]>
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. | |
ea04106b | 24 | * Copyright (c) 2012, 2014 by Delphix. All rights reserved. |
428870ff BB |
25 | */ |
26 | ||
27 | #include <sys/zfs_context.h> | |
28 | #include <sys/spa.h> | |
29 | #include <sys/spa_impl.h> | |
30 | #include <sys/zio.h> | |
31 | #include <sys/ddt.h> | |
32 | #include <sys/zap.h> | |
33 | #include <sys/dmu_tx.h> | |
34 | #include <sys/arc.h> | |
35 | #include <sys/dsl_pool.h> | |
36 | #include <sys/zio_checksum.h> | |
37 | #include <sys/zio_compress.h> | |
38 | #include <sys/dsl_scan.h> | |
39 | ||
a08ee875 LG |
40 | static kmem_cache_t *ddt_cache; |
41 | static kmem_cache_t *ddt_entry_cache; | |
42 | ||
572e2857 BB |
43 | /* |
44 | * Enable/disable prefetching of dedup-ed blocks which are going to be freed. | |
45 | */ | |
ea04106b | 46 | int zfs_dedup_prefetch = 0; |
572e2857 | 47 | |
428870ff BB |
48 | static const ddt_ops_t *ddt_ops[DDT_TYPES] = { |
49 | &ddt_zap_ops, | |
50 | }; | |
51 | ||
52 | static const char *ddt_class_name[DDT_CLASSES] = { | |
53 | "ditto", | |
54 | "duplicate", | |
55 | "unique", | |
56 | }; | |
57 | ||
58 | static void | |
59 | ddt_object_create(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
60 | dmu_tx_t *tx) | |
61 | { | |
62 | spa_t *spa = ddt->ddt_spa; | |
63 | objset_t *os = ddt->ddt_os; | |
64 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
65 | boolean_t prehash = zio_checksum_table[ddt->ddt_checksum].ci_dedup; | |
66 | char name[DDT_NAMELEN]; | |
67 | ||
68 | ddt_object_name(ddt, type, class, name); | |
69 | ||
70 | ASSERT(*objectp == 0); | |
71 | VERIFY(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash) == 0); | |
72 | ASSERT(*objectp != 0); | |
73 | ||
74 | VERIFY(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name, | |
75 | sizeof (uint64_t), 1, objectp, tx) == 0); | |
76 | ||
77 | VERIFY(zap_add(os, spa->spa_ddt_stat_object, name, | |
78 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
79 | &ddt->ddt_histogram[type][class], tx) == 0); | |
80 | } | |
81 | ||
82 | static void | |
83 | ddt_object_destroy(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
84 | dmu_tx_t *tx) | |
85 | { | |
86 | spa_t *spa = ddt->ddt_spa; | |
87 | objset_t *os = ddt->ddt_os; | |
88 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
e8fd45a0 | 89 | uint64_t count; |
428870ff BB |
90 | char name[DDT_NAMELEN]; |
91 | ||
92 | ddt_object_name(ddt, type, class, name); | |
93 | ||
94 | ASSERT(*objectp != 0); | |
428870ff | 95 | ASSERT(ddt_histogram_empty(&ddt->ddt_histogram[type][class])); |
e8fd45a0 | 96 | VERIFY(ddt_object_count(ddt, type, class, &count) == 0 && count == 0); |
428870ff BB |
97 | VERIFY(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx) == 0); |
98 | VERIFY(zap_remove(os, spa->spa_ddt_stat_object, name, tx) == 0); | |
99 | VERIFY(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx) == 0); | |
100 | bzero(&ddt->ddt_object_stats[type][class], sizeof (ddt_object_t)); | |
101 | ||
102 | *objectp = 0; | |
103 | } | |
104 | ||
105 | static int | |
106 | ddt_object_load(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
107 | { | |
108 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
109 | dmu_object_info_t doi; | |
e8fd45a0 | 110 | uint64_t count; |
428870ff BB |
111 | char name[DDT_NAMELEN]; |
112 | int error; | |
113 | ||
114 | ddt_object_name(ddt, type, class, name); | |
115 | ||
116 | error = zap_lookup(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, name, | |
117 | sizeof (uint64_t), 1, &ddt->ddt_object[type][class]); | |
118 | ||
ea04106b | 119 | if (error != 0) |
428870ff BB |
120 | return (error); |
121 | ||
ea04106b | 122 | VERIFY0(zap_lookup(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, |
428870ff | 123 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), |
ea04106b | 124 | &ddt->ddt_histogram[type][class])); |
428870ff BB |
125 | |
126 | /* | |
127 | * Seed the cached statistics. | |
128 | */ | |
2a4a9dc2 BB |
129 | error = ddt_object_info(ddt, type, class, &doi); |
130 | if (error) | |
131 | return (error); | |
428870ff | 132 | |
e8fd45a0 BB |
133 | error = ddt_object_count(ddt, type, class, &count); |
134 | if (error) | |
135 | return (error); | |
136 | ||
137 | ddo->ddo_count = count; | |
428870ff BB |
138 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
139 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
140 | ||
ea04106b | 141 | return (0); |
428870ff BB |
142 | } |
143 | ||
144 | static void | |
145 | ddt_object_sync(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
146 | dmu_tx_t *tx) | |
147 | { | |
148 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
149 | dmu_object_info_t doi; | |
e8fd45a0 | 150 | uint64_t count; |
428870ff BB |
151 | char name[DDT_NAMELEN]; |
152 | ||
153 | ddt_object_name(ddt, type, class, name); | |
154 | ||
155 | VERIFY(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, | |
156 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
157 | &ddt->ddt_histogram[type][class], tx) == 0); | |
158 | ||
159 | /* | |
