xref: /illumos-gate/usr/src/uts/common/fs/zfs/bpobj.c (revision eb633035)
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) 2011, 2016 by Delphix. All rights reserved.
24  * Copyright (c) 2014 Integros [integros.com]
25  * Copyright (c) 2017 Datto Inc.
26  */
27 
28 #include <sys/bpobj.h>
29 #include <sys/zfs_context.h>
30 #include <sys/refcount.h>
31 #include <sys/dsl_pool.h>
32 #include <sys/zfeature.h>
33 #include <sys/zap.h>
34 
35 /*
36  * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
37  */
38 uint64_t
bpobj_alloc_empty(objset_t * os,int blocksize,dmu_tx_t * tx)39 bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx)
40 {
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, SPA_FEATURE_EMPTY_BPOBJ)) {
45 		if (!spa_feature_is_active(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
46 			ASSERT0(dp->dp_empty_bpobj);
47 			dp->dp_empty_bpobj =
48 			    bpobj_alloc(os, SPA_OLD_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, SPA_FEATURE_EMPTY_BPOBJ, 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
bpobj_decr_empty(objset_t * os,dmu_tx_t * tx)63 bpobj_decr_empty(objset_t *os, dmu_tx_t *tx)
64 {
65 	dsl_pool_t *dp = dmu_objset_pool(os);
66 
67 	spa_feature_decr(dmu_objset_spa(os), SPA_FEATURE_EMPTY_BPOBJ, tx);
68 	if (!spa_feature_is_active(dmu_objset_spa(os),
69 	    SPA_FEATURE_EMPTY_BPOBJ)) {
70 		VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
71 		    DMU_POOL_DIRECTORY_OBJECT,
72 		    DMU_POOL_EMPTY_BPOBJ, tx));
73 		VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
74 		dp->dp_empty_bpobj = 0;
75 	}
76 }
77 
78 uint64_t
bpobj_alloc(objset_t * os,int blocksize,dmu_tx_t * tx)79 bpobj_alloc(objset_t *os, int blocksize, dmu_tx_t *tx)
80 {
81 	int size;
82 
83 	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
84 		size = BPOBJ_SIZE_V0;
85 	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
86 		size = BPOBJ_SIZE_V1;
87 	else
88 		size = sizeof (bpobj_phys_t);
89 
90 	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
91 	    DMU_OT_BPOBJ_HDR, size, tx));
92 }
93 
94 void
bpobj_free(objset_t * os,uint64_t obj,dmu_tx_t * tx)95 bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx)
96 {
97 	int64_t i;
98 	bpobj_t bpo;
99 	dmu_object_info_t doi;
100 	int epb;
101 	dmu_buf_t *dbuf = NULL;
102 
103 	ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj);
104 	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
105 
106 	mutex_enter(&bpo.bpo_lock);
107 
108 	if (!bpo.bpo_havesubobj || bpo.bpo_phys->bpo_subobjs == 0)
109 		goto out;
110 
111 	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
112 	epb = doi.doi_data_block_size / sizeof (uint64_t);
113 
114 	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
115 		uint64_t *objarray;
116 		uint64_t offset, blkoff;
117 
118 		offset = i * sizeof (uint64_t);
119 		blkoff = P2PHASE(i, epb);
120 
121 		if (dbuf == NULL || dbuf->db_offset > offset) {
122 			if (dbuf)
123 				dmu_buf_rele(dbuf, FTAG);
124 			VERIFY3U(0, ==, dmu_buf_hold(os,
125 			    bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
126 		}
127 
128 		ASSERT3U(offset, >=, dbuf->db_offset);
129 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
130 
131 		objarray = dbuf->db_data;
132 		bpobj_free(os, objarray[blkoff], tx);
133 	}
134 	if (dbuf) {
135 		dmu_buf_rele(dbuf, FTAG);
136 		dbuf = NULL;
137 	}
138 	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
139 
140 out:
141 	mutex_exit(&bpo.bpo_lock);
142 	bpobj_close(&bpo);
