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, 2014 by Delphix. All rights reserved.
24  * Copyright (c) 2014 Integros [integros.com]
25  */
26 
27 #include <sys/bpobj.h>
28 #include <sys/zfs_context.h>
29 #include <sys/refcount.h>
30 #include <sys/dsl_pool.h>
31 #include <sys/zfeature.h>
32 #include <sys/zap.h>
33 
34 /*
35  * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
36  */
37 uint64_t
bpobj_alloc_empty(objset_t * os,int blocksize,dmu_tx_t * tx)38 bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx)
39 {
40 	spa_t *spa = dmu_objset_spa(os);
41 	dsl_pool_t *dp = dmu_objset_pool(os);
42 
43 	if (spa_feature_is_enabled(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
44 		if (!spa_feature_is_active(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
45 			ASSERT0(dp->dp_empty_bpobj);
46 			dp->dp_empty_bpobj =
47 			    bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx);
48 			VERIFY(zap_add(os,
49 			    DMU_POOL_DIRECTORY_OBJECT,
50 			    DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
51 			    &dp->dp_empty_bpobj, tx) == 0);
52 		}
53 		spa_feature_incr(spa, SPA_FEATURE_EMPTY_BPOBJ, tx);
54 		ASSERT(dp->dp_empty_bpobj != 0);
55 		return (dp->dp_empty_bpobj);
56 	} else {
57 		return (bpobj_alloc(os, blocksize, tx));
58 	}
59 }
60 
61 void
bpobj_decr_empty(objset_t * os,dmu_tx_t * tx)62 bpobj_decr_empty(objset_t *os, dmu_tx_t *tx)
63 {
64 	dsl_pool_t *dp = dmu_objset_pool(os);
65 
66 	spa_feature_decr(dmu_objset_spa(os), SPA_FEATURE_EMPTY_BPOBJ, tx);
67 	if (!spa_feature_is_active(dmu_objset_spa(os),
68 	    SPA_FEATURE_EMPTY_BPOBJ)) {
69 		VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
70 		    DMU_POOL_DIRECTORY_OBJECT,
71 		    DMU_POOL_EMPTY_BPOBJ, tx));
72 		VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
73 		dp->dp_empty_bpobj = 0;
74 	}
75 }
76 
77 uint64_t
bpobj_alloc(objset_t * os,int blocksize,dmu_tx_t * tx)78 bpobj_alloc(objset_t *os, int blocksize, dmu_tx_t *tx)
79 {
80 	int size;
81 
82 	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
83 		size = BPOBJ_SIZE_V0;
84 	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
85 		size = BPOBJ_SIZE_V1;
86 	else
87 		size = sizeof (bpobj_phys_t);
88 
89 	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
90 	    DMU_OT_BPOBJ_HDR, size, tx));
91 }
92 
93 void
bpobj_free(objset_t * os,uint64_t obj,dmu_tx_t * tx)94 bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx)
95 {
96 	int64_t i;
97 	bpobj_t bpo;
98 	dmu_object_info_t doi;
99 	int epb;
100 	dmu_buf_t *dbuf = NULL;
101 
102 	ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj);
103 	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
104 
105 	mutex_enter(&bpo.bpo_lock);
106 
107 	if (!bpo.bpo_havesubobj || bpo.bpo_phys->bpo_subobjs == 0)
108 		goto out;
109 
110 	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
111 	epb = doi.doi_data_block_size / sizeof (uint64_t);
112 
113 	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
114 		uint64_t *objarray;
115 		uint64_t offset, blkoff;
116 
117 		offset = i * sizeof (uint64_t);
118 		blkoff = P2PHASE(i, epb);
119 
120 		if (dbuf == NULL || dbuf->db_offset > offset) {
121 			if (dbuf)
122 				dmu_buf_rele(dbuf, FTAG);
123 			VERIFY3U(0, ==, dmu_buf_hold(os,
124 			    bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
125 		}
126 
127 		ASSERT3U(offset, >=, dbuf->db_offset);
128 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
129 
130 		objarray = dbuf->db_data;
