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, 2016 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 40 extern "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 /* 62 * If we ever increase this value beyond 20, we need to revisit all logic that 63 * does x << level * ebps to handle overflow. With a 1M indirect block size, 64 * 4 levels of indirect blocks would not be able to guarantee addressing an 65 * entire object, so 5 levels will be used, but 5 * (20 - 7) = 65. 66 */ 67 #define DN_MAX_INDBLKSHIFT 17 /* 128k */ 68 #define DNODE_BLOCK_SHIFT 14 /* 16k */ 69 #define DNODE_CORE_SIZE 64 /* 64 bytes for dnode sans blkptrs */ 70 #define DN_MAX_OBJECT_SHIFT 48 /* 256 trillion (zfs_fid_t limit) */ 71 #define DN_MAX_OFFSET_SHIFT 64 /* 2^64 bytes in a dnode */ 72 73 /* 74 * dnode id flags 75 * 76 * Note: a file will never ever have its 77 * ids moved from bonus->spill 78 * and only in a crypto environment would it be on spill 79 */ 80 #define DN_ID_CHKED_BONUS 0x1 81 #define DN_ID_CHKED_SPILL 0x2 82 #define DN_ID_OLD_EXIST 0x4 83 #define DN_ID_NEW_EXIST 0x8 84 85 /* 86 * Derived constants. 87 */ 88 #define DNODE_SIZE (1 << DNODE_SHIFT) 89 #define DN_MAX_NBLKPTR ((DNODE_SIZE - DNODE_CORE_SIZE) >> SPA_BLKPTRSHIFT) 90 #define DN_MAX_BONUSLEN (DNODE_SIZE - DNODE_CORE_SIZE - (1 << SPA_BLKPTRSHIFT)) 91 #define DN_MAX_OBJECT (1ULL << DN_MAX_OBJECT_SHIFT) 92 #define DN_ZERO_BONUSLEN (DN_MAX_BONUSLEN + 1) 93 #define DN_KILL_SPILLBLK (1) 94 95 #define DNODES_PER_BLOCK_SHIFT (DNODE_BLOCK_SHIFT - DNODE_SHIFT) 96 #define DNODES_PER_BLOCK (1ULL << DNODES_PER_BLOCK_SHIFT) 97 98 /* 99 * This is inaccurate if the indblkshift of the particular object is not the 100 * max. But it's only used by userland to calculate the zvol reservation. 101 */ 102 #define DNODES_PER_LEVEL_SHIFT (DN_MAX_INDBLKSHIFT - SPA_BLKPTRSHIFT) 103 #define DNODES_PER_LEVEL (1ULL << DNODES_PER_LEVEL_SHIFT) 104 105 /* The +2 here is a cheesy way to round up */ 106 #define DN_MAX_LEVELS (2 + ((DN_MAX_OFFSET_SHIFT - SPA_MINBLOCKSHIFT) / \ 107 (DN_MIN_INDBLKSHIFT - SPA_BLKPTRSHIFT))) 108 109 #define DN_BONUS(dnp) ((void*)((dnp)->dn_bonus + \ 110 (((dnp)->dn_nblkptr - 1) * sizeof (blkptr_t)))) 111 112 #define DN_USED_BYTES(dnp) (((dnp)->dn_flags & DNODE_FLAG_USED_BYTES) ? \ 113 (dnp)->dn_used : (dnp)->dn_used << SPA_MINBLOCKSHIFT) 114 115 #define EPB(blkshift, typeshift) (1 << (blkshift - typeshift)) 116 117 struct dmu_buf_impl; 118 struct objset; 119 struct zio; 120 121 enum dnode_dirtycontext { 122 DN_UNDIRTIED, 123 DN_DIRTY_OPEN, 124 DN_DIRTY_SYNC 125 }; 126 127 /* Is dn_used in bytes? if not, it's in multiples of SPA_MINBLOCKSIZE */ 128 #define DNODE_FLAG_USED_BYTES (1<<0) 129 #define DNODE_FLAG_USERUSED_ACCOUNTED (1<<1) 130 131 /* Does dnode have a SA spill blkptr in bonus? */ 132 #define DNODE_FLAG_SPILL_BLKPTR (1<<2) 133 134 typedef struct dnode_phys { 135 uint8_t dn_type; /* dmu_object_type_t */ 136 uint8_t dn_indblkshift; /* ln2(indirect block size) */ 137 uint8_t dn_nlevels; /* 1=dn_blkptr->data blocks */ 138 uint8_t dn_nblkptr; /* length of dn_blkptr */ 139 uint8_t dn_bonustype; /* type of data in bonus buffer */ 140 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 141 uint8_t dn_compress; /* ZIO_COMPRESS type */ 142 uint8_t dn_flags; /* DNODE_FLAG_* */ 143 uint16_t dn_datablkszsec; /* data block size in 512b sectors */ 144 uint16_t dn_bonuslen; /* length of dn_bonus */ 145 uint8_t dn_pad2[4]; 146 147 /* accounting is protected by dn_dirty_mtx */ 148 uint64_t dn_maxblkid; /* largest allocated block ID */ 149 uint64_t dn_used; /* bytes (or sectors) of disk space */ 150 151 uint64_t dn_pad3[4]; 152 153 blkptr_t dn_blkptr[1]; 154 uint8_t dn_bonus[DN_MAX_BONUSLEN - sizeof (blkptr_t)]; 155 blkptr_t dn_spill; 156 } dnode_phys_t; 157 158 struct dnode { 159 /* 