xref: /titanic_52/usr/src/uts/common/fs/smbsrv/smb_lock.c (revision 2bc987325e3ded1865bff043128661815c4690b9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /*
26  * This module provides range lock functionality for CIFS/SMB clients.
27  * Lock range service functions process SMB lock and and unlock
28  * requests for a file by applying lock rules and marks file range
29  * as locked if the lock is successful otherwise return proper
30  * error code.
31  */
32 
33 #include <smbsrv/smb_kproto.h>
34 #include <smbsrv/smb_fsops.h>
35 #include <sys/nbmlock.h>
36 #include <sys/param.h>
37 
38 extern caller_context_t smb_ct;
39 
40 static void smb_lock_posix_unlock(smb_node_t *, smb_lock_t *, cred_t *);
41 static boolean_t smb_is_range_unlocked(uint64_t, uint64_t, uint32_t,
42     smb_llist_t *, uint64_t *);
43 static int smb_lock_range_overlap(smb_lock_t *, uint64_t, uint64_t);
44 static uint32_t smb_lock_range_lckrules(smb_request_t *, smb_ofile_t *,
45     smb_node_t *, smb_lock_t *, smb_lock_t **);
46 static clock_t smb_lock_wait(smb_request_t *, smb_lock_t *, smb_lock_t *);
47 static uint32_t smb_lock_range_ulckrules(smb_request_t *, smb_node_t *,
48     uint64_t, uint64_t, smb_lock_t **nodelock);
49 static smb_lock_t *smb_lock_create(smb_request_t *, uint64_t, uint64_t,
50     uint32_t, uint32_t);
51 static void smb_lock_destroy(smb_lock_t *);
52 static void smb_lock_free(smb_lock_t *);
53 
54 /*
55  * Return the number of range locks on the specified ofile.
56  */
57 uint32_t
58 smb_lock_get_lock_count(smb_node_t *node, smb_ofile_t *of)
59 {
60 	smb_lock_t 	*lock;
61 	smb_llist_t	*llist;
62 	uint32_t	count = 0;
63 
64 	SMB_NODE_VALID(node);
65 	SMB_OFILE_VALID(of);
66 
67 	llist = &node->n_lock_list;
68 
69 	smb_llist_enter(llist, RW_READER);
70 	for (lock = smb_llist_head(llist);
71 	    lock != NULL;
72 	    lock = smb_llist_next(llist, lock)) {
73 		if (lock->l_file == of)
74 			++count;
75 	}
76 	smb_llist_exit(llist);
77 
78 	return (count);
79 }
80 
81 /*
82  * smb_unlock_range
83  *
84  * locates lock range performed for corresponding to unlock request.
85  *
86  * NT_STATUS_SUCCESS - Lock range performed successfully.
87  * !NT_STATUS_SUCCESS - Error in unlock range operation.
88  */
89 uint32_t
90 smb_unlock_range(
91     smb_request_t	*sr,
92     smb_node_t		*node,
93     uint64_t		start,
94     uint64_t		length)
95 {
96 	smb_lock_t	*lock = NULL;
97 	uint32_t	status;
98 
99 	/* Apply unlocking rules */
100 	smb_llist_enter(&node->n_lock_list, RW_WRITER);
101 	status = smb_lock_range_ulckrules(sr, node, start, length, &lock);
102 	if (status != NT_STATUS_SUCCESS) {
103 		/*
104 		 * If lock range is not matching in the list
105 		 * return error.
106 		 */
107 		ASSERT(lock == NULL);
108 		smb_llist_exit(&node->n_lock_list);
109 		return (status);
110 	}
111 
112 	smb_llist_remove(&node->n_lock_list, lock);
113 	smb_lock_posix_unlock(node, lock, sr->user_cr);
114 	smb_llist_exit(&node->n_lock_list);
115 	smb_lock_destroy(lock);
116 
117 	return (status);
118 }
119 
120 /*
121  * smb_lock_range
122  *
123  * Checks for integrity of file lock operation for the given range of file data.
