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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 */
25
26 #include <sys/bpobj.h>
27 #include <sys/zfs_context.h>
28 #include <sys/refcount.h>
29 #include <sys/dsl_pool.h>
30 #include <sys/zfeature.h>
31 #include <sys/zap.h>
32
33 /*
34 * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
35 */
36 uint64_t
37 bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx)
38 {
39 zfeature_info_t *empty_bpobj_feat =
40 &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ];
41 spa_t *spa = dmu_objset_spa(os);
42 dsl_pool_t *dp = dmu_objset_pool(os);
43
44 if (spa_feature_is_enabled(spa, empty_bpobj_feat)) {
45 if (!spa_feature_is_active(spa, empty_bpobj_feat)) {
46 ASSERT0(dp->dp_empty_bpobj);
47 dp->dp_empty_bpobj =
48 bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx);
49 VERIFY(zap_add(os,
50 DMU_POOL_DIRECTORY_OBJECT,
51 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
52 &dp->dp_empty_bpobj, tx) == 0);
53 }
54 spa_feature_incr(spa, empty_bpobj_feat, tx);
55 ASSERT(dp->dp_empty_bpobj != 0);
56 return (dp->dp_empty_bpobj);
57 } else {
58 return (bpobj_alloc(os, blocksize, tx));
59 }
60 }
61
62 void
63 bpobj_decr_empty(objset_t *os, dmu_tx_t *tx)
64 {
65 zfeature_info_t *empty_bpobj_feat =
66 &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ];
67 dsl_pool_t *dp = dmu_objset_pool(os);
68
69 spa_feature_decr(dmu_objset_spa(os), empty_bpobj_feat, tx);
70 if (!spa_feature_is_active(dmu_objset_spa(os), empty_bpobj_feat)) {
71 VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
72 DMU_POOL_DIRECTORY_OBJECT,
73 DMU_POOL_EMPTY_BPOBJ, tx));
74 VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
75 dp->dp_empty_bpobj = 0;
76 }
77 }
78
79 uint64_t
80 bpobj_alloc(objset_t *os, int blocksize, dmu_tx_t *tx)
81 {
82 int size;
83
84 if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
85 size = BPOBJ_SIZE_V0;
86 else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
87 size = BPOBJ_SIZE_V1;
88 else
89 size = sizeof (bpobj_phys_t);
90
91 return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
92 DMU_OT_BPOBJ_HDR, size, tx));
93 }
94
95 void
96 bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx)
97 {
98 int64_t i;
99 bpobj_t bpo;
100 dmu_object_info_t doi;
101 int epb;
102 dmu_buf_t *dbuf = NULL;
103
104 ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj);
105 VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
106
107 mutex_enter(&bpo.bpo_lock);
108
109 if (!bpo.bpo_havesubobj || bpo.bpo_phys->bpo_subobjs == 0)
110 goto out;
111
112 VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
113 epb = doi.doi_data_block_size / sizeof (uint64_t);
114
115 for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
116 uint64_t *objarray;
117 uint64_t offset, blkoff;
118
119 offset = i * sizeof (uint64_t);
120 blkoff = P2PHASE(i, epb);
121
122 if (dbuf == NULL || dbuf->db_offset > offset) {
123 if (dbuf)
124 dmu_buf_rele(dbuf, FTAG);
125 VERIFY3U(0, ==, dmu_buf_hold(os,
126 bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
127 }
128
129 ASSERT3U(offset, >=, dbuf->db_offset);
130 ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
131
132 objarray = dbuf->db_data;
133 bpobj_free(os, objarray[blkoff], tx);
134 }
135 if (dbuf) {
136 dmu_buf_rele(dbuf, FTAG);
137 dbuf = NULL;
138 }
139 VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
140
141 out:
142 mutex_exit(&bpo.bpo_lock);
143 bpobj_close(&bpo);
144
145 VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
146 }
147
148 int
149 bpobj_open(bpobj_t *bpo, objset_t *os, uint64_t object)
150 {
151 dmu_object_info_t doi;
152 int err;
153
154 err = dmu_object_info(os, object, &doi);
155 if (err)
156 return (err);
157
158 bzero(bpo, sizeof (*bpo));
159 mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