160 * Protects the structure of the dnode, including the number of levels 161 * of indirection (dn_nlevels), dn_maxblkid, and dn_next_* 162 */ 163 krwlock_t dn_struct_rwlock; 164 165 /* Our link on dn_objset->os_dnodes list; protected by os_lock. */ 166 list_node_t dn_link; 167 168 /* immutable: */ 169 struct objset *dn_objset; 170 uint64_t dn_object; 171 struct dmu_buf_impl *dn_dbuf; 172 struct dnode_handle *dn_handle; 173 dnode_phys_t *dn_phys; /* pointer into dn->dn_dbuf->db.db_data */ 174 175 /* 176 * Copies of stuff in dn_phys. They're valid in the open 177 * context (eg. even before the dnode is first synced). 178 * Where necessary, these are protected by dn_struct_rwlock. 179 */ 180 dmu_object_type_t dn_type; /* object type */ 181 uint16_t dn_bonuslen; /* bonus length */ 182 uint8_t dn_bonustype; /* bonus type */ 183 uint8_t dn_nblkptr; /* number of blkptrs (immutable) */ 184 uint8_t dn_checksum; /* ZIO_CHECKSUM type */ 185 uint8_t dn_compress; /* ZIO_COMPRESS type */ 186 uint8_t dn_nlevels; 187 uint8_t dn_indblkshift; 188 uint8_t dn_datablkshift; /* zero if blksz not power of 2! */ 189 uint8_t dn_moved; /* Has this dnode been moved? */ 190 uint16_t dn_datablkszsec; /* in 512b sectors */ 191 uint32_t dn_datablksz; /* in bytes */ 192 uint64_t dn_maxblkid; 193 uint8_t dn_next_type[TXG_SIZE]; 194 uint8_t dn_next_nblkptr[TXG_SIZE]; 195 uint8_t dn_next_nlevels[TXG_SIZE]; 196 uint8_t dn_next_indblkshift[TXG_SIZE]; 197 uint8_t dn_next_bonustype[TXG_SIZE]; 198 uint8_t dn_rm_spillblk[TXG_SIZE]; /* for removing spill blk */ 199 uint16_t dn_next_bonuslen[TXG_SIZE]; 200 uint32_t dn_next_blksz[TXG_SIZE]; /* next block size in bytes */ 201 202 /* protected by dn_dbufs_mtx; declared here to fill 32-bit hole */ 203 uint32_t dn_dbufs_count; /* count of dn_dbufs */ 204 205 /* protected by os_lock: */ 206 list_node_t dn_dirty_link[TXG_SIZE]; /* next on dataset's dirty */ 207 208 /* protected by dn_mtx: */ 209 kmutex_t dn_mtx; 210 list_t dn_dirty_records[TXG_SIZE]; 211 struct range_tree *dn_free_ranges[TXG_SIZE]; 212 uint64_t dn_allocated_txg; 213 uint64_t dn_free_txg; 214 uint64_t dn_assigned_txg; 215 kcondvar_t dn_notxholds; 216 enum dnode_dirtycontext dn_dirtyctx; 217 uint8_t *dn_dirtyctx_firstset; /* dbg: contents meaningless */ 218 219 /* protected by own devices */ 220 refcount_t dn_tx_holds; 221 refcount_t dn_holds; 222 223 kmutex_t dn_dbufs_mtx; 224 /* 225 * Descendent dbufs, ordered by dbuf_compare. Note that dn_dbufs 226 * can contain multiple dbufs of the same (level, blkid) when a 227 * dbuf is marked DB_EVICTING without being removed from 228 * dn_dbufs. To maintain the avl invariant that there cannot be 229 * duplicate entries, we order the dbufs by an arbitrary value - 230 * their address in memory. This means that dn_dbufs cannot be used to 231 * directly look up a dbuf. Instead, callers must use avl_walk, have 232 * a reference to the dbuf, or look up a non-existant node with 233 * db_state = DB_SEARCH (see dbuf_free_range for an example). 234 */ 235 avl_tree_t dn_dbufs; 236 237 /* protected by dn_struct_rwlock */ 238 struct dmu_buf_impl *dn_bonus; /* bonus buffer dbuf */ 239 240 boolean_t dn_have_spill; /* have spill or are spilling */ 241 242 /* parent IO for current sync write */ 243 zio_t *dn_zio; 244 245 /* used in syncing context */ 246 uint64_t dn_oldused; /* old phys used bytes */ 247 uint64_t dn_oldflags; /* old phys dn_flags */ 248 uint64_t dn_olduid, dn_oldgid; 249 uint64_t dn_newuid, dn_newgid; 250 int dn_id_flags; 251 252 /* holds prefetch structure */ 253 struct zfetch dn_zfetch; 254 }; 255 256 /* 257 * Adds a level of indirection between the dbuf and the dnode to avoid 258 * iterating descendent dbufs in dnode_move(). Handles are not allocated 259 * individually, but as an array of child dnodes in dnode_hold_impl(). 