xref: /freebsd/sys/kern/subr_acl_nfs4.c (revision 3ef51c5fb9163f2aafb1c14729e06a8bf0c4d113)
1 /*-
2  * Copyright (c) 2008-2010 Edward Tomasz Napierała <trasz@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 /*
28  * ACL support routines specific to NFSv4 access control lists.  These are
29  * utility routines for code common across file systems implementing NFSv4
30  * ACLs.
31  */
32 
33 #ifdef _KERNEL
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include <sys/param.h>
38 #include <sys/kernel.h>
39 #include <sys/module.h>
40 #include <sys/systm.h>
41 #include <sys/mount.h>
42 #include <sys/priv.h>
43 #include <sys/vnode.h>
44 #include <sys/errno.h>
45 #include <sys/stat.h>
46 #include <sys/sysctl.h>
47 #include <sys/acl.h>
48 #else
49 #include <errno.h>
50 #include <assert.h>
51 #include <sys/acl.h>
52 #include <sys/stat.h>
53 #define KASSERT(a, b) assert(a)
54 #define CTASSERT(a)
55 
56 #endif /* !_KERNEL */
57 
58 #ifdef _KERNEL
59 
60 static void	acl_nfs4_trivial_from_mode(struct acl *aclp, mode_t mode);
61 
62 static int	acl_nfs4_old_semantics = 0;
63 
64 SYSCTL_INT(_vfs, OID_AUTO, acl_nfs4_old_semantics, CTLFLAG_RW,
65     &acl_nfs4_old_semantics, 0, "Use pre-PSARC/2010/029 NFSv4 ACL semantics");
66 
67 static struct {
68 	accmode_t accmode;
69 	int mask;
70 } accmode2mask[] = {{VREAD, ACL_READ_DATA},
71 		    {VWRITE, ACL_WRITE_DATA},
72 		    {VAPPEND, ACL_APPEND_DATA},
73 		    {VEXEC, ACL_EXECUTE},
74 		    {VREAD_NAMED_ATTRS, ACL_READ_NAMED_ATTRS},
75 		    {VWRITE_NAMED_ATTRS, ACL_WRITE_NAMED_ATTRS},
76 		    {VDELETE_CHILD, ACL_DELETE_CHILD},
77 		    {VREAD_ATTRIBUTES, ACL_READ_ATTRIBUTES},
78 		    {VWRITE_ATTRIBUTES, ACL_WRITE_ATTRIBUTES},
79 		    {VDELETE, ACL_DELETE},
80 		    {VREAD_ACL, ACL_READ_ACL},
81 		    {VWRITE_ACL, ACL_WRITE_ACL},
82 		    {VWRITE_OWNER, ACL_WRITE_OWNER},
83 		    {VSYNCHRONIZE, ACL_SYNCHRONIZE},
84 		    {0, 0}};
85 
86 static int
87 _access_mask_from_accmode(accmode_t accmode)
88 {
89 	int access_mask = 0, i;
90 
91 	for (i = 0; accmode2mask[i].accmode != 0; i++) {
92 		if (accmode & accmode2mask[i].accmode)
93 			access_mask |= accmode2mask[i].mask;
94 	}
95 
96 	/*
97 	 * VAPPEND is just a modifier for VWRITE; if the caller asked
98 	 * for 'VAPPEND | VWRITE', we want to check for ACL_APPEND_DATA only.
99 	 */
100 	if (access_mask & ACL_APPEND_DATA)
101 		access_mask &= ~ACL_WRITE_DATA;
102 
103 	return (access_mask);
104 }
105 
106 /*
107  * Return 0, iff access is allowed, 1 otherwise.
108  */
109 static int
110 _acl_denies(const struct acl *aclp, int access_mask, struct ucred *cred,
111     int file_uid, int file_gid, int *denied_explicitly)
112 {
113 	int i;
114 	const struct acl_entry *entry;
115 
116 	if (denied_explicitly != NULL)
117 		*denied_explicitly = 0;
118 
119 	KASSERT(aclp->acl_cnt <= ACL_MAX_ENTRIES,
120 	    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
121 
122 	for (i = 0; i < aclp->acl_cnt; i++) {
123 		entry = &(aclp->acl_entry[i]);
124 
125 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
126 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
127 			continue;
128 		if (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY)
129 			continue;
130 		switch (entry->ae_tag) {
131 		case ACL_USER_OBJ:
132 			if (file_uid != cred->cr_uid)
133 				continue;
134 			break;
135 		case ACL_USER:
136 			if (entry->ae_id != cred->cr_uid)
137 				continue;
138 			break;
139 		case ACL_GROUP_OBJ:
140 			if (!groupmember(file_gid, cred))
141 				continue;
142 			break;
143 		case ACL_GROUP:
144 			if (!groupmember(entry->ae_id, cred))
145 				continue;
146 			break;
147 		default:
148 			KASSERT(entry->ae_tag == ACL_EVERYONE,
149 			    ("entry->ae_tag == ACL_EVERYONE"));
150 		}
151 
152 		if (entry->ae_entry_type == ACL_ENTRY_TYPE_DENY) {
153 			if (entry->ae_perm & access_mask) {
154 				if (denied_explicitly != NULL)
155 					*denied_explicitly = 1;
156 				return (1);
157 			}
158 		}
159 
160 		access_mask &= ~(entry->ae_perm);
161 		if (access_mask == 0)
162 			return (0);
163 	}
164 
165 	return (1);
166 }
167 
168 int
169 vaccess_acl_nfs4(enum vtype type, uid_t file_uid, gid_t file_gid,
170     struct acl *aclp, accmode_t accmode, struct ucred *cred, int *privused)
171 {
172 	accmode_t priv_granted = 0;
173 	int denied, explicitly_denied, access_mask, is_directory,
174 	    must_be_owner = 0;
175 	mode_t file_mode = 0;
176 
177 	KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND |
178 	    VEXPLICIT_DENY | VREAD_NAMED_ATTRS | VWRITE_NAMED_ATTRS |
179 	    VDELETE_CHILD | VREAD_ATTRIBUTES | VWRITE_ATTRIBUTES | VDELETE |
180 	    VREAD_ACL | VWRITE_ACL | VWRITE_OWNER | VSYNCHRONIZE)) == 0,
181 	    ("invalid bit in accmode"));
182 	KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
183 	    	("VAPPEND without VWRITE"));
184 
185 	if (privused != NULL)
186 		*privused = 0;
187 
188 	if (accmode & VADMIN)
189 		must_be_owner = 1;
190 
191 	/*
192 	 * Ignore VSYNCHRONIZE permission.
193 	 */
194 	accmode &= ~VSYNCHRONIZE;
195 
196 	access_mask = _access_mask_from_accmode(accmode);
197 
198 	if (type == VDIR)
199 		is_directory = 1;
200 	else
201 		is_directory = 0;
202 
203 	/*
204 	 * File owner is always allowed to read and write the ACL
205 	 * and basic attributes.  This is to prevent a situation
206 	 * where user would change ACL in a way that prevents him
207 	 * from undoing the change.
