xref: /freebsd/sys/contrib/openzfs/module/zfs/dsl_deadlist.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * CDDL HEADER START
4  *
5  * The contents of this file are subject to the terms of the
6  * Common Development and Distribution License (the "License").
7  * You may not use this file except in compliance with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or https://opensource.org/licenses/CDDL-1.0.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
25  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26  */
27 
28 #include <sys/dmu.h>
29 #include <sys/zap.h>
30 #include <sys/zfs_context.h>
31 #include <sys/dsl_pool.h>
32 #include <sys/dsl_dataset.h>
33 
34 /*
35  * Deadlist concurrency:
36  *
37  * Deadlists can only be modified from the syncing thread.
38  *
39  * Except for dsl_deadlist_insert(), it can only be modified with the
40  * dp_config_rwlock held with RW_WRITER.
41  *
42  * The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
43  * be called concurrently, from open context, with the dl_config_rwlock held
44  * with RW_READER.
45  *
46  * Therefore, we only need to provide locking between dsl_deadlist_insert() and
47  * the accessors, protecting:
48  *     dl_phys->dl_used,comp,uncomp
49  *     and protecting the dl_tree from being loaded.
50  * The locking is provided by dl_lock.  Note that locking on the bpobj_t
51  * provides its own locking, and dl_oldfmt is immutable.
52  */
53 
54 /*
55  * Livelist Overview
56  * ================
57  *
58  * Livelists use the same 'deadlist_t' struct as deadlists and are also used
59  * to track blkptrs over the lifetime of a dataset. Livelists however, belong
60  * to clones and track the blkptrs that are clone-specific (were born after
61  * the clone's creation). The exception is embedded block pointers which are
62  * not included in livelists because they do not need to be freed.
63  *
64  * When it comes time to delete the clone, the livelist provides a quick
65  * reference as to what needs to be freed. For this reason, livelists also track
66  * when clone-specific blkptrs are freed before deletion to prevent double
67  * frees. Each blkptr in a livelist is marked as a FREE or an ALLOC and the
68  * deletion algorithm iterates backwards over the livelist, matching
69  * FREE/ALLOC pairs and then freeing those ALLOCs which remain. livelists
70  * are also updated in the case when blkptrs are remapped: the old version
71  * of the blkptr is cancelled out with a FREE and the new version is tracked
72  * with an ALLOC.
73  *
74  * To bound the amount of memory required for deletion, livelists over a
75  * certain size are spread over multiple entries. Entries are grouped by
76  * birth txg so we can be sure the ALLOC/FREE pair for a given blkptr will
77  * be in the same entry. This allows us to delete livelists incrementally
78  * over multiple syncs, one entry at a time.
79  *
80  * During the lifetime of the clone, livelists can get extremely large.
81  * Their size is managed by periodic condensing (preemptively cancelling out
82  * FREE/ALLOC pairs). Livelists are disabled when a clone is promoted or when
83  * the shared space between the clone and its origin is so small that it
84  * doesn't make sense to use livelists anymore.
85  */
86 
87 /*
88  * The threshold sublist size at which we create a new sub-livelist for the
89  * next txg. However, since blkptrs of the same transaction group must be in
90  * the same sub-list, the actual sublist size may exceed this. When picking the
91  * size we had to balance the fact that larger sublists mean fewer sublists
92  * (decreasing the cost of insertion) against the consideration that sublists
93  * will be loaded into memory and shouldn't take up an inordinate amount of
94  * space. We settled on ~500000 entries, corresponding to roughly 128M.
95  */
96 uint64_t zfs_livelist_max_entries = 500000;
97 
98 /*
99  * We can approximate how much of a performance gain a livelist will give us
100  * based on the percentage of blocks shared between the clone and its origin.
101  * 0 percent shared means that the clone has completely diverged and that the
102  * old method is maximally effective: every read from the block tree will
103  * result in lots of frees. Livelists give us gains when they track blocks
104  * scattered across the tree, when one read in the old method might only
105  * result in a few frees. Once the clone has been overwritten enough,
106  * writes are no longer sparse and we'll no longer get much of a benefit from
107  * tracking them with a livelist. We chose a lower limit of 75 percent shared
108  * (25 percent overwritten). This means that 1/4 of all block pointers will be
109  * freed (e.g. each read frees 256, out of a max of 1024) so we expect livelists
110  * to make deletion 4x faster. Once the amount of shared space drops below this
111  * threshold, the clone will revert to the old deletion method.
112  */
113 int zfs_livelist_min_percent_shared = 75;
114 
115 static int
dsl_deadlist_compare(const void * arg1,const void * arg2)116 dsl_deadlist_compare(const void *arg1, const void *arg2)
117 {
118 	const dsl_deadlist_entry_t *dle1 = arg1;
119 	const dsl_deadlist_entry_t *dle2 = arg2;
120 
121 	return (TREE_CMP(dle1->dle_mintxg, dle2->dle_mintxg));
122 }
123 
124 static int
dsl_deadlist_cache_compare(const void * arg1,const void * arg2)125 dsl_deadlist_cache_compare(const void *arg1, const void *arg2)
126 {
127 	const dsl_deadlist_cache_entry_t *dlce1 = arg1;
128 	const dsl_deadlist_cache_entry_t *dlce2 = arg2;
129 
130 	return (TREE_CMP(dlce1->dlce_mintxg, dlce2->dlce_mintxg));
131 }
132 
133 static void
dsl_deadlist_load_tree(dsl_deadlist_t * dl)134 dsl_deadlist_load_tree(dsl_deadlist_t *dl)
135 {
136 	zap_cursor_t zc;
137 	zap_attribute_t *za;
138 	int error;
139 
140 	ASSERT(MUTEX_HELD(&dl->dl_lock));
141 
142 	ASSERT(!dl->dl_oldfmt);
143 	if (dl->dl_havecache) {
144 		/*
145 		 * After loading the tree, the caller may modify the tree,
146 		 * e.g. to add or remove nodes, or to make a node no longer
147 		 * refer to the empty_bpobj.  These changes would make the
148 		 * dl_cache incorrect.  Therefore we discard the cache here,
149 		 * so that it can't become incorrect.
