]>
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 | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
428870ff BB |
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. | |
a6255b7f | 24 | * Copyright (c) 2012, 2016 by Delphix. All rights reserved. |
67a1b037 | 25 | * Copyright (c) 2022 by Pawel Jakub Dawidek |
428870ff BB |
26 | */ |
27 | ||
28 | #include <sys/zfs_context.h> | |
29 | #include <sys/spa.h> | |
30 | #include <sys/spa_impl.h> | |
31 | #include <sys/zio.h> | |
32 | #include <sys/ddt.h> | |
33 | #include <sys/zap.h> | |
34 | #include <sys/dmu_tx.h> | |
35 | #include <sys/arc.h> | |
36 | #include <sys/dsl_pool.h> | |
37 | #include <sys/zio_checksum.h> | |
428870ff | 38 | #include <sys/dsl_scan.h> |
a6255b7f | 39 | #include <sys/abd.h> |
428870ff | 40 | |
ecf3d9b8 JL |
41 | static kmem_cache_t *ddt_cache; |
42 | static kmem_cache_t *ddt_entry_cache; | |
43 | ||
572e2857 BB |
44 | /* |
45 | * Enable/disable prefetching of dedup-ed blocks which are going to be freed. | |
46 | */ | |
0dfc7324 | 47 | int zfs_dedup_prefetch = 0; |
572e2857 | 48 | |
861166b0 | 49 | static const ddt_ops_t *const ddt_ops[DDT_TYPES] = { |
428870ff BB |
50 | &ddt_zap_ops, |
51 | }; | |
52 | ||
861166b0 | 53 | static const char *const ddt_class_name[DDT_CLASSES] = { |
428870ff BB |
54 | "ditto", |
55 | "duplicate", | |
56 | "unique", | |
57 | }; | |
58 | ||
59 | static void | |
60 | ddt_object_create(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
61 | dmu_tx_t *tx) | |
62 | { | |
63 | spa_t *spa = ddt->ddt_spa; | |
64 | objset_t *os = ddt->ddt_os; | |
65 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
3c67d83a TH |
66 | boolean_t prehash = zio_checksum_table[ddt->ddt_checksum].ci_flags & |
67 | ZCHECKSUM_FLAG_DEDUP; | |
428870ff BB |
68 | char name[DDT_NAMELEN]; |
69 | ||
70 | ddt_object_name(ddt, type, class, name); | |
71 | ||
d3bafe45 RN |
72 | ASSERT3U(*objectp, ==, 0); |
73 | VERIFY0(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash)); | |
74 | ASSERT3U(*objectp, !=, 0); | |
428870ff | 75 | |
d3bafe45 RN |
76 | VERIFY0(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name, |
77 | sizeof (uint64_t), 1, objectp, tx)); | |
428870ff | 78 | |
d3bafe45 | 79 | VERIFY0(zap_add(os, spa->spa_ddt_stat_object, name, |
428870ff | 80 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), |
d3bafe45 | 81 | &ddt->ddt_histogram[type][class], tx)); |
428870ff BB |
82 | } |
83 | ||
84 | static void | |
85 | ddt_object_destroy(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
86 | dmu_tx_t *tx) | |
87 | { | |
88 | spa_t *spa = ddt->ddt_spa; | |
89 | objset_t *os = ddt->ddt_os; | |
90 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
e8fd45a0 | 91 | uint64_t count; |
428870ff BB |
92 | char name[DDT_NAMELEN]; |
93 | ||
94 | ddt_object_name(ddt, type, class, name); | |
95 | ||
d3bafe45 | 96 | ASSERT3U(*objectp, !=, 0); |
428870ff | 97 | ASSERT(ddt_histogram_empty(&ddt->ddt_histogram[type][class])); |
d3bafe45 RN |
98 | VERIFY0(ddt_object_count(ddt, type, class, &count)); |
99 | VERIFY0(count); | |
100 | VERIFY0(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx)); | |
101 | VERIFY0(zap_remove(os, spa->spa_ddt_stat_object, name, tx)); | |
102 | VERIFY0(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx)); | |
861166b0 | 103 | memset(&ddt->ddt_object_stats[type][class], 0, sizeof (ddt_object_t)); |
428870ff BB |
104 | |
105 | *objectp = 0; | |
106 | } | |
107 | ||
108 | static int | |
109 | ddt_object_load(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
110 | { | |
111 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
112 | dmu_object_info_t doi; | |
e8fd45a0 | 113 | uint64_t count; |
428870ff BB |
114 | char name[DDT_NAMELEN]; |
115 | int error; | |
116 | ||
117 | ddt_object_name(ddt, type, class, name); | |
118 | ||
119 | error = zap_lookup(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, name, | |
120 | sizeof (uint64_t), 1, &ddt->ddt_object[type][class]); | |
b0bc7a84 | 121 | if (error != 0) |
428870ff BB |
122 | return (error); |
123 | ||
ff9b1d07 | 124 | error = zap_lookup(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, |
428870ff | 125 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), |
ff9b1d07 BB |
126 | &ddt->ddt_histogram[type][class]); |
127 | if (error != 0) | |
128 | return (error); | |
428870ff BB |
129 | |
130 | /* | |
131 | * Seed the cached statistics. | |
132 | */ | |
2a4a9dc2 BB |
133 | error = ddt_object_info(ddt, type, class, &doi); |
