]>
Commit | Line | Data |
---|---|---|
428870ff BB |
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 | /* | |
23 | * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | */ | |
25 | ||
26 | #include <sys/zfs_context.h> | |
27 | #include <sys/spa.h> | |
28 | #include <sys/spa_impl.h> | |
29 | #include <sys/zio.h> | |
30 | #include <sys/ddt.h> | |
31 | #include <sys/zap.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/arc.h> | |
34 | #include <sys/dsl_pool.h> | |
35 | #include <sys/zio_checksum.h> | |
36 | #include <sys/zio_compress.h> | |
37 | #include <sys/dsl_scan.h> | |
38 | ||
572e2857 BB |
39 | /* |
40 | * Enable/disable prefetching of dedup-ed blocks which are going to be freed. | |
41 | */ | |
42 | int zfs_dedup_prefetch = 1; | |
43 | ||
428870ff BB |
44 | static const ddt_ops_t *ddt_ops[DDT_TYPES] = { |
45 | &ddt_zap_ops, | |
46 | }; | |
47 | ||
48 | static const char *ddt_class_name[DDT_CLASSES] = { | |
49 | "ditto", | |
50 | "duplicate", | |
51 | "unique", | |
52 | }; | |
53 | ||
54 | static void | |
55 | ddt_object_create(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
56 | dmu_tx_t *tx) | |
57 | { | |
58 | spa_t *spa = ddt->ddt_spa; | |
59 | objset_t *os = ddt->ddt_os; | |
60 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
61 | boolean_t prehash = zio_checksum_table[ddt->ddt_checksum].ci_dedup; | |
62 | char name[DDT_NAMELEN]; | |
63 | ||
64 | ddt_object_name(ddt, type, class, name); | |
65 | ||
66 | ASSERT(*objectp == 0); | |
67 | VERIFY(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash) == 0); | |
68 | ASSERT(*objectp != 0); | |
69 | ||
70 | VERIFY(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name, | |
71 | sizeof (uint64_t), 1, objectp, tx) == 0); | |
72 | ||
73 | VERIFY(zap_add(os, spa->spa_ddt_stat_object, name, | |
74 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
75 | &ddt->ddt_histogram[type][class], tx) == 0); | |
76 | } | |
77 | ||
78 | static void | |
79 | ddt_object_destroy(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
80 | dmu_tx_t *tx) | |
81 | { | |
82 | spa_t *spa = ddt->ddt_spa; | |
83 | objset_t *os = ddt->ddt_os; | |
84 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
e8fd45a0 | 85 | uint64_t count; |
428870ff BB |
86 | char name[DDT_NAMELEN]; |
87 | ||
88 | ddt_object_name(ddt, type, class, name); | |
89 | ||
90 | ASSERT(*objectp != 0); | |
428870ff | 91 | ASSERT(ddt_histogram_empty(&ddt->ddt_histogram[type][class])); |
e8fd45a0 | 92 | VERIFY(ddt_object_count(ddt, type, class, &count) == 0 && count == 0); |
428870ff BB |
93 | VERIFY(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx) == 0); |
94 | VERIFY(zap_remove(os, spa->spa_ddt_stat_object, name, tx) == 0); | |
95 | VERIFY(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx) == 0); | |
96 | bzero(&ddt->ddt_object_stats[type][class], sizeof (ddt_object_t)); | |
97 | ||
98 | *objectp = 0; | |
99 | } | |
100 | ||
101 | static int | |
102 | ddt_object_load(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
103 | { | |
104 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
105 | dmu_object_info_t doi; | |
e8fd45a0 | 106 | uint64_t count; |
428870ff BB |
107 | char name[DDT_NAMELEN]; |
108 | int error; | |
109 | ||
110 | ddt_object_name(ddt, type, class, name); | |
111 | ||
112 | error = zap_lookup(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, name, | |
113 | sizeof (uint64_t), 1, &ddt->ddt_object[type][class]); | |
114 | ||
115 | if (error) | |
116 | return (error); | |
117 | ||
118 | error = zap_lookup(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, | |
119 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
120 | &ddt->ddt_histogram[type][class]); | |
121 | ||
122 | /* | |
123 | * Seed the cached statistics. | |
124 | */ | |
2a4a9dc2 BB |
125 | error = ddt_object_info(ddt, type, class, &doi); |
126 | if (error) | |
127 | return (error); | |
428870ff | 128 | |
e8fd45a0 BB |
129 | error = ddt_object_count(ddt, type, class, &count); |
130 | if (error) | |
131 | return (error); | |
132 | ||
133 | ddo->ddo_count = count; | |
428870ff BB |
134 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
135 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
136 | ||
137 | ASSERT(error == 0); | |
138 | return (error); | |
139 | } | |
140 | ||
141 | static void | |
142 | ddt_object_sync(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
143 | dmu_tx_t *tx) | |
144 | { | |
145 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
146 | dmu_object_info_t doi; | |
e8fd45a0 | 147 | uint64_t count; |
428870ff BB |
148 | char name[DDT_NAMELEN]; |
149 | ||
150 | ddt_object_name(ddt, type, class, name); | |
151 | ||
152 | VERIFY(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, | |
153 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
154 | &ddt->ddt_histogram[type][class], tx) == 0); | |
155 | ||
156 | /* | |
157 | * Cache DDT statistics; this is the only time they'll change. | |
