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 by Delphix. All rights reserved.
24 */
25
26 #include <sys/dsl_dataset.h>
27 #include <sys/dmu.h>
28 #include <sys/refcount.h>
29 #include <sys/zap.h>
30 #include <sys/zfs_context.h>
31 #include <sys/dsl_pool.h>
32
33 /*
34 * Deadlist concurrency:
35 *
36 * Deadlists can only be modified from the syncing thread.
37 *
38 * Except for dsl_deadlist_insert(), it can only be modified with the
39 * dp_config_rwlock held with RW_WRITER.
40 *
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
43 * with RW_READER.
44 *
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.
51 */
52
53 static int
54 dsl_deadlist_compare(const void *arg1, const void *arg2)
55 {
56 const dsl_deadlist_entry_t *dle1 = arg1;
57 const dsl_deadlist_entry_t *dle2 = arg2;
58
59 if (dle1->dle_mintxg < dle2->dle_mintxg)
60 return (-1);
61 else if (dle1->dle_mintxg > dle2->dle_mintxg)
62 return (+1);
63 else
64 return (0);
65 }
66
67 static void
68 dsl_deadlist_load_tree(dsl_deadlist_t *dl)
69 {
70 zap_cursor_t zc;
71 zap_attribute_t za;
72
73 ASSERT(!dl->dl_oldfmt);
74 if (dl->dl_havetree)
75 return;
76
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);
88 }
89 zap_cursor_fini(&zc);
90 dl->dl_havetree = B_TRUE;
91 }
92
93 void
94 dsl_deadlist_open(dsl_deadlist_t *dl, objset_t *os, uint64_t object)
95 {
96 dmu_object_info_t doi;
97
98 mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
99 dl->dl_os = os;
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);
105 dl->dl_dbuf = NULL;
106 dl->dl_oldfmt = B_TRUE;
107 VERIFY3U(0, ==, bpobj_open(&dl->dl_bpobj, os, object));
108 return;
109 }
110
111 dl->dl_oldfmt = B_FALSE;
112 dl->dl_phys = dl->dl_dbuf->db_data;
113 dl->dl_havetree = B_FALSE;
114 }
115
116 void
117 dsl_deadlist_close(dsl_deadlist_t *dl)
118 {
119 void *cookie = NULL;
120 dsl_deadlist_entry_t *dle;
121
122 if (dl->dl_oldfmt) {
123 dl->dl_oldfmt = B_FALSE;
124 bpobj_close(&dl->dl_bpobj);
125 return;
126 }
127
128 if (dl->dl_havetree) {
129 while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
130 != NULL) {
131 bpobj_close(&dle->dle_bpobj);
132 kmem_free(dle, sizeof (*dle));
133 }
134 avl_destroy(&dl->dl_tree);
135 }
136 dmu_buf_rele(dl->dl_dbuf, dl);
137 mutex_destroy(&dl->dl_lock);
138 dl->dl_dbuf = NULL;
139 dl->dl_phys = NULL;
140 }
141
142 uint64_t
143 dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx)
144 {
145 if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
146 return (bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx));
147 return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
148 sizeof (dsl_deadlist_phys_t), tx));
149 }
150
151 void
152 dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx)
153 {
154 dmu_object_info_t doi;
155 zap_cursor_t zc;
156 zap_attribute_t za;
157
158 VERIFY3U(0, ==, dmu_object_info(os, dlobj, &doi));
159 if (doi.doi_type == DMU_OT_BPOBJ) {
160 bpobj_free(os, dlobj, tx);
161 return;
162 }
163
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);
170 else
171 bpobj_free(os, obj, tx);
172 }
173 zap_cursor_fini(&zc);
174 VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx));
175 }
176
177 static void
178 dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
179 const blkptr_t *bp, dmu_tx_t *tx)
180 {
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_OLD_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));
189 }
190 bpobj_enqueue(&dle->dle_bpobj, bp, tx);
191 }
192
193 static void
194 dle_enqueue_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
195 uint64_t obj, dmu_tx_t *tx)
196 {
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);
200 } else {
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));
206 }
207 }
208
209 void
210 dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx)
211 {
212 dsl_deadlist_entry_t dle_tofind;
213 dsl_deadlist_entry_t *dle;
214 avl_index_t where;
215
216 if (dl->dl_oldfmt) {
217 bpobj_enqueue(&dl->dl_bpobj, bp, tx);
218 return;
219 }
220
221 dsl_deadlist_load_tree(dl);
222
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);
230
231 dle_tofind.dle_mintxg = bp->blk_birth;
232 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
233 if (dle == NULL)
234 dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
235 else
236 dle = AVL_PREV(&dl->dl_tree, dle);
237 dle_enqueue(dl, dle, bp, tx);
238 }
239
240 /*
241 * Insert new key in deadlist, which must be > all current entries.