143 
144 	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
145 }
146 
147 int
bpobj_open(bpobj_t * bpo,objset_t * os,uint64_t object)148 bpobj_open(bpobj_t *bpo, objset_t *os, uint64_t object)
149 {
150 	dmu_object_info_t doi;
151 	int err;
152 
153 	err = dmu_object_info(os, object, &doi);
154 	if (err)
155 		return (err);
156 
157 	bzero(bpo, sizeof (*bpo));
158 	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
159 
160 	ASSERT(bpo->bpo_dbuf == NULL);
161 	ASSERT(bpo->bpo_phys == NULL);
162 	ASSERT(object != 0);
163 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
164 	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
165 
166 	err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
167 	if (err)
168 		return (err);
169 
170 	bpo->bpo_os = os;
171 	bpo->bpo_object = object;
172 	bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
173 	bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
174 	bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
175 	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
176 	return (0);
177 }
178 
179 boolean_t
bpobj_is_open(const bpobj_t * bpo)180 bpobj_is_open(const bpobj_t *bpo)
181 {
182 	return (bpo->bpo_object != 0);
183 }
184 
185 void
bpobj_close(bpobj_t * bpo)186 bpobj_close(bpobj_t *bpo)
187 {
188 	/* Lame workaround for closing a bpobj that was never opened. */
189 	if (bpo->bpo_object == 0)
190 		return;
191 
192 	dmu_buf_rele(bpo->bpo_dbuf, bpo);
193 	if (bpo->bpo_cached_dbuf != NULL)
194 		dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
195 	bpo->bpo_dbuf = NULL;
196 	bpo->bpo_phys = NULL;
197 	bpo->bpo_cached_dbuf = NULL;
198 	bpo->bpo_object = 0;
199 
200 	mutex_destroy(&bpo->bpo_lock);
201 }
202 
203 boolean_t
bpobj_is_empty(bpobj_t * bpo)204 bpobj_is_empty(bpobj_t *bpo)
205 {
206 	return (bpo->bpo_phys->bpo_num_blkptrs == 0 &&
207 	    (!bpo->bpo_havesubobj || bpo->bpo_phys->bpo_num_subobjs == 0));
208 }
209 
210 static int
bpobj_iterate_impl(bpobj_t * bpo,bpobj_itor_t func,void * arg,dmu_tx_t * tx,boolean_t free)211 bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
212     boolean_t free)
213 {
214 	dmu_object_info_t doi;
215 	int epb;
216 	int64_t i;
217 	int err = 0;
218 	dmu_buf_t *dbuf = NULL;
219 
220 	ASSERT(bpobj_is_open(bpo));
221 	mutex_enter(&bpo->bpo_lock);
222 
223 	if (free)
224 		dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
225 
226 	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
227 		blkptr_t *bparray;
228 		blkptr_t *bp;
229 		uint64_t offset, blkoff;
230 
231 		offset = i * sizeof (blkptr_t);
232 		blkoff = P2PHASE(i, bpo->bpo_epb);
233 
234 		if (dbuf == NULL || dbuf->db_offset > offset) {
235 			if (dbuf)
236 				dmu_buf_rele(dbuf, FTAG);
237 			err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
238 			    FTAG, &dbuf, 0);
239 			if (err)
240 				break;
241 		}
242 
243 		ASSERT3U(offset, >=, dbuf->db_offset);
244 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
245 
246 		bparray = dbuf->db_data;
247 		bp = &bparray[blkoff];
248 		err = func(arg, bp, tx);
249 		if (err)
250 			break;
251 		if (free) {
252 			bpo->bpo_phys->bpo_bytes -=
253 			    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
254 			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
255 			if (bpo->bpo_havecomp) {
256 				bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
257 				bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
258 			}
259 			bpo->bpo_phys->bpo_num_blkptrs--;