158 | */ | |
159 | VERIFY(ddt_object_info(ddt, type, class, &doi) == 0); | |
e8fd45a0 | 160 | VERIFY(ddt_object_count(ddt, type, class, &count) == 0); |
428870ff | 161 | |
e8fd45a0 | 162 | ddo->ddo_count = count; |
428870ff BB |
163 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
164 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
165 | } | |
166 | ||
167 | static int | |
168 | ddt_object_lookup(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
169 | ddt_entry_t *dde) | |
170 | { | |
171 | if (!ddt_object_exists(ddt, type, class)) | |
172 | return (ENOENT); | |
173 | ||
174 | return (ddt_ops[type]->ddt_op_lookup(ddt->ddt_os, | |
175 | ddt->ddt_object[type][class], dde)); | |
176 | } | |
177 | ||
178 | static void | |
179 | ddt_object_prefetch(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
180 | ddt_entry_t *dde) | |
181 | { | |
182 | if (!ddt_object_exists(ddt, type, class)) | |
183 | return; | |
184 | ||
185 | ddt_ops[type]->ddt_op_prefetch(ddt->ddt_os, | |
186 | ddt->ddt_object[type][class], dde); | |
187 | } | |
188 | ||
189 | int | |
190 | ddt_object_update(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
191 | ddt_entry_t *dde, dmu_tx_t *tx) | |
192 | { | |
193 | ASSERT(ddt_object_exists(ddt, type, class)); | |
194 | ||
195 | return (ddt_ops[type]->ddt_op_update(ddt->ddt_os, | |
196 | ddt->ddt_object[type][class], dde, tx)); | |
197 | } | |
198 | ||
199 | static int | |
200 | ddt_object_remove(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
201 | ddt_entry_t *dde, dmu_tx_t *tx) | |
202 | { | |
203 | ASSERT(ddt_object_exists(ddt, type, class)); | |
204 | ||
205 | return (ddt_ops[type]->ddt_op_remove(ddt->ddt_os, | |
206 | ddt->ddt_object[type][class], dde, tx)); | |
207 | } | |
208 | ||
209 | int | |
210 | ddt_object_walk(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
211 | uint64_t *walk, ddt_entry_t *dde) | |
212 | { | |
213 | ASSERT(ddt_object_exists(ddt, type, class)); | |
214 | ||
215 | return (ddt_ops[type]->ddt_op_walk(ddt->ddt_os, | |
216 | ddt->ddt_object[type][class], dde, walk)); | |
217 | } | |
218 | ||
e8fd45a0 BB |
219 | int |
220 | ddt_object_count(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
221 | uint64_t *count) | |
428870ff BB |
222 | { |
223 | ASSERT(ddt_object_exists(ddt, type, class)); | |
224 | ||
225 | return (ddt_ops[type]->ddt_op_count(ddt->ddt_os, | |
e8fd45a0 | 226 | ddt->ddt_object[type][class], count)); |
428870ff BB |
227 | } |
228 | ||
229 | int | |
230 | ddt_object_info(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
231 | dmu_object_info_t *doi) | |
232 | { | |
233 | if (!ddt_object_exists(ddt, type, class)) | |
234 | return (ENOENT); | |
235 | ||
236 | return (dmu_object_info(ddt->ddt_os, ddt->ddt_object[type][class], | |
237 | doi)); | |
238 | } | |
239 | ||
240 | boolean_t | |
241 | ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
242 | { | |
243 | return (!!ddt->ddt_object[type][class]); | |
244 | } | |
245 | ||
246 | void | |
247 | ddt_object_name(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
248 | char *name) | |
249 | { | |
250 | (void) sprintf(name, DMU_POOL_DDT, | |
251 | zio_checksum_table[ddt->ddt_checksum].ci_name, | |
252 | ddt_ops[type]->ddt_op_name, ddt_class_name[class]); | |
253 | } | |
254 | ||
255 | void | |
256 | ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, uint64_t txg) | |
257 | { | |
d6320ddb | 258 | int d; |
428870ff BB |
259 | ASSERT(txg != 0); |
260 | ||
d6320ddb | 261 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
262 | bp->blk_dva[d] = ddp->ddp_dva[d]; |
263 | BP_SET_BIRTH(bp, txg, ddp->ddp_phys_birth); | |
264 | } | |
265 | ||
266 | void | |
267 | ddt_bp_create(enum zio_checksum checksum, | |
268 | const ddt_key_t *ddk, const ddt_phys_t *ddp, blkptr_t *bp) | |
269 | { | |
270 | BP_ZERO(bp); | |
271 | ||
272 | if (ddp != NULL) | |
273 | ddt_bp_fill(ddp, bp, ddp->ddp_phys_birth); | |
274 | ||
275 | bp->blk_cksum = ddk->ddk_cksum; | |
276 | bp->blk_fill = 1; | |
277 | ||
278 | BP_SET_LSIZE(bp, DDK_GET_LSIZE(ddk)); | |
279 | BP_SET_PSIZE(bp, DDK_GET_PSIZE(ddk)); | |
280 | BP_SET_COMPRESS(bp, DDK_GET_COMPRESS(ddk)); | |
281 | BP_SET_CHECKSUM(bp, checksum); | |
282 | BP_SET_TYPE(bp, DMU_OT_DEDUP); | |
283 | BP_SET_LEVEL(bp, 0); | |
284 | BP_SET_DEDUP(bp, 0); | |
285 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); | |
286 | } | |
287 | ||
288 | void | |
289 | ddt_key_fill(ddt_key_t *ddk, const blkptr_t *bp) | |
290 | { | |
291 | ddk->ddk_cksum = bp->blk_cksum; | |
292 | ddk->ddk_prop = 0; | |
293 | ||
294 | DDK_SET_LSIZE(ddk, BP_GET_LSIZE(bp)); | |
295 | DDK_SET_PSIZE(ddk, BP_GET_PSIZE(bp)); | |
296 | DDK_SET_COMPRESS(ddk, BP_GET_COMPRESS(bp)); | |
297 | } | |
298 | ||
299 | void | |
300 | ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp) | |
301 | { | |
d6320ddb | 302 | int d; |
428870ff BB |
303 | ASSERT(ddp->ddp_phys_birth == 0); |
304 | ||