124  * This is performed by applying lock rules with all the elements of the node
125  * lock list.
126  *
127  * Break shared (levelII) oplocks. If there is an exclusive oplock, it is
128  * owned by this ofile and therefore should not be broken.
129  *
130  * The function returns with new lock added if lock request is non-conflicting
131  * with existing range lock for the file. Otherwise smb request is filed
132  * without returning.
133  *
134  * NT_STATUS_SUCCESS - Lock range performed successfully.
135  * !NT_STATUS_SUCCESS - Error in lock range operation.
136  */
137 uint32_t
138 smb_lock_range(
139     smb_request_t	*sr,
140     uint64_t		start,
141     uint64_t		length,
142     uint32_t		timeout,
143     uint32_t		locktype)
144 {
145 	smb_ofile_t	*file = sr->fid_ofile;
146 	smb_node_t	*node = file->f_node;
147 	smb_lock_t	*lock;
148 	smb_lock_t	*clock = NULL;
149 	uint32_t	result = NT_STATUS_SUCCESS;
150 	boolean_t	lock_has_timeout = (timeout != 0);
151 
152 	lock = smb_lock_create(sr, start, length, locktype, timeout);
153 
154 	smb_llist_enter(&node->n_lock_list, RW_WRITER);
155 	for (;;) {
156 		clock_t	rc;
157 
158 		/* Apply locking rules */
159 		result = smb_lock_range_lckrules(sr, file, node, lock, &clock);
160 
161 		if ((result == NT_STATUS_CANCELLED) ||
162 		    (result == NT_STATUS_SUCCESS) ||
163 		    (result == NT_STATUS_RANGE_NOT_LOCKED)) {
164 			ASSERT(clock == NULL);
165 			break;
166 		} else if (timeout == 0) {
167 			break;
168 		}
169 
170 		ASSERT(result == NT_STATUS_LOCK_NOT_GRANTED);
171 		ASSERT(clock);
172 		/*
173 		 * Call smb_lock_wait holding write lock for
174 		 * node lock list.  smb_lock_wait will release
175 		 * this lock if it blocks.
176 		 */
177 		ASSERT(node == clock->l_file->f_node);
178 
179 		rc = smb_lock_wait(sr, lock, clock);
180 		if (rc == 0) {
181 			result = NT_STATUS_CANCELLED;
182 			break;
183 		}
184 		if (rc == -1)
185 			timeout = 0;
186 
187 		clock = NULL;
188 	}
189 
190 	lock->l_blocked_by = NULL;
191 
192 	if (result != NT_STATUS_SUCCESS) {
193 		/*
194 		 * Under certain conditions NT_STATUS_FILE_LOCK_CONFLICT
195 		 * should be returned instead of NT_STATUS_LOCK_NOT_GRANTED.
196 		 */
197 		if (result == NT_STATUS_LOCK_NOT_GRANTED) {
198 			/*
199 			 * Locks with timeouts always return
200 			 * NT_STATUS_FILE_LOCK_CONFLICT
201 			 */
202 			if (lock_has_timeout)
203 				result = NT_STATUS_FILE_LOCK_CONFLICT;
204 
205 			/*
206 			 * Locks starting higher than 0xef000000 that do not
207 			 * have the MSB set always return
208 			 * NT_STATUS_FILE_LOCK_CONFLICT
209 			 */
210 			if ((lock->l_start >= 0xef000000) &&
211 			    !(lock->l_start & (1ULL << 63))) {
212 				result = NT_STATUS_FILE_LOCK_CONFLICT;
213 			}
214 
215 			/*
216 			 * If the last lock attempt to fail on this file handle
217 			 * started at the same offset as this one then return
218 			 * NT_STATUS_FILE_LOCK_CONFLICT
219 			 */
