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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _SYS_DNODE_H 27 #define _SYS_DNODE_H 28 29 #include <sys/zfs_context.h> 30 #include <sys/avl.h> 31 #include <sys/spa.h> 32 #include <sys/txg.h> 33 #include <sys/zio.h> 34 #include <sys/refcount.h> 35 #include <sys/dmu_zfetch.h> 36 37 #ifdef __cplusplus 38 extern "C" { 39 #endif 40 41 /* 42 * dnode_hold() flags. 43 */ 44 #define DNODE_MUST_BE_ALLOCATED 1 45 #define DNODE_MUST_BE_FREE 2 46 47 /* 48 * dnode_next_offset() flags. 49 */ 50 #define DNODE_FIND_HOLE 1 51 #define DNODE_FIND_BACKWARDS 2 52 #define DNODE_FIND_HAVELOCK 4 53 54 /* 55 * Fixed constants. 56 */ 57 #define DNODE_SHIFT 9 /* 512 bytes */ 58 #define DN_MIN_INDBLKSHIFT 10 /* 1k */ 59 #define DN_MAX_INDBLKSHIFT 14 /* 16k */ 60 #define DNODE_BLOCK_SHIFT 14 /* 16k */ 61 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ 62 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ 63 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ 64 65 /* 66 * Derived constants. 67 */ 68 #define DNODE_SIZE (1 << DNODE_SHIFT) 69 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT) 70 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT)) 71 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT) 72 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1) 73 74 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) 75 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) 76 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) 77 78 /* The +2 here is a cheesy way to round up */ 79 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ 80 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) 81 82 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ 83 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) 84 85 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ 86 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) 87 88 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) 89 90 struct dmu_buf_impl; 91 struct objset_impl; 92 struct zio; 93 94 enum dnode_dirtycontext { 95 DN_UNDIRTIED, 96 DN_DIRTY_OPEN, 97 DN_DIRTY_SYNC 98 }; 99 100 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ 101 #define DNODE_FLAG_USED_BYTES (1<<0) 102 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) 103 104 typedef struct dnode_phys { 105 uint8_t dn_type; /* dmu_object_type_t */ 106 uint8_t dn_indblkshift; /* ln2(indirect block size) */ 107 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ 108 uint8_t dn_nblkptr; /* length of dn_blkptr */ 109 uint8_t dn_bonustype; /* type of data in bonus buffer */ 110 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 111 uint8_t dn_compress; /* ZIO_COMPRESS type */ 112 uint8_t dn_flags; /* DNODE_FLAG_* */ 113 uint16_t dn_datablkszsec; /* data block size in 512b sectors */ 114 uint16_t dn_bonuslen; /* length of dn_bonus */ 115 uint8_t dn_pad2[4]; 116 117 /* accounting is protected by dn_dirty_mtx */ 118 uint64_t dn_maxblkid; /* largest allocated block ID */ 119 uint64_t dn_used; /* bytes (or sectors) of disk space */ 120 121 uint64_t dn_pad3[4]; 122 123 blkptr_t dn_blkptr[1]; 124 uint8_t dn_bonus[DN_MAX_BONUSLEN]; 125 } dnode_phys_t; 126 127 typedef struct dnode { 128 /* 129 * dn_struct_rwlock protects the structure of the dnode, 130 * including the number of levels of indirection (dn_nlevels), 131 * dn_maxblkid, and dn_next_* 132 */ 133 krwlock_t dn_struct_rwlock; 134 135 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 136 list_node_t dn_link; 137 138 /* immutable: */ 139 struct objset_impl *dn_objset; 140 uint64_t dn_object; 141 struct dmu_buf_impl *dn_dbuf; 142 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 143 144 /* 145 * Copies of stuff in dn_phys. They're valid in the open 146 * context (eg. even before the dnode is first synced). 147 * Where necessary, these are protected by dn_struct_rwlock. 