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 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _SYS_DBUF_H 27 #define _SYS_DBUF_H 28 29 #pragma ident "%Z%%M% %I% %E% SMI" 30 31 #include <sys/dmu.h> 32 #include <sys/spa.h> 33 #include <sys/txg.h> 34 #include <sys/zio.h> 35 #include <sys/arc.h> 36 #include <sys/zfs_context.h> 37 #include <sys/refcount.h> 38 39 #ifdef __cplusplus 40 extern "C" { 41 #endif 42 43 #define DB_BONUS_BLKID (-1ULL) 44 #define IN_DMU_SYNC ((blkptr_t *)-1) 45 46 /* 47 * define flags for dbuf_read 48 */ 49 50 #define DB_RF_MUST_SUCCEED (1 << 0) 51 #define DB_RF_CANFAIL (1 << 1) 52 #define DB_RF_HAVESTRUCT (1 << 2) 53 #define DB_RF_NOPREFETCH (1 << 3) 54 #define DB_RF_NEVERWAIT (1 << 4) 55 #define DB_RF_CACHED (1 << 5) 56 57 /* 58 * The state transition diagram for dbufs looks like: 59 * 60 * +----> READ ----+ 61 * | | 62 * | V 63 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free) 64 * | ^ 65 * | | 66 * +----> FILL ----+ 67 */ 68 typedef enum dbuf_states { 69 DB_UNCACHED, 70 DB_FILL, 71 DB_READ, 72 DB_CACHED, 73 DB_EVICTING 74 } dbuf_states_t; 75 76 struct objset_impl; 77 struct dnode; 78 struct dmu_tx; 79 80 /* 81 * level = 0 means the user data 82 * level = 1 means the single indirect block 83 * etc. 84 */ 85 86 #define LIST_LINK_INACTIVE(link) \ 87 ((link)->list_next == NULL && (link)->list_prev == NULL) 88 89 typedef struct dmu_buf_impl { 90 /* 91 * The following members are immutable, with the exception of 92 * db.db_data, which is protected by db_mtx. 93 */ 94 95 /* the publicly visible structure */ 96 dmu_buf_t db; 97 98 /* the objset we belong to */ 99 struct objset_impl *db_objset; 100 101 /* 102 * the dnode we belong to (NULL when evicted) 103 */ 104 struct dnode *db_dnode; 105 106 /* 107 * our parent buffer; if the dnode points to us directly, 108 * db_parent == db_dnode->dn_dbuf 109 * only accessed by sync thread ??? 110 * (NULL when evicted) 111 */ 112 struct dmu_buf_impl *db_parent; 113 114 /* 115 * link for hash table of all dmu_buf_impl_t's 116 */ 117 struct dmu_buf_impl *db_hash_next; 118 119 /* our block number */ 120 uint64_t db_blkid; 121 122 /* 123 * Pointer to the blkptr_t which points to us. May be NULL if we 124 * don't have one yet. (NULL when evicted) 125 */ 126 blkptr_t *db_blkptr; 127 128 /* 129 * Our indirection level. Data buffers have db_level==0. 130 * Indirect buffers which point to data buffers have 131 * db_level==1. etc. Buffers which contain dnodes have 132 * db_level==0, since the dnodes are stored in a file. 133 */ 134 uint8_t db_level; 135 136 /* db_mtx protects the members below */ 137 kmutex_t db_mtx; 138 139 /* 140 * Current state of the buffer 141 */ 142 dbuf_states_t db_state; 143 144 /* 145 * Refcount accessed by dmu_buf_{hold,rele}. 146 * If nonzero, the buffer can't be destroyed. 147 * Protected by db_mtx. 148 */ 149 refcount_t db_holds; 150 151 /* buffer holding our data */ 152 arc_buf_t *db_buf; 153 154 kcondvar_t db_changed; 155 arc_buf_t *db_data_pending; 156 157 /* 158 * Last time (transaction group) this buffer was dirtied. 159 */ 160 uint64_t db_dirtied; 161 162 /* 163 * If db_dnode != NULL, our link on the owner dnodes's dn_dbufs list. 164 * Protected by its dn_dbufs_mtx. 165 */ 166 list_node_t db_link; 167 168 /* Our link on dn_dirty_dbufs[txg] */ 169 list_node_t db_dirty_node[TXG_SIZE]; 170 uint8_t db_dirtycnt; 171 172 /* 173 * Data which is unique to data (leaf) blocks: 174 */ 175 struct { 176 /* stuff we store for the user (see dmu_buf_set_user) */ 177 void *db_user_ptr; 178 void **db_user_data_ptr_ptr; 179 dmu_buf_evict_func_t *db_evict_func; 180 uint8_t db_immediate_evict; 181 uint8_t db_freed_in_flight; 182 183 /* 184 * db_data_old[txg&TXG_MASK] is set when we 185 * dirty the buffer, so that we can retain the 186 * pointer even if it gets COW'd in a subsequent 187 * transaction group. 188 * 189 * If the buffer is dirty in any txg, it can't 190 * be destroyed. 