208 	 */
209 	if (file_uid == cred->cr_uid)
210 		access_mask &= ~(ACL_READ_ACL | ACL_WRITE_ACL |
211 		    ACL_READ_ATTRIBUTES | ACL_WRITE_ATTRIBUTES);
212 
213 	/*
214 	 * Ignore append permission for regular files; use write
215 	 * permission instead.
216 	 */
217 	if (!is_directory && (access_mask & ACL_APPEND_DATA)) {
218 		access_mask &= ~ACL_APPEND_DATA;
219 		access_mask |= ACL_WRITE_DATA;
220 	}
221 
222 	denied = _acl_denies(aclp, access_mask, cred, file_uid, file_gid,
223 	    &explicitly_denied);
224 
225 	if (must_be_owner) {
226 		if (file_uid != cred->cr_uid)
227 			denied = EPERM;
228 	}
229 
230 	/*
231 	 * For VEXEC, ensure that at least one execute bit is set for
232 	 * non-directories. We have to check the mode here to stay
233 	 * consistent with execve(2). See the test in
234 	 * exec_check_permissions().
235 	 */
236 	acl_nfs4_sync_mode_from_acl(&file_mode, aclp);
237 	if (!denied && !is_directory && (accmode & VEXEC) &&
238 	    (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)
239 		denied = EACCES;
240 
241 	if (!denied)
242 		return (0);
243 
244 	/*
245 	 * Access failed.  Iff it was not denied explicitly and
246 	 * VEXPLICIT_DENY flag was specified, allow access.
247 	 */
248 	if ((accmode & VEXPLICIT_DENY) && explicitly_denied == 0)
249 		return (0);
250 
251 	accmode &= ~VEXPLICIT_DENY;
252 
253 	/*
254 	 * No match.  Try to use privileges, if there are any.
255 	 */
256 	if (is_directory) {
257 		if ((accmode & VEXEC) && !priv_check_cred(cred,
258 		    PRIV_VFS_LOOKUP, 0))
259 			priv_granted |= VEXEC;
260 	} else {
261 		/*
262 		 * Ensure that at least one execute bit is on. Otherwise,
263 		 * a privileged user will always succeed, and we don't want
264 		 * this to happen unless the file really is executable.
265 		 */
266 		if ((accmode & VEXEC) && (file_mode &
267 		    (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
268 		    !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
269 			priv_granted |= VEXEC;
270 	}
271 
272 	if ((accmode & VREAD) && !priv_check_cred(cred, PRIV_VFS_READ, 0))
273 		priv_granted |= VREAD;
274 
275 	if ((accmode & (VWRITE | VAPPEND | VDELETE_CHILD)) &&
276 	    !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
277 		priv_granted |= (VWRITE | VAPPEND | VDELETE_CHILD);
278 
279 	if ((accmode & VADMIN_PERMS) &&
280 	    !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
281 		priv_granted |= VADMIN_PERMS;
282 
283 	if ((accmode & VSTAT_PERMS) &&
284 	    !priv_check_cred(cred, PRIV_VFS_STAT, 0))
285 		priv_granted |= VSTAT_PERMS;
286 
287 	if ((accmode & priv_granted) == accmode) {
288 		if (privused != NULL)
289 			*privused = 1;
290 
291 		return (0);
292 	}
293 
294 	if (accmode & (VADMIN_PERMS | VDELETE_CHILD | VDELETE))
295 		denied = EPERM;
296 	else
297 		denied = EACCES;
298 
299 	return (denied);
300 }
301 #endif /* _KERNEL */
302 
303 static int
304 _acl_entry_matches(struct acl_entry *entry, acl_tag_t tag, acl_perm_t perm,
305     acl_entry_type_t entry_type)
306 {
307 	if (entry->ae_tag != tag)
308 		return (0);
309 
310 	if (entry->ae_id != ACL_UNDEFINED_ID)
311 		return (0);
312 
313 	if (entry->ae_perm != perm)
314 		return (0);
315 
316 	if (entry->ae_entry_type != entry_type)
317 		return (0);
318 
319 	if (entry->ae_flags != 0)
320 		return (0);
321 
322 	return (1);
323 }
324 
325 static struct acl_entry *
326 _acl_append(struct acl *aclp, acl_tag_t tag, acl_perm_t perm,
327     acl_entry_type_t entry_type)
328 {
329 	struct acl_entry *entry;
330 
331 	KASSERT(aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
332 	    ("aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
333 
334 	entry = &(aclp->acl_entry[aclp->acl_cnt]);
335 	aclp->acl_cnt++;
336 
337 	entry->ae_tag = tag;
338 	entry->ae_id = ACL_UNDEFINED_ID;
339 	entry->ae_perm = perm;
340 	entry->ae_entry_type = entry_type;
341 	entry->ae_flags = 0;
342 
343 	return (entry);
344 }
345 
346 static struct acl_entry *
347 _acl_duplicate_entry(struct acl *aclp, int entry_index)
348 {
349 	int i;
350 
351 	KASSERT(aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
352 	    ("aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
353 
354 	for (i = aclp->acl_cnt; i > entry_index; i--)
355 		aclp->acl_entry[i] = aclp->acl_entry[i - 1];
356 
357 	aclp->acl_cnt++;
358 
359 	return (&(aclp->acl_entry[entry_index + 1]));
360 }
361 
362 static void
363 acl_nfs4_sync_acl_from_mode_draft(struct acl *aclp, mode_t mode,
364     int file_owner_id)
365 {
366 	int i, meets, must_append;
367 	struct acl_entry *entry, *copy, *previous,
368 	    *a1, *a2, *a3, *a4, *a5, *a6;
369 	mode_t amode;
370 	const int READ = 04;
371 	const int WRITE = 02;
372 	const int EXEC = 01;
373 
374 	KASSERT(aclp->acl_cnt <= ACL_MAX_ENTRIES,
375 	    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
376 
377 	/*
378 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
379 	 *
380 	 * 3.16.6.3. Applying a Mode to an Existing ACL
381 	 */
382 
383 	/*
384 	 * 1. For each ACE:
385 	 */
386 	for (i = 0; i < aclp->acl_cnt; i++) {
387 		entry = &(aclp->acl_entry[i]);
388 
389 		/*
390 		 * 1.1. If the type is neither ALLOW or DENY - skip.
391 		 */
392 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
393 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
394 			continue;
395 
396 		/*
397 		 * 1.2. If ACL_ENTRY_INHERIT_ONLY is set - skip.
398 		 */
399 		if (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY)
400 			continue;
401 
402 		/*
403 		 * 1.3. If ACL_ENTRY_FILE_INHERIT or ACL_ENTRY_DIRECTORY_INHERIT
404 		 *      are set:
405 		 */
406 		if (entry->ae_flags &
407 		    (ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT)) {
408 			/*
409 			 * 1.3.1. A copy of the current ACE is made, and placed
410 			 *        in the ACL immediately following the current
411 			 *        ACE.