134 | if (error) | |
135 | return (error); | |
428870ff | 136 | |
e8fd45a0 BB |
137 | error = ddt_object_count(ddt, type, class, &count); |
138 | if (error) | |
139 | return (error); | |
140 | ||
141 | ddo->ddo_count = count; | |
428870ff BB |
142 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
143 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
144 | ||
b0bc7a84 | 145 | return (0); |
428870ff BB |
146 | } |
147 | ||
148 | static void | |
149 | ddt_object_sync(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
150 | dmu_tx_t *tx) | |
151 | { | |
152 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
153 | dmu_object_info_t doi; | |
e8fd45a0 | 154 | uint64_t count; |
428870ff BB |
155 | char name[DDT_NAMELEN]; |
156 | ||
157 | ddt_object_name(ddt, type, class, name); | |
158 | ||
d3bafe45 | 159 | VERIFY0(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, |
428870ff | 160 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), |
d3bafe45 | 161 | &ddt->ddt_histogram[type][class], tx)); |
428870ff BB |
162 | |
163 | /* | |
164 | * Cache DDT statistics; this is the only time they'll change. | |
165 | */ | |
d3bafe45 RN |
166 | VERIFY0(ddt_object_info(ddt, type, class, &doi)); |
167 | VERIFY0(ddt_object_count(ddt, type, class, &count)); | |
428870ff | 168 | |
e8fd45a0 | 169 | ddo->ddo_count = count; |
428870ff BB |
170 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
171 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
172 | } | |
173 | ||
174 | static int | |
175 | ddt_object_lookup(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
176 | ddt_entry_t *dde) | |
177 | { | |
178 | if (!ddt_object_exists(ddt, type, class)) | |
2e528b49 | 179 | return (SET_ERROR(ENOENT)); |
428870ff BB |
180 | |
181 | return (ddt_ops[type]->ddt_op_lookup(ddt->ddt_os, | |
182 | ddt->ddt_object[type][class], dde)); | |
183 | } | |
184 | ||
185 | static void | |
186 | ddt_object_prefetch(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
187 | ddt_entry_t *dde) | |
188 | { | |
189 | if (!ddt_object_exists(ddt, type, class)) | |
190 | return; | |
191 | ||
192 | ddt_ops[type]->ddt_op_prefetch(ddt->ddt_os, | |
193 | ddt->ddt_object[type][class], dde); | |
194 | } | |
195 | ||
196 | int | |
197 | ddt_object_update(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
198 | ddt_entry_t *dde, dmu_tx_t *tx) | |
199 | { | |
200 | ASSERT(ddt_object_exists(ddt, type, class)); | |
201 | ||
202 | return (ddt_ops[type]->ddt_op_update(ddt->ddt_os, | |
203 | ddt->ddt_object[type][class], dde, tx)); | |
204 | } | |
205 | ||
206 | static int | |
207 | ddt_object_remove(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
208 | ddt_entry_t *dde, dmu_tx_t *tx) | |
209 | { | |
210 | ASSERT(ddt_object_exists(ddt, type, class)); | |
211 | ||
212 | return (ddt_ops[type]->ddt_op_remove(ddt->ddt_os, | |
213 | ddt->ddt_object[type][class], dde, tx)); | |
214 | } | |
215 | ||
216 | int | |
217 | ddt_object_walk(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
218 | uint64_t *walk, ddt_entry_t *dde) | |
219 | { | |
220 | ASSERT(ddt_object_exists(ddt, type, class)); | |
221 | ||
222 | return (ddt_ops[type]->ddt_op_walk(ddt->ddt_os, | |
223 | ddt->ddt_object[type][class], dde, walk)); | |
224 | } | |
225 | ||
e8fd45a0 BB |
226 | int |
227 | ddt_object_count(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
228 | uint64_t *count) | |
428870ff BB |
229 | { |
230 | ASSERT(ddt_object_exists(ddt, type, class)); | |
231 | ||
232 | return (ddt_ops[type]->ddt_op_count(ddt->ddt_os, | |
e8fd45a0 | 233 | ddt->ddt_object[type][class], count)); |
428870ff BB |
234 | } |
235 | ||
236 | int | |
237 | ddt_object_info(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
238 | dmu_object_info_t *doi) | |
239 | { | |
240 | if (!ddt_object_exists(ddt, type, class)) | |
2e528b49 | 241 | return (SET_ERROR(ENOENT)); |
428870ff BB |
242 | |
243 | return (dmu_object_info(ddt->ddt_os, ddt->ddt_object[type][class], | |
244 | doi)); | |
245 | } | |
246 | ||
247 | boolean_t | |
248 | ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
249 | { | |
250 | return (!!ddt->ddt_object[type][class]); | |
251 | } | |
252 | ||
253 | void | |
254 | ddt_object_name(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
255 | char *name) | |
256 | { | |
c9e319fa | 257 | (void) snprintf(name, DDT_NAMELEN, DMU_POOL_DDT, |
428870ff BB |
258 | zio_checksum_table[ddt->ddt_checksum].ci_name, |
259 | ddt_ops[type]->ddt_op_name, ddt_class_name[class]); | |