150 		 */
151 		dsl_deadlist_cache_entry_t *dlce;
152 		void *cookie = NULL;
153 		while ((dlce = avl_destroy_nodes(&dl->dl_cache, &cookie))
154 		    != NULL) {
155 			kmem_free(dlce, sizeof (*dlce));
156 		}
157 		avl_destroy(&dl->dl_cache);
158 		dl->dl_havecache = B_FALSE;
159 	}
160 	if (dl->dl_havetree)
161 		return;
162 
163 	za = zap_attribute_alloc();
164 	avl_create(&dl->dl_tree, dsl_deadlist_compare,
165 	    sizeof (dsl_deadlist_entry_t),
166 	    offsetof(dsl_deadlist_entry_t, dle_node));
167 	for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
168 	    (error = zap_cursor_retrieve(&zc, za)) == 0;
169 	    zap_cursor_advance(&zc)) {
170 		dsl_deadlist_entry_t *dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
171 		dle->dle_mintxg = zfs_strtonum(za->za_name, NULL);
172 
173 		/*
174 		 * Prefetch all the bpobj's so that we do that i/o
175 		 * in parallel.  Then open them all in a second pass.
176 		 */
177 		dle->dle_bpobj.bpo_object = za->za_first_integer;
178 		dmu_prefetch_dnode(dl->dl_os, dle->dle_bpobj.bpo_object,
179 		    ZIO_PRIORITY_SYNC_READ);
180 
181 		avl_add(&dl->dl_tree, dle);
182 	}
183 	VERIFY3U(error, ==, ENOENT);
184 	zap_cursor_fini(&zc);
185 	zap_attribute_free(za);
186 
187 	for (dsl_deadlist_entry_t *dle = avl_first(&dl->dl_tree);
188 	    dle != NULL; dle = AVL_NEXT(&dl->dl_tree, dle)) {
189 		VERIFY0(bpobj_open(&dle->dle_bpobj, dl->dl_os,
190 		    dle->dle_bpobj.bpo_object));
191 	}
192 	dl->dl_havetree = B_TRUE;
193 }
194 
195 /*
196  * Load only the non-empty bpobj's into the dl_cache.  The cache is an analog
197  * of the dl_tree, but contains only non-empty_bpobj nodes from the ZAP. It
198  * is used only for gathering space statistics.  The dl_cache has two
199  * advantages over the dl_tree:
200  *
201  * 1. Loading the dl_cache is ~5x faster than loading the dl_tree (if it's
202  * mostly empty_bpobj's), due to less CPU overhead to open the empty_bpobj
203  * many times and to inquire about its (zero) space stats many times.
204  *
205  * 2. The dl_cache uses less memory than the dl_tree.  We only need to load
206  * the dl_tree of snapshots when deleting a snapshot, after which we free the
207  * dl_tree with dsl_deadlist_discard_tree
208  */
209 static void
dsl_deadlist_load_cache(dsl_deadlist_t * dl)210 dsl_deadlist_load_cache(dsl_deadlist_t *dl)
211 {
212 	zap_cursor_t zc;
213 	zap_attribute_t *za;
214 	int error;
215 
216 	ASSERT(MUTEX_HELD(&dl->dl_lock));
217 
218 	ASSERT(!dl->dl_oldfmt);
219 	if (dl->dl_havecache)
220 		return;
221 
222 	uint64_t empty_bpobj = dmu_objset_pool(dl->dl_os)->dp_empty_bpobj;
223 
224 	avl_create(&dl->dl_cache, dsl_deadlist_cache_compare,
225 	    sizeof (dsl_deadlist_cache_entry_t),
226 	    offsetof(dsl_deadlist_cache_entry_t, dlce_node));
227 	za = zap_attribute_alloc();
228 	for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
229 	    (error = zap_cursor_retrieve(&zc, za)) == 0;
230 	    zap_cursor_advance(&zc)) {
231 		if (za->za_first_integer == empty_bpobj)
232 			continue;
233 		dsl_deadlist_cache_entry_t *dlce =
234 		    kmem_zalloc(sizeof (*dlce), KM_SLEEP);
235 		dlce->dlce_mintxg = zfs_strtonum(za->za_name, NULL);
236 
237 		/*
238 		 * Prefetch all the bpobj's so that we do that i/o
239 		 * in parallel.  Then open them all in a second pass.
240 		 */
241 		dlce->dlce_bpobj = za->za_first_integer;
242 		dmu_prefetch_dnode(dl->dl_os, dlce->dlce_bpobj,
243 		    ZIO_PRIORITY_SYNC_READ);
244 		avl_add(&dl->dl_cache, dlce);
245 	}
246 	VERIFY3U(error, ==, ENOENT);
247 	zap_cursor_fini(&zc);
248 	zap_attribute_free(za);
249 
250 	for (dsl_deadlist_cache_entry_t *dlce = avl_first(&dl->dl_cache);
251 	    dlce != NULL; dlce = AVL_NEXT(&dl->dl_cache, dlce)) {
252 		bpobj_t bpo;
253 		VERIFY0(bpobj_open(&bpo, dl->dl_os, dlce->dlce_bpobj));
254 
255 		VERIFY0(bpobj_space(&bpo,
256 		    &dlce->dlce_bytes, &dlce->dlce_comp, &dlce->dlce_uncomp));
257 		bpobj_close(&bpo);
258 	}
259 	dl->dl_havecache = B_TRUE;
260 }
261 
262 /*
263  * Discard the tree to save memory.