220 			mutex_enter(&file->f_mutex);
221 			if ((file->f_flags & SMB_OFLAGS_LLF_POS_VALID) &&
222 			    (lock->l_start == file->f_llf_pos)) {
223 				result = NT_STATUS_FILE_LOCK_CONFLICT;
224 			}
225 			mutex_exit(&file->f_mutex);
226 		}
227 
228 		/* Update last lock failed offset */
229 		mutex_enter(&file->f_mutex);
230 		file->f_llf_pos = lock->l_start;
231 		file->f_flags |= SMB_OFLAGS_LLF_POS_VALID;
232 		mutex_exit(&file->f_mutex);
233 
234 		smb_lock_free(lock);
235 	} else {
236 		/*
237 		 * don't insert into the CIFS lock list unless the
238 		 * posix lock worked
239 		 */
240 		if (smb_fsop_frlock(node, lock, B_FALSE, sr->user_cr))
241 			result = NT_STATUS_FILE_LOCK_CONFLICT;
242 		else
243 			smb_llist_insert_tail(&node->n_lock_list, lock);
244 	}
245 	smb_llist_exit(&node->n_lock_list);
246 
247 	if (result == NT_STATUS_SUCCESS)
248 		smb_oplock_break_levelII(node);
249 
250 	return (result);
251 }
252 
253 
254 /*
255  * smb_lock_range_access
256  *
257  * scans node lock list
258  * to check if there is any overlapping lock. Overlapping
259  * lock is allowed only under same session and client pid.
260  *
261  * Return values
262  *	NT_STATUS_SUCCESS		lock access granted.
263  *	NT_STATUS_FILE_LOCK_CONFLICT 	access denied due to lock conflict.
264  */
265 int
266 smb_lock_range_access(
267     smb_request_t	*sr,
268     smb_node_t		*node,
269     uint64_t		start,
270     uint64_t		length,
271     boolean_t		will_write)
272 {
273 	smb_lock_t	*lock;
274 	smb_llist_t	*llist;
275 	int		status = NT_STATUS_SUCCESS;
276 
277 	llist = &node->n_lock_list;
278 	smb_llist_enter(llist, RW_READER);
279 	/* Search for any applicable lock */
280 	for (lock = smb_llist_head(llist);
281 	    lock != NULL;
282 	    lock = smb_llist_next(llist, lock)) {
283 
284 		if (!smb_lock_range_overlap(lock, start, length))
285 			/* Lock does not overlap */
286 			continue;
287 
288 		if (lock->l_type == SMB_LOCK_TYPE_READONLY && !will_write)
289 			continue;
290 
291 		if (lock->l_type == SMB_LOCK_TYPE_READWRITE &&
292 		    lock->l_session_kid == sr->session->s_kid &&
293 		    lock->l_pid == sr->smb_pid)
294 			continue;
295 
296 		status = NT_STATUS_FILE_LOCK_CONFLICT;
297 		break;
298 	}
299 	smb_llist_exit(llist);
300 	return (status);
301 }
302 
303 void
304 smb_node_destroy_lock_by_ofile(smb_node_t *node, smb_ofile_t *file)
305 {
306 	smb_lock_t	*lock;
307 	smb_lock_t	*nxtl;
308 	list_t		destroy_list;
309 
310 	SMB_NODE_VALID(node);
311 	ASSERT(node->n_refcnt);
312 
313 	/*
314 	 * Move locks matching the specified file from the node->n_lock_list
315 	 * to a temporary list (holding the lock the entire time) then
316 	 * destroy all the matching locks.  We can't call smb_lock_destroy
317 	 * while we are holding the lock for node->n_lock_list because we will
318 	 * deadlock and we can't drop the lock because the list contents might
319 	 * change (for example nxtl might get removed on another thread).