160 | * Cache DDT statistics; this is the only time they'll change. | |
161 | */ | |
162 | VERIFY(ddt_object_info(ddt, type, class, &doi) == 0); | |
e8fd45a0 | 163 | VERIFY(ddt_object_count(ddt, type, class, &count) == 0); |
428870ff | 164 | |
e8fd45a0 | 165 | ddo->ddo_count = count; |
428870ff BB |
166 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
167 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
168 | } | |
169 | ||
170 | static int | |
171 | ddt_object_lookup(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
172 | ddt_entry_t *dde) | |
173 | { | |
174 | if (!ddt_object_exists(ddt, type, class)) | |
a08ee875 | 175 | return (SET_ERROR(ENOENT)); |
428870ff BB |
176 | |
177 | return (ddt_ops[type]->ddt_op_lookup(ddt->ddt_os, | |
178 | ddt->ddt_object[type][class], dde)); | |
179 | } | |
180 | ||
181 | static void | |
182 | ddt_object_prefetch(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
183 | ddt_entry_t *dde) | |
184 | { | |
185 | if (!ddt_object_exists(ddt, type, class)) | |
186 | return; | |
187 | ||
188 | ddt_ops[type]->ddt_op_prefetch(ddt->ddt_os, | |
189 | ddt->ddt_object[type][class], dde); | |
190 | } | |
191 | ||
192 | int | |
193 | ddt_object_update(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
194 | ddt_entry_t *dde, dmu_tx_t *tx) | |
195 | { | |
196 | ASSERT(ddt_object_exists(ddt, type, class)); | |
197 | ||
198 | return (ddt_ops[type]->ddt_op_update(ddt->ddt_os, | |
199 | ddt->ddt_object[type][class], dde, tx)); | |
200 | } | |
201 | ||
202 | static int | |
203 | ddt_object_remove(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
204 | ddt_entry_t *dde, dmu_tx_t *tx) | |
205 | { | |
206 | ASSERT(ddt_object_exists(ddt, type, class)); | |
207 | ||
208 | return (ddt_ops[type]->ddt_op_remove(ddt->ddt_os, | |
209 | ddt->ddt_object[type][class], dde, tx)); | |
210 | } | |
211 | ||
212 | int | |
213 | ddt_object_walk(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
214 | uint64_t *walk, ddt_entry_t *dde) | |
215 | { | |
216 | ASSERT(ddt_object_exists(ddt, type, class)); | |
217 | ||
218 | return (ddt_ops[type]->ddt_op_walk(ddt->ddt_os, | |
219 | ddt->ddt_object[type][class], dde, walk)); | |
220 | } | |
221 | ||
e8fd45a0 BB |
222 | int |
223 | ddt_object_count(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
224 | uint64_t *count) | |
428870ff BB |
225 | { |
226 | ASSERT(ddt_object_exists(ddt, type, class)); | |
227 | ||
228 | return (ddt_ops[type]->ddt_op_count(ddt->ddt_os, | |
e8fd45a0 | 229 | ddt->ddt_object[type][class], count)); |
428870ff BB |
230 | } |
231 | ||
232 | int | |
233 | ddt_object_info(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
234 | dmu_object_info_t *doi) | |
235 | { | |
236 | if (!ddt_object_exists(ddt, type, class)) | |
a08ee875 | 237 | return (SET_ERROR(ENOENT)); |
428870ff BB |
238 | |
239 | return (dmu_object_info(ddt->ddt_os, ddt->ddt_object[type][class], | |
240 | doi)); | |
241 | } | |
242 | ||
243 | boolean_t | |
244 | ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
245 | { | |
246 | return (!!ddt->ddt_object[type][class]); | |
247 | } | |
248 | ||
249 | void | |
250 | ddt_object_name(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
251 | char *name) | |
252 | { | |
253 | (void) sprintf(name, DMU_POOL_DDT, | |
254 | zio_checksum_table[ddt->ddt_checksum].ci_name, | |
255 | ddt_ops[type]->ddt_op_name, ddt_class_name[class]); | |
256 | } | |
257 | ||
258 | void | |
259 | ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, uint64_t txg) | |
260 | { | |
d6320ddb | 261 | int d; |
428870ff BB |
262 | ASSERT(txg != 0); |
263 | ||
d6320ddb | 264 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
265 | bp->blk_dva[d] = ddp->ddp_dva[d]; |
266 | BP_SET_BIRTH(bp, txg, ddp->ddp_phys_birth); | |
267 | } | |
268 | ||
269 | void | |
270 | ddt_bp_create(enum zio_checksum checksum, | |
271 | const ddt_key_t *ddk, const ddt_phys_t *ddp, blkptr_t *bp) | |
272 | { | |
273 | BP_ZERO(bp); | |
274 | ||
275 | if (ddp != NULL) | |
276 | ddt_bp_fill(ddp, bp, ddp->ddp_phys_birth); | |
277 | ||
278 | bp->blk_cksum = ddk->ddk_cksum; | |
279 | bp->blk_fill = 1; | |
280 | ||
281 | BP_SET_LSIZE(bp, DDK_GET_LSIZE(ddk)); | |
282 | BP_SET_PSIZE(bp, DDK_GET_PSIZE(ddk)); | |
283 | BP_SET_COMPRESS(bp, DDK_GET_COMPRESS(ddk)); | |
284 | BP_SET_CHECKSUM(bp, checksum); | |
285 | BP_SET_TYPE(bp, DMU_OT_DEDUP); | |
286 | BP_SET_LEVEL(bp, 0); | |
287 | BP_SET_DEDUP(bp, 0); | |
288 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); | |
289 | } | |
290 | ||
291 | void | |
292 | ddt_key_fill(ddt_key_t *ddk, const blkptr_t *bp) | |
293 | { | |
294 | ddk->ddk_cksum = bp->blk_cksum; | |
295 | ddk->ddk_prop = 0; | |
296 | ||
297 | DDK_SET_LSIZE(ddk, BP_GET_LSIZE(bp)); | |
298 | DDK_SET_PSIZE(ddk, BP_GET_PSIZE(bp)); | |
299 | DDK_SET_COMPRESS(ddk, BP_GET_COMPRESS(bp)); | |
300 | } | |
301 | ||
302 | void | |
303 | ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp) | |
304 | { | |
d6320ddb | 305 | int d; |
428870ff BB |
306 | ASSERT(ddp->ddp_phys_birth == 0); |
307 | ||