412 			 */
413 			copy = _acl_duplicate_entry(aclp, i);
414 
415 			/*
416 			 * 1.3.2. In the first ACE, the flag
417 			 *        ACL_ENTRY_INHERIT_ONLY is set.
418 			 */
419 			entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
420 
421 			/*
422 			 * 1.3.3. In the second ACE, the following flags
423 			 *        are cleared:
424 			 *        ACL_ENTRY_FILE_INHERIT,
425 			 *        ACL_ENTRY_DIRECTORY_INHERIT,
426 			 *        ACL_ENTRY_NO_PROPAGATE_INHERIT.
427 			 */
428 			copy->ae_flags &= ~(ACL_ENTRY_FILE_INHERIT |
429 			    ACL_ENTRY_DIRECTORY_INHERIT |
430 			    ACL_ENTRY_NO_PROPAGATE_INHERIT);
431 
432 			/*
433 			 * The algorithm continues on with the second ACE.
434 			 */
435 			i++;
436 			entry = copy;
437 		}
438 
439 		/*
440 		 * 1.4. If it's owner@, group@ or everyone@ entry, clear
441 		 *      ACL_READ_DATA, ACL_WRITE_DATA, ACL_APPEND_DATA
442 		 *      and ACL_EXECUTE.  Continue to the next entry.
443 		 */
444 		if (entry->ae_tag == ACL_USER_OBJ ||
445 		    entry->ae_tag == ACL_GROUP_OBJ ||
446 		    entry->ae_tag == ACL_EVERYONE) {
447 			entry->ae_perm &= ~(ACL_READ_DATA | ACL_WRITE_DATA |
448 			    ACL_APPEND_DATA | ACL_EXECUTE);
449 			continue;
450 		}
451 
452 		/*
453 		 * 1.5. Otherwise, if the "who" field did not match one
454 		 *      of OWNER@, GROUP@, EVERYONE@:
455 		 *
456 		 * 1.5.1. If the type is ALLOW, check the preceding ACE.
457 		 *        If it does not meet all of the following criteria:
458 		 */
459 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW)
460 			continue;
461 
462 		meets = 0;
463 		if (i > 0) {
464 			meets = 1;
465 			previous = &(aclp->acl_entry[i - 1]);
466 
467 			/*
468 			 * 1.5.1.1. The type field is DENY,
469 			 */
470 			if (previous->ae_entry_type != ACL_ENTRY_TYPE_DENY)
471 				meets = 0;
472 
473 			/*
474 			 * 1.5.1.2. The "who" field is the same as the current
475 			 *          ACE,
476 			 *
477 			 * 1.5.1.3. The flag bit ACE4_IDENTIFIER_GROUP
478 			 *          is the same as it is in the current ACE,
479 			 *          and no other flag bits are set,
480 			 */
481 			if (previous->ae_id != entry->ae_id ||
482 			    previous->ae_tag != entry->ae_tag)
483 				meets = 0;
484 
485 			if (previous->ae_flags)
486 				meets = 0;
487 
488 			/*
489 			 * 1.5.1.4. The mask bits are a subset of the mask bits
490 			 *          of the current ACE, and are also subset of
491 			 *          the following: ACL_READ_DATA,
492 			 *          ACL_WRITE_DATA, ACL_APPEND_DATA, ACL_EXECUTE
493 			 */
494 			if (previous->ae_perm & ~(entry->ae_perm))
495 				meets = 0;
496 
497 			if (previous->ae_perm & ~(ACL_READ_DATA |
498 			    ACL_WRITE_DATA | ACL_APPEND_DATA | ACL_EXECUTE))
499 				meets = 0;
500 		}
501 
502 		if (!meets) {
503 			/*
504 		 	 * Then the ACE of type DENY, with a who equal
505 			 * to the current ACE, flag bits equal to
506 			 * (<current ACE flags> & <ACE_IDENTIFIER_GROUP>)
507 			 * and no mask bits, is prepended.
508 			 */
509 			previous = entry;
510 			entry = _acl_duplicate_entry(aclp, i);
511 
512 			/* Adjust counter, as we've just added an entry. */
513 			i++;
514 
515 			previous->ae_tag = entry->ae_tag;
516 			previous->ae_id = entry->ae_id;
517 			previous->ae_flags = entry->ae_flags;
518 			previous->ae_perm = 0;
519 			previous->ae_entry_type = ACL_ENTRY_TYPE_DENY;
520 		}
521 
522 		/*
523 		 * 1.5.2. The following modifications are made to the prepended
524 		 *        ACE.  The intent is to mask the following ACE
525 		 *        to disallow ACL_READ_DATA, ACL_WRITE_DATA,
526 		 *        ACL_APPEND_DATA, or ACL_EXECUTE, based upon the group
527 		 *        permissions of the new mode.  As a special case,
528 		 *        if the ACE matches the current owner of the file,
529 		 *        the owner bits are used, rather than the group bits.
530 		 *        This is reflected in the algorithm below.
531 		 */
532 		amode = mode >> 3;
533 
534 		/*
535 		 * If ACE4_IDENTIFIER_GROUP is not set, and the "who" field
536 		 * in ACE matches the owner of the file, we shift amode three
537 		 * more bits, in order to have the owner permission bits
538 		 * placed in the three low order bits of amode.
539 		 */
540 		if (entry->ae_tag == ACL_USER && entry->ae_id == file_owner_id)
541 			amode = amode >> 3;
542 
543 		if (entry->ae_perm & ACL_READ_DATA) {
544 			if (amode & READ)
545 				previous->ae_perm &= ~ACL_READ_DATA;
546 			else
547 				previous->ae_perm |= ACL_READ_DATA;
548 		}
549 
550 		if (entry->ae_perm & ACL_WRITE_DATA) {
551 			if (amode & WRITE)
552 				previous->ae_perm &= ~ACL_WRITE_DATA;
553 			else
554 				previous->ae_perm |= ACL_WRITE_DATA;
555 		}
556 
557 		if (entry->ae_perm & ACL_APPEND_DATA) {
558 			if (amode & WRITE)
559 				previous->ae_perm &= ~ACL_APPEND_DATA;
560 			else
561 				previous->ae_perm |= ACL_APPEND_DATA;
562 		}
563 
564 		if (entry->ae_perm & ACL_EXECUTE) {
565 			if (amode & EXEC)
566 				previous->ae_perm &= ~ACL_EXECUTE;
567 			else
568 				previous->ae_perm |= ACL_EXECUTE;
569 		}
570 
571 		/*
572 		 * 1.5.3. If ACE4_IDENTIFIER_GROUP is set in the flags
573 		 *        of the ALLOW ace:
574 		 *
575 		 * XXX: This point is not there in the Falkner's draft.
