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 VERIFY0(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 VERIFY0(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 VERIFY0(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_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 VERIFY0(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 bpobj_free(os, za.za_first_integer, tx);
168 zap_cursor_fini(&zc);
169 VERIFY0(dmu_object_free(os, dlobj, tx));
170 }
171
172 void
173 dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx)
174 {
175 dsl_deadlist_entry_t dle_tofind;
176 dsl_deadlist_entry_t *dle;
177 avl_index_t where;
178
179 if (dl->dl_oldfmt) {
180 bpobj_enqueue(&dl->dl_bpobj, bp, tx);
181 return;
182 }
183
184 dsl_deadlist_load_tree(dl);
185
186 dmu_buf_will_dirty(dl->dl_dbuf, tx);
187 mutex_enter(&dl->dl_lock);
188 dl->dl_phys->dl_used +=
189 bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
190 dl->dl_phys->dl_comp += BP_GET_PSIZE(bp);
191 dl->dl_phys->dl_uncomp += BP_GET_UCSIZE(bp);
192 mutex_exit(&dl->dl_lock);
193
194 dle_tofind.dle_mintxg = bp->blk_birth;
195 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
196 if (dle == NULL)
197 dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
198 else
199 dle = AVL_PREV(&dl->dl_tree, dle);
200 bpobj_enqueue(&dle->dle_bpobj, bp, tx);
201 }
202
203 /*
204 * Insert new key in deadlist, which must be > all current entries.
205 * mintxg is not inclusive.
206 */
207 void
208 dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
209 {
210 uint64_t obj;
211 dsl_deadlist_entry_t *dle;
212
213 if (dl->dl_oldfmt)
214 return;
215
216 dsl_deadlist_load_tree(dl);
217
218 dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
219 dle->dle_mintxg = mintxg;
220 obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
221 VERIFY0(bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
222 avl_add(&dl->dl_tree, dle);
223
224 VERIFY0(zap_add_int_key(dl->dl_os, dl->dl_object,
225 mintxg, obj, tx));
226 }
227
228 /*
229 * Remove this key, merging its entries into the previous key.
230 */
231 void
232 dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
233 {
234 dsl_deadlist_entry_t dle_tofind;
235 dsl_deadlist_entry_t *dle, *dle_prev;
236
237 if (dl->dl_oldfmt)
238 return;
239
240 dsl_deadlist_load_tree(dl);
241
242 dle_tofind.dle_mintxg = mintxg;
243 dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
244 dle_prev = AVL_PREV(&dl->dl_tree, dle);
245
246 bpobj_enqueue_subobj(&dle_prev->dle_bpobj,
247 dle->dle_bpobj.bpo_object, tx);
248
249 avl_remove(&dl->dl_tree, dle);
250 bpobj_close(&dle->dle_bpobj);
251 kmem_free(dle, sizeof (*dle));
252
253 VERIFY0(zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
254 }
255
256 /*
257 * Walk ds's snapshots to regenerate generate ZAP & AVL.
258 */
259 static void
260 dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
261 uint64_t mrs_obj, dmu_tx_t *tx)
262 {
263 dsl_deadlist_t dl;
264 dsl_pool_t *dp = dmu_objset_pool(os);
265
266 dsl_deadlist_open(&dl, os, dlobj);
267 if (dl.dl_oldfmt) {
268 dsl_deadlist_close(&dl);
269 return;
270 }
271
272 while (mrs_obj != 0) {
273 dsl_dataset_t *ds;
274 VERIFY0(dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
275 dsl_deadlist_add_key(&dl, ds->ds_phys->ds_prev_snap_txg, tx);
276 mrs_obj = ds->ds_phys->ds_prev_snap_obj;
277 dsl_dataset_rele(ds, FTAG);
278 }
279 dsl_deadlist_close(&dl);
280 }
281
282 uint64_t
283 dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
284 uint64_t mrs_obj, dmu_tx_t *tx)
285 {
286 dsl_deadlist_entry_t *dle;
287 uint64_t newobj;
288
289 newobj = dsl_deadlist_alloc(dl->dl_os, tx);
290
291 if (dl->dl_oldfmt) {
292 dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
293 return (newobj);
294 }
295
296 dsl_deadlist_load_tree(dl);
297
298 for (dle = avl_first(&dl->dl_tree); dle;
299 dle = AVL_NEXT(&dl->dl_tree, dle)) {
300 uint64_t obj;
301
302 if (dle->dle_mintxg >= maxtxg)
303 break;
304
305 obj = bpobj_alloc(dl->dl_os, SPA_MAXBLOCKSIZE, tx);
306 VERIFY0(zap_add_int_key(dl->dl_os, newobj,
307 dle->dle_mintxg, obj, tx));
308 }
309 return (newobj);
310 }
311
312 void
313 dsl_deadlist_space(dsl_deadlist_t *dl,
314 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
315 {
316 if (dl->dl_oldfmt) {
317 VERIFY0(bpobj_space(&dl->dl_bpobj,
318 usedp, compp, uncompp));
319 return;
320 }
321
322 mutex_enter(&dl->dl_lock);
323 *usedp = dl->dl_phys->dl_used;
324 *compp = dl->dl_phys->dl_comp;
325 *uncompp = dl->dl_phys->dl_uncomp;
326 mutex_exit(&dl->dl_lock);
327 }
328
329 /*
330 * return space used in the range (mintxg, maxtxg].
331 * Includes maxtxg, does not include mintxg.
332 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
333 * larger than any bp in the deadlist (eg. UINT64_MAX)).