d6320ddb | 308 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
309 | ddp->ddp_dva[d] = bp->blk_dva[d]; |
310 | ddp->ddp_phys_birth = BP_PHYSICAL_BIRTH(bp); | |
311 | } | |
312 | ||
313 | void | |
314 | ddt_phys_clear(ddt_phys_t *ddp) | |
315 | { | |
316 | bzero(ddp, sizeof (*ddp)); | |
317 | } | |
318 | ||
319 | void | |
320 | ddt_phys_addref(ddt_phys_t *ddp) | |
321 | { | |
322 | ddp->ddp_refcnt++; | |
323 | } | |
324 | ||
325 | void | |
326 | ddt_phys_decref(ddt_phys_t *ddp) | |
327 | { | |
5dc6af0e | 328 | if (ddp) { |
c06d4368 | 329 | ASSERT(ddp->ddp_refcnt > 0); |
5dc6af0e BB |
330 | ddp->ddp_refcnt--; |
331 | } | |
428870ff BB |
332 | } |
333 | ||
334 | void | |
335 | ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, uint64_t txg) | |
336 | { | |
337 | blkptr_t blk; | |
338 | ||
339 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
340 | ddt_phys_clear(ddp); | |
341 | zio_free(ddt->ddt_spa, txg, &blk); | |
342 | } | |
343 | ||
344 | ddt_phys_t * | |
345 | ddt_phys_select(const ddt_entry_t *dde, const blkptr_t *bp) | |
346 | { | |
347 | ddt_phys_t *ddp = (ddt_phys_t *)dde->dde_phys; | |
d6320ddb | 348 | int p; |
428870ff | 349 | |
d6320ddb | 350 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
351 | if (DVA_EQUAL(BP_IDENTITY(bp), &ddp->ddp_dva[0]) && |
352 | BP_PHYSICAL_BIRTH(bp) == ddp->ddp_phys_birth) | |
353 | return (ddp); | |
354 | } | |
355 | return (NULL); | |
356 | } | |
357 | ||
358 | uint64_t | |
359 | ddt_phys_total_refcnt(const ddt_entry_t *dde) | |
360 | { | |
361 | uint64_t refcnt = 0; | |
d6320ddb | 362 | int p; |
428870ff | 363 | |
d6320ddb | 364 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) |
428870ff BB |
365 | refcnt += dde->dde_phys[p].ddp_refcnt; |
366 | ||
367 | return (refcnt); | |
368 | } | |
369 | ||
370 | static void | |
371 | ddt_stat_generate(ddt_t *ddt, ddt_entry_t *dde, ddt_stat_t *dds) | |
372 | { | |
373 | spa_t *spa = ddt->ddt_spa; | |
374 | ddt_phys_t *ddp = dde->dde_phys; | |
375 | ddt_key_t *ddk = &dde->dde_key; | |
376 | uint64_t lsize = DDK_GET_LSIZE(ddk); | |
377 | uint64_t psize = DDK_GET_PSIZE(ddk); | |
d6320ddb | 378 | int p, d; |
428870ff BB |
379 | |
380 | bzero(dds, sizeof (*dds)); | |
381 | ||
d6320ddb | 382 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
383 | uint64_t dsize = 0; |
384 | uint64_t refcnt = ddp->ddp_refcnt; | |
385 | ||
386 | if (ddp->ddp_phys_birth == 0) | |
387 | continue; | |
388 | ||
d6320ddb | 389 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
390 | dsize += dva_get_dsize_sync(spa, &ddp->ddp_dva[d]); |
391 | ||
392 | dds->dds_blocks += 1; | |
393 | dds->dds_lsize += lsize; | |
394 | dds->dds_psize += psize; | |
395 | dds->dds_dsize += dsize; | |
396 | ||
397 | dds->dds_ref_blocks += refcnt; | |
398 | dds->dds_ref_lsize += lsize * refcnt; | |
399 | dds->dds_ref_psize += psize * refcnt; | |
400 | dds->dds_ref_dsize += dsize * refcnt; | |
401 | } | |
402 | } | |
403 | ||
404 | void | |
405 | ddt_stat_add(ddt_stat_t *dst, const ddt_stat_t *src, uint64_t neg) | |
406 | { | |
407 | const uint64_t *s = (const uint64_t *)src; | |
408 | uint64_t *d = (uint64_t *)dst; | |
409 | uint64_t *d_end = (uint64_t *)(dst + 1); | |
410 | ||
411 | ASSERT(neg == 0 || neg == -1ULL); /* add or subtract */ | |
412 | ||
413 | while (d < d_end) | |
414 | *d++ += (*s++ ^ neg) - neg; | |
415 | } | |
416 | ||
417 | static void | |
418 | ddt_stat_update(ddt_t *ddt, ddt_entry_t *dde, uint64_t neg) | |
419 | { | |
420 | ddt_stat_t dds; | |
421 | ddt_histogram_t *ddh; | |
422 | int bucket; | |
423 | ||
424 | ddt_stat_generate(ddt, dde, &dds); | |
425 | ||
ea04106b | 426 | bucket = highbit64(dds.dds_ref_blocks) - 1; |
428870ff BB |
427 | ASSERT(bucket >= 0); |
428 | ||
429 | ddh = &ddt->ddt_histogram[dde->dde_type][dde->dde_class]; | |
430 | ||
431 | ddt_stat_add(&ddh->ddh_stat[bucket], &dds, neg); | |
432 | } | |
433 | ||
434 | void | |
435 | ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src) | |
436 | { | |
d6320ddb BB |
437 | int h; |
438 | ||
439 | for (h = 0; h < 64; h++) | |
428870ff BB |
440 | ddt_stat_add(&dst->ddh_stat[h], &src->ddh_stat[h], 0); |
441 | } | |
442 | ||
443 | void | |
444 | ddt_histogram_stat(ddt_stat_t *dds, const ddt_histogram_t *ddh) | |
445 | { | |
d6320ddb BB |
446 | int h; |
447 | ||
428870ff BB |
448 | bzero(dds, sizeof (*dds)); |
449 | ||
d6320ddb | 450 | for (h = 0; h < 64; h++) |
428870ff BB |
451 | ddt_stat_add(dds, &ddh->ddh_stat[h], 0); |
452 | } | |
453 | ||
454 | boolean_t | |
455 | ddt_histogram_empty(const ddt_histogram_t *ddh) | |
456 | { | |
457 | const uint64_t *s = (const uint64_t *)ddh; | |
458 | const uint64_t *s_end = (const uint64_t *)(ddh + 1); | |
459 | ||
460 | while (s < s_end) | |
461 | if (*s++ != 0) | |
462 | return (B_FALSE); | |
463 | ||
464 | return (B_TRUE); | |
465 | } | |
466 | ||
467 | void | |
468 | ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo_total) | |
469 | { | |
d6320ddb BB |
470 | enum zio_checksum c; |
471 | enum ddt_type type; | |
472 | enum ddt_class class; | |
473 | ||
