xref: /freebsd/sys/contrib/openzfs/module/zfs/zfs_rlock.c (revision d0b2dbfa0ecf2bbc9709efc5e20baf8e4b44bbbf)
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 https://opensource.org/licenses/CDDL-1.0.
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 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 /*
26  * Copyright (c) 2012, 2018 by Delphix. All rights reserved.
27  */
28 
29 /*
30  * This file contains the code to implement file range locking in
31  * ZFS, although there isn't much specific to ZFS (all that comes to mind is
32  * support for growing the blocksize).
33  *
34  * Interface
35  * ---------
36  * Defined in zfs_rlock.h but essentially:
37  *	lr = rangelock_enter(zp, off, len, lock_type);
38  *	rangelock_reduce(lr, off, len); // optional
39  *	rangelock_exit(lr);
40  *
41  * Range locking rules
42  * --------------------
43  * 1. When truncating a file (zfs_create, zfs_setattr, zfs_space) the whole
44  *    file range needs to be locked as RL_WRITER. Only then can the pages be
45  *    freed etc and zp_size reset. zp_size must be set within range lock.
46  * 2. For writes and punching holes (zfs_write & zfs_space) just the range
47  *    being written or freed needs to be locked as RL_WRITER.
48  *    Multiple writes at the end of the file must coordinate zp_size updates
49  *    to ensure data isn't lost. A compare and swap loop is currently used
50  *    to ensure the file size is at least the offset last written.
51  * 3. For reads (zfs_read, zfs_get_data & zfs_putapage) just the range being
52  *    read needs to be locked as RL_READER. A check against zp_size can then
53  *    be made for reading beyond end of file.
54  *
55  * AVL tree
56  * --------
57  * An AVL tree is used to maintain the state of the existing ranges
58  * that are locked for exclusive (writer) or shared (reader) use.
59  * The starting range offset is used for searching and sorting the tree.
60  *
61  * Common case
62  * -----------
63  * The (hopefully) usual case is of no overlaps or contention for locks. On
64  * entry to rangelock_enter(), a locked_range_t is allocated; the tree
65  * searched that finds no overlap, and *this* locked_range_t is placed in the
66  * tree.
67  *
68  * Overlaps/Reference counting/Proxy locks
69  * ---------------------------------------
70  * The avl code only allows one node at a particular offset. Also it's very
71  * inefficient to search through all previous entries looking for overlaps
72  * (because the very 1st in the ordered list might be at offset 0 but
73  * cover the whole file).
74  * So this implementation uses reference counts and proxy range locks.
75  * Firstly, only reader locks use reference counts and proxy locks,
76  * because writer locks are exclusive.
77  * When a reader lock overlaps with another then a proxy lock is created
78  * for that range and replaces the original lock. If the overlap
79  * is exact then the reference count of the proxy is simply incremented.
80  * Otherwise, the proxy lock is split into smaller lock ranges and
81  * new proxy locks created for non overlapping ranges.
82  * The reference counts are adjusted accordingly.
83  * Meanwhile, the original lock is kept around (this is the callers handle)
84  * and its offset and length are used when releasing the lock.
85  *
86  * Thread coordination
87  * -------------------
88  * In order to make wakeups efficient and to ensure multiple continuous
89  * readers on a range don't starve a writer for the same range lock,
90  * two condition variables are allocated in each rl_t.
91  * If a writer (or reader) can't get a range it initialises the writer
92  * (or reader) cv; sets a flag saying there's a writer (or reader) waiting;
93  * and waits on that cv. When a thread unlocks that range it wakes up all
94  * writers then all readers before destroying the lock.
95  *
96  * Append mode writes
97  * ------------------
98  * Append mode writes need to lock a range at the end of a file.
99  * The offset of the end of the file is determined under the
100  * range locking mutex, and the lock type converted from RL_APPEND to
101  * RL_WRITER and the range locked.
102  *
103  * Grow block handling
104  * -------------------
105  * ZFS supports multiple block sizes, up to 16MB. The smallest
106  * block size is used for the file which is grown as needed. During this
107  * growth all other writers and readers must be excluded.
