dnode.h revision 288549
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 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 25 */ 26 27#ifndef _SYS_DNODE_H 28#define _SYS_DNODE_H 29 30#include <sys/zfs_context.h> 31#include <sys/avl.h> 32#include <sys/spa.h> 33#include <sys/txg.h> 34#include <sys/zio.h> 35#include <sys/refcount.h> 36#include <sys/dmu_zfetch.h> 37#include <sys/zrlock.h> 38 39#ifdef __cplusplus 40extern "C" { 41#endif 42 43/* 44 * dnode_hold() flags. 45 */ 46#define DNODE_MUST_BE_ALLOCATED 1 47#define DNODE_MUST_BE_FREE 2 48 49/* 50 * dnode_next_offset() flags. 51 */ 52#define DNODE_FIND_HOLE 1 53#define DNODE_FIND_BACKWARDS 2 54#define DNODE_FIND_HAVELOCK 4 55 56/* 57 * Fixed constants. 58 */ 59#define DNODE_SHIFT 9 /* 512 bytes */ 60#define DN_MIN_INDBLKSHIFT 12 /* 4k */ 61#define DN_MAX_INDBLKSHIFT 14 /* 16k */ 62#define DNODE_BLOCK_SHIFT 14 /* 16k */ 63#define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ 64#define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ 65#define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ 66 67/* 68 * dnode id flags 69 * 70 * Note: a file will never ever have its 71 * ids moved from bonus->spill 72 * and only in a crypto environment would it be on spill 73 */ 74#define DN_ID_CHKED_BONUS 0x1 75#define DN_ID_CHKED_SPILL 0x2 76#define DN_ID_OLD_EXIST 0x4 77#define DN_ID_NEW_EXIST 0x8 78 79/* 80 * Derived constants. 81 */ 82#define DNODE_SIZE (1 << DNODE_SHIFT) 83#define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT) 84#define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT)) 85#define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT) 86#define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1) 87#define DN_KILL_SPILLBLK (1) 88 89#define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) 90#define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) 91#define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) 92#define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT) 93 94/* The +2 here is a cheesy way to round up */ 95#define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ 96 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) 97 98#define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ 99 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) 100 101#define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ 102 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) 103 104#define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) 105 106struct dmu_buf_impl; 107struct objset; 108struct zio; 109 110enum dnode_dirtycontext { 111 DN_UNDIRTIED, 112 DN_DIRTY_OPEN, 113 DN_DIRTY_SYNC 114}; 115 116/* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ 117#define DNODE_FLAG_USED_BYTES (1<<0) 118#define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) 119 120/* Does dnode have a SA spill blkptr in bonus? */ 121#define DNODE_FLAG_SPILL_BLKPTR (1<<2) 122 123typedef struct dnode_phys { 124 uint8_t dn_type; /* dmu_object_type_t */ 125 uint8_t dn_indblkshift; /* ln2(indirect block size) */ 126 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ 127 uint8_t dn_nblkptr; /* length of dn_blkptr */ 128 uint8_t dn_bonustype; /* type of data in bonus buffer */ 129 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 130 uint8_t dn_compress; /* ZIO_COMPRESS type */ 131 uint8_t dn_flags; /* DNODE_FLAG_* */ 132 uint16_t dn_datablkszsec; /* data block size in 512b sectors */ 133 uint16_t dn_bonuslen; /* length of dn_bonus */ 134 uint8_t dn_pad2[4]; 135 136 /* accounting is protected by dn_dirty_mtx */ 137 uint64_t dn_maxblkid; /* largest allocated block ID */ 138 uint64_t dn_used; /* bytes (or sectors) of disk space */ 139 140 uint64_t dn_pad3[4]; 141 142 blkptr_t dn_blkptr[1]; 143 uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)]; 144 blkptr_t dn_spill; 145} dnode_phys_t; 146 147typedef struct dnode { 148 /* 149 * Protects the structure of the dnode, including the number of levels 150 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* 151 */ 152 krwlock_t dn_struct_rwlock; 153 154 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 155 list_node_t dn_link; 156 157 /* immutable: */ 158 struct objset *dn_objset; 159 uint64_t dn_object; 160 struct dmu_buf_impl *dn_dbuf; 161 struct dnode_handle *dn_handle; 162 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 163 164 /* 165 * Copies of stuff in dn_phys. They're valid in the open 166 * context (eg. even before the dnode is first synced). 167 * Where necessary, these are protected by dn_struct_rwlock. 168 */ 169 dmu_object_type_t dn_type; /* object type */ 170 uint16_t dn_bonuslen; /* bonus length */ 171 uint8_t dn_bonustype; /* bonus type */ 172 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 173 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 174 uint8_t dn_compress; /* ZIO_COMPRESS type */ 175 uint8_t dn_nlevels; 176 uint8_t dn_indblkshift; 177 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 178 uint8_t dn_moved; /* Has this dnode been moved? */ 179 uint16_t dn_datablkszsec; /* in 512b sectors */ 180 uint32_t dn_datablksz; /* in bytes */ 181 uint64_t dn_maxblkid; 182 uint8_t dn_next_type[TXG_SIZE]; 183 uint8_t dn_next_nblkptr[TXG_SIZE]; 184 uint8_t dn_next_nlevels[TXG_SIZE]; 185 uint8_t dn_next_indblkshift[TXG_SIZE]; 186 uint8_t dn_next_bonustype[TXG_SIZE]; 187 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ 188 uint16_t dn_next_bonuslen[TXG_SIZE]; 189 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 190 191 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ 192 uint32_t dn_dbufs_count; /* count of dn_dbufs */ 193 /* There are no level-0 blocks of this blkid or higher in dn_dbufs */ 194 uint64_t dn_unlisted_l0_blkid; 195 196 /* protected by os_lock: */ 197 list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 198 199 /* protected by dn_mtx: */ 200 kmutex_t dn_mtx; 201 list_t dn_dirty_records[TXG_SIZE]; 202 struct range_tree *dn_free_ranges[TXG_SIZE]; 203 uint64_t dn_allocated_txg; 204 uint64_t dn_free_txg; 205 uint64_t dn_assigned_txg; 206 kcondvar_t dn_notxholds; 207 enum dnode_dirtycontext dn_dirtyctx; 208 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ 209 210 /* protected by own devices */ 211 refcount_t dn_tx_holds; 212 refcount_t dn_holds; 213 214 kmutex_t dn_dbufs_mtx; 215 /* 216 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs 217 * can contain multiple dbufs of the same (level, blkid) when a 218 * dbuf is marked DB_EVICTING without being removed from 219 * dn_dbufs. To maintain the avl invariant that there cannot be 220 * duplicate entries, we order the dbufs by an arbitrary value - 221 * their address in memory. This means that dn_dbufs cannot be used to 222 * directly look up a dbuf. Instead, callers must use avl_walk, have 223 * a reference to the dbuf, or look up a non-existant node with 224 * db_state = DB_SEARCH (see dbuf_free_range for an example). 225 */ 226 avl_tree_t dn_dbufs; 227 228 /* protected by dn_struct_rwlock */ 229 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 230 231 boolean_t dn_have_spill; /* have spill or are spilling */ 232 233 /* parent IO for current sync write */ 234 zio_t *dn_zio; 235 236 /* used in syncing context */ 237 uint64_t dn_oldused; /* old phys used bytes */ 238 uint64_t dn_oldflags; /* old phys dn_flags */ 239 uint64_t dn_olduid, dn_oldgid; 240 uint64_t dn_newuid, dn_newgid; 241 int dn_id_flags; 242 243 /* holds prefetch structure */ 244 struct zfetch dn_zfetch; 245} dnode_t; 246 247/* 248 * Adds a level of indirection between the dbuf and the dnode to avoid 249 * iterating descendent dbufs in dnode_move(). Handles are not allocated 250 * individually, but as an array of child dnodes in dnode_hold_impl(). 251 */ 252typedef struct dnode_handle { 253 /* Protects dnh_dnode from modification by dnode_move(). */ 254 zrlock_t dnh_zrlock; 255 dnode_t *dnh_dnode; 256} dnode_handle_t; 257 258typedef struct dnode_children { 259 dmu_buf_user_t dnc_dbu; /* User evict data */ 260 size_t dnc_count; /* number of children */ 261 dnode_handle_t dnc_children[]; /* sized dynamically */ 262} dnode_children_t; 263 264typedef struct free_range { 265 avl_node_t fr_node; 266 uint64_t fr_blkid; 267 uint64_t fr_nblks; 268} free_range_t; 269 270void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, 271 uint64_t object, dnode_handle_t *dnh); 272void dnode_special_close(dnode_handle_t *dnh); 273 274void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 275void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); 276void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); 277 278int dnode_hold(struct objset *dd, uint64_t object, 279 void *ref, dnode_t **dnp); 280int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, 281 void *ref, dnode_t **dnp); 282boolean_t dnode_add_ref(dnode_t *dn, void *ref); 283void dnode_rele(dnode_t *dn, void *ref); 284void dnode_rele_and_unlock(dnode_t *dn, void *tag); 285void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 286void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 287void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 288 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 289void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 290 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 291void dnode_free(dnode_t *dn, dmu_tx_t *tx); 292void dnode_byteswap(dnode_phys_t *dnp); 293void dnode_buf_byteswap(void *buf, size_t size); 294void dnode_verify(dnode_t *dn); 295int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 296void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 297void dnode_diduse_space(dnode_t *dn, int64_t space); 298void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx); 299void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); 300uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 301void dnode_init(void); 302void dnode_fini(void); 303int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 304 int minlvl, uint64_t blkfill, uint64_t txg); 305void dnode_evict_dbufs(dnode_t *dn); 306void dnode_evict_bonus(dnode_t *dn); 307 308#ifdef ZFS_DEBUG 309 310/* 311 * There should be a ## between the string literal and fmt, to make it 312 * clear that we're joining two strings together, but that piece of shit 313 * gcc doesn't support that preprocessor token. 314 */ 315#define dprintf_dnode(dn, fmt, ...) do { \ 316 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 317 char __db_buf[32]; \ 318 uint64_t __db_obj = (dn)->dn_object; \ 319 if (__db_obj == DMU_META_DNODE_OBJECT) \ 320 (void) strcpy(__db_buf, "mdn"); \ 321 else \ 322 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 323 (u_longlong_t)__db_obj);\ 324 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 325 __db_buf, __VA_ARGS__); \ 326 } \ 327_NOTE(CONSTCOND) } while (0) 328 329#define DNODE_VERIFY(dn) dnode_verify(dn) 330#define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 331 332#else 333 334#define dprintf_dnode(db, fmt, ...) 335#define DNODE_VERIFY(dn) 336#define FREE_VERIFY(db, start, end, tx) 337 338#endif 339 340#ifdef __cplusplus 341} 342#endif 343 344#endif /* _SYS_DNODE_H */ 345