1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * CDDL HEADER START 4 * 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or https://opensource.org/licenses/CDDL-1.0. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2016 by Delphix. All rights reserved. 25 * Copyright (c) 2023, Klara Inc. 26 */ 27 28 #ifndef _SYS_DDT_H 29 #define _SYS_DDT_H 30 31 #include <sys/sysmacros.h> 32 #include <sys/types.h> 33 #include <sys/fs/zfs.h> 34 #include <sys/zio.h> 35 #include <sys/dmu.h> 36 37 #ifdef __cplusplus 38 extern "C" { 39 #endif 40 41 struct abd; 42 43 /* 44 * DDT-wide feature flags. These are set in ddt_flags by ddt_configure(). 45 */ 46 #define DDT_FLAG_FLAT (1 << 0) /* single extensible phys */ 47 #define DDT_FLAG_LOG (1 << 1) /* dedup log (journal) */ 48 #define DDT_FLAG_MASK (DDT_FLAG_FLAT|DDT_FLAG_LOG) 49 50 /* 51 * DDT on-disk storage object types. Each one corresponds to specific 52 * implementation, see ddt_ops_t. The value itself is not stored on disk. 53 * 54 * When searching for an entry, objects types will be searched in this order. 55 * 56 * Note that DDT_TYPES is used as the "no type" for new entries that have not 57 * yet been written to a storage object. 58 */ 59 typedef enum { 60 DDT_TYPE_ZAP = 0, /* ZAP storage object, ddt_zap */ 61 DDT_TYPES 62 } ddt_type_t; 63 64 _Static_assert(DDT_TYPES <= UINT8_MAX, 65 "ddt_type_t must fit in a uint8_t"); 66 67 /* New and updated entries recieve this type, see ddt_sync_entry() */ 68 #define DDT_TYPE_DEFAULT (DDT_TYPE_ZAP) 69 70 /* 71 * DDT storage classes. Each class has a separate storage object for each type. 72 * The value itself is not stored on disk. 73 * 74 * When search for an entry, object classes will be searched in this order. 75 * 76 * Note that DDT_CLASSES is used as the "no class" for new entries that have not 77 * yet been written to a storage object. 78 */ 79 typedef enum { 80 DDT_CLASS_DITTO = 0, /* entry has ditto blocks (obsolete) */ 81 DDT_CLASS_DUPLICATE, /* entry has multiple references */ 82 DDT_CLASS_UNIQUE, /* entry has a single reference */ 83 DDT_CLASSES 84 } ddt_class_t; 85 86 _Static_assert(DDT_CLASSES < UINT8_MAX, 87 "ddt_class_t must fit in a uint8_t"); 88 89 /* 90 * The "key" part of an on-disk entry. This is the unique "name" for a block, 91 * that is, that parts of the block pointer that will always be the same for 92 * the same data. 93 */ 94 typedef struct { 95 zio_cksum_t ddk_cksum; /* 256-bit block checksum */ 96 /* 97 * Encoded with logical & physical size, encryption, and compression, 98 * as follows: 99 * +-------+-------+-------+-------+-------+-------+-------+-------+ 100 * | 0 | 0 | 0 |X| comp| PSIZE | LSIZE | 101 * +-------+-------+-------+-------+-------+-------+-------+-------+ 102 */ 103 uint64_t ddk_prop; 104 } ddt_key_t; 105 106 /* 107 * Macros for accessing parts of a ddt_key_t. These are similar to their BP_* 108 * counterparts. 109 */ 110 #define DDK_GET_LSIZE(ddk) \ 111 BF64_GET_SB((ddk)->ddk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1) 112 #define DDK_SET_LSIZE(ddk, x) \ 113 BF64_SET_SB((ddk)->ddk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1, x) 114 115 #define DDK_GET_PSIZE(ddk) \ 116 BF64_GET_SB((ddk)->ddk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1) 117 #define DDK_SET_PSIZE(ddk, x) \ 118 BF64_SET_SB((ddk)->ddk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1, x) 119 120 #define DDK_GET_COMPRESS(ddk) BF64_GET((ddk)->ddk_prop, 32, 7) 121 #define DDK_SET_COMPRESS(ddk, x) BF64_SET((ddk)->ddk_prop, 32, 7, x) 122 123 #define DDK_GET_CRYPT(ddk) BF64_GET((ddk)->ddk_prop, 39, 1) 124 #define DDK_SET_CRYPT(ddk, x) BF64_SET((ddk)->ddk_prop, 39, 1, x) 125 126 /* 127 * The "value" part for an on-disk entry. These are the "physical" 128 * characteristics of the stored block, such as its location on disk (DVAs), 129 * birth txg and ref count. 