d6320ddb | 305 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
306 | ddp->ddp_dva[d] = bp->blk_dva[d]; |
307 | ddp->ddp_phys_birth = BP_PHYSICAL_BIRTH(bp); | |
308 | } | |
309 | ||
310 | void | |
311 | ddt_phys_clear(ddt_phys_t *ddp) | |
312 | { | |
313 | bzero(ddp, sizeof (*ddp)); | |
314 | } | |
315 | ||
316 | void | |
317 | ddt_phys_addref(ddt_phys_t *ddp) | |
318 | { | |
319 | ddp->ddp_refcnt++; | |
320 | } | |
321 | ||
322 | void | |
323 | ddt_phys_decref(ddt_phys_t *ddp) | |
324 | { | |
325 | ASSERT((int64_t)ddp->ddp_refcnt > 0); | |
326 | ddp->ddp_refcnt--; | |
327 | } | |
328 | ||
329 | void | |
330 | ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, uint64_t txg) | |
331 | { | |
332 | blkptr_t blk; | |
333 | ||
334 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
335 | ddt_phys_clear(ddp); | |
336 | zio_free(ddt->ddt_spa, txg, &blk); | |
337 | } | |
338 | ||
339 | ddt_phys_t * | |
340 | ddt_phys_select(const ddt_entry_t *dde, const blkptr_t *bp) | |
341 | { | |
342 | ddt_phys_t *ddp = (ddt_phys_t *)dde->dde_phys; | |
d6320ddb | 343 | int p; |
428870ff | 344 | |
d6320ddb | 345 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
346 | if (DVA_EQUAL(BP_IDENTITY(bp), &ddp->ddp_dva[0]) && |
347 | BP_PHYSICAL_BIRTH(bp) == ddp->ddp_phys_birth) | |
348 | return (ddp); | |
349 | } | |
350 | return (NULL); | |
351 | } | |
352 | ||
353 | uint64_t | |
354 | ddt_phys_total_refcnt(const ddt_entry_t *dde) | |
355 | { | |
356 | uint64_t refcnt = 0; | |
d6320ddb | 357 | int p; |
428870ff | 358 | |
d6320ddb | 359 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) |
428870ff BB |
360 | refcnt += dde->dde_phys[p].ddp_refcnt; |
361 | ||
362 | return (refcnt); | |
363 | } | |
364 | ||
365 | static void | |
366 | ddt_stat_generate(ddt_t *ddt, ddt_entry_t *dde, ddt_stat_t *dds) | |
367 | { | |
368 | spa_t *spa = ddt->ddt_spa; | |
369 | ddt_phys_t *ddp = dde->dde_phys; | |
370 | ddt_key_t *ddk = &dde->dde_key; | |
371 | uint64_t lsize = DDK_GET_LSIZE(ddk); | |
372 | uint64_t psize = DDK_GET_PSIZE(ddk); | |
d6320ddb | 373 | int p, d; |
428870ff BB |
374 | |
375 | bzero(dds, sizeof (*dds)); | |
376 | ||
d6320ddb | 377 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
378 | uint64_t dsize = 0; |
379 | uint64_t refcnt = ddp->ddp_refcnt; | |
380 | ||
381 | if (ddp->ddp_phys_birth == 0) | |
382 | continue; | |
383 | ||
d6320ddb | 384 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
385 | dsize += dva_get_dsize_sync(spa, &ddp->ddp_dva[d]); |
386 | ||
387 | dds->dds_blocks += 1; | |
388 | dds->dds_lsize += lsize; | |
389 | dds->dds_psize += psize; | |
390 | dds->dds_dsize += dsize; | |
391 | ||
392 | dds->dds_ref_blocks += refcnt; | |
393 | dds->dds_ref_lsize += lsize * refcnt; | |
394 | dds->dds_ref_psize += psize * refcnt; | |
395 | dds->dds_ref_dsize += dsize * refcnt; | |
396 | } | |
397 | } | |
398 | ||
399 | void | |
400 | ddt_stat_add(ddt_stat_t *dst, const ddt_stat_t *src, uint64_t neg) | |
401 | { | |
402 | const uint64_t *s = (const uint64_t *)src; | |
403 | uint64_t *d = (uint64_t *)dst; | |
404 | uint64_t *d_end = (uint64_t *)(dst + 1); | |
405 | ||
406 | ASSERT(neg == 0 || neg == -1ULL); /* add or subtract */ | |
407 | ||
408 | while (d < d_end) | |
409 | *d++ += (*s++ ^ neg) - neg; | |
410 | } | |
411 | ||
412 | static void | |
413 | ddt_stat_update(ddt_t *ddt, ddt_entry_t *dde, uint64_t neg) | |
414 | { | |
415 | ddt_stat_t dds; | |
416 | ddt_histogram_t *ddh; | |
417 | int bucket; | |
418 | ||
419 | ddt_stat_generate(ddt, dde, &dds); | |
420 | ||
421 | bucket = highbit(dds.dds_ref_blocks) - 1; | |
422 | ASSERT(bucket >= 0); | |
423 | ||
424 | ddh = &ddt->ddt_histogram[dde->dde_type][dde->dde_class]; | |
425 | ||
426 | ddt_stat_add(&ddh->ddh_stat[bucket], &dds, neg); | |
427 | } | |
428 | ||
429 | void | |
430 | ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src) | |
431 | { | |
d6320ddb BB |
432 | int h; |
433 | ||
434 | for (h = 0; h < 64; h++) | |
428870ff BB |
435 | ddt_stat_add(&dst->ddh_stat[h], &src->ddh_stat[h], 0); |
436 | } | |
437 | ||
438 | void | |
439 | ddt_histogram_stat(ddt_stat_t *dds, const ddt_histogram_t *ddh) | |
440 | { | |
d6320ddb BB |
441 | int h; |
442 | ||
428870ff BB |
443 | bzero(dds, sizeof (*dds)); |
444 | ||
d6320ddb | 445 | for (h = 0; h < 64; h++) |
428870ff BB |
446 | ddt_stat_add(dds, &ddh->ddh_stat[h], 0); |
447 | } | |
448 | ||
449 | boolean_t | |
450 | ddt_histogram_empty(const ddt_histogram_t *ddh) | |
451 | { | |
452 | const uint64_t *s = (const uint64_t *)ddh; | |
453 | const uint64_t *s_end = (const uint64_t *)(ddh + 1); | |
454 | ||
455 | while (s < s_end) | |
456 | if (*s++ != 0) | |
457 | return (B_FALSE); | |
458 | ||
459 | return (B_TRUE); | |
460 | } | |
461 | ||
462 | void | |
463 | ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo_total) | |
464 | { | |
d6320ddb BB |
465 | enum zio_checksum c; |
466 | enum ddt_type type; | |
467 | enum ddt_class class; | |
