xref: /illumos-gate/usr/src/uts/common/fs/smbsrv/smb_lock.c (revision 0f1702c5201310f0529cd5abb77652e5e9b241b6)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"@(#)smb_lock.c	1.10	08/08/04 SMI"
27 
28 /*
29  * This module provides range lock functionality for CIFS/SMB clients.
30  * Lock range service functions process SMB lock and and unlock
31  * requests for a file by applying lock rules and marks file range
32  * as locked if the lock is successful otherwise return proper
33  * error code.
34  */
35 
36 #include <smbsrv/smb_incl.h>
37 #include <smbsrv/smb_fsops.h>
38 #include <sys/nbmlock.h>
39 #include <sys/param.h>
40 
41 extern caller_context_t smb_ct;
42 
43 static void smb_lock_posix_unlock(smb_node_t *, smb_lock_t *, cred_t *);
44 static boolean_t smb_is_range_unlocked(uint64_t, uint64_t, uint32_t,
45     smb_llist_t *, uint64_t *);
46 static int smb_lock_range_overlap(smb_lock_t *, uint64_t, uint64_t);
47 static uint32_t smb_lock_range_lckrules(smb_request_t *, smb_ofile_t *,
48     smb_node_t *, smb_lock_t *, smb_lock_t **);
49 static clock_t smb_lock_wait(smb_request_t *, smb_lock_t *, smb_lock_t *);
50 static uint32_t smb_lock_range_ulckrules(smb_request_t *, smb_node_t *,
51     uint64_t, uint64_t, smb_lock_t **nodelock);
52 static smb_lock_t *smb_lock_create(smb_request_t *, uint64_t, uint64_t,
53     uint32_t, uint32_t);
54 static void smb_lock_destroy(smb_lock_t *);
55 static void smb_lock_free(smb_lock_t *);
56 
57 
58 
59 /*
60  * smb_unlock_range
61  *
62  * locates lock range performed for corresponding to unlock request.
63  *
64  * NT_STATUS_SUCCESS - Lock range performed successfully.
65  * !NT_STATUS_SUCCESS - Error in unlock range operation.
66  */
67 uint32_t
68 smb_unlock_range(
69     smb_request_t	*sr,
70     smb_node_t		*node,
71     uint64_t		start,
72     uint64_t		length)
73 {
74 	smb_lock_t	*lock = NULL;
75 	uint32_t	status;
76 
77 	/* Apply unlocking rules */
78 	smb_llist_enter(&node->n_lock_list, RW_WRITER);
79 	status = smb_lock_range_ulckrules(sr, node, start, length, &lock);
80 	if (status != NT_STATUS_SUCCESS) {
81 		/*
82 		 * If lock range is not matching in the list
83 		 * return error.
84 		 */
85 		ASSERT(lock == NULL);
86 		smb_llist_exit(&node->n_lock_list);
87 		return (status);
88 	}
89 
90 	smb_llist_remove(&node->n_lock_list, lock);
91 	smb_lock_posix_unlock(node, lock, sr->user_cr);
92 	smb_llist_exit(&node->n_lock_list);
93 	smb_lock_destroy(lock);
94 
95 	return (status);
96 }
97 
98 /*
99  * smb_lock_range
100  *
101  * checks for integrity of file lock operation for the given range of file data.
102  * This is performed by applying lock rules with all the elements of the node
103  * lock list.
104  *
105  * The function returns with new lock added if lock request is non-conflicting
106  * with existing range lock for the file. Otherwise smb request is filed
107  * without returning.
108  *
109  * NT_STATUS_SUCCESS - Lock range performed successfully.
110  * !NT_STATUS_SUCCESS - Error in lock range operation.