264  */
265 void
dsl_deadlist_discard_tree(dsl_deadlist_t * dl)266 dsl_deadlist_discard_tree(dsl_deadlist_t *dl)
267 {
268 	mutex_enter(&dl->dl_lock);
269 
270 	if (!dl->dl_havetree) {
271 		mutex_exit(&dl->dl_lock);
272 		return;
273 	}
274 	dsl_deadlist_entry_t *dle;
275 	void *cookie = NULL;
276 	while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie)) != NULL) {
277 		bpobj_close(&dle->dle_bpobj);
278 		kmem_free(dle, sizeof (*dle));
279 	}
280 	avl_destroy(&dl->dl_tree);
281 
282 	dl->dl_havetree = B_FALSE;
283 	mutex_exit(&dl->dl_lock);
284 }
285 
286 void
dsl_deadlist_iterate(dsl_deadlist_t * dl,deadlist_iter_t func,void * args)287 dsl_deadlist_iterate(dsl_deadlist_t *dl, deadlist_iter_t func, void *args)
288 {
289 	dsl_deadlist_entry_t *dle;
290 
291 	ASSERT(dsl_deadlist_is_open(dl));
292 
293 	mutex_enter(&dl->dl_lock);
294 	dsl_deadlist_load_tree(dl);
295 	mutex_exit(&dl->dl_lock);
296 	for (dle = avl_first(&dl->dl_tree); dle != NULL;
297 	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
298 		if (func(args, dle) != 0)
299 			break;
300 	}
301 }
302 
303 int
dsl_deadlist_open(dsl_deadlist_t * dl,objset_t * os,uint64_t object)304 dsl_deadlist_open(dsl_deadlist_t *dl, objset_t *os, uint64_t object)
305 {
306 	dmu_object_info_t doi;
307 	int err;
308 
309 	ASSERT(!dsl_deadlist_is_open(dl));
310 
311 	mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
312 	dl->dl_os = os;
313 	dl->dl_object = object;
314 	err = dmu_bonus_hold(os, object, dl, &dl->dl_dbuf);
315 	if (err != 0)
316 		return (err);
317 	dmu_object_info_from_db(dl->dl_dbuf, &doi);
318 	if (doi.doi_type == DMU_OT_BPOBJ) {
319 		dmu_buf_rele(dl->dl_dbuf, dl);
320 		dl->dl_dbuf = NULL;
321 		dl->dl_oldfmt = B_TRUE;
322 		return (bpobj_open(&dl->dl_bpobj, os, object));
323 	}
324 
325 	dl->dl_oldfmt = B_FALSE;
326 	dl->dl_phys = dl->dl_dbuf->db_data;
327 	dl->dl_havetree = B_FALSE;
328 	dl->dl_havecache = B_FALSE;
329 	return (0);
330 }
331 
332 boolean_t
dsl_deadlist_is_open(dsl_deadlist_t * dl)333 dsl_deadlist_is_open(dsl_deadlist_t *dl)
334 {
335 	return (dl->dl_os != NULL);
336 }
337 
338 void
dsl_deadlist_close(dsl_deadlist_t * dl)339 dsl_deadlist_close(dsl_deadlist_t *dl)
340 {
341 	ASSERT(dsl_deadlist_is_open(dl));
342 	mutex_destroy(&dl->dl_lock);
343 
344 	if (dl->dl_oldfmt) {
345 		dl->dl_oldfmt = B_FALSE;
346 		bpobj_close(&dl->dl_bpobj);
347 		dl->dl_os = NULL;
348 		dl->dl_object = 0;
349 		return;
350 	}
351 
352 	if (dl->dl_havetree) {
353 		dsl_deadlist_entry_t *dle;
354 		void *cookie = NULL;
355 		while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
356 		    != NULL) {
357 			bpobj_close(&dle->dle_bpobj);
358 			kmem_free(dle, sizeof (*dle));
359 		}
360 		avl_destroy(&dl->dl_tree);
361 	}
362 	if (dl->dl_havecache) {
363 		dsl_deadlist_cache_entry_t *dlce;
364 		void *cookie = NULL;
365 		while ((dlce = avl_destroy_nodes(&dl->dl_cache, &cookie))
366 		    != NULL) {
367 			kmem_free(dlce, sizeof (*dlce));
368 		}
369 		avl_destroy(&dl->dl_cache);
370 	}
371 	dmu_buf_rele(dl->dl_dbuf, dl);
372 	dl->dl_dbuf = NULL;
373 	dl->dl_phys = NULL;
374 	dl->dl_os = NULL;
375 	dl->dl_object = 0;
376 }
377 
378 uint64_t
dsl_deadlist_alloc(objset_t * os,dmu_tx_t * tx)379 dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx)
380 {
381 	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
382 		return (bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx));
383 	return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
384 	    sizeof (dsl_deadlist_phys_t), tx));
385 }
386 
387 void
dsl_deadlist_free(objset_t * os,uint64_t dlobj,dmu_tx_t * tx)388 dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx)
389 {
390 	dmu_object_info_t doi;
391 	zap_cursor_t zc;
392 	zap_attribute_t *za;
393 	int error;
394 
395 	VERIFY0(dmu_object_info(os, dlobj, &doi));
396 	if (doi.doi_type == DMU_OT_BPOBJ) {
397 		bpobj_free(os, dlobj, tx);
398 		return;
399 	}
400 
401 	za = zap_attribute_alloc();
402 	for (zap_cursor_init(&zc, os, dlobj);
403 	    (error = zap_cursor_retrieve(&zc, za)) == 0;
404 	    zap_cursor_advance(&zc)) {
405 		uint64_t obj = za->za_first_integer;
406 		if (obj == dmu_objset_pool(os)->dp_empty_bpobj)
407 			bpobj_decr_empty(os, tx);
408 		else
409 			bpobj_free(os, obj, tx);
410 	}
411 	VERIFY3U(error, ==, ENOENT);
412 	zap_cursor_fini(&zc);
413 	zap_attribute_free(za);
414 	VERIFY0(dmu_object_free(os, dlobj, tx));
415 }
416 
417 static void
dle_enqueue(dsl_deadlist_t * dl,dsl_deadlist_entry_t * dle,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)418 dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
419     const blkptr_t *bp, boolean_t bp_freed, dmu_tx_t *tx)
420 {
421 	ASSERT(MUTEX_HELD(&dl->dl_lock));
422 	if (dle->dle_bpobj.bpo_object ==
423 	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
424 		uint64_t obj = bpobj_alloc(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
425 		bpobj_close(&dle->dle_bpobj);
426 		bpobj_decr_empty(dl->dl_os, tx);
427 		VERIFY0(bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
428 		VERIFY0(zap_update_int_key(dl->dl_os, dl->dl_object,
429 		    dle->dle_mintxg, obj, tx));
430 	}
431 	bpobj_enqueue(&dle->dle_bpobj, bp, bp_freed, tx);
432 }
433 
434 static void
dle_enqueue_subobj(dsl_deadlist_t * dl,dsl_deadlist_entry_t * dle,uint64_t obj,dmu_tx_t * tx)435 dle_enqueue_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
436     uint64_t obj, dmu_tx_t *tx)
437 {
438 	ASSERT(MUTEX_HELD(&dl->dl_lock));
439 	if (dle->dle_bpobj.bpo_object !=
440 	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
441 		bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
442 	} else {
443 		bpobj_close(&dle->dle_bpobj);
444 		bpobj_decr_empty(dl->dl_os, tx);
445 		VERIFY0(bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
446 		VERIFY0(zap_update_int_key(dl->dl_os, dl->dl_object,
447 		    dle->dle_mintxg, obj, tx));
448 	}
449 }
450 
451 /*
452  * Prefetch metadata required for dle_enqueue_subobj().