260 | } | |
261 | ||
262 | void | |
263 | ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, uint64_t txg) | |
264 | { | |
d3bafe45 | 265 | ASSERT3U(txg, !=, 0); |
428870ff | 266 | |
1c27024e | 267 | for (int d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
268 | bp->blk_dva[d] = ddp->ddp_dva[d]; |
269 | BP_SET_BIRTH(bp, txg, ddp->ddp_phys_birth); | |
270 | } | |
271 | ||
b5256303 TC |
272 | /* |
273 | * The bp created via this function may be used for repairs and scrub, but it | |
274 | * will be missing the salt / IV required to do a full decrypting read. | |
275 | */ | |
428870ff BB |
276 | void |
277 | ddt_bp_create(enum zio_checksum checksum, | |
278 | const ddt_key_t *ddk, const ddt_phys_t *ddp, blkptr_t *bp) | |
279 | { | |
280 | BP_ZERO(bp); | |
281 | ||
282 | if (ddp != NULL) | |
283 | ddt_bp_fill(ddp, bp, ddp->ddp_phys_birth); | |
284 | ||
285 | bp->blk_cksum = ddk->ddk_cksum; | |
428870ff BB |
286 | |
287 | BP_SET_LSIZE(bp, DDK_GET_LSIZE(ddk)); | |
288 | BP_SET_PSIZE(bp, DDK_GET_PSIZE(ddk)); | |
289 | BP_SET_COMPRESS(bp, DDK_GET_COMPRESS(ddk)); | |
b5256303 TC |
290 | BP_SET_CRYPT(bp, DDK_GET_CRYPT(ddk)); |
291 | BP_SET_FILL(bp, 1); | |
428870ff BB |
292 | BP_SET_CHECKSUM(bp, checksum); |
293 | BP_SET_TYPE(bp, DMU_OT_DEDUP); | |
294 | BP_SET_LEVEL(bp, 0); | |
52b68423 | 295 | BP_SET_DEDUP(bp, 1); |
428870ff BB |
296 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); |
297 | } | |
298 | ||
299 | void | |
300 | ddt_key_fill(ddt_key_t *ddk, const blkptr_t *bp) | |
301 | { | |
302 | ddk->ddk_cksum = bp->blk_cksum; | |
303 | ddk->ddk_prop = 0; | |
304 | ||
b5256303 TC |
305 | ASSERT(BP_IS_ENCRYPTED(bp) || !BP_USES_CRYPT(bp)); |
306 | ||
428870ff BB |
307 | DDK_SET_LSIZE(ddk, BP_GET_LSIZE(bp)); |
308 | DDK_SET_PSIZE(ddk, BP_GET_PSIZE(bp)); | |
309 | DDK_SET_COMPRESS(ddk, BP_GET_COMPRESS(bp)); | |
b5256303 | 310 | DDK_SET_CRYPT(ddk, BP_USES_CRYPT(bp)); |
428870ff BB |
311 | } |
312 | ||
313 | void | |
314 | ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp) | |
315 | { | |
d3bafe45 | 316 | ASSERT0(ddp->ddp_phys_birth); |
428870ff | 317 | |
1c27024e | 318 | for (int d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
319 | ddp->ddp_dva[d] = bp->blk_dva[d]; |
320 | ddp->ddp_phys_birth = BP_PHYSICAL_BIRTH(bp); | |
321 | } | |
322 | ||
323 | void | |
324 | ddt_phys_clear(ddt_phys_t *ddp) | |
325 | { | |
861166b0 | 326 | memset(ddp, 0, sizeof (*ddp)); |
428870ff BB |
327 | } |
328 | ||
329 | void | |
330 | ddt_phys_addref(ddt_phys_t *ddp) | |
331 | { | |
332 | ddp->ddp_refcnt++; | |
333 | } | |
334 | ||
335 | void | |
336 | ddt_phys_decref(ddt_phys_t *ddp) | |
337 | { | |
5dc6af0e | 338 | if (ddp) { |
d3bafe45 | 339 | ASSERT3U(ddp->ddp_refcnt, >, 0); |
5dc6af0e BB |
340 | ddp->ddp_refcnt--; |
341 | } | |
428870ff BB |
342 | } |
343 | ||
344 | void | |
345 | ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, uint64_t txg) | |
346 | { | |
347 | blkptr_t blk; | |
348 | ||
349 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
52b68423 BB |
350 | |
351 | /* | |
352 | * We clear the dedup bit so that zio_free() will actually free the | |
353 | * space, rather than just decrementing the refcount in the DDT. | |
354 | */ | |
355 | BP_SET_DEDUP(&blk, 0); | |
356 | ||
428870ff BB |
357 | ddt_phys_clear(ddp); |
358 | zio_free(ddt->ddt_spa, txg, &blk); | |
359 | } | |
360 | ||
361 | ddt_phys_t * | |
362 | ddt_phys_select(const ddt_entry_t *dde, const blkptr_t *bp) | |
363 | { | |
364 | ddt_phys_t *ddp = (ddt_phys_t *)dde->dde_phys; | |
365 | ||
1c27024e | 366 | for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
367 | if (DVA_EQUAL(BP_IDENTITY(bp), &ddp->ddp_dva[0]) && |
368 | BP_PHYSICAL_BIRTH(bp) == ddp->ddp_phys_birth) | |
369 | return (ddp); | |
370 | } | |
371 | return (NULL); | |
372 | } | |
373 | ||
374 | uint64_t | |
375 | ddt_phys_total_refcnt(const ddt_entry_t *dde) | |
376 | { | |
377 | uint64_t refcnt = 0; | |
378 | ||
1c27024e | 379 | for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) |
428870ff BB |
380 | refcnt += dde->dde_phys[p].ddp_refcnt; |
381 | ||
382 | return (refcnt); | |
383 | } | |
384 | ||
428870ff BB |
385 | ddt_t * |
386 | ddt_select(spa_t *spa, const blkptr_t *bp) | |
387 | { | |
388 | return (spa->spa_ddt[BP_GET_CHECKSUM(bp)]); | |
389 | } | |
390 | ||
391 | void | |
392 | ddt_enter(ddt_t *ddt) | |
393 | { | |
394 | mutex_enter(&ddt->ddt_lock); | |
395 | } | |
396 | ||
397 | void | |
398 | ddt_exit(ddt_t *ddt) | |