130 * 131 * The "traditional" entry has an array of four, one for each number of DVAs 132 * (copies= property) and another for additional "ditto" copies. Users of the 133 * traditional struct will specify the variant (index) of the one they want. 134 * 135 * The newer "flat" entry has only a single form that is specified using the 136 * DDT_PHYS_FLAT variant. 137 * 138 * Since the value size varies, use one of the size macros when interfacing 139 * with the ddt zap. 140 */ 141 142 #define DDT_PHYS_MAX (4) 143 144 /* 145 * Note - this can be used in a flexible array and allocated for 146 * a specific size (ddp_trad or ddp_flat). So be careful not to 147 * copy using "=" assignment but instead use ddt_phys_copy(). 148 */ 149 typedef union { 150 /* 151 * Traditional physical payload value for DDT zap (256 bytes) 152 */ 153 struct { 154 dva_t ddp_dva[SPA_DVAS_PER_BP]; 155 uint64_t ddp_refcnt; 156 uint64_t ddp_phys_birth; 157 } ddp_trad[DDT_PHYS_MAX]; 158 159 /* 160 * Flat physical payload value for DDT zap (72 bytes) 161 */ 162 struct { 163 dva_t ddp_dva[SPA_DVAS_PER_BP]; 164 uint64_t ddp_refcnt; 165 uint64_t ddp_phys_birth; /* txg based from BP */ 166 uint64_t ddp_class_start; /* in realtime seconds */ 167 } ddp_flat; 168 } ddt_univ_phys_t; 169 170 /* 171 * This enum denotes which variant of a ddt_univ_phys_t to target. For 172 * a traditional DDT entry, it represents the indexes into the ddp_trad 173 * array. Any consumer of a ddt_univ_phys_t needs to know which variant 174 * is being targeted. 175 * 176 * Note, we no longer generate new DDT_PHYS_DITTO-type blocks. However, 177 * we maintain the ability to free existing dedup-ditto blocks. 178 */ 179 180 typedef enum { 181 DDT_PHYS_DITTO = 0, 182 DDT_PHYS_SINGLE = 1, 183 DDT_PHYS_DOUBLE = 2, 184 DDT_PHYS_TRIPLE = 3, 185 DDT_PHYS_FLAT = 4, 186 DDT_PHYS_NONE = 5 187 } ddt_phys_variant_t; 188 189 #define DDT_PHYS_VARIANT(ddt, p) \ 190 (ASSERT((p) < DDT_PHYS_NONE), \ 191 ((ddt)->ddt_flags & DDT_FLAG_FLAT ? DDT_PHYS_FLAT : (p))) 192 193 #define DDT_TRAD_PHYS_SIZE sizeof (((ddt_univ_phys_t *)0)->ddp_trad) 194 #define DDT_FLAT_PHYS_SIZE sizeof (((ddt_univ_phys_t *)0)->ddp_flat) 195 196 #define _DDT_PHYS_SWITCH(ddt, flat, trad) \ 197 (((ddt)->ddt_flags & DDT_FLAG_FLAT) ? (flat) : (trad)) 198 199 #define DDT_PHYS_SIZE(ddt) _DDT_PHYS_SWITCH(ddt, \ 200 DDT_FLAT_PHYS_SIZE, DDT_TRAD_PHYS_SIZE) 201 202 #define DDT_NPHYS(ddt) _DDT_PHYS_SWITCH(ddt, 1, DDT_PHYS_MAX) 203 #define DDT_PHYS_FOR_COPIES(ddt, p) _DDT_PHYS_SWITCH(ddt, 0, p) 204 #define DDT_PHYS_IS_DITTO(ddt, p) _DDT_PHYS_SWITCH(ddt, 0, (p == 0)) 205 206 /* 207 * A "live" entry, holding changes to an entry made this txg, and other data to 208 * support loading, updating and repairing the entry. 209 */ 210 211 /* State flags for dde_flags */ 212 #define DDE_FLAG_LOADED (1 << 0) /* entry ready for use */ 213 #define DDE_FLAG_OVERQUOTA (1 << 1) /* entry unusable, no space */ 214 #define DDE_FLAG_LOGGED (1 << 2) /* loaded from log */ 215 216 /* 217 * Additional data to support entry update or repair. This is fixed size 218 * because its relatively rarely used. 219 */ 220 typedef struct { 221 /* copy of data after a repair read, to be rewritten */ 222 abd_t *dde_repair_abd; 223 224 /* original phys contents before update, for error handling */ 225 ddt_univ_phys_t dde_orig_phys; 226 227 /* in-flight update IOs */ 228 zio_t *dde_lead_zio[DDT_PHYS_MAX]; 229 } ddt_entry_io_t; 