453  */
454 static void
dle_prefetch_subobj(dsl_deadlist_t * dl,dsl_deadlist_entry_t * dle,uint64_t obj)455 dle_prefetch_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
456     uint64_t obj)
457 {
458 	if (dle->dle_bpobj.bpo_object !=
459 	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj)
460 		bpobj_prefetch_subobj(&dle->dle_bpobj, obj);
461 }
462 
463 void
dsl_deadlist_insert(dsl_deadlist_t * dl,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)464 dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, boolean_t bp_freed,
465     dmu_tx_t *tx)
466 {
467 	dsl_deadlist_entry_t dle_tofind;
468 	dsl_deadlist_entry_t *dle;
469 	avl_index_t where;
470 
471 	if (dl->dl_oldfmt) {
472 		bpobj_enqueue(&dl->dl_bpobj, bp, bp_freed, tx);
473 		return;
474 	}
475 
476 	mutex_enter(&dl->dl_lock);
477 	dsl_deadlist_load_tree(dl);
478 
479 	dmu_buf_will_dirty(dl->dl_dbuf, tx);
480 
481 	int sign = bp_freed ? -1 : +1;
482 	dl->dl_phys->dl_used +=
483 	    sign * bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
484 	dl->dl_phys->dl_comp += sign * BP_GET_PSIZE(bp);
485 	dl->dl_phys->dl_uncomp += sign * BP_GET_UCSIZE(bp);
486 
487 	dle_tofind.dle_mintxg = BP_GET_LOGICAL_BIRTH(bp);
488 	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
489 	if (dle == NULL)
490 		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
491 	else
492 		dle = AVL_PREV(&dl->dl_tree, dle);
493 
494 	if (dle == NULL) {
495 		zfs_panic_recover("blkptr at %p has invalid BLK_BIRTH %llu",
496 		    bp, (longlong_t)BP_GET_LOGICAL_BIRTH(bp));
497 		dle = avl_first(&dl->dl_tree);
498 	}
499 
500 	ASSERT3P(dle, !=, NULL);
501 	dle_enqueue(dl, dle, bp, bp_freed, tx);
502 	mutex_exit(&dl->dl_lock);
503 }
504 
505 int
dsl_deadlist_insert_alloc_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)506 dsl_deadlist_insert_alloc_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
507 {
508 	dsl_deadlist_t *dl = arg;
509 	dsl_deadlist_insert(dl, bp, B_FALSE, tx);
510 	return (0);
511 }
512 
513 int
dsl_deadlist_insert_free_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)514 dsl_deadlist_insert_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
515 {
516 	dsl_deadlist_t *dl = arg;
517 	dsl_deadlist_insert(dl, bp, B_TRUE, tx);
518 	return (0);
519 }
520 
521 /*
522  * Insert new key in deadlist, which must be > all current entries.
523  * mintxg is not inclusive.
524  */
525 void
dsl_deadlist_add_key(dsl_deadlist_t * dl,uint64_t mintxg,dmu_tx_t * tx)526 dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
527 {
528 	uint64_t obj;
529 	dsl_deadlist_entry_t *dle;
530 
531 	if (dl->dl_oldfmt)
532 		return;
533 
534 	dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
535 	dle->dle_mintxg = mintxg;
536 
537 	mutex_enter(&dl->dl_lock);
538 	dsl_deadlist_load_tree(dl);
539 
540 	obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
541 	VERIFY0(bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
542 	avl_add(&dl->dl_tree, dle);
543 
544 	VERIFY0(zap_add_int_key(dl->dl_os, dl->dl_object,
545 	    mintxg, obj, tx));
546 	mutex_exit(&dl->dl_lock);
547 }
548 
549 /*
550  * Remove this key, merging its entries into the previous key.
551  */
552 void
dsl_deadlist_remove_key(dsl_deadlist_t * dl,uint64_t mintxg,dmu_tx_t * tx)553 dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
554 {
555 	dsl_deadlist_entry_t dle_tofind;
556 	dsl_deadlist_entry_t *dle, *dle_prev;
557 
558 	if (dl->dl_oldfmt)
559 		return;
560 	mutex_enter(&dl->dl_lock);
561 	dsl_deadlist_load_tree(dl);
562 
563 	dle_tofind.dle_mintxg = mintxg;
564 	dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
565 	ASSERT3P(dle, !=, NULL);
566 	dle_prev = AVL_PREV(&dl->dl_tree, dle);
567 	ASSERT3P(dle_prev, !=, NULL);
568 
569 	dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx);
570 
571 	avl_remove(&dl->dl_tree, dle);
572 	bpobj_close(&dle->dle_bpobj);
573 	kmem_free(dle, sizeof (*dle));
574 
575 	VERIFY0(zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
576 	mutex_exit(&dl->dl_lock);
577 }
578 
579 /*
580  * Remove a deadlist entry and all of its contents by removing the entry from
581  * the deadlist's avl tree, freeing the entry's bpobj and adjusting the
582  * deadlist's space accounting accordingly.
