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 */
24
25#include <sys/dsl_dataset.h>
26#include <sys/dmu.h>
27#include <sys/refcount.h>
28#include <sys/zap.h>
29#include <sys/zfs_context.h>
30#include <sys/dsl_pool.h>
31
32static int
33dsl_deadlist_compare(const void *arg1, const void *arg2)
34{
35	const dsl_deadlist_entry_t *dle1 = arg1;
36	const dsl_deadlist_entry_t *dle2 = arg2;
37
38	if (dle1->dle_mintxg < dle2->dle_mintxg)
39		return (-1);
40	else if (dle1->dle_mintxg > dle2->dle_mintxg)
41		return (+1);
42	else
43		return (0);
44}
45
46static void
47dsl_deadlist_load_tree(dsl_deadlist_t *dl)
48{
49	zap_cursor_t zc;
50	zap_attribute_t za;
51
52	ASSERT(!dl->dl_oldfmt);
53	if (dl->dl_havetree)
54		return;
55
56	avl_create(&dl->dl_tree, dsl_deadlist_compare,
57	    sizeof (dsl_deadlist_entry_t),
58	    offsetof(dsl_deadlist_entry_t, dle_node));
59	for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
60	    zap_cursor_retrieve(&zc, &za) == 0;
61	    zap_cursor_advance(&zc)) {
62		dsl_deadlist_entry_t *dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
63		dle->dle_mintxg = strtonum(za.za_name, NULL);
64		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os,
65		    za.za_first_integer));
66		avl_add(&dl->dl_tree, dle);
67	}
68	zap_cursor_fini(&zc);
69	dl->dl_havetree = B_TRUE;
70}
71
72void
73dsl_deadlist_open(dsl_deadlist_t *dl, objset_t *os, uint64_t object)
74{
75	dmu_object_info_t doi;
76
77	mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
78	dl->dl_os = os;
79	dl->dl_object = object;
80	VERIFY3U(0, ==, dmu_bonus_hold(os, object, dl, &dl->dl_dbuf));
81	dmu_object_info_from_db(dl->dl_dbuf, &doi);
82	if (doi.doi_type == DMU_OT_BPOBJ) {
83		dmu_buf_rele(dl->dl_dbuf, dl);
84		dl->dl_dbuf = NULL;
85		dl->dl_oldfmt = B_TRUE;
86		VERIFY3U(0, ==, bpobj_open(&dl->dl_bpobj, os, object));
87		return;
88	}
89
90	dl->dl_oldfmt = B_FALSE;
91	dl->dl_phys = dl->dl_dbuf->db_data;
92	dl->dl_havetree = B_FALSE;
93}
94
95void
96dsl_deadlist_close(dsl_deadlist_t *dl)
97{
98	void *cookie = NULL;
99	dsl_deadlist_entry_t *dle;
100
101	if (dl->dl_oldfmt) {
102		dl->dl_oldfmt = B_FALSE;
103		bpobj_close(&dl->dl_bpobj);
104		return;
105	}
106
107	if (dl->dl_havetree) {
108		while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
109		    != NULL) {
110			bpobj_close(&dle->dle_bpobj);
111			kmem_free(dle, sizeof (*dle));
112		}
113		avl_destroy(&dl->dl_tree);
114	}
115	dmu_buf_rele(dl->dl_dbuf, dl);
116	mutex_destroy(&dl->dl_lock);
117	dl->dl_dbuf = NULL;
118	dl->dl_phys = NULL;
119}
120
121uint64_t
122dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx)
123{
124	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
125		return (bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx));
126	return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
127	    sizeof (dsl_deadlist_phys_t), tx));
128}
129
130void
131dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx)
132{
133	dmu_object_info_t doi;
134	zap_cursor_t zc;
135	zap_attribute_t za;
136
137	VERIFY3U(0, ==, dmu_object_info(os, dlobj, &doi));
138	if (doi.doi_type == DMU_OT_BPOBJ) {
139		bpobj_free(os, dlobj, tx);
140		return;
141	}
142
143	for (zap_cursor_init(&zc, os, dlobj);
144	    zap_cursor_retrieve(&zc, &za) == 0;
145	    zap_cursor_advance(&zc))
146		bpobj_free(os, za.za_first_integer, tx);
147	zap_cursor_fini(&zc);
148	VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx));
149}
150
151void
152dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx)
153{
154	dsl_deadlist_entry_t dle_tofind;
155	dsl_deadlist_entry_t *dle;
156	avl_index_t where;
157
158	if (dl->dl_oldfmt) {
159		bpobj_enqueue(&dl->dl_bpobj, bp, tx);
160		return;
161	}
162
163	dsl_deadlist_load_tree(dl);
164
165	dmu_buf_will_dirty(dl->dl_dbuf, tx);
166	mutex_enter(&dl->dl_lock);
167	dl->dl_phys->dl_used +=
168	    bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
169	dl->dl_phys->dl_comp += BP_GET_PSIZE(bp);
170	dl->dl_phys->dl_uncomp += BP_GET_UCSIZE(bp);
171	mutex_exit(&dl->dl_lock);
172
173	dle_tofind.dle_mintxg = bp->blk_birth;
174	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
175	if (dle == NULL)
176		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
177	else
178		dle = AVL_PREV(&dl->dl_tree, dle);
179	bpobj_enqueue(&dle->dle_bpobj, bp, tx);
180}
181
182/*
183 * Insert new key in deadlist, which must be > all current entries.