428870ff | 474 | /* Sum the statistics we cached in ddt_object_sync(). */ |
d6320ddb | 475 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 476 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
477 | for (type = 0; type < DDT_TYPES; type++) { |
478 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
479 | class++) { |
480 | ddt_object_t *ddo = | |
481 | &ddt->ddt_object_stats[type][class]; | |
482 | ddo_total->ddo_count += ddo->ddo_count; | |
483 | ddo_total->ddo_dspace += ddo->ddo_dspace; | |
484 | ddo_total->ddo_mspace += ddo->ddo_mspace; | |
485 | } | |
486 | } | |
487 | } | |
488 | ||
489 | /* ... and compute the averages. */ | |
490 | if (ddo_total->ddo_count != 0) { | |
491 | ddo_total->ddo_dspace /= ddo_total->ddo_count; | |
492 | ddo_total->ddo_mspace /= ddo_total->ddo_count; | |
428870ff BB |
493 | } |
494 | } | |
495 | ||
496 | void | |
497 | ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh) | |
498 | { | |
d6320ddb BB |
499 | enum zio_checksum c; |
500 | enum ddt_type type; | |
501 | enum ddt_class class; | |
502 | ||
503 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { | |
428870ff | 504 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
505 | for (type = 0; type < DDT_TYPES; type++) { |
506 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
507 | class++) { |
508 | ddt_histogram_add(ddh, | |
509 | &ddt->ddt_histogram_cache[type][class]); | |
510 | } | |
511 | } | |
512 | } | |
513 | } | |
514 | ||
515 | void | |
516 | ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total) | |
517 | { | |
518 | ddt_histogram_t *ddh_total; | |
519 | ||
ea04106b | 520 | ddh_total = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP); |
428870ff BB |
521 | ddt_get_dedup_histogram(spa, ddh_total); |
522 | ddt_histogram_stat(dds_total, ddh_total); | |
523 | kmem_free(ddh_total, sizeof (ddt_histogram_t)); | |
524 | } | |
525 | ||
526 | uint64_t | |
527 | ddt_get_dedup_dspace(spa_t *spa) | |
528 | { | |
529 | ddt_stat_t dds_total = { 0 }; | |
530 | ||
531 | ddt_get_dedup_stats(spa, &dds_total); | |
532 | return (dds_total.dds_ref_dsize - dds_total.dds_dsize); | |
533 | } | |
534 | ||
535 | uint64_t | |
536 | ddt_get_pool_dedup_ratio(spa_t *spa) | |
537 | { | |
538 | ddt_stat_t dds_total = { 0 }; | |
539 | ||
540 | ddt_get_dedup_stats(spa, &dds_total); | |
541 | if (dds_total.dds_dsize == 0) | |
542 | return (100); | |
543 | ||
544 | return (dds_total.dds_ref_dsize * 100 / dds_total.dds_dsize); | |
545 | } | |
546 | ||
547 | int | |
548 | ddt_ditto_copies_needed(ddt_t *ddt, ddt_entry_t *dde, ddt_phys_t *ddp_willref) | |
549 | { | |
550 | spa_t *spa = ddt->ddt_spa; | |
551 | uint64_t total_refcnt = 0; | |
552 | uint64_t ditto = spa->spa_dedup_ditto; | |
553 | int total_copies = 0; | |
554 | int desired_copies = 0; | |
d6320ddb | 555 | int p; |
428870ff | 556 | |
d6320ddb | 557 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
558 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
559 | zio_t *zio = dde->dde_lead_zio[p]; | |
560 | uint64_t refcnt = ddp->ddp_refcnt; /* committed refs */ | |
561 | if (zio != NULL) | |
562 | refcnt += zio->io_parent_count; /* pending refs */ | |
563 | if (ddp == ddp_willref) | |
564 | refcnt++; /* caller's ref */ | |
565 | if (refcnt != 0) { | |
566 | total_refcnt += refcnt; | |
567 | total_copies += p; | |
568 | } | |
569 | } | |
570 | ||
571 | if (ditto == 0 || ditto > UINT32_MAX) | |
572 | ditto = UINT32_MAX; | |
573 | ||
574 | if (total_refcnt >= 1) | |
575 | desired_copies++; | |
576 | if (total_refcnt >= ditto) | |
577 | desired_copies++; | |
578 | if (total_refcnt >= ditto * ditto) | |
579 | desired_copies++; | |
580 | ||
581 | return (MAX(desired_copies, total_copies) - total_copies); | |
582 | } | |
583 | ||
584 | int | |
585 | ddt_ditto_copies_present(ddt_entry_t *dde) | |
586 | { | |
587 | ddt_phys_t *ddp = &dde->dde_phys[DDT_PHYS_DITTO]; | |
588 | dva_t *dva = ddp->ddp_dva; | |
589 | int copies = 0 - DVA_GET_GANG(dva); | |
d6320ddb | 590 | int d; |
428870ff | 591 | |
d6320ddb | 592 | for (d = 0; d < SPA_DVAS_PER_BP; d++, dva++) |
428870ff BB |
593 | if (DVA_IS_VALID(dva)) |
594 | copies++; | |
595 | ||
596 | ASSERT(copies >= 0 && copies < SPA_DVAS_PER_BP); | |
597 | ||
598 | return (copies); | |
599 | } | |
600 | ||
601 | size_t | |
602 | ddt_compress(void *src, uchar_t *dst, size_t s_len, size_t d_len) | |
603 | { | |
604 | uchar_t *version = dst++; | |
605 | int cpfunc = ZIO_COMPRESS_ZLE; | |
606 | zio_compress_info_t *ci = &zio_compress_table[cpfunc]; | |
607 | size_t c_len; | |
608 | ||
609 | ASSERT(d_len >= s_len + 1); /* no compression plus version byte */ | |
610 | ||
611 | c_len = ci->ci_compress(src, dst, s_len, d_len - 1, ci->ci_level); | |
612 | ||
613 | if (c_len == s_len) { | |
614 | cpfunc = ZIO_COMPRESS_OFF; | |
615 | bcopy(src, dst, s_len); | |
616 | } | |
617 | ||
ea04106b AX |
618 | *version = cpfunc; |
619 | /* CONSTCOND */ | |
620 | if (ZFS_HOST_BYTEORDER) | |
621 | *version |= DDT_COMPRESS_BYTEORDER_MASK; | |
428870ff BB |
622 | |
623 | return (c_len + 1); | |
624 | } | |