583  */
584 void
dsl_deadlist_remove_entry(dsl_deadlist_t * dl,uint64_t mintxg,dmu_tx_t * tx)585 dsl_deadlist_remove_entry(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
586 {
587 	uint64_t used, comp, uncomp;
588 	dsl_deadlist_entry_t dle_tofind;
589 	dsl_deadlist_entry_t *dle;
590 	objset_t *os = dl->dl_os;
591 
592 	if (dl->dl_oldfmt)
593 		return;
594 
595 	mutex_enter(&dl->dl_lock);
596 	dsl_deadlist_load_tree(dl);
597 
598 	dle_tofind.dle_mintxg = mintxg;
599 	dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
600 	VERIFY3P(dle, !=, NULL);
601 
602 	avl_remove(&dl->dl_tree, dle);
603 	VERIFY0(zap_remove_int(os, dl->dl_object, mintxg, tx));
604 	VERIFY0(bpobj_space(&dle->dle_bpobj, &used, &comp, &uncomp));
605 	dmu_buf_will_dirty(dl->dl_dbuf, tx);
606 	dl->dl_phys->dl_used -= used;
607 	dl->dl_phys->dl_comp -= comp;
608 	dl->dl_phys->dl_uncomp -= uncomp;
609 	if (dle->dle_bpobj.bpo_object == dmu_objset_pool(os)->dp_empty_bpobj) {
610 		bpobj_decr_empty(os, tx);
611 	} else {
612 		bpobj_free(os, dle->dle_bpobj.bpo_object, tx);
613 	}
614 	bpobj_close(&dle->dle_bpobj);
615 	kmem_free(dle, sizeof (*dle));
616 	mutex_exit(&dl->dl_lock);
617 }
618 
619 /*
620  * Clear out the contents of a deadlist_entry by freeing its bpobj,
621  * replacing it with an empty bpobj and adjusting the deadlist's
622  * space accounting
623  */
624 void
dsl_deadlist_clear_entry(dsl_deadlist_entry_t * dle,dsl_deadlist_t * dl,dmu_tx_t * tx)625 dsl_deadlist_clear_entry(dsl_deadlist_entry_t *dle, dsl_deadlist_t *dl,
626     dmu_tx_t *tx)
627 {
628 	uint64_t new_obj, used, comp, uncomp;
629 	objset_t *os = dl->dl_os;
630 
631 	mutex_enter(&dl->dl_lock);
632 	VERIFY0(zap_remove_int(os, dl->dl_object, dle->dle_mintxg, tx));
633 	VERIFY0(bpobj_space(&dle->dle_bpobj, &used, &comp, &uncomp));
634 	dmu_buf_will_dirty(dl->dl_dbuf, tx);
635 	dl->dl_phys->dl_used -= used;
636 	dl->dl_phys->dl_comp -= comp;
637 	dl->dl_phys->dl_uncomp -= uncomp;
638 	if (dle->dle_bpobj.bpo_object == dmu_objset_pool(os)->dp_empty_bpobj)
639 		bpobj_decr_empty(os, tx);
640 	else
641 		bpobj_free(os, dle->dle_bpobj.bpo_object, tx);
642 	bpobj_close(&dle->dle_bpobj);
643 	new_obj = bpobj_alloc_empty(os, SPA_OLD_MAXBLOCKSIZE, tx);
644 	VERIFY0(bpobj_open(&dle->dle_bpobj, os, new_obj));
645 	VERIFY0(zap_add_int_key(os, dl->dl_object, dle->dle_mintxg,
646 	    new_obj, tx));
647 	ASSERT(bpobj_is_empty(&dle->dle_bpobj));
648 	mutex_exit(&dl->dl_lock);
649 }
650 
651 /*
652  * Return the first entry in deadlist's avl tree
653  */
654 dsl_deadlist_entry_t *
dsl_deadlist_first(dsl_deadlist_t * dl)655 dsl_deadlist_first(dsl_deadlist_t *dl)
656 {
657 	dsl_deadlist_entry_t *dle;
658 
659 	mutex_enter(&dl->dl_lock);
660 	dsl_deadlist_load_tree(dl);
661 	dle = avl_first(&dl->dl_tree);
662 	mutex_exit(&dl->dl_lock);
663 
664 	return (dle);
665 }
666 
667 /*
668  * Return the last entry in deadlist's avl tree
669  */
670 dsl_deadlist_entry_t *
dsl_deadlist_last(dsl_deadlist_t * dl)671 dsl_deadlist_last(dsl_deadlist_t *dl)
672 {
673 	dsl_deadlist_entry_t *dle;
674 
675 	mutex_enter(&dl->dl_lock);
676 	dsl_deadlist_load_tree(dl);
677 	dle = avl_last(&dl->dl_tree);
678 	mutex_exit(&dl->dl_lock);
679 
680 	return (dle);
681 }
682 
683 /*
684  * Walk ds's snapshots to regenerate generate ZAP & AVL.
