1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5ea8dc4b6Seschrock * Common Development and Distribution License (the "License"). 6ea8dc4b6Seschrock * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 21fa9e4066Sahrens /* 22c1449561SEric Taylor * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 23fa9e4066Sahrens * Use is subject to license terms. 24fa9e4066Sahrens */ 25fa9e4066Sahrens 26fa9e4066Sahrens #ifndef _SYS_DNODE_H 27fa9e4066Sahrens #define _SYS_DNODE_H 28fa9e4066Sahrens 29fa9e4066Sahrens #include <sys/zfs_context.h> 30fa9e4066Sahrens #include <sys/avl.h> 31fa9e4066Sahrens #include <sys/spa.h> 32fa9e4066Sahrens #include <sys/txg.h> 33c717a561Smaybee #include <sys/zio.h> 34fa9e4066Sahrens #include <sys/refcount.h> 35fa9e4066Sahrens #include <sys/dmu_zfetch.h> 36fa9e4066Sahrens 37fa9e4066Sahrens #ifdef __cplusplus 38fa9e4066Sahrens extern "C" { 39fa9e4066Sahrens #endif 40fa9e4066Sahrens 41fa9e4066Sahrens /* 42cdb0ab79Smaybee * dnode_hold() flags. 43fa9e4066Sahrens */ 44fa9e4066Sahrens #define DNODE_MUST_BE_ALLOCATED 1 45fa9e4066Sahrens #define DNODE_MUST_BE_FREE 2 46fa9e4066Sahrens 47fa9e4066Sahrens /* 48cdb0ab79Smaybee * dnode_next_offset() flags. 49cdb0ab79Smaybee */ 50cdb0ab79Smaybee #define DNODE_FIND_HOLE 1 51cdb0ab79Smaybee #define DNODE_FIND_BACKWARDS 2 52cdb0ab79Smaybee #define DNODE_FIND_HAVELOCK 4 53cdb0ab79Smaybee 54cdb0ab79Smaybee /* 55fa9e4066Sahrens * Fixed constants. 56fa9e4066Sahrens */ 57fa9e4066Sahrens #define DNODE_SHIFT 9 /* 512 bytes */ 58fa9e4066Sahrens #define DN_MIN_INDBLKSHIFT 10 /* 1k */ 59fa9e4066Sahrens #define DN_MAX_INDBLKSHIFT 14 /* 16k */ 60fa9e4066Sahrens #define DNODE_BLOCK_SHIFT 14 /* 16k */ 61fa9e4066Sahrens #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ 62fa9e4066Sahrens #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ 63fa9e4066Sahrens #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ 64fa9e4066Sahrens 65fa9e4066Sahrens /* 66*0a586ceaSMark Shellenbaum * dnode id flags 67*0a586ceaSMark Shellenbaum * 68*0a586ceaSMark Shellenbaum * Note: a file will never ever have its 69*0a586ceaSMark Shellenbaum * ids moved from bonus->spill 70*0a586ceaSMark Shellenbaum * and only in a crypto environment would it be on spill 71*0a586ceaSMark Shellenbaum */ 72*0a586ceaSMark Shellenbaum #define DN_ID_CHKED_BONUS 0x1 73*0a586ceaSMark Shellenbaum #define DN_ID_CHKED_SPILL 0x2 74*0a586ceaSMark Shellenbaum #define DN_ID_OLD_EXIST 0x4 75*0a586ceaSMark Shellenbaum #define DN_ID_NEW_EXIST 0x8 76*0a586ceaSMark Shellenbaum #define DN_ID_SYNC 0x10 77*0a586ceaSMark Shellenbaum 78*0a586ceaSMark Shellenbaum /* 79fa9e4066Sahrens * Derived constants. 80fa9e4066Sahrens */ 81fa9e4066Sahrens #define DNODE_SIZE (1 << DNODE_SHIFT) 82fa9e4066Sahrens #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT) 83fa9e4066Sahrens #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT)) 84ea8dc4b6Seschrock #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT) 851934e92fSmaybee #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1) 86*0a586ceaSMark Shellenbaum #define DN_KILL_SPILLBLK (1) 87fa9e4066Sahrens 88fa9e4066Sahrens #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) 89fa9e4066Sahrens #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) 90fa9e4066Sahrens #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) 91c1449561SEric Taylor #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT) 92fa9e4066Sahrens 93fa9e4066Sahrens /* The +2 here is a cheesy way to round up */ 94fa9e4066Sahrens #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ 95fa9e4066Sahrens (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) 