625 | ||
626 | void | |
627 | ddt_decompress(uchar_t *src, void *dst, size_t s_len, size_t d_len) | |
628 | { | |
629 | uchar_t version = *src++; | |
630 | int cpfunc = version & DDT_COMPRESS_FUNCTION_MASK; | |
631 | zio_compress_info_t *ci = &zio_compress_table[cpfunc]; | |
632 | ||
633 | if (ci->ci_decompress != NULL) | |
634 | (void) ci->ci_decompress(src, dst, s_len, d_len, ci->ci_level); | |
635 | else | |
636 | bcopy(src, dst, d_len); | |
637 | ||
ea04106b AX |
638 | if (((version & DDT_COMPRESS_BYTEORDER_MASK) != 0) != |
639 | (ZFS_HOST_BYTEORDER != 0)) | |
428870ff BB |
640 | byteswap_uint64_array(dst, d_len); |
641 | } | |
642 | ||
643 | ddt_t * | |
644 | ddt_select_by_checksum(spa_t *spa, enum zio_checksum c) | |
645 | { | |
646 | return (spa->spa_ddt[c]); | |
647 | } | |
648 | ||
649 | ddt_t * | |
650 | ddt_select(spa_t *spa, const blkptr_t *bp) | |
651 | { | |
652 | return (spa->spa_ddt[BP_GET_CHECKSUM(bp)]); | |
653 | } | |
654 | ||
655 | void | |
656 | ddt_enter(ddt_t *ddt) | |
657 | { | |
658 | mutex_enter(&ddt->ddt_lock); | |
659 | } | |
660 | ||
661 | void | |
662 | ddt_exit(ddt_t *ddt) | |
663 | { | |
664 | mutex_exit(&ddt->ddt_lock); | |
665 | } | |
666 | ||
a08ee875 LG |
667 | void |
668 | ddt_init(void) | |
669 | { | |
670 | ddt_cache = kmem_cache_create("ddt_cache", | |
671 | sizeof (ddt_t), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
672 | ddt_entry_cache = kmem_cache_create("ddt_entry_cache", | |
673 | sizeof (ddt_entry_t), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
674 | } | |
675 | ||
676 | void | |
677 | ddt_fini(void) | |
678 | { | |
679 | kmem_cache_destroy(ddt_entry_cache); | |
680 | kmem_cache_destroy(ddt_cache); | |
681 | } | |
682 | ||
428870ff BB |
683 | static ddt_entry_t * |
684 | ddt_alloc(const ddt_key_t *ddk) | |
685 | { | |
686 | ddt_entry_t *dde; | |
687 | ||
ea04106b | 688 | dde = kmem_cache_alloc(ddt_entry_cache, KM_SLEEP); |
a08ee875 | 689 | bzero(dde, sizeof (ddt_entry_t)); |
428870ff BB |
690 | cv_init(&dde->dde_cv, NULL, CV_DEFAULT, NULL); |
691 | ||
692 | dde->dde_key = *ddk; | |
693 | ||
694 | return (dde); | |
695 | } | |
696 | ||
697 | static void | |
698 | ddt_free(ddt_entry_t *dde) | |
699 | { | |
d6320ddb BB |
700 | int p; |
701 | ||
428870ff BB |
702 | ASSERT(!dde->dde_loading); |
703 | ||
d6320ddb | 704 | for (p = 0; p < DDT_PHYS_TYPES; p++) |
428870ff BB |
705 | ASSERT(dde->dde_lead_zio[p] == NULL); |
706 | ||
707 | if (dde->dde_repair_data != NULL) | |
708 | zio_buf_free(dde->dde_repair_data, | |
709 | DDK_GET_PSIZE(&dde->dde_key)); | |
710 | ||
711 | cv_destroy(&dde->dde_cv); | |
a08ee875 | 712 | kmem_cache_free(ddt_entry_cache, dde); |
428870ff BB |
713 | } |
714 | ||
715 | void | |
716 | ddt_remove(ddt_t *ddt, ddt_entry_t *dde) | |
717 | { | |
718 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
719 | ||
720 | avl_remove(&ddt->ddt_tree, dde); | |
721 | ddt_free(dde); | |
722 | } | |
723 | ||
724 | ddt_entry_t * | |
725 | ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add) | |
726 | { | |
727 | ddt_entry_t *dde, dde_search; | |
728 | enum ddt_type type; | |
729 | enum ddt_class class; | |
730 | avl_index_t where; | |
731 | int error; | |
732 | ||
733 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
734 | ||
735 | ddt_key_fill(&dde_search.dde_key, bp); | |
736 | ||
737 | dde = avl_find(&ddt->ddt_tree, &dde_search, &where); | |
738 | if (dde == NULL) { | |
739 | if (!add) | |
740 | return (NULL); | |
741 | dde = ddt_alloc(&dde_search.dde_key); | |
742 | avl_insert(&ddt->ddt_tree, dde, where); | |
743 | } | |
744 | ||
745 | while (dde->dde_loading) | |
746 | cv_wait(&dde->dde_cv, &ddt->ddt_lock); | |
747 | ||
748 | if (dde->dde_loaded) | |
749 | return (dde); | |
750 | ||
751 | dde->dde_loading = B_TRUE; | |
752 | ||
753 | ddt_exit(ddt); | |
754 | ||
755 | error = ENOENT; | |
756 | ||
757 | for (type = 0; type < DDT_TYPES; type++) { | |
758 | for (class = 0; class < DDT_CLASSES; class++) { | |
759 | error = ddt_object_lookup(ddt, type, class, dde); | |
760 | if (error != ENOENT) | |
761 | break; | |
762 | } | |
763 | if (error != ENOENT) | |
764 | break; | |
765 | } | |
766 | ||
767 | ASSERT(error == 0 || error == ENOENT); | |
768 | ||
769 | ddt_enter(ddt); | |
770 | ||
771 | ASSERT(dde->dde_loaded == B_FALSE); | |
772 | ASSERT(dde->dde_loading == B_TRUE); | |
773 | ||
774 | dde->dde_type = type; /* will be DDT_TYPES if no entry found */ | |
775 | dde->dde_class = class; /* will be DDT_CLASSES if no entry found */ | |
776 | dde->dde_loaded = B_TRUE; | |
777 | dde->dde_loading = B_FALSE; | |
778 | ||
779 | if (error == 0) | |
780 | ddt_stat_update(ddt, dde, -1ULL); | |
781 | ||
782 | cv_broadcast(&dde->dde_cv); | |
783 | ||
784 | return (dde); | |
785 | } | |
786 | ||
787 | void | |
788 | ddt_prefetch(spa_t *spa, const blkptr_t *bp) | |
789 | { | |
790 | ddt_t *ddt; | |
791 | ddt_entry_t dde; | |
d6320ddb BB |
792 | enum ddt_type type; |
793 | enum ddt_class class; | |
428870ff | 794 | |
572e2857 | 795 | if (!zfs_dedup_prefetch || bp == NULL || !BP_GET_DEDUP(bp)) |
