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.
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.
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]
22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
26 #include <sys/dsl_dataset.h>
28 #include <sys/refcount.h>
30 #include <sys/zfs_context.h>
31 #include <sys/dsl_pool.h>
34 * Deadlist concurrency:
36 * Deadlists can only be modified from the syncing thread.
38 * Except for dsl_deadlist_insert(), it can only be modified with the
39 * dp_config_rwlock held with RW_WRITER.
41 * The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
42 * be called concurrently, from open context, with the dl_config_rwlock held
45 * Therefore, we only need to provide locking between dsl_deadlist_insert() and
46 * the accessors, protecting:
47 * dl_phys->dl_used,comp,uncomp
48 * and protecting the dl_tree from being loaded.
49 * The locking is provided by dl_lock. Note that locking on the bpobj_t
50 * provides its own locking, and dl_oldfmt is immutable.
54 dsl_deadlist_compare(const void *arg1, const void *arg2)
56 const dsl_deadlist_entry_t *dle1 = arg1;
57 const dsl_deadlist_entry_t *dle2 = arg2;
59 if (dle1->dle_mintxg < dle2->dle_mintxg)
61 else if (dle1->dle_mintxg > dle2->dle_mintxg)
68 dsl_deadlist_load_tree(dsl_deadlist_t *dl)
73 ASSERT(!dl->dl_oldfmt);
77 avl_create(&dl->dl_tree, dsl_deadlist_compare,
78 sizeof (dsl_deadlist_entry_t),
79 offsetof(dsl_deadlist_entry_t, dle_node));
80 for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
81 zap_cursor_retrieve(&zc, &za) == 0;
82 zap_cursor_advance(&zc)) {
83 dsl_deadlist_entry_t *dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
84 dle->dle_mintxg = strtonum(za.za_name, NULL);
85 VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os,
86 za.za_first_integer));
87 avl_add(&dl->dl_tree, dle);
90 dl->dl_havetree = B_TRUE;
94 dsl_deadlist_open(dsl_deadlist_t *dl, objset_t *os, uint64_t object)
96 dmu_object_info_t doi;
98 mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
100 dl->dl_object = object;
101 VERIFY3U(0, ==, dmu_bonus_hold(os, object, dl, &dl->dl_dbuf));
102 dmu_object_info_from_db(dl->dl_dbuf, &doi);
103 if (doi.doi_type == DMU_OT_BPOBJ) {
104 dmu_buf_rele(dl->dl_dbuf, dl);
106 dl->dl_oldfmt = B_TRUE;
107 VERIFY3U(0, ==, bpobj_open(&dl->dl_bpobj, os, object));
111 dl->dl_oldfmt = B_FALSE;
112 dl->dl_phys = dl->dl_dbuf->db_data;
113 dl->dl_havetree = B_FALSE;
117 dsl_deadlist_close(dsl_deadlist_t *dl)
120 dsl_deadlist_entry_t *dle;
123 dl->dl_oldfmt = B_FALSE;
124 bpobj_close(&dl->dl_bpobj);
128 if (dl->dl_havetree) {
129 while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
131 bpobj_close(&dle->dle_bpobj);
132 kmem_free(dle, sizeof (*dle));
134 avl_destroy(&dl->dl_tree);
136 dmu_buf_rele(dl->dl_dbuf, dl);
137 mutex_destroy(&dl->dl_lock);
143 dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx)
145 if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
146 return (bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx));
147 return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
148 sizeof (dsl_deadlist_phys_t), tx));
152 dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx)
154 dmu_object_info_t doi;
158 VERIFY3U(0, ==, dmu_object_info(os, dlobj, &doi));
159 if (doi.doi_type == DMU_OT_BPOBJ) {
160 bpobj_free(os, dlobj, tx);
164 for (zap_cursor_init(&zc, os, dlobj);
165 zap_cursor_retrieve(&zc, &za) == 0;
166 zap_cursor_advance(&zc)) {
167 uint64_t obj = za.za_first_integer;
168 if (obj == dmu_objset_pool(os)->dp_empty_bpobj)
169 bpobj_decr_empty(os, tx);
171 bpobj_free(os, obj, tx);
173 zap_cursor_fini(&zc);
174 VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx));
178 dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
179 const blkptr_t *bp, dmu_tx_t *tx)
181 if (dle->dle_bpobj.bpo_object ==
182 dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
183 uint64_t obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
184 bpobj_close(&dle->dle_bpobj);
185 bpobj_decr_empty(dl->dl_os, tx);