399 | { | |
400 | mutex_exit(&ddt->ddt_lock); | |
401 | } | |
402 | ||
ecf3d9b8 JL |
403 | void |
404 | ddt_init(void) | |
405 | { | |
406 | ddt_cache = kmem_cache_create("ddt_cache", | |
407 | sizeof (ddt_t), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
408 | ddt_entry_cache = kmem_cache_create("ddt_entry_cache", | |
409 | sizeof (ddt_entry_t), 0, NULL, NULL, NULL, NULL, NULL, 0); | |
410 | } | |
411 | ||
412 | void | |
413 | ddt_fini(void) | |
414 | { | |
415 | kmem_cache_destroy(ddt_entry_cache); | |
416 | kmem_cache_destroy(ddt_cache); | |
417 | } | |
418 | ||
428870ff BB |
419 | static ddt_entry_t * |
420 | ddt_alloc(const ddt_key_t *ddk) | |
421 | { | |
422 | ddt_entry_t *dde; | |
423 | ||
79c76d5b | 424 | dde = kmem_cache_alloc(ddt_entry_cache, KM_SLEEP); |
861166b0 | 425 | memset(dde, 0, sizeof (ddt_entry_t)); |
428870ff BB |
426 | cv_init(&dde->dde_cv, NULL, CV_DEFAULT, NULL); |
427 | ||
428 | dde->dde_key = *ddk; | |
429 | ||
430 | return (dde); | |
431 | } | |
432 | ||
433 | static void | |
434 | ddt_free(ddt_entry_t *dde) | |
435 | { | |
436 | ASSERT(!dde->dde_loading); | |
437 | ||
1c27024e | 438 | for (int p = 0; p < DDT_PHYS_TYPES; p++) |
d3bafe45 | 439 | ASSERT3P(dde->dde_lead_zio[p], ==, NULL); |
428870ff | 440 | |
a6255b7f DQ |
441 | if (dde->dde_repair_abd != NULL) |
442 | abd_free(dde->dde_repair_abd); | |
428870ff BB |
443 | |
444 | cv_destroy(&dde->dde_cv); | |
ecf3d9b8 | 445 | kmem_cache_free(ddt_entry_cache, dde); |
428870ff BB |
446 | } |
447 | ||
448 | void | |
449 | ddt_remove(ddt_t *ddt, ddt_entry_t *dde) | |
450 | { | |
451 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
452 | ||
453 | avl_remove(&ddt->ddt_tree, dde); | |
454 | ddt_free(dde); | |
455 | } | |
456 | ||
457 | ddt_entry_t * | |
458 | ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add) | |
459 | { | |
0cb1ef60 RN |
460 | ddt_key_t search; |
461 | ddt_entry_t *dde; | |
428870ff BB |
462 | enum ddt_type type; |
463 | enum ddt_class class; | |
464 | avl_index_t where; | |
465 | int error; | |
466 | ||
467 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
468 | ||
0cb1ef60 | 469 | ddt_key_fill(&search, bp); |
428870ff | 470 | |
0cb1ef60 | 471 | dde = avl_find(&ddt->ddt_tree, &search, &where); |
428870ff BB |
472 | if (dde == NULL) { |
473 | if (!add) | |
474 | return (NULL); | |
0cb1ef60 | 475 | dde = ddt_alloc(&search); |
428870ff BB |
476 | avl_insert(&ddt->ddt_tree, dde, where); |
477 | } | |
478 | ||
479 | while (dde->dde_loading) | |
480 | cv_wait(&dde->dde_cv, &ddt->ddt_lock); | |
481 | ||
482 | if (dde->dde_loaded) | |
483 | return (dde); | |
484 | ||
485 | dde->dde_loading = B_TRUE; | |
486 | ||
487 | ddt_exit(ddt); | |
488 | ||
489 | error = ENOENT; | |
490 | ||
491 | for (type = 0; type < DDT_TYPES; type++) { | |
492 | for (class = 0; class < DDT_CLASSES; class++) { | |
493 | error = ddt_object_lookup(ddt, type, class, dde); | |
a1d477c2 MA |
494 | if (error != ENOENT) { |
495 | ASSERT0(error); | |
428870ff | 496 | break; |
a1d477c2 | 497 | } |
428870ff BB |
498 | } |
499 | if (error != ENOENT) | |
500 | break; | |
501 | } | |
502 | ||
428870ff BB |
503 | ddt_enter(ddt); |
504 | ||
d3bafe45 RN |
505 | ASSERT(!dde->dde_loaded); |
506 | ASSERT(dde->dde_loading); | |
428870ff BB |
507 | |
508 | dde->dde_type = type; /* will be DDT_TYPES if no entry found */ | |
509 | dde->dde_class = class; /* will be DDT_CLASSES if no entry found */ | |
510 | dde->dde_loaded = B_TRUE; | |
511 | dde->dde_loading = B_FALSE; | |
512 | ||
513 | if (error == 0) | |
514 | ddt_stat_update(ddt, dde, -1ULL); | |
515 | ||
516 | cv_broadcast(&dde->dde_cv); | |
517 | ||
518 | return (dde); | |
519 | } | |
520 | ||
521 | void | |
522 | ddt_prefetch(spa_t *spa, const blkptr_t *bp) | |
523 | { | |
524 | ddt_t *ddt; | |
525 | ddt_entry_t dde; | |
526 | ||
572e2857 | 527 | if (!zfs_dedup_prefetch || bp == NULL || !BP_GET_DEDUP(bp)) |
428870ff BB |
528 | return; |
529 | ||
530 | /* | |
572e2857 BB |
531 | * We only remove the DDT once all tables are empty and only |
532 | * prefetch dedup blocks when there are entries in the DDT. | |
533 | * Thus no locking is required as the DDT can't disappear on us. | |
428870ff BB |
534 | */ |
535 | ddt = ddt_select(spa, bp); | |
536 | ddt_key_fill(&dde.dde_key, bp); | |
537 | ||
1c27024e DB |
538 | for (enum ddt_type type = 0; type < DDT_TYPES; type++) { |
539 | for (enum ddt_class class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
