESS))
1365 			return (EINVAL);
1366 
1367 		/* Disallow flags not allowed for ordinary files. */
1368 		if (!is_directory) {
1369 			if (entry->ae_flags & (ACL_ENTRY_FILE_INHERIT |
1370 			    ACL_ENTRY_DIRECTORY_INHERIT |
1371 			    ACL_ENTRY_NO_PROPAGATE_INHERIT | ACL_ENTRY_INHERIT_ONLY))
1372 				return (EINVAL);
1373 		}
1374 	}
1375 
1376 	return (0);
1377 }
1378 
1379 #ifdef	_KERNEL
1380 static int
1381 acl_nfs4_modload(module_t module, int what, void *arg)
1382 {
1383 	int ret;
1384 
1385 	ret = 0;
1386 
1387 	switch (what) {
1388 	case MOD_LOAD:
1389 	case MOD_SHUTDOWN:
1390 		break;
1391 
1392 	case MOD_QUIESCE:
1393 		/* XXX TODO */
1394 		ret = 0;
1395 		break;
1396 
1397 	case MOD_UNLOAD:
1398 		/* XXX TODO */
1399 		ret = 0;
1400 		break;
1401 	default:
1402 		ret = EINVAL;
1403 		break;
1404 	}
1405 
1406 	return (ret);
1407 }
1408 
1409 static moduledata_t acl_nfs4_mod = {
1410 	"acl_nfs4",
1411 	acl_nfs4_modload,
1412 	NULL
1413 };
1414 
1415 /*
1416  * XXX TODO: which subsystem, order?
1417  */
1418 DECLARE_MODULE(acl_nfs4, acl_nfs4_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1419 MODULE_VERSION(acl_nfs4, 1);
1420 #endif	/* _KERNEL */
1421