468 | ||
428870ff | 469 | /* Sum the statistics we cached in ddt_object_sync(). */ |
d6320ddb | 470 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 471 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
472 | for (type = 0; type < DDT_TYPES; type++) { |
473 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
474 | class++) { |
475 | ddt_object_t *ddo = | |
476 | &ddt->ddt_object_stats[type][class]; | |
477 | ddo_total->ddo_count += ddo->ddo_count; | |
478 | ddo_total->ddo_dspace += ddo->ddo_dspace; | |
479 | ddo_total->ddo_mspace += ddo->ddo_mspace; | |
480 | } | |
481 | } | |
482 | } | |
483 | ||
484 | /* ... and compute the averages. */ | |
485 | if (ddo_total->ddo_count != 0) { | |
486 | ddo_total->ddo_dspace /= ddo_total->ddo_count; | |
487 | ddo_total->ddo_mspace /= ddo_total->ddo_count; | |
428870ff BB |
488 | } |
489 | } | |
490 | ||
491 | void | |
492 | ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh) | |
493 | { | |
d6320ddb BB |
494 | enum zio_checksum c; |
495 | enum ddt_type type; | |
496 | enum ddt_class class; | |
497 | ||
498 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { | |
428870ff | 499 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
500 | for (type = 0; type < DDT_TYPES; type++) { |
501 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
502 | class++) { |
503 | ddt_histogram_add(ddh, | |
504 | &ddt->ddt_histogram_cache[type][class]); | |
505 | } | |
506 | } | |
507 | } | |
508 | } | |
509 | ||
510 | void | |
511 | ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total) | |
512 | { | |
513 | ddt_histogram_t *ddh_total; | |
514 | ||
00b46022 | 515 | /* XXX: Move to a slab */ |
b8d06fca | 516 | ddh_total = kmem_zalloc(sizeof (ddt_histogram_t), KM_PUSHPAGE); |
428870ff BB |
517 | ddt_get_dedup_histogram(spa, ddh_total); |
518 | ddt_histogram_stat(dds_total, ddh_total); | |
519 | kmem_free(ddh_total, sizeof (ddt_histogram_t)); | |
520 | } | |
521 | ||
522 | uint64_t | |
523 | ddt_get_dedup_dspace(spa_t *spa) | |
524 | { | |
525 | ddt_stat_t dds_total = { 0 }; | |
526 | ||
527 | ddt_get_dedup_stats(spa, &dds_total); | |
528 | return (dds_total.dds_ref_dsize - dds_total.dds_dsize); | |
529 | } | |
530 | ||
531 | uint64_t | |
532 | ddt_get_pool_dedup_ratio(spa_t *spa) | |
533 | { | |
534 | ddt_stat_t dds_total = { 0 }; | |
535 | ||
536 | ddt_get_dedup_stats(spa, &dds_total); | |
537 | if (dds_total.dds_dsize == 0) | |
538 | return (100); | |
539 | ||
540 | return (dds_total.dds_ref_dsize * 100 / dds_total.dds_dsize); | |
541 | } | |
542 | ||
543 | int | |
544 | ddt_ditto_copies_needed(ddt_t *ddt, ddt_entry_t *dde, ddt_phys_t *ddp_willref) | |
545 | { | |
546 | spa_t *spa = ddt->ddt_spa; | |
547 | uint64_t total_refcnt = 0; | |
548 | uint64_t ditto = spa->spa_dedup_ditto; | |
549 | int total_copies = 0; | |
550 | int desired_copies = 0; | |
d6320ddb | 551 | int p; |
428870ff | 552 | |
d6320ddb | 553 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
554 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
555 | zio_t *zio = dde->dde_lead_zio[p]; | |
556 | uint64_t refcnt = ddp->ddp_refcnt; /* committed refs */ | |
557 | if (zio != NULL) | |
558 | refcnt += zio->io_parent_count; /* pending refs */ | |
559 | if (ddp == ddp_willref) | |
560 | refcnt++; /* caller's ref */ | |
561 | if (refcnt != 0) { | |
562 | total_refcnt += refcnt; | |
563 | total_copies += p; | |
564 | } | |
565 | } | |
566 | ||
567 | if (ditto == 0 || ditto > UINT32_MAX) | |
568 | ditto = UINT32_MAX; | |
569 | ||
570 | if (total_refcnt >= 1) | |
571 | desired_copies++; | |
572 | if (total_refcnt >= ditto) | |
573 | desired_copies++; | |
574 | if (total_refcnt >= ditto * ditto) | |
575 | desired_copies++; | |
576 | ||
577 | return (MAX(desired_copies, total_copies) - total_copies); | |
578 | } | |
579 | ||
580 | int | |
581 | ddt_ditto_copies_present(ddt_entry_t *dde) | |
582 | { | |
583 | ddt_phys_t *ddp = &dde->dde_phys[DDT_PHYS_DITTO]; | |
584 | dva_t *dva = ddp->ddp_dva; | |
585 | int copies = 0 - DVA_GET_GANG(dva); | |
d6320ddb | 586 | int d; |
428870ff | 587 | |
d6320ddb | 588 | for (d = 0; d < SPA_DVAS_PER_BP; d++, dva++) |
428870ff BB |
589 | if (DVA_IS_VALID(dva)) |
590 | copies++; | |
591 | ||
592 | ASSERT(copies >= 0 && copies < SPA_DVAS_PER_BP); | |
593 | ||
594 | return (copies); | |
595 | } | |
596 | ||
597 | size_t | |
598 | ddt_compress(void *src, uchar_t *dst, size_t s_len, size_t d_len) | |
599 | { | |
600 | uchar_t *version = dst++; | |
601 | int cpfunc = ZIO_COMPRESS_ZLE; | |
602 | zio_compress_info_t *ci = &zio_compress_table[cpfunc]; | |
603 | size_t c_len; | |
604 | ||
605 | ASSERT(d_len >= s_len + 1); /* no compression plus version byte */ | |
606 | ||
607 | c_len = ci->ci_compress(src, dst, s_len, d_len - 1, ci->ci_level); | |
608 | ||
609 | if (c_len == s_len) { | |
610 | cpfunc = ZIO_COMPRESS_OFF; | |
611 | bcopy(src, dst, s_len); | |
612 | } | |
613 | ||