111  */
112 uint32_t
113 smb_lock_range(
114     smb_request_t	*sr,
115     uint64_t		start,
116     uint64_t		length,
117     uint32_t		timeout,
118     uint32_t		locktype)
119 {
120 	smb_ofile_t	*file = sr->fid_ofile;
121 	smb_node_t	*node = file->f_node;
122 	smb_lock_t	*lock;
123 	smb_lock_t	*clock = NULL;
124 	uint32_t	result = NT_STATUS_SUCCESS;
125 	boolean_t	lock_has_timeout = (timeout != 0);
126 
127 	lock = smb_lock_create(sr, start, length, locktype, timeout);
128 
129 	smb_llist_enter(&node->n_lock_list, RW_WRITER);
130 	for (;;) {
131 		clock_t	rc;
132 
133 		/* Apply locking rules */
134 		result = smb_lock_range_lckrules(sr, file, node, lock, &clock);
135 
136 		if ((result == NT_STATUS_CANCELLED) ||
137 		    (result == NT_STATUS_SUCCESS) ||
138 		    (result == NT_STATUS_RANGE_NOT_LOCKED)) {
139 			ASSERT(clock == NULL);
140 			break;
141 		} else if (timeout == 0) {
142 			break;
143 		}
144 
145 		ASSERT(result == NT_STATUS_LOCK_NOT_GRANTED);
146 		ASSERT(clock);
147 		/*
148 		 * Call smb_lock_wait holding write lock for
149 		 * node lock list.  smb_lock_wait will release
150 		 * this lock if it blocks.
151 		 */
152 		ASSERT(node == clock->l_file->f_node);
153 
154 		rc = smb_lock_wait(sr, lock, clock);
155 		if (rc == 0) {
156 			result = NT_STATUS_CANCELLED;
157 			break;
158 		}
159 		if (rc == -1)
160 			timeout = 0;
161 
162 		clock = NULL;
163 	}
164 
165 	lock->l_blocked_by = NULL;
166 
167 	if (result != NT_STATUS_SUCCESS) {
168 		/*
169 		 * Under certain conditions NT_STATUS_FILE_LOCK_CONFLICT
170 		 * should be returned instead of NT_STATUS_LOCK_NOT_GRANTED.
171 		 */
172 		if (result == NT_STATUS_LOCK_NOT_GRANTED) {
173 			/*
174 			 * Locks with timeouts always return
175 			 * NT_STATUS_FILE_LOCK_CONFLICT
176 			 */
177 			if (lock_has_timeout)
178 				result = NT_STATUS_FILE_LOCK_CONFLICT;
179 
180 			/*
181 			 * Locks starting higher than 0xef000000 that do not
182 			 * have the MSB set always return
183 			 * NT_STATUS_FILE_LOCK_CONFLICT
184 			 */
185 			if ((lock->l_start >= 0xef000000) &&
186 			    !(lock->l_start & (1ULL << 63))) {
187 				result = NT_STATUS_FILE_LOCK_CONFLICT;
188 			}
189 
190 			/*
191 			 * If the last lock attempt to fail on this file handle
192 			 * started at the same offset as this one then return
193 			 * NT_STATUS_FILE_LOCK_CONFLICT
194 			 */
195 			mutex_enter(&file->f_mutex);
196 			if ((file->f_flags & SMB_OFLAGS_LLF_POS_VALID) &&
197 			    (lock->l_start == file->f_llf_pos)) {
198 				result = NT_STATUS_FILE_LOCK_CONFLICT;
199 			}
200 			mutex_exit(&file->f_mutex);
201 		}
202 
203 		/* Update last lock failed offset */
204 		mutex_enter(&file->f_mutex);
205 		file->f_llf_pos = lock->l_start;
206 		file->f_flags |= SMB_OFLAGS_LLF_POS_VALID;
207 		mutex_exit(&file->f_mutex);
208 
209 		smb_lock_free(lock);
210 	} else {
211 		/*
212 		 * don't insert into the CIFS lock list unless the
213 		 * posix lock worked
214 		 */
215 		if (smb_fsop_frlock(node, lock, B_FALSE, sr->user_cr))
216 			result = NT_STATUS_FILE_LOCK_CONFLICT;
217 		else
218 			smb_llist_insert_tail(&node->n_lock_list, lock);
219 	}
220 	smb_llist_exit(&node->n_lock_list);
221 
222 	return (result);
223 }
224 
225 
226 /*
227  * smb_lock_range_access
228  *
229  * scans node lock list
230  * to check if there is any overlapping lock. Overlapping
231  * lock is allowed only under same session and client pid.