260 */ 261 typedef struct dnode_handle { 262 /* Protects dnh_dnode from modification by dnode_move(). */ 263 zrlock_t dnh_zrlock; 264 dnode_t *dnh_dnode; 265 } dnode_handle_t; 266 267 typedef struct dnode_children { 268 dmu_buf_user_t dnc_dbu; /* User evict data */ 269 size_t dnc_count; /* number of children */ 270 dnode_handle_t dnc_children[]; /* sized dynamically */ 271 } dnode_children_t; 272 273 typedef struct free_range { 274 avl_node_t fr_node; 275 uint64_t fr_blkid; 276 uint64_t fr_nblks; 277 } free_range_t; 278 279 void dnode_special_open(struct objset *dd, dnode_phys_t *dnp, 280 uint64_t object, dnode_handle_t *dnh); 281 void dnode_special_close(dnode_handle_t *dnh); 282 283 void dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx); 284 void dnode_setbonus_type(dnode_t *dn, dmu_object_type_t, dmu_tx_t *tx); 285 void dnode_rm_spill(dnode_t *dn, dmu_tx_t *tx); 286 287 int dnode_hold(struct objset *dd, uint64_t object, 288 void *ref, dnode_t **dnp); 289 int dnode_hold_impl(struct objset *dd, uint64_t object, int flag, 290 void *ref, dnode_t **dnp); 291 boolean_t dnode_add_ref(dnode_t *dn, void *ref); 292 void dnode_rele(dnode_t *dn, void *ref); 293 void dnode_rele_and_unlock(dnode_t *dn, void *tag); 294 void dnode_setdirty(dnode_t *dn, dmu_tx_t *tx); 295 void dnode_sync(dnode_t *dn, dmu_tx_t *tx); 296 void dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs, 297 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 298 void dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, 299 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx); 300 void dnode_free(dnode_t *dn, dmu_tx_t *tx); 301 void dnode_byteswap(dnode_phys_t *dnp); 302 void dnode_buf_byteswap(void *buf, size_t size); 303 void dnode_verify(dnode_t *dn); 304 int dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx); 305 void dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx); 306 void dnode_diduse_space(dnode_t *dn, int64_t space); 307 void dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t); 308 uint64_t dnode_block_freed(dnode_t *dn, uint64_t blkid); 309 void dnode_init(void); 310 void dnode_fini(void); 311 int dnode_next_offset(dnode_t *dn, int flags, uint64_t *off, 312 int minlvl, uint64_t blkfill, uint64_t txg); 313 void dnode_evict_dbufs(dnode_t *dn); 314 void dnode_evict_bonus(dnode_t *dn); 315 316 #define DNODE_IS_CACHEABLE(_dn) \ 317 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 318 (DMU_OT_IS_METADATA((_dn)->dn_type) && \ 319 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA)) 320 321 #define DNODE_META_IS_CACHEABLE(_dn) \ 322 ((_dn)->dn_objset->os_primary_cache == ZFS_CACHE_ALL || \ 323 (_dn)->dn_objset->os_primary_cache == ZFS_CACHE_METADATA) 324 325 #ifdef ZFS_DEBUG 326 327 /* 328 * There should be a ## between the string literal and fmt, to make it 329 * clear that we're joining two strings together, but that piece of shit 330 * gcc doesn't support that preprocessor token. 331 */ 332 #define dprintf_dnode(dn, fmt, ...) do { \ 333 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 334 char __db_buf[32]; \ 335 uint64_t __db_obj = (dn)->dn_object; \ 336 if (__db_obj == DMU_META_DNODE_OBJECT) \ 337 (void) strcpy(__db_buf, "mdn"); \ 338 else \ 339 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 340 (u_longlong_t)__db_obj);\ 341 dprintf_ds((dn)->dn_objset->os_dsl_dataset, "obj=%s " fmt, \ 342 __db_buf, __VA_ARGS__); \ 343 } \ 344 _NOTE(CONSTCOND) } while (0) 345 346 #define DNODE_VERIFY(dn) dnode_verify(dn) 347 #define FREE_VERIFY(db, start, end, tx) free_verify(db, start, end, tx) 348 349 #else 350 351 #define dprintf_dnode(db, fmt, ...) 352 #define DNODE_VERIFY(dn) 353 #define FREE_VERIFY(db, start, end, tx) 354 355 #endif 356 357 #ifdef __cplusplus 358 } 359 #endif 360 361 #endif /* _SYS_DNODE_H */ 362