320 	 */
321 	list_create(&destroy_list, sizeof (smb_lock_t),
322 	    offsetof(smb_lock_t, l_lnd));
323 
324 	smb_llist_enter(&node->n_lock_list, RW_WRITER);
325 	lock = smb_llist_head(&node->n_lock_list);
326 	while (lock) {
327 		nxtl = smb_llist_next(&node->n_lock_list, lock);
328 		if (lock->l_file == file) {
329 			smb_llist_remove(&node->n_lock_list, lock);
330 			smb_lock_posix_unlock(node, lock, file->f_user->u_cred);
331 			list_insert_tail(&destroy_list, lock);
332 		}
333 		lock = nxtl;
334 	}
335 	smb_llist_exit(&node->n_lock_list);
336 
337 	lock = list_head(&destroy_list);
338 	while (lock) {
339 		nxtl = list_next(&destroy_list, lock);
340 		list_remove(&destroy_list, lock);
341 		smb_lock_destroy(lock);
342 		lock = nxtl;
343 	}
344 
345 	list_destroy(&destroy_list);
346 }
347 
348 void
349 smb_lock_range_error(smb_request_t *sr, uint32_t status32)
350 {
351 	uint16_t errcode;
352 
353 	if (status32 == NT_STATUS_CANCELLED)
354 		errcode = ERROR_OPERATION_ABORTED;
355 	else
356 		errcode = ERRlock;
357 
358 	smbsr_error(sr, status32, ERRDOS, errcode);
359 }
360 
361 /*
362  * smb_range_check()
363  *
364  * Perform range checking.  First check for internal CIFS range conflicts
365  * and then check for external conflicts, for example, with NFS or local
366  * access.
367  *
368  * If nbmand is enabled, this function must be called from within an nbmand
369  * critical region
370  */
371 
372 DWORD
373 smb_range_check(smb_request_t *sr, smb_node_t *node, uint64_t start,
374     uint64_t length, boolean_t will_write)
375 {
376 	smb_error_t smberr;
377 	int svmand;
378 	int nbl_op;
379 	int rc;
380 
381 	SMB_NODE_VALID(node);
382 
383 	ASSERT(smb_node_in_crit(node));
384 
385 	if (smb_node_is_dir(node))
386 		return (NT_STATUS_SUCCESS);
387 
388 	rc = smb_lock_range_access(sr, node, start, length, will_write);
389 	if (rc)
390 		return (NT_STATUS_FILE_LOCK_CONFLICT);
391 
392 	if ((rc = nbl_svmand(node->vp, kcred, &svmand)) != 0) {
393 		smbsr_map_errno(rc, &smberr);
394 		return (smberr.status);
395 	}
396 
397 	nbl_op = (will_write) ? NBL_WRITE : NBL_READ;
398 
399 	if (nbl_lock_conflict(node->vp, nbl_op, start, length, svmand, &smb_ct))
400 		return (NT_STATUS_FILE_LOCK_CONFLICT);
401 
402 	return (NT_STATUS_SUCCESS);
403 }
404 
405 /*
406  * smb_lock_posix_unlock
407  *
408  * checks if the current unlock request is in another lock and repeatedly calls
409  * smb_is_range_unlocked on a sliding basis to unlock all bits of the lock
410  * that are not in other locks
411  *
412  */
413 static void
414 smb_lock_posix_unlock(smb_node_t *node, smb_lock_t *lock, cred_t *cr)
415 {
416 	uint64_t	new_mark;
417 	uint64_t	unlock_start;
418 	uint64_t	unlock_end;
419 	smb_lock_t	new_unlock;
420 	smb_llist_t	*llist;
421 	boolean_t	can_unlock;
422 
423 	new_mark = 0;
424 	unlock_start = lock->l_start;
425 	unlock_end = unlock_start + lock->l_length;
426 	llist = &node->n_lock_list;
427 
428 	for (;;) {
429 		can_unlock = smb_is_range_unlocked(unlock_start, unlock_end,
430 		    lock->l_file->f_uniqid, llist, &new_mark);
431 		if (can_unlock) {
432 			if (new_mark) {
433 				new_unlock = *lock;
434 				new_unlock.l_start = unlock_start;
435 				new_unlock.l_length = new_mark - unlock_start;
436 				(void) smb_fsop_frlock(node, &new_unlock,
437 				    B_TRUE, cr);
438 				unlock_start = new_mark;
439 			} else {
440 				new_unlock = *lock;
441 				new_unlock.l_start = unlock_start;
442 				new_unlock.l_length = unlock_end - unlock_start;
443 				(void) smb_fsop_frlock(node, &new_unlock,
444 				    B_TRUE, cr);
445 				break;
446 			}
447 		} else if (new_mark) {
448 			unlock_start = new_mark;
449 		} else {
450 			break;
451 		}
452 	}
453 }
454 
455 /*
456  * smb_lock_range_overlap
457  *
458  * Checks if lock range(start, length) overlaps range in lock structure.