576 		 */
577 		if (entry->ae_tag == ACL_GROUP &&
578 		    entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW) {
579 			mode_t extramode, ownermode;
580 			extramode = (mode >> 3) & 07;
581 			ownermode = mode >> 6;
582 			extramode &= ~ownermode;
583 
584 			if (extramode) {
585 				if (extramode & READ) {
586 					entry->ae_perm &= ~ACL_READ_DATA;
587 					previous->ae_perm &= ~ACL_READ_DATA;
588 				}
589 
590 				if (extramode & WRITE) {
591 					entry->ae_perm &=
592 					    ~(ACL_WRITE_DATA | ACL_APPEND_DATA);
593 					previous->ae_perm &=
594 					    ~(ACL_WRITE_DATA | ACL_APPEND_DATA);
595 				}
596 
597 				if (extramode & EXEC) {
598 					entry->ae_perm &= ~ACL_EXECUTE;
599 					previous->ae_perm &= ~ACL_EXECUTE;
600 				}
601 			}
602 		}
603 	}
604 
605 	/*
606 	 * 2. If there at least six ACEs, the final six ACEs are examined.
607 	 *    If they are not equal to what we want, append six ACEs.
608 	 */
609 	must_append = 0;
610 	if (aclp->acl_cnt < 6) {
611 		must_append = 1;
612 	} else {
613 		a6 = &(aclp->acl_entry[aclp->acl_cnt - 1]);
614 		a5 = &(aclp->acl_entry[aclp->acl_cnt - 2]);
615 		a4 = &(aclp->acl_entry[aclp->acl_cnt - 3]);
616 		a3 = &(aclp->acl_entry[aclp->acl_cnt - 4]);
617 		a2 = &(aclp->acl_entry[aclp->acl_cnt - 5]);
618 		a1 = &(aclp->acl_entry[aclp->acl_cnt - 6]);
619 
620 		if (!_acl_entry_matches(a1, ACL_USER_OBJ, 0,
621 		    ACL_ENTRY_TYPE_DENY))
622 			must_append = 1;
623 		if (!_acl_entry_matches(a2, ACL_USER_OBJ, ACL_WRITE_ACL |
624 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
625 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_ALLOW))
626 			must_append = 1;
627 		if (!_acl_entry_matches(a3, ACL_GROUP_OBJ, 0,
628 		    ACL_ENTRY_TYPE_DENY))
629 			must_append = 1;
630 		if (!_acl_entry_matches(a4, ACL_GROUP_OBJ, 0,
631 		    ACL_ENTRY_TYPE_ALLOW))
632 			must_append = 1;
633 		if (!_acl_entry_matches(a5, ACL_EVERYONE, ACL_WRITE_ACL |
634 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
635 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_DENY))
636 			must_append = 1;
637 		if (!_acl_entry_matches(a6, ACL_EVERYONE, ACL_READ_ACL |
638 		    ACL_READ_ATTRIBUTES | ACL_READ_NAMED_ATTRS |
639 		    ACL_SYNCHRONIZE, ACL_ENTRY_TYPE_ALLOW))
640 			must_append = 1;
641 	}
642 
643 	if (must_append) {
644 		KASSERT(aclp->acl_cnt + 6 <= ACL_MAX_ENTRIES,
645 		    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
646 
647 		a1 = _acl_append(aclp, ACL_USER_OBJ, 0, ACL_ENTRY_TYPE_DENY);
648 		a2 = _acl_append(aclp, ACL_USER_OBJ, ACL_WRITE_ACL |
649 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
650 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_ALLOW);
651 		a3 = _acl_append(aclp, ACL_GROUP_OBJ, 0, ACL_ENTRY_TYPE_DENY);
652 		a4 = _acl_append(aclp, ACL_GROUP_OBJ, 0, ACL_ENTRY_TYPE_ALLOW);
653 		a5 = _acl_append(aclp, ACL_EVERYONE, ACL_WRITE_ACL |
654 		    ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
655 		    ACL_WRITE_NAMED_ATTRS, ACL_ENTRY_TYPE_DENY);
656 		a6 = _acl_append(aclp, ACL_EVERYONE, ACL_READ_ACL |
657 		    ACL_READ_ATTRIBUTES | ACL_READ_NAMED_ATTRS |
658 		    ACL_SYNCHRONIZE, ACL_ENTRY_TYPE_ALLOW);
659 
660 		KASSERT(a1 != NULL && a2 != NULL && a3 != NULL && a4 != NULL &&
661 		    a5 != NULL && a6 != NULL, ("couldn't append to ACL."));
662 	}
663 
664 	/*
665 	 * 3. The final six ACEs are adjusted according to the incoming mode.
666 	 */
667 	if (mode & S_IRUSR)
668 		a2->ae_perm |= ACL_READ_DATA;
669 	else
670 		a1->ae_perm |= ACL_READ_DATA;
671 	if (mode & S_IWUSR)
672 		a2->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
673 	else
674 		a1->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
675 	if (mode & S_IXUSR)
676 		a2->ae_perm |= ACL_EXECUTE;
677 	else
678 		a1->ae_perm |= ACL_EXECUTE;
679 
680 	if (mode & S_IRGRP)
681 		a4->ae_perm |= ACL_READ_DATA;
682 	else
683 		a3->ae_perm |= ACL_READ_DATA;
684 	if (mode & S_IWGRP)
685 		a4->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
686 	else
687 		a3->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
688 	if (mode & S_IXGRP)
689 		a4->ae_perm |= ACL_EXECUTE;
690 	else
691 		a3->ae_perm |= ACL_EXECUTE;
692 
693 	if (mode & S_IROTH)
694 		a6->ae_perm |= ACL_READ_DATA;
695 	else
696 		a5->ae_perm |= ACL_READ_DATA;
697 	if (mode & S_IWOTH)
698 		a6->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
699 	else
700 		a5->ae_perm |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
701 	if (mode & S_IXOTH)
702 		a6->ae_perm |= ACL_EXECUTE;
703 	else
704 		a5->ae_perm |= ACL_EXECUTE;
705 }
706 
707 #ifdef _KERNEL
708 void
709 acl_nfs4_sync_acl_from_mode(struct acl *aclp, mode_t mode,
710     int file_owner_id)
711 {
712 
713 	if (acl_nfs4_old_semantics)
714 		acl_nfs4_sync_acl_from_mode_draft(aclp, mode, file_owner_id);
715 	else
716 		acl_nfs4_trivial_from_mode(aclp, mode);
717 }
718 #endif /* _KERNEL */
719 
720 void
721 acl_nfs4_sync_mode_from_acl(mode_t *_mode, const struct acl *aclp)
722 {
723 	int i;
724 	mode_t old_mode = *_mode, mode = 0, seen = 0;
725 	const struct acl_entry *entry;
726 
727 	KASSERT(aclp->acl_cnt <= ACL_MAX_ENTRIES,
728 	    ("aclp->acl_cnt <= ACL_MAX_ENTRIES"));
729 
730 	/*
731 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
732 	 *
733 	 * 3.16.6.1. Recomputing mode upon SETATTR of ACL
734 	 */
735 
736 	for (i = 0; i < aclp->acl_cnt; i++) {
737 		entry = &(aclp->acl_entry[i]);
738 
739 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
740 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
741 			continue;
742 
743 		if (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY)
744 			continue;
745 
746 		if (entry->ae_tag == ACL_USER_OBJ) {
747 			if ((entry->ae_perm & ACL_READ_DATA) &&
748 			    ((seen & S_IRUSR) == 0)) {
749 				seen |= S_IRUSR;
750 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