96fa9e4066Sahrens 97fa9e4066Sahrens #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ 98fa9e4066Sahrens (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) 99fa9e4066Sahrens 10099653d4eSeschrock #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ 10199653d4eSeschrock (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) 10299653d4eSeschrock 103fa9e4066Sahrens #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) 104fa9e4066Sahrens 105fa9e4066Sahrens struct dmu_buf_impl; 106503ad85cSMatthew Ahrens struct objset; 107fa9e4066Sahrens struct zio; 108fa9e4066Sahrens 109fa9e4066Sahrens enum dnode_dirtycontext { 110fa9e4066Sahrens DN_UNDIRTIED, 111fa9e4066Sahrens DN_DIRTY_OPEN, 112fa9e4066Sahrens DN_DIRTY_SYNC 113fa9e4066Sahrens }; 114fa9e4066Sahrens 11599653d4eSeschrock /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ 11699653d4eSeschrock #define DNODE_FLAG_USED_BYTES (1<<0) 11714843421SMatthew Ahrens #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) 11899653d4eSeschrock 119*0a586ceaSMark Shellenbaum /* Does dnode have a SA spill blkptr in bonus? */ 120*0a586ceaSMark Shellenbaum #define DNODE_FLAG_SPILL_BLKPTR (1<<2) 121*0a586ceaSMark Shellenbaum 122fa9e4066Sahrens typedef struct dnode_phys { 123fa9e4066Sahrens uint8_t dn_type; /* dmu_object_type_t */ 124fa9e4066Sahrens uint8_t dn_indblkshift; /* ln2(indirect block size) */ 125fa9e4066Sahrens uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ 126fa9e4066Sahrens uint8_t dn_nblkptr; /* length of dn_blkptr */ 127fa9e4066Sahrens uint8_t dn_bonustype; /* type of data in bonus buffer */ 128fa9e4066Sahrens uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 129fa9e4066Sahrens uint8_t dn_compress; /* ZIO_COMPRESS type */ 13099653d4eSeschrock uint8_t dn_flags; /* DNODE_FLAG_* */ 131fa9e4066Sahrens uint16_t dn_datablkszsec; /* data block size in 512b sectors */ 132fa9e4066Sahrens uint16_t dn_bonuslen; /* length of dn_bonus */ 133fa9e4066Sahrens uint8_t dn_pad2[4]; 134fa9e4066Sahrens 135fa9e4066Sahrens /* accounting is protected by dn_dirty_mtx */ 136fa9e4066Sahrens uint64_t dn_maxblkid; /* largest allocated block ID */ 13799653d4eSeschrock uint64_t dn_used; /* bytes (or sectors) of disk space */ 138fa9e4066Sahrens 139fa9e4066Sahrens uint64_t dn_pad3[4]; 140fa9e4066Sahrens 141fa9e4066Sahrens blkptr_t dn_blkptr[1]; 142*0a586ceaSMark Shellenbaum uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)]; 143*0a586ceaSMark Shellenbaum blkptr_t dn_spill; 144fa9e4066Sahrens } dnode_phys_t; 145fa9e4066Sahrens 146fa9e4066Sahrens typedef struct dnode { 147fa9e4066Sahrens /* 148c543ec06Sahrens * dn_struct_rwlock protects the structure of the dnode, 149c543ec06Sahrens * including the number of levels of indirection (dn_nlevels), 150c543ec06Sahrens * dn_maxblkid, and dn_next_* 151fa9e4066Sahrens */ 152fa9e4066Sahrens krwlock_t dn_struct_rwlock; 153fa9e4066Sahrens 15414843421SMatthew Ahrens /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 155fa9e4066Sahrens list_node_t dn_link; 156fa9e4066Sahrens 157fa9e4066Sahrens /* immutable: */ 158503ad85cSMatthew Ahrens struct objset *dn_objset; 159fa9e4066Sahrens uint64_t dn_object; 160fa9e4066Sahrens struct dmu_buf_impl *dn_dbuf; 161fa9e4066Sahrens dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 162fa9e4066Sahrens 163fa9e4066Sahrens /* 164c543ec06Sahrens * Copies of stuff in dn_phys. They're valid in the open 165c543ec06Sahrens * context (eg. even before the dnode is first synced). 166c543ec06Sahrens * Where necessary, these are protected by dn_struct_rwlock. 