428870ff BB |
796 | return; |
797 | ||
798 | /* | |
572e2857 BB |
799 | * We only remove the DDT once all tables are empty and only |
800 | * prefetch dedup blocks when there are entries in the DDT. | |
801 | * Thus no locking is required as the DDT can't disappear on us. | |
428870ff BB |
802 | */ |
803 | ddt = ddt_select(spa, bp); | |
804 | ddt_key_fill(&dde.dde_key, bp); | |
805 | ||
d6320ddb BB |
806 | for (type = 0; type < DDT_TYPES; type++) { |
807 | for (class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
808 | ddt_object_prefetch(ddt, type, class, &dde); |
809 | } | |
810 | } | |
811 | } | |
812 | ||
813 | int | |
814 | ddt_entry_compare(const void *x1, const void *x2) | |
815 | { | |
816 | const ddt_entry_t *dde1 = x1; | |
817 | const ddt_entry_t *dde2 = x2; | |
818 | const uint64_t *u1 = (const uint64_t *)&dde1->dde_key; | |
819 | const uint64_t *u2 = (const uint64_t *)&dde2->dde_key; | |
d6320ddb | 820 | int i; |
428870ff | 821 | |
d6320ddb | 822 | for (i = 0; i < DDT_KEY_WORDS; i++) { |
428870ff BB |
823 | if (u1[i] < u2[i]) |
824 | return (-1); | |
825 | if (u1[i] > u2[i]) | |
826 | return (1); | |
827 | } | |
828 | ||
829 | return (0); | |
830 | } | |
831 | ||
832 | static ddt_t * | |
833 | ddt_table_alloc(spa_t *spa, enum zio_checksum c) | |
834 | { | |
835 | ddt_t *ddt; | |
836 | ||
ea04106b | 837 | ddt = kmem_cache_alloc(ddt_cache, KM_SLEEP); |
a08ee875 | 838 | bzero(ddt, sizeof (ddt_t)); |
428870ff BB |
839 | |
840 | mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL); | |
841 | avl_create(&ddt->ddt_tree, ddt_entry_compare, | |
842 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); | |
843 | avl_create(&ddt->ddt_repair_tree, ddt_entry_compare, | |
844 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); | |
845 | ddt->ddt_checksum = c; | |
846 | ddt->ddt_spa = spa; | |
847 | ddt->ddt_os = spa->spa_meta_objset; | |
848 | ||
849 | return (ddt); | |
850 | } | |
851 | ||
852 | static void | |
853 | ddt_table_free(ddt_t *ddt) | |
854 | { | |
855 | ASSERT(avl_numnodes(&ddt->ddt_tree) == 0); | |
856 | ASSERT(avl_numnodes(&ddt->ddt_repair_tree) == 0); | |
857 | avl_destroy(&ddt->ddt_tree); | |
858 | avl_destroy(&ddt->ddt_repair_tree); | |
859 | mutex_destroy(&ddt->ddt_lock); | |
a08ee875 | 860 | kmem_cache_free(ddt_cache, ddt); |
428870ff BB |
861 | } |
862 | ||
863 | void | |
864 | ddt_create(spa_t *spa) | |
865 | { | |
d6320ddb BB |
866 | enum zio_checksum c; |
867 | ||
428870ff BB |
868 | spa->spa_dedup_checksum = ZIO_DEDUPCHECKSUM; |
869 | ||
d6320ddb | 870 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) |
428870ff BB |
871 | spa->spa_ddt[c] = ddt_table_alloc(spa, c); |
872 | } | |
873 | ||
874 | int | |
875 | ddt_load(spa_t *spa) | |
876 | { | |
d6320ddb BB |
877 | enum zio_checksum c; |
878 | enum ddt_type type; | |
879 | enum ddt_class class; | |
428870ff BB |
880 | int error; |
881 | ||
882 | ddt_create(spa); | |
883 | ||
884 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
885 | DMU_POOL_DDT_STATS, sizeof (uint64_t), 1, | |
886 | &spa->spa_ddt_stat_object); | |
887 | ||
888 | if (error) | |
889 | return (error == ENOENT ? 0 : error); | |
890 | ||
d6320ddb | 891 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 892 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
893 | for (type = 0; type < DDT_TYPES; type++) { |
894 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
895 | class++) { |
896 | error = ddt_object_load(ddt, type, class); | |
897 | if (error != 0 && error != ENOENT) | |
898 | return (error); | |
899 | } | |
900 | } | |
901 | ||
902 | /* | |
903 | * Seed the cached histograms. | |
904 | */ | |
905 | bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache, | |
906 | sizeof (ddt->ddt_histogram)); | |
907 | } | |
908 | ||
909 | return (0); | |
910 | } | |
911 | ||
912 | void | |
913 | ddt_unload(spa_t *spa) | |
914 | { | |
d6320ddb BB |
915 | enum zio_checksum c; |
916 | ||
917 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { | |
428870ff BB |
918 | if (spa->spa_ddt[c]) { |
919 | ddt_table_free(spa->spa_ddt[c]); | |
920 | spa->spa_ddt[c] = NULL; | |
921 | } | |
922 | } | |
923 | } | |
924 | ||
925 | boolean_t | |
926 | ddt_class_contains(spa_t *spa, enum ddt_class max_class, const blkptr_t *bp) | |
927 | { | |
928 | ddt_t *ddt; | |
e95b3bdc | 929 | ddt_entry_t *dde; |
d6320ddb BB |
930 | enum ddt_type type; |
931 | enum ddt_class class; | |
428870ff BB |
932 | |
933 | if (!BP_GET_DEDUP(bp)) | |
934 | return (B_FALSE); | |
935 | ||
936 | if (max_class == DDT_CLASS_UNIQUE) | |
937 | return (B_TRUE); | |
938 | ||
939 | ddt = spa->spa_ddt[BP_GET_CHECKSUM(bp)]; | |
ea04106b | 940 | dde = kmem_cache_alloc(ddt_entry_cache, KM_SLEEP); |
428870ff | 941 | |
e95b3bdc | 942 | ddt_key_fill(&(dde->dde_key), bp); |
428870ff | 943 | |
e95b3bdc BB |
944 | for (type = 0; type < DDT_TYPES; type++) { |
945 | for (class = 0; class <= max_class; class++) { | |
946 | if (ddt_object_lookup(ddt, type, class, dde) == 0) { | |
a08ee875 | 947 | kmem_cache_free(ddt_entry_cache, dde); |