232  *
233  * Return values
234  *	NT_STATUS_SUCCESS		lock access granted.
235  *	NT_STATUS_FILE_LOCK_CONFLICT 	access denied due to lock conflict.
236  */
237 int
238 smb_lock_range_access(
239     smb_request_t	*sr,
240     smb_node_t		*node,
241     uint64_t		start,
242     uint64_t		length,
243     boolean_t		will_write)
244 {
245 	smb_lock_t	*lock;
246 	smb_llist_t	*llist;
247 	int		status = NT_STATUS_SUCCESS;
248 
249 	llist = &node->n_lock_list;
250 	smb_llist_enter(llist, RW_READER);
251 	/* Search for any applicable lock */
252 	for (lock = smb_llist_head(llist);
253 	    lock != NULL;
254 	    lock = smb_llist_next(llist, lock)) {
255 
256 		if (!smb_lock_range_overlap(lock, start, length))
257 			/* Lock does not overlap */
258 			continue;
259 
260 		if (lock->l_type == SMB_LOCK_TYPE_READONLY && !will_write)
261 			continue;
262 
263 		if (lock->l_type == SMB_LOCK_TYPE_READWRITE &&
264 		    lock->l_session_kid == sr->session->s_kid &&
265 		    lock->l_pid == sr->smb_pid)
266 			continue;
267 
268 		status = NT_STATUS_FILE_LOCK_CONFLICT;
269 		break;
270 	}
271 	smb_llist_exit(llist);
272 	return (status);
273 }
274 
275 void
276 smb_node_destroy_lock_by_ofile(smb_node_t *node, smb_ofile_t *file)
277 {
278 	smb_lock_t	*lock;
279 	smb_lock_t	*nxtl;
280 	list_t		destroy_list;
281 
282 	ASSERT(node);
283 	ASSERT(node->n_magic == SMB_NODE_MAGIC);
284 	ASSERT(node->n_state == SMB_NODE_STATE_AVAILABLE);
285 	ASSERT(node->n_refcnt);
286 
287 	/*
288 	 * Move locks matching the specified file from the node->n_lock_list
289 	 * to a temporary list (holding the lock the entire time) then
290 	 * destroy all the matching locks.  We can't call smb_lock_destroy
291 	 * while we are holding the lock for node->n_lock_list because we will
292 	 * deadlock and we can't drop the lock because the list contents might
293 	 * change (for example nxtl might get removed on another thread).
294 	 */
295 	list_create(&destroy_list, sizeof (smb_lock_t),
296 	    offsetof(smb_lock_t, l_lnd));
297 
298 	smb_llist_enter(&node->n_lock_list, RW_WRITER);
299 	lock = smb_llist_head(&node->n_lock_list);
300 	while (lock) {
301 		nxtl = smb_llist_next(&node->n_lock_list, lock);
302 		if (lock->l_file == file) {
303 			smb_llist_remove(&node->n_lock_list, lock);
304 			smb_lock_posix_unlock(node, lock, file->f_user->u_cred);
305 			list_insert_tail(&destroy_list, lock);
306 		}
307 		lock = nxtl;
308 	}
309 	smb_llist_exit(&node->n_lock_list);
310 
311 	lock = list_head(&destroy_list);
312 	while (lock) {
313 		nxtl = list_next(&destroy_list, lock);
314 		list_remove(&destroy_list, lock);
315 		smb_lock_destroy(lock);
316 		lock = nxtl;
317 	}
318 
319 	list_destroy(&destroy_list);
320 }
321 
322 void
323 smb_lock_range_error(smb_request_t *sr, uint32_t status32)
324 {
325 	uint16_t errcode;
326 
327 	if (status32 == NT_STATUS_CANCELLED)
328 		errcode = ERROR_OPERATION_ABORTED;
329 	else
330 		errcode = ERRlock;
331 
332 	smbsr_error(sr, status32, ERRDOS, errcode);
333 }
334 
335 /*
336  * smb_range_check()
337  *
338  * Perform range checking.  First check for internal CIFS range conflicts
339  * and then check for external conflicts, for example, with NFS or local
340  * access.
