1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28#include <sys/dsl_dataset.h>
29#include <sys/dmu.h>
30#include <sys/refcount.h>
31#include <sys/zap.h>
32#include <sys/zfs_context.h>
33#include <sys/dsl_pool.h>
34
35/*
36 * Deadlist concurrency:
37 *
38 * Deadlists can only be modified from the syncing thread.
39 *
40 * Except for dsl_deadlist_insert(), it can only be modified with the
41 * dp_config_rwlock held with RW_WRITER.
42 *
43 * The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
44 * be called concurrently, from open context, with the dl_config_rwlock held
45 * with RW_READER.
46 *
47 * Therefore, we only need to provide locking between dsl_deadlist_insert() and
48 * the accessors, protecting:
49 *     dl_phys->dl_used,comp,uncomp
50 *     and protecting the dl_tree from being loaded.
51 * The locking is provided by dl_lock.  Note that locking on the bpobj_t
52 * provides its own locking, and dl_oldfmt is immutable.
53 */
54
55static int
56dsl_deadlist_compare(const void *arg1, const void *arg2)
57{
58	const dsl_deadlist_entry_t *dle1 = (const dsl_deadlist_entry_t *)arg1;
59	const dsl_deadlist_entry_t *dle2 = (const dsl_deadlist_entry_t *)arg2;
60
61	return (AVL_CMP(dle1->dle_mintxg, dle2->dle_mintxg));
62}
63
64static void
65dsl_deadlist_load_tree(dsl_deadlist_t *dl)
66{
67	zap_cursor_t zc;
68	zap_attribute_t za;
69
70	ASSERT(MUTEX_HELD(&dl->dl_lock));
71
72	ASSERT(!dl->dl_oldfmt);
73	if (dl->dl_havetree)
74		return;
75
76	avl_create(&dl->dl_tree, dsl_deadlist_compare,
77	    sizeof (dsl_deadlist_entry_t),
78	    offsetof(dsl_deadlist_entry_t, dle_node));
79	for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
80	    zap_cursor_retrieve(&zc, &za) == 0;
81	    zap_cursor_advance(&zc)) {
82		dsl_deadlist_entry_t *dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
83		dle->dle_mintxg = zfs_strtonum(za.za_name, NULL);
84		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os,
85		    za.za_first_integer));
86		avl_add(&dl->dl_tree, dle);
87	}
88	zap_cursor_fini(&zc);
89	dl->dl_havetree = B_TRUE;
90}
91
92void
93dsl_deadlist_open(dsl_deadlist_t *dl, objset_t *os, uint64_t object)
94{
95	dmu_object_info_t doi;
96
97	ASSERT(!dsl_deadlist_is_open(dl));
98
99	mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
100	dl->dl_os = os;
101	dl->dl_object = object;
102	VERIFY3U(0, ==, dmu_bonus_hold(os, object, dl, &dl->dl_dbuf));
103	dmu_object_info_from_db(dl->dl_dbuf, &doi);
104	if (doi.doi_type == DMU_OT_BPOBJ) {
105		dmu_buf_rele(dl->dl_dbuf, dl);
106		dl->dl_dbuf = NULL;
107		dl->dl_oldfmt = B_TRUE;
108		VERIFY3U(0, ==, bpobj_open(&dl->dl_bpobj, os, object));
109		return;
110	}
111
112	dl->dl_oldfmt = B_FALSE;
113	dl->dl_phys = dl->dl_dbuf->db_data;
114	dl->dl_havetree = B_FALSE;
115}
116
117boolean_t
118dsl_deadlist_is_open(dsl_deadlist_t *dl)
119{
120	return (dl->dl_os != NULL);
121}
122
123void
124dsl_deadlist_close(dsl_deadlist_t *dl)
125{
126	void *cookie = NULL;
127	dsl_deadlist_entry_t *dle;
128
129	ASSERT(dsl_deadlist_is_open(dl));
130
131	if (dl->dl_oldfmt) {
132		dl->dl_oldfmt = B_FALSE;
133		bpobj_close(&dl->dl_bpobj);
134		dl->dl_os = NULL;
135		dl->dl_object = 0;
136		return;
137	}
138
139	if (dl->dl_havetree) {
140		while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
141		    != NULL) {
142			bpobj_close(&dle->dle_bpobj);
143			kmem_free(dle, sizeof (*dle));
144		}
145		avl_destroy(&dl->dl_tree);
146	}
147	dmu_buf_rele(dl->dl_dbuf, dl);
148	mutex_destroy(&dl->dl_lock);
149	dl->dl_dbuf = NULL;
150	dl->dl_phys = NULL;
151	dl->dl_os = NULL;
152	dl->dl_object = 0;
153}
154
155uint64_t
156dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx)
157{
158	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
