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 2008 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 2 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 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 } dl; 136 } dt; 137 } dbuf_dirty_record_t; 138 139 typedef struct dmu_buf_impl { 140 /* 141 * The following members are immutable, with the exception of 142 * db.db_data, which is protected by db_mtx. 143 */ 144 145 /* the publicly visible structure */ 146 dmu_buf_t db; 147 148 /* the objset we belong to */ 149 struct objset_impl *db_objset; 150 151 /* 152 * the dnode we belong to (NULL when evicted) 153 */ 154 struct dnode *db_dnode; 155 156 /* 157 * our parent buffer; if the dnode points to us directly, 158 * db_parent == db_dnode->dn_dbuf 159 * only accessed by sync thread ??? 160 * (NULL when evicted) 161 */ 162 struct dmu_buf_impl *db_parent; 163 164 /* 165 * link for hash table of all dmu_buf_impl_t's 166 */ 167 struct dmu_buf_impl *db_hash_next; 168 169 /* our block number */ 170 uint64_t db_blkid; 171 172 /* 173 * Pointer to the blkptr_t which points to us. May be NULL if we 174 * don't have one yet. (NULL when evicted) 175 */ 176 blkptr_t *db_blkptr; 177 178 /* 179 * Our indirection level. Data buffers have db_level==0. 180 * Indirect buffers which point to data buffers have 181 * db_level==1. etc. Buffers which contain dnodes have 182 * db_level==0, since the dnodes are stored in a file. 183 */ 184 uint8_t db_level; 185 186 /* db_mtx protects the members below */ 187 kmutex_t db_mtx; 188 189 /* 190 * Current state of the buffer 191 */ 192 dbuf_states_t db_state; 193 194 /* 195 * Refcount accessed by dmu_buf_{hold,rele}. 196 * If nonzero, the buffer can't be destroyed. 197 * Protected by db_mtx. 198 */ 199 refcount_t db_holds; 200 201 /* buffer holding our data */ 202 arc_buf_t *db_buf; 203 204 kcondvar_t db_changed; 205 dbuf_dirty_record_t *db_data_pending; 206 207 /* pointer to most recent dirty record for this buffer */ 208 dbuf_dirty_record_t *db_last_dirty; 209 210 /* 211 * Our link on the owner dnodes's dn_dbufs list. 212 * Protected by its dn_dbufs_mtx. 213 */ 214 list_node_t db_link; 215 216 /* Data which is unique to data (leaf) blocks: */ 217 218 /* stuff we store for the user (see dmu_buf_set_user) */ 219 void *db_user_ptr; 220 void **db_user_data_ptr_ptr; 221 dmu_buf_evict_func_t *db_evict_func; 222 223 uint8_t db_immediate_evict; 224 uint8_t db_freed_in_flight; 225 226 uint8_t db_dirtycnt; 227 } dmu_buf_impl_t; 228 229 /* Note: the dbuf hash table is exposed only for the mdb module */ 230 #define DBUF_MUTEXES 256 231 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) 232 typedef struct dbuf_hash_table { 233 uint64_t hash_table_mask; 234 dmu_buf_impl_t **hash_table; 235 kmutex_t hash_mutexes[DBUF_MUTEXES]; 236 } dbuf_hash_table_t; 237 238 239 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset); 240 241 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data); 242 void dbuf_create_bonus(struct dnode *dn); 243 244 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag); 245 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, 246 void *tag); 247 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create, 248 void *tag, dmu_buf_impl_t **dbp); 249 250 void dbuf_prefetch(struct dnode *dn, uint64_t blkid); 251 252 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag); 253 uint64_t dbuf_refcount(dmu_buf_impl_t *db); 254 255 void dbuf_rele(dmu_buf_impl_t *db, void *tag); 256 257 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid); 258 259 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); 260 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 261 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 262 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx); 263 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 264 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); 265 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 266 267 void dbuf_clear(dmu_buf_impl_t *db); 268 void dbuf_evict(dmu_buf_impl_t *db); 269 270 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 271 void dbuf_unoverride(dbuf_dirty_record_t *dr); 272 void dbuf_sync_list(list_t *list, dmu_tx_t *tx); 273 274 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, 275 struct dmu_tx *); 276 277 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); 278 279 void dbuf_init(void); 280 void dbuf_fini(void); 281 282 #define DBUF_GET_BUFC_TYPE(db) \ 283 ((((db)->db_level > 0) || \ 284 (dmu_ot[(db)->db_dnode->dn_type].ot_metadata)) ? \ 285 ARC_BUFC_METADATA : ARC_BUFC_DATA); 286 287 #ifdef ZFS_DEBUG 288 289 /* 290 * There should be a ## between the string literal and fmt, to make it 291 * clear that we're joining two strings together, but gcc does not 292 * support that preprocessor token. 293 */ 294 #define dprintf_dbuf(dbuf, fmt, ...) do { \ 295 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 296 char __db_buf[32]; \ 297 uint64_t __db_obj = (dbuf)->db.db_object; \ 298 if (__db_obj == DMU_META_DNODE_OBJECT) \ 299 (void) strcpy(__db_buf, "mdn"); \ 300 else \ 301 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 302 (u_longlong_t)__db_obj); \ 303 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ 304 "obj=%s lvl=%u blkid=%lld " fmt, \ 305 __db_buf, (dbuf)->db_level, \ 306 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ 307 } \ 308 _NOTE(CONSTCOND) } while (0) 309 310 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ 311 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 312 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ 313 sprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ 314 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ 315 kmem_free(__blkbuf, BP_SPRINTF_LEN); \ 316 } \ 317 _NOTE(CONSTCOND) } while (0) 318 319 #define DBUF_VERIFY(db) dbuf_verify(db) 320 321 #else 322 323 #define dprintf_dbuf(db, fmt, ...) 324 #define dprintf_dbuf_bp(db, bp, fmt, ...) 325 #define DBUF_VERIFY(db) 326 327 #endif 328 329 330 #ifdef __cplusplus 331 } 332 #endif 333 334 #endif /* _SYS_DBUF_H */ 335