341  *
342  * If nbmand is enabled, this function must be called from within an nbmand
343  * critical region
344  */
345 
346 DWORD
347 smb_range_check(smb_request_t *sr, smb_node_t *node, uint64_t start,
348     uint64_t length, boolean_t will_write)
349 {
350 	smb_error_t smberr;
351 	int svmand;
352 	int nbl_op;
353 	int rc;
354 
355 	ASSERT(node);
356 	ASSERT(node->n_magic == SMB_NODE_MAGIC);
357 	ASSERT(node->n_state == SMB_NODE_STATE_AVAILABLE);
358 
359 	ASSERT(smb_node_in_crit(node));
360 
361 	if (node->attr.sa_vattr.va_type == VDIR)
362 		return (NT_STATUS_SUCCESS);
363 
364 	rc = smb_lock_range_access(sr, node, start, length, will_write);
365 	if (rc)
366 		return (NT_STATUS_FILE_LOCK_CONFLICT);
367 
368 	if ((rc = nbl_svmand(node->vp, kcred, &svmand)) != 0) {
369 		smbsr_map_errno(rc, &smberr);
370 		return (smberr.status);
371 	}
372 
373 	nbl_op = (will_write) ? NBL_WRITE : NBL_READ;
374 
375 	if (nbl_lock_conflict(node->vp, nbl_op, start, length, svmand, &smb_ct))
376 		return (NT_STATUS_FILE_LOCK_CONFLICT);
377 
378 	return (NT_STATUS_SUCCESS);
379 }
380 
381 /*
382  * smb_lock_posix_unlock
383  *
384  * checks if the current unlock request is in another lock and repeatedly calls
385  * smb_is_range_unlocked on a sliding basis to unlock all bits of the lock
386  * that are not in other locks
387  *
388  */
389 static void
390 smb_lock_posix_unlock(smb_node_t *node, smb_lock_t *lock, cred_t *cr)
391 {
392 	uint64_t	new_mark;
393 	uint64_t	unlock_start;
394 	uint64_t	unlock_end;
395 	smb_lock_t	new_unlock;
396 	smb_llist_t	*llist;
397 	boolean_t	can_unlock;
398 
399 	new_mark = 0;
400 	unlock_start = lock->l_start;
401 	unlock_end = unlock_start + lock->l_length;
402 	llist = &node->n_lock_list;
403 
404 	for (;;) {
405 		can_unlock = smb_is_range_unlocked(unlock_start, unlock_end,
406 		    lock->l_file->f_uniqid, llist, &new_mark);
407 		if (can_unlock) {
408 			if (new_mark) {
409 				new_unlock = *lock;
410 				new_unlock.l_start = unlock_start;
411 				new_unlock.l_length = new_mark - unlock_start;
412 				(void) smb_fsop_frlock(node, &new_unlock,
413 				    B_TRUE, cr);
414 				unlock_start = new_mark;
415 			} else {
416 				new_unlock = *lock;
417 				new_unlock.l_start = unlock_start;
418 				new_unlock.l_length = unlock_end - unlock_start;
419 				(void) smb_fsop_frlock(node, &new_unlock,
420 				    B_TRUE, cr);
421 				break;
422 			}
423 		} else if (new_mark) {
424 			unlock_start = new_mark;
425 		} else {
426 			break;
427 		}
428 	}
429 }
430 
431 /*
432  * smb_lock_range_overlap
433  *
434  * Checks if lock range(start, length) overlaps range in lock structure.
435  *
436  * Zero-length byte range locks actually affect no single byte of the stream,
437  * meaning they can still be accessed even with such locks in place. However,
438  * they do conflict with other ranges in the following manner:
439  *  conflict will only exist if the positive-length range contains the
440  *  zero-length range's offset but doesn't start at it
441  *
442  * return values:
443  *	0 - Lock range doesn't overlap
444  *	1 - Lock range overlaps.