334 */
335 void
336 dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
337 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
338 {
339 dsl_deadlist_entry_t *dle;
340 dsl_deadlist_entry_t dle_tofind;
341 avl_index_t where;
342
343 if (dl->dl_oldfmt) {
344 VERIFY0(bpobj_space_range(&dl->dl_bpobj,
345 mintxg, maxtxg, usedp, compp, uncompp));
346 return;
347 }
348
349 *usedp = *compp = *uncompp = 0;
350
351 mutex_enter(&dl->dl_lock);
352 dsl_deadlist_load_tree(dl);
353 dle_tofind.dle_mintxg = mintxg;
354 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
355 /*
356 * If we don't find this mintxg, there shouldn't be anything
357 * after it either.
358 */
359 ASSERT(dle != NULL ||
360 avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);
361
362 for (; dle && dle->dle_mintxg < maxtxg;
363 dle = AVL_NEXT(&dl->dl_tree, dle)) {
364 uint64_t used, comp, uncomp;
365
366 VERIFY0(bpobj_space(&dle->dle_bpobj,
367 &used, &comp, &uncomp));
368
369 *usedp += used;
370 *compp += comp;
371 *uncompp += uncomp;
372 }
373 mutex_exit(&dl->dl_lock);
374 }
375
376 static void
377 dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
378 dmu_tx_t *tx)
379 {
380 dsl_deadlist_entry_t dle_tofind;
381 dsl_deadlist_entry_t *dle;
382 avl_index_t where;
383 uint64_t used, comp, uncomp;
384 bpobj_t bpo;
385
386 VERIFY0(bpobj_open(&bpo, dl->dl_os, obj));
387 VERIFY0(bpobj_space(&bpo, &used, &comp, &uncomp));
388 bpobj_close(&bpo);
389
390 dsl_deadlist_load_tree(dl);
391
392 dmu_buf_will_dirty(dl->dl_dbuf, tx);
393 mutex_enter(&dl->dl_lock);
394 dl->dl_phys->dl_used += used;
395 dl->dl_phys->dl_comp += comp;
396 dl->dl_phys->dl_uncomp += uncomp;
397 mutex_exit(&dl->dl_lock);
398
399 dle_tofind.dle_mintxg = birth;
400 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
401 if (dle == NULL)
402 dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
403 bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
404 }
405
406 static int
407 dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
408 {
409 dsl_deadlist_t *dl = arg;
410 dsl_deadlist_insert(dl, bp, tx);
411 return (0);
412 }
413
414 /*
415 * Merge the deadlist pointed to by 'obj' into dl. obj will be left as
416 * an empty deadlist.
417 */
418 void
419 dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
420 {
421 zap_cursor_t zc;
422 zap_attribute_t za;
423 dmu_buf_t *bonus;
424 dsl_deadlist_phys_t *dlp;
425 dmu_object_info_t doi;
426
427 VERIFY0(dmu_object_info(dl->dl_os, obj, &doi));
428 if (doi.doi_type == DMU_OT_BPOBJ) {
429 bpobj_t bpo;
430 VERIFY0(bpobj_open(&bpo, dl->dl_os, obj));
431 VERIFY0(bpobj_iterate(&bpo,
432 dsl_deadlist_insert_cb, dl, tx));
433 bpobj_close(&bpo);
434 return;
435 }
436
437 for (zap_cursor_init(&zc, dl->dl_os, obj);
438 zap_cursor_retrieve(&zc, &za) == 0;
439 zap_cursor_advance(&zc)) {
440 uint64_t mintxg = strtonum(za.za_name, NULL);
441 dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
442 VERIFY0(zap_remove_int(dl->dl_os, obj, mintxg, tx));
443 }
444 zap_cursor_fini(&zc);
445
446 VERIFY0(dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
447 dlp = bonus->db_data;
448 dmu_buf_will_dirty(bonus, tx);
449 bzero(dlp, sizeof (*dlp));
450 dmu_buf_rele(bonus, FTAG);
451 }
452
453 /*
454 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
455 */
456 void
457 dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
458 dmu_tx_t *tx)
459 {
460 dsl_deadlist_entry_t dle_tofind;
461 dsl_deadlist_entry_t *dle;
462 avl_index_t where;
463
464 ASSERT(!dl->dl_oldfmt);
465 dmu_buf_will_dirty(dl->dl_dbuf, tx);
466 dsl_deadlist_load_tree(dl);
467
468 dle_tofind.dle_mintxg = mintxg;
469 dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
470 if (dle == NULL)
471 dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
472 while (dle) {
473 uint64_t used, comp, uncomp;
474 dsl_deadlist_entry_t *dle_next;
475
476 bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
477
478 VERIFY0(bpobj_space(&dle->dle_bpobj,
479 &used, &comp, &uncomp));
480 mutex_enter(&dl->dl_lock);
481 ASSERT3U(dl->dl_phys->dl_used, >=, used);
482 ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
483 ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
484 dl->dl_phys->dl_used -= used;
485 dl->dl_phys->dl_comp -= comp;
486 dl->dl_phys->dl_uncomp -= uncomp;
487 mutex_exit(&dl->dl_lock);
488
489 VERIFY0(zap_remove_int(dl->dl_os, dl->dl_object,
490 dle->dle_mintxg, tx));
491
492 dle_next = AVL_NEXT(&dl->dl_tree, dle);
493 avl_remove(&dl->dl_tree, dle);
494 bpobj_close(&dle->dle_bpobj);
495 kmem_free(dle, sizeof (*dle));
496 dle = dle_next;
497 }
498 }