]>
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. | |
9ae529ec | 24 | * Copyright (c) 2012 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 | ||
572e2857 BB |
40 | /* |
41 | * Enable/disable prefetching of dedup-ed blocks which are going to be freed. | |
42 | */ | |
43 | int zfs_dedup_prefetch = 1; | |
44 | ||
428870ff BB |
45 | static const ddt_ops_t *ddt_ops[DDT_TYPES] = { |
46 | &ddt_zap_ops, | |
47 | }; | |
48 | ||
49 | static const char *ddt_class_name[DDT_CLASSES] = { | |
50 | "ditto", | |
51 | "duplicate", | |
52 | "unique", | |
53 | }; | |
54 | ||
55 | static void | |
56 | ddt_object_create(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
57 | dmu_tx_t *tx) | |
58 | { | |
59 | spa_t *spa = ddt->ddt_spa; | |
60 | objset_t *os = ddt->ddt_os; | |
61 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
62 | boolean_t prehash = zio_checksum_table[ddt->ddt_checksum].ci_dedup; | |
63 | char name[DDT_NAMELEN]; | |
64 | ||
65 | ddt_object_name(ddt, type, class, name); | |
66 | ||
67 | ASSERT(*objectp == 0); | |
68 | VERIFY(ddt_ops[type]->ddt_op_create(os, objectp, tx, prehash) == 0); | |
69 | ASSERT(*objectp != 0); | |
70 | ||
71 | VERIFY(zap_add(os, DMU_POOL_DIRECTORY_OBJECT, name, | |
72 | sizeof (uint64_t), 1, objectp, tx) == 0); | |
73 | ||
74 | VERIFY(zap_add(os, spa->spa_ddt_stat_object, name, | |
75 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
76 | &ddt->ddt_histogram[type][class], tx) == 0); | |
77 | } | |
78 | ||
79 | static void | |
80 | ddt_object_destroy(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
81 | dmu_tx_t *tx) | |
82 | { | |
83 | spa_t *spa = ddt->ddt_spa; | |
84 | objset_t *os = ddt->ddt_os; | |
85 | uint64_t *objectp = &ddt->ddt_object[type][class]; | |
e8fd45a0 | 86 | uint64_t count; |
428870ff BB |
87 | char name[DDT_NAMELEN]; |
88 | ||
89 | ddt_object_name(ddt, type, class, name); | |
90 | ||
91 | ASSERT(*objectp != 0); | |
428870ff | 92 | ASSERT(ddt_histogram_empty(&ddt->ddt_histogram[type][class])); |
e8fd45a0 | 93 | VERIFY(ddt_object_count(ddt, type, class, &count) == 0 && count == 0); |
428870ff BB |
94 | VERIFY(zap_remove(os, DMU_POOL_DIRECTORY_OBJECT, name, tx) == 0); |
95 | VERIFY(zap_remove(os, spa->spa_ddt_stat_object, name, tx) == 0); | |
96 | VERIFY(ddt_ops[type]->ddt_op_destroy(os, *objectp, tx) == 0); | |
97 | bzero(&ddt->ddt_object_stats[type][class], sizeof (ddt_object_t)); | |
98 | ||
99 | *objectp = 0; | |
100 | } | |
101 | ||
102 | static int | |
103 | ddt_object_load(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
104 | { | |
105 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
106 | dmu_object_info_t doi; | |
e8fd45a0 | 107 | uint64_t count; |
428870ff BB |
108 | char name[DDT_NAMELEN]; |
109 | int error; | |
110 | ||
111 | ddt_object_name(ddt, type, class, name); | |
112 | ||
113 | error = zap_lookup(ddt->ddt_os, DMU_POOL_DIRECTORY_OBJECT, name, | |
114 | sizeof (uint64_t), 1, &ddt->ddt_object[type][class]); | |
115 | ||
116 | if (error) | |
117 | return (error); | |
118 | ||
119 | error = zap_lookup(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, | |
120 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
121 | &ddt->ddt_histogram[type][class]); | |
122 | ||
123 | /* | |
124 | * Seed the cached statistics. | |
125 | */ | |
2a4a9dc2 BB |
126 | error = ddt_object_info(ddt, type, class, &doi); |
127 | if (error) | |
128 | return (error); | |
428870ff | 129 | |
e8fd45a0 BB |
130 | error = ddt_object_count(ddt, type, class, &count); |
131 | if (error) | |
132 | return (error); | |
133 | ||
134 | ddo->ddo_count = count; | |
428870ff BB |
135 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
136 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
137 | ||
138 | ASSERT(error == 0); | |
139 | return (error); | |
140 | } | |
141 | ||
142 | static void | |
143 | ddt_object_sync(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
144 | dmu_tx_t *tx) | |
145 | { | |
146 | ddt_object_t *ddo = &ddt->ddt_object_stats[type][class]; | |
147 | dmu_object_info_t doi; | |
e8fd45a0 | 148 | uint64_t count; |
428870ff BB |
149 | char name[DDT_NAMELEN]; |
150 | ||
151 | ddt_object_name(ddt, type, class, name); | |
152 | ||
153 | VERIFY(zap_update(ddt->ddt_os, ddt->ddt_spa->spa_ddt_stat_object, name, | |
154 | sizeof (uint64_t), sizeof (ddt_histogram_t) / sizeof (uint64_t), | |
155 | &ddt->ddt_histogram[type][class], tx) == 0); | |
156 | ||
157 | /* | |
158 | * Cache DDT statistics; this is the only time they'll change. | |
159 | */ | |
160 | VERIFY(ddt_object_info(ddt, type, class, &doi) == 0); | |
e8fd45a0 | 161 | VERIFY(ddt_object_count(ddt, type, class, &count) == 0); |
428870ff | 162 | |
e8fd45a0 | 163 | ddo->ddo_count = count; |
428870ff BB |
164 | ddo->ddo_dspace = doi.doi_physical_blocks_512 << 9; |
165 | ddo->ddo_mspace = doi.doi_fill_count * doi.doi_data_block_size; | |
166 | } | |