191 */ 192 /* 193 * XXX Protected by db_mtx and dn_dirty_mtx. 194 * db_mtx must be held to read db_dirty[], and 195 * both db_mtx and dn_dirty_mtx must be held to 196 * modify (dirty or clean). db_mtx must be held 197 * before dn_dirty_mtx. 198 */ 199 arc_buf_t *db_data_old[TXG_SIZE]; 200 blkptr_t *db_overridden_by[TXG_SIZE]; 201 } db_d; 202 } dmu_buf_impl_t; 203 204 /* Note: the dbuf hash table is exposed only for the mdb module */ 205 #define DBUF_MUTEXES 256 206 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) 207 typedef struct dbuf_hash_table { 208 uint64_t hash_table_mask; 209 dmu_buf_impl_t **hash_table; 210 kmutex_t hash_mutexes[DBUF_MUTEXES]; 211 } dbuf_hash_table_t; 212 213 214 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset); 215 216 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data); 217 dmu_buf_impl_t *dbuf_create_bonus(struct dnode *dn); 218 219 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag); 220 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, 221 void *tag); 222 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create, 223 void *tag, dmu_buf_impl_t **dbp); 224 225 void dbuf_prefetch(struct dnode *dn, uint64_t blkid); 226 227 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag); 228 uint64_t dbuf_refcount(dmu_buf_impl_t *db); 229 230 void dbuf_rele(dmu_buf_impl_t *db, void *tag); 231 232 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid); 233 234 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); 235 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 236 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 237 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx); 238 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 239 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); 240 void dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 241 242 void dbuf_clear(dmu_buf_impl_t *db); 243 void dbuf_evict(dmu_buf_impl_t *db); 244 245 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 246 void dbuf_sync(dmu_buf_impl_t *db, zio_t *zio, dmu_tx_t *tx); 247 void dbuf_unoverride(dmu_buf_impl_t *db, uint64_t txg); 248 249 void dbuf_free_range(struct dnode *dn, uint64_t blkid, uint64_t nblks, 250 struct dmu_tx *); 251 252 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); 253 254 void dbuf_init(void); 255 void dbuf_fini(void); 256 257 #define DBUF_GET_BUFC_TYPE(db) \ 258 ((((db)->db_level > 0) || \ 259 (dmu_ot[(db)->db_dnode->dn_type].ot_metadata)) ? \ 260 ARC_BUFC_METADATA : ARC_BUFC_DATA); 261 262 #ifdef ZFS_DEBUG 263 264 /* 265 * There should be a ## between the string literal and fmt, to make it 266 * clear that we're joining two strings together, but gcc does not 267 * support that preprocessor token. 268 */ 269 #define dprintf_dbuf(dbuf, fmt, ...) do { \ 270 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 271 char __db_buf[32]; \ 272 uint64_t __db_obj = (dbuf)->db.db_object; \ 273 if (__db_obj == DMU_META_DNODE_OBJECT) \ 274 (void) strcpy(__db_buf, "mdn"); \ 275 else \ 276 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 277 (u_longlong_t)__db_obj); \ 278 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ 279 "obj=%s lvl=%u blkid=%lld " fmt, \ 280 __db_buf, (dbuf)->db_level, \ 281 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ 282 } \ 283 _NOTE(CONSTCOND) } while (0) 284 285 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ 286 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 287 char __blkbuf[BP_SPRINTF_LEN]; \ 288 sprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ 289 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ 290 } \ 291 _NOTE(CONSTCOND) } while (0) 292 293 #define DBUF_VERIFY(db) dbuf_verify(db) 294 295 #else 296 297 #define dprintf_dbuf(db, fmt, ...) 298 #define dprintf_dbuf_bp(db, bp, fmt, ...) 299 #define DBUF_VERIFY(db) 300 301 #endif 302 303 304 #ifdef __cplusplus 305 } 306 #endif 307 308 #endif /* _SYS_DBUF_H */ 309