614 | *version = (ZFS_HOST_BYTEORDER & DDT_COMPRESS_BYTEORDER_MASK) | cpfunc; | |
615 | ||
616 | return (c_len + 1); | |
617 | } | |
618 | ||
619 | void | |
620 | ddt_decompress(uchar_t *src, void *dst, size_t s_len, size_t d_len) | |
621 | { | |
622 | uchar_t version = *src++; | |
623 | int cpfunc = version & DDT_COMPRESS_FUNCTION_MASK; | |
624 | zio_compress_info_t *ci = &zio_compress_table[cpfunc]; | |
625 | ||
626 | if (ci->ci_decompress != NULL) | |
627 | (void) ci->ci_decompress(src, dst, s_len, d_len, ci->ci_level); | |
628 | else | |
629 | bcopy(src, dst, d_len); | |
630 | ||
631 | if ((version ^ ZFS_HOST_BYTEORDER) & DDT_COMPRESS_BYTEORDER_MASK) | |
632 | byteswap_uint64_array(dst, d_len); | |
633 | } | |
634 | ||
635 | ddt_t * | |
636 | ddt_select_by_checksum(spa_t *spa, enum zio_checksum c) | |
637 | { | |
638 | return (spa->spa_ddt[c]); | |
639 | } | |
640 | ||
641 | ddt_t * | |
642 | ddt_select(spa_t *spa, const blkptr_t *bp) | |
643 | { | |
644 | return (spa->spa_ddt[BP_GET_CHECKSUM(bp)]); | |
645 | } | |
646 | ||
647 | void | |
648 | ddt_enter(ddt_t *ddt) | |
649 | { | |
650 | mutex_enter(&ddt->ddt_lock); | |
651 | } | |
652 | ||
653 | void | |
654 | ddt_exit(ddt_t *ddt) | |
655 | { | |
656 | mutex_exit(&ddt->ddt_lock); | |
657 | } | |
658 | ||
659 | static ddt_entry_t * | |
660 | ddt_alloc(const ddt_key_t *ddk) | |
661 | { | |
662 | ddt_entry_t *dde; | |
663 | ||
00b46022 | 664 | /* XXX: Move to a slab */ |
95fd8c9a | 665 | dde = kmem_zalloc(sizeof (ddt_entry_t), KM_PUSHPAGE); |
428870ff BB |
666 | cv_init(&dde->dde_cv, NULL, CV_DEFAULT, NULL); |
667 | ||
668 | dde->dde_key = *ddk; | |
669 | ||
670 | return (dde); | |
671 | } | |
672 | ||
673 | static void | |
674 | ddt_free(ddt_entry_t *dde) | |
675 | { | |
d6320ddb BB |
676 | int p; |
677 | ||
428870ff BB |
678 | ASSERT(!dde->dde_loading); |
679 | ||
d6320ddb | 680 | for (p = 0; p < DDT_PHYS_TYPES; p++) |
428870ff BB |
681 | ASSERT(dde->dde_lead_zio[p] == NULL); |
682 | ||
683 | if (dde->dde_repair_data != NULL) | |
684 | zio_buf_free(dde->dde_repair_data, | |
685 | DDK_GET_PSIZE(&dde->dde_key)); | |
686 | ||
687 | cv_destroy(&dde->dde_cv); | |
688 | kmem_free(dde, sizeof (*dde)); | |
689 | } | |
690 | ||
691 | void | |
692 | ddt_remove(ddt_t *ddt, ddt_entry_t *dde) | |
693 | { | |
694 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
695 | ||
696 | avl_remove(&ddt->ddt_tree, dde); | |
697 | ddt_free(dde); | |
698 | } | |
699 | ||
700 | ddt_entry_t * | |
701 | ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add) | |
702 | { | |
703 | ddt_entry_t *dde, dde_search; | |
704 | enum ddt_type type; | |
705 | enum ddt_class class; | |
706 | avl_index_t where; | |
707 | int error; | |
708 | ||
709 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
710 | ||
711 | ddt_key_fill(&dde_search.dde_key, bp); | |
712 | ||
713 | dde = avl_find(&ddt->ddt_tree, &dde_search, &where); | |
714 | if (dde == NULL) { | |
715 | if (!add) | |
716 | return (NULL); | |
717 | dde = ddt_alloc(&dde_search.dde_key); | |
718 | avl_insert(&ddt->ddt_tree, dde, where); | |
719 | } | |
720 | ||
721 | while (dde->dde_loading) | |
722 | cv_wait(&dde->dde_cv, &ddt->ddt_lock); | |
723 | ||
724 | if (dde->dde_loaded) | |
725 | return (dde); | |
726 | ||
727 | dde->dde_loading = B_TRUE; | |
728 | ||
729 | ddt_exit(ddt); | |
730 | ||
731 | error = ENOENT; | |
732 | ||
733 | for (type = 0; type < DDT_TYPES; type++) { | |
734 | for (class = 0; class < DDT_CLASSES; class++) { | |
735 | error = ddt_object_lookup(ddt, type, class, dde); | |
736 | if (error != ENOENT) | |
737 | break; | |
738 | } | |
739 | if (error != ENOENT) | |
740 | break; | |
741 | } | |
742 | ||
743 | ASSERT(error == 0 || error == ENOENT); | |
744 | ||
745 | ddt_enter(ddt); | |
746 | ||
747 | ASSERT(dde->dde_loaded == B_FALSE); | |
748 | ASSERT(dde->dde_loading == B_TRUE); | |
749 | ||
750 | dde->dde_type = type; /* will be DDT_TYPES if no entry found */ | |
751 | dde->dde_class = class; /* will be DDT_CLASSES if no entry found */ | |
752 | dde->dde_loaded = B_TRUE; | |
753 | dde->dde_loading = B_FALSE; | |
754 | ||
755 | if (error == 0) | |
756 | ddt_stat_update(ddt, dde, -1ULL); | |
757 | ||
758 | cv_broadcast(&dde->dde_cv); | |
759 | ||
760 | return (dde); | |
761 | } | |
762 | ||
763 | void | |
764 | ddt_prefetch(spa_t *spa, const blkptr_t *bp) | |
765 | { | |
766 | ddt_t *ddt; | |
767 | ddt_entry_t dde; | |
d6320ddb BB |
768 | enum ddt_type type; |
769 | enum ddt_class class; | |
428870ff | 770 | |
572e2857 | 771 | if (!zfs_dedup_prefetch || bp == NULL || !BP_GET_DEDUP(bp)) |
428870ff BB |
772 | return; |
773 | ||
774 | /* | |
572e2857 BB |
775 | * We only remove the DDT once all tables are empty and only |
776 | * prefetch dedup blocks when there are entries in the DDT. | |
777 | * Thus no locking is required as the DDT can't disappear on us. | |
428870ff BB |
778 | */ |
779 | ddt = ddt_select(spa, bp); | |
780 | ddt_key_fill(&dde.dde_key, bp); | |