260 			ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
261 		}
262 	}
263 	if (dbuf) {
264 		dmu_buf_rele(dbuf, FTAG);
265 		dbuf = NULL;
266 	}
267 	if (free) {
268 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
269 		    (i + 1) * sizeof (blkptr_t), DMU_OBJECT_END, tx));
270 	}
271 	if (err || !bpo->bpo_havesubobj || bpo->bpo_phys->bpo_subobjs == 0)
272 		goto out;
273 
274 	ASSERT(bpo->bpo_havecomp);
275 	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
276 	if (err) {
277 		mutex_exit(&bpo->bpo_lock);
278 		return (err);
279 	}
280 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
281 	epb = doi.doi_data_block_size / sizeof (uint64_t);
282 
283 	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
284 		uint64_t *objarray;
285 		uint64_t offset, blkoff;
286 		bpobj_t sublist;
287 		uint64_t used_before, comp_before, uncomp_before;
288 		uint64_t used_after, comp_after, uncomp_after;
289 
290 		offset = i * sizeof (uint64_t);
291 		blkoff = P2PHASE(i, epb);
292 
293 		if (dbuf == NULL || dbuf->db_offset > offset) {
294 			if (dbuf)
295 				dmu_buf_rele(dbuf, FTAG);
296 			err = dmu_buf_hold(bpo->bpo_os,
297 			    bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
298 			if (err)
299 				break;
300 		}
301 
302 		ASSERT3U(offset, >=, dbuf->db_offset);
303 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
304 
305 		objarray = dbuf->db_data;
306 		err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
307 		if (err)
308 			break;
309 		if (free) {
310 			err = bpobj_space(&sublist,
311 			    &used_before, &comp_before, &uncomp_before);
312 			if (err != 0) {
313 				bpobj_close(&sublist);
314 				break;
315 			}
316 		}
317 		err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
318 		if (free) {
319 			VERIFY3U(0, ==, bpobj_space(&sublist,
320 			    &used_after, &comp_after, &uncomp_after));
321 			bpo->bpo_phys->bpo_bytes -= used_before - used_after;
322 			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
323 			bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
324 			bpo->bpo_phys->bpo_uncomp -=
325 			    uncomp_before - uncomp_after;
326 		}
327 
328 		bpobj_close(&sublist);
329 		if (err)
330 			break;
331 		if (free) {
332 			err = dmu_object_free(bpo->bpo_os,
333 			    objarray[blkoff], tx);
334 			if (err)
335 				break;
336 			bpo->bpo_phys->bpo_num_subobjs--;
337 			ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
338 		}
339 	}
340 	if (dbuf) {
341 		dmu_buf_rele(dbuf, FTAG);
342 		dbuf = NULL;
343 	}
344 	if (free) {
345 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
346 		    bpo->bpo_phys->bpo_subobjs,
347 		    (i + 1) * sizeof (uint64_t), DMU_OBJECT_END, tx));
348 	}
349 
350 out:
351 	/* If there are no entries, there should be no bytes. */
352 	if (bpobj_is_empty(bpo)) {
353 		ASSERT0(bpo->bpo_phys->bpo_bytes);
354 		ASSERT0(bpo->bpo_phys->bpo_comp);
355 		ASSERT0(bpo->bpo_phys->bpo_uncomp);
356 	}
357 
358 	mutex_exit(&bpo->bpo_lock);
359 	return (err);
360 }
361 
362 /*
363  * Iterate and remove the entries.  If func returns nonzero, iteration
364  * will stop and that entry will not be removed.
365  */
366 int
bpobj_iterate(bpobj_t * bpo,bpobj_itor_t func,void * arg,dmu_tx_t * tx)367 bpobj_iterate(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
368 {
369 	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
370 }
371 
372 /*
373  * Iterate the entries.  If func returns nonzero, iteration will stop.