230 231 typedef struct { 232 /* key must be first for ddt_key_compare */ 233 ddt_key_t dde_key; /* ddt_tree key */ 234 avl_node_t dde_node; /* ddt_tree_node */ 235 236 /* storage type and class the entry was loaded from */ 237 ddt_type_t dde_type; 238 ddt_class_t dde_class; 239 240 uint8_t dde_flags; /* load state flags */ 241 kcondvar_t dde_cv; /* signaled when load completes */ 242 uint64_t dde_waiters; /* count of waiters on dde_cv */ 243 244 ddt_entry_io_t *dde_io; /* IO support, when required */ 245 246 ddt_univ_phys_t dde_phys[]; /* flexible -- allocated size varies */ 247 } ddt_entry_t; 248 249 /* 250 * A lightweight entry is for short-lived or transient uses, like iterating or 251 * inspecting, when you don't care where it came from. 252 */ 253 typedef struct { 254 ddt_key_t ddlwe_key; 255 ddt_type_t ddlwe_type; 256 ddt_class_t ddlwe_class; 257 ddt_univ_phys_t ddlwe_phys; 258 } ddt_lightweight_entry_t; 259 260 /* 261 * In-core DDT log. A separate struct to make it easier to switch between the 262 * appending and flushing logs. 263 */ 264 typedef struct { 265 avl_tree_t ddl_tree; /* logged entries */ 266 uint32_t ddl_flags; /* flags for this log */ 267 uint64_t ddl_object; /* log object id */ 268 uint64_t ddl_length; /* on-disk log size */ 269 uint64_t ddl_first_txg; /* txg log became active */ 270 ddt_key_t ddl_checkpoint; /* last checkpoint */ 271 } ddt_log_t; 272 273 /* 274 * In-core DDT object. This covers all entries and stats for a the whole pool 275 * for a given checksum type. 276 */ 277 typedef struct { 278 kmutex_t ddt_lock; /* protects changes to all fields */ 279 280 avl_tree_t ddt_tree; /* "live" (changed) entries this txg */ 281 avl_tree_t ddt_log_tree; /* logged entries */ 282 283 avl_tree_t ddt_repair_tree; /* entries being repaired */ 284 285 ddt_log_t ddt_log[2]; /* active/flushing logs */ 286 ddt_log_t *ddt_log_active; /* pointers into ddt_log */ 287 ddt_log_t *ddt_log_flushing; /* swapped when flush starts */ 288 289 int32_t ddt_log_ingest_rate; /* rolling log ingest rate */ 290 int32_t ddt_log_flush_rate; /* rolling log flush rate */ 291 int32_t ddt_log_flush_time_rate; /* avg time spent flushing */ 292 uint32_t ddt_log_flush_pressure; /* pressure to apply for cap */ 293 uint32_t ddt_log_flush_prev_backlog; /* prev backlog size */ 294 295 uint64_t ddt_flush_force_txg; /* flush hard before this txg */ 296 297 kstat_t *ddt_ksp; /* kstats context */ 298 299 enum zio_checksum ddt_checksum; /* checksum algorithm in use */ 300 spa_t *ddt_spa; /* pool this ddt is on */ 301 objset_t *ddt_os; /* ddt objset (always MOS) */ 302 303 uint64_t ddt_dir_object; /* MOS dir holding ddt objects */ 304 uint64_t ddt_version; /* DDT version */ 305 uint64_t ddt_flags; /* FDT option flags */ 306 307 /* per-type/per-class entry store objects */ 308 uint64_t ddt_object[DDT_TYPES][DDT_CLASSES]; 309 310 /* object ids for stored, logged and per-type/per-class stats */ 311 uint64_t ddt_stat_object; 312 ddt_object_t ddt_log_stats; 313 ddt_object_t ddt_object_stats[DDT_TYPES][DDT_CLASSES]; 314 315 /* type/class stats by power-2-sized referenced blocks */ 316 ddt_histogram_t ddt_histogram[DDT_TYPES][DDT_CLASSES]; 317 ddt_histogram_t ddt_histogram_cache[DDT_TYPES][DDT_CLASSES]; 318 319 /* log stats power-2-sized referenced blocks */ 320 ddt_histogram_t ddt_log_histogram; 321 } ddt_t; 322 323 /* 324 * In-core and on-disk bookmark for DDT walks. This is a cursor for ddt_walk(), 325 * and is stable across calls, even if the DDT is updated, the pool is 326 * restarted or loaded on another system, or OpenZFS is upgraded. 