751 					mode |= S_IRUSR;
752 			}
753 			if ((entry->ae_perm & ACL_WRITE_DATA) &&
754 			     ((seen & S_IWUSR) == 0)) {
755 				seen |= S_IWUSR;
756 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
757 					mode |= S_IWUSR;
758 			}
759 			if ((entry->ae_perm & ACL_EXECUTE) &&
760 			    ((seen & S_IXUSR) == 0)) {
761 				seen |= S_IXUSR;
762 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
763 					mode |= S_IXUSR;
764 			}
765 		} else if (entry->ae_tag == ACL_GROUP_OBJ) {
766 			if ((entry->ae_perm & ACL_READ_DATA) &&
767 			    ((seen & S_IRGRP) == 0)) {
768 				seen |= S_IRGRP;
769 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
770 					mode |= S_IRGRP;
771 			}
772 			if ((entry->ae_perm & ACL_WRITE_DATA) &&
773 			    ((seen & S_IWGRP) == 0)) {
774 				seen |= S_IWGRP;
775 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
776 					mode |= S_IWGRP;
777 			}
778 			if ((entry->ae_perm & ACL_EXECUTE) &&
779 			    ((seen & S_IXGRP) == 0)) {
780 				seen |= S_IXGRP;
781 				if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
782 					mode |= S_IXGRP;
783 			}
784 		} else if (entry->ae_tag == ACL_EVERYONE) {
785 			if (entry->ae_perm & ACL_READ_DATA) {
786 				if ((seen & S_IRUSR) == 0) {
787 					seen |= S_IRUSR;
788 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
789 						mode |= S_IRUSR;
790 				}
791 				if ((seen & S_IRGRP) == 0) {
792 					seen |= S_IRGRP;
793 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
794 						mode |= S_IRGRP;
795 				}
796 				if ((seen & S_IROTH) == 0) {
797 					seen |= S_IROTH;
798 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
799 						mode |= S_IROTH;
800 				}
801 			}
802 			if (entry->ae_perm & ACL_WRITE_DATA) {
803 				if ((seen & S_IWUSR) == 0) {
804 					seen |= S_IWUSR;
805 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
806 						mode |= S_IWUSR;
807 				}
808 				if ((seen & S_IWGRP) == 0) {
809 					seen |= S_IWGRP;
810 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
811 						mode |= S_IWGRP;
812 				}
813 				if ((seen & S_IWOTH) == 0) {
814 					seen |= S_IWOTH;
815 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
816 						mode |= S_IWOTH;
817 				}
818 			}
819 			if (entry->ae_perm & ACL_EXECUTE) {
820 				if ((seen & S_IXUSR) == 0) {
821 					seen |= S_IXUSR;
822 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
823 						mode |= S_IXUSR;
824 				}
825 				if ((seen & S_IXGRP) == 0) {
826 					seen |= S_IXGRP;
827 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
828 						mode |= S_IXGRP;
829 				}
830 				if ((seen & S_IXOTH) == 0) {
831 					seen |= S_IXOTH;
832 					if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
833 						mode |= S_IXOTH;
834 				}
835 			}
836 		}
837 	}
838 
839 	*_mode = mode | (old_mode & ACL_PRESERVE_MASK);
840 }
841 
842 #ifdef _KERNEL
843 /*
844  * Calculate inherited ACL in a manner compatible with NFSv4 Minor Version 1,
845  * draft-ietf-nfsv4-minorversion1-03.txt.
846  */
847 static void
848 acl_nfs4_compute_inherited_acl_draft(const struct acl *parent_aclp,
849     struct acl *child_aclp, mode_t mode, int file_owner_id,
850     int is_directory)
851 {
852 	int i, flags;
853 	const struct acl_entry *parent_entry;
854 	struct acl_entry *entry, *copy;
855 
856 	KASSERT(child_aclp->acl_cnt == 0, ("child_aclp->acl_cnt == 0"));
857 	KASSERT(parent_aclp->acl_cnt <= ACL_MAX_ENTRIES,
858 	    ("parent_aclp->acl_cnt <= ACL_MAX_ENTRIES"));
859 
860 	/*
861 	 * NFSv4 Minor Version 1, draft-ietf-nfsv4-minorversion1-03.txt
862 	 *
863 	 * 3.16.6.2. Applying the mode given to CREATE or OPEN
864 	 *           to an inherited ACL
865 	 */
866 
867 	/*
868 	 * 1. Form an ACL that is the concatenation of all inheritable ACEs.
869 	 */
870 	for (i = 0; i < parent_aclp->acl_cnt; i++) {
871 		parent_entry = &(parent_aclp->acl_entry[i]);
872 		flags = parent_entry->ae_flags;
873 
874 		/*
875 		 * Entry is not inheritable at all.
876 		 */
877 		if ((flags & (ACL_ENTRY_DIRECTORY_INHERIT |
878 		    ACL_ENTRY_FILE_INHERIT)) == 0)
879 			continue;
880 
881 		/*
882 		 * We're creating a file, but entry is not inheritable
883 		 * by files.
884 		 */
885 		if (!is_directory && (flags & ACL_ENTRY_FILE_INHERIT) == 0)
886 			continue;
887 
888 		/*
889 		 * Entry is inheritable only by files, but has NO_PROPAGATE
890 		 * flag set, and we're creating a directory, so it wouldn't
891 		 * propagate to any file in that directory anyway.
892 		 */
893 		if (is_directory &&
894 		    (flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0 &&
895 		    (flags & ACL_ENTRY_NO_PROPAGATE_INHERIT))
896 			continue;
897 
898 		KASSERT(child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
899 		    ("child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
900 		child_aclp->acl_entry[child_aclp->acl_cnt] = *parent_entry;
901 		child_aclp->acl_cnt++;
902 	}
903 
904 	/*
905 	 * 2. For each entry in the new ACL, adjust its flags, possibly
906 	 *    creating two entries in place of one.
907 	 */
908 	for (i = 0; i < child_aclp->acl_cnt; i++) {
909 		entry = &(child_aclp->acl_entry[i]);
910 
911 		/*
912 		 * This is not in the specification, but SunOS
913 		 * apparently does that.
914 		 */
915 		if (((entry->ae_flags & ACL_ENTRY_NO_PROPAGATE_INHERIT) ||
916 		    !is_directory) &&
917 		    entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
918 			entry->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER);
919 
920 		/*
921 		 * 2.A. If the ACL_ENTRY_NO_PROPAGATE_INHERIT is set, or if the object
922 		 *      being created is not a directory, then clear the
923 		 *      following flags: ACL_ENTRY_NO_PROPAGATE_INHERIT,
924 		 *      ACL_ENTRY_FILE_INHERIT, ACL_ENTRY_DIRECTORY_INHERIT,
925 		 *      ACL_ENTRY_INHERIT_ONLY.