131 		bpobj_free(os, objarray[blkoff], tx);
132 	}
133 	if (dbuf) {
134 		dmu_buf_rele(dbuf, FTAG);
135 		dbuf = NULL;
136 	}
137 	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
138 
139 out:
140 	mutex_exit(&bpo.bpo_lock);
141 	bpobj_close(&bpo);
142 
143 	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
144 }
145 
146 int
bpobj_open(bpobj_t * bpo,objset_t * os,uint64_t object)147 bpobj_open(bpobj_t *bpo, objset_t *os, uint64_t object)
148 {
149 	dmu_object_info_t doi;
150 	int err;
151 
152 	err = dmu_object_info(os, object, &doi);
153 	if (err)
154 		return (err);
155 
156 	bzero(bpo, sizeof (*bpo));
157 	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
158 
159 	ASSERT(bpo->bpo_dbuf == NULL);
160 	ASSERT(bpo->bpo_phys == NULL);
161 	ASSERT(object != 0);
162 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
163 	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
164 
165 	err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
166 	if (err)
167 		return (err);
168 
169 	bpo->bpo_os = os;
170 	bpo->bpo_object = object;
171 	bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
172 	bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
173 	bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
174 	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
175 	return (0);
176 }
177 
178 void
bpobj_close(bpobj_t * bpo)179 bpobj_close(bpobj_t *bpo)
180 {
181 	/* Lame workaround for closing a bpobj that was never opened. */
182 	if (bpo->bpo_object == 0)
183 		return;
184 
185 	dmu_buf_rele(bpo->bpo_dbuf, bpo);
186 	if (bpo->bpo_cached_dbuf != NULL)
187 		dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
188 	bpo->bpo_dbuf = NULL;
189 	bpo->bpo_phys = NULL;
190 	bpo->bpo_cached_dbuf = NULL;
191 	bpo->bpo_object = 0;
192 
193 	mutex_destroy(&bpo->bpo_lock);
194 }
195 
196 static boolean_t
bpobj_hasentries(bpobj_t * bpo)197 bpobj_hasentries(bpobj_t *bpo)
198 {
199 	return (bpo->bpo_phys->bpo_num_blkptrs != 0 ||
200 	    (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs != 0));
201 }
202 
203 static int
bpobj_iterate_impl(bpobj_t * bpo,bpobj_itor_t func,void * arg,dmu_tx_t * tx,boolean_t free)204 bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
205     boolean_t free)
206 {
207 	dmu_object_info_t doi;
208 	int epb;
209 	int64_t i;
210 	int err = 0;
211 	dmu_buf_t *dbuf = NULL;
212 
213 	mutex_enter(&bpo->bpo_lock);
214 
215 	if (free)
216 		dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
217 
218 	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
219 		blkptr_t *bparray;
220 		blkptr_t *bp;
221 		uint64_t offset, blkoff;
222 
223 		offset = i * sizeof (blkptr_t);
224 		blkoff = P2PHASE(i, bpo->bpo_epb);
225 
226 		if (dbuf == NULL || dbuf->db_offset > offset) {
227 			if (dbuf)
228 				dmu_buf_rele(dbuf, FTAG);
229 			err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
230 			    FTAG, &dbuf, 0);
231 			if (err)
232 				break;
233 		}
234 
235 		ASSERT3U(offset, >=, dbuf->db_offset);
236 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
237 
238 		bparray = dbuf->db_data;
239 		bp = &bparray[blkoff];
240 		err = func(arg, bp, tx);
241 		if (err)
242 			break;
243 		if (free) {
244 			bpo->bpo_phys->bpo_bytes -=
245 			    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
246 			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
247 			if (bpo->bpo_havecomp) {
248 				bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
249 				bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