160
161 ASSERT(bpo->bpo_dbuf == NULL);
162 ASSERT(bpo->bpo_phys == NULL);
163 ASSERT(object != 0);
164 ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
165 ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
166
167 err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
168 if (err)
169 return (err);
170
171 bpo->bpo_os = os;
172 bpo->bpo_object = object;
173 bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
174 bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
175 bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
176 bpo->bpo_phys = bpo->bpo_dbuf->db_data;
177 return (0);
178 }
179
180 void
181 bpobj_close(bpobj_t *bpo)
182 {
183 /* Lame workaround for closing a bpobj that was never opened. */
184 if (bpo->bpo_object == 0)
185 return;
186
187 dmu_buf_rele(bpo->bpo_dbuf, bpo);
188 if (bpo->bpo_cached_dbuf != NULL)
189 dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
190 bpo->bpo_dbuf = NULL;
191 bpo->bpo_phys = NULL;
192 bpo->bpo_cached_dbuf = NULL;
193 bpo->bpo_object = 0;
194
195 mutex_destroy(&bpo->bpo_lock);
196 }
197
198 static int
199 bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
200 boolean_t free)
201 {
202 dmu_object_info_t doi;
203 int epb;
204 int64_t i;
205 int err = 0;
206 dmu_buf_t *dbuf = NULL;
207
208 mutex_enter(&bpo->bpo_lock);
209
210 if (free)
211 dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
212
213 for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
214 blkptr_t *bparray;
215 blkptr_t *bp;
216 uint64_t offset, blkoff;
217
218 offset = i * sizeof (blkptr_t);
219 blkoff = P2PHASE(i, bpo->bpo_epb);
220
221 if (dbuf == NULL || dbuf->db_offset > offset) {
222 if (dbuf)
223 dmu_buf_rele(dbuf, FTAG);
224 err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
225 FTAG, &dbuf, 0);
226 if (err)
227 break;
228 }
229
230 ASSERT3U(offset, >=, dbuf->db_offset);
231 ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
232
233 bparray = dbuf->db_data;
234 bp = &bparray[blkoff];
235 err = func(arg, bp, tx);
236 if (err)
237 break;
238 if (free) {
239 bpo->bpo_phys->bpo_bytes -=
240 bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
241 ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
242 if (bpo->bpo_havecomp) {
243 bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
244 bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
245 }
246 bpo->bpo_phys->bpo_num_blkptrs--;
247 ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
248 }
249 }
250 if (dbuf) {
251 dmu_buf_rele(dbuf, FTAG);
252 dbuf = NULL;
253 }
254 if (free) {
255 i++;
256 VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
257 i * sizeof (blkptr_t), -1ULL, tx));
258 }
259 if (err || !bpo->bpo_havesubobj || bpo->bpo_phys->bpo_subobjs == 0)
260 goto out;
261
262 ASSERT(bpo->bpo_havecomp);
263 err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
264 if (err) {
265 mutex_exit(&bpo->bpo_lock);
266 return (err);
267 }
268 epb = doi.doi_data_block_size / sizeof (uint64_t);
269
270 for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
271 uint64_t *objarray;
272 uint64_t offset, blkoff;
273 bpobj_t sublist;
274 uint64_t used_before, comp_before, uncomp_before;
275 uint64_t used_after, comp_after, uncomp_after;
276
277 offset = i * sizeof (uint64_t);
278 blkoff = P2PHASE(i, epb);
279
280 if (dbuf == NULL || dbuf->db_offset > offset) {
281 if (dbuf)
282 dmu_buf_rele(dbuf, FTAG);
283 err = dmu_buf_hold(bpo->bpo_os,
284 bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
285 if (err)
286 break;
287 }
288
289 ASSERT3U(offset, >=, dbuf->db_offset);
290 ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
291
292 objarray = dbuf->db_data;
293 err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
294 if (err)
295 break;
296 if (free) {
297 err = bpobj_space(&sublist,