148 */ 149 dmu_object_type_t dn_type; /* object type */ 150 uint16_t dn_bonuslen; /* bonus length */ 151 uint8_t dn_bonustype; /* bonus type */ 152 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 153 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 154 uint8_t dn_compress; /* ZIO_COMPRESS type */ 155 uint8_t dn_nlevels; 156 uint8_t dn_indblkshift; 157 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 158 uint16_t dn_datablkszsec; /* in 512b sectors */ 159 uint32_t dn_datablksz; /* in bytes */ 160 uint64_t dn_maxblkid; 161 uint8_t dn_next_nblkptr[TXG_SIZE]; 162 uint8_t dn_next_nlevels[TXG_SIZE]; 163 uint8_t dn_next_indblkshift[TXG_SIZE]; 164 uint16_t dn_next_bonuslen[TXG_SIZE]; 165 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 166 167 /* protected by os_lock: */ 168 list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 169 170 /* protected by dn_mtx: */ 171 kmutex_t dn_mtx; 172 list_t dn_dirty_records[TXG_SIZE]; 173 avl_tree_t dn_ranges[TXG_SIZE]; 174 uint64_t dn_allocated_txg; 175 uint64_t dn_free_txg; 176 uint64_t dn_assigned_txg; 177 kcondvar_t dn_notxholds; 178 enum dnode_dirtycontext dn_dirtyctx; 179 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ 180 181 /* protected by own devices */ 182 refcount_t dn_tx_holds; 183 refcount_t dn_holds; 184 185 kmutex_t dn_dbufs_mtx; 186 list_t dn_dbufs; /* linked list of descendent dbuf_t's */ 187 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 188 189 /* parent IO for current sync write */ 190 zio_t *dn_zio; 191 192 /* used in syncing context */ 193 dnode_phys_t *dn_oldphys; 194 195 /* holds prefetch structure */ 196 struct zfetch dn_zfetch; 197 } dnode_t; 198 199 typedef struct free_range { 200 avl_node_t fr_node; 201 uint64_t fr_blkid; 202 uint64_t fr_nblks; 203 } free_range_t; 204 205 dnode_t *dnode_special_open(struct objset_impl *dd, dnode_phys_t *dnp, 206 uint64_t object); 207 void dnode_special_close(dnode_t *dn); 208 209 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 210 int dnode_hold(struct objset_impl *dd, uint64_t object, 211 void *ref, dnode_t **dnp); 212 int dnode_hold_impl(struct objset_impl *dd, uint64_t object, int flag, 213 void *ref, dnode_t **dnp); 214 boolean_t dnode_add_ref(dnode_t *dn, void *ref); 215 void dnode_rele(dnode_t *dn, void *ref); 216 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 217 void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 218 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 219 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 220 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 221 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 222 void dnode_free(dnode_t *dn, dmu_tx_t *tx); 223 void dnode_byteswap(dnode_phys_t *dnp); 224 void dnode_buf_byteswap(void *buf, size_t size); 225 void dnode_verify(dnode_t *dn); 226 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 227 uint64_t dnode_current_max_length(dnode_t *dn); 228 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 229 void dnode_clear_range(dnode_t *dn, uint64_t blkid, 230 uint64_t nblks, dmu_tx_t *tx); 231 void dnode_diduse_space(dnode_t *dn, int64_t space); 232 void dnode_willuse_space(dnode_t *dn, int64_t space, dmu_tx_t *tx); 233 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); 234 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 235 void dnode_init(void); 236 void dnode_fini(void); 237 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 238 int minlvl, uint64_t blkfill, uint64_t txg); 239 void dnode_evict_dbufs(dnode_t *dn); 240 241 #ifdef ZFS_DEBUG 242 243 /* 244 * There should be a ## between the string literal and fmt, to make it 245 * clear that we're joining two strings together, but that piece of shit 246 * gcc doesn't support that preprocessor token. 247 */ 248 #define dprintf_dnode(dn, fmt, ...) do { \ 249 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 250 char __db_buf[32]; \ 251 uint64_t __db_obj = (dn)->dn_object; \ 252 if (__db_obj == DMU_META_DNODE_OBJECT) \ 253 (void) strcpy(__db_buf, "mdn"); \ 254 else \ 255 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 256 (u_longlong_t)__db_obj);\ 257 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 258 __db_buf, __VA_ARGS__); \ 259 } \ 260 _NOTE(CONSTCOND) } while (0) 261 262 #define DNODE_VERIFY(dn) dnode_verify(dn) 263 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 264 265 #else 266 267 #define dprintf_dnode(db, fmt, ...) 268 #define DNODE_VERIFY(dn) 269 #define FREE_VERIFY(db, start, end, tx) 270 271 #endif 272 273 #ifdef __cplusplus 274 } 275 #endif 276 277 #endif /* _SYS_DNODE_H */ 278