428870ff | 948 | return (B_TRUE); |
e95b3bdc BB |
949 | } |
950 | } | |
951 | } | |
428870ff | 952 | |
a08ee875 | 953 | kmem_cache_free(ddt_entry_cache, dde); |
428870ff BB |
954 | return (B_FALSE); |
955 | } | |
956 | ||
957 | ddt_entry_t * | |
958 | ddt_repair_start(ddt_t *ddt, const blkptr_t *bp) | |
959 | { | |
960 | ddt_key_t ddk; | |
961 | ddt_entry_t *dde; | |
d6320ddb BB |
962 | enum ddt_type type; |
963 | enum ddt_class class; | |
428870ff BB |
964 | |
965 | ddt_key_fill(&ddk, bp); | |
966 | ||
967 | dde = ddt_alloc(&ddk); | |
968 | ||
d6320ddb BB |
969 | for (type = 0; type < DDT_TYPES; type++) { |
970 | for (class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
971 | /* |
972 | * We can only do repair if there are multiple copies | |
973 | * of the block. For anything in the UNIQUE class, | |
974 | * there's definitely only one copy, so don't even try. | |
975 | */ | |
976 | if (class != DDT_CLASS_UNIQUE && | |
977 | ddt_object_lookup(ddt, type, class, dde) == 0) | |
978 | return (dde); | |
979 | } | |
980 | } | |
981 | ||
982 | bzero(dde->dde_phys, sizeof (dde->dde_phys)); | |
983 | ||
984 | return (dde); | |
985 | } | |
986 | ||
987 | void | |
988 | ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde) | |
989 | { | |
990 | avl_index_t where; | |
991 | ||
992 | ddt_enter(ddt); | |
993 | ||
994 | if (dde->dde_repair_data != NULL && spa_writeable(ddt->ddt_spa) && | |
995 | avl_find(&ddt->ddt_repair_tree, dde, &where) == NULL) | |
996 | avl_insert(&ddt->ddt_repair_tree, dde, where); | |
997 | else | |
998 | ddt_free(dde); | |
999 | ||
1000 | ddt_exit(ddt); | |
1001 | } | |
1002 | ||
1003 | static void | |
1004 | ddt_repair_entry_done(zio_t *zio) | |
1005 | { | |
1006 | ddt_entry_t *rdde = zio->io_private; | |
1007 | ||
1008 | ddt_free(rdde); | |
1009 | } | |
1010 | ||
1011 | static void | |
1012 | ddt_repair_entry(ddt_t *ddt, ddt_entry_t *dde, ddt_entry_t *rdde, zio_t *rio) | |
1013 | { | |
1014 | ddt_phys_t *ddp = dde->dde_phys; | |
1015 | ddt_phys_t *rddp = rdde->dde_phys; | |
1016 | ddt_key_t *ddk = &dde->dde_key; | |
1017 | ddt_key_t *rddk = &rdde->dde_key; | |
1018 | zio_t *zio; | |
1019 | blkptr_t blk; | |
d6320ddb | 1020 | int p; |
428870ff BB |
1021 | |
1022 | zio = zio_null(rio, rio->io_spa, NULL, | |
1023 | ddt_repair_entry_done, rdde, rio->io_flags); | |
1024 | ||
d6320ddb | 1025 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++, rddp++) { |
428870ff BB |
1026 | if (ddp->ddp_phys_birth == 0 || |
1027 | ddp->ddp_phys_birth != rddp->ddp_phys_birth || | |
1028 | bcmp(ddp->ddp_dva, rddp->ddp_dva, sizeof (ddp->ddp_dva))) | |
1029 | continue; | |
1030 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
1031 | zio_nowait(zio_rewrite(zio, zio->io_spa, 0, &blk, | |
1032 | rdde->dde_repair_data, DDK_GET_PSIZE(rddk), NULL, NULL, | |
1033 | ZIO_PRIORITY_SYNC_WRITE, ZIO_DDT_CHILD_FLAGS(zio), NULL)); | |
1034 | } | |
1035 | ||
1036 | zio_nowait(zio); | |
1037 | } | |
1038 | ||
1039 | static void | |
1040 | ddt_repair_table(ddt_t *ddt, zio_t *rio) | |
1041 | { | |
1042 | spa_t *spa = ddt->ddt_spa; | |
1043 | ddt_entry_t *dde, *rdde_next, *rdde; | |
1044 | avl_tree_t *t = &ddt->ddt_repair_tree; | |
1045 | blkptr_t blk; | |
1046 | ||
1047 | if (spa_sync_pass(spa) > 1) | |
1048 | return; | |
1049 | ||
1050 | ddt_enter(ddt); | |
1051 | for (rdde = avl_first(t); rdde != NULL; rdde = rdde_next) { | |
1052 | rdde_next = AVL_NEXT(t, rdde); | |
1053 | avl_remove(&ddt->ddt_repair_tree, rdde); | |
1054 | ddt_exit(ddt); | |
1055 | ddt_bp_create(ddt->ddt_checksum, &rdde->dde_key, NULL, &blk); | |
1056 | dde = ddt_repair_start(ddt, &blk); | |
1057 | ddt_repair_entry(ddt, dde, rdde, rio); | |
1058 | ddt_repair_done(ddt, dde); | |
1059 | ddt_enter(ddt); | |
1060 | } | |
1061 | ddt_exit(ddt); | |
1062 | } | |
1063 | ||
1064 | static void | |
1065 | ddt_sync_entry(ddt_t *ddt, ddt_entry_t *dde, dmu_tx_t *tx, uint64_t txg) | |
1066 | { | |
1067 | dsl_pool_t *dp = ddt->ddt_spa->spa_dsl_pool; | |
1068 | ddt_phys_t *ddp = dde->dde_phys; | |
1069 | ddt_key_t *ddk = &dde->dde_key; | |
1070 | enum ddt_type otype = dde->dde_type; | |
1071 | enum ddt_type ntype = DDT_TYPE_CURRENT; | |
1072 | enum ddt_class oclass = dde->dde_class; | |
1073 | enum ddt_class nclass; | |
1074 | uint64_t total_refcnt = 0; | |
d6320ddb | 1075 | int p; |
428870ff BB |
1076 | |
1077 | ASSERT(dde->dde_loaded); | |
1078 | ASSERT(!dde->dde_loading); | |
1079 | ||
d6320ddb | 1080 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff | 1081 | ASSERT(dde->dde_lead_zio[p] == NULL); |
428870ff BB |
1082 | if (ddp->ddp_phys_birth == 0) { |
1083 | ASSERT(ddp->ddp_refcnt == 0); | |
1084 | continue; | |
1085 | } | |
1086 | if (p == DDT_PHYS_DITTO) { | |
1087 | if (ddt_ditto_copies_needed(ddt, dde, NULL) == 0) | |
1088 | ddt_phys_free(ddt, ddk, ddp, txg); | |
1089 | continue; | |
1090 | } | |
1091 | if (ddp->ddp_refcnt == 0) | |
1092 | ddt_phys_free(ddt, ddk, ddp, txg); | |
1093 | total_refcnt += ddp->ddp_refcnt; | |
1094 | } | |
1095 | ||
1096 | if (dde->dde_phys[DDT_PHYS_DITTO].ddp_phys_birth != 0) | |
