1219089Spjd/* 2219089Spjd * CDDL HEADER START 3219089Spjd * 4219089Spjd * The contents of this file are subject to the terms of the 5219089Spjd * Common Development and Distribution License (the "License"). 6219089Spjd * You may not use this file except in compliance with the License. 7219089Spjd * 8219089Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9219089Spjd * or http://www.opensolaris.org/os/licensing. 10219089Spjd * See the License for the specific language governing permissions 11219089Spjd * and limitations under the License. 12219089Spjd * 13219089Spjd * When distributing Covered Code, include this CDDL HEADER in each 14219089Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15219089Spjd * If applicable, add the following below this CDDL HEADER, with the 16219089Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17219089Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18219089Spjd * 19219089Spjd * CDDL HEADER END 20219089Spjd */ 21219089Spjd/* 22219089Spjd * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved. 23219089Spjd */ 24219089Spjd 25219089Spjd#ifndef _SYS_DDT_H 26219089Spjd#define _SYS_DDT_H 27219089Spjd 28219089Spjd#include <sys/sysmacros.h> 29219089Spjd#include <sys/types.h> 30219089Spjd#include <sys/fs/zfs.h> 31219089Spjd#include <sys/zio.h> 32219089Spjd#include <sys/dmu.h> 33219089Spjd 34219089Spjd#ifdef __cplusplus 35219089Spjdextern "C" { 36219089Spjd#endif 37219089Spjd 38219089Spjd/* 39219089Spjd * On-disk DDT formats, in the desired search order (newest version first). 40219089Spjd */ 41219089Spjdenum ddt_type { 42219089Spjd DDT_TYPE_ZAP = 0, 43219089Spjd DDT_TYPES 44219089Spjd}; 45219089Spjd 46219089Spjd/* 47219089Spjd * DDT classes, in the desired search order (highest replication level first). 48219089Spjd */ 49219089Spjdenum ddt_class { 50219089Spjd DDT_CLASS_DITTO = 0, 51219089Spjd DDT_CLASS_DUPLICATE, 52219089Spjd DDT_CLASS_UNIQUE, 53219089Spjd DDT_CLASSES 54219089Spjd}; 55219089Spjd 56219089Spjd#define DDT_TYPE_CURRENT 0 57219089Spjd 58219089Spjd#define DDT_COMPRESS_BYTEORDER_MASK 0x80 59219089Spjd#define DDT_COMPRESS_FUNCTION_MASK 0x7f 60219089Spjd 61219089Spjd/* 62219089Spjd * On-disk ddt entry: key (name) and physical storage (value). 63219089Spjd */ 64219089Spjdtypedef struct ddt_key { 65219089Spjd zio_cksum_t ddk_cksum; /* 256-bit block checksum */ 66251631Sdelphij /* 67251631Sdelphij * Encoded with logical & physical size, and compression, as follows: 68251631Sdelphij * +-------+-------+-------+-------+-------+-------+-------+-------+ 69251631Sdelphij * | 0 | 0 | 0 | comp | PSIZE | LSIZE | 70251631Sdelphij * +-------+-------+-------+-------+-------+-------+-------+-------+ 71251631Sdelphij */ 72251631Sdelphij uint64_t ddk_prop; 73219089Spjd} ddt_key_t; 74219089Spjd 75219089Spjd#define DDK_GET_LSIZE(ddk) \ 76219089Spjd BF64_GET_SB((ddk)->ddk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1) 77219089Spjd#define DDK_SET_LSIZE(ddk, x) \ 78219089Spjd BF64_SET_SB((ddk)->ddk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1, x) 79219089Spjd 80219089Spjd#define DDK_GET_PSIZE(ddk) \ 81219089Spjd BF64_GET_SB((ddk)->ddk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1) 82219089Spjd#define DDK_SET_PSIZE(ddk, x) \ 83219089Spjd BF64_SET_SB((ddk)->ddk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1, x) 84219089Spjd 85219089Spjd#define DDK_GET_COMPRESS(ddk) BF64_GET((ddk)->ddk_prop, 32, 8) 86219089Spjd#define DDK_SET_COMPRESS(ddk, x) BF64_SET((ddk)->ddk_prop, 32, 8, x) 87219089Spjd 88219089Spjd#define DDT_KEY_WORDS (sizeof (ddt_key_t) / sizeof (uint64_t)) 89219089Spjd 