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 2010 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 #include <sys/dmu.h> 30 #include <sys/spa.h> 31 #include <sys/txg.h> 32 #include <sys/zio.h> 33 #include <sys/arc.h> 34 #include <sys/zfs_context.h> 35 #include <sys/refcount.h> 36 37 #ifdef __cplusplus 38 extern "C" { 39 #endif 40 41 #define IN_DMU_SYNC 2 42 43 /* 44 * define flags for dbuf_read 45 */ 46 47 #define DB_RF_MUST_SUCCEED (1 << 0) 48 #define DB_RF_CANFAIL (1 << 1) 49 #define DB_RF_HAVESTRUCT (1 << 2) 50 #define DB_RF_NOPREFETCH (1 << 3) 51 #define DB_RF_NEVERWAIT (1 << 4) 52 #define DB_RF_CACHED (1 << 5) 53 54 /* 55 * The simplified state transition diagram for dbufs looks like: 56 * 57 * +----> READ ----+ 58 * | | 59 * | V 60 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free) 61 * | ^ ^ 62 * | | | 63 * +----> FILL ----+ | 64 * | | 65 * | | 66 * +--------> NOFILL -------+ 67 */ 68 typedef enum dbuf_states { 69 DB_UNCACHED, 70 DB_FILL, 71 DB_NOFILL, 72 DB_READ, 73 DB_CACHED, 74 DB_EVICTING 75 } dbuf_states_t; 76 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 struct dmu_buf_impl; 90 91 typedef enum override_states { 92 DR_NOT_OVERRIDDEN, 93 DR_IN_DMU_SYNC, 94 DR_OVERRIDDEN 95 } override_states_t; 96 97 typedef struct dbuf_dirty_record { 98 /* link on our parents dirty list */ 99 list_node_t dr_dirty_node; 100 101 /* transaction group this data will sync in */ 102 uint64_t dr_txg; 103 104 /* zio of outstanding write IO */ 105 zio_t *dr_zio; 106 107 /* pointer back to our dbuf */ 108 struct dmu_buf_impl *dr_dbuf; 109 110 /* pointer to next dirty record */ 111 struct dbuf_dirty_record *dr_next; 112 113 /* pointer to parent dirty record */ 114 struct dbuf_dirty_record *dr_parent; 115 116 union dirty_types { 117 struct dirty_indirect { 118 119 /* protect access to list */ 120 kmutex_t dr_mtx; 121 122 /* Our list of dirty children */ 123 list_t dr_children; 124 } di; 125 struct dirty_leaf { 126 127 /* 128 * dr_data is set when we dirty the buffer 129 * so that we can retain the pointer even if it 130 * gets COW'd in a subsequent transaction group. 131 */ 132 arc_buf_t *dr_data; 133 blkptr_t dr_overridden_by; 134 override_states_t dr_override_state; 135 uint8_t dr_copies; 136 } dl; 137 } dt; 138 } dbuf_dirty_record_t; 139 140 typedef struct dmu_buf_impl { 141 /* 142 * The following members are immutable, with the exception of 143 * db.db_data, which is protected by db_mtx. 144 */ 145 146 /* the publicly visible structure */ 147 dmu_buf_t db; 148 149 /* the objset we belong to */ 150 struct objset *db_objset; 151 152 /* 153 * the dnode we belong to (NULL when evicted) 154 */ 155 struct dnode *db_dnode; 156 157 /* 158 * our parent buffer; if the dnode points to us directly, 159 * db_parent == db_dnode->dn_dbuf 160 * only accessed by sync thread ??? 161 * (NULL when evicted) 162 */ 163 struct dmu_buf_impl *db_parent; 164 165 /* 166 * link for hash table of all dmu_buf_impl_t's 167 */ 168 struct dmu_buf_impl *db_hash_next; 169 170 /* our block number */ 171 uint64_t db_blkid; 172 173 /* 174 * Pointer to the blkptr_t which points to us. May be NULL if we 175 * don't have one yet. (NULL when evicted) 176 */ 177 blkptr_t *db_blkptr; 178 179 /* 180 * Our indirection level. Data buffers have db_level==0. 181 * Indirect buffers which point to data buffers have 182 * db_level==1. etc. Buffers which contain dnodes have 183 * db_level==0, since the dnodes are stored in a file. 184 */ 185 uint8_t db_level; 186 187 /* db_mtx protects the members below */ 188 kmutex_t db_mtx; 189 190 /* 191 * Current state of the buffer 192 */ 193 dbuf_states_t db_state; 194 195 /* 196 * Refcount accessed by dmu_buf_{hold,rele}. 197 * If nonzero, the buffer can't be destroyed. 198 * Protected by db_mtx. 199 */ 200 refcount_t db_holds; 201 202 /* buffer holding our data */ 203 arc_buf_t *db_buf; 204 205 kcondvar_t db_changed; 206 dbuf_dirty_record_t *db_data_pending; 207 208 /* pointer to most recent dirty record for this buffer */ 209 dbuf_dirty_record_t *db_last_dirty; 210 211 /* 212 * Our link on the owner dnodes's dn_dbufs list. 213 * Protected by its dn_dbufs_mtx. 214 */ 215 list_node_t db_link; 216 217 /* Data which is unique to data (leaf) blocks: */ 218 219 /* stuff we store for the user (see dmu_buf_set_user) */ 220 void *db_user_ptr; 221 void **db_user_data_ptr_ptr; 222 dmu_buf_evict_func_t *db_evict_func; 223 224 uint8_t db_immediate_evict; 225 uint8_t