445  */
446 
447 #define	RANGE_NO_OVERLAP	0
448 #define	RANGE_OVERLAP		1
449 
450 static int
451 smb_lock_range_overlap(struct smb_lock *lock, uint64_t start, uint64_t length)
452 {
453 	if (length == 0) {
454 		if ((lock->l_start < start) &&
455 		    ((lock->l_start + lock->l_length) > start))
456 			return (RANGE_OVERLAP);
457 
458 		return (RANGE_NO_OVERLAP);
459 	}
460 
461 	/* The following test is intended to catch roll over locks. */
462 	if ((start == lock->l_start) && (length == lock->l_length))
463 		return (RANGE_OVERLAP);
464 
465 	if (start < lock->l_start) {
466 		if (start + length > lock->l_start)
467 			return (RANGE_OVERLAP);
468 	} else if (start < lock->l_start + lock->l_length)
469 		return (RANGE_OVERLAP);
470 
471 	return (RANGE_NO_OVERLAP);
472 }
473 
474 /*
475  * smb_lock_range_lckrules
476  *
477  * Lock range rules:
478  *	1. Overlapping read locks are allowed if the
479  *	   current locks in the region are only read locks
480  *	   irrespective of pid of smb client issuing lock request.
481  *
482  *	2. Read lock in the overlapped region of write lock
483  *	   are allowed if the pervious lock is performed by the
484  *	   same pid and connection.
485  *
486  * return status:
487  *	NT_STATUS_SUCCESS - Input lock range adapts to lock rules.
488  *	NT_STATUS_LOCK_NOT_GRANTED - Input lock conflicts lock rules.
489  *	NT_STATUS_CANCELLED - Error in processing lock rules
490  */
491 static uint32_t
492 smb_lock_range_lckrules(
493     smb_request_t	*sr,
494     smb_ofile_t		*file,
495     smb_node_t		*node,
496     smb_lock_t		*dlock,
497     smb_lock_t		**clockp)
498 {
499 	smb_lock_t	*lock;
500 	uint32_t	status = NT_STATUS_SUCCESS;
501 
502 	/* Check if file is closed */
503 	if (!smb_ofile_is_open(file)) {
504 		return (NT_STATUS_RANGE_NOT_LOCKED);
505 	}
506 
507 	/* Caller must hold lock for node->n_lock_list */
508 	for (lock = smb_llist_head(&node->n_lock_list);
509 	    lock != NULL;
510 	    lock = smb_llist_next(&node->n_lock_list, lock)) {
511 
512 		if (!smb_lock_range_overlap(lock, dlock->l_start,
513 		    dlock->l_length))
514 			continue;
515 
516 		/*
517 		 * Check to see if lock in the overlapping record
518 		 * is only read lock. Current finding is read
519 		 * locks can overlapped irrespective of pids.
520 		 */
521 		if ((lock->l_type == SMB_LOCK_TYPE_READONLY) &&
522 		    (dlock->l_type == SMB_LOCK_TYPE_READONLY)) {
523 			continue;
524 		}
525 
526 		/*
527 		 * When the read lock overlaps write lock, check if
528 		 * allowed.
529 		 */
530 		if ((dlock->l_type == SMB_LOCK_TYPE_READONLY) &&
531 		    !(lock->l_type == SMB_LOCK_TYPE_READONLY)) {
532 			if (lock->l_file == sr->fid_ofile &&
533 			    lock->l_session_kid == sr->session->s_kid &&
534 			    lock->l_pid == sr->smb_pid &&
535 			    lock->l_uid == sr->smb_uid) {
536 				continue;
537 			}
538 		}
539 
540 		/* Conflict in overlapping lock element */
541 		*clockp = lock;
542 		status = NT_STATUS_LOCK_NOT_GRANTED;
543 		break;
544 	}
545 
546 	return (status);
547 }
548 
549 /*
550  * smb_lock_wait
551  *
552  * Wait operation for smb overlapping lock to be released.  Caller must hold
553  * write lock for node->n_lock_list so that the set of active locks can't
554  * change unexpectedly.  The lock for node->n_lock_list  will be released
555  * within this function during the sleep after the lock dependency has
556  * been recorded.