540 | ddt_object_prefetch(ddt, type, class, &dde); |
541 | } | |
542 | } | |
543 | } | |
544 | ||
ee36c709 | 545 | /* |
0cb1ef60 RN |
546 | * Key comparison. Any struct wanting to make use of this function must have |
547 | * the key as the first element. | |
ee36c709 GN |
548 | */ |
549 | #define DDT_KEY_CMP_LEN (sizeof (ddt_key_t) / sizeof (uint16_t)) | |
550 | ||
551 | typedef struct ddt_key_cmp { | |
552 | uint16_t u16[DDT_KEY_CMP_LEN]; | |
553 | } ddt_key_cmp_t; | |
554 | ||
428870ff | 555 | int |
0cb1ef60 | 556 | ddt_key_compare(const void *x1, const void *x2) |
428870ff | 557 | { |
0cb1ef60 RN |
558 | const ddt_key_cmp_t *k1 = (const ddt_key_cmp_t *)x1; |
559 | const ddt_key_cmp_t *k2 = (const ddt_key_cmp_t *)x2; | |
ee36c709 | 560 | int32_t cmp = 0; |
428870ff | 561 | |
1c27024e | 562 | for (int i = 0; i < DDT_KEY_CMP_LEN; i++) { |
ee36c709 GN |
563 | cmp = (int32_t)k1->u16[i] - (int32_t)k2->u16[i]; |
564 | if (likely(cmp)) | |
565 | break; | |
428870ff BB |
566 | } |
567 | ||
ca577779 | 568 | return (TREE_ISIGN(cmp)); |
428870ff BB |
569 | } |
570 | ||
571 | static ddt_t * | |
572 | ddt_table_alloc(spa_t *spa, enum zio_checksum c) | |
573 | { | |
574 | ddt_t *ddt; | |
575 | ||
79c76d5b | 576 | ddt = kmem_cache_alloc(ddt_cache, KM_SLEEP); |
861166b0 | 577 | memset(ddt, 0, sizeof (ddt_t)); |
428870ff BB |
578 | |
579 | mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL); | |
0cb1ef60 | 580 | avl_create(&ddt->ddt_tree, ddt_key_compare, |
428870ff | 581 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); |
0cb1ef60 | 582 | avl_create(&ddt->ddt_repair_tree, ddt_key_compare, |
428870ff BB |
583 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); |
584 | ddt->ddt_checksum = c; | |
585 | ddt->ddt_spa = spa; | |
586 | ddt->ddt_os = spa->spa_meta_objset; | |
587 | ||
588 | return (ddt); | |
589 | } | |
590 | ||
591 | static void | |
592 | ddt_table_free(ddt_t *ddt) | |
593 | { | |
d3bafe45 RN |
594 | ASSERT0(avl_numnodes(&ddt->ddt_tree)); |
595 | ASSERT0(avl_numnodes(&ddt->ddt_repair_tree)); | |
428870ff BB |
596 | avl_destroy(&ddt->ddt_tree); |
597 | avl_destroy(&ddt->ddt_repair_tree); | |
598 | mutex_destroy(&ddt->ddt_lock); | |
ecf3d9b8 | 599 | kmem_cache_free(ddt_cache, ddt); |
428870ff BB |
600 | } |
601 | ||
602 | void | |
603 | ddt_create(spa_t *spa) | |
604 | { | |
605 | spa->spa_dedup_checksum = ZIO_DEDUPCHECKSUM; | |
606 | ||
1c27024e | 607 | for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) |
428870ff BB |
608 | spa->spa_ddt[c] = ddt_table_alloc(spa, c); |
609 | } | |
610 | ||
611 | int | |
612 | ddt_load(spa_t *spa) | |
613 | { | |
614 | int error; | |
615 | ||
616 | ddt_create(spa); | |
617 | ||
618 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
619 | DMU_POOL_DDT_STATS, sizeof (uint64_t), 1, | |
620 | &spa->spa_ddt_stat_object); | |
621 | ||
622 | if (error) | |
623 | return (error == ENOENT ? 0 : error); | |
624 | ||
1c27024e | 625 | for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 626 | ddt_t *ddt = spa->spa_ddt[c]; |
1c27024e DB |
627 | for (enum ddt_type type = 0; type < DDT_TYPES; type++) { |
628 | for (enum ddt_class class = 0; class < DDT_CLASSES; | |
428870ff BB |
629 | class++) { |
630 | error = ddt_object_load(ddt, type, class); | |
631 | if (error != 0 && error != ENOENT) | |
632 | return (error); | |
633 | } | |
634 | } | |
635 | ||
636 | /* | |
637 | * Seed the cached histograms. | |
638 | */ | |
861166b0 | 639 | memcpy(&ddt->ddt_histogram_cache, ddt->ddt_histogram, |
428870ff | 640 | sizeof (ddt->ddt_histogram)); |
e8a20144 | 641 | spa->spa_dedup_dspace = ~0ULL; |
428870ff BB |
642 | } |
643 | ||
644 | return (0); | |
645 | } | |
646 | ||
647 | void | |
648 | ddt_unload(spa_t *spa) | |
649 | { | |
1c27024e | 650 | for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff BB |
651 | if (spa->spa_ddt[c]) { |
652 | ddt_table_free(spa->spa_ddt[c]); | |
653 | spa->spa_ddt[c] = NULL; | |
654 | } | |
655 | } | |
656 | } | |
657 | ||
658 | boolean_t | |
659 | ddt_class_contains(spa_t *spa, enum ddt_class max_class, const blkptr_t *bp) | |
660 | { | |
661 | ddt_t *ddt; | |
e95b3bdc | 662 | ddt_entry_t *dde; |
428870ff BB |
663 | |
664 | if (!BP_GET_DEDUP(bp)) | |
665 | return (B_FALSE); | |
666 | ||
667 | if (max_class == DDT_CLASS_UNIQUE) | |
668 | return (B_TRUE); | |
669 | ||
670 | ddt = spa->spa_ddt[BP_GET_CHECKSUM(bp)]; | |
79c76d5b | 671 | dde = kmem_cache_alloc(ddt_entry_cache, KM_SLEEP); |
428870ff | 672 | |
e95b3bdc | 673 | ddt_key_fill(&(dde->dde_key), bp); |