685  */
686 static void
dsl_deadlist_regenerate(objset_t * os,uint64_t dlobj,uint64_t mrs_obj,dmu_tx_t * tx)687 dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
688     uint64_t mrs_obj, dmu_tx_t *tx)
689 {
690 	dsl_deadlist_t dl = { 0 };
691 	dsl_pool_t *dp = dmu_objset_pool(os);
692 
693 	VERIFY0(dsl_deadlist_open(&dl, os, dlobj));
694 	if (dl.dl_oldfmt) {
695 		dsl_deadlist_close(&dl);
696 		return;
697 	}
698 
699 	while (mrs_obj != 0) {
700 		dsl_dataset_t *ds;
701 		VERIFY0(dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
702 		dsl_deadlist_add_key(&dl,
703 		    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
704 		mrs_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
705 		dsl_dataset_rele(ds, FTAG);
706 	}
707 	dsl_deadlist_close(&dl);
708 }
709 
710 uint64_t
dsl_deadlist_clone(dsl_deadlist_t * dl,uint64_t maxtxg,uint64_t mrs_obj,dmu_tx_t * tx)711 dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
712     uint64_t mrs_obj, dmu_tx_t *tx)
713 {
714 	dsl_deadlist_entry_t *dle;
715 	uint64_t newobj;
716 
717 	newobj = dsl_deadlist_alloc(dl->dl_os, tx);
718 
719 	if (dl->dl_oldfmt) {
720 		dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
721 		return (newobj);
722 	}
723 
724 	mutex_enter(&dl->dl_lock);
725 	dsl_deadlist_load_tree(dl);
726 
727 	for (dle = avl_first(&dl->dl_tree); dle;
728 	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
729 		uint64_t obj;
730 
731 		if (dle->dle_mintxg >= maxtxg)
732 			break;
733 
734 		obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
735 		VERIFY0(zap_add_int_key(dl->dl_os, newobj,
736 		    dle->dle_mintxg, obj, tx));
737 	}
738 	mutex_exit(&dl->dl_lock);
739 	return (newobj);
740 }
741 
742 void
dsl_deadlist_space(dsl_deadlist_t * dl,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)743 dsl_deadlist_space(dsl_deadlist_t *dl,
744     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
745 {
746 	ASSERT(dsl_deadlist_is_open(dl));
747 	if (dl->dl_oldfmt) {
748 		VERIFY0(bpobj_space(&dl->dl_bpobj,
749 		    usedp, compp, uncompp));
750 		return;
751 	}
752 
753 	mutex_enter(&dl->dl_lock);
754 	*usedp = dl->dl_phys->dl_used;
755 	*compp = dl->dl_phys->dl_comp;
756 	*uncompp = dl->dl_phys->dl_uncomp;
757 	mutex_exit(&dl->dl_lock);
758 }
759 
760 /*
761  * return space used in the range (mintxg, maxtxg].
762  * Includes maxtxg, does not include mintxg.
763  * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
764  * UINT64_MAX).
765  */
766 void
dsl_deadlist_space_range(dsl_deadlist_t * dl,uint64_t mintxg,uint64_t maxtxg,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)767 dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
768     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
769 {
770 	dsl_deadlist_cache_entry_t *dlce;
771 	dsl_deadlist_cache_entry_t dlce_tofind;
772 	avl_index_t where;
773 
774 	if (dl->dl_oldfmt) {
775 		VERIFY0(bpobj_space_range(&dl->dl_bpobj,
776 		    mintxg, maxtxg, usedp, compp, uncompp));
777 		return;
778 	}
779 
780 	*usedp = *compp = *uncompp = 0;
781 
782 	mutex_enter(&dl->dl_lock);
783 	dsl_deadlist_load_cache(dl);
784 	dlce_tofind.dlce_mintxg = mintxg;
785 	dlce = avl_find(&dl->dl_cache, &dlce_tofind, &where);
786 
787 	/*
788 	 * If this mintxg doesn't exist, it may be an empty_bpobj which
789 	 * is omitted from the sparse tree.  Start at the next non-empty
790 	 * entry.
791 	 */
792 	if (dlce == NULL)
793 		dlce = avl_nearest(&dl->dl_cache, where, AVL_AFTER);
794 
795 	for (; dlce && dlce->dlce_mintxg < maxtxg;
796 	    dlce = AVL_NEXT(&dl->dl_tree, dlce)) {
797 		*usedp += dlce->dlce_bytes;
798 		*compp += dlce->dlce_comp;
799 		*uncompp += dlce->dlce_uncomp;
800 	}
801 
802 	mutex_exit(&dl->dl_lock);
803 }
804 
805 static void
dsl_deadlist_insert_bpobj(dsl_deadlist_t * dl,uint64_t obj,uint64_t birth,dmu_tx_t * tx)806 dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
807     dmu_tx_t *tx)
808 {
809 	dsl_deadlist_entry_t dle_tofind;
810 	dsl_deadlist_entry_t *dle;
811 	avl_index_t where;
812 	uint64_t used, comp, uncomp;
813 	bpobj_t bpo;
814 
815 	ASSERT(MUTEX_HELD(&dl->dl_lock));
816 
817 	VERIFY0(bpobj_open(&bpo, dl->dl_os, obj));
818 	VERIFY0(bpobj_space(&bpo, &used, &comp, &uncomp));
819 	bpobj_close(&bpo);
820 
821 	dsl_deadlist_load_tree(dl);
822 
823 	dmu_buf_will_dirty(dl->dl_dbuf, tx);
824 	dl->dl_phys->dl_used += used;
825 	dl->dl_phys->dl_comp += comp;
826 	dl->dl_phys->dl_uncomp += uncomp;
827 
828 	dle_tofind.dle_mintxg = birth;
829 	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
830 	if (dle == NULL)
831 		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
832 	dle_enqueue_subobj(dl, dle, obj, tx);
833 }
834 
835 /*
836  * Prefetch metadata required for dsl_deadlist_insert_bpobj().
837  */
838 static void
dsl_deadlist_prefetch_bpobj(dsl_deadlist_t * dl,uint64_t obj,uint64_t birth)839 dsl_deadlist_prefetch_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth)
840 {
841 	dsl_deadlist_entry_t dle_tofind;
842 	dsl_deadlist_entry_t *dle;
843 	avl_index_t where;
844 
845 	ASSERT(MUTEX_HELD(&dl->dl_lock));
846 
847 	dsl_deadlist_load_tree(dl);
848 
849 	dle_tofind.dle_mintxg = birth;
850 	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
851 	if (dle == NULL)
852 		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
853 	dle_prefetch_subobj(dl, dle, obj);
854 }
855 
856 static int
dsl_deadlist_insert_cb(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)857 dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, boolean_t bp_freed,
858     dmu_tx_t *tx)
859 {
860 	dsl_deadlist_t *dl = arg;
861 	dsl_deadlist_insert(dl, bp, bp_freed, tx);
862 	return (0);
863 }
864 
865 /*
866  * Merge the deadlist pointed to by 'obj' into dl.  obj will be left as
867  * an empty deadlist.