557  *
558  * return value
559  *
560  *	0	The request was canceled.
561  *	-1	The timeout was reached.
562  *	>0	Condition met.
563  */
564 static clock_t
565 smb_lock_wait(smb_request_t *sr, smb_lock_t *b_lock, smb_lock_t *c_lock)
566 {
567 	clock_t		rc;
568 
569 	ASSERT(sr->sr_awaiting == NULL);
570 
571 	mutex_enter(&sr->sr_mutex);
572 
573 	switch (sr->sr_state) {
574 	case SMB_REQ_STATE_ACTIVE:
575 		/*
576 		 * Wait up till the timeout time keeping track of actual
577 		 * time waited for possible retry failure.
578 		 */
579 		sr->sr_state = SMB_REQ_STATE_WAITING_LOCK;
580 		sr->sr_awaiting = c_lock;
581 		mutex_exit(&sr->sr_mutex);
582 
583 		mutex_enter(&c_lock->l_mutex);
584 		/*
585 		 * The conflict list (l_conflict_list) for a lock contains
586 		 * all the locks that are blocked by and in conflict with
587 		 * that lock.  Add the new lock to the conflict list for the
588 		 * active lock.
589 		 *
590 		 * l_conflict_list is currently a fancy way of representing
591 		 * the references/dependencies on a lock.  It could be
592 		 * replaced with a reference count but this approach
593 		 * has the advantage that MDB can display the lock
594 		 * dependencies at any point in time.  In the future
595 		 * we should be able to leverage the list to implement
596 		 * an asynchronous locking model.
597 		 *
598 		 * l_blocked_by is the reverse of the conflict list.  It
599 		 * points to the lock that the new lock conflicts with.
600 		 * As currently implemented this value is purely for
601 		 * debug purposes -- there are windows of time when
602 		 * l_blocked_by may be non-NULL even though there is no
603 		 * conflict list
604 		 */
605 		b_lock->l_blocked_by = c_lock;
606 		smb_slist_insert_tail(&c_lock->l_conflict_list, b_lock);
607 		smb_llist_exit(&c_lock->l_file->f_node->n_lock_list);
608 
609 		/*
610 		 * XXX Hack.. drop s_lock to avoid blocking subsequent SMBs
611 		 * that might affect the state of this lock (i.e.
612 		 * smb_com_close).  We shouldn't sleep while holding
613 		 * locks anyway.
614 		 */
615 		smb_rwx_rwexit(&sr->session->s_lock);
616 
617 		if (SMB_LOCK_INDEFINITE_WAIT(b_lock)) {
618 			cv_wait(&c_lock->l_cv, &c_lock->l_mutex);
619 		} else {
620 			rc = cv_timedwait(&c_lock->l_cv,
621 			    &c_lock->l_mutex, b_lock->l_end_time);
622 		}
623 
624 		/*
625 		 * XXX Hack continued from above... re-acquire s_lock
626 		 * OK to hardcode RW_READER since this is just a hack and
627 		 * we really should yank it out and do something else.
628 		 */
629 		smb_rwx_rwenter(&sr->session->s_lock, RW_READER);
630 
631 		mutex_exit(&c_lock->l_mutex);
632 
633 		smb_llist_enter(&c_lock->l_file->f_node->n_lock_list,
634 		    RW_WRITER);
635 		smb_slist_remove(&c_lock->l_conflict_list, b_lock);
636 
637 		mutex_enter(&sr->sr_mutex);
638 		sr->sr_awaiting = NULL;
639 		if (sr->sr_state == SMB_REQ_STATE_CANCELED) {
640 			rc = 0;
641 		} else {
642 			sr->sr_state = SMB_REQ_STATE_ACTIVE;
643 		}
644 		break;
645 
646 	default:
647 		ASSERT(sr->sr_state == SMB_REQ_STATE_CANCELED);
648 		rc = 0;
649 		break;
650 	}
651 	mutex_exit(&sr->sr_mutex);
652 
653 	return (rc);
654 }
655 
656 /*
657  * smb_lock_range_ulckrules
658  *
659  *	1. Unlock should be performed at exactly matching ends.