167 | ||
168 | static int | |
169 | ddt_object_lookup(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
170 | ddt_entry_t *dde) | |
171 | { | |
172 | if (!ddt_object_exists(ddt, type, class)) | |
173 | return (ENOENT); | |
174 | ||
175 | return (ddt_ops[type]->ddt_op_lookup(ddt->ddt_os, | |
176 | ddt->ddt_object[type][class], dde)); | |
177 | } | |
178 | ||
179 | static void | |
180 | ddt_object_prefetch(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
181 | ddt_entry_t *dde) | |
182 | { | |
183 | if (!ddt_object_exists(ddt, type, class)) | |
184 | return; | |
185 | ||
186 | ddt_ops[type]->ddt_op_prefetch(ddt->ddt_os, | |
187 | ddt->ddt_object[type][class], dde); | |
188 | } | |
189 | ||
190 | int | |
191 | ddt_object_update(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
192 | ddt_entry_t *dde, dmu_tx_t *tx) | |
193 | { | |
194 | ASSERT(ddt_object_exists(ddt, type, class)); | |
195 | ||
196 | return (ddt_ops[type]->ddt_op_update(ddt->ddt_os, | |
197 | ddt->ddt_object[type][class], dde, tx)); | |
198 | } | |
199 | ||
200 | static int | |
201 | ddt_object_remove(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
202 | ddt_entry_t *dde, dmu_tx_t *tx) | |
203 | { | |
204 | ASSERT(ddt_object_exists(ddt, type, class)); | |
205 | ||
206 | return (ddt_ops[type]->ddt_op_remove(ddt->ddt_os, | |
207 | ddt->ddt_object[type][class], dde, tx)); | |
208 | } | |
209 | ||
210 | int | |
211 | ddt_object_walk(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
212 | uint64_t *walk, ddt_entry_t *dde) | |
213 | { | |
214 | ASSERT(ddt_object_exists(ddt, type, class)); | |
215 | ||
216 | return (ddt_ops[type]->ddt_op_walk(ddt->ddt_os, | |
217 | ddt->ddt_object[type][class], dde, walk)); | |
218 | } | |
219 | ||
e8fd45a0 BB |
220 | int |
221 | ddt_object_count(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
222 | uint64_t *count) | |
428870ff BB |
223 | { |
224 | ASSERT(ddt_object_exists(ddt, type, class)); | |
225 | ||
226 | return (ddt_ops[type]->ddt_op_count(ddt->ddt_os, | |
e8fd45a0 | 227 | ddt->ddt_object[type][class], count)); |
428870ff BB |
228 | } |
229 | ||
230 | int | |
231 | ddt_object_info(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
232 | dmu_object_info_t *doi) | |
233 | { | |
234 | if (!ddt_object_exists(ddt, type, class)) | |
235 | return (ENOENT); | |
236 | ||
237 | return (dmu_object_info(ddt->ddt_os, ddt->ddt_object[type][class], | |
238 | doi)); | |
239 | } | |
240 | ||
241 | boolean_t | |
242 | ddt_object_exists(ddt_t *ddt, enum ddt_type type, enum ddt_class class) | |
243 | { | |
244 | return (!!ddt->ddt_object[type][class]); | |
245 | } | |
246 | ||
247 | void | |
248 | ddt_object_name(ddt_t *ddt, enum ddt_type type, enum ddt_class class, | |
249 | char *name) | |
250 | { | |
251 | (void) sprintf(name, DMU_POOL_DDT, | |
252 | zio_checksum_table[ddt->ddt_checksum].ci_name, | |
253 | ddt_ops[type]->ddt_op_name, ddt_class_name[class]); | |
254 | } | |
255 | ||
256 | void | |
257 | ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, uint64_t txg) | |
258 | { | |
d6320ddb | 259 | int d; |
428870ff BB |
260 | ASSERT(txg != 0); |
261 | ||
d6320ddb | 262 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
263 | bp->blk_dva[d] = ddp->ddp_dva[d]; |
264 | BP_SET_BIRTH(bp, txg, ddp->ddp_phys_birth); | |
265 | } | |
266 | ||
267 | void | |
268 | ddt_bp_create(enum zio_checksum checksum, | |
269 | const ddt_key_t *ddk, const ddt_phys_t *ddp, blkptr_t *bp) | |
270 | { | |
271 | BP_ZERO(bp); | |
272 | ||
273 | if (ddp != NULL) | |
274 | ddt_bp_fill(ddp, bp, ddp->ddp_phys_birth); | |
275 | ||
276 | bp->blk_cksum = ddk->ddk_cksum; | |
277 | bp->blk_fill = 1; | |
278 | ||
279 | BP_SET_LSIZE(bp, DDK_GET_LSIZE(ddk)); | |
280 | BP_SET_PSIZE(bp, DDK_GET_PSIZE(ddk)); | |
281 | BP_SET_COMPRESS(bp, DDK_GET_COMPRESS(ddk)); | |
282 | BP_SET_CHECKSUM(bp, checksum); | |
283 | BP_SET_TYPE(bp, DMU_OT_DEDUP); | |
284 | BP_SET_LEVEL(bp, 0); | |
285 | BP_SET_DEDUP(bp, 0); | |
286 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); | |
287 | } | |
288 | ||
289 | void | |
290 | ddt_key_fill(ddt_key_t *ddk, const blkptr_t *bp) | |
291 | { | |
292 | ddk->ddk_cksum = bp->blk_cksum; | |
293 | ddk->ddk_prop = 0; | |
294 | ||
295 | DDK_SET_LSIZE(ddk, BP_GET_LSIZE(bp)); | |
296 | DDK_SET_PSIZE(ddk, BP_GET_PSIZE(bp)); | |
297 | DDK_SET_COMPRESS(ddk, BP_GET_COMPRESS(bp)); | |
298 | } | |
299 | ||
300 | void | |
301 | ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp) | |
302 | { | |
d6320ddb | 303 | int d; |
428870ff BB |
304 | ASSERT(ddp->ddp_phys_birth == 0); |
305 | ||
d6320ddb | 306 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
307 | ddp->ddp_dva[d] = bp->blk_dva[d]; |
308 | ddp->ddp_phys_birth = BP_PHYSICAL_BIRTH(bp); | |
309 | } | |
310 | ||
311 | void | |
312 | ddt_phys_clear(ddt_phys_t *ddp) | |
313 | { | |
314 | bzero(ddp, sizeof (*ddp)); | |