159		return (bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx));
160	return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
161	    sizeof (dsl_deadlist_phys_t), tx));
162}
163
164void
165dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx)
166{
167	dmu_object_info_t doi;
168	zap_cursor_t zc;
169	zap_attribute_t za;
170
171	VERIFY3U(0, ==, dmu_object_info(os, dlobj, &doi));
172	if (doi.doi_type == DMU_OT_BPOBJ) {
173		bpobj_free(os, dlobj, tx);
174		return;
175	}
176
177	for (zap_cursor_init(&zc, os, dlobj);
178	    zap_cursor_retrieve(&zc, &za) == 0;
179	    zap_cursor_advance(&zc)) {
180		uint64_t obj = za.za_first_integer;
181		if (obj == dmu_objset_pool(os)->dp_empty_bpobj)
182			bpobj_decr_empty(os, tx);
183		else
184			bpobj_free(os, obj, tx);
185	}
186	zap_cursor_fini(&zc);
187	VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx));
188}
189
190static void
191dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
192    const blkptr_t *bp, dmu_tx_t *tx)
193{
194	ASSERT(MUTEX_HELD(&dl->dl_lock));
195	if (dle->dle_bpobj.bpo_object ==
196	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
197		uint64_t obj = bpobj_alloc(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
198		bpobj_close(&dle->dle_bpobj);
199		bpobj_decr_empty(dl->dl_os, tx);
200		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
201		VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
202		    dle->dle_mintxg, obj, tx));
203	}
204	bpobj_enqueue(&dle->dle_bpobj, bp, tx);
205}
206
207static void
208dle_enqueue_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
209    uint64_t obj, dmu_tx_t *tx)
210{
211	ASSERT(MUTEX_HELD(&dl->dl_lock));
212	if (dle->dle_bpobj.bpo_object !=
213	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
214		bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
215	} else {
216		bpobj_close(&dle->dle_bpobj);
217		bpobj_decr_empty(dl->dl_os, tx);
218		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
219		VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
220		    dle->dle_mintxg, obj, tx));
221	}
222}
223
224void
225dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx)
226{
227	dsl_deadlist_entry_t dle_tofind;
228	dsl_deadlist_entry_t *dle;
229	avl_index_t where;
230
231	if (dl->dl_oldfmt) {
232		bpobj_enqueue(&dl->dl_bpobj, bp, tx);
233		return;
234	}
235
236	mutex_enter(&dl->dl_lock);
237	dsl_deadlist_load_tree(dl);
238
239	dmu_buf_will_dirty(dl->dl_dbuf, tx);
240	dl->dl_phys->dl_used +=
241	    bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
242	dl->dl_phys->dl_comp += BP_GET_PSIZE(bp);
243	dl->dl_phys->dl_uncomp += BP_GET_UCSIZE(bp);
244
245	dle_tofind.dle_mintxg = bp->blk_birth;
246	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
247	if (dle == NULL)
248		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
249	else
250		dle = AVL_PREV(&dl->dl_tree, dle);
251	dle_enqueue(dl, dle, bp, tx);
252	mutex_exit(&dl->dl_lock);
253}
254
255/*
256 * Insert new key in deadlist, which must be > all current entries.
257 * mintxg is not inclusive.
258 */
259void
260dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
261{
262	uint64_t obj;
263	dsl_deadlist_entry_t *dle;
264
265	if (dl->dl_oldfmt)
266		return;
267
268	dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
269	dle->dle_mintxg = mintxg;
270
271	mutex_enter(&dl->dl_lock);
272	dsl_deadlist_load_tree(dl);
273
274	obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
275	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
276	avl_add(&dl->dl_tree, dle);
277
278	VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, dl->dl_object,
279	    mintxg, obj, tx));
280	mutex_exit(&dl->dl_lock);
281}
282
283/*
284 * Remove this key, merging its entries into the previous key.