186 VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
187 VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
188 dle->dle_mintxg, obj, tx));
190 bpobj_enqueue(&dle->dle_bpobj, bp, tx);
194 dle_enqueue_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
195 uint64_t obj, dmu_tx_t *tx)
197 if (dle->dle_bpobj.bpo_object !=
198 dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
199 bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
201 bpobj_close(&dle->dle_bpobj);
202 bpobj_decr_empty(dl->dl_os, tx);
203 VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
204 VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
205 dle->dle_mintxg, obj, tx));
210 dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx)
212 dsl_deadlist_entry_t dle_tofind;
213 dsl_deadlist_entry_t *dle;
217 bpobj_enqueue(&dl->dl_bpobj, bp, tx);
221 dsl_deadlist_load_tree(dl);
223 dmu_buf_will_dirty(dl->dl_dbuf, tx);
224 mutex_enter(&dl->dl_lock);
225 dl->dl_phys->dl_used +=
226 bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
227 dl->dl_phys->dl_comp += BP_GET_PSIZE(bp);
228 dl->dl_phys->dl_uncomp += BP_GET_UCSIZE(bp);
229 mutex_exit(&dl->dl_lock);
231 dle_tofind.dle_mintxg = bp->blk_birth;
232 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
234 dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
236 dle = AVL_PREV(&dl->dl_tree, dle);
237 dle_enqueue(dl, dle, bp, tx);
241 * Insert new key in deadlist, which must be > all current entries.
242 * mintxg is not inclusive.
245 dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
248 dsl_deadlist_entry_t *dle;
253 dsl_deadlist_load_tree(dl);
255 dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
256 dle->dle_mintxg = mintxg;
257 obj = bpobj_alloc_empty(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
258 VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
259 avl_add(&dl->dl_tree, dle);
261 VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, dl->dl_object,
266 * Remove this key, merging its entries into the previous key.
269 dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
271 dsl_deadlist_entry_t dle_tofind;
272 dsl_deadlist_entry_t *dle, *dle_prev;
277 dsl_deadlist_load_tree(dl);
279 dle_tofind.dle_mintxg = mintxg;
280 dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
281 dle_prev = AVL_PREV(&dl->dl_tree, dle);
283 dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx);
285 avl_remove(&dl->dl_tree, dle);
286 bpobj_close(&dle->dle_bpobj);
287 kmem_free(dle, sizeof (*dle));
289 VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
293 * Walk ds's snapshots to regenerate generate ZAP & AVL.
296 dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
297 uint64_t mrs_obj, dmu_tx_t *tx)
300 dsl_pool_t *dp = dmu_objset_pool(os);
302 dsl_deadlist_open(&dl, os, dlobj);
304 dsl_deadlist_close(&dl);
308 while (mrs_obj != 0) {
310 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
311 dsl_deadlist_add_key(&dl, ds->ds_phys->ds_prev_snap_txg, tx);
312 mrs_obj = ds->ds_phys->ds_prev_snap_obj;
313 dsl_dataset_rele(ds, FTAG);
315 dsl_deadlist_close(&dl);
319 dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
320 uint64_t mrs_obj, dmu_tx_t *tx)
322 dsl_deadlist_entry_t *dle;
325 newobj = dsl_deadlist_alloc(dl->dl_os, tx);
328 dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
332 dsl_deadlist_load_tree(dl);
334 for (dle = avl_first(&dl->dl_tree); dle;
335 dle = AVL_NEXT(&dl->dl_tree, dle)) {
338 if (dle->dle_mintxg >= maxtxg)
341 obj = bpobj_alloc_empty(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
342 VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj,
343 dle->dle_mintxg, obj, tx));
349 dsl_deadlist_space(dsl_deadlist_t *dl,
350 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
353 VERIFY3U(0, ==, bpobj_space(&dl->dl_bpobj,
354 usedp, compp, uncompp));
358 mutex_enter(&dl->dl_lock);
359 *usedp = dl->dl_phys->dl_used;
360 *compp = dl->dl_phys->dl_comp;
361 *uncompp = dl->dl_phys->dl_uncomp;
362 mutex_exit(&dl->dl_lock);
366 * return space used in the range (mintxg, maxtxg].