660  *	   This has been changed because overlapping ends is
661  *	   allowed and there is no other precise way of locating
662  *	   lock entity in node lock list.
663  *
664  *	2. Unlock is failed if there is no corresponding lock exists.
665  *
666  * Return values
667  *
668  *	NT_STATUS_SUCCESS		Unlock request matches lock record
669  *					pointed by 'nodelock' lock structure.
670  *
671  *	NT_STATUS_RANGE_NOT_LOCKED	Unlock request doen't match any
672  *					of lock record in node lock request or
673  *					error in unlock range processing.
674  */
675 static uint32_t
676 smb_lock_range_ulckrules(
677     smb_request_t	*sr,
678     smb_node_t		*node,
679     uint64_t		start,
680     uint64_t		length,
681     smb_lock_t		**nodelock)
682 {
683 	smb_lock_t	*lock;
684 	uint32_t	status = NT_STATUS_RANGE_NOT_LOCKED;
685 
686 	/* Caller must hold lock for node->n_lock_list */
687 	for (lock = smb_llist_head(&node->n_lock_list);
688 	    lock != NULL;
689 	    lock = smb_llist_next(&node->n_lock_list, lock)) {
690 
691 		if ((start == lock->l_start) &&
692 		    (length == lock->l_length) &&
693 		    lock->l_file == sr->fid_ofile &&
694 		    lock->l_session_kid == sr->session->s_kid &&
695 		    lock->l_pid == sr->smb_pid &&
696 		    lock->l_uid == sr->smb_uid) {
697 			*nodelock = lock;
698 			status = NT_STATUS_SUCCESS;
699 			break;
700 		}
701 	}
702 
703 	return (status);
704 }
705 
706 static smb_lock_t *
707 smb_lock_create(
708     smb_request_t *sr,
709     uint64_t start,
710     uint64_t length,
711     uint32_t locktype,
712     uint32_t timeout)
713 {
714 	smb_lock_t *lock;
715 
716 	ASSERT(locktype == SMB_LOCK_TYPE_READWRITE ||
717 	    locktype == SMB_LOCK_TYPE_READONLY);
718 
719 	lock = kmem_zalloc(sizeof (smb_lock_t), KM_SLEEP);
720 	lock->l_magic = SMB_LOCK_MAGIC;
721 	lock->l_sr = sr; /* Invalid after lock is active */
722 	lock->l_session_kid = sr->session->s_kid;
723 	lock->l_session = sr->session;
724 	lock->l_file = sr->fid_ofile;
725 	lock->l_uid = sr->smb_uid;
726 	lock->l_pid = sr->smb_pid;
727 	lock->l_type = locktype;
728 	lock->l_start = start;
729 	lock->l_length = length;
730 	/*
731 	 * Calculate the absolute end time so that we can use it
732 	 * in cv_timedwait.
733 	 */
734 	lock->l_end_time = lbolt + MSEC_TO_TICK(timeout);
735 	if (timeout == UINT_MAX)
736 		lock->l_flags |= SMB_LOCK_FLAG_INDEFINITE;
737 
738 	mutex_init(&lock->l_mutex, NULL, MUTEX_DEFAULT, NULL);
739 	cv_init(&lock->l_cv, NULL, CV_DEFAULT, NULL);
740 	smb_slist_constructor(&lock->l_conflict_list, sizeof (smb_lock_t),
741 	    offsetof(smb_lock_t, l_conflict_lnd));
742 
743 	return (lock);
744 }
745 
746 static void
747 smb_lock_free(smb_lock_t *lock)
748 {
749 	smb_slist_destructor(&lock->l_conflict_list);
750 	cv_destroy(&lock->l_cv);
751 	mutex_destroy(&lock->l_mutex);
752 
753 	kmem_free(lock, sizeof (smb_lock_t));
754 }
755 
756 /*
757  * smb_lock_destroy
758  *
759  * Caller must hold node->n_lock_list
760  */
761 static void
762 smb_lock_destroy(smb_lock_t *lock)
763 {
764 	/*
765 	 * Caller must hold node->n_lock_list lock.