428870ff | 674 | |
1c27024e DB |
675 | for (enum ddt_type type = 0; type < DDT_TYPES; type++) { |
676 | for (enum ddt_class class = 0; class <= max_class; class++) { | |
e95b3bdc | 677 | if (ddt_object_lookup(ddt, type, class, dde) == 0) { |
ecf3d9b8 | 678 | kmem_cache_free(ddt_entry_cache, dde); |
428870ff | 679 | return (B_TRUE); |
e95b3bdc BB |
680 | } |
681 | } | |
682 | } | |
428870ff | 683 | |
ecf3d9b8 | 684 | kmem_cache_free(ddt_entry_cache, dde); |
428870ff BB |
685 | return (B_FALSE); |
686 | } | |
687 | ||
688 | ddt_entry_t * | |
689 | ddt_repair_start(ddt_t *ddt, const blkptr_t *bp) | |
690 | { | |
691 | ddt_key_t ddk; | |
692 | ddt_entry_t *dde; | |
693 | ||
694 | ddt_key_fill(&ddk, bp); | |
695 | ||
696 | dde = ddt_alloc(&ddk); | |
697 | ||
1c27024e DB |
698 | for (enum ddt_type type = 0; type < DDT_TYPES; type++) { |
699 | for (enum ddt_class class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
700 | /* |
701 | * We can only do repair if there are multiple copies | |
702 | * of the block. For anything in the UNIQUE class, | |
703 | * there's definitely only one copy, so don't even try. | |
704 | */ | |
705 | if (class != DDT_CLASS_UNIQUE && | |
706 | ddt_object_lookup(ddt, type, class, dde) == 0) | |
707 | return (dde); | |
708 | } | |
709 | } | |
710 | ||
861166b0 | 711 | memset(dde->dde_phys, 0, sizeof (dde->dde_phys)); |
428870ff BB |
712 | |
713 | return (dde); | |
714 | } | |
715 | ||
716 | void | |
717 | ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde) | |
718 | { | |
719 | avl_index_t where; | |
720 | ||
721 | ddt_enter(ddt); | |
722 | ||
a6255b7f | 723 | if (dde->dde_repair_abd != NULL && spa_writeable(ddt->ddt_spa) && |
428870ff BB |
724 | avl_find(&ddt->ddt_repair_tree, dde, &where) == NULL) |
725 | avl_insert(&ddt->ddt_repair_tree, dde, where); | |
726 | else | |
727 | ddt_free(dde); | |
728 | ||
729 | ddt_exit(ddt); | |
730 | } | |
731 | ||
732 | static void | |
733 | ddt_repair_entry_done(zio_t *zio) | |
734 | { | |
735 | ddt_entry_t *rdde = zio->io_private; | |
736 | ||
737 | ddt_free(rdde); | |
738 | } | |
739 | ||
740 | static void | |
741 | ddt_repair_entry(ddt_t *ddt, ddt_entry_t *dde, ddt_entry_t *rdde, zio_t *rio) | |
742 | { | |
743 | ddt_phys_t *ddp = dde->dde_phys; | |
744 | ddt_phys_t *rddp = rdde->dde_phys; | |
745 | ddt_key_t *ddk = &dde->dde_key; | |
746 | ddt_key_t *rddk = &rdde->dde_key; | |
747 | zio_t *zio; | |
748 | blkptr_t blk; | |
749 | ||
750 | zio = zio_null(rio, rio->io_spa, NULL, | |
751 | ddt_repair_entry_done, rdde, rio->io_flags); | |
752 | ||
1c27024e | 753 | for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++, rddp++) { |
428870ff BB |
754 | if (ddp->ddp_phys_birth == 0 || |
755 | ddp->ddp_phys_birth != rddp->ddp_phys_birth || | |
861166b0 | 756 | memcmp(ddp->ddp_dva, rddp->ddp_dva, sizeof (ddp->ddp_dva))) |
428870ff BB |
757 | continue; |
758 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
759 | zio_nowait(zio_rewrite(zio, zio->io_spa, 0, &blk, | |
a6255b7f | 760 | rdde->dde_repair_abd, DDK_GET_PSIZE(rddk), NULL, NULL, |
428870ff BB |
761 | ZIO_PRIORITY_SYNC_WRITE, ZIO_DDT_CHILD_FLAGS(zio), NULL)); |
762 | } | |
763 | ||
764 | zio_nowait(zio); | |
765 | } | |
766 | ||
767 | static void | |
768 | ddt_repair_table(ddt_t *ddt, zio_t *rio) | |
769 | { | |
770 | spa_t *spa = ddt->ddt_spa; | |
771 | ddt_entry_t *dde, *rdde_next, *rdde; | |
772 | avl_tree_t *t = &ddt->ddt_repair_tree; | |
773 | blkptr_t blk; | |
774 | ||
775 | if (spa_sync_pass(spa) > 1) | |
776 | return; | |
777 | ||
778 | ddt_enter(ddt); | |
779 | for (rdde = avl_first(t); rdde != NULL; rdde = rdde_next) { | |
780 | rdde_next = AVL_NEXT(t, rdde); | |
781 | avl_remove(&ddt->ddt_repair_tree, rdde); | |
782 | ddt_exit(ddt); | |
783 | ddt_bp_create(ddt->ddt_checksum, &rdde->dde_key, NULL, &blk); | |
784 | dde = ddt_repair_start(ddt, &blk); | |
785 | ddt_repair_entry(ddt, dde, rdde, rio); | |
786 | ddt_repair_done(ddt, dde); | |
787 | ddt_enter(ddt); | |
788 | } | |
789 | ddt_exit(ddt); | |
790 | } | |
791 | ||
792 | static void | |
793 | ddt_sync_entry(ddt_t *ddt, ddt_entry_t *dde, dmu_tx_t *tx, uint64_t txg) | |
794 | { | |
795 | dsl_pool_t *dp = ddt->ddt_spa->spa_dsl_pool; | |
796 | ddt_phys_t *ddp = dde->dde_phys; | |
797 | ddt_key_t *ddk = &dde->dde_key; | |
798 | enum ddt_type otype = dde->dde_type; | |
799 | enum ddt_type ntype = DDT_TYPE_CURRENT; | |
800 | enum ddt_class oclass = dde->dde_class; | |
801 | enum ddt_class nclass; | |