242 * mintxg is not inclusive.
243 */
244 void
245 dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
246 {
247 uint64_t obj;
248 dsl_deadlist_entry_t *dle;
249
250 if (dl->dl_oldfmt)
251 return;
252
253 dsl_deadlist_load_tree(dl);
254
255 dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
256 dle->dle_mintxg = mintxg;
257 obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
258 VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
259 avl_add(&dl->dl_tree, dle);
260
261 VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, dl->dl_object,
262 mintxg, obj, tx));
263 }
264
265 /*
266 * Remove this key, merging its entries into the previous key.
267 */
268 void
269 dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
270 {
271 dsl_deadlist_entry_t dle_tofind;
272 dsl_deadlist_entry_t *dle, *dle_prev;
273
274 if (dl->dl_oldfmt)
275 return;
276
277 dsl_deadlist_load_tree(dl);
278
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);
282
283 dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx);
284
285 avl_remove(&dl->dl_tree, dle);
286 bpobj_close(&dle->dle_bpobj);
287 kmem_free(dle, sizeof (*dle));
288
289 VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
290 }
291
292 /*
293 * Walk ds's snapshots to regenerate generate ZAP & AVL.
294 */
295 static void
296 dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
297 uint64_t mrs_obj, dmu_tx_t *tx)
298 {
299 dsl_deadlist_t dl;
300 dsl_pool_t *dp = dmu_objset_pool(os);
301
302 dsl_deadlist_open(&dl, os, dlobj);
303 if (dl.dl_oldfmt) {
304 dsl_deadlist_close(&dl);
305 return;
306 }
307
308 while (mrs_obj != 0) {
309 dsl_dataset_t *ds;
310 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
311 dsl_deadlist_add_key(&dl,
312 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
313 mrs_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
314 dsl_dataset_rele(ds, FTAG);
315 }
316 dsl_deadlist_close(&dl);
317 }
318
319 uint64_t
320 dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
321 uint64_t mrs_obj, dmu_tx_t *tx)
322 {
323 dsl_deadlist_entry_t *dle;
324 uint64_t newobj;
325
326 newobj = dsl_deadlist_alloc(dl->dl_os, tx);
327
328 if (dl->dl_oldfmt) {
329 dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
330 return (newobj);
331 }
332
333 dsl_deadlist_load_tree(dl);
334
335 for (dle = avl_first(&dl->dl_tree); dle;
336 dle = AVL_NEXT(&dl->dl_tree, dle)) {
337 uint64_t obj;
338
339 if (dle->dle_mintxg >= maxtxg)
340 break;
341
342 obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
343 VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj,
344 dle->dle_mintxg, obj, tx));
345 }
346 return (newobj);
347 }
348
349 void
350 dsl_deadlist_space(dsl_deadlist_t *dl,
351 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
352 {
353 if (dl->dl_oldfmt) {
354 VERIFY3U(0, ==, bpobj_space(&dl->dl_bpobj,
355 usedp, compp, uncompp));
356 return;
357 }
358
359 mutex_enter(&dl->dl_lock);
360 *usedp = dl->dl_phys->dl_used;
361 *compp = dl->dl_phys->dl_comp;
362 *uncompp = dl->dl_phys->dl_uncomp;
363 mutex_exit(&dl->dl_lock);
364 }
365
366 /*
367 * return space used in the range (mintxg, maxtxg].
368 * Includes maxtxg, does not include mintxg.
369 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
370 * larger than any bp in the deadlist (eg. UINT64_MAX)).
371 */
372 void
373 dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
374 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
375 {
376 dsl_deadlist_entry_t *dle;
377 dsl_deadlist_entry_t dle_tofind;
378 avl_index_t where;
379
380 if (dl->dl_oldfmt) {
381 VERIFY3U(0, ==, bpobj_space_range(&dl->dl_bpobj,
382 mintxg, maxtxg, usedp, compp, uncompp));
383 return;
384 }
385
386 *usedp = *compp = *uncompp = 0;
387
388 mutex_enter(&dl->dl_lock);
389 dsl_deadlist_load_tree(dl);
390 dle_tofind.dle_mintxg = mintxg;
391 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
392 /*
393 * If we don't find this mintxg, there shouldn't be anything
394 * after it either.