250 			}
251 			bpo->bpo_phys->bpo_num_blkptrs--;
252 			ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
253 		}
254 	}
255 	if (dbuf) {
256 		dmu_buf_rele(dbuf, FTAG);
257 		dbuf = NULL;
258 	}
259 	if (free) {
260 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
261 		    (i + 1) * sizeof (blkptr_t), -1ULL, tx));
262 	}
263 	if (err || !bpo->bpo_havesubobj || bpo->bpo_phys->bpo_subobjs == 0)
264 		goto out;
265 
266 	ASSERT(bpo->bpo_havecomp);
267 	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
268 	if (err) {
269 		mutex_exit(&bpo->bpo_lock);
270 		return (err);
271 	}
272 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
273 	epb = doi.doi_data_block_size / sizeof (uint64_t);
274 
275 	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
276 		uint64_t *objarray;
277 		uint64_t offset, blkoff;
278 		bpobj_t sublist;
279 		uint64_t used_before, comp_before, uncomp_before;
280 		uint64_t used_after, comp_after, uncomp_after;
281 
282 		offset = i * sizeof (uint64_t);
283 		blkoff = P2PHASE(i, epb);
284 
285 		if (dbuf == NULL || dbuf->db_offset > offset) {
286 			if (dbuf)
287 				dmu_buf_rele(dbuf, FTAG);
288 			err = dmu_buf_hold(bpo->bpo_os,
289 			    bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
290 			if (err)
291 				break;
292 		}
293 
294 		ASSERT3U(offset, >=, dbuf->db_offset);
295 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
296 
297 		objarray = dbuf->db_data;
298 		err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
299 		if (err)
300 			break;
301 		if (free) {
302 			err = bpobj_space(&sublist,
303 			    &used_before, &comp_before, &uncomp_before);
304 			if (err != 0) {
305 				bpobj_close(&sublist);
306 				break;
307 			}
308 		}
309 		err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
310 		if (free) {
311 			VERIFY3U(0, ==, bpobj_space(&sublist,
312 			    &used_after, &comp_after, &uncomp_after));
313 			bpo->bpo_phys->bpo_bytes -= used_before - used_after;
314 			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
315 			bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
316 			bpo->bpo_phys->bpo_uncomp -=
317 			    uncomp_before - uncomp_after;
318 		}
319 
320 		bpobj_close(&sublist);
321 		if (err)
322 			break;
323 		if (free) {
324 			err = dmu_object_free(bpo->bpo_os,
325 			    objarray[blkoff], tx);
326 			if (err)
327 				break;
328 			bpo->bpo_phys->bpo_num_subobjs--;
329 			ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
330 		}
331 	}
332 	if (dbuf) {
333 		dmu_buf_rele(dbuf, FTAG);
334 		dbuf = NULL;
335 	}
336 	if (free) {
337 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
338 		    bpo->bpo_phys->bpo_subobjs,
339 		    (i + 1) * sizeof (uint64_t), -1ULL, tx));
340 	}
341 
342 out:
343 	/* If there are no entries, there should be no bytes. */
344 	if (!bpobj_hasentries(bpo)) {
345 		ASSERT0(bpo->bpo_phys->bpo_bytes);
346 		ASSERT0(bpo->bpo_phys->bpo_comp);
347 		ASSERT0(bpo->bpo_phys->bpo_uncomp);
348 	}
349 
350 	mutex_exit(&bpo->bpo_lock);
351 	return (err);
352 }
353 
354 /*
355  * Iterate and remove the entries.  If func returns nonzero, iteration
356  * will stop and that entry will not be removed.
357  */
358 int
bpobj_iterate(bpobj_t * bpo,bpobj_itor_t func,void * arg,dmu_tx_t * tx)359 bpobj_iterate(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
360 {
361 	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
362 }
363 
364 /*
365  * Iterate the entries.  If func returns nonzero, iteration will stop.