802 | uint64_t total_refcnt = 0; | |
803 | ||
804 | ASSERT(dde->dde_loaded); | |
805 | ASSERT(!dde->dde_loading); | |
806 | ||
1c27024e | 807 | for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
d3bafe45 | 808 | ASSERT3P(dde->dde_lead_zio[p], ==, NULL); |
428870ff | 809 | if (ddp->ddp_phys_birth == 0) { |
d3bafe45 | 810 | ASSERT0(ddp->ddp_refcnt); |
428870ff BB |
811 | continue; |
812 | } | |
813 | if (p == DDT_PHYS_DITTO) { | |
050d720c MA |
814 | /* |
815 | * Note, we no longer create DDT-DITTO blocks, but we | |
816 | * don't want to leak any written by older software. | |
817 | */ | |
818 | ddt_phys_free(ddt, ddk, ddp, txg); | |
428870ff BB |
819 | continue; |
820 | } | |
821 | if (ddp->ddp_refcnt == 0) | |
822 | ddt_phys_free(ddt, ddk, ddp, txg); | |
823 | total_refcnt += ddp->ddp_refcnt; | |
824 | } | |
825 | ||
050d720c MA |
826 | /* We do not create new DDT-DITTO blocks. */ |
827 | ASSERT0(dde->dde_phys[DDT_PHYS_DITTO].ddp_phys_birth); | |
828 | if (total_refcnt > 1) | |
428870ff BB |
829 | nclass = DDT_CLASS_DUPLICATE; |
830 | else | |
831 | nclass = DDT_CLASS_UNIQUE; | |
832 | ||
833 | if (otype != DDT_TYPES && | |
834 | (otype != ntype || oclass != nclass || total_refcnt == 0)) { | |
d3bafe45 RN |
835 | VERIFY0(ddt_object_remove(ddt, otype, oclass, dde, tx)); |
836 | ASSERT3U( | |
837 | ddt_object_lookup(ddt, otype, oclass, dde), ==, ENOENT); | |
428870ff BB |
838 | } |
839 | ||
840 | if (total_refcnt != 0) { | |
841 | dde->dde_type = ntype; | |
842 | dde->dde_class = nclass; | |
843 | ddt_stat_update(ddt, dde, 0); | |
844 | if (!ddt_object_exists(ddt, ntype, nclass)) | |
845 | ddt_object_create(ddt, ntype, nclass, tx); | |
d3bafe45 | 846 | VERIFY0(ddt_object_update(ddt, ntype, nclass, dde, tx)); |
428870ff BB |
847 | |
848 | /* | |
849 | * If the class changes, the order that we scan this bp | |
850 | * changes. If it decreases, we could miss it, so | |
851 | * scan it right now. (This covers both class changing | |
852 | * while we are doing ddt_walk(), and when we are | |
853 | * traversing.) | |
854 | */ | |
855 | if (nclass < oclass) { | |
856 | dsl_scan_ddt_entry(dp->dp_scan, | |
857 | ddt->ddt_checksum, dde, tx); | |
858 | } | |
859 | } | |
860 | } | |
861 | ||
862 | static void | |
863 | ddt_sync_table(ddt_t *ddt, dmu_tx_t *tx, uint64_t txg) | |
864 | { | |
865 | spa_t *spa = ddt->ddt_spa; | |
866 | ddt_entry_t *dde; | |
867 | void *cookie = NULL; | |
868 | ||
869 | if (avl_numnodes(&ddt->ddt_tree) == 0) | |
870 | return; | |
871 | ||
d3bafe45 | 872 | ASSERT3U(spa->spa_uberblock.ub_version, >=, SPA_VERSION_DEDUP); |
428870ff BB |
873 | |
874 | if (spa->spa_ddt_stat_object == 0) { | |
9ae529ec CS |
875 | spa->spa_ddt_stat_object = zap_create_link(ddt->ddt_os, |
876 | DMU_OT_DDT_STATS, DMU_POOL_DIRECTORY_OBJECT, | |
877 | DMU_POOL_DDT_STATS, tx); | |
428870ff BB |
878 | } |
879 | ||
880 | while ((dde = avl_destroy_nodes(&ddt->ddt_tree, &cookie)) != NULL) { | |
881 | ddt_sync_entry(ddt, dde, tx, txg); | |
882 | ddt_free(dde); | |
883 | } | |
884 | ||
1c27024e | 885 | for (enum ddt_type type = 0; type < DDT_TYPES; type++) { |
e8fd45a0 | 886 | uint64_t add, count = 0; |
1c27024e | 887 | for (enum ddt_class class = 0; class < DDT_CLASSES; class++) { |
572e2857 BB |
888 | if (ddt_object_exists(ddt, type, class)) { |
889 | ddt_object_sync(ddt, type, class, tx); | |
d3bafe45 RN |
890 | VERIFY0(ddt_object_count(ddt, type, class, |
891 | &add)); | |
e8fd45a0 | 892 | count += add; |
572e2857 BB |
893 | } |
894 | } | |
1c27024e | 895 | for (enum ddt_class class = 0; class < DDT_CLASSES; class++) { |
572e2857 | 896 | if (count == 0 && ddt_object_exists(ddt, type, class)) |
428870ff BB |
897 | ddt_object_destroy(ddt, type, class, tx); |
898 | } | |
899 | } | |
900 | ||
861166b0 | 901 | memcpy(&ddt->ddt_histogram_cache, ddt->ddt_histogram, |
428870ff | 902 | sizeof (ddt->ddt_histogram)); |
e8a20144 | 903 | spa->spa_dedup_dspace = ~0ULL; |
428870ff BB |
904 | } |
905 | ||
906 | void | |
907 | ddt_sync(spa_t *spa, uint64_t txg) | |
908 | { | |
d4a72f23 | 909 | dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; |
428870ff | 910 | dmu_tx_t *tx; |
d4a72f23 | 911 | zio_t *rio; |
428870ff | 912 | |
d3bafe45 | 913 | ASSERT3U(spa_syncing_txg(spa), ==, txg); |
428870ff BB |
914 | |
915 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
916 | ||
d4a72f23 | 917 | rio = zio_root(spa, NULL, NULL, |
a1d477c2 | 918 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SELF_HEAL); |