395 */
396 ASSERT(dle != NULL ||
397 avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);
398
399 for (; dle && dle->dle_mintxg < maxtxg;
400 dle = AVL_NEXT(&dl->dl_tree, dle)) {
401 uint64_t used, comp, uncomp;
402
403 VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
404 &used, &comp, &uncomp));
405
406 *usedp += used;
407 *compp += comp;
408 *uncompp += uncomp;
409 }
410 mutex_exit(&dl->dl_lock);
411 }
412
413 static void
414 dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
415 dmu_tx_t *tx)
416 {
417 dsl_deadlist_entry_t dle_tofind;
418 dsl_deadlist_entry_t *dle;
419 avl_index_t where;
420 uint64_t used, comp, uncomp;
421 bpobj_t bpo;
422
423 VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
424 VERIFY3U(0, ==, bpobj_space(&bpo, &used, &comp, &uncomp));
425 bpobj_close(&bpo);
426
427 dsl_deadlist_load_tree(dl);
428
429 dmu_buf_will_dirty(dl->dl_dbuf, tx);
430 mutex_enter(&dl->dl_lock);
431 dl->dl_phys->dl_used += used;
432 dl->dl_phys->dl_comp += comp;
433 dl->dl_phys->dl_uncomp += uncomp;
434 mutex_exit(&dl->dl_lock);
435
436 dle_tofind.dle_mintxg = birth;
437 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
438 if (dle == NULL)
439 dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
440 dle_enqueue_subobj(dl, dle, obj, tx);
441 }
442
443 static int
444 dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
445 {
446 dsl_deadlist_t *dl = arg;
447 dsl_deadlist_insert(dl, bp, tx);
448 return (0);
449 }
450
451 /*
452 * Merge the deadlist pointed to by 'obj' into dl. obj will be left as
453 * an empty deadlist.
454 */
455 void
456 dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
457 {
458 zap_cursor_t zc;
459 zap_attribute_t za;
460 dmu_buf_t *bonus;
461 dsl_deadlist_phys_t *dlp;
462 dmu_object_info_t doi;
463
464 VERIFY3U(0, ==, dmu_object_info(dl->dl_os, obj, &doi));
465 if (doi.doi_type == DMU_OT_BPOBJ) {
466 bpobj_t bpo;
467 VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
468 VERIFY3U(0, ==, bpobj_iterate(&bpo,
469 dsl_deadlist_insert_cb, dl, tx));
470 bpobj_close(&bpo);
471 return;
472 }
473
474 for (zap_cursor_init(&zc, dl->dl_os, obj);
475 zap_cursor_retrieve(&zc, &za) == 0;
476 zap_cursor_advance(&zc)) {
477 uint64_t mintxg = strtonum(za.za_name, NULL);
478 dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
479 VERIFY3U(0, ==, zap_remove_int(dl->dl_os, obj, mintxg, tx));
480 }
481 zap_cursor_fini(&zc);
482
483 VERIFY3U(0, ==, dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
484 dlp = bonus->db_data;
485 dmu_buf_will_dirty(bonus, tx);
486 bzero(dlp, sizeof (*dlp));
487 dmu_buf_rele(bonus, FTAG);
488 }
489
490 /*
491 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
492 */
493 void
494 dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
495 dmu_tx_t *tx)
496 {
497 dsl_deadlist_entry_t dle_tofind;
498 dsl_deadlist_entry_t *dle;
499 avl_index_t where;
500
501 ASSERT(!dl->dl_oldfmt);
502 dmu_buf_will_dirty(dl->dl_dbuf, tx);
503 dsl_deadlist_load_tree(dl);
504
505 dle_tofind.dle_mintxg = mintxg;
506 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
507 if (dle == NULL)
508 dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
509 while (dle) {
510 uint64_t used, comp, uncomp;
511 dsl_deadlist_entry_t *dle_next;
512
513 bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
514
515 VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
516 &used, &comp, &uncomp));
517 mutex_enter(&dl->dl_lock);
518 ASSERT3U(dl->dl_phys->dl_used, >=, used);
519 ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
520 ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
521 dl->dl_phys->dl_used -= used;
522 dl->dl_phys->dl_comp -= comp;
523 dl->dl_phys->dl_uncomp -= uncomp;
524 mutex_exit(&dl->dl_lock);
525
526 VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object,
527 dle->dle_mintxg, tx));
528
529 dle_next = AVL_NEXT(&dl->dl_tree, dle);
530 avl_remove(&dl->dl_tree, dle);
531 bpobj_close(&dle->dle_bpobj);
532 kmem_free(dle, sizeof (*dle));
533 dle = dle_next;
534 }
535 }