xref: /titanic_41/usr/src/uts/common/fs/zfs/bpobj.c (revision a1bf6e2eba3dc9acc46299eac9c6ef1bc1e15c2b)
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  */
24 
25 #include <sys/bpobj.h>
26 #include <sys/zfs_context.h>
27 #include <sys/refcount.h>
28 
29 uint64_t
30 bpobj_alloc(objset_t *os, int blocksize, dmu_tx_t *tx)
31 {
32 	int size;
33 
34 	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
35 		size = BPOBJ_SIZE_V0;
36 	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
37 		size = BPOBJ_SIZE_V1;
38 	else
39 		size = sizeof (bpobj_phys_t);
40 
41 	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
42 	    DMU_OT_BPOBJ_HDR, size, tx));
43 }
44 
45 void
46 bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx)
47 {
48 	int64_t i;
49 	bpobj_t bpo;
50 	dmu_object_info_t doi;
51 	int epb;
52 	dmu_buf_t *dbuf = NULL;
53 
54 	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
55 
56 	mutex_enter(&bpo.bpo_lock);
57 
58 	if (!bpo.bpo_havesubobj || bpo.bpo_phys->bpo_subobjs == 0)
59 		goto out;
60 
61 	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
62 	epb = doi.doi_data_block_size / sizeof (uint64_t);
63 
64 	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
65 		uint64_t *objarray;
66 		uint64_t offset, blkoff;
67 
68 		offset = i * sizeof (uint64_t);
69 		blkoff = P2PHASE(i, epb);
70 
71 		if (dbuf == NULL || dbuf->db_offset > offset) {
72 			if (dbuf)
73 				dmu_buf_rele(dbuf, FTAG);
74 			VERIFY3U(0, ==, dmu_buf_hold(os,
75 			    bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
76 		}
77 
78 		ASSERT3U(offset, >=, dbuf->db_offset);
79 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
80 
81 		objarray = dbuf->db_data;
82 		bpobj_free(os, objarray[blkoff], tx);
83 	}
84 	if (dbuf) {
85 		dmu_buf_rele(dbuf, FTAG);
86 		dbuf = NULL;
87 	}
88 	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
89 
90 out:
91 	mutex_exit(&bpo.bpo_lock);
92 	bpobj_close(&bpo);
93 
94 	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
95 }
96 
97 int
98 bpobj_open(bpobj_t *bpo, objset_t *os, uint64_t object)
99 {
100 	dmu_object_info_t doi;
101 	int err;
102 
103 	err = dmu_object_info(os, object, &doi);
104 	if (err)
105 		return (err);
106 
107 	bzero(bpo, sizeof (*bpo));
108 	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
109 
110 	ASSERT(bpo->bpo_dbuf == NULL);
111 	ASSERT(bpo->bpo_phys == NULL);
112 	ASSERT(object != 0);
113 	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
114 	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
115 
116 	bpo->bpo_os = os;
117 	bpo->bpo_object = object;
118 	bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
119 	bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
120 	bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
121 
122 	err = dmu_bonus_hold(bpo->bpo_os,
123 	    bpo->bpo_object, bpo, &bpo->bpo_dbuf);
124 	if (err)
125 		return (err);
126 	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
127 	return (0);
128 }
129 
130 void
131 bpobj_close(bpobj_t *bpo)
132 {
133 	/* Lame workaround for closing a bpobj that was never opened. */
134 	if (bpo->bpo_object == 0)
135 		return;
136 
137 	dmu_buf_rele(bpo->bpo_dbuf, bpo);
138 	if (bpo->bpo_cached_dbuf != NULL)
139 		dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
140 	bpo->bpo_dbuf = NULL;
141 	bpo->bpo_phys = NULL;
142 	bpo->bpo_cached_dbuf = NULL;
143 
144 	mutex_destroy(&bpo->bpo_lock);
145 }
146 
147 static int
148 bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
149     boolean_t free)
150 {
151 	dmu_object_info_t doi;
152 	int epb;
153 	int64_t i;
154 	int err = 0;
155 	dmu_buf_t *dbuf = NULL;
156 
157 	mutex_enter(&bpo->bpo_lock);
158 
159 	if (free)
160 		dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
161 
162 	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
163 		blkptr_t *bparray;
164 		blkptr_t *bp;
165 		uint64_t offset, blkoff;
166 
167 		offset = i * sizeof (blkptr_t);
168 		blkoff = P2PHASE(i, bpo->bpo_epb);
169 
170 		if (dbuf == NULL || dbuf->db_offset > offset) {