315 | } | |
316 | ||
317 | void | |
318 | ddt_phys_addref(ddt_phys_t *ddp) | |
319 | { | |
320 | ddp->ddp_refcnt++; | |
321 | } | |
322 | ||
323 | void | |
324 | ddt_phys_decref(ddt_phys_t *ddp) | |
325 | { | |
326 | ASSERT((int64_t)ddp->ddp_refcnt > 0); | |
327 | ddp->ddp_refcnt--; | |
328 | } | |
329 | ||
330 | void | |
331 | ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, uint64_t txg) | |
332 | { | |
333 | blkptr_t blk; | |
334 | ||
335 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
336 | ddt_phys_clear(ddp); | |
337 | zio_free(ddt->ddt_spa, txg, &blk); | |
338 | } | |
339 | ||
340 | ddt_phys_t * | |
341 | ddt_phys_select(const ddt_entry_t *dde, const blkptr_t *bp) | |
342 | { | |
343 | ddt_phys_t *ddp = (ddt_phys_t *)dde->dde_phys; | |
d6320ddb | 344 | int p; |
428870ff | 345 | |
d6320ddb | 346 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
347 | if (DVA_EQUAL(BP_IDENTITY(bp), &ddp->ddp_dva[0]) && |
348 | BP_PHYSICAL_BIRTH(bp) == ddp->ddp_phys_birth) | |
349 | return (ddp); | |
350 | } | |
351 | return (NULL); | |
352 | } | |
353 | ||
354 | uint64_t | |
355 | ddt_phys_total_refcnt(const ddt_entry_t *dde) | |
356 | { | |
357 | uint64_t refcnt = 0; | |
d6320ddb | 358 | int p; |
428870ff | 359 | |
d6320ddb | 360 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) |
428870ff BB |
361 | refcnt += dde->dde_phys[p].ddp_refcnt; |
362 | ||
363 | return (refcnt); | |
364 | } | |
365 | ||
366 | static void | |
367 | ddt_stat_generate(ddt_t *ddt, ddt_entry_t *dde, ddt_stat_t *dds) | |
368 | { | |
369 | spa_t *spa = ddt->ddt_spa; | |
370 | ddt_phys_t *ddp = dde->dde_phys; | |
371 | ddt_key_t *ddk = &dde->dde_key; | |
372 | uint64_t lsize = DDK_GET_LSIZE(ddk); | |
373 | uint64_t psize = DDK_GET_PSIZE(ddk); | |
d6320ddb | 374 | int p, d; |
428870ff BB |
375 | |
376 | bzero(dds, sizeof (*dds)); | |
377 | ||
d6320ddb | 378 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
379 | uint64_t dsize = 0; |
380 | uint64_t refcnt = ddp->ddp_refcnt; | |
381 | ||
382 | if (ddp->ddp_phys_birth == 0) | |
383 | continue; | |
384 | ||
d6320ddb | 385 | for (d = 0; d < SPA_DVAS_PER_BP; d++) |
428870ff BB |
386 | dsize += dva_get_dsize_sync(spa, &ddp->ddp_dva[d]); |
387 | ||
388 | dds->dds_blocks += 1; | |
389 | dds->dds_lsize += lsize; | |
390 | dds->dds_psize += psize; | |
391 | dds->dds_dsize += dsize; | |
392 | ||
393 | dds->dds_ref_blocks += refcnt; | |
394 | dds->dds_ref_lsize += lsize * refcnt; | |
395 | dds->dds_ref_psize += psize * refcnt; | |
396 | dds->dds_ref_dsize += dsize * refcnt; | |
397 | } | |
398 | } | |
399 | ||
400 | void | |
401 | ddt_stat_add(ddt_stat_t *dst, const ddt_stat_t *src, uint64_t neg) | |
402 | { | |
403 | const uint64_t *s = (const uint64_t *)src; | |
404 | uint64_t *d = (uint64_t *)dst; | |
405 | uint64_t *d_end = (uint64_t *)(dst + 1); | |
406 | ||
407 | ASSERT(neg == 0 || neg == -1ULL); /* add or subtract */ | |
408 | ||
409 | while (d < d_end) | |
410 | *d++ += (*s++ ^ neg) - neg; | |
411 | } | |
412 | ||
413 | static void | |
414 | ddt_stat_update(ddt_t *ddt, ddt_entry_t *dde, uint64_t neg) | |
415 | { | |
416 | ddt_stat_t dds; | |
417 | ddt_histogram_t *ddh; | |
418 | int bucket; | |
419 | ||
420 | ddt_stat_generate(ddt, dde, &dds); | |
421 | ||
422 | bucket = highbit(dds.dds_ref_blocks) - 1; | |
423 | ASSERT(bucket >= 0); | |
424 | ||
425 | ddh = &ddt->ddt_histogram[dde->dde_type][dde->dde_class]; | |
426 | ||
427 | ddt_stat_add(&ddh->ddh_stat[bucket], &dds, neg); | |
428 | } | |
429 | ||
430 | void | |
431 | ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src) | |
432 | { | |
d6320ddb BB |
433 | int h; |
434 | ||
435 | for (h = 0; h < 64; h++) | |
428870ff BB |
436 | ddt_stat_add(&dst->ddh_stat[h], &src->ddh_stat[h], 0); |
437 | } | |
438 | ||
439 | void | |
440 | ddt_histogram_stat(ddt_stat_t *dds, const ddt_histogram_t *ddh) | |
441 | { | |
d6320ddb BB |
442 | int h; |
443 | ||
428870ff BB |
444 | bzero(dds, sizeof (*dds)); |
445 | ||
d6320ddb | 446 | for (h = 0; h < 64; h++) |
428870ff BB |
447 | ddt_stat_add(dds, &ddh->ddh_stat[h], 0); |
448 | } | |
449 | ||
450 | boolean_t | |
451 | ddt_histogram_empty(const ddt_histogram_t *ddh) | |
452 | { | |
453 | const uint64_t *s = (const uint64_t *)ddh; | |
454 | const uint64_t *s_end = (const uint64_t *)(ddh + 1); | |
455 | ||
456 | while (s < s_end) | |
457 | if (*s++ != 0) | |
458 | return (B_FALSE); | |
459 | ||
460 | return (B_TRUE); | |
461 | } | |
462 | ||
463 | void | |
464 | ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo_total) | |
465 | { | |
d6320ddb BB |
466 | enum zio_checksum c; |
467 | enum ddt_type type; | |
468 | enum ddt_class class; | |
469 | ||
428870ff | 470 | /* Sum the statistics we cached in ddt_object_sync(). */ |
d6320ddb | 471 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 472 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