781 | ||
d6320ddb BB |
782 | for (type = 0; type < DDT_TYPES; type++) { |
783 | for (class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
784 | ddt_object_prefetch(ddt, type, class, &dde); |
785 | } | |
786 | } | |
787 | } | |
788 | ||
789 | int | |
790 | ddt_entry_compare(const void *x1, const void *x2) | |
791 | { | |
792 | const ddt_entry_t *dde1 = x1; | |
793 | const ddt_entry_t *dde2 = x2; | |
794 | const uint64_t *u1 = (const uint64_t *)&dde1->dde_key; | |
795 | const uint64_t *u2 = (const uint64_t *)&dde2->dde_key; | |
d6320ddb | 796 | int i; |
428870ff | 797 | |
d6320ddb | 798 | for (i = 0; i < DDT_KEY_WORDS; i++) { |
428870ff BB |
799 | if (u1[i] < u2[i]) |
800 | return (-1); | |
801 | if (u1[i] > u2[i]) | |
802 | return (1); | |
803 | } | |
804 | ||
805 | return (0); | |
806 | } | |
807 | ||
808 | static ddt_t * | |
809 | ddt_table_alloc(spa_t *spa, enum zio_checksum c) | |
810 | { | |
811 | ddt_t *ddt; | |
812 | ||
00b46022 | 813 | /* XXX: Move to a slab */ |
95fd8c9a | 814 | ddt = kmem_zalloc(sizeof (*ddt), KM_PUSHPAGE | KM_NODEBUG); |
428870ff BB |
815 | |
816 | mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL); | |
817 | avl_create(&ddt->ddt_tree, ddt_entry_compare, | |
818 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); | |
819 | avl_create(&ddt->ddt_repair_tree, ddt_entry_compare, | |
820 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); | |
821 | ddt->ddt_checksum = c; | |
822 | ddt->ddt_spa = spa; | |
823 | ddt->ddt_os = spa->spa_meta_objset; | |
824 | ||
825 | return (ddt); | |
826 | } | |
827 | ||
828 | static void | |
829 | ddt_table_free(ddt_t *ddt) | |
830 | { | |
831 | ASSERT(avl_numnodes(&ddt->ddt_tree) == 0); | |
832 | ASSERT(avl_numnodes(&ddt->ddt_repair_tree) == 0); | |
833 | avl_destroy(&ddt->ddt_tree); | |
834 | avl_destroy(&ddt->ddt_repair_tree); | |
835 | mutex_destroy(&ddt->ddt_lock); | |
836 | kmem_free(ddt, sizeof (*ddt)); | |
837 | } | |
838 | ||
839 | void | |
840 | ddt_create(spa_t *spa) | |
841 | { | |
d6320ddb BB |
842 | enum zio_checksum c; |
843 | ||
428870ff BB |
844 | spa->spa_dedup_checksum = ZIO_DEDUPCHECKSUM; |
845 | ||
d6320ddb | 846 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) |
428870ff BB |
847 | spa->spa_ddt[c] = ddt_table_alloc(spa, c); |
848 | } | |
849 | ||
850 | int | |
851 | ddt_load(spa_t *spa) | |
852 | { | |
d6320ddb BB |
853 | enum zio_checksum c; |
854 | enum ddt_type type; | |
855 | enum ddt_class class; | |
428870ff BB |
856 | int error; |
857 | ||
858 | ddt_create(spa); | |
859 | ||
860 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
861 | DMU_POOL_DDT_STATS, sizeof (uint64_t), 1, | |
862 | &spa->spa_ddt_stat_object); | |
863 | ||
864 | if (error) | |
865 | return (error == ENOENT ? 0 : error); | |
866 | ||
d6320ddb | 867 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 868 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
869 | for (type = 0; type < DDT_TYPES; type++) { |
870 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
871 | class++) { |
872 | error = ddt_object_load(ddt, type, class); | |
873 | if (error != 0 && error != ENOENT) | |
874 | return (error); | |
875 | } | |
876 | } | |
877 | ||
878 | /* | |
879 | * Seed the cached histograms. | |
880 | */ | |
881 | bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache, | |
882 | sizeof (ddt->ddt_histogram)); | |
883 | } | |
884 | ||
885 | return (0); | |
886 | } | |
887 | ||
888 | void | |
889 | ddt_unload(spa_t *spa) | |
890 | { | |
d6320ddb BB |
891 | enum zio_checksum c; |
892 | ||
893 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { | |
428870ff BB |
894 | if (spa->spa_ddt[c]) { |
895 | ddt_table_free(spa->spa_ddt[c]); | |
896 | spa->spa_ddt[c] = NULL; | |
897 | } | |
898 | } | |
899 | } | |
900 | ||
901 | boolean_t | |
902 | ddt_class_contains(spa_t *spa, enum ddt_class max_class, const blkptr_t *bp) | |
903 | { | |
904 | ddt_t *ddt; | |
e95b3bdc | 905 | ddt_entry_t *dde; |
d6320ddb BB |
906 | enum ddt_type type; |
907 | enum ddt_class class; | |
428870ff BB |
908 | |
909 | if (!BP_GET_DEDUP(bp)) | |
910 | return (B_FALSE); | |
911 | ||
912 | if (max_class == DDT_CLASS_UNIQUE) | |
913 | return (B_TRUE); | |
914 | ||
915 | ddt = spa->spa_ddt[BP_GET_CHECKSUM(bp)]; | |
95fd8c9a | 916 | dde = kmem_alloc(sizeof(ddt_entry_t), KM_PUSHPAGE); |
428870ff | 917 | |
e95b3bdc | 918 | ddt_key_fill(&(dde->dde_key), bp); |
428870ff | 919 | |
e95b3bdc BB |
920 | for (type = 0; type < DDT_TYPES; type++) { |
921 | for (class = 0; class <= max_class; class++) { | |
922 | if (ddt_object_lookup(ddt, type, class, dde) == 0) { | |
923 | kmem_free(dde, sizeof(ddt_entry_t)); | |
428870ff | 924 | return (B_TRUE); |
e95b3bdc BB |
925 | } |
926 | } | |
927 | } | |
428870ff | 928 | |
e95b3bdc | 929 | kmem_free(dde, sizeof(ddt_entry_t)); |
428870ff BB |
930 | return (B_FALSE); |
931 | } | |
932 | ||
933 | ddt_entry_t * | |
934 | ddt_repair_start(ddt_t *ddt, const blkptr_t *bp) | |