459  *
460  * Zero-length byte range locks actually affect no single byte of the stream,
461  * meaning they can still be accessed even with such locks in place. However,
462  * they do conflict with other ranges in the following manner:
463  *  conflict will only exist if the positive-length range contains the
464  *  zero-length range's offset but doesn't start at it
465  *
466  * return values:
467  *	0 - Lock range doesn't overlap
468  *	1 - Lock range overlaps.
469  */
470 
471 #define	RANGE_NO_OVERLAP	0
472 #define	RANGE_OVERLAP		1
473 
474 static int
475 smb_lock_range_overlap(struct smb_lock *lock, uint64_t start, uint64_t length)
476 {
477 	if (length == 0) {
478 		if ((lock->l_start < start) &&
479 		    ((lock->l_start + lock->l_length) > start))
480 			return (RANGE_OVERLAP);
481 
482 		return (RANGE_NO_OVERLAP);
483 	}
484 
485 	/* The following test is intended to catch roll over locks. */
486 	if ((start == lock->l_start) && (length == lock->l_length))
487 		return (RANGE_OVERLAP);
488 
489 	if (start < lock->l_start) {
490 		if (start + length > lock->l_start)
491 			return (RANGE_OVERLAP);
492 	} else if (start < lock->l_start + lock->l_length)
493 		return (RANGE_OVERLAP);
494 
495 	return (RANGE_NO_OVERLAP);
496 }
497 
498 /*
499  * smb_lock_range_lckrules
500  *
501  * Lock range rules:
502  *	1. Overlapping read locks are allowed if the
503  *	   current locks in the region are only read locks
504  *	   irrespective of pid of smb client issuing lock request.
505  *
506  *	2. Read lock in the overlapped region of write lock
507  *	   are allowed if the pervious lock is performed by the
508  *	   same pid and connection.
509  *
510  * return status:
511  *	NT_STATUS_SUCCESS - Input lock range adapts to lock rules.
512  *	NT_STATUS_LOCK_NOT_GRANTED - Input lock conflicts lock rules.
513  *	NT_STATUS_CANCELLED - Error in processing lock rules
514  */
515 static uint32_t
516 smb_lock_range_lckrules(
517     smb_request_t	*sr,
518     smb_ofile_t		*file,
519     smb_node_t		*node,
520     smb_lock_t		*dlock,
521     smb_lock_t		**clockp)
522 {
523 	smb_lock_t	*lock;
524 	uint32_t	status = NT_STATUS_SUCCESS;
525 
526 	/* Check if file is closed */
527 	if (!smb_ofile_is_open(file)) {
528 		return (NT_STATUS_RANGE_NOT_LOCKED);
529 	}
530 
531 	/* Caller must hold lock for node->n_lock_list */
532 	for (lock = smb_llist_head(&node->n_lock_list);
533 	    lock != NULL;
534 	    lock = smb_llist_next(&node->n_lock_list, lock)) {
535 
536 		if (!smb_lock_range_overlap(lock, dlock->l_start,
537 		    dlock->l_length))
538 			continue;
539 
540 		/*
541 		 * Check to see if lock in the overlapping record
542 		 * is only read lock. Current finding is read
543 		 * locks can overlapped irrespective of pids.
544 		 */
545 		if ((lock->l_type == SMB_LOCK_TYPE_READONLY) &&
546 		    (dlock->l_type == SMB_LOCK_TYPE_READONLY)) {
547 			continue;
548 		}
549 
550 		/*
551 		 * When the read lock overlaps write lock, check if
552 		 * allowed.