473 | for (type = 0; type < DDT_TYPES; type++) { |
474 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
475 | class++) { |
476 | ddt_object_t *ddo = | |
477 | &ddt->ddt_object_stats[type][class]; | |
478 | ddo_total->ddo_count += ddo->ddo_count; | |
479 | ddo_total->ddo_dspace += ddo->ddo_dspace; | |
480 | ddo_total->ddo_mspace += ddo->ddo_mspace; | |
481 | } | |
482 | } | |
483 | } | |
484 | ||
485 | /* ... and compute the averages. */ | |
486 | if (ddo_total->ddo_count != 0) { | |
487 | ddo_total->ddo_dspace /= ddo_total->ddo_count; | |
488 | ddo_total->ddo_mspace /= ddo_total->ddo_count; | |
428870ff BB |
489 | } |
490 | } | |
491 | ||
492 | void | |
493 | ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh) | |
494 | { | |
d6320ddb BB |
495 | enum zio_checksum c; |
496 | enum ddt_type type; | |
497 | enum ddt_class class; | |
498 | ||
499 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { | |
428870ff | 500 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
501 | for (type = 0; type < DDT_TYPES; type++) { |
502 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
503 | class++) { |
504 | ddt_histogram_add(ddh, | |
505 | &ddt->ddt_histogram_cache[type][class]); | |
506 | } | |
507 | } | |
508 | } | |
509 | } | |
510 | ||
511 | void | |
512 | ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total) | |
513 | { | |
514 | ddt_histogram_t *ddh_total; | |
515 | ||
00b46022 | 516 | /* XXX: Move to a slab */ |
b8d06fca | 517 | ddh_total = kmem_zalloc(sizeof (ddt_histogram_t), KM_PUSHPAGE); |
428870ff BB |
518 | ddt_get_dedup_histogram(spa, ddh_total); |
519 | ddt_histogram_stat(dds_total, ddh_total); | |
520 | kmem_free(ddh_total, sizeof (ddt_histogram_t)); | |
521 | } | |
522 | ||
523 | uint64_t | |
524 | ddt_get_dedup_dspace(spa_t *spa) | |
525 | { | |
526 | ddt_stat_t dds_total = { 0 }; | |
527 | ||
528 | ddt_get_dedup_stats(spa, &dds_total); | |
529 | return (dds_total.dds_ref_dsize - dds_total.dds_dsize); | |
530 | } | |
531 | ||
532 | uint64_t | |
533 | ddt_get_pool_dedup_ratio(spa_t *spa) | |
534 | { | |
535 | ddt_stat_t dds_total = { 0 }; | |
536 | ||
537 | ddt_get_dedup_stats(spa, &dds_total); | |
538 | if (dds_total.dds_dsize == 0) | |
539 | return (100); | |
540 | ||
541 | return (dds_total.dds_ref_dsize * 100 / dds_total.dds_dsize); | |
542 | } | |
543 | ||
544 | int | |
545 | ddt_ditto_copies_needed(ddt_t *ddt, ddt_entry_t *dde, ddt_phys_t *ddp_willref) | |
546 | { | |
547 | spa_t *spa = ddt->ddt_spa; | |
548 | uint64_t total_refcnt = 0; | |
549 | uint64_t ditto = spa->spa_dedup_ditto; | |
550 | int total_copies = 0; | |
551 | int desired_copies = 0; | |
d6320ddb | 552 | int p; |
428870ff | 553 | |
d6320ddb | 554 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
555 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
556 | zio_t *zio = dde->dde_lead_zio[p]; | |
557 | uint64_t refcnt = ddp->ddp_refcnt; /* committed refs */ | |
558 | if (zio != NULL) | |
559 | refcnt += zio->io_parent_count; /* pending refs */ | |
560 | if (ddp == ddp_willref) | |
561 | refcnt++; /* caller's ref */ | |
562 | if (refcnt != 0) { | |
563 | total_refcnt += refcnt; | |
564 | total_copies += p; | |
565 | } | |
566 | } | |
567 | ||
568 | if (ditto == 0 || ditto > UINT32_MAX) | |
569 | ditto = UINT32_MAX; | |
570 | ||
571 | if (total_refcnt >= 1) | |
572 | desired_copies++; | |
573 | if (total_refcnt >= ditto) | |
574 | desired_copies++; | |
575 | if (total_refcnt >= ditto * ditto) | |
576 | desired_copies++; | |
577 | ||
578 | return (MAX(desired_copies, total_copies) - total_copies); | |
579 | } | |
580 | ||
581 | int | |
582 | ddt_ditto_copies_present(ddt_entry_t *dde) | |
583 | { | |
584 | ddt_phys_t *ddp = &dde->dde_phys[DDT_PHYS_DITTO]; | |
585 | dva_t *dva = ddp->ddp_dva; | |
586 | int copies = 0 - DVA_GET_GANG(dva); | |
d6320ddb | 587 | int d; |
428870ff | 588 | |
d6320ddb | 589 | for (d = 0; d < SPA_DVAS_PER_BP; d++, dva++) |
428870ff BB |
590 | if (DVA_IS_VALID(dva)) |
591 | copies++; | |
592 | ||
593 | ASSERT(copies >= 0 && copies < SPA_DVAS_PER_BP); | |
594 | ||
595 | return (copies); | |
596 | } | |
597 | ||
598 | size_t | |
599 | ddt_compress(void *src, uchar_t *dst, size_t s_len, size_t d_len) | |
600 | { | |
601 | uchar_t *version = dst++; | |
602 | int cpfunc = ZIO_COMPRESS_ZLE; | |
603 | zio_compress_info_t *ci = &zio_compress_table[cpfunc]; | |
604 | size_t c_len; | |
605 | ||
606 | ASSERT(d_len >= s_len + 1); /* no compression plus version byte */ | |
607 | ||
608 | c_len = ci->ci_compress(src, dst, s_len, d_len - 1, ci->ci_level); | |
609 | ||
610 | if (c_len == s_len) { | |
611 | cpfunc = ZIO_COMPRESS_OFF; | |
612 | bcopy(src, dst, s_len); | |
613 | } | |
614 | ||
615 | *version = (ZFS_HOST_BYTEORDER & DDT_COMPRESS_BYTEORDER_MASK) | cpfunc; | |
616 | ||
617 | return (c_len + 1); | |
618 | } | |
619 | ||
620 | void | |