90219089Spjdtypedef struct ddt_phys { 91219089Spjd dva_t ddp_dva[SPA_DVAS_PER_BP]; 92219089Spjd uint64_t ddp_refcnt; 93219089Spjd uint64_t ddp_phys_birth; 94219089Spjd} ddt_phys_t; 95219089Spjd 96219089Spjdenum ddt_phys_type { 97219089Spjd DDT_PHYS_DITTO = 0, 98219089Spjd DDT_PHYS_SINGLE = 1, 99219089Spjd DDT_PHYS_DOUBLE = 2, 100219089Spjd DDT_PHYS_TRIPLE = 3, 101219089Spjd DDT_PHYS_TYPES 102219089Spjd}; 103219089Spjd 104219089Spjd/* 105219089Spjd * In-core ddt entry 106219089Spjd */ 107219089Spjdstruct ddt_entry { 108219089Spjd ddt_key_t dde_key; 109219089Spjd ddt_phys_t dde_phys[DDT_PHYS_TYPES]; 110219089Spjd zio_t *dde_lead_zio[DDT_PHYS_TYPES]; 111219089Spjd void *dde_repair_data; 112219089Spjd enum ddt_type dde_type; 113219089Spjd enum ddt_class dde_class; 114219089Spjd uint8_t dde_loading; 115219089Spjd uint8_t dde_loaded; 116219089Spjd kcondvar_t dde_cv; 117219089Spjd avl_node_t dde_node; 118219089Spjd}; 119219089Spjd 120219089Spjd/* 121219089Spjd * In-core ddt 122219089Spjd */ 123219089Spjdstruct ddt { 124219089Spjd kmutex_t ddt_lock; 125219089Spjd avl_tree_t ddt_tree; 126219089Spjd avl_tree_t ddt_repair_tree; 127219089Spjd enum zio_checksum ddt_checksum; 128219089Spjd spa_t *ddt_spa; 129219089Spjd objset_t *ddt_os; 130219089Spjd uint64_t ddt_stat_object; 131219089Spjd uint64_t ddt_object[DDT_TYPES][DDT_CLASSES]; 132219089Spjd ddt_histogram_t ddt_histogram[DDT_TYPES][DDT_CLASSES]; 133219089Spjd ddt_histogram_t ddt_histogram_cache[DDT_TYPES][DDT_CLASSES]; 134219089Spjd ddt_object_t ddt_object_stats[DDT_TYPES][DDT_CLASSES]; 135219089Spjd avl_node_t ddt_node; 136219089Spjd}; 137219089Spjd 138219089Spjd/* 139219089Spjd * In-core and on-disk bookmark for DDT walks 140219089Spjd */ 141219089Spjdtypedef struct ddt_bookmark { 142219089Spjd uint64_t ddb_class; 143219089Spjd uint64_t ddb_type; 144219089Spjd uint64_t ddb_checksum; 145219089Spjd uint64_t ddb_cursor; 146219089Spjd} ddt_bookmark_t; 147219089Spjd 148219089Spjd/* 149219089Spjd * Ops vector to access a specific DDT object type. 150219089Spjd */ 151219089Spjdtypedef struct ddt_ops { 152219089Spjd char ddt_op_name[32]; 153219089Spjd int (*ddt_op_create)(objset_t *os, uint64_t *object, dmu_tx_t *tx, 154219089Spjd boolean_t prehash); 155219089Spjd int (*ddt_op_destroy)(objset_t *os, uint64_t object, dmu_tx_t *tx); 156219089Spjd int (*ddt_op_lookup)(objset_t *os, uint64_t object, ddt_entry_t *dde); 157219089Spjd void (*ddt_op_prefetch)(objset_t *os, uint64_t object, 158219089Spjd ddt_entry_t *dde); 159219089Spjd int (*ddt_op_update)(objset_t *os, uint64_t object, ddt_entry_t *dde, 160219089Spjd dmu_tx_t *tx); 161219089Spjd int (*ddt_op_remove)(objset_t *os, uint64_t object, ddt_entry_t *dde, 162219089Spjd dmu_tx_t *tx); 163219089Spjd int (*ddt_op_walk)(objset_t *os, uint64_t object, ddt_entry_t *dde, 164219089Spjd uint64_t *walk); 165245264Sdelphij int (*ddt_op_count)(objset_t *os, uint64_t object, uint64_t *count); 166219089Spjd} ddt_ops_t; 167219089Spjd 168219089Spjd#define DDT_NAMELEN 80 169219089Spjd 170219089Spjdextern void ddt_object_name(ddt_t *ddt, enum ddt_type type, 171222950Sgibbs enum ddt_class cls, char *name); 172219089Spjdextern int ddt_object_walk(ddt_t *ddt, enum ddt_type type, 173222950Sgibbs enum ddt_class cls, uint64_t *walk, ddt_entry_t *dde); 174245264Sdelphijextern int ddt_object_count(ddt_t *ddt, enum ddt_type type, 175253643Smav enum ddt_class cls, uint64_t *count); 176219089Spjdextern int ddt_object_info(ddt_t *ddt, enum ddt_type type, 177222950Sgibbs enum ddt_class cls, dmu_object_info_t *); 178219089Spjdextern boolean_t ddt_object_exists(ddt_t *ddt, enum ddt_type type, 179222950Sgibbs enum ddt_class cls); 180219089Spjd 181219089Spjdextern void ddt_bp_fill(const ddt_phys_t *ddp, blkptr_t *bp, 182219089Spjd uint64_t txg); 183219089Spjdextern void ddt_bp_create(enum zio_checksum checksum, const ddt_key_t *ddk, 184219089Spjd const ddt_phys_t *ddp, blkptr_t *bp); 185219089Spjd 186219089Spjdextern void ddt_key_fill(ddt_key_t *ddk, const blkptr_t *bp); 187219089Spjd 188219089Spjdextern void ddt_phys_fill(ddt_phys_t *ddp, const blkptr_t *bp); 189219089Spjdextern void ddt_phys_clear(ddt_phys_t *ddp); 190219089Spjdextern void ddt_phys_addref(ddt_phys_t *ddp); 191219089Spjdextern void ddt_phys_decref(ddt_phys_t *ddp); 192219089Spjdextern void ddt_phys_free(ddt_t *ddt, ddt_key_t *ddk, ddt_phys_t *ddp, 193219089Spjd uint64_t txg); 194219089Spjdextern ddt_phys_t *ddt_phys_select(const ddt_entry_t *dde, const blkptr_t *bp); 195219089Spjdextern uint64_t ddt_phys_total_refcnt(const ddt_entry_t *dde); 196219089Spjd 197219089Spjdextern void ddt_stat_add(ddt_stat_t *dst, const ddt_stat_t *src, uint64_t neg); 198219089Spjd 199219089Spjdextern void ddt_histogram_add(ddt_histogram_t *dst, const ddt_histogram_t *src); 200219089Spjdextern void ddt_histogram_stat(ddt_stat_t *dds, const ddt_histogram_t *ddh); 201219089Spjdextern boolean_t ddt_histogram_empty(const ddt_histogram_t *ddh); 202219089Spjdextern void ddt_get_dedup_object_stats(spa_t *spa, ddt_object_t *ddo); 203219089Spjdextern void ddt_get_dedup_histogram(spa_t *spa, ddt_histogram_t *ddh); 204219089Spjdextern void ddt_get_dedup_stats(spa_t *spa, ddt_stat_t *dds_total); 205219089Spjd 206219089Spjdextern uint64_t ddt_get_dedup_dspace(spa_t *spa); 207219089Spjdextern uint64_t ddt_get_pool_dedup_ratio(spa_t *spa); 208219089Spjd 209219089Spjdextern int ddt_ditto_copies_needed(ddt_t *ddt, ddt_entry_t *dde, 210219089Spjd ddt_phys_t *ddp_willref); 211219089Spjdextern int ddt_ditto_copies_present(ddt_entry_t *dde); 212219089Spjd 213219089Spjdextern size_t ddt_compress(void *src, uchar_t *dst, size_t s_len, size_t d_len); 214219089Spjdextern void ddt_decompress(uchar_t *src, void *dst, size_t s_len, size_t d_len); 215219089Spjd 216219089Spjdextern ddt_t *ddt_select(spa_t *spa, const blkptr_t *bp); 217219089Spjdextern void ddt_enter(ddt_t *ddt); 218219089Spjdextern void ddt_exit(ddt_t *ddt); 219219089Spjdextern ddt_entry_t *ddt_lookup(ddt_t *ddt, const blkptr_t *bp, boolean_t add); 220219089Spjdextern void ddt_prefetch(spa_t *spa, const blkptr_t *bp); 221219089Spjdextern void ddt_remove(ddt_t *ddt, ddt_entry_t *dde); 222219089Spjd 223219089Spjdextern boolean_t ddt_class_contains(spa_t *spa, enum ddt_class max_class, 224219089Spjd const blkptr_t *bp); 225219089Spjd 226219089Spjdextern ddt_entry_t *ddt_repair_start(ddt_t *ddt, const blkptr_t *bp); 227219089Spjdextern void ddt_repair_done(ddt_t *ddt, ddt_entry_t *dde); 228219089Spjd 229219089Spjdextern int ddt_entry_compare(const void *x1, const void *x2); 230219089Spjd 231219089Spjdextern void ddt_create(spa_t *spa); 232219089Spjdextern int ddt_load(spa_t *spa); 233219089Spjdextern void ddt_unload(spa_t *spa); 234219089Spjdextern void ddt_sync(spa_t *spa, uint64_t txg); 235219089Spjdextern int ddt_walk(spa_t *spa, ddt_bookmark_t *ddb, ddt_entry_t *dde); 236219089Spjdextern int ddt_object_update(ddt_t *ddt, enum ddt_type type, 237222950Sgibbs enum ddt_class cls, ddt_entry_t *dde, dmu_tx_t *tx); 238219089Spjd 239219089Spjdextern const ddt_ops_t ddt_zap_ops; 240219089Spjd 241219089Spjd#ifdef __cplusplus 242219089Spjd} 243219089Spjd#endif 244219089Spjd 245219089Spjd#endif /* _SYS_DDT_H */ 246