298 &used_before, &comp_before, &uncomp_before);
299 if (err)
300 break;
301 }
302 err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
303 if (free) {
304 VERIFY3U(0, ==, bpobj_space(&sublist,
305 &used_after, &comp_after, &uncomp_after));
306 bpo->bpo_phys->bpo_bytes -= used_before - used_after;
307 ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
308 bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
309 bpo->bpo_phys->bpo_uncomp -=
310 uncomp_before - uncomp_after;
311 }
312
313 bpobj_close(&sublist);
314 if (err)
315 break;
316 if (free) {
317 err = dmu_object_free(bpo->bpo_os,
318 objarray[blkoff], tx);
319 if (err)
320 break;
321 bpo->bpo_phys->bpo_num_subobjs--;
322 ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
323 }
324 }
325 if (dbuf) {
326 dmu_buf_rele(dbuf, FTAG);
327 dbuf = NULL;
328 }
329 if (free) {
330 VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
331 bpo->bpo_phys->bpo_subobjs,
332 (i + 1) * sizeof (uint64_t), -1ULL, tx));
333 }
334
335 out:
336 /* If there are no entries, there should be no bytes. */
337 ASSERT(bpo->bpo_phys->bpo_num_blkptrs > 0 ||
338 (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs > 0) ||
339 bpo->bpo_phys->bpo_bytes == 0);
340
341 mutex_exit(&bpo->bpo_lock);
342 return (err);
343 }
344
345 /*
346 * Iterate and remove the entries. If func returns nonzero, iteration
347 * will stop and that entry will not be removed.
348 */
349 int
350 bpobj_iterate(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
351 {
352 return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
353 }
354
355 /*
356 * Iterate the entries. If func returns nonzero, iteration will stop.
357 */
358 int
359 bpobj_iterate_nofree(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
360 {
361 return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
362 }
363
364 void
365 bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
366 {
367 bpobj_t subbpo;
368 uint64_t used, comp, uncomp, subsubobjs;
369
370 ASSERT(bpo->bpo_havesubobj);
371 ASSERT(bpo->bpo_havecomp);
372 ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
373
374 if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
375 bpobj_decr_empty(bpo->bpo_os, tx);
376 return;
377 }
378
379 VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
380 VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
381
382 if (used == 0) {
383 /* No point in having an empty subobj. */
384 bpobj_close(&subbpo);
385 bpobj_free(bpo->bpo_os, subobj, tx);
386 return;
387 }
388
389 dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
390 if (bpo->bpo_phys->bpo_subobjs == 0) {
391 bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
392 DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
393 }
394
395 dmu_object_info_t doi;
396 ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
397 ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
398
399 mutex_enter(&bpo->bpo_lock);
400 dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
401 bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
402 sizeof (subobj), &subobj, tx);
403 bpo->bpo_phys->bpo_num_subobjs++;
404
405 /*
406 * If subobj has only one block of subobjs, then move subobj's
407 * subobjs to bpo's subobj list directly. This reduces
408 * recursion in bpobj_iterate due to nested subobjs.
409 */
410 subsubobjs = subbpo.bpo_phys->bpo_subobjs;
411 if (subsubobjs != 0) {
412 dmu_object_info_t doi;
413
414 VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
415 if (doi.doi_max_offset == doi.doi_data_block_size) {
416 dmu_buf_t *subdb;
417 uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
418
419 VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
420 0, FTAG, &subdb, 0));
421 /*
422 * Make sure that we are not asking dmu_write()
423 * to write more data than we have in our buffer.