621 | ddt_decompress(uchar_t *src, void *dst, size_t s_len, size_t d_len) | |
622 | { | |
623 | uchar_t version = *src++; | |
624 | int cpfunc = version & DDT_COMPRESS_FUNCTION_MASK; | |
625 | zio_compress_info_t *ci = &zio_compress_table[cpfunc]; | |
626 | ||
627 | if (ci->ci_decompress != NULL) | |
628 | (void) ci->ci_decompress(src, dst, s_len, d_len, ci->ci_level); | |
629 | else | |
630 | bcopy(src, dst, d_len); | |
631 | ||
632 | if ((version ^ ZFS_HOST_BYTEORDER) & DDT_COMPRESS_BYTEORDER_MASK) | |
633 | byteswap_uint64_array(dst, d_len); | |
634 | } | |
635 | ||
636 | ddt_t * | |
637 | ddt_select_by_checksum(spa_t *spa, enum zio_checksum c) | |
638 | { | |
639 | return (spa->spa_ddt[c]); | |
640 | } | |
641 | ||
642 | ddt_t * | |
643 | ddt_select(spa_t *spa, const blkptr_t *bp) | |
644 | { | |
645 | return (spa->spa_ddt[BP_GET_CHECKSUM(bp)]); | |
646 | } | |
647 | ||
648 | void | |
649 | ddt_enter(ddt_t *ddt) | |
650 | { | |
651 | mutex_enter(&ddt->ddt_lock); | |
652 | } | |
653 | ||
654 | void | |
655 | ddt_exit(ddt_t *ddt) | |
656 | { | |
657 | mutex_exit(&ddt->ddt_lock); | |
658 | } | |
659 | ||
660 | static ddt_entry_t * | |
661 | ddt_alloc(const ddt_key_t *ddk) | |
662 | { | |
663 | ddt_entry_t *dde; | |
664 | ||
00b46022 | 665 | /* XXX: Move to a slab */ |
95fd8c9a | 666 | dde = kmem_zalloc(sizeof (ddt_entry_t), KM_PUSHPAGE); |
428870ff BB |
667 | cv_init(&dde->dde_cv, NULL, CV_DEFAULT, NULL); |
668 | ||
669 | dde->dde_key = *ddk; | |
670 | ||
671 | return (dde); | |
672 | } | |
673 | ||
674 | static void | |
675 | ddt_free(ddt_entry_t *dde) | |
676 | { | |
d6320ddb BB |
677 | int p; |
678 | ||
428870ff BB |
679 | ASSERT(!dde->dde_loading); |
680 | ||
d6320ddb | 681 | for (p = 0; p < DDT_PHYS_TYPES; p++) |
428870ff BB |
682 | ASSERT(dde->dde_lead_zio[p] == NULL); |
683 | ||
684 | if (dde->dde_repair_data != NULL) | |
685 | zio_buf_free(dde->dde_repair_data, | |
686 | DDK_GET_PSIZE(&dde->dde_key)); | |
687 | ||
688 | cv_destroy(&dde->dde_cv); | |
689 | kmem_free(dde, sizeof (*dde)); | |
690 | } | |
691 | ||
692 | void | |
693 | ddt_remove(ddt_t *ddt, ddt_entry_t *dde) | |
694 | { | |
695 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
696 | ||
697 | avl_remove(&ddt->ddt_tree, dde); | |
698 | ddt_free(dde); | |
699 | } | |
700 | ||
701 | ddt_entry_t * | |
702 | ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add) | |
703 | { | |
704 | ddt_entry_t *dde, dde_search; | |
705 | enum ddt_type type; | |
706 | enum ddt_class class; | |
707 | avl_index_t where; | |
708 | int error; | |
709 | ||
710 | ASSERT(MUTEX_HELD(&ddt->ddt_lock)); | |
711 | ||
712 | ddt_key_fill(&dde_search.dde_key, bp); | |
713 | ||
714 | dde = avl_find(&ddt->ddt_tree, &dde_search, &where); | |
715 | if (dde == NULL) { | |
716 | if (!add) | |
717 | return (NULL); | |
718 | dde = ddt_alloc(&dde_search.dde_key); | |
719 | avl_insert(&ddt->ddt_tree, dde, where); | |
720 | } | |
721 | ||
722 | while (dde->dde_loading) | |
723 | cv_wait(&dde->dde_cv, &ddt->ddt_lock); | |
724 | ||
725 | if (dde->dde_loaded) | |
726 | return (dde); | |
727 | ||
728 | dde->dde_loading = B_TRUE; | |
729 | ||
730 | ddt_exit(ddt); | |
731 | ||
732 | error = ENOENT; | |
733 | ||
734 | for (type = 0; type < DDT_TYPES; type++) { | |
735 | for (class = 0; class < DDT_CLASSES; class++) { | |
736 | error = ddt_object_lookup(ddt, type, class, dde); | |
737 | if (error != ENOENT) | |
738 | break; | |
739 | } | |
740 | if (error != ENOENT) | |
741 | break; | |
742 | } | |
743 | ||
744 | ASSERT(error == 0 || error == ENOENT); | |
745 | ||
746 | ddt_enter(ddt); | |
747 | ||
748 | ASSERT(dde->dde_loaded == B_FALSE); | |
749 | ASSERT(dde->dde_loading == B_TRUE); | |
750 | ||
751 | dde->dde_type = type; /* will be DDT_TYPES if no entry found */ | |
752 | dde->dde_class = class; /* will be DDT_CLASSES if no entry found */ | |
753 | dde->dde_loaded = B_TRUE; | |
754 | dde->dde_loading = B_FALSE; | |
755 | ||
756 | if (error == 0) | |
757 | ddt_stat_update(ddt, dde, -1ULL); | |
758 | ||
759 | cv_broadcast(&dde->dde_cv); | |
760 | ||
761 | return (dde); | |
762 | } | |
763 | ||
764 | void | |
765 | ddt_prefetch(spa_t *spa, const blkptr_t *bp) | |
766 | { | |
767 | ddt_t *ddt; | |
768 | ddt_entry_t dde; | |
d6320ddb BB |
769 | enum ddt_type type; |
770 | enum ddt_class class; | |
428870ff | 771 | |
572e2857 | 772 | if (!zfs_dedup_prefetch || bp == NULL || !BP_GET_DEDUP(bp)) |
428870ff BB |
773 | return; |
774 | ||
775 | /* | |
572e2857 BB |
776 | * We only remove the DDT once all tables are empty and only |
777 | * prefetch dedup blocks when there are entries in the DDT. | |
778 | * Thus no locking is required as the DDT can't disappear on us. | |
428870ff BB |
779 | */ |
780 | ddt = ddt_select(spa, bp); | |
781 | ddt_key_fill(&dde.dde_key, bp); | |
782 | ||
d6320ddb BB |
783 | for (type = 0; type < DDT_TYPES; type++) { |
784 | for (class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