926 		 */
927 		if (entry->ae_flags & ACL_ENTRY_NO_PROPAGATE_INHERIT ||
928 		    !is_directory) {
929 			entry->ae_flags &= ~(ACL_ENTRY_NO_PROPAGATE_INHERIT |
930 			ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT |
931 			ACL_ENTRY_INHERIT_ONLY);
932 
933 			/*
934 			 * Continue on to the next ACE.
935 			 */
936 			continue;
937 		}
938 
939 		/*
940 		 * 2.B. If the object is a directory and ACL_ENTRY_FILE_INHERIT
941 		 *      is set, but ACL_ENTRY_NO_PROPAGATE_INHERIT is not set, ensure
942 		 *      that ACL_ENTRY_INHERIT_ONLY is set.  Continue to the
943 		 *      next ACE.  Otherwise...
944 		 */
945 		/*
946 		 * XXX: Read it again and make sure what does the "otherwise"
947 		 *      apply to.
948 		 */
949 		if (is_directory &&
950 		    (entry->ae_flags & ACL_ENTRY_FILE_INHERIT) &&
951 		    ((entry->ae_flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0)) {
952 			entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
953 			continue;
954 		}
955 
956 		/*
957 		 * 2.C. If the type of the ACE is neither ALLOW nor deny,
958 		 *      then continue.
959 		 */
960 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
961 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
962 			continue;
963 
964 		/*
965 		 * 2.D. Copy the original ACE into a second, adjacent ACE.
966 		 */
967 		copy = _acl_duplicate_entry(child_aclp, i);
968 
969 		/*
970 		 * 2.E. On the first ACE, ensure that ACL_ENTRY_INHERIT_ONLY
971 		 *      is set.
972 		 */
973 		entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
974 
975 		/*
976 		 * 2.F. On the second ACE, clear the following flags:
977 		 *      ACL_ENTRY_NO_PROPAGATE_INHERIT, ACL_ENTRY_FILE_INHERIT,
978 		 *      ACL_ENTRY_DIRECTORY_INHERIT, ACL_ENTRY_INHERIT_ONLY.
979 		 */
980 		copy->ae_flags &= ~(ACL_ENTRY_NO_PROPAGATE_INHERIT |
981 		    ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT |
982 		    ACL_ENTRY_INHERIT_ONLY);
983 
984 		/*
985 		 * 2.G. On the second ACE, if the type is ALLOW,
986 		 *      an implementation MAY clear the following
987 		 *      mask bits: ACL_WRITE_ACL, ACL_WRITE_OWNER.
988 		 */
989 		if (copy->ae_entry_type == ACL_ENTRY_TYPE_ALLOW)
990 			copy->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER);
991 
992 		/*
993 		 * Increment the counter to skip the copied entry.
994 		 */
995 		i++;
996 	}
997 
998 	/*
999 	 * 3. To ensure that the mode is honored, apply the algorithm describe
1000 	 *    in Section 2.16.6.3, using the mode that is to be used for file
1001 	 *    creation.
1002 	 */
1003 	acl_nfs4_sync_acl_from_mode(child_aclp, mode, file_owner_id);
1004 }
1005 #endif /* _KERNEL */
1006 
1007 /*
1008  * Populate the ACL with entries inherited from parent_aclp.
1009  */
1010 static void
1011 acl_nfs4_inherit_entries(const struct acl *parent_aclp,
1012     struct acl *child_aclp, mode_t mode, int file_owner_id,
1013     int is_directory)
1014 {
1015 	int i, flags, tag;
1016 	const struct acl_entry *parent_entry;
1017 	struct acl_entry *entry;
1018 
1019 	KASSERT(parent_aclp->acl_cnt <= ACL_MAX_ENTRIES,
1020 	    ("parent_aclp->acl_cnt <= ACL_MAX_ENTRIES"));
1021 
1022 	for (i = 0; i < parent_aclp->acl_cnt; i++) {
1023 		parent_entry = &(parent_aclp->acl_entry[i]);
1024 		flags = parent_entry->ae_flags;
1025 		tag = parent_entry->ae_tag;
1026 
1027 		/*
1028 		 * Don't inherit owner@, group@, or everyone@ entries.
1029 		 */
1030 		if (tag == ACL_USER_OBJ || tag == ACL_GROUP_OBJ ||
1031 		    tag == ACL_EVERYONE)
1032 			continue;
1033 
1034 		/*
1035 		 * Entry is not inheritable at all.
1036 		 */
1037 		if ((flags & (ACL_ENTRY_DIRECTORY_INHERIT |
1038 		    ACL_ENTRY_FILE_INHERIT)) == 0)
1039 			continue;
1040 
1041 		/*
1042 		 * We're creating a file, but entry is not inheritable
1043 		 * by files.
1044 		 */
1045 		if (!is_directory && (flags & ACL_ENTRY_FILE_INHERIT) == 0)
1046 			continue;
1047 
1048 		/*
1049 		 * Entry is inheritable only by files, but has NO_PROPAGATE
1050 		 * flag set, and we're creating a directory, so it wouldn't
1051 		 * propagate to any file in that directory anyway.
1052 		 */
1053 		if (is_directory &&
1054 		    (flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0 &&
1055 		    (flags & ACL_ENTRY_NO_PROPAGATE_INHERIT))
1056 			continue;
1057 
1058 		/*
1059 		 * Entry qualifies for being inherited.
1060 		 */
1061 		KASSERT(child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES,
1062 		    ("child_aclp->acl_cnt + 1 <= ACL_MAX_ENTRIES"));
1063 		entry = &(child_aclp->acl_entry[child_aclp->acl_cnt]);
1064 		*entry = *parent_entry;
1065 		child_aclp->acl_cnt++;
1066 
1067 		entry->ae_flags &= ~ACL_ENTRY_INHERIT_ONLY;
1068 
1069 		/*
1070 		 * If the type of the ACE is neither ALLOW nor DENY,
1071 		 * then leave it as it is and proceed to the next one.
1072 		 */
1073 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
1074 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
1075 			continue;
1076 
1077 		/*
1078 		 * If the ACL_ENTRY_NO_PROPAGATE_INHERIT is set, or if
1079 		 * the object being created is not a directory, then clear
1080 		 * the following flags: ACL_ENTRY_NO_PROPAGATE_INHERIT,
1081 		 * ACL_ENTRY_FILE_INHERIT, ACL_ENTRY_DIRECTORY_INHERIT,
1082 		 * ACL_ENTRY_INHERIT_ONLY.
1083 		 */
1084 		if (entry->ae_flags & ACL_ENTRY_NO_PROPAGATE_INHERIT ||
1085 		    !is_directory) {
1086 			entry->ae_flags &= ~(ACL_ENTRY_NO_PROPAGATE_INHERIT |
1087 			ACL_ENTRY_FILE_INHERIT | ACL_ENTRY_DIRECTORY_INHERIT |
1088 			ACL_ENTRY_INHERIT_ONLY);
1089 		}
1090 
1091 		/*
1092 		 * If the object is a directory and ACL_ENTRY_FILE_INHERIT
1093 		 * is set, but ACL_ENTRY_DIRECTORY_INHERIT is not set, ensure
1094 		 * that ACL_ENTRY_INHERIT_ONLY is set.