553 		 */
554 		if ((dlock->l_type == SMB_LOCK_TYPE_READONLY) &&
555 		    !(lock->l_type == SMB_LOCK_TYPE_READONLY)) {
556 			if (lock->l_file == sr->fid_ofile &&
557 			    lock->l_session_kid == sr->session->s_kid &&
558 			    lock->l_pid == sr->smb_pid &&
559 			    lock->l_uid == sr->smb_uid) {
560 				continue;
561 			}
562 		}
563 
564 		/* Conflict in overlapping lock element */
565 		*clockp = lock;
566 		status = NT_STATUS_LOCK_NOT_GRANTED;
567 		break;
568 	}
569 
570 	return (status);
571 }
572 
573 /*
574  * smb_lock_wait
575  *
576  * Wait operation for smb overlapping lock to be released.  Caller must hold
577  * write lock for node->n_lock_list so that the set of active locks can't
578  * change unexpectedly.  The lock for node->n_lock_list  will be released
579  * within this function during the sleep after the lock dependency has
580  * been recorded.
581  *
582  * return value
583  *
584  *	0	The request was canceled.
585  *	-1	The timeout was reached.
586  *	>0	Condition met.
587  */
588 static clock_t
589 smb_lock_wait(smb_request_t *sr, smb_lock_t *b_lock, smb_lock_t *c_lock)
590 {
591 	clock_t		rc;
592 
593 	ASSERT(sr->sr_awaiting == NULL);
594 
595 	mutex_enter(&sr->sr_mutex);
596 
597 	switch (sr->sr_state) {
598 	case SMB_REQ_STATE_ACTIVE:
599 		/*
600 		 * Wait up till the timeout time keeping track of actual
601 		 * time waited for possible retry failure.
602 		 */
603 		sr->sr_state = SMB_REQ_STATE_WAITING_LOCK;
604 		sr->sr_awaiting = c_lock;
605 		mutex_exit(&sr->sr_mutex);
606 
607 		mutex_enter(&c_lock->l_mutex);
608 		/*
609 		 * The conflict list (l_conflict_list) for a lock contains
610 		 * all the locks that are blocked by and in conflict with
611 		 * that lock.  Add the new lock to the conflict list for the
612 		 * active lock.
613 		 *
614 		 * l_conflict_list is currently a fancy way of representing
615 		 * the references/dependencies on a lock.  It could be
616 		 * replaced with a reference count but this approach
617 		 * has the advantage that MDB can display the lock
618 		 * dependencies at any point in time.  In the future
619 		 * we should be able to leverage the list to implement
620 		 * an asynchronous locking model.
621 		 *
622 		 * l_blocked_by is the reverse of the conflict list.  It
623 		 * points to the lock that the new lock conflicts with.
624 		 * As currently implemented this value is purely for
625 		 * debug purposes -- there are windows of time when
626 		 * l_blocked_by may be non-NULL even though there is no
627 		 * conflict list
628 		 */
629 		b_lock->l_blocked_by = c_lock;
630 		smb_slist_insert_tail(&c_lock->l_conflict_list, b_lock);
631 		smb_llist_exit(&c_lock->l_file->f_node->n_lock_list);
632 
633 		if (SMB_LOCK_INDEFINITE_WAIT(b_lock)) {
634 			cv_wait(&c_lock->l_cv, &c_lock->l_mutex);
635 		} else {
636 			rc = cv_timedwait(&c_lock->l_cv,
637 			    &c_lock->l_mutex, b_lock->l_end_time);
638 		}
639 
640 		mutex_exit(&c_lock->l_mutex);
641 
642 		smb_llist_enter(&c_lock->l_file->f_node->n_lock_list,
643 		    RW_WRITER);
644 		smb_slist_remove(&c_lock->l_conflict_list, b_lock);
645 
646 		mutex_enter(&sr->sr_mutex);
647 		sr->sr_awaiting = NULL;
648 		if (sr->sr_state == SMB_REQ_STATE_CANCELED) {
649 			rc = 0;
650 		} else {
651 			sr->sr_state = SMB_REQ_STATE_ACTIVE;
652 		}
653 		break;
654 
655 	default:
656 		ASSERT(sr->sr_state == SMB_REQ_STATE_CANCELED);
657 		rc = 0;
658 		break;
659 	}
660 	mutex_exit(&sr->sr_mutex);
661 
662 	return (rc);
663 }
664 
665 /*
666  * smb_lock_range_ulckrules
667  *
668  *	1. Unlock should be performed at exactly matching ends.