171 			if (dbuf)
172 				dmu_buf_rele(dbuf, FTAG);
173 			err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
174 			    FTAG, &dbuf, 0);
175 			if (err)
176 				break;
177 		}
178 
179 		ASSERT3U(offset, >=, dbuf->db_offset);
180 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
181 
182 		bparray = dbuf->db_data;
183 		bp = &bparray[blkoff];
184 		err = func(arg, bp, tx);
185 		if (err)
186 			break;
187 		if (free) {
188 			bpo->bpo_phys->bpo_bytes -=
189 			    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
190 			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
191 			if (bpo->bpo_havecomp) {
192 				bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
193 				bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
194 			}
195 			bpo->bpo_phys->bpo_num_blkptrs--;
196 			ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
197 		}
198 	}
199 	if (dbuf) {
200 		dmu_buf_rele(dbuf, FTAG);
201 		dbuf = NULL;
202 	}
203 	if (free) {
204 		i++;
205 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
206 		    i * sizeof (blkptr_t), -1ULL, tx));
207 	}
208 	if (err || !bpo->bpo_havesubobj || bpo->bpo_phys->bpo_subobjs == 0)
209 		goto out;
210 
211 	ASSERT(bpo->bpo_havecomp);
212 	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
213 	if (err)
214 		return (err);
215 	epb = doi.doi_data_block_size / sizeof (uint64_t);
216 
217 	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
218 		uint64_t *objarray;
219 		uint64_t offset, blkoff;
220 		bpobj_t sublist;
221 		uint64_t used_before, comp_before, uncomp_before;
222 		uint64_t used_after, comp_after, uncomp_after;
223 
224 		offset = i * sizeof (uint64_t);
225 		blkoff = P2PHASE(i, epb);
226 
227 		if (dbuf == NULL || dbuf->db_offset > offset) {
228 			if (dbuf)
229 				dmu_buf_rele(dbuf, FTAG);
230 			err = dmu_buf_hold(bpo->bpo_os,
231 			    bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
232 			if (err)
233 				break;
234 		}
235 
236 		ASSERT3U(offset, >=, dbuf->db_offset);
237 		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
238 
239 		objarray = dbuf->db_data;
240 		err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
241 		if (err)
242 			break;
243 		if (free) {
244 			err = bpobj_space(&sublist,
245 			    &used_before, &comp_before, &uncomp_before);
246 			if (err)
247 				break;
248 		}
249 		err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
250 		if (free) {
251 			VERIFY3U(0, ==, bpobj_space(&sublist,
252 			    &used_after, &comp_after, &uncomp_after));
253 			bpo->bpo_phys->bpo_bytes -= used_before - used_after;
254 			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
255 			bpo->bpo_phys->bpo_comp -= comp_before - used_after;
256 			bpo->bpo_phys->bpo_uncomp -=
257 			    uncomp_before - uncomp_after;
258 		}
259 
260 		bpobj_close(&sublist);
261 		if (err)
262 			break;
263 		if (free) {
264 			err = dmu_object_free(bpo->bpo_os,
265 			    objarray[blkoff], tx);
266 			if (err)
267 				break;
268 			bpo->bpo_phys->bpo_num_subobjs--;
269 			ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
270 		}
271 	}
272 	if (dbuf) {
273 		dmu_buf_rele(dbuf, FTAG);
274 		dbuf = NULL;
275 	}
276 	if (free) {
277 		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
278 		    bpo->bpo_phys->bpo_subobjs,
279 		    (i + 1) * sizeof (uint64_t), -1ULL, tx));
280 	}
281 
282 out:
283 	/* If there are no entries, there should be no bytes. */
284 	ASSERT(bpo->bpo_phys->bpo_num_blkptrs > 0 ||
285 	    (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs > 0) ||
286 	    bpo->bpo_phys->bpo_bytes == 0);
287 
288 	mutex_exit(&bpo->bpo_lock);
289 	return (err);
290 }
291 
292 /*
293  * Iterate and remove the entries.  If func returns nonzero, iteration
294  * will stop and that entry will not be removed.
295  */
296 int
297 bpobj_iterate(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
298 {
299 	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
300 }
301 
302 /*
303  * Iterate the entries.  If func returns nonzero, iteration will stop.