935 | { | |
936 | ddt_key_t ddk; | |
937 | ddt_entry_t *dde; | |
d6320ddb BB |
938 | enum ddt_type type; |
939 | enum ddt_class class; | |
428870ff BB |
940 | |
941 | ddt_key_fill(&ddk, bp); | |
942 | ||
943 | dde = ddt_alloc(&ddk); | |
944 | ||
d6320ddb BB |
945 | for (type = 0; type < DDT_TYPES; type++) { |
946 | for (class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
947 | /* |
948 | * We can only do repair if there are multiple copies | |
949 | * of the block. For anything in the UNIQUE class, | |
950 | * there's definitely only one copy, so don't even try. | |
951 | */ | |
952 | if (class != DDT_CLASS_UNIQUE && | |
953 | ddt_object_lookup(ddt, type, class, dde) == 0) | |
954 | return (dde); | |
955 | } | |
956 | } | |
957 | ||
958 | bzero(dde->dde_phys, sizeof (dde->dde_phys)); | |
959 | ||
960 | return (dde); | |
961 | } | |
962 | ||
963 | void | |
964 | ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde) | |
965 | { | |
966 | avl_index_t where; | |
967 | ||
968 | ddt_enter(ddt); | |
969 | ||
970 | if (dde->dde_repair_data != NULL && spa_writeable(ddt->ddt_spa) && | |
971 | avl_find(&ddt->ddt_repair_tree, dde, &where) == NULL) | |
972 | avl_insert(&ddt->ddt_repair_tree, dde, where); | |
973 | else | |
974 | ddt_free(dde); | |
975 | ||
976 | ddt_exit(ddt); | |
977 | } | |
978 | ||
979 | static void | |
980 | ddt_repair_entry_done(zio_t *zio) | |
981 | { | |
982 | ddt_entry_t *rdde = zio->io_private; | |
983 | ||
984 | ddt_free(rdde); | |
985 | } | |
986 | ||
987 | static void | |
988 | ddt_repair_entry(ddt_t *ddt, ddt_entry_t *dde, ddt_entry_t *rdde, zio_t *rio) | |
989 | { | |
990 | ddt_phys_t *ddp = dde->dde_phys; | |
991 | ddt_phys_t *rddp = rdde->dde_phys; | |
992 | ddt_key_t *ddk = &dde->dde_key; | |
993 | ddt_key_t *rddk = &rdde->dde_key; | |
994 | zio_t *zio; | |
995 | blkptr_t blk; | |
d6320ddb | 996 | int p; |
428870ff BB |
997 | |
998 | zio = zio_null(rio, rio->io_spa, NULL, | |
999 | ddt_repair_entry_done, rdde, rio->io_flags); | |
1000 | ||
d6320ddb | 1001 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++, rddp++) { |
428870ff BB |
1002 | if (ddp->ddp_phys_birth == 0 || |
1003 | ddp->ddp_phys_birth != rddp->ddp_phys_birth || | |
1004 | bcmp(ddp->ddp_dva, rddp->ddp_dva, sizeof (ddp->ddp_dva))) | |
1005 | continue; | |
1006 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
1007 | zio_nowait(zio_rewrite(zio, zio->io_spa, 0, &blk, | |
1008 | rdde->dde_repair_data, DDK_GET_PSIZE(rddk), NULL, NULL, | |
1009 | ZIO_PRIORITY_SYNC_WRITE, ZIO_DDT_CHILD_FLAGS(zio), NULL)); | |
1010 | } | |
1011 | ||
1012 | zio_nowait(zio); | |
1013 | } | |
1014 | ||
1015 | static void | |
1016 | ddt_repair_table(ddt_t *ddt, zio_t *rio) | |
1017 | { | |
1018 | spa_t *spa = ddt->ddt_spa; | |
1019 | ddt_entry_t *dde, *rdde_next, *rdde; | |
1020 | avl_tree_t *t = &ddt->ddt_repair_tree; | |
1021 | blkptr_t blk; | |
1022 | ||
1023 | if (spa_sync_pass(spa) > 1) | |
1024 | return; | |
1025 | ||
1026 | ddt_enter(ddt); | |
1027 | for (rdde = avl_first(t); rdde != NULL; rdde = rdde_next) { | |
1028 | rdde_next = AVL_NEXT(t, rdde); | |
1029 | avl_remove(&ddt->ddt_repair_tree, rdde); | |
1030 | ddt_exit(ddt); | |
1031 | ddt_bp_create(ddt->ddt_checksum, &rdde->dde_key, NULL, &blk); | |
1032 | dde = ddt_repair_start(ddt, &blk); | |
1033 | ddt_repair_entry(ddt, dde, rdde, rio); | |
1034 | ddt_repair_done(ddt, dde); | |
1035 | ddt_enter(ddt); | |
1036 | } | |
1037 | ddt_exit(ddt); | |
1038 | } | |
1039 | ||
1040 | static void | |
1041 | ddt_sync_entry(ddt_t *ddt, ddt_entry_t *dde, dmu_tx_t *tx, uint64_t txg) | |
1042 | { | |
1043 | dsl_pool_t *dp = ddt->ddt_spa->spa_dsl_pool; | |
1044 | ddt_phys_t *ddp = dde->dde_phys; | |
1045 | ddt_key_t *ddk = &dde->dde_key; | |
1046 | enum ddt_type otype = dde->dde_type; | |
1047 | enum ddt_type ntype = DDT_TYPE_CURRENT; | |
1048 | enum ddt_class oclass = dde->dde_class; | |
1049 | enum ddt_class nclass; | |
1050 | uint64_t total_refcnt = 0; | |
d6320ddb | 1051 | int p; |
428870ff BB |
1052 | |
1053 | ASSERT(dde->dde_loaded); | |
1054 | ASSERT(!dde->dde_loading); | |
1055 | ||
d6320ddb | 1056 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
1057 | ASSERT(dde->dde_lead_zio[p] == NULL); |
1058 | ASSERT((int64_t)ddp->ddp_refcnt >= 0); | |
1059 | if (ddp->ddp_phys_birth == 0) { | |
1060 | ASSERT(ddp->ddp_refcnt == 0); | |
1061 | continue; | |
1062 | } | |
1063 | if (p == DDT_PHYS_DITTO) { | |
1064 | if (ddt_ditto_copies_needed(ddt, dde, NULL) == 0) | |
1065 | ddt_phys_free(ddt, ddk, ddp, txg); | |
1066 | continue; | |
1067 | } | |
1068 | if (ddp->ddp_refcnt == 0) | |
1069 | ddt_phys_free(ddt, ddk, ddp, txg); | |
1070 | total_refcnt += ddp->ddp_refcnt; | |
1071 | } | |
1072 | ||
1073 | if (dde->dde_phys[DDT_PHYS_DITTO].ddp_phys_birth != 0) | |
1074 | nclass = DDT_CLASS_DITTO; | |
1075 | else if (total_refcnt > 1) | |
1076 | nclass = DDT_CLASS_DUPLICATE; | |
1077 | else | |