184 * mintxg is not inclusive.
185 */
186void
187dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
188{
189	uint64_t obj;
190	dsl_deadlist_entry_t *dle;
191
192	if (dl->dl_oldfmt)
193		return;
194
195	dsl_deadlist_load_tree(dl);
196
197	dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
198	dle->dle_mintxg = mintxg;
199	obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
200	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
201	avl_add(&dl->dl_tree, dle);
202
203	VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, dl->dl_object,
204	    mintxg, obj, tx));
205}
206
207/*
208 * Remove this key, merging its entries into the previous key.
209 */
210void
211dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
212{
213	dsl_deadlist_entry_t dle_tofind;
214	dsl_deadlist_entry_t *dle, *dle_prev;
215
216	if (dl->dl_oldfmt)
217		return;
218
219	dsl_deadlist_load_tree(dl);
220
221	dle_tofind.dle_mintxg = mintxg;
222	dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
223	dle_prev = AVL_PREV(&dl->dl_tree, dle);
224
225	bpobj_enqueue_subobj(&dle_prev->dle_bpobj,
226	    dle->dle_bpobj.bpo_object, tx);
227
228	avl_remove(&dl->dl_tree, dle);
229	bpobj_close(&dle->dle_bpobj);
230	kmem_free(dle, sizeof (*dle));
231
232	VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
233}
234
235/*
236 * Walk ds's snapshots to regenerate generate ZAP & AVL.
237 */
238static void
239dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
240    uint64_t mrs_obj, dmu_tx_t *tx)
241{
242	dsl_deadlist_t dl;
243	dsl_pool_t *dp = dmu_objset_pool(os);
244
245	dsl_deadlist_open(&dl, os, dlobj);
246	if (dl.dl_oldfmt) {
247		dsl_deadlist_close(&dl);
248		return;
249	}
250
251	while (mrs_obj != 0) {
252		dsl_dataset_t *ds;
253		VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
254		dsl_deadlist_add_key(&dl, ds->ds_phys->ds_prev_snap_txg, tx);
255		mrs_obj = ds->ds_phys->ds_prev_snap_obj;
256		dsl_dataset_rele(ds, FTAG);
257	}
258	dsl_deadlist_close(&dl);
259}
260
261uint64_t
262dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
263    uint64_t mrs_obj, dmu_tx_t *tx)
264{
265	dsl_deadlist_entry_t *dle;
266	uint64_t newobj;
267
268	newobj = dsl_deadlist_alloc(dl->dl_os, tx);
269
270	if (dl->dl_oldfmt) {
271		dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
272		return (newobj);
273	}
274
275	dsl_deadlist_load_tree(dl);
276
277	for (dle = avl_first(&dl->dl_tree); dle;
278	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
279		uint64_t obj;
280
281		if (dle->dle_mintxg >= maxtxg)
282			break;
283
284		obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
285		VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj,
286		    dle->dle_mintxg, obj, tx));
287	}
288	return (newobj);
289}
290
291void
292dsl_deadlist_space(dsl_deadlist_t *dl,
293    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
294{
295	if (dl->dl_oldfmt) {
296		VERIFY3U(0, ==, bpobj_space(&dl->dl_bpobj,
297		    usedp, compp, uncompp));
298		return;
299	}
300
301	mutex_enter(&dl->dl_lock);
302	*usedp = dl->dl_phys->dl_used;
303	*compp = dl->dl_phys->dl_comp;
304	*uncompp = dl->dl_phys->dl_uncomp;
305	mutex_exit(&dl->dl_lock);
306}
307
308/*
309 * return space used in the range (mintxg, maxtxg].
310 * Includes maxtxg, does not include mintxg.
311 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
312 * UINT64_MAX).
313 */
314void
315dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
316    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
317{
318	dsl_deadlist_entry_t dle_tofind;
319	dsl_deadlist_entry_t *dle;
320	avl_index_t where;
321
322	if (dl->dl_oldfmt) {
323		VERIFY3U(0, ==, bpobj_space_range(&dl->dl_bpobj,
324		    mintxg, maxtxg, usedp, compp, uncompp));
325		return;
326	}
327
328	dsl_deadlist_load_tree(dl);
329	*usedp = *compp = *uncompp = 0;
330
331	dle_tofind.dle_mintxg = mintxg;
332	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
333	/*
334	 * If we don't find this mintxg, there shouldn't be anything
335	 * after it either.