108  * So if the block size needs to be grown then the whole file is
109  * exclusively locked, then later the caller will reduce the lock
110  * range to just the range to be written using rangelock_reduce().
111  */
112 
113 #include <sys/zfs_context.h>
114 #include <sys/zfs_rlock.h>
115 
116 
117 /*
118  * AVL comparison function used to order range locks
119  * Locks are ordered on the start offset of the range.
120  */
121 static int
122 zfs_rangelock_compare(const void *arg1, const void *arg2)
123 {
124 	const zfs_locked_range_t *rl1 = (const zfs_locked_range_t *)arg1;
125 	const zfs_locked_range_t *rl2 = (const zfs_locked_range_t *)arg2;
126 
127 	return (TREE_CMP(rl1->lr_offset, rl2->lr_offset));
128 }
129 
130 /*
131  * The callback is invoked when acquiring a RL_WRITER or RL_APPEND lock.
132  * It must convert RL_APPEND to RL_WRITER (starting at the end of the file),
133  * and may increase the range that's locked for RL_WRITER.
134  */
135 void
136 zfs_rangelock_init(zfs_rangelock_t *rl, zfs_rangelock_cb_t *cb, void *arg)
137 {
138 	mutex_init(&rl->rl_lock, NULL, MUTEX_DEFAULT, NULL);
139 	avl_create(&rl->rl_tree, zfs_rangelock_compare,
140 	    sizeof (zfs_locked_range_t), offsetof(zfs_locked_range_t, lr_node));
141 	rl->rl_cb = cb;
142 	rl->rl_arg = arg;
143 }
144 
145 void
146 zfs_rangelock_fini(zfs_rangelock_t *rl)
147 {
148 	mutex_destroy(&rl->rl_lock);
149 	avl_destroy(&rl->rl_tree);
150 }
151 
152 /*
153  * Check if a write lock can be grabbed.  If not, fail immediately or sleep and
154  * recheck until available, depending on the value of the "nonblock" parameter.
155  */
156 static boolean_t
157 zfs_rangelock_enter_writer(zfs_rangelock_t *rl, zfs_locked_range_t *new,
158     boolean_t nonblock)
159 {
160 	avl_tree_t *tree = &rl->rl_tree;
161 	zfs_locked_range_t *lr;
162 	avl_index_t where;
163 	uint64_t orig_off = new->lr_offset;
164 	uint64_t orig_len = new->lr_length;
165 	zfs_rangelock_type_t orig_type = new->lr_type;
166 
167 	for (;;) {
168 		/*
169 		 * Call callback which can modify new->r_off,len,type.
170 		 * Note, the callback is used by the ZPL to handle appending
171 		 * and changing blocksizes.  It isn't needed for zvols.
172 		 */
173 		if (rl->rl_cb != NULL) {
174 			rl->rl_cb(new, rl->rl_arg);
175 		}
176 
177 		/*
178 		 * If the type was APPEND, the callback must convert it to
179 		 * WRITER.
180 		 */
181 		ASSERT3U(new->lr_type, ==, RL_WRITER);
182 
183 		/*
184 		 * First check for the usual case of no locks
185 		 */
186 		if (avl_numnodes(tree) == 0) {
187 			avl_add(tree, new);
188 			return (B_TRUE);
189 		}
190 
191 		/*
192 		 * Look for any locks in the range.