327 */ 328 typedef struct { 329 uint64_t ddb_class; 330 uint64_t ddb_type; 331 uint64_t ddb_checksum; 332 uint64_t ddb_cursor; 333 } ddt_bookmark_t; 334 335 extern void ddt_bp_fill(const ddt_univ_phys_t *ddp, ddt_phys_variant_t v, 336 blkptr_t *bp, uint64_t txg); 337 extern void ddt_bp_create(enum zio_checksum checksum, const ddt_key_t *ddk, 338 const ddt_univ_phys_t *ddp, ddt_phys_variant_t v, blkptr_t *bp); 339 340 extern void ddt_phys_extend(ddt_univ_phys_t *ddp, ddt_phys_variant_t v, 341 const blkptr_t *bp); 342 extern void ddt_phys_copy(ddt_univ_phys_t *dst, const ddt_univ_phys_t *src, 343 ddt_phys_variant_t v); 344 extern void ddt_phys_clear(ddt_univ_phys_t *ddp, ddt_phys_variant_t v); 345 extern void ddt_phys_addref(ddt_univ_phys_t *ddp, ddt_phys_variant_t v); 346 extern uint64_t ddt_phys_decref(ddt_univ_phys_t *ddp, ddt_phys_variant_t v); 347 extern uint64_t ddt_phys_refcnt(const ddt_univ_phys_t *ddp, 348 ddt_phys_variant_t v); 349 extern ddt_phys_variant_t ddt_phys_select(const ddt_t *ddt, 350 const ddt_entry_t *dde, const blkptr_t *bp); 351 extern uint64_t ddt_phys_birth(const ddt_univ_phys_t *ddp, 352 ddt_phys_variant_t v); 353 extern int ddt_phys_dva_count(const ddt_univ_phys_t *ddp, ddt_phys_variant_t v, 354 boolean_t encrypted); 355 356 extern void ddt_histogram_add_entry(ddt_t *ddt, ddt_histogram_t *ddh, 357 const ddt_lightweight_entry_t *ddlwe); 358 extern void ddt_histogram_sub_entry(ddt_t *ddt, ddt_histogram_t *ddh, 359 const ddt_lightweight_entry_t *ddlwe); 360 361 extern void ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src); 362 extern void ddt_histogram_total(ddt_stat_t *dds, const ddt_histogram_t *ddh); 363 extern boolean_t ddt_histogram_empty(const ddt_histogram_t *ddh); 364 365 extern void ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo); 366 extern uint64_t ddt_get_ddt_dsize(spa_t *spa); 367 extern void ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh); 368 extern void ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total); 369 370 extern uint64_t ddt_get_dedup_dspace(spa_t *spa); 371 extern uint64_t ddt_get_pool_dedup_ratio(spa_t *spa); 372 extern int ddt_get_pool_dedup_cached(spa_t *spa, uint64_t *psize); 373 374 extern ddt_t *ddt_select(spa_t *spa, const blkptr_t *bp); 375 extern void ddt_enter(ddt_t *ddt); 376 extern void ddt_exit(ddt_t *ddt); 377 extern void ddt_init(void); 378 extern void ddt_fini(void); 379 extern ddt_entry_t *ddt_lookup(ddt_t *ddt, const blkptr_t *bp, 380 boolean_t verify); 381 extern void ddt_remove(ddt_t *ddt, ddt_entry_t *dde); 382 extern void ddt_prefetch(spa_t *spa, const blkptr_t *bp); 383 extern void ddt_prefetch_all(spa_t *spa); 384 385 extern boolean_t ddt_class_contains(spa_t *spa, ddt_class_t max_class, 386 const blkptr_t *bp); 387 388 extern void ddt_alloc_entry_io(ddt_entry_t *dde); 389 390 extern ddt_entry_t *ddt_repair_start(ddt_t *ddt, const blkptr_t *bp); 391 extern void ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde); 392 393 extern int ddt_key_compare(const void *x1, const void *x2); 394 395 extern void ddt_create(spa_t *spa); 396 extern int ddt_load(spa_t *spa); 397 extern void ddt_unload(spa_t *spa); 398 extern void ddt_sync(spa_t *spa, uint64_t txg); 399 400 extern void ddt_walk_init(spa_t *spa, uint64_t txg); 401 extern boolean_t ddt_walk_ready(spa_t *spa); 402 extern int ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, 403 ddt_lightweight_entry_t *ddlwe); 404 405 extern boolean_t ddt_addref(spa_t *spa, const blkptr_t *bp); 406 407 extern int ddt_prune_unique_entries(spa_t *spa, zpool_ddt_prune_unit_t unit, 408 uint64_t amount); 409 410 #ifdef __cplusplus 411 } 412 #endif 413 414 #endif /* _SYS_DDT_H */ 415