336	 */
337	ASSERT(dle != NULL ||
338	    avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);
339	for (; dle && dle->dle_mintxg < maxtxg;
340	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
341		uint64_t used, comp, uncomp;
342
343		VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
344		    &used, &comp, &uncomp));
345
346		*usedp += used;
347		*compp += comp;
348		*uncompp += uncomp;
349	}
350}
351
352static void
353dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
354    dmu_tx_t *tx)
355{
356	dsl_deadlist_entry_t dle_tofind;
357	dsl_deadlist_entry_t *dle;
358	avl_index_t where;
359	uint64_t used, comp, uncomp;
360	bpobj_t bpo;
361
362	VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
363	VERIFY3U(0, ==, bpobj_space(&bpo, &used, &comp, &uncomp));
364	bpobj_close(&bpo);
365
366	dsl_deadlist_load_tree(dl);
367
368	dmu_buf_will_dirty(dl->dl_dbuf, tx);
369	mutex_enter(&dl->dl_lock);
370	dl->dl_phys->dl_used += used;
371	dl->dl_phys->dl_comp += comp;
372	dl->dl_phys->dl_uncomp += uncomp;
373	mutex_exit(&dl->dl_lock);
374
375	dle_tofind.dle_mintxg = birth;
376	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
377	if (dle == NULL)
378		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
379	bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
380}
381
382static int
383dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
384{
385	dsl_deadlist_t *dl = arg;
386	dsl_deadlist_insert(dl, bp, tx);
387	return (0);
388}
389
390/*
391 * Merge the deadlist pointed to by 'obj' into dl.  obj will be left as
392 * an empty deadlist.
393 */
394void
395dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
396{
397	zap_cursor_t zc;
398	zap_attribute_t za;
399	dmu_buf_t *bonus;
400	dsl_deadlist_phys_t *dlp;
401	dmu_object_info_t doi;
402
403	VERIFY3U(0, ==, dmu_object_info(dl->dl_os, obj, &doi));
404	if (doi.doi_type == DMU_OT_BPOBJ) {
405		bpobj_t bpo;
406		VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
407		VERIFY3U(0, ==, bpobj_iterate(&bpo,
408		    dsl_deadlist_insert_cb, dl, tx));
409		bpobj_close(&bpo);
410		return;
411	}
412
413	for (zap_cursor_init(&zc, dl->dl_os, obj);
414	    zap_cursor_retrieve(&zc, &za) == 0;
415	    zap_cursor_advance(&zc)) {
416		uint64_t mintxg = strtonum(za.za_name, NULL);
417		dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
418		VERIFY3U(0, ==, zap_remove_int(dl->dl_os, obj, mintxg, tx));
419	}
420	zap_cursor_fini(&zc);
421
422	VERIFY3U(0, ==, dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
423	dlp = bonus->db_data;
424	dmu_buf_will_dirty(bonus, tx);
425	bzero(dlp, sizeof (*dlp));
426	dmu_buf_rele(bonus, FTAG);
427}
428
429/*
430 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
431 */
432void
433dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
434    dmu_tx_t *tx)
435{
436	dsl_deadlist_entry_t dle_tofind;
437	dsl_deadlist_entry_t *dle;
438	avl_index_t where;
439
440	ASSERT(!dl->dl_oldfmt);
441	dmu_buf_will_dirty(dl->dl_dbuf, tx);
442	dsl_deadlist_load_tree(dl);
443
444	dle_tofind.dle_mintxg = mintxg;
445	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
446	if (dle == NULL)
447		dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
448	while (dle) {
449		uint64_t used, comp, uncomp;
450		dsl_deadlist_entry_t *dle_next;
451
452		bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
453
454		VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
455		    &used, &comp, &uncomp));
456		mutex_enter(&dl->dl_lock);
457		ASSERT3U(dl->dl_phys->dl_used, >=, used);
458		ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
459		ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
460		dl->dl_phys->dl_used -= used;
461		dl->dl_phys->dl_comp -= comp;
462		dl->dl_phys->dl_uncomp -= uncomp;
463		mutex_exit(&dl->dl_lock);
464
465		VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object,
466		    dle->dle_mintxg, tx));
467
468		dle_next = AVL_NEXT(&dl->dl_tree, dle);
469		avl_remove(&dl->dl_tree, dle);
470		bpobj_close(&dle->dle_bpobj);
471		kmem_free(dle, sizeof (*dle));
472		dle = dle_next;
473	}
474}
475