374  */
375 int
bpobj_iterate_nofree(bpobj_t * bpo,bpobj_itor_t func,void * arg,dmu_tx_t * tx)376 bpobj_iterate_nofree(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
377 {
378 	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
379 }
380 
381 void
bpobj_enqueue_subobj(bpobj_t * bpo,uint64_t subobj,dmu_tx_t * tx)382 bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
383 {
384 	bpobj_t subbpo;
385 	uint64_t used, comp, uncomp, subsubobjs;
386 
387 	ASSERT(bpobj_is_open(bpo));
388 	ASSERT(subobj != 0);
389 	ASSERT(bpo->bpo_havesubobj);
390 	ASSERT(bpo->bpo_havecomp);
391 	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
392 
393 	if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
394 		bpobj_decr_empty(bpo->bpo_os, tx);
395 		return;
396 	}
397 
398 	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
399 	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
400 
401 	if (bpobj_is_empty(&subbpo)) {
402 		/* No point in having an empty subobj. */
403 		bpobj_close(&subbpo);
404 		bpobj_free(bpo->bpo_os, subobj, tx);
405 		return;
406 	}
407 
408 	mutex_enter(&bpo->bpo_lock);
409 	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
410 	if (bpo->bpo_phys->bpo_subobjs == 0) {
411 		bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
412 		    DMU_OT_BPOBJ_SUBOBJ, SPA_OLD_MAXBLOCKSIZE,
413 		    DMU_OT_NONE, 0, tx);
414 	}
415 
416 	dmu_object_info_t doi;
417 	ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
418 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
419 
420 	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
421 	    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
422 	    sizeof (subobj), &subobj, tx);
423 	bpo->bpo_phys->bpo_num_subobjs++;
424 
425 	/*
426 	 * If subobj has only one block of subobjs, then move subobj's
427 	 * subobjs to bpo's subobj list directly.  This reduces
428 	 * recursion in bpobj_iterate due to nested subobjs.
429 	 */
430 	subsubobjs = subbpo.bpo_phys->bpo_subobjs;
431 	if (subsubobjs != 0) {
432 		dmu_object_info_t doi;
433 
434 		VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
435 		if (doi.doi_max_offset == doi.doi_data_block_size) {
436 			dmu_buf_t *subdb;
437 			uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
438 
439 			VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
440 			    0, FTAG, &subdb, 0));
441 			/*
442 			 * Make sure that we are not asking dmu_write()
443 			 * to write more data than we have in our buffer.
444 			 */
445 			VERIFY3U(subdb->db_size, >=,
446 			    numsubsub * sizeof (subobj));
447 			dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
448 			    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
449 			    numsubsub * sizeof (subobj), subdb->db_data, tx);
450 			dmu_buf_rele(subdb, FTAG);
451 			bpo->bpo_phys->bpo_num_subobjs += numsubsub;
452 
453 			dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
454 			subbpo.bpo_phys->bpo_subobjs = 0;
455 			VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
456 			    subsubobjs, tx));
457 		}
458 	}
459 	bpo->bpo_phys->bpo_bytes += used;
460 	bpo->bpo_phys->bpo_comp += comp;
461 	bpo->bpo_phys->bpo_uncomp += uncomp;
462 	mutex_exit(&bpo->bpo_lock);
463 
464 	bpobj_close(&subbpo);
465 }
466 
467 void
bpobj_enqueue(bpobj_t * bpo,const blkptr_t * bp,dmu_tx_t * tx)468 bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
469 {
470 	blkptr_t stored_bp = *bp;
471 	uint64_t offset;
472 	int blkoff;
473 	blkptr_t *bparray;
474 
475 	ASSERT(bpobj_is_open(bpo));
476 	ASSERT(!BP_IS_HOLE(bp));
477 	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
478 
479 	if (BP_IS_EMBEDDED(bp)) {
480 		/*
481 		 * The bpobj will compress better without the payload.
482 		 *
483 		 * Note that we store EMBEDDED bp's because they have an
484 		 * uncompressed size, which must be accounted for.  An
485 		 * alternative would be to add their size to bpo_uncomp
486 		 * without storing the bp, but that would create additional
487 		 * complications: bpo_uncomp would be inconsistent with the
488 		 * set of BP's stored, and bpobj_iterate() wouldn't visit
489 		 * all the space accounted for in the bpobj.