669  *	   This has been changed because overlapping ends is
670  *	   allowed and there is no other precise way of locating
671  *	   lock entity in node lock list.
672  *
673  *	2. Unlock is failed if there is no corresponding lock exists.
674  *
675  * Return values
676  *
677  *	NT_STATUS_SUCCESS		Unlock request matches lock record
678  *					pointed by 'nodelock' lock structure.
679  *
680  *	NT_STATUS_RANGE_NOT_LOCKED	Unlock request doen't match any
681  *					of lock record in node lock request or
682  *					error in unlock range processing.
683  */
684 static uint32_t
685 smb_lock_range_ulckrules(
686     smb_request_t	*sr,
687     smb_node_t		*node,
688     uint64_t		start,
689     uint64_t		length,
690     smb_lock_t		**nodelock)
691 {
692 	smb_lock_t	*lock;
693 	uint32_t	status = NT_STATUS_RANGE_NOT_LOCKED;
694 
695 	/* Caller must hold lock for node->n_lock_list */
696 	for (lock = smb_llist_head(&node->n_lock_list);
697 	    lock != NULL;
698 	    lock = smb_llist_next(&node->n_lock_list, lock)) {
699 
700 		if ((start == lock->l_start) &&
701 		    (length == lock->l_length) &&
702 		    lock->l_file == sr->fid_ofile &&
703 		    lock->l_session_kid == sr->session->s_kid &&
704 		    lock->l_pid == sr->smb_pid &&
705 		    lock->l_uid == sr->smb_uid) {
706 			*nodelock = lock;
707 			status = NT_STATUS_SUCCESS;
708 			break;
709 		}
710 	}
711 
712 	return (status);
713 }
714 
715 static smb_lock_t *
716 smb_lock_create(
717     smb_request_t *sr,
718     uint64_t start,
719     uint64_t length,
720     uint32_t locktype,
721     uint32_t timeout)
722 {
723 	smb_lock_t *lock;
724 
725 	ASSERT(locktype == SMB_LOCK_TYPE_READWRITE ||
726 	    locktype == SMB_LOCK_TYPE_READONLY);
727 
728 	lock = kmem_zalloc(sizeof (smb_lock_t), KM_SLEEP);
729 	lock->l_magic = SMB_LOCK_MAGIC;
730 	lock->l_sr = sr; /* Invalid after lock is active */
731 	lock->l_session_kid = sr->session->s_kid;
732 	lock->l_session = sr->session;
733 	lock->l_file = sr->fid_ofile;
734 	lock->l_uid = sr->smb_uid;
735 	lock->l_pid = sr->smb_pid;
736 	lock->l_type = locktype;
737 	lock->l_start = start;
738 	lock->l_length = length;
739 	/*
740 	 * Calculate the absolute end time so that we can use it
741 	 * in cv_timedwait.
742 	 */
743 	lock->l_end_time = ddi_get_lbolt() + MSEC_TO_TICK(timeout);
744 	if (timeout == UINT_MAX)
745 		lock->l_flags |= SMB_LOCK_FLAG_INDEFINITE;
746 
747 	mutex_init(&lock->l_mutex, NULL, MUTEX_DEFAULT, NULL);
748 	cv_init(&lock->l_cv, NULL, CV_DEFAULT, NULL);
749 	smb_slist_constructor(&lock->l_conflict_list, sizeof (smb_lock_t),
750 	    offsetof(smb_lock_t, l_conflict_lnd));
751 
752 	return (lock);
753 }
754 
755 static void
756 smb_lock_free(smb_lock_t *lock)
757 {
758 	smb_slist_destructor(&lock->l_conflict_list);
759 	cv_destroy(&lock->l_cv);
760 	mutex_destroy(&lock->l_mutex);
761 
762 	kmem_free(lock, sizeof (smb_lock_t));
763 }
764 
765 /*
766  * smb_lock_destroy
767  *
768  * Caller must hold node->n_lock_list
769  */
770 static void
771 smb_lock_destroy(smb_lock_t *lock)
772 {
773 	/*
774 	 * Caller must hold node->n_lock_list lock.