366  */
367 int
bpobj_iterate_nofree(bpobj_t * bpo,bpobj_itor_t func,void * arg,dmu_tx_t * tx)368 bpobj_iterate_nofree(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
369 {
370 	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
371 }
372 
373 void
bpobj_enqueue_subobj(bpobj_t * bpo,uint64_t subobj,dmu_tx_t * tx)374 bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
375 {
376 	bpobj_t subbpo;
377 	uint64_t used, comp, uncomp, subsubobjs;
378 
379 	ASSERT(bpo->bpo_havesubobj);
380 	ASSERT(bpo->bpo_havecomp);
381 	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
382 
383 	if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
384 		bpobj_decr_empty(bpo->bpo_os, tx);
385 		return;
386 	}
387 
388 	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
389 	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
390 
391 	if (!bpobj_hasentries(&subbpo)) {
392 		/* No point in having an empty subobj. */
393 		bpobj_close(&subbpo);
394 		bpobj_free(bpo->bpo_os, subobj, tx);
395 		return;
396 	}
397 
398 	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
399 	if (bpo->bpo_phys->bpo_subobjs == 0) {
400 		bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
401 		    DMU_OT_BPOBJ_SUBOBJ, SPA_OLD_MAXBLOCKSIZE,
402 		    DMU_OT_NONE, 0, tx);
403 	}
404 
405 	dmu_object_info_t doi;
406 	ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
407 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
408 
409 	mutex_enter(&bpo->bpo_lock);
410 	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
411 	    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
412 	    sizeof (subobj), &subobj, tx);
413 	bpo->bpo_phys->bpo_num_subobjs++;
414 
415 	/*
416 	 * If subobj has only one block of subobjs, then move subobj's
417 	 * subobjs to bpo's subobj list directly.  This reduces
418 	 * recursion in bpobj_iterate due to nested subobjs.
419 	 */
420 	subsubobjs = subbpo.bpo_phys->bpo_subobjs;
421 	if (subsubobjs != 0) {
422 		dmu_object_info_t doi;
423 
424 		VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
425 		if (doi.doi_max_offset == doi.doi_data_block_size) {
426 			dmu_buf_t *subdb;
427 			uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
428 
429 			VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
430 			    0, FTAG, &subdb, 0));
431 			/*
432 			 * Make sure that we are not asking dmu_write()
433 			 * to write more data than we have in our buffer.
434 			 */
435 			VERIFY3U(subdb->db_size, >=,
436 			    numsubsub * sizeof (subobj));
437 			dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
438 			    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
439 			    numsubsub * sizeof (subobj), subdb->db_data, tx);
440 			dmu_buf_rele(subdb, FTAG);
441 			bpo->bpo_phys->bpo_num_subobjs += numsubsub;
442 
443 			dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
444 			subbpo.bpo_phys->bpo_subobjs = 0;
445 			VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
446 			    subsubobjs, tx));
447 		}
448 	}
449 	bpo->bpo_phys->bpo_bytes += used;
450 	bpo->bpo_phys->bpo_comp += comp;
451 	bpo->bpo_phys->bpo_uncomp += uncomp;
452 	mutex_exit(&bpo->bpo_lock);
453 
454 	bpobj_close(&subbpo);
455 }
456 
457 void
bpobj_enqueue(bpobj_t * bpo,const blkptr_t * bp,dmu_tx_t * tx)458 bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
459 {
460 	blkptr_t stored_bp = *bp;
461 	uint64_t offset;
462 	int blkoff;
463 	blkptr_t *bparray;
464 
465 	ASSERT(!BP_IS_HOLE(bp));
466 	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
467 
468 	if (BP_IS_EMBEDDED(bp)) {
469 		/*
470 		 * The bpobj will compress better without the payload.
471 		 *
472 		 * Note that we store EMBEDDED bp's because they have an
473 		 * uncompressed size, which must be accounted for.  An
474 		 * alternative would be to add their size to bpo_uncomp
475 		 * without storing the bp, but that would create additional
476 		 * complications: bpo_uncomp would be inconsistent with the
477 		 * set of BP's stored, and bpobj_iterate() wouldn't visit
478 		 * all the space accounted for in the bpobj.