193 		 */
194 		lr = avl_find(tree, new, &where);
195 		if (lr != NULL)
196 			goto wait; /* already locked at same offset */
197 
198 		lr = avl_nearest(tree, where, AVL_AFTER);
199 		if (lr != NULL &&
200 		    lr->lr_offset < new->lr_offset + new->lr_length)
201 			goto wait;
202 
203 		lr = avl_nearest(tree, where, AVL_BEFORE);
204 		if (lr != NULL &&
205 		    lr->lr_offset + lr->lr_length > new->lr_offset)
206 			goto wait;
207 
208 		avl_insert(tree, new, where);
209 		return (B_TRUE);
210 wait:
211 		if (nonblock)
212 			return (B_FALSE);
213 		if (!lr->lr_write_wanted) {
214 			cv_init(&lr->lr_write_cv, NULL, CV_DEFAULT, NULL);
215 			lr->lr_write_wanted = B_TRUE;
216 		}
217 		cv_wait(&lr->lr_write_cv, &rl->rl_lock);
218 
219 		/* reset to original */
220 		new->lr_offset = orig_off;
221 		new->lr_length = orig_len;
222 		new->lr_type = orig_type;
223 	}
224 }
225 
226 /*
227  * If this is an original (non-proxy) lock then replace it by
228  * a proxy and return the proxy.
229  */
230 static zfs_locked_range_t *
231 zfs_rangelock_proxify(avl_tree_t *tree, zfs_locked_range_t *lr)
232 {
233 	zfs_locked_range_t *proxy;
234 
235 	if (lr->lr_proxy)
236 		return (lr); /* already a proxy */
237 
238 	ASSERT3U(lr->lr_count, ==, 1);
239 	ASSERT(lr->lr_write_wanted == B_FALSE);
240 	ASSERT(lr->lr_read_wanted == B_FALSE);
241 	avl_remove(tree, lr);
242 	lr->lr_count = 0;
243 
244 	/* create a proxy range lock */
245 	proxy = kmem_alloc(sizeof (zfs_locked_range_t), KM_SLEEP);
246 	proxy->lr_offset = lr->lr_offset;
247 	proxy->lr_length = lr->lr_length;
248 	proxy->lr_count = 1;
249 	proxy->lr_type = RL_READER;
250 	proxy->lr_proxy = B_TRUE;
251 	proxy->lr_write_wanted = B_FALSE;
252 	proxy->lr_read_wanted = B_FALSE;
253 	avl_add(tree, proxy);
254 
255 	return (proxy);
256 }
257 
258 /*
259  * Split the range lock at the supplied offset
260  * returning the *front* proxy.
261  */
262 static zfs_locked_range_t *
263 zfs_rangelock_split(avl_tree_t *tree, zfs_locked_range_t *lr, uint64_t off)
264 {
265 	zfs_locked_range_t *rear;
266 
267 	ASSERT3U(lr->lr_length, >, 1);
268 	ASSERT3U(off, >, lr->lr_offset);
269 	ASSERT3U(off, <, lr->lr_offset + lr->lr_length);
270 	ASSERT(lr->lr_write_wanted == B_FALSE);
271 	ASSERT(lr->lr_read_wanted == B_FALSE);
272 
273 	/* create the rear proxy range lock */
274 	rear = kmem_alloc(sizeof (zfs_locked_range_t), KM_SLEEP);
275 	rear->lr_offset = off;
276 	rear->lr_length = lr->lr_offset + lr->lr_length - off;
277 	rear->lr_count = lr->lr_count;
278 	rear->lr_type = RL_READER;
279 	rear->lr_proxy = B_TRUE;
280 	rear->lr_write_wanted = B_FALSE;
281 	rear->lr_read_wanted = B_FALSE;
282 
283 	zfs_locked_range_t *front = zfs_rangelock_proxify(tree, lr);
284 	front->lr_length = off - lr->lr_offset;
285 
286 	avl_insert_here(tree, rear, front, AVL_AFTER);
287 	return (front);
288 }
289 
290 /*
291  * Create and add a new proxy range lock for the supplied range.