868  */
869 void
dsl_deadlist_merge(dsl_deadlist_t * dl,uint64_t obj,dmu_tx_t * tx)870 dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
871 {
872 	zap_cursor_t zc, pzc;
873 	zap_attribute_t *za, *pza;
874 	dmu_buf_t *bonus;
875 	dsl_deadlist_phys_t *dlp;
876 	dmu_object_info_t doi;
877 	int error, perror, i;
878 
879 	VERIFY0(dmu_object_info(dl->dl_os, obj, &doi));
880 	if (doi.doi_type == DMU_OT_BPOBJ) {
881 		bpobj_t bpo;
882 		VERIFY0(bpobj_open(&bpo, dl->dl_os, obj));
883 		VERIFY0(bpobj_iterate(&bpo, dsl_deadlist_insert_cb, dl, tx));
884 		bpobj_close(&bpo);
885 		return;
886 	}
887 
888 	za = zap_attribute_alloc();
889 	pza = zap_attribute_alloc();
890 
891 	mutex_enter(&dl->dl_lock);
892 	/*
893 	 * Prefetch up to 128 deadlists first and then more as we progress.
894 	 * The limit is a balance between ARC use and diminishing returns.
895 	 */
896 	for (zap_cursor_init(&pzc, dl->dl_os, obj), i = 0;
897 	    (perror = zap_cursor_retrieve(&pzc, pza)) == 0 && i < 128;
898 	    zap_cursor_advance(&pzc), i++) {
899 		dsl_deadlist_prefetch_bpobj(dl, pza->za_first_integer,
900 		    zfs_strtonum(pza->za_name, NULL));
901 	}
902 	for (zap_cursor_init(&zc, dl->dl_os, obj);
903 	    (error = zap_cursor_retrieve(&zc, za)) == 0;
904 	    zap_cursor_advance(&zc)) {
905 		dsl_deadlist_insert_bpobj(dl, za->za_first_integer,
906 		    zfs_strtonum(za->za_name, NULL), tx);
907 		VERIFY0(zap_remove(dl->dl_os, obj, za->za_name, tx));
908 		if (perror == 0) {
909 			dsl_deadlist_prefetch_bpobj(dl, pza->za_first_integer,
910 			    zfs_strtonum(pza->za_name, NULL));
911 			zap_cursor_advance(&pzc);
912 			perror = zap_cursor_retrieve(&pzc, pza);
913 		}
914 	}
915 	VERIFY3U(error, ==, ENOENT);
916 	zap_cursor_fini(&zc);
917 	zap_cursor_fini(&pzc);
918 
919 	VERIFY0(dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
920 	dlp = bonus->db_data;
921 	dmu_buf_will_dirty(bonus, tx);
922 	memset(dlp, 0, sizeof (*dlp));
923 	dmu_buf_rele(bonus, FTAG);
924 	mutex_exit(&dl->dl_lock);
925 
926 	zap_attribute_free(za);
927 	zap_attribute_free(pza);
928 }
929 
930 /*
931  * Remove entries on dl that are born > mintxg, and put them on the bpobj.
932  */
933 void
dsl_deadlist_move_bpobj(dsl_deadlist_t * dl,bpobj_t * bpo,uint64_t mintxg,dmu_tx_t * tx)934 dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
935     dmu_tx_t *tx)
936 {
937 	dsl_deadlist_entry_t dle_tofind;
938 	dsl_deadlist_entry_t *dle, *pdle;
939 	avl_index_t where;
940 	int i;
941 
942 	ASSERT(!dl->dl_oldfmt);
943 
944 	mutex_enter(&dl->dl_lock);
945 	dmu_buf_will_dirty(dl->dl_dbuf, tx);
946 	dsl_deadlist_load_tree(dl);
947 
948 	dle_tofind.dle_mintxg = mintxg;
949 	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
950 	if (dle == NULL)
951 		dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
952 	/*
953 	 * Prefetch up to 128 deadlists first and then more as we progress.
954 	 * The limit is a balance between ARC use and diminishing returns.
955 	 */
956 	for (pdle = dle, i = 0; pdle && i < 128; i++) {
957 		bpobj_prefetch_subobj(bpo, pdle->dle_bpobj.bpo_object);
958 		pdle = AVL_NEXT(&dl->dl_tree, pdle);
959 	}
960 	while (dle) {
961 		uint64_t used, comp, uncomp;
962 		dsl_deadlist_entry_t *dle_next;
963 
964 		bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
965 		if (pdle) {
966 			bpobj_prefetch_subobj(bpo, pdle->dle_bpobj.bpo_object);
967 			pdle = AVL_NEXT(&dl->dl_tree, pdle);
968 		}
969 
970 		VERIFY0(bpobj_space(&dle->dle_bpobj,
971 		    &used, &comp, &uncomp));
972 		ASSERT3U(dl->dl_phys->dl_used, >=, used);
973 		ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
974 		ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
975 		dl->dl_phys->dl_used -= used;
976 		dl->dl_phys->dl_comp -= comp;
977 		dl->dl_phys->dl_uncomp -= uncomp;
978 
979 		VERIFY0(zap_remove_int(dl->dl_os, dl->dl_object,
980 		    dle->dle_mintxg, tx));
981 
982 		dle_next = AVL_NEXT(&dl->dl_tree, dle);
983 		avl_remove(&dl->dl_tree, dle);
984 		bpobj_close(&dle->dle_bpobj);
985 		kmem_free(dle, sizeof (*dle));
986 		dle = dle_next;
987 	}
988 	mutex_exit(&dl->dl_lock);
989 }
990 
991 typedef struct livelist_entry {
992 	blkptr_t le_bp;
993 	uint32_t le_refcnt;
994 	avl_node_t le_node;
995 } livelist_entry_t;
996 
997 static int
livelist_compare(const void * larg,const void * rarg)998 livelist_compare(const void *larg, const void *rarg)
999 {
1000 	const blkptr_t *l = &((livelist_entry_t *)larg)->le_bp;
1001 	const blkptr_t *r = &((livelist_entry_t *)rarg)->le_bp;
1002 
1003 	/* Sort them according to dva[0] */
1004 	uint64_t l_dva0_vdev = DVA_GET_VDEV(&l->blk_dva[0]);
1005 	uint64_t r_dva0_vdev = DVA_GET_VDEV(&r->blk_dva[0]);
1006 
1007 	if (l_dva0_vdev != r_dva0_vdev)
1008 		return (TREE_CMP(l_dva0_vdev, r_dva0_vdev));
1009 
1010 	/* if vdevs are equal, sort by offsets. */
1011 	uint64_t l_dva0_offset = DVA_GET_OFFSET(&l->blk_dva[0]);
1012 	uint64_t r_dva0_offset = DVA_GET_OFFSET(&r->blk_dva[0]);
1013 	return (TREE_CMP(l_dva0_offset, r_dva0_offset));
1014 }
1015 
1016 struct livelist_iter_arg {
1017 	avl_tree_t *avl;
1018 	bplist_t *to_free;
1019 	zthr_t *t;
1020 };
1021 
1022 /*
1023  * Expects an AVL tree which is incrementally filled will FREE blkptrs
1024  * and used to match up ALLOC/FREE pairs. ALLOC'd blkptrs without a
1025  * corresponding FREE are stored in the supplied bplist.