1095 		 */
1096 		if (is_directory &&
1097 		    (entry->ae_flags & ACL_ENTRY_FILE_INHERIT) &&
1098 		    ((entry->ae_flags & ACL_ENTRY_DIRECTORY_INHERIT) == 0)) {
1099 			entry->ae_flags |= ACL_ENTRY_INHERIT_ONLY;
1100 		}
1101 
1102 		if (entry->ae_entry_type == ACL_ENTRY_TYPE_ALLOW &&
1103 		    (entry->ae_flags & ACL_ENTRY_INHERIT_ONLY) == 0) {
1104 			/*
1105 			 * Some permissions must never be inherited.
1106 			 */
1107 			entry->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER |
1108 			    ACL_WRITE_NAMED_ATTRS | ACL_WRITE_ATTRIBUTES);
1109 
1110 			/*
1111 			 * Others must be masked according to the file mode.
1112 			 */
1113 			if ((mode & S_IRGRP) == 0)
1114 				entry->ae_perm &= ~ACL_READ_DATA;
1115 			if ((mode & S_IWGRP) == 0)
1116 				entry->ae_perm &=
1117 				    ~(ACL_WRITE_DATA | ACL_APPEND_DATA);
1118 			if ((mode & S_IXGRP) == 0)
1119 				entry->ae_perm &= ~ACL_EXECUTE;
1120 		}
1121 	}
1122 }
1123 
1124 /*
1125  * Calculate inherited ACL in a manner compatible with PSARC/2010/029.
1126  * It's also being used to calculate a trivial ACL, by inheriting from
1127  * a NULL ACL.
1128  */
1129 static void
1130 acl_nfs4_compute_inherited_acl_psarc(const struct acl *parent_aclp,
1131     struct acl *aclp, mode_t mode, int file_owner_id, int is_directory)
1132 {
1133 	acl_perm_t user_allow_first = 0, user_deny = 0, group_deny = 0;
1134 	acl_perm_t user_allow, group_allow, everyone_allow;
1135 
1136 	KASSERT(aclp->acl_cnt == 0, ("aclp->acl_cnt == 0"));
1137 
1138 	user_allow = group_allow = everyone_allow = ACL_READ_ACL |
1139 	    ACL_READ_ATTRIBUTES | ACL_READ_NAMED_ATTRS | ACL_SYNCHRONIZE;
1140 	user_allow |= ACL_WRITE_ACL | ACL_WRITE_OWNER | ACL_WRITE_ATTRIBUTES |
1141 	    ACL_WRITE_NAMED_ATTRS;
1142 
1143 	if (mode & S_IRUSR)
1144 		user_allow |= ACL_READ_DATA;
1145 	if (mode & S_IWUSR)
1146 		user_allow |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
1147 	if (mode & S_IXUSR)
1148 		user_allow |= ACL_EXECUTE;
1149 
1150 	if (mode & S_IRGRP)
1151 		group_allow |= ACL_READ_DATA;
1152 	if (mode & S_IWGRP)
1153 		group_allow |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
1154 	if (mode & S_IXGRP)
1155 		group_allow |= ACL_EXECUTE;
1156 
1157 	if (mode & S_IROTH)
1158 		everyone_allow |= ACL_READ_DATA;
1159 	if (mode & S_IWOTH)
1160 		everyone_allow |= (ACL_WRITE_DATA | ACL_APPEND_DATA);
1161 	if (mode & S_IXOTH)
1162 		everyone_allow |= ACL_EXECUTE;
1163 
1164 	user_deny = ((group_allow | everyone_allow) & ~user_allow);
1165 	group_deny = everyone_allow & ~group_allow;
1166 	user_allow_first = group_deny & ~user_deny;
1167 
1168 	if (user_allow_first != 0)
1169 		_acl_append(aclp, ACL_USER_OBJ, user_allow_first,
1170 		    ACL_ENTRY_TYPE_ALLOW);
1171 	if (user_deny != 0)
1172 		_acl_append(aclp, ACL_USER_OBJ, user_deny,
1173 		    ACL_ENTRY_TYPE_DENY);
1174 	if (group_deny != 0)
1175 		_acl_append(aclp, ACL_GROUP_OBJ, group_deny,
1176 		    ACL_ENTRY_TYPE_DENY);
1177 
1178 	if (parent_aclp != NULL)
1179 		acl_nfs4_inherit_entries(parent_aclp, aclp, mode,
1180 		    file_owner_id, is_directory);
1181 
1182 	_acl_append(aclp, ACL_USER_OBJ, user_allow, ACL_ENTRY_TYPE_ALLOW);
1183 	_acl_append(aclp, ACL_GROUP_OBJ, group_allow, ACL_ENTRY_TYPE_ALLOW);
1184 	_acl_append(aclp, ACL_EVERYONE, everyone_allow, ACL_ENTRY_TYPE_ALLOW);
1185 }
1186 
1187 #ifdef _KERNEL
1188 void
1189 acl_nfs4_compute_inherited_acl(const struct acl *parent_aclp,
1190     struct acl *child_aclp, mode_t mode, int file_owner_id,
1191     int is_directory)
1192 {
1193 
1194 	if (acl_nfs4_old_semantics)
1195 		acl_nfs4_compute_inherited_acl_draft(parent_aclp, child_aclp,
1196 		    mode, file_owner_id, is_directory);
1197 	else
1198 		acl_nfs4_compute_inherited_acl_psarc(parent_aclp, child_aclp,
1199 		    mode, file_owner_id, is_directory);
1200 }
1201 #endif /* _KERNEL */
1202 
1203 /*
1204  * Calculate trivial ACL in a manner compatible with PSARC/2010/029.
1205  * Note that this results in an ACL different from (but semantically
1206  * equal to) the "canonical six" trivial ACL computed using algorithm
1207  * described in draft-ietf-nfsv4-minorversion1-03.txt, 3.16.6.2.
1208  */
1209 static void
1210 acl_nfs4_trivial_from_mode(struct acl *aclp, mode_t mode)
1211 {
1212 
1213 	aclp->acl_cnt = 0;
1214 	acl_nfs4_compute_inherited_acl_psarc(NULL, aclp, mode, -1, -1);
1215 }
1216 
1217 #ifndef _KERNEL
1218 /*
1219  * This routine is used by libc to implement acl_strip_np(3)
1220  * and acl_is_trivial_np(3).