292  */
293 static void
294 zfs_rangelock_new_proxy(avl_tree_t *tree, uint64_t off, uint64_t len)
295 {
296 	zfs_locked_range_t *lr;
297 
298 	ASSERT(len != 0);
299 	lr = kmem_alloc(sizeof (zfs_locked_range_t), KM_SLEEP);
300 	lr->lr_offset = off;
301 	lr->lr_length = len;
302 	lr->lr_count = 1;
303 	lr->lr_type = RL_READER;
304 	lr->lr_proxy = B_TRUE;
305 	lr->lr_write_wanted = B_FALSE;
306 	lr->lr_read_wanted = B_FALSE;
307 	avl_add(tree, lr);
308 }
309 
310 static void
311 zfs_rangelock_add_reader(avl_tree_t *tree, zfs_locked_range_t *new,
312     zfs_locked_range_t *prev, avl_index_t where)
313 {
314 	zfs_locked_range_t *next;
315 	uint64_t off = new->lr_offset;
316 	uint64_t len = new->lr_length;
317 
318 	/*
319 	 * prev arrives either:
320 	 * - pointing to an entry at the same offset
321 	 * - pointing to the entry with the closest previous offset whose
322 	 *   range may overlap with the new range
323 	 * - null, if there were no ranges starting before the new one
324 	 */
325 	if (prev != NULL) {
326 		if (prev->lr_offset + prev->lr_length <= off) {
327 			prev = NULL;
328 		} else if (prev->lr_offset != off) {
329 			/*
330 			 * convert to proxy if needed then
331 			 * split this entry and bump ref count
332 			 */
333 			prev = zfs_rangelock_split(tree, prev, off);
334 			prev = AVL_NEXT(tree, prev); /* move to rear range */
335 		}
336 	}
337 	ASSERT((prev == NULL) || (prev->lr_offset == off));
338 
339 	if (prev != NULL)
340 		next = prev;
341 	else
342 		next = avl_nearest(tree, where, AVL_AFTER);
343 
344 	if (next == NULL || off + len <= next->lr_offset) {
345 		/* no overlaps, use the original new rl_t in the tree */
346 		avl_insert(tree, new, where);
347 		return;
348 	}
349 
350 	if (off < next->lr_offset) {
351 		/* Add a proxy for initial range before the overlap */
352 		zfs_rangelock_new_proxy(tree, off, next->lr_offset - off);
353 	}
354 
355 	new->lr_count = 0; /* will use proxies in tree */
356 	/*
357 	 * We now search forward through the ranges, until we go past the end
358 	 * of the new range. For each entry we make it a proxy if it
359 	 * isn't already, then bump its reference count. If there's any
360 	 * gaps between the ranges then we create a new proxy range.
361 	 */
362 	for (prev = NULL; next; prev = next, next = AVL_NEXT(tree, next)) {
363 		if (off + len <= next->lr_offset)
364 			break;
365 		if (prev != NULL && prev->lr_offset + prev->lr_length <
366 		    next->lr_offset) {
367 			/* there's a gap */
368 			ASSERT3U(next->lr_offset, >,
369 			    prev->lr_offset + prev->lr_length);
370 			zfs_rangelock_new_proxy(tree,
371 			    prev->lr_offset + prev->lr_length,
372 			    next->lr_offset -
373 			    (prev->lr_offset + prev->lr_length));
374 		}
375 		if (off + len == next->lr_offset + next->lr_length) {
376 			/* exact overlap with end */
377 			next = zfs_rangelock_proxify(tree, next);
378 			next->lr_count++;
379 			return;
380 		}
381 		if (off + len < next->lr_offset + next->lr_length) {
382 			/* new range ends in the middle of this block */
383 			next = zfs_rangelock_split(tree, next, off + len);
384 			next->lr_count++;
385 			return;
386 		}
387 		ASSERT3U(off + len, >, next->lr_offset + next->lr_length);
388 		next = zfs_rangelock_proxify(tree, next);
389 		next->lr_count++;
390 	}
391 
392 	/* Add the remaining end range. */
393 	zfs_rangelock_new_proxy(tree, prev->lr_offset + prev->lr_length,
394 	    (off + len) - (prev->lr_offset + prev->lr_length));
395 }
396 
397 /*
398  * Check if a reader lock can be grabbed.  If not, fail immediately or sleep and
399  * recheck until available, depending on the value of the "nonblock" parameter.