db_freed_in_flight; 226 227 uint8_t db_dirtycnt; 228 } dmu_buf_impl_t; 229 230 /* Note: the dbuf hash table is exposed only for the mdb module */ 231 #define DBUF_MUTEXES 256 232 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) 233 typedef struct dbuf_hash_table { 234 uint64_t hash_table_mask; 235 dmu_buf_impl_t **hash_table; 236 kmutex_t hash_mutexes[DBUF_MUTEXES]; 237 } dbuf_hash_table_t; 238 239 240 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset); 241 242 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data); 243 void dbuf_create_bonus(struct dnode *dn); 244 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx); 245 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag); 246 247 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx); 248 249 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag); 250 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, 251 void *tag); 252 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create, 253 void *tag, dmu_buf_impl_t **dbp); 254 255 void dbuf_prefetch(struct dnode *dn, uint64_t blkid); 256 257 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag); 258 uint64_t dbuf_refcount(dmu_buf_impl_t *db); 259 260 void dbuf_rele(dmu_buf_impl_t *db, void *tag); 261 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag); 262 263 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid); 264 265 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); 266 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 267 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx); 268 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx); 269 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 270 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); 271 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx); 272 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 273 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db); 274 275 void dbuf_clear(dmu_buf_impl_t *db); 276 void dbuf_evict(dmu_buf_impl_t *db); 277 278 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 279 void dbuf_unoverride(dbuf_dirty_record_t *dr); 280 void dbuf_sync_list(list_t *list, dmu_tx_t *tx); 281 282 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, 283 struct dmu_tx *); 284 285 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); 286 287 void dbuf_init(void); 288 void dbuf_fini(void); 289 290 #define DBUF_IS_METADATA(db) \ 291 ((db)->db_level > 0 || dmu_ot[(db)->db_dnode->dn_type].ot_metadata) 292 293 #define DBUF_GET_BUFC_TYPE(db) \ 294 (DBUF_IS_METADATA(db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) 295 296 #define DBUF_IS_CACHEABLE(db) \ 297 ((db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ 298 (DBUF_IS_METADATA(db) && \ 299 ((db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) 300 301 #define DBUF_IS_L2CACHEABLE(db) \ 302 ((db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \ 303 (DBUF_IS_METADATA(db) && \ 304 ((db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA))) 305 306 #ifdef ZFS_DEBUG 307 308 /* 309 * There should be a ## between the string literal and fmt, to make it 310 * clear that we're joining two strings together, but gcc does not 311 * support that preprocessor token. 312 */ 313 #define dprintf_dbuf(dbuf, fmt, ...) do { \ 314 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 315 char __db_buf[32]; \ 316 uint64_t __db_obj = (dbuf)->db.db_object; \ 317 if (__db_obj == DMU_META_DNODE_OBJECT) \ 318 (void) strcpy(__db_buf, "mdn"); \ 319 else \ 320 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 321 (u_longlong_t)__db_obj); \ 322 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ 323 "obj=%s lvl=%u blkid=%lld " fmt, \ 324 __db_buf, (dbuf)->db_level, \ 325 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ 326 } \ 327 _NOTE(CONSTCOND) } while (0) 328 329 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ 330 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 331 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ 332 sprintf_blkptr(__blkbuf, bp); \ 333 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ 334 kmem_free(__blkbuf, BP_SPRINTF_LEN); \ 335 } \ 336 _NOTE(CONSTCOND) } while (0) 337 338 #define DBUF_VERIFY(db) dbuf_verify(db) 339 340 #else 341 342 #define dprintf_dbuf(db, fmt, ...) 343 #define dprintf_dbuf_bp(db, bp, fmt, ...) 344 #define DBUF_VERIFY(db) 345 346 #endif 347 348 349 #ifdef __cplusplus 350 } 351 #endif 352 353 #endif /* _SYS_DBUF_H */ 354