785 | ddt_object_prefetch(ddt, type, class, &dde); |
786 | } | |
787 | } | |
788 | } | |
789 | ||
790 | int | |
791 | ddt_entry_compare(const void *x1, const void *x2) | |
792 | { | |
793 | const ddt_entry_t *dde1 = x1; | |
794 | const ddt_entry_t *dde2 = x2; | |
795 | const uint64_t *u1 = (const uint64_t *)&dde1->dde_key; | |
796 | const uint64_t *u2 = (const uint64_t *)&dde2->dde_key; | |
d6320ddb | 797 | int i; |
428870ff | 798 | |
d6320ddb | 799 | for (i = 0; i < DDT_KEY_WORDS; i++) { |
428870ff BB |
800 | if (u1[i] < u2[i]) |
801 | return (-1); | |
802 | if (u1[i] > u2[i]) | |
803 | return (1); | |
804 | } | |
805 | ||
806 | return (0); | |
807 | } | |
808 | ||
809 | static ddt_t * | |
810 | ddt_table_alloc(spa_t *spa, enum zio_checksum c) | |
811 | { | |
812 | ddt_t *ddt; | |
813 | ||
00b46022 | 814 | /* XXX: Move to a slab */ |
95fd8c9a | 815 | ddt = kmem_zalloc(sizeof (*ddt), KM_PUSHPAGE | KM_NODEBUG); |
428870ff BB |
816 | |
817 | mutex_init(&ddt->ddt_lock, NULL, MUTEX_DEFAULT, NULL); | |
818 | avl_create(&ddt->ddt_tree, ddt_entry_compare, | |
819 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); | |
820 | avl_create(&ddt->ddt_repair_tree, ddt_entry_compare, | |
821 | sizeof (ddt_entry_t), offsetof(ddt_entry_t, dde_node)); | |
822 | ddt->ddt_checksum = c; | |
823 | ddt->ddt_spa = spa; | |
824 | ddt->ddt_os = spa->spa_meta_objset; | |
825 | ||
826 | return (ddt); | |
827 | } | |
828 | ||
829 | static void | |
830 | ddt_table_free(ddt_t *ddt) | |
831 | { | |
832 | ASSERT(avl_numnodes(&ddt->ddt_tree) == 0); | |
833 | ASSERT(avl_numnodes(&ddt->ddt_repair_tree) == 0); | |
834 | avl_destroy(&ddt->ddt_tree); | |
835 | avl_destroy(&ddt->ddt_repair_tree); | |
836 | mutex_destroy(&ddt->ddt_lock); | |
837 | kmem_free(ddt, sizeof (*ddt)); | |
838 | } | |
839 | ||
840 | void | |
841 | ddt_create(spa_t *spa) | |
842 | { | |
d6320ddb BB |
843 | enum zio_checksum c; |
844 | ||
428870ff BB |
845 | spa->spa_dedup_checksum = ZIO_DEDUPCHECKSUM; |
846 | ||
d6320ddb | 847 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) |
428870ff BB |
848 | spa->spa_ddt[c] = ddt_table_alloc(spa, c); |
849 | } | |
850 | ||
851 | int | |
852 | ddt_load(spa_t *spa) | |
853 | { | |
d6320ddb BB |
854 | enum zio_checksum c; |
855 | enum ddt_type type; | |
856 | enum ddt_class class; | |
428870ff BB |
857 | int error; |
858 | ||
859 | ddt_create(spa); | |
860 | ||
861 | error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
862 | DMU_POOL_DDT_STATS, sizeof (uint64_t), 1, | |
863 | &spa->spa_ddt_stat_object); | |
864 | ||
865 | if (error) | |
866 | return (error == ENOENT ? 0 : error); | |
867 | ||
d6320ddb | 868 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff | 869 | ddt_t *ddt = spa->spa_ddt[c]; |
d6320ddb BB |
870 | for (type = 0; type < DDT_TYPES; type++) { |
871 | for (class = 0; class < DDT_CLASSES; | |
428870ff BB |
872 | class++) { |
873 | error = ddt_object_load(ddt, type, class); | |
874 | if (error != 0 && error != ENOENT) | |
875 | return (error); | |
876 | } | |
877 | } | |
878 | ||
879 | /* | |
880 | * Seed the cached histograms. | |
881 | */ | |
882 | bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache, | |
883 | sizeof (ddt->ddt_histogram)); | |
884 | } | |
885 | ||
886 | return (0); | |
887 | } | |
888 | ||
889 | void | |
890 | ddt_unload(spa_t *spa) | |
891 | { | |
d6320ddb BB |
892 | enum zio_checksum c; |
893 | ||
894 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { | |
428870ff BB |
895 | if (spa->spa_ddt[c]) { |
896 | ddt_table_free(spa->spa_ddt[c]); | |
897 | spa->spa_ddt[c] = NULL; | |
898 | } | |
899 | } | |
900 | } | |
901 | ||
902 | boolean_t | |
903 | ddt_class_contains(spa_t *spa, enum ddt_class max_class, const blkptr_t *bp) | |
904 | { | |
905 | ddt_t *ddt; | |
e95b3bdc | 906 | ddt_entry_t *dde; |
d6320ddb BB |
907 | enum ddt_type type; |
908 | enum ddt_class class; | |
428870ff BB |
909 | |
910 | if (!BP_GET_DEDUP(bp)) | |
911 | return (B_FALSE); | |
912 | ||
913 | if (max_class == DDT_CLASS_UNIQUE) | |
914 | return (B_TRUE); | |
915 | ||
916 | ddt = spa->spa_ddt[BP_GET_CHECKSUM(bp)]; | |
95fd8c9a | 917 | dde = kmem_alloc(sizeof(ddt_entry_t), KM_PUSHPAGE); |
428870ff | 918 | |
e95b3bdc | 919 | ddt_key_fill(&(dde->dde_key), bp); |
428870ff | 920 | |
e95b3bdc BB |
921 | for (type = 0; type < DDT_TYPES; type++) { |
922 | for (class = 0; class <= max_class; class++) { | |
923 | if (ddt_object_lookup(ddt, type, class, dde) == 0) { | |
924 | kmem_free(dde, sizeof(ddt_entry_t)); | |
428870ff | 925 | return (B_TRUE); |
e95b3bdc BB |
926 | } |
927 | } | |
928 | } | |
428870ff | 929 | |
e95b3bdc | 930 | kmem_free(dde, sizeof(ddt_entry_t)); |
428870ff BB |
931 | return (B_FALSE); |
932 | } | |
933 | ||
934 | ddt_entry_t * | |
935 | ddt_repair_start(ddt_t *ddt, const blkptr_t *bp) | |
936 | { | |
937 | ddt_key_t ddk; | |