400  */
401 static boolean_t
402 zfs_rangelock_enter_reader(zfs_rangelock_t *rl, zfs_locked_range_t *new,
403     boolean_t nonblock)
404 {
405 	avl_tree_t *tree = &rl->rl_tree;
406 	zfs_locked_range_t *prev, *next;
407 	avl_index_t where;
408 	uint64_t off = new->lr_offset;
409 	uint64_t len = new->lr_length;
410 
411 	/*
412 	 * Look for any writer locks in the range.
413 	 */
414 retry:
415 	prev = avl_find(tree, new, &where);
416 	if (prev == NULL)
417 		prev = avl_nearest(tree, where, AVL_BEFORE);
418 
419 	/*
420 	 * Check the previous range for a writer lock overlap.
421 	 */
422 	if (prev && (off < prev->lr_offset + prev->lr_length)) {
423 		if ((prev->lr_type == RL_WRITER) || (prev->lr_write_wanted)) {
424 			if (nonblock)
425 				return (B_FALSE);
426 			if (!prev->lr_read_wanted) {
427 				cv_init(&prev->lr_read_cv,
428 				    NULL, CV_DEFAULT, NULL);
429 				prev->lr_read_wanted = B_TRUE;
430 			}
431 			cv_wait(&prev->lr_read_cv, &rl->rl_lock);
432 			goto retry;
433 		}
434 		if (off + len < prev->lr_offset + prev->lr_length)
435 			goto got_lock;
436 	}
437 
438 	/*
439 	 * Search through the following ranges to see if there's
440 	 * write lock any overlap.
441 	 */
442 	if (prev != NULL)
443 		next = AVL_NEXT(tree, prev);
444 	else
445 		next = avl_nearest(tree, where, AVL_AFTER);
446 	for (; next != NULL; next = AVL_NEXT(tree, next)) {
447 		if (off + len <= next->lr_offset)
448 			goto got_lock;
449 		if ((next->lr_type == RL_WRITER) || (next->lr_write_wanted)) {
450 			if (nonblock)
451 				return (B_FALSE);
452 			if (!next->lr_read_wanted) {
453 				cv_init(&next->lr_read_cv,
454 				    NULL, CV_DEFAULT, NULL);
455 				next->lr_read_wanted = B_TRUE;
456 			}
457 			cv_wait(&next->lr_read_cv, &rl->rl_lock);
458 			goto retry;
459 		}
460 		if (off + len <= next->lr_offset + next->lr_length)
461 			goto got_lock;
462 	}
463 
464 got_lock:
465 	/*
466 	 * Add the read lock, which may involve splitting existing
467 	 * locks and bumping ref counts (r_count).
468 	 */
469 	zfs_rangelock_add_reader(tree, new, prev, where);
470 	return (B_TRUE);
471 }
472 
473 /*
474  * Lock a range (offset, length) as either shared (RL_READER) or exclusive
475  * (RL_WRITER or RL_APPEND).  If RL_APPEND is specified, rl_cb() will convert
476  * it to a RL_WRITER lock (with the offset at the end of the file).  Returns
477  * the range lock structure for later unlocking (or reduce range if the
478  * entire file is locked as RL_WRITER), or NULL if nonblock is true and the
479  * lock could not be acquired immediately.