766 	 */
767 	mutex_enter(&lock->l_mutex);
768 	cv_broadcast(&lock->l_cv);
769 	mutex_exit(&lock->l_mutex);
770 
771 	/*
772 	 * The cv_broadcast above should wake up any locks that previous
773 	 * had conflicts with this lock.  Wait for the locking threads
774 	 * to remove their references to this lock.
775 	 */
776 	smb_slist_wait_for_empty(&lock->l_conflict_list);
777 
778 	smb_lock_free(lock);
779 }
780 
781 /*
782  * smb_is_range_unlocked
783  *
784  * Checks if the current unlock byte range request overlaps another lock
785  * This function is used to determine where POSIX unlocks should be
786  * applied.
787  *
788  * The return code and the value of new_mark must be interpreted as
789  * follows:
790  *
791  * B_TRUE and (new_mark == 0):
792  *   This is the last or only lock left to be unlocked
793  *
794  * B_TRUE and (new_mark > 0):
795  *   The range from start to new_mark can be unlocked
796  *
797  * B_FALSE and (new_mark == 0):
798  *   The unlock can't be performed and we are done
799  *
800  * B_FALSE and (new_mark > 0),
801  *   The range from start to new_mark can't be unlocked
802  *   Start should be reset to new_mark for the next pass
803  */
804 
805 static boolean_t
806 smb_is_range_unlocked(uint64_t start, uint64_t end, uint32_t uniqid,
807     smb_llist_t *llist_head, uint64_t *new_mark)
808 {
809 	struct smb_lock *lk = NULL;
810 	uint64_t low_water_mark = MAXOFFSET_T;
811 	uint64_t lk_start;
812 	uint64_t lk_end;
813 
814 	*new_mark = 0;
815 	lk = smb_llist_head(llist_head);
816 	while (lk) {
817 		if (lk->l_length == 0) {
818 			lk = smb_llist_next(llist_head, lk);
819 			continue;
820 		}
821 
822 		if (lk->l_file->f_uniqid != uniqid) {
823 			lk = smb_llist_next(llist_head, lk);
824 			continue;
825 		}
826 
827 		lk_end = lk->l_start + lk->l_length - 1;
828 		lk_start = lk->l_start;
829 
830 		/*
831 		 * there is no overlap for the first 2 cases
832 		 * check next node
833 		 */
834 		if (lk_end < start) {
835 			lk = smb_llist_next(llist_head, lk);
836 			continue;
837 		}
838 		if (lk_start > end) {
839 			lk = smb_llist_next(llist_head, lk);
840 			continue;
841 		}
842 
843 		/* this range is completely locked */
844 		if ((lk_start <= start) && (lk_end >= end)) {
845 			return (B_FALSE);
846 		}
847 
848 		/* the first part of this range is locked */
849 		if ((start >= lk_start) && (start <= lk_end)) {
850 			if (end > lk_end)
851 				*new_mark = lk_end + 1;
852 			return (B_FALSE);
853 		}
854 
855 		/* this piece is unlocked */
856 		if ((lk_start >= start) && (lk_start <= end)) {
857 			if (low_water_mark > lk_start)
858 				low_water_mark  = lk_start;
859 		}
860 
861 		lk = smb_llist_next(llist_head, lk);
862 	}
863 
864 	if (low_water_mark != MAXOFFSET_T) {
865 		*new_mark = low_water_mark;
866 		return (B_TRUE);
867 	}
868 	/* the range is completely unlocked */
869 	return (B_TRUE);
870 }
871