424 */
425 VERIFY3U(subdb->db_size, >=,
426 numsubsub * sizeof (subobj));
427 dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
428 bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
429 numsubsub * sizeof (subobj), subdb->db_data, tx);
430 dmu_buf_rele(subdb, FTAG);
431 bpo->bpo_phys->bpo_num_subobjs += numsubsub;
432
433 dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
434 subbpo.bpo_phys->bpo_subobjs = 0;
435 VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
436 subsubobjs, tx));
437 }
438 }
439 bpo->bpo_phys->bpo_bytes += used;
440 bpo->bpo_phys->bpo_comp += comp;
441 bpo->bpo_phys->bpo_uncomp += uncomp;
442 mutex_exit(&bpo->bpo_lock);
443
444 bpobj_close(&subbpo);
445 }
446
447 void
448 bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
449 {
450 blkptr_t stored_bp = *bp;
451 uint64_t offset;
452 int blkoff;
453 blkptr_t *bparray;
454
455 ASSERT(!BP_IS_HOLE(bp));
456 ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
457
458 /* We never need the fill count. */
459 stored_bp.blk_fill = 0;
460
461 /* The bpobj will compress better if we can leave off the checksum */
462 if (!BP_GET_DEDUP(bp))
463 bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
464
465 mutex_enter(&bpo->bpo_lock);
466
467 offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
468 blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
469
470 if (bpo->bpo_cached_dbuf == NULL ||
471 offset < bpo->bpo_cached_dbuf->db_offset ||
472 offset >= bpo->bpo_cached_dbuf->db_offset +
473 bpo->bpo_cached_dbuf->db_size) {
474 if (bpo->bpo_cached_dbuf)
475 dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
476 VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
477 offset, bpo, &bpo->bpo_cached_dbuf, 0));
478 }
479
480 dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
481 bparray = bpo->bpo_cached_dbuf->db_data;
482 bparray[blkoff] = stored_bp;
483
484 dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
485 bpo->bpo_phys->bpo_num_blkptrs++;
486 bpo->bpo_phys->bpo_bytes +=
487 bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
488 if (bpo->bpo_havecomp) {
489 bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
490 bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
491 }
492 mutex_exit(&bpo->bpo_lock);
493 }
494
495 struct space_range_arg {
496 spa_t *spa;
497 uint64_t mintxg;
498 uint64_t maxtxg;
499 uint64_t used;
500 uint64_t comp;
501 uint64_t uncomp;
502 };
503
504 /* ARGSUSED */
505 static int
506 space_range_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
507 {
508 struct space_range_arg *sra = arg;
509
510 if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
511 if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
512 sra->used += bp_get_dsize_sync(sra->spa, bp);
513 else
514 sra->used += bp_get_dsize(sra->spa, bp);
515 sra->comp += BP_GET_PSIZE(bp);
516 sra->uncomp += BP_GET_UCSIZE(bp);
517 }
518 return (0);
519 }
520
521 int
522 bpobj_space(bpobj_t *bpo, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
523 {
524 mutex_enter(&bpo->bpo_lock);
525
526 *usedp = bpo->bpo_phys->bpo_bytes;
527 if (bpo->bpo_havecomp) {
528 *compp = bpo->bpo_phys->bpo_comp;
529 *uncompp = bpo->bpo_phys->bpo_uncomp;
530 mutex_exit(&bpo->bpo_lock);
531 return (0);
532 } else {
533 mutex_exit(&bpo->bpo_lock);
534 return (bpobj_space_range(bpo, 0, UINT64_MAX,
535 usedp, compp, uncompp));
536 }
537 }
538
539 /*
540 * Return the amount of space in the bpobj which is:
541 * mintxg < blk_birth <= maxtxg
542 */
543 int
544 bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
545 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
546 {
547 struct space_range_arg sra = { 0 };
548 int err;
549
550 /*
551 * As an optimization, if they want the whole txg range, just
552 * get bpo_bytes rather than iterating over the bps.
553 */
554 if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
555 return (bpobj_space(bpo, usedp, compp, uncompp));
556
557 sra.spa = dmu_objset_spa(bpo->bpo_os);
558 sra.mintxg = mintxg;
559 sra.maxtxg = maxtxg;
560
561 err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
562 *usedp = sra.used;
563 *compp = sra.comp;
564 *uncompp = sra.uncomp;
565 return (err);
566 }