d4a72f23 TC |
919 | |
920 | /* | |
921 | * This function may cause an immediate scan of ddt blocks (see | |
922 | * the comment above dsl_scan_ddt() for details). We set the | |
923 | * scan's root zio here so that we can wait for any scan IOs in | |
924 | * addition to the regular ddt IOs. | |
925 | */ | |
926 | ASSERT3P(scn->scn_zio_root, ==, NULL); | |
927 | scn->scn_zio_root = rio; | |
928 | ||
1c27024e | 929 | for (enum zio_checksum c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff BB |
930 | ddt_t *ddt = spa->spa_ddt[c]; |
931 | if (ddt == NULL) | |
932 | continue; | |
933 | ddt_sync_table(ddt, tx, txg); | |
934 | ddt_repair_table(ddt, rio); | |
935 | } | |
936 | ||
937 | (void) zio_wait(rio); | |
d4a72f23 | 938 | scn->scn_zio_root = NULL; |
428870ff BB |
939 | |
940 | dmu_tx_commit(tx); | |
941 | } | |
942 | ||
943 | int | |
944 | ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, ddt_entry_t *dde) | |
945 | { | |
946 | do { | |
947 | do { | |
948 | do { | |
949 | ddt_t *ddt = spa->spa_ddt[ddb->ddb_checksum]; | |
950 | int error = ENOENT; | |
951 | if (ddt_object_exists(ddt, ddb->ddb_type, | |
952 | ddb->ddb_class)) { | |
953 | error = ddt_object_walk(ddt, | |
954 | ddb->ddb_type, ddb->ddb_class, | |
955 | &ddb->ddb_cursor, dde); | |
956 | } | |
957 | dde->dde_type = ddb->ddb_type; | |
958 | dde->dde_class = ddb->ddb_class; | |
959 | if (error == 0) | |
960 | return (0); | |
961 | if (error != ENOENT) | |
962 | return (error); | |
963 | ddb->ddb_cursor = 0; | |
964 | } while (++ddb->ddb_checksum < ZIO_CHECKSUM_FUNCTIONS); | |
965 | ddb->ddb_checksum = 0; | |
966 | } while (++ddb->ddb_type < DDT_TYPES); | |
967 | ddb->ddb_type = 0; | |
968 | } while (++ddb->ddb_class < DDT_CLASSES); | |
969 | ||
2e528b49 | 970 | return (SET_ERROR(ENOENT)); |
428870ff | 971 | } |
c409e464 | 972 | |
67a1b037 PJD |
973 | /* |
974 | * This function is used by Block Cloning (brt.c) to increase reference | |
975 | * counter for the DDT entry if the block is already in DDT. | |
976 | * | |
977 | * Return false if the block, despite having the D bit set, is not present | |
978 | * in the DDT. Currently this is not possible but might be in the future. | |
979 | * See the comment below. | |
980 | */ | |
981 | boolean_t | |
982 | ddt_addref(spa_t *spa, const blkptr_t *bp) | |
983 | { | |
984 | ddt_t *ddt; | |
985 | ddt_entry_t *dde; | |
986 | boolean_t result; | |
987 | ||
988 | spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER); | |
989 | ddt = ddt_select(spa, bp); | |
990 | ddt_enter(ddt); | |
991 | ||
992 | dde = ddt_lookup(ddt, bp, B_TRUE); | |
d3bafe45 | 993 | ASSERT3P(dde, !=, NULL); |
67a1b037 PJD |
994 | |
995 | if (dde->dde_type < DDT_TYPES) { | |
996 | ddt_phys_t *ddp; | |
997 | ||
998 | ASSERT3S(dde->dde_class, <, DDT_CLASSES); | |
999 | ||
1000 | ddp = &dde->dde_phys[BP_GET_NDVAS(bp)]; | |
61ab05ca RN |
1001 | |
1002 | /* | |
1003 | * This entry already existed (dde_type is real), so it must | |
1004 | * have refcnt >0 at the start of this txg. We are called from | |
1005 | * brt_pending_apply(), before frees are issued, so the refcnt | |
1006 | * can't be lowered yet. Therefore, it must be >0. We assert | |
1007 | * this because if the order of BRT and DDT interactions were | |
1008 | * ever to change and the refcnt was ever zero here, then | |
1009 | * likely further action is required to fill out the DDT entry, | |
1010 | * and this is a place that is likely to be missed in testing. | |
1011 | */ | |
1012 | ASSERT3U(ddp->ddp_refcnt, >, 0); | |
1013 | ||
67a1b037 PJD |
1014 | ddt_phys_addref(ddp); |
1015 | result = B_TRUE; | |
1016 | } else { | |
1017 | /* | |
1018 | * At the time of implementating this if the block has the | |
1019 | * DEDUP flag set it must exist in the DEDUP table, but | |
1020 | * there are many advocates that want ability to remove | |
1021 | * entries from DDT with refcnt=1. If this will happen, | |
1022 | * we may have a block with the DEDUP set, but which doesn't | |
1023 | * have a corresponding entry in the DDT. Be ready. | |
1024 | */ | |
1025 | ASSERT3S(dde->dde_class, ==, DDT_CLASSES); | |
1026 | ddt_remove(ddt, dde); | |
1027 | result = B_FALSE; | |
1028 | } | |
1029 | ||
1030 | ddt_exit(ddt); | |
1031 | spa_config_exit(spa, SCL_ZIO, FTAG); | |
1032 | ||
1033 | return (result); | |
1034 | } | |
1035 | ||
a7929f31 | 1036 | ZFS_MODULE_PARAM(zfs_dedup, zfs_dedup_, prefetch, INT, ZMOD_RW, |
03fdcb9a | 1037 | "Enable prefetching dedup-ed blks"); |