938 | ddt_entry_t *dde; | |
d6320ddb BB |
939 | enum ddt_type type; |
940 | enum ddt_class class; | |
428870ff BB |
941 | |
942 | ddt_key_fill(&ddk, bp); | |
943 | ||
944 | dde = ddt_alloc(&ddk); | |
945 | ||
d6320ddb BB |
946 | for (type = 0; type < DDT_TYPES; type++) { |
947 | for (class = 0; class < DDT_CLASSES; class++) { | |
428870ff BB |
948 | /* |
949 | * We can only do repair if there are multiple copies | |
950 | * of the block. For anything in the UNIQUE class, | |
951 | * there's definitely only one copy, so don't even try. | |
952 | */ | |
953 | if (class != DDT_CLASS_UNIQUE && | |
954 | ddt_object_lookup(ddt, type, class, dde) == 0) | |
955 | return (dde); | |
956 | } | |
957 | } | |
958 | ||
959 | bzero(dde->dde_phys, sizeof (dde->dde_phys)); | |
960 | ||
961 | return (dde); | |
962 | } | |
963 | ||
964 | void | |
965 | ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde) | |
966 | { | |
967 | avl_index_t where; | |
968 | ||
969 | ddt_enter(ddt); | |
970 | ||
971 | if (dde->dde_repair_data != NULL && spa_writeable(ddt->ddt_spa) && | |
972 | avl_find(&ddt->ddt_repair_tree, dde, &where) == NULL) | |
973 | avl_insert(&ddt->ddt_repair_tree, dde, where); | |
974 | else | |
975 | ddt_free(dde); | |
976 | ||
977 | ddt_exit(ddt); | |
978 | } | |
979 | ||
980 | static void | |
981 | ddt_repair_entry_done(zio_t *zio) | |
982 | { | |
983 | ddt_entry_t *rdde = zio->io_private; | |
984 | ||
985 | ddt_free(rdde); | |
986 | } | |
987 | ||
988 | static void | |
989 | ddt_repair_entry(ddt_t *ddt, ddt_entry_t *dde, ddt_entry_t *rdde, zio_t *rio) | |
990 | { | |
991 | ddt_phys_t *ddp = dde->dde_phys; | |
992 | ddt_phys_t *rddp = rdde->dde_phys; | |
993 | ddt_key_t *ddk = &dde->dde_key; | |
994 | ddt_key_t *rddk = &rdde->dde_key; | |
995 | zio_t *zio; | |
996 | blkptr_t blk; | |
d6320ddb | 997 | int p; |
428870ff BB |
998 | |
999 | zio = zio_null(rio, rio->io_spa, NULL, | |
1000 | ddt_repair_entry_done, rdde, rio->io_flags); | |
1001 | ||
d6320ddb | 1002 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++, rddp++) { |
428870ff BB |
1003 | if (ddp->ddp_phys_birth == 0 || |
1004 | ddp->ddp_phys_birth != rddp->ddp_phys_birth || | |
1005 | bcmp(ddp->ddp_dva, rddp->ddp_dva, sizeof (ddp->ddp_dva))) | |
1006 | continue; | |
1007 | ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk); | |
1008 | zio_nowait(zio_rewrite(zio, zio->io_spa, 0, &blk, | |
1009 | rdde->dde_repair_data, DDK_GET_PSIZE(rddk), NULL, NULL, | |
1010 | ZIO_PRIORITY_SYNC_WRITE, ZIO_DDT_CHILD_FLAGS(zio), NULL)); | |
1011 | } | |
1012 | ||
1013 | zio_nowait(zio); | |
1014 | } | |
1015 | ||
1016 | static void | |
1017 | ddt_repair_table(ddt_t *ddt, zio_t *rio) | |
1018 | { | |
1019 | spa_t *spa = ddt->ddt_spa; | |
1020 | ddt_entry_t *dde, *rdde_next, *rdde; | |
1021 | avl_tree_t *t = &ddt->ddt_repair_tree; | |
1022 | blkptr_t blk; | |
1023 | ||
1024 | if (spa_sync_pass(spa) > 1) | |
1025 | return; | |
1026 | ||
1027 | ddt_enter(ddt); | |
1028 | for (rdde = avl_first(t); rdde != NULL; rdde = rdde_next) { | |
1029 | rdde_next = AVL_NEXT(t, rdde); | |
1030 | avl_remove(&ddt->ddt_repair_tree, rdde); | |
1031 | ddt_exit(ddt); | |
1032 | ddt_bp_create(ddt->ddt_checksum, &rdde->dde_key, NULL, &blk); | |
1033 | dde = ddt_repair_start(ddt, &blk); | |
1034 | ddt_repair_entry(ddt, dde, rdde, rio); | |
1035 | ddt_repair_done(ddt, dde); | |
1036 | ddt_enter(ddt); | |
1037 | } | |
1038 | ddt_exit(ddt); | |
1039 | } | |
1040 | ||
1041 | static void | |
1042 | ddt_sync_entry(ddt_t *ddt, ddt_entry_t *dde, dmu_tx_t *tx, uint64_t txg) | |
1043 | { | |
1044 | dsl_pool_t *dp = ddt->ddt_spa->spa_dsl_pool; | |
1045 | ddt_phys_t *ddp = dde->dde_phys; | |
1046 | ddt_key_t *ddk = &dde->dde_key; | |
1047 | enum ddt_type otype = dde->dde_type; | |
1048 | enum ddt_type ntype = DDT_TYPE_CURRENT; | |
1049 | enum ddt_class oclass = dde->dde_class; | |
1050 | enum ddt_class nclass; | |
1051 | uint64_t total_refcnt = 0; | |
d6320ddb | 1052 | int p; |
428870ff BB |
1053 | |
1054 | ASSERT(dde->dde_loaded); | |
1055 | ASSERT(!dde->dde_loading); | |
1056 | ||
d6320ddb | 1057 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
1058 | ASSERT(dde->dde_lead_zio[p] == NULL); |
1059 | ASSERT((int64_t)ddp->ddp_refcnt >= 0); | |
1060 | if (ddp->ddp_phys_birth == 0) { | |
1061 | ASSERT(ddp->ddp_refcnt == 0); | |
1062 | continue; | |
1063 | } | |
1064 | if (p == DDT_PHYS_DITTO) { | |
1065 | if (ddt_ditto_copies_needed(ddt, dde, NULL) == 0) | |
1066 | ddt_phys_free(ddt, ddk, ddp, txg); | |
1067 | continue; | |
1068 | } | |
1069 | if (ddp->ddp_refcnt == 0) | |
1070 | ddt_phys_free(ddt, ddk, ddp, txg); | |
1071 | total_refcnt += ddp->ddp_refcnt; | |
1072 | } | |
1073 | ||
1074 | if (dde->dde_phys[DDT_PHYS_DITTO].ddp_phys_birth != 0) | |
1075 | nclass = DDT_CLASS_DITTO; | |
1076 | else if (total_refcnt > 1) | |
1077 | nclass = DDT_CLASS_DUPLICATE; | |
1078 | else | |