480  */
481 static zfs_locked_range_t *
482 zfs_rangelock_enter_impl(zfs_rangelock_t *rl, uint64_t off, uint64_t len,
483     zfs_rangelock_type_t type, boolean_t nonblock)
484 {
485 	zfs_locked_range_t *new;
486 
487 	ASSERT(type == RL_READER || type == RL_WRITER || type == RL_APPEND);
488 
489 	new = kmem_alloc(sizeof (zfs_locked_range_t), KM_SLEEP);
490 	new->lr_rangelock = rl;
491 	new->lr_offset = off;
492 	if (len + off < off)	/* overflow */
493 		len = UINT64_MAX - off;
494 	new->lr_length = len;
495 	new->lr_count = 1; /* assume it's going to be in the tree */
496 	new->lr_type = type;
497 	new->lr_proxy = B_FALSE;
498 	new->lr_write_wanted = B_FALSE;
499 	new->lr_read_wanted = B_FALSE;
500 
501 	mutex_enter(&rl->rl_lock);
502 	if (type == RL_READER) {
503 		/*
504 		 * First check for the usual case of no locks
505 		 */
506 		if (avl_numnodes(&rl->rl_tree) == 0) {
507 			avl_add(&rl->rl_tree, new);
508 		} else if (!zfs_rangelock_enter_reader(rl, new, nonblock)) {
509 			kmem_free(new, sizeof (*new));
510 			new = NULL;
511 		}
512 	} else if (!zfs_rangelock_enter_writer(rl, new, nonblock)) {
513 		kmem_free(new, sizeof (*new));
514 		new = NULL;
515 	}
516 	mutex_exit(&rl->rl_lock);
517 	return (new);
518 }
519 
520 zfs_locked_range_t *
521 zfs_rangelock_enter(zfs_rangelock_t *rl, uint64_t off, uint64_t len,
522     zfs_rangelock_type_t type)
523 {
524 	return (zfs_rangelock_enter_impl(rl, off, len, type, B_FALSE));
525 }
526 
527 zfs_locked_range_t *
528 zfs_rangelock_tryenter(zfs_rangelock_t *rl, uint64_t off, uint64_t len,
529     zfs_rangelock_type_t type)
530 {
531 	return (zfs_rangelock_enter_impl(rl, off, len, type, B_TRUE));
532 }
533 
534 /*
535  * Safely free the zfs_locked_range_t.
536  */
537 static void
538 zfs_rangelock_free(zfs_locked_range_t *lr)
539 {
540 	if (lr->lr_write_wanted)
541 		cv_destroy(&lr->lr_write_cv);
542 
543 	if (lr->lr_read_wanted)
544 		cv_destroy(&lr->lr_read_cv);
545 
546 	kmem_free(lr, sizeof (zfs_locked_range_t));
547 }
548 
549 /*
550  * Unlock a reader lock
551  */
552 static void
553 zfs_rangelock_exit_reader(zfs_rangelock_t *rl, zfs_locked_range_t *remove,
554     list_t *free_list)
555 {
556 	avl_tree_t *tree = &rl->rl_tree;
557 	uint64_t len;
558 
559 	/*
560 	 * The common case is when the remove entry is in the tree
561 	 * (cnt == 1) meaning there's been no other reader locks overlapping
562 	 * with this one. Otherwise the remove entry will have been
563 	 * removed from the tree and replaced by proxies (one or
564 	 * more ranges mapping to the entire range).
565 	 */
566 	if (remove->lr_count == 1) {
567 		avl_remove(tree, remove);
568 		if (remove->lr_write_wanted)
569 			cv_broadcast(&remove->lr_write_cv);
570 		if (remove->lr_read_wanted)
571 			cv_broadcast(&remove->lr_read_cv);
572 		list_insert_tail(free_list, remove);
573 	} else {
574 		ASSERT0(remove->lr_count);
575 		ASSERT0(remove->lr_write_wanted);
576 		ASSERT0(remove->lr_read_wanted);
577 		/*
578 		 * Find start proxy representing this reader lock,
579 		 * then decrement ref count on all proxies
580 		 * that make up this range, freeing them as needed.
581 		 */
582 		zfs_locked_range_t *lr = avl_find(tree, remove, NULL);
583 		ASSERT3P(lr, !=, NULL);
584 		ASSERT3U(lr->lr_count, !=, 0);
585 		ASSERT3U(lr->lr_type, ==, RL_READER);
586 		zfs_locked_range_t *next = NULL;
587 		for (len = remove->lr_length; len != 0; lr = next) {
588 			len -= lr->lr_length;
589 			if (len != 0) {
590 				next = AVL_NEXT(tree, lr);
591 				ASSERT3P(next, !=, NULL);
592 				ASSERT3U(lr->lr_offset + lr->lr_length, ==,
593 				    next->lr_offset);
594 				ASSERT3U(next->lr_count, !=, 0);
595 				ASSERT3U(next->lr_type, ==, RL_READER);
596 			}
597 			lr->lr_count--;
598 			if (lr->lr_count == 0) {
599 				avl_remove(tree, lr);
600 				if (lr->lr_write_wanted)
601 					cv_broadcast(&lr->lr_write_cv);
602 				if (lr->lr_read_wanted)
603 					cv_broadcast(&lr->lr_read_cv);
604 				list_insert_tail(free_list, lr);
605 			}
606 		}
607 		kmem_free(remove, sizeof (zfs_locked_range_t));
608 	}
609 }
610 
611 /*
612  * Unlock range and destroy range lock structure.