1078 | nclass = DDT_CLASS_UNIQUE; | |
1079 | ||
1080 | if (otype != DDT_TYPES && | |
1081 | (otype != ntype || oclass != nclass || total_refcnt == 0)) { | |
1082 | VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0); | |
1083 | ASSERT(ddt_object_lookup(ddt, otype, oclass, dde) == ENOENT); | |
1084 | } | |
1085 | ||
1086 | if (total_refcnt != 0) { | |
1087 | dde->dde_type = ntype; | |
1088 | dde->dde_class = nclass; | |
1089 | ddt_stat_update(ddt, dde, 0); | |
1090 | if (!ddt_object_exists(ddt, ntype, nclass)) | |
1091 | ddt_object_create(ddt, ntype, nclass, tx); | |
1092 | VERIFY(ddt_object_update(ddt, ntype, nclass, dde, tx) == 0); | |
1093 | ||
1094 | /* | |
1095 | * If the class changes, the order that we scan this bp | |
1096 | * changes. If it decreases, we could miss it, so | |
1097 | * scan it right now. (This covers both class changing | |
1098 | * while we are doing ddt_walk(), and when we are | |
1099 | * traversing.) | |
1100 | */ | |
1101 | if (nclass < oclass) { | |
1102 | dsl_scan_ddt_entry(dp->dp_scan, | |
1103 | ddt->ddt_checksum, dde, tx); | |
1104 | } | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | static void | |
1109 | ddt_sync_table(ddt_t *ddt, dmu_tx_t *tx, uint64_t txg) | |
1110 | { | |
1111 | spa_t *spa = ddt->ddt_spa; | |
1112 | ddt_entry_t *dde; | |
1113 | void *cookie = NULL; | |
d6320ddb BB |
1114 | enum ddt_type type; |
1115 | enum ddt_class class; | |
428870ff BB |
1116 | |
1117 | if (avl_numnodes(&ddt->ddt_tree) == 0) | |
1118 | return; | |
1119 | ||
1120 | ASSERT(spa->spa_uberblock.ub_version >= SPA_VERSION_DEDUP); | |
1121 | ||
1122 | if (spa->spa_ddt_stat_object == 0) { | |
1123 | spa->spa_ddt_stat_object = zap_create(ddt->ddt_os, | |
1124 | DMU_OT_DDT_STATS, DMU_OT_NONE, 0, tx); | |
1125 | VERIFY(zap_add(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, | |
1126 | DMU_POOL_DDT_STATS, sizeof (uint64_t), 1, | |
1127 | &spa->spa_ddt_stat_object, tx) == 0); | |
1128 | } | |
1129 | ||
1130 | while ((dde = avl_destroy_nodes(&ddt->ddt_tree, &cookie)) != NULL) { | |
1131 | ddt_sync_entry(ddt, dde, tx, txg); | |
1132 | ddt_free(dde); | |
1133 | } | |
1134 | ||
d6320ddb | 1135 | for (type = 0; type < DDT_TYPES; type++) { |
e8fd45a0 | 1136 | uint64_t add, count = 0; |
d6320ddb | 1137 | for (class = 0; class < DDT_CLASSES; class++) { |
572e2857 BB |
1138 | if (ddt_object_exists(ddt, type, class)) { |
1139 | ddt_object_sync(ddt, type, class, tx); | |
e8fd45a0 BB |
1140 | VERIFY(ddt_object_count(ddt, type, class, |
1141 | &add) == 0); | |
1142 | count += add; | |
572e2857 BB |
1143 | } |
1144 | } | |
d6320ddb | 1145 | for (class = 0; class < DDT_CLASSES; class++) { |
572e2857 | 1146 | if (count == 0 && ddt_object_exists(ddt, type, class)) |
428870ff BB |
1147 | ddt_object_destroy(ddt, type, class, tx); |
1148 | } | |
1149 | } | |
1150 | ||
1151 | bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache, | |
1152 | sizeof (ddt->ddt_histogram)); | |
1153 | } | |
1154 | ||
1155 | void | |
1156 | ddt_sync(spa_t *spa, uint64_t txg) | |
1157 | { | |
1158 | dmu_tx_t *tx; | |
1159 | zio_t *rio = zio_root(spa, NULL, NULL, | |
1160 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
d6320ddb | 1161 | enum zio_checksum c; |
428870ff BB |
1162 | |
1163 | ASSERT(spa_syncing_txg(spa) == txg); | |
1164 | ||
1165 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
1166 | ||
d6320ddb | 1167 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff BB |
1168 | ddt_t *ddt = spa->spa_ddt[c]; |
1169 | if (ddt == NULL) | |
1170 | continue; | |
1171 | ddt_sync_table(ddt, tx, txg); | |
1172 | ddt_repair_table(ddt, rio); | |
1173 | } | |
1174 | ||
1175 | (void) zio_wait(rio); | |
1176 | ||
1177 | dmu_tx_commit(tx); | |
1178 | } | |
1179 | ||
1180 | int | |
1181 | ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, ddt_entry_t *dde) | |
1182 | { | |
1183 | do { | |
1184 | do { | |
1185 | do { | |
1186 | ddt_t *ddt = spa->spa_ddt[ddb->ddb_checksum]; | |
1187 | int error = ENOENT; | |
1188 | if (ddt_object_exists(ddt, ddb->ddb_type, | |
1189 | ddb->ddb_class)) { | |
1190 | error = ddt_object_walk(ddt, | |
1191 | ddb->ddb_type, ddb->ddb_class, | |
1192 | &ddb->ddb_cursor, dde); | |
1193 | } | |
1194 | dde->dde_type = ddb->ddb_type; | |
1195 | dde->dde_class = ddb->ddb_class; | |
1196 | if (error == 0) | |
1197 | return (0); | |
1198 | if (error != ENOENT) | |
1199 | return (error); | |
1200 | ddb->ddb_cursor = 0; | |
1201 | } while (++ddb->ddb_checksum < ZIO_CHECKSUM_FUNCTIONS); | |
1202 | ddb->ddb_checksum = 0; | |
1203 | } while (++ddb->ddb_type < DDT_TYPES); | |
1204 | ddb->ddb_type = 0; | |
1205 | } while (++ddb->ddb_class < DDT_CLASSES); | |
1206 | ||
1207 | return (ENOENT); | |
1208 | } | |
c409e464 BB |
1209 | |
1210 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
1211 | module_param(zfs_dedup_prefetch, int, 0644); | |
1212 | MODULE_PARM_DESC(zfs_dedup_prefetch,"Enable prefetching dedup-ed blks"); | |
1213 | #endif |