1097 | nclass = DDT_CLASS_DITTO; | |
1098 | else if (total_refcnt > 1) | |
1099 | nclass = DDT_CLASS_DUPLICATE; | |
1100 | else | |
1101 | nclass = DDT_CLASS_UNIQUE; | |
1102 | ||
1103 | if (otype != DDT_TYPES && | |
1104 | (otype != ntype || oclass != nclass || total_refcnt == 0)) { | |
1105 | VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0); | |
1106 | ASSERT(ddt_object_lookup(ddt, otype, oclass, dde) == ENOENT); | |
1107 | } | |
1108 | ||
1109 | if (total_refcnt != 0) { | |
1110 | dde->dde_type = ntype; | |
1111 | dde->dde_class = nclass; | |
1112 | ddt_stat_update(ddt, dde, 0); | |
1113 | if (!ddt_object_exists(ddt, ntype, nclass)) | |
1114 | ddt_object_create(ddt, ntype, nclass, tx); | |
1115 | VERIFY(ddt_object_update(ddt, ntype, nclass, dde, tx) == 0); | |
1116 | ||
1117 | /* | |
1118 | * If the class changes, the order that we scan this bp | |
1119 | * changes. If it decreases, we could miss it, so | |
1120 | * scan it right now. (This covers both class changing | |
1121 | * while we are doing ddt_walk(), and when we are | |
1122 | * traversing.) | |
1123 | */ | |
1124 | if (nclass < oclass) { | |
1125 | dsl_scan_ddt_entry(dp->dp_scan, | |
1126 | ddt->ddt_checksum, dde, tx); | |
1127 | } | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | static void | |
1132 | ddt_sync_table(ddt_t *ddt, dmu_tx_t *tx, uint64_t txg) | |
1133 | { | |
1134 | spa_t *spa = ddt->ddt_spa; | |
1135 | ddt_entry_t *dde; | |
1136 | void *cookie = NULL; | |
d6320ddb BB |
1137 | enum ddt_type type; |
1138 | enum ddt_class class; | |
428870ff BB |
1139 | |
1140 | if (avl_numnodes(&ddt->ddt_tree) == 0) | |
1141 | return; | |
1142 | ||
1143 | ASSERT(spa->spa_uberblock.ub_version >= SPA_VERSION_DEDUP); | |
1144 | ||
1145 | if (spa->spa_ddt_stat_object == 0) { | |
9ae529ec CS |
1146 | spa->spa_ddt_stat_object = zap_create_link(ddt->ddt_os, |
1147 | DMU_OT_DDT_STATS, DMU_POOL_DIRECTORY_OBJECT, | |
1148 | DMU_POOL_DDT_STATS, tx); | |
428870ff BB |
1149 | } |
1150 | ||
1151 | while ((dde = avl_destroy_nodes(&ddt->ddt_tree, &cookie)) != NULL) { | |
1152 | ddt_sync_entry(ddt, dde, tx, txg); | |
1153 | ddt_free(dde); | |
1154 | } | |
1155 | ||
d6320ddb | 1156 | for (type = 0; type < DDT_TYPES; type++) { |
e8fd45a0 | 1157 | uint64_t add, count = 0; |
d6320ddb | 1158 | for (class = 0; class < DDT_CLASSES; class++) { |
572e2857 BB |
1159 | if (ddt_object_exists(ddt, type, class)) { |
1160 | ddt_object_sync(ddt, type, class, tx); | |
e8fd45a0 BB |
1161 | VERIFY(ddt_object_count(ddt, type, class, |
1162 | &add) == 0); | |
1163 | count += add; | |
572e2857 BB |
1164 | } |
1165 | } | |
d6320ddb | 1166 | for (class = 0; class < DDT_CLASSES; class++) { |
572e2857 | 1167 | if (count == 0 && ddt_object_exists(ddt, type, class)) |
428870ff BB |
1168 | ddt_object_destroy(ddt, type, class, tx); |
1169 | } | |
1170 | } | |
1171 | ||
1172 | bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache, | |
1173 | sizeof (ddt->ddt_histogram)); | |
1174 | } | |
1175 | ||
1176 | void | |
1177 | ddt_sync(spa_t *spa, uint64_t txg) | |
1178 | { | |
1179 | dmu_tx_t *tx; | |
1180 | zio_t *rio = zio_root(spa, NULL, NULL, | |
1181 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
d6320ddb | 1182 | enum zio_checksum c; |
428870ff BB |
1183 | |
1184 | ASSERT(spa_syncing_txg(spa) == txg); | |
1185 | ||
1186 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
1187 | ||
d6320ddb | 1188 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff BB |
1189 | ddt_t *ddt = spa->spa_ddt[c]; |
1190 | if (ddt == NULL) | |
1191 | continue; | |
1192 | ddt_sync_table(ddt, tx, txg); | |
1193 | ddt_repair_table(ddt, rio); | |
1194 | } | |
1195 | ||
1196 | (void) zio_wait(rio); | |
1197 | ||
1198 | dmu_tx_commit(tx); | |
1199 | } | |
1200 | ||
1201 | int | |
1202 | ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, ddt_entry_t *dde) | |
1203 | { | |
1204 | do { | |
1205 | do { | |
1206 | do { | |
1207 | ddt_t *ddt = spa->spa_ddt[ddb->ddb_checksum]; | |
1208 | int error = ENOENT; | |
1209 | if (ddt_object_exists(ddt, ddb->ddb_type, | |
1210 | ddb->ddb_class)) { | |
1211 | error = ddt_object_walk(ddt, | |
1212 | ddb->ddb_type, ddb->ddb_class, | |
1213 | &ddb->ddb_cursor, dde); | |
1214 | } | |
1215 | dde->dde_type = ddb->ddb_type; | |
1216 | dde->dde_class = ddb->ddb_class; | |
1217 | if (error == 0) | |
1218 | return (0); | |
1219 | if (error != ENOENT) | |
1220 | return (error); | |
1221 | ddb->ddb_cursor = 0; | |
1222 | } while (++ddb->ddb_checksum < ZIO_CHECKSUM_FUNCTIONS); | |
1223 | ddb->ddb_checksum = 0; | |
1224 | } while (++ddb->ddb_type < DDT_TYPES); | |
1225 | ddb->ddb_type = 0; | |
1226 | } while (++ddb->ddb_class < DDT_CLASSES); | |
1227 | ||
a08ee875 | 1228 | return (SET_ERROR(ENOENT)); |
428870ff | 1229 | } |
c409e464 BB |
1230 | |
1231 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
1232 | module_param(zfs_dedup_prefetch, int, 0644); | |
a08ee875 | 1233 | MODULE_PARM_DESC(zfs_dedup_prefetch, "Enable prefetching dedup-ed blks"); |
c409e464 | 1234 | #endif |