304  */
305 int
306 bpobj_iterate_nofree(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
307 {
308 	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
309 }
310 
311 void
312 bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
313 {
314 	bpobj_t subbpo;
315 	uint64_t used, comp, uncomp;
316 
317 	ASSERT(bpo->bpo_havesubobj);
318 	ASSERT(bpo->bpo_havecomp);
319 
320 	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
321 	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
322 	bpobj_close(&subbpo);
323 
324 	if (used == 0) {
325 		/* No point in having an empty subobj. */
326 		bpobj_free(bpo->bpo_os, subobj, tx);
327 		return;
328 	}
329 
330 	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
331 	if (bpo->bpo_phys->bpo_subobjs == 0) {
332 		bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
333 		    DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
334 	}
335 
336 	mutex_enter(&bpo->bpo_lock);
337 	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
338 	    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
339 	    sizeof (subobj), &subobj, tx);
340 	bpo->bpo_phys->bpo_num_subobjs++;
341 	bpo->bpo_phys->bpo_bytes += used;
342 	bpo->bpo_phys->bpo_comp += comp;
343 	bpo->bpo_phys->bpo_uncomp += uncomp;
344 	mutex_exit(&bpo->bpo_lock);
345 }
346 
347 void
348 bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
349 {
350 	blkptr_t stored_bp = *bp;
351 	uint64_t offset;
352 	int blkoff;
353 	blkptr_t *bparray;
354 
355 	ASSERT(!BP_IS_HOLE(bp));
356 
357 	/* We never need the fill count. */
358 	stored_bp.blk_fill = 0;
359 
360 	/* The bpobj will compress better if we can leave off the checksum */
361 	if (!BP_GET_DEDUP(bp))
362 		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
363 
364 	mutex_enter(&bpo->bpo_lock);
365 
366 	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
367 	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
368 
369 	if (bpo->bpo_cached_dbuf == NULL ||
370 	    offset < bpo->bpo_cached_dbuf->db_offset ||
371 	    offset >= bpo->bpo_cached_dbuf->db_offset +
372 	    bpo->bpo_cached_dbuf->db_size) {
373 		if (bpo->bpo_cached_dbuf)
374 			dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
375 		VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
376 		    offset, bpo, &bpo->bpo_cached_dbuf, 0));
377 	}
378 
379 	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
380 	bparray = bpo->bpo_cached_dbuf->db_data;
381 	bparray[blkoff] = stored_bp;
382 
383 	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
384 	bpo->bpo_phys->bpo_num_blkptrs++;
385 	bpo->bpo_phys->bpo_bytes +=
386 	    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
387 	if (bpo->bpo_havecomp) {
388 		bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
389 		bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
390 	}
391 	mutex_exit(&bpo->bpo_lock);
392 }
393 
394 struct space_range_arg {
395 	spa_t *spa;
396 	uint64_t mintxg;
397 	uint64_t maxtxg;
398 	uint64_t used;
399 	uint64_t comp;
400 	uint64_t uncomp;
401 };
402 
403 /* ARGSUSED */
404 static int
405 space_range_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
406 {
407 	struct space_range_arg *sra = arg;
408 
409 	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
410 		sra->used += bp_get_dsize_sync(sra->spa, bp);
411 		sra->comp += BP_GET_PSIZE(bp);
412 		sra->uncomp += BP_GET_UCSIZE(bp);
413 	}
414 	return (0);
415 }
416 
417 int
418 bpobj_space(bpobj_t *bpo, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
419 {
420 	mutex_enter(&bpo->bpo_lock);
421 
422 	*usedp = bpo->bpo_phys->bpo_bytes;
423 	if (bpo->bpo_havecomp) {
424 		*compp = bpo->bpo_phys->bpo_comp;
425 		*uncompp = bpo->bpo_phys->bpo_uncomp;
426 		mutex_exit(&bpo->bpo_lock);
427 		return (0);
428 	} else {
429 		mutex_exit(&bpo->bpo_lock);
430 		return (bpobj_space_range(bpo, 0, UINT64_MAX,
431 		    usedp, compp, uncompp));
432 	}
433 }
434 
435 /*
436  * Return the amount of space in the bpobj which is:
437  * mintxg < blk_birth <= maxtxg
438  */
439 int
440 bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
441     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
442 {
443 	struct space_range_arg sra = { 0 };
444 	int err;
445 
446 	/*
447 	 * As an optimization, if they want the whole txg range, just
448 	 * get bpo_bytes rather than iterating over the bps.
449 	 */
450 	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
451 		return (bpobj_space(bpo, usedp, compp, uncompp));
452 
453 	sra.spa = dmu_objset_spa(bpo->bpo_os);
454 	sra.mintxg = mintxg;
455 	sra.maxtxg = maxtxg;
456 
457 	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
458 	*usedp = sra.used;
459 	*compp = sra.comp;
460 	*uncompp = sra.uncomp;
461 	return (err);
462 }
463