1026  *
1027  * Note that multiple FREE and ALLOC entries for the same blkptr may be
1028  * encountered when dedup or block cloning is involved.  For this reason we
1029  * keep a refcount for all the FREE entries of each blkptr and ensure that
1030  * each of those FREE entries has a corresponding ALLOC preceding it.
1031  */
1032 static int
dsl_livelist_iterate(void * arg,const blkptr_t * bp,boolean_t bp_freed,dmu_tx_t * tx)1033 dsl_livelist_iterate(void *arg, const blkptr_t *bp, boolean_t bp_freed,
1034     dmu_tx_t *tx)
1035 {
1036 	struct livelist_iter_arg *lia = arg;
1037 	avl_tree_t *avl = lia->avl;
1038 	bplist_t *to_free = lia->to_free;
1039 	zthr_t *t = lia->t;
1040 	ASSERT(tx == NULL);
1041 
1042 	if ((t != NULL) && (zthr_has_waiters(t) || zthr_iscancelled(t)))
1043 		return (SET_ERROR(EINTR));
1044 
1045 	livelist_entry_t node;
1046 	node.le_bp = *bp;
1047 	livelist_entry_t *found = avl_find(avl, &node, NULL);
1048 	if (found) {
1049 		ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(&found->le_bp));
1050 		ASSERT3U(BP_GET_CHECKSUM(bp), ==,
1051 		    BP_GET_CHECKSUM(&found->le_bp));
1052 		ASSERT3U(BP_GET_BIRTH(bp), ==, BP_GET_BIRTH(&found->le_bp));
1053 	}
1054 	if (bp_freed) {
1055 		if (found == NULL) {
1056 			/* first free entry for this blkptr */
1057 			livelist_entry_t *e =
1058 			    kmem_alloc(sizeof (livelist_entry_t), KM_SLEEP);
1059 			e->le_bp = *bp;
1060 			e->le_refcnt = 1;
1061 			avl_add(avl, e);
1062 		} else {
1063 			/*
1064 			 * Deduped or cloned block free.  We could assert D bit
1065 			 * for dedup, but there is no such one for cloning.
1066 			 */
1067 			ASSERT3U(found->le_refcnt + 1, >, found->le_refcnt);
1068 			found->le_refcnt++;
1069 		}
1070 	} else {
1071 		if (found == NULL) {
1072 			/* block is currently marked as allocated */
1073 			bplist_append(to_free, bp);
1074 		} else {
1075 			/* alloc matches a free entry */
1076 			ASSERT3U(found->le_refcnt, !=, 0);
1077 			found->le_refcnt--;
1078 			if (found->le_refcnt == 0) {
1079 				/* all tracked free pairs have been matched */
1080 				avl_remove(avl, found);
1081 				kmem_free(found, sizeof (livelist_entry_t));
1082 			}
1083 		}
1084 	}
1085 	return (0);
1086 }
1087 
1088 /*
1089  * Accepts a bpobj and a bplist. Will insert into the bplist the blkptrs
1090  * which have an ALLOC entry but no matching FREE
1091  */
1092 int
dsl_process_sub_livelist(bpobj_t * bpobj,bplist_t * to_free,zthr_t * t,uint64_t * size)1093 dsl_process_sub_livelist(bpobj_t *bpobj, bplist_t *to_free, zthr_t *t,
1094     uint64_t *size)
1095 {
1096 	avl_tree_t avl;
1097 	avl_create(&avl, livelist_compare, sizeof (livelist_entry_t),
1098 	    offsetof(livelist_entry_t, le_node));
1099 
1100 	/* process the sublist */
1101 	struct livelist_iter_arg arg = {
1102 	    .avl = &avl,
1103 	    .to_free = to_free,
1104 	    .t = t
1105 	};
1106 	int err = bpobj_iterate_nofree(bpobj, dsl_livelist_iterate, &arg, size);
1107 	VERIFY(err != 0 || avl_numnodes(&avl) == 0);
1108 
1109 	void *cookie = NULL;
1110 	livelist_entry_t *le = NULL;
1111 	while ((le = avl_destroy_nodes(&avl, &cookie)) != NULL) {
1112 		kmem_free(le, sizeof (livelist_entry_t));
1113 	}
1114 	avl_destroy(&avl);
1115 	return (err);
1116 }
1117 
1118 ZFS_MODULE_PARAM(zfs_livelist, zfs_livelist_, max_entries, U64, ZMOD_RW,
1119 	"Size to start the next sub-livelist in a livelist");
1120 
1121 ZFS_MODULE_PARAM(zfs_livelist, zfs_livelist_, min_percent_shared, INT, ZMOD_RW,
1122 	"Threshold at which livelist is disabled");
1123