479 		 */
480 		bzero(&stored_bp, sizeof (stored_bp));
481 		stored_bp.blk_prop = bp->blk_prop;
482 		stored_bp.blk_birth = bp->blk_birth;
483 	} else if (!BP_GET_DEDUP(bp)) {
484 		/* The bpobj will compress better without the checksum */
485 		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
486 	}
487 
488 	/* We never need the fill count. */
489 	stored_bp.blk_fill = 0;
490 
491 	mutex_enter(&bpo->bpo_lock);
492 
493 	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
494 	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
495 
496 	if (bpo->bpo_cached_dbuf == NULL ||
497 	    offset < bpo->bpo_cached_dbuf->db_offset ||
498 	    offset >= bpo->bpo_cached_dbuf->db_offset +
499 	    bpo->bpo_cached_dbuf->db_size) {
500 		if (bpo->bpo_cached_dbuf)
501 			dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
502 		VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
503 		    offset, bpo, &bpo->bpo_cached_dbuf, 0));
504 	}
505 
506 	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
507 	bparray = bpo->bpo_cached_dbuf->db_data;
508 	bparray[blkoff] = stored_bp;
509 
510 	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
511 	bpo->bpo_phys->bpo_num_blkptrs++;
512 	bpo->bpo_phys->bpo_bytes +=
513 	    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
514 	if (bpo->bpo_havecomp) {
515 		bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
516 		bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
517 	}
518 	mutex_exit(&bpo->bpo_lock);
519 }
520 
521 struct space_range_arg {
522 	spa_t *spa;
523 	uint64_t mintxg;
524 	uint64_t maxtxg;
525 	uint64_t used;
526 	uint64_t comp;
527 	uint64_t uncomp;
528 };
529 
530 /* ARGSUSED */
531 static int
space_range_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)532 space_range_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
533 {
534 	struct space_range_arg *sra = arg;
535 
536 	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
537 		if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
538 			sra->used += bp_get_dsize_sync(sra->spa, bp);
539 		else
540 			sra->used += bp_get_dsize(sra->spa, bp);
541 		sra->comp += BP_GET_PSIZE(bp);
542 		sra->uncomp += BP_GET_UCSIZE(bp);
543 	}
544 	return (0);
545 }
546 
547 int
bpobj_space(bpobj_t * bpo,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)548 bpobj_space(bpobj_t *bpo, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
549 {
550 	mutex_enter(&bpo->bpo_lock);
551 
552 	*usedp = bpo->bpo_phys->bpo_bytes;
553 	if (bpo->bpo_havecomp) {
554 		*compp = bpo->bpo_phys->bpo_comp;
555 		*uncompp = bpo->bpo_phys->bpo_uncomp;
556 		mutex_exit(&bpo->bpo_lock);
557 		return (0);
558 	} else {
559 		mutex_exit(&bpo->bpo_lock);
560 		return (bpobj_space_range(bpo, 0, UINT64_MAX,
561 		    usedp, compp, uncompp));
562 	}
563 }
564 
565 /*
566  * Return the amount of space in the bpobj which is:
567  * mintxg < blk_birth <= maxtxg
568  */
569 int
bpobj_space_range(bpobj_t * bpo,uint64_t mintxg,uint64_t maxtxg,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)570 bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
571     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
572 {
573 	struct space_range_arg sra = { 0 };
574 	int err;
575 
576 	/*
577 	 * As an optimization, if they want the whole txg range, just
578 	 * get bpo_bytes rather than iterating over the bps.
579 	 */
580 	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
581 		return (bpobj_space(bpo, usedp, compp, uncompp));
582 
583 	sra.spa = dmu_objset_spa(bpo->bpo_os);
584 	sra.mintxg = mintxg;
585 	sra.maxtxg = maxtxg;
586 
587 	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
588 	*usedp = sra.used;
589 	*compp = sra.comp;
590 	*uncompp = sra.uncomp;
591 	return (err);
592 }
593