285 */
286void
287dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
288{
289	dsl_deadlist_entry_t dle_tofind;
290	dsl_deadlist_entry_t *dle, *dle_prev;
291
292	if (dl->dl_oldfmt)
293		return;
294
295	mutex_enter(&dl->dl_lock);
296	dsl_deadlist_load_tree(dl);
297
298	dle_tofind.dle_mintxg = mintxg;
299	dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
300	dle_prev = AVL_PREV(&dl->dl_tree, dle);
301
302	dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx);
303
304	avl_remove(&dl->dl_tree, dle);
305	bpobj_close(&dle->dle_bpobj);
306	kmem_free(dle, sizeof (*dle));
307
308	VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
309	mutex_exit(&dl->dl_lock);
310}
311
312/*
313 * Walk ds's snapshots to regenerate generate ZAP & AVL.
314 */
315static void
316dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
317    uint64_t mrs_obj, dmu_tx_t *tx)
318{
319	dsl_deadlist_t dl = { 0 };
320	dsl_pool_t *dp = dmu_objset_pool(os);
321
322	dsl_deadlist_open(&dl, os, dlobj);
323	if (dl.dl_oldfmt) {
324		dsl_deadlist_close(&dl);
325		return;
326	}
327
328	while (mrs_obj != 0) {
329		dsl_dataset_t *ds;
330		VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
331		dsl_deadlist_add_key(&dl,
332		    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
333		mrs_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
334		dsl_dataset_rele(ds, FTAG);
335	}
336	dsl_deadlist_close(&dl);
337}
338
339uint64_t
340dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
341    uint64_t mrs_obj, dmu_tx_t *tx)
342{
343	dsl_deadlist_entry_t *dle;
344	uint64_t newobj;
345
346	newobj = dsl_deadlist_alloc(dl->dl_os, tx);
347
348	if (dl->dl_oldfmt) {
349		dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
350		return (newobj);
351	}
352
353	mutex_enter(&dl->dl_lock);
354	dsl_deadlist_load_tree(dl);
355
356	for (dle = avl_first(&dl->dl_tree); dle;
357	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
358		uint64_t obj;
359
360		if (dle->dle_mintxg >= maxtxg)
361			break;
362
363		obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
364		VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj,
365		    dle->dle_mintxg, obj, tx));
366	}
367	mutex_exit(&dl->dl_lock);
368	return (newobj);
369}
370
371void
372dsl_deadlist_space(dsl_deadlist_t *dl,
373    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
374{
375	ASSERT(dsl_deadlist_is_open(dl));
376	if (dl->dl_oldfmt) {
377		VERIFY3U(0, ==, bpobj_space(&dl->dl_bpobj,
378		    usedp, compp, uncompp));
379		return;
380	}
381
382	mutex_enter(&dl->dl_lock);
383	*usedp = dl->dl_phys->dl_used;
384	*compp = dl->dl_phys->dl_comp;
385	*uncompp = dl->dl_phys->dl_uncomp;
386	mutex_exit(&dl->dl_lock);
387}
388
389/*
390 * return space used in the range (mintxg, maxtxg].
391 * Includes maxtxg, does not include mintxg.
392 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
393 * larger than any bp in the deadlist (eg. UINT64_MAX)).
394 */
395void
396dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
397    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
398{
399	dsl_deadlist_entry_t *dle;
400	dsl_deadlist_entry_t dle_tofind;
401	avl_index_t where;
402
403	if (dl->dl_oldfmt) {
404		VERIFY3U(0, ==, bpobj_space_range(&dl->dl_bpobj,
405		    mintxg, maxtxg, usedp, compp, uncompp));
406		return;
407	}
408
409	*usedp = *compp = *uncompp = 0;
410
411	mutex_enter(&dl->dl_lock);
412	dsl_deadlist_load_tree(dl);
413	dle_tofind.dle_mintxg = mintxg;
414	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
415	/*
416	 * If we don't find this mintxg, there shouldn't be anything
417	 * after it either.