1221  */
1222 void
1223 acl_nfs4_trivial_from_mode_libc(struct acl *aclp, int mode, int canonical_six)
1224 {
1225 
1226 	aclp->acl_cnt = 0;
1227 	if (canonical_six)
1228 		acl_nfs4_sync_acl_from_mode_draft(aclp, mode, -1);
1229 	else
1230 		acl_nfs4_trivial_from_mode(aclp, mode);
1231 }
1232 #endif /* !_KERNEL */
1233 
1234 #ifdef _KERNEL
1235 static int
1236 _acls_are_equal(const struct acl *a, const struct acl *b)
1237 {
1238 	int i;
1239 	const struct acl_entry *entrya, *entryb;
1240 
1241 	if (a->acl_cnt != b->acl_cnt)
1242 		return (0);
1243 
1244 	for (i = 0; i < b->acl_cnt; i++) {
1245 		entrya = &(a->acl_entry[i]);
1246 		entryb = &(b->acl_entry[i]);
1247 
1248 		if (entrya->ae_tag != entryb->ae_tag ||
1249 		    entrya->ae_id != entryb->ae_id ||
1250 		    entrya->ae_perm != entryb->ae_perm ||
1251 		    entrya->ae_entry_type != entryb->ae_entry_type ||
1252 		    entrya->ae_flags != entryb->ae_flags)
1253 			return (0);
1254 	}
1255 
1256 	return (1);
1257 }
1258 
1259 /*
1260  * This routine is used to determine whether to remove extended attribute
1261  * that stores ACL contents.
1262  */
1263 int
1264 acl_nfs4_is_trivial(const struct acl *aclp, int file_owner_id)
1265 {
1266 	int trivial;
1267 	mode_t tmpmode = 0;
1268 	struct acl *tmpaclp;
1269 
1270 	if (aclp->acl_cnt > 6)
1271 		return (0);
1272 
1273 	/*
1274 	 * Compute the mode from the ACL, then compute new ACL from that mode.
1275 	 * If the ACLs are identical, then the ACL is trivial.
1276 	 *
1277 	 * XXX: I guess there is a faster way to do this.  However, even
1278 	 *      this slow implementation significantly speeds things up
1279 	 *      for files that don't have non-trivial ACLs - it's critical
1280 	 *      for performance to not use EA when they are not needed.
1281 	 *
1282 	 * First try the PSARC/2010/029 semantics.
1283 	 */
1284 	tmpaclp = acl_alloc(M_WAITOK | M_ZERO);
1285 	acl_nfs4_sync_mode_from_acl(&tmpmode, aclp);
1286 	acl_nfs4_trivial_from_mode(tmpaclp, tmpmode);
1287 	trivial = _acls_are_equal(aclp, tmpaclp);
1288 	if (trivial) {
1289 		acl_free(tmpaclp);
1290 		return (trivial);
1291 	}
1292 
1293 	/*
1294 	 * Check if it's a draft-ietf-nfsv4-minorversion1-03.txt trivial ACL.
1295 	 */
1296 	tmpaclp->acl_cnt = 0;
1297 	acl_nfs4_sync_acl_from_mode_draft(tmpaclp, tmpmode, file_owner_id);
1298 	trivial = _acls_are_equal(aclp, tmpaclp);
1299 	acl_free(tmpaclp);
1300 
1301 	return (trivial);
1302 }
1303 #endif /* _KERNEL */
1304 
1305 int
1306 acl_nfs4_check(const struct acl *aclp, int is_directory)
1307 {
1308 	int i;
1309 	const struct acl_entry *entry;
1310 
1311 	/*
1312 	 * The spec doesn't seem to say anything about ACL validity.
1313 	 * It seems there is not much to do here.  There is even no need
1314 	 * to count "owner@" or "everyone@" (ACL_USER_OBJ and ACL_EVERYONE)
1315 	 * entries, as there can be several of them and that's perfectly
1316 	 * valid.  There can be none of them too.  Really.
1317 	 */
1318 
1319 	if (aclp->acl_cnt > ACL_MAX_ENTRIES || aclp->acl_cnt <= 0)
1320 		return (EINVAL);
1321 
1322 	for (i = 0; i < aclp->acl_cnt; i++) {
1323 		entry = &(aclp->acl_entry[i]);
1324 
1325 		switch (entry->ae_tag) {
1326 		case ACL_USER_OBJ:
1327 		case ACL_GROUP_OBJ:
1328 		case ACL_EVERYONE:
1329 			if (entry->ae_id != ACL_UNDEFINED_ID)
1330 				return (EINVAL);
1331 			break;
1332 
1333 		case ACL_USER:
1334 		case ACL_GROUP:
1335 			if (entry->ae_id == ACL_UNDEFINED_ID)
1336 				return (EINVAL);
1337 			break;
1338 
1339 		default:
1340 			return (EINVAL);
1341 		}
1342 
1343 		if ((entry->ae_perm | ACL_NFS4_PERM_BITS) != ACL_NFS4_PERM_BITS)
1344 			return (EINVAL);
1345 
1346 		/*
1347 		 * Disallow ACL_ENTRY_TYPE_AUDIT and ACL_ENTRY_TYPE_ALARM for now.
1348 		 */
1349 		if (entry->ae_entry_type != ACL_ENTRY_TYPE_ALLOW &&
1350 		    entry->ae_entry_type != ACL_ENTRY_TYPE_DENY)
1351 			return (EINVAL);
1352 
1353 		if ((entry->ae_flags | ACL_FLAGS_BITS) != ACL_FLAGS_BITS)
1354 			return (EINVAL);
1355 
1356 		/* Disallow unimplemented flags. */
1357 		if (entry->ae_flags & (ACL_ENTRY_SUCCESSFUL_ACCESS |
1358 		    ACL_ENTRY_FAILED_ACCESS))
1359 			return (EINVAL);
1360 
1361 		/* Disallow flags not allowed for ordinary files. */
1362 		if (!is_directory) {
1363 			if (entry->ae_flags & (ACL_ENTRY_FILE_INHERIT |
1364 			    ACL_ENTRY_DIRECTORY_INHERIT |
1365 			    ACL_ENTRY_NO_PROPAGATE_INHERIT | ACL_ENTRY_INHERIT_ONLY))
1366 				return (EINVAL);
1367 		}
1368 	}
1369 
1370 	return (0);
1371 }
1372 
1373 #ifdef	_KERNEL
1374 static int
1375 acl_nfs4_modload(module_t module, int what, void *arg)
1376 {
1377 	int ret;
1378 
1379 	ret = 0;
1380 
1381 	switch (what) {
1382 	case MOD_LOAD:
1383 	case MOD_SHUTDOWN:
1384 		break;
1385 
1386 	case MOD_QUIESCE:
1387 		/* XXX TODO */
1388 		ret = 0;
1389 		break;
1390 
1391 	case MOD_UNLOAD:
1392 		/* XXX TODO */
1393 		ret = 0;
1394 		break;
1395 	default:
1396 		ret = EINVAL;
1397 		break;
1398 	}
1399 
1400 	return (ret);
1401 }
1402 
1403 static moduledata_t acl_nfs4_mod = {
1404 	"acl_nfs4",
1405 	acl_nfs4_modload,
1406 	NULL
1407 };
1408 
1409 /*
1410  * XXX TODO: which subsystem, order?
1411  */
1412 DECLARE_MODULE(acl_nfs4, acl_nfs4_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1413 MODULE_VERSION(acl_nfs4, 1);
1414 #endif	/* _KERNEL */
1415