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
zfs_rangelock_compare(const void * arg1,const void * arg2)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
zfs_rangelock_init(zfs_rangelock_t * rl,zfs_rangelock_cb_t * cb,void * arg)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
zfs_rangelock_fini(zfs_rangelock_t * rl)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
zfs_rangelock_enter_writer(zfs_rangelock_t * rl,zfs_locked_range_t * new,boolean_t nonblock)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 *
zfs_rangelock_proxify(avl_tree_t * tree,zfs_locked_range_t * lr)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 *
zfs_rangelock_split(avl_tree_t * tree,zfs_locked_range_t * lr,uint64_t off)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
zfs_rangelock_new_proxy(avl_tree_t * tree,uint64_t off,uint64_t len)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
zfs_rangelock_add_reader(avl_tree_t * tree,zfs_locked_range_t * new,zfs_locked_range_t * prev,avl_index_t where)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
zfs_rangelock_enter_reader(zfs_rangelock_t * rl,zfs_locked_range_t * new,boolean_t nonblock)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 *
zfs_rangelock_enter_impl(zfs_rangelock_t * rl,uint64_t off,uint64_t len,zfs_rangelock_type_t type,boolean_t nonblock)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 *
zfs_rangelock_enter(zfs_rangelock_t * rl,uint64_t off,uint64_t len,zfs_rangelock_type_t type)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 *
zfs_rangelock_tryenter(zfs_rangelock_t * rl,uint64_t off,uint64_t len,zfs_rangelock_type_t type)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
zfs_rangelock_free(zfs_locked_range_t * lr)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
zfs_rangelock_exit_reader(zfs_rangelock_t * rl,zfs_locked_range_t * remove,list_t * free_list)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
zfs_rangelock_exit(zfs_locked_range_t * lr)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
zfs_rangelock_reduce(zfs_locked_range_t * lr,uint64_t off,uint64_t len)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