490 		 */
491 		bzero(&stored_bp, sizeof (stored_bp));
492 		stored_bp.blk_prop = bp->blk_prop;
493 		stored_bp.blk_birth = bp->blk_birth;
494 	} else if (!BP_GET_DEDUP(bp)) {
495 		/* The bpobj will compress better without the checksum */
496 		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
497 	}
498 
499 	/* We never need the fill count. */
500 	stored_bp.blk_fill = 0;
501 
502 	mutex_enter(&bpo->bpo_lock);
503 
504 	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
505 	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
506 
507 	if (bpo->bpo_cached_dbuf == NULL ||
508 	    offset < bpo->bpo_cached_dbuf->db_offset ||
509 	    offset >= bpo->bpo_cached_dbuf->db_offset +
510 	    bpo->bpo_cached_dbuf->db_size) {
511 		if (bpo->bpo_cached_dbuf)
512 			dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
513 		VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
514 		    offset, bpo, &bpo->bpo_cached_dbuf, 0));
515 	}
516 
517 	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
518 	bparray = bpo->bpo_cached_dbuf->db_data;
519 	bparray[blkoff] = stored_bp;
520 
521 	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
522 	bpo->bpo_phys->bpo_num_blkptrs++;
523 	bpo->bpo_phys->bpo_bytes +=
524 	    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
525 	if (bpo->bpo_havecomp) {
526 		bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
527 		bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
528 	}
529 	mutex_exit(&bpo->bpo_lock);
530 }
531 
532 struct space_range_arg {
533 	spa_t *spa;
534 	uint64_t mintxg;
535 	uint64_t maxtxg;
536 	uint64_t used;
537 	uint64_t comp;
538 	uint64_t uncomp;
539 };
540 
541 /* ARGSUSED */
542 static int
space_range_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)543 space_range_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
544 {
545 	struct space_range_arg *sra = arg;
546 
547 	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
548 		if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
549 			sra->used += bp_get_dsize_sync(sra->spa, bp);
550 		else
551 			sra->used += bp_get_dsize(sra->spa, bp);
552 		sra->comp += BP_GET_PSIZE(bp);
553 		sra->uncomp += BP_GET_UCSIZE(bp);
554 	}
555 	return (0);
556 }
557 
558 int
bpobj_space(bpobj_t * bpo,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)559 bpobj_space(bpobj_t *bpo, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
560 {
561 	ASSERT(bpobj_is_open(bpo));
562 	mutex_enter(&bpo->bpo_lock);
563 
564 	*usedp = bpo->bpo_phys->bpo_bytes;
565 	if (bpo->bpo_havecomp) {
566 		*compp = bpo->bpo_phys->bpo_comp;
567 		*uncompp = bpo->bpo_phys->bpo_uncomp;
568 		mutex_exit(&bpo->bpo_lock);
569 		return (0);
570 	} else {
571 		mutex_exit(&bpo->bpo_lock);
572 		return (bpobj_space_range(bpo, 0, UINT64_MAX,
573 		    usedp, compp, uncompp));
574 	}
575 }
576 
577 /*
578  * Return the amount of space in the bpobj which is:
579  * mintxg < blk_birth <= maxtxg
580  */
581 int
bpobj_space_range(bpobj_t * bpo,uint64_t mintxg,uint64_t maxtxg,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)582 bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
583     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
584 {
585 	struct space_range_arg sra = { 0 };
586 	int err;
587 
588 	ASSERT(bpobj_is_open(bpo));
589 
590 	/*
591 	 * As an optimization, if they want the whole txg range, just
592 	 * get bpo_bytes rather than iterating over the bps.
593 	 */
594 	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
595 		return (bpobj_space(bpo, usedp, compp, uncompp));
596 
597 	sra.spa = dmu_objset_spa(bpo->bpo_os);
598 	sra.mintxg = mintxg;
599 	sra.maxtxg = maxtxg;
600 
601 	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
602 	*usedp = sra.used;
603 	*compp = sra.comp;
604 	*uncompp = sra.uncomp;
605 	return (err);
606 }
607