775 	 */
776 	mutex_enter(&lock->l_mutex);
777 	cv_broadcast(&lock->l_cv);
778 	mutex_exit(&lock->l_mutex);
779 
780 	/*
781 	 * The cv_broadcast above should wake up any locks that previous
782 	 * had conflicts with this lock.  Wait for the locking threads
783 	 * to remove their references to this lock.
784 	 */
785 	smb_slist_wait_for_empty(&lock->l_conflict_list);
786 
787 	smb_lock_free(lock);
788 }
789 
790 /*
791  * smb_is_range_unlocked
792  *
793  * Checks if the current unlock byte range request overlaps another lock
794  * This function is used to determine where POSIX unlocks should be
795  * applied.
796  *
797  * The return code and the value of new_mark must be interpreted as
798  * follows:
799  *
800  * B_TRUE and (new_mark == 0):
801  *   This is the last or only lock left to be unlocked
802  *
803  * B_TRUE and (new_mark > 0):
804  *   The range from start to new_mark can be unlocked
805  *
806  * B_FALSE and (new_mark == 0):
807  *   The unlock can't be performed and we are done
808  *
809  * B_FALSE and (new_mark > 0),
810  *   The range from start to new_mark can't be unlocked
811  *   Start should be reset to new_mark for the next pass
812  */
813 
814 static boolean_t
815 smb_is_range_unlocked(uint64_t start, uint64_t end, uint32_t uniqid,
816     smb_llist_t *llist_head, uint64_t *new_mark)
817 {
818 	struct smb_lock *lk = NULL;
819 	uint64_t low_water_mark = MAXOFFSET_T;
820 	uint64_t lk_start;
821 	uint64_t lk_end;
822 
823 	*new_mark = 0;
824 	lk = smb_llist_head(llist_head);
825 	while (lk) {
826 		if (lk->l_length == 0) {
827 			lk = smb_llist_next(llist_head, lk);
828 			continue;
829 		}
830 
831 		if (lk->l_file->f_uniqid != uniqid) {
832 			lk = smb_llist_next(llist_head, lk);
833 			continue;
834 		}
835 
836 		lk_end = lk->l_start + lk->l_length - 1;
837 		lk_start = lk->l_start;
838 
839 		/*
840 		 * there is no overlap for the first 2 cases
841 		 * check next node
842 		 */
843 		if (lk_end < start) {
844 			lk = smb_llist_next(llist_head, lk);
845 			continue;
846 		}
847 		if (lk_start > end) {
848 			lk = smb_llist_next(llist_head, lk);
849 			continue;
850 		}
851 
852 		/* this range is completely locked */
853 		if ((lk_start <= start) && (lk_end >= end)) {
854 			return (B_FALSE);
855 		}
856 
857 		/* the first part of this range is locked */
858 		if ((start >= lk_start) && (start <= lk_end)) {
859 			if (end > lk_end)
860 				*new_mark = lk_end + 1;
861 			return (B_FALSE);
862 		}
863 
864 		/* this piece is unlocked */
865 		if ((lk_start >= start) && (lk_start <= end)) {
866 			if (low_water_mark > lk_start)
867 				low_water_mark  = lk_start;
868 		}
869 
870 		lk = smb_llist_next(llist_head, lk);
871 	}
872 
873 	if (low_water_mark != MAXOFFSET_T) {
874 		*new_mark = low_water_mark;
875 		return (B_TRUE);
876 	}
877 	/* the range is completely unlocked */
878 	return (B_TRUE);
879 }
880