613  */
614 void
615 zfs_rangelock_exit(zfs_locked_range_t *lr)
616 {
617 	zfs_rangelock_t *rl = lr->lr_rangelock;
618 	list_t free_list;
619 	zfs_locked_range_t *free_lr;
620 
621 	ASSERT(lr->lr_type == RL_WRITER || lr->lr_type == RL_READER);
622 	ASSERT(lr->lr_count == 1 || lr->lr_count == 0);
623 	ASSERT(!lr->lr_proxy);
624 
625 	/*
626 	 * The free list is used to defer the cv_destroy() and
627 	 * subsequent kmem_free until after the mutex is dropped.
628 	 */
629 	list_create(&free_list, sizeof (zfs_locked_range_t),
630 	    offsetof(zfs_locked_range_t, lr_node));
631 
632 	mutex_enter(&rl->rl_lock);
633 	if (lr->lr_type == RL_WRITER) {
634 		/* writer locks can't be shared or split */
635 		avl_remove(&rl->rl_tree, lr);
636 		if (lr->lr_write_wanted)
637 			cv_broadcast(&lr->lr_write_cv);
638 		if (lr->lr_read_wanted)
639 			cv_broadcast(&lr->lr_read_cv);
640 		list_insert_tail(&free_list, lr);
641 	} else {
642 		/*
643 		 * lock may be shared, let rangelock_exit_reader()
644 		 * release the lock and free the zfs_locked_range_t.
645 		 */
646 		zfs_rangelock_exit_reader(rl, lr, &free_list);
647 	}
648 	mutex_exit(&rl->rl_lock);
649 
650 	while ((free_lr = list_remove_head(&free_list)) != NULL)
651 		zfs_rangelock_free(free_lr);
652 
653 	list_destroy(&free_list);
654 }
655 
656 /*
657  * Reduce range locked as RL_WRITER from whole file to specified range.
658  * Asserts the whole file is exclusively locked and so there's only one
659  * entry in the tree.
660  */
661 void
662 zfs_rangelock_reduce(zfs_locked_range_t *lr, uint64_t off, uint64_t len)
663 {
664 	zfs_rangelock_t *rl = lr->lr_rangelock;
665 
666 	/* Ensure there are no other locks */
667 	ASSERT3U(avl_numnodes(&rl->rl_tree), ==, 1);
668 	ASSERT3U(lr->lr_offset, ==, 0);
669 	ASSERT3U(lr->lr_type, ==, RL_WRITER);
670 	ASSERT(!lr->lr_proxy);
671 	ASSERT3U(lr->lr_length, ==, UINT64_MAX);
672 	ASSERT3U(lr->lr_count, ==, 1);
673 
674 	mutex_enter(&rl->rl_lock);
675 	lr->lr_offset = off;
676 	lr->lr_length = len;
677 	mutex_exit(&rl->rl_lock);
678 	if (lr->lr_write_wanted)
679 		cv_broadcast(&lr->lr_write_cv);
680 	if (lr->lr_read_wanted)
681 		cv_broadcast(&lr->lr_read_cv);
682 }
683 
684 #if defined(_KERNEL)
685 EXPORT_SYMBOL(zfs_rangelock_init);
686 EXPORT_SYMBOL(zfs_rangelock_fini);
687 EXPORT_SYMBOL(zfs_rangelock_enter);
688 EXPORT_SYMBOL(zfs_rangelock_tryenter);
689 EXPORT_SYMBOL(zfs_rangelock_exit);
690 EXPORT_SYMBOL(zfs_rangelock_reduce);
691 #endif
692