1079 | nclass = DDT_CLASS_UNIQUE; | |
1080 | ||
1081 | if (otype != DDT_TYPES && | |
1082 | (otype != ntype || oclass != nclass || total_refcnt == 0)) { | |
1083 | VERIFY(ddt_object_remove(ddt, otype, oclass, dde, tx) == 0); | |
1084 | ASSERT(ddt_object_lookup(ddt, otype, oclass, dde) == ENOENT); | |
1085 | } | |
1086 | ||
1087 | if (total_refcnt != 0) { | |
1088 | dde->dde_type = ntype; | |
1089 | dde->dde_class = nclass; | |
1090 | ddt_stat_update(ddt, dde, 0); | |
1091 | if (!ddt_object_exists(ddt, ntype, nclass)) | |
1092 | ddt_object_create(ddt, ntype, nclass, tx); | |
1093 | VERIFY(ddt_object_update(ddt, ntype, nclass, dde, tx) == 0); | |
1094 | ||
1095 | /* | |
1096 | * If the class changes, the order that we scan this bp | |
1097 | * changes. If it decreases, we could miss it, so | |
1098 | * scan it right now. (This covers both class changing | |
1099 | * while we are doing ddt_walk(), and when we are | |
1100 | * traversing.) | |
1101 | */ | |
1102 | if (nclass < oclass) { | |
1103 | dsl_scan_ddt_entry(dp->dp_scan, | |
1104 | ddt->ddt_checksum, dde, tx); | |
1105 | } | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | static void | |
1110 | ddt_sync_table(ddt_t *ddt, dmu_tx_t *tx, uint64_t txg) | |
1111 | { | |
1112 | spa_t *spa = ddt->ddt_spa; | |
1113 | ddt_entry_t *dde; | |
1114 | void *cookie = NULL; | |
d6320ddb BB |
1115 | enum ddt_type type; |
1116 | enum ddt_class class; | |
428870ff BB |
1117 | |
1118 | if (avl_numnodes(&ddt->ddt_tree) == 0) | |
1119 | return; | |
1120 | ||
1121 | ASSERT(spa->spa_uberblock.ub_version >= SPA_VERSION_DEDUP); | |
1122 | ||
1123 | if (spa->spa_ddt_stat_object == 0) { | |
9ae529ec CS |
1124 | spa->spa_ddt_stat_object = zap_create_link(ddt->ddt_os, |
1125 | DMU_OT_DDT_STATS, DMU_POOL_DIRECTORY_OBJECT, | |
1126 | DMU_POOL_DDT_STATS, tx); | |
428870ff BB |
1127 | } |
1128 | ||
1129 | while ((dde = avl_destroy_nodes(&ddt->ddt_tree, &cookie)) != NULL) { | |
1130 | ddt_sync_entry(ddt, dde, tx, txg); | |
1131 | ddt_free(dde); | |
1132 | } | |
1133 | ||
d6320ddb | 1134 | for (type = 0; type < DDT_TYPES; type++) { |
e8fd45a0 | 1135 | uint64_t add, count = 0; |
d6320ddb | 1136 | for (class = 0; class < DDT_CLASSES; class++) { |
572e2857 BB |
1137 | if (ddt_object_exists(ddt, type, class)) { |
1138 | ddt_object_sync(ddt, type, class, tx); | |
e8fd45a0 BB |
1139 | VERIFY(ddt_object_count(ddt, type, class, |
1140 | &add) == 0); | |
1141 | count += add; | |
572e2857 BB |
1142 | } |
1143 | } | |
d6320ddb | 1144 | for (class = 0; class < DDT_CLASSES; class++) { |
572e2857 | 1145 | if (count == 0 && ddt_object_exists(ddt, type, class)) |
428870ff BB |
1146 | ddt_object_destroy(ddt, type, class, tx); |
1147 | } | |
1148 | } | |
1149 | ||
1150 | bcopy(ddt->ddt_histogram, &ddt->ddt_histogram_cache, | |
1151 | sizeof (ddt->ddt_histogram)); | |
1152 | } | |
1153 | ||
1154 | void | |
1155 | ddt_sync(spa_t *spa, uint64_t txg) | |
1156 | { | |
1157 | dmu_tx_t *tx; | |
1158 | zio_t *rio = zio_root(spa, NULL, NULL, | |
1159 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE); | |
d6320ddb | 1160 | enum zio_checksum c; |
428870ff BB |
1161 | |
1162 | ASSERT(spa_syncing_txg(spa) == txg); | |
1163 | ||
1164 | tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); | |
1165 | ||
d6320ddb | 1166 | for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) { |
428870ff BB |
1167 | ddt_t *ddt = spa->spa_ddt[c]; |
1168 | if (ddt == NULL) | |
1169 | continue; | |
1170 | ddt_sync_table(ddt, tx, txg); | |
1171 | ddt_repair_table(ddt, rio); | |
1172 | } | |
1173 | ||
1174 | (void) zio_wait(rio); | |
1175 | ||
1176 | dmu_tx_commit(tx); | |
1177 | } | |
1178 | ||
1179 | int | |
1180 | ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, ddt_entry_t *dde) | |
1181 | { | |
1182 | do { | |
1183 | do { | |
1184 | do { | |
1185 | ddt_t *ddt = spa->spa_ddt[ddb->ddb_checksum]; | |
1186 | int error = ENOENT; | |
1187 | if (ddt_object_exists(ddt, ddb->ddb_type, | |
1188 | ddb->ddb_class)) { | |
1189 | error = ddt_object_walk(ddt, | |
1190 | ddb->ddb_type, ddb->ddb_class, | |
1191 | &ddb->ddb_cursor, dde); | |
1192 | } | |
1193 | dde->dde_type = ddb->ddb_type; | |
1194 | dde->dde_class = ddb->ddb_class; | |
1195 | if (error == 0) | |
1196 | return (0); | |
1197 | if (error != ENOENT) | |
1198 | return (error); | |
1199 | ddb->ddb_cursor = 0; | |
1200 | } while (++ddb->ddb_checksum < ZIO_CHECKSUM_FUNCTIONS); | |
1201 | ddb->ddb_checksum = 0; | |
1202 | } while (++ddb->ddb_type < DDT_TYPES); | |
1203 | ddb->ddb_type = 0; | |
1204 | } while (++ddb->ddb_class < DDT_CLASSES); | |
1205 | ||
1206 | return (ENOENT); | |
1207 | } | |
c409e464 BB |
1208 | |
1209 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
1210 | module_param(zfs_dedup_prefetch, int, 0644); | |
1211 | MODULE_PARM_DESC(zfs_dedup_prefetch,"Enable prefetching dedup-ed blks"); | |
1212 | #endif |