418	 */
419	ASSERT(dle != NULL ||
420	    avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);
421
422	for (; dle && dle->dle_mintxg < maxtxg;
423	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
424		uint64_t used, comp, uncomp;
425
426		VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
427		    &used, &comp, &uncomp));
428
429		*usedp += used;
430		*compp += comp;
431		*uncompp += uncomp;
432	}
433	mutex_exit(&dl->dl_lock);
434}
435
436static void
437dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
438    dmu_tx_t *tx)
439{
440	dsl_deadlist_entry_t dle_tofind;
441	dsl_deadlist_entry_t *dle;
442	avl_index_t where;
443	uint64_t used, comp, uncomp;
444	bpobj_t bpo;
445
446	ASSERT(MUTEX_HELD(&dl->dl_lock));
447
448	VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
449	VERIFY3U(0, ==, bpobj_space(&bpo, &used, &comp, &uncomp));
450	bpobj_close(&bpo);
451
452	dsl_deadlist_load_tree(dl);
453
454	dmu_buf_will_dirty(dl->dl_dbuf, tx);
455	dl->dl_phys->dl_used += used;
456	dl->dl_phys->dl_comp += comp;
457	dl->dl_phys->dl_uncomp += uncomp;
458
459	dle_tofind.dle_mintxg = birth;
460	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
461	if (dle == NULL)
462		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
463	dle_enqueue_subobj(dl, dle, obj, tx);
464}
465
466static int
467dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
468{
469	dsl_deadlist_t *dl = arg;
470	dsl_deadlist_insert(dl, bp, tx);
471	return (0);
472}
473
474/*
475 * Merge the deadlist pointed to by 'obj' into dl.  obj will be left as
476 * an empty deadlist.
477 */
478void
479dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
480{
481	zap_cursor_t zc;
482	zap_attribute_t za;
483	dmu_buf_t *bonus;
484	dsl_deadlist_phys_t *dlp;
485	dmu_object_info_t doi;
486
487	VERIFY3U(0, ==, dmu_object_info(dl->dl_os, obj, &doi));
488	if (doi.doi_type == DMU_OT_BPOBJ) {
489		bpobj_t bpo;
490		VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
491		VERIFY3U(0, ==, bpobj_iterate(&bpo,
492		    dsl_deadlist_insert_cb, dl, tx));
493		bpobj_close(&bpo);
494		return;
495	}
496
497	mutex_enter(&dl->dl_lock);
498	for (zap_cursor_init(&zc, dl->dl_os, obj);
499	    zap_cursor_retrieve(&zc, &za) == 0;
500	    zap_cursor_advance(&zc)) {
501		uint64_t mintxg = zfs_strtonum(za.za_name, NULL);
502		dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
503		VERIFY3U(0, ==, zap_remove_int(dl->dl_os, obj, mintxg, tx));
504	}
505	zap_cursor_fini(&zc);
506
507	VERIFY3U(0, ==, dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
508	dlp = bonus->db_data;
509	dmu_buf_will_dirty(bonus, tx);
510	bzero(dlp, sizeof (*dlp));
511	dmu_buf_rele(bonus, FTAG);
512	mutex_exit(&dl->dl_lock);
513}
514
515/*
516 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
517 */
518void
519dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
520    dmu_tx_t *tx)
521{
522	dsl_deadlist_entry_t dle_tofind;
523	dsl_deadlist_entry_t *dle;
524	avl_index_t where;
525
526	ASSERT(!dl->dl_oldfmt);
527
528	mutex_enter(&dl->dl_lock);
529	dmu_buf_will_dirty(dl->dl_dbuf, tx);
530	dsl_deadlist_load_tree(dl);
531
532	dle_tofind.dle_mintxg = mintxg;
533	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
534	if (dle == NULL)
535		dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
536	while (dle) {
537		uint64_t used, comp, uncomp;
538		dsl_deadlist_entry_t *dle_next;
539
540		bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
541
542		VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
543		    &used, &comp, &uncomp));
544		ASSERT3U(dl->dl_phys->dl_used, >=, used);
545		ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
546		ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
547		dl->dl_phys->dl_used -= used;
548		dl->dl_phys->dl_comp -= comp;
549		dl->dl_phys->dl_uncomp -= uncomp;
550
551		VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object,
552		    dle->dle_mintxg, tx));
553
554		dle_next = AVL_NEXT(&dl->dl_tree, dle);
555		avl_remove(&dl->dl_tree, dle);
556		bpobj_close(&dle->dle_bpobj);
557		kmem_free(dle, sizeof (*dle));
558		dle = dle_next;
559	}
560	mutex_exit(&dl->dl_lock);
561}
562