167fa9e4066Sahrens */ 168c543ec06Sahrens dmu_object_type_t dn_type; /* object type */ 169c543ec06Sahrens uint16_t dn_bonuslen; /* bonus length */ 170c543ec06Sahrens uint8_t dn_bonustype; /* bonus type */ 171fa9e4066Sahrens uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 172fa9e4066Sahrens uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 173fa9e4066Sahrens uint8_t dn_compress; /* ZIO_COMPRESS type */ 174fa9e4066Sahrens uint8_t dn_nlevels; 175fa9e4066Sahrens uint8_t dn_indblkshift; 176c543ec06Sahrens uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 177c543ec06Sahrens uint16_t dn_datablkszsec; /* in 512b sectors */ 178c543ec06Sahrens uint32_t dn_datablksz; /* in bytes */ 179c543ec06Sahrens uint64_t dn_maxblkid; 180da03de99SMark Maybee uint8_t dn_next_nblkptr[TXG_SIZE]; 181fa9e4066Sahrens uint8_t dn_next_nlevels[TXG_SIZE]; 182fa9e4066Sahrens uint8_t dn_next_indblkshift[TXG_SIZE]; 183*0a586ceaSMark Shellenbaum uint8_t dn_next_bonustype[TXG_SIZE]; 184*0a586ceaSMark Shellenbaum uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ 1851934e92fSmaybee uint16_t dn_next_bonuslen[TXG_SIZE]; 186c543ec06Sahrens uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 187fa9e4066Sahrens 188fa9e4066Sahrens /* protected by os_lock: */ 189fa9e4066Sahrens list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 190fa9e4066Sahrens 191fa9e4066Sahrens /* protected by dn_mtx: */ 192fa9e4066Sahrens kmutex_t dn_mtx; 193c717a561Smaybee list_t dn_dirty_records[TXG_SIZE]; 194fa9e4066Sahrens avl_tree_t dn_ranges[TXG_SIZE]; 195fa9e4066Sahrens uint64_t dn_allocated_txg; 196fa9e4066Sahrens uint64_t dn_free_txg; 197fa9e4066Sahrens uint64_t dn_assigned_txg; 198fa9e4066Sahrens kcondvar_t dn_notxholds; 199fa9e4066Sahrens enum dnode_dirtycontext dn_dirtyctx; 200fa9e4066Sahrens uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ 201fa9e4066Sahrens 202fa9e4066Sahrens /* protected by own devices */ 203fa9e4066Sahrens refcount_t dn_tx_holds; 204fa9e4066Sahrens refcount_t dn_holds; 205fa9e4066Sahrens 206fa9e4066Sahrens kmutex_t dn_dbufs_mtx; 207fa9e4066Sahrens list_t dn_dbufs; /* linked list of descendent dbuf_t's */ 208ea8dc4b6Seschrock struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 209*0a586ceaSMark Shellenbaum boolean_t dn_have_spill; /* have spill or are spilling */ 210fa9e4066Sahrens 211c717a561Smaybee /* parent IO for current sync write */ 212c717a561Smaybee zio_t *dn_zio; 213c717a561Smaybee 21414843421SMatthew Ahrens /* used in syncing context */ 215*0a586ceaSMark Shellenbaum uint64_t dn_oldused; /* old phys used bytes */ 216*0a586ceaSMark Shellenbaum uint64_t dn_oldflags; /* old phys dn_flags */ 217*0a586ceaSMark Shellenbaum uint64_t dn_olduid, dn_oldgid; 218*0a586ceaSMark Shellenbaum uint64_t dn_newuid, dn_newgid; 219*0a586ceaSMark Shellenbaum int dn_id_flags; 22014843421SMatthew Ahrens 221fa9e4066Sahrens /* holds prefetch structure */ 222fa9e4066Sahrens struct zfetch dn_zfetch; 223fa9e4066Sahrens } dnode_t; 224fa9e4066Sahrens 225fa9e4066Sahrens typedef struct free_range { 226fa9e4066Sahrens avl_node_t fr_node; 227fa9e4066Sahrens uint64_t fr_blkid; 228fa9e4066Sahrens uint64_t fr_nblks; 229fa9e4066Sahrens } free_range_t; 230fa9e4066Sahrens 231503ad85cSMatthew Ahrens dnode_t *dnode_special_open(struct objset *dd, dnode_phys_t *dnp, 232fa9e4066Sahrens uint64_t object); 233fa9e4066Sahrens void dnode_special_close(dnode_t *dn); 234fa9e4066Sahrens 2351934e92fSmaybee void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 236*0a586ceaSMark Shellenbaum void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); 237*0a586ceaSMark Shellenbaum void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); 238*0a586ceaSMark Shellenbaum 239503ad85cSMatthew Ahrens int