367 * Includes maxtxg, does not include mintxg.
368 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
369 * larger than any bp in the deadlist (eg. UINT64_MAX)).
372 dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
373 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
375 dsl_deadlist_entry_t *dle;
376 dsl_deadlist_entry_t dle_tofind;
380 VERIFY3U(0, ==, bpobj_space_range(&dl->dl_bpobj,
381 mintxg, maxtxg, usedp, compp, uncompp));
385 *usedp = *compp = *uncompp = 0;
387 mutex_enter(&dl->dl_lock);
388 dsl_deadlist_load_tree(dl);
389 dle_tofind.dle_mintxg = mintxg;
390 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
392 * If we don't find this mintxg, there shouldn't be anything
395 ASSERT(dle != NULL ||
396 avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);
398 for (; dle && dle->dle_mintxg < maxtxg;
399 dle = AVL_NEXT(&dl->dl_tree, dle)) {
400 uint64_t used, comp, uncomp;
402 VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
403 &used, &comp, &uncomp));
409 mutex_exit(&dl->dl_lock);
413 dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
416 dsl_deadlist_entry_t dle_tofind;
417 dsl_deadlist_entry_t *dle;
419 uint64_t used, comp, uncomp;
422 VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
423 VERIFY3U(0, ==, bpobj_space(&bpo, &used, &comp, &uncomp));
426 dsl_deadlist_load_tree(dl);
428 dmu_buf_will_dirty(dl->dl_dbuf, tx);
429 mutex_enter(&dl->dl_lock);
430 dl->dl_phys->dl_used += used;
431 dl->dl_phys->dl_comp += comp;
432 dl->dl_phys->dl_uncomp += uncomp;
433 mutex_exit(&dl->dl_lock);
435 dle_tofind.dle_mintxg = birth;
436 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
438 dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
439 dle_enqueue_subobj(dl, dle, obj, tx);
443 dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
445 dsl_deadlist_t *dl = arg;
446 dsl_deadlist_insert(dl, bp, tx);
451 * Merge the deadlist pointed to by 'obj' into dl. obj will be left as
455 dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
460 dsl_deadlist_phys_t *dlp;
461 dmu_object_info_t doi;
463 VERIFY3U(0, ==, dmu_object_info(dl->dl_os, obj, &doi));
464 if (doi.doi_type == DMU_OT_BPOBJ) {
466 VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
467 VERIFY3U(0, ==, bpobj_iterate(&bpo,
468 dsl_deadlist_insert_cb, dl, tx));
473 for (zap_cursor_init(&zc, dl->dl_os, obj);
474 zap_cursor_retrieve(&zc, &za) == 0;
475 zap_cursor_advance(&zc)) {
476 uint64_t mintxg = strtonum(za.za_name, NULL);
477 dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
478 VERIFY3U(0, ==, zap_remove_int(dl->dl_os, obj, mintxg, tx));
480 zap_cursor_fini(&zc);
482 VERIFY3U(0, ==, dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
483 dlp = bonus->db_data;
484 dmu_buf_will_dirty(bonus, tx);
485 bzero(dlp, sizeof (*dlp));
486 dmu_buf_rele(bonus, FTAG);
490 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
493 dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
496 dsl_deadlist_entry_t dle_tofind;
497 dsl_deadlist_entry_t *dle;
500 ASSERT(!dl->dl_oldfmt);
501 dmu_buf_will_dirty(dl->dl_dbuf, tx);
502 dsl_deadlist_load_tree(dl);
504 dle_tofind.dle_mintxg = mintxg;
505 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
507 dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
509 uint64_t used, comp, uncomp;
510 dsl_deadlist_entry_t *dle_next;
512 bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
514 VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
515 &used, &comp, &uncomp));
516 mutex_enter(&dl->dl_lock);
517 ASSERT3U(dl->dl_phys->dl_used, >=, used);
518 ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
519 ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
520 dl->dl_phys->dl_used -= used;
521 dl->dl_phys->dl_comp -= comp;
522 dl->dl_phys->dl_uncomp -= uncomp;
523 mutex_exit(&dl->dl_lock);
525 VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object,
526 dle->dle_mintxg, tx));
528 dle_next = AVL_NEXT(&dl->dl_tree, dle);
529 avl_remove(&dl->dl_tree, dle);
530 bpobj_close(&dle->dle_bpobj);
531 kmem_free(dle, sizeof (*dle));