dnode_hold(struct objset *dd, uint64_t object, 240ea8dc4b6Seschrock void *ref, dnode_t **dnp); 241503ad85cSMatthew Ahrens int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, 242ea8dc4b6Seschrock void *ref, dnode_t **dnp); 2431934e92fSmaybee boolean_t dnode_add_ref(dnode_t *dn, void *ref); 244fa9e4066Sahrens void dnode_rele(dnode_t *dn, void *ref); 245fa9e4066Sahrens void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 246c717a561Smaybee void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 247fa9e4066Sahrens void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 248fa9e4066Sahrens dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 249fa9e4066Sahrens void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 250fa9e4066Sahrens dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 251fa9e4066Sahrens void dnode_free(dnode_t *dn, dmu_tx_t *tx); 252fa9e4066Sahrens void dnode_byteswap(dnode_phys_t *dnp); 253fa9e4066Sahrens void dnode_buf_byteswap(void *buf, size_t size); 254fa9e4066Sahrens void dnode_verify(dnode_t *dn); 255fa9e4066Sahrens int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 256fa9e4066Sahrens uint64_t dnode_current_max_length(dnode_t *dn); 257fa9e4066Sahrens void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 258fa9e4066Sahrens void dnode_clear_range(dnode_t *dn, uint64_t blkid, 259fa9e4066Sahrens uint64_t nblks, dmu_tx_t *tx); 260fa9e4066Sahrens void dnode_diduse_space(dnode_t *dn, int64_t space); 261fa9e4066Sahrens void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx); 2628346f03fSJonathan W Adams void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); 263fa9e4066Sahrens uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 264fa9e4066Sahrens void dnode_init(void); 265fa9e4066Sahrens void dnode_fini(void); 266cdb0ab79Smaybee int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 267cdb0ab79Smaybee int minlvl, uint64_t blkfill, uint64_t txg); 2681934e92fSmaybee void dnode_evict_dbufs(dnode_t *dn); 269fa9e4066Sahrens 270fa9e4066Sahrens #ifdef ZFS_DEBUG 271fa9e4066Sahrens 272fa9e4066Sahrens /* 273fa9e4066Sahrens * There should be a ## between the string literal and fmt, to make it 274fa9e4066Sahrens * clear that we're joining two strings together, but that piece of shit 275fa9e4066Sahrens * gcc doesn't support that preprocessor token. 276fa9e4066Sahrens */ 277fa9e4066Sahrens #define dprintf_dnode(dn, fmt, ...) do { \ 278fa9e4066Sahrens if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 279fa9e4066Sahrens char __db_buf[32]; \ 280fa9e4066Sahrens uint64_t __db_obj = (dn)->dn_object; \ 281fa9e4066Sahrens if (__db_obj == DMU_META_DNODE_OBJECT) \ 282fa9e4066Sahrens (void) strcpy(__db_buf, "mdn"); \ 283fa9e4066Sahrens else \ 284fa9e4066Sahrens (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 285fa9e4066Sahrens (u_longlong_t)__db_obj);\ 286fa9e4066Sahrens dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 287fa9e4066Sahrens __db_buf, __VA_ARGS__); \ 288fa9e4066Sahrens } \ 289fa9e4066Sahrens _NOTE(CONSTCOND) } while (0) 290fa9e4066Sahrens 2919c9dc39aSek110237 #define DNODE_VERIFY(dn) dnode_verify(dn) 2929c9dc39aSek110237 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 2939c9dc39aSek110237 294fa9e4066Sahrens #else 295fa9e4066Sahrens 296fa9e4066Sahrens #define dprintf_dnode(db, fmt, ...) 2979c9dc39aSek110237 #define DNODE_VERIFY(dn) 2989c9dc39aSek110237 #define FREE_VERIFY(db, start, end, tx) 299fa9e4066Sahrens 300fa9e4066Sahrens #endif 301fa9e4066Sahrens 302fa9e4066Sahrens #ifdef __cplusplus 303fa9e4066Sahrens } 304fa9e4066Sahrens #endif 305fa9e4066Sahrens 306fa9e4066Sahrens #endif /* _SYS_DNODE_H */ 307