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, 2015 by Delphix. All rights reserved. 24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. 25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. 26 */ 27 28 #ifndef _SYS_DBUF_H 29 #define _SYS_DBUF_H 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 #include <sys/zrlock.h> 39 40 #ifdef __cplusplus 41 extern "C" { 42 #endif 43 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 simplified 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 * | | 69 * +--------> NOFILL -------+ 70 * 71 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range 72 * to find all dbufs in a range of a dnode and must be less than any other 73 * dbuf_states_t (see comment on dn_dbufs in dnode.h). 74 */ 75 typedef enum dbuf_states { 76 DB_SEARCH = -1, 77 DB_UNCACHED, 78 DB_FILL, 79 DB_NOFILL, 80 DB_READ, 81 DB_CACHED, 82 DB_EVICTING 83 } dbuf_states_t; 84 85 struct dnode; 86 struct dmu_tx; 87 88 /* 89 * level = 0 means the user data 90 * level = 1 means the single indirect block 91 * etc. 92 */ 93 94 struct dmu_buf_impl; 95 96 typedef enum override_states { 97 DR_NOT_OVERRIDDEN, 98 DR_IN_DMU_SYNC, 99 DR_OVERRIDDEN 100 } override_states_t; 101 102 typedef struct dbuf_dirty_record { 103 /* link on our parents dirty list */ 104 list_node_t dr_dirty_node; 105 106 /* transaction group this data will sync in */ 107 uint64_t dr_txg; 108 109 /* zio of outstanding write IO */ 110 zio_t *dr_zio; 111 112 /* pointer back to our dbuf */ 113 struct dmu_buf_impl *dr_dbuf; 114 115 /* pointer to next dirty record */ 116 struct dbuf_dirty_record *dr_next; 117 118 /* pointer to parent dirty record */ 119 struct dbuf_dirty_record *dr_parent; 120 121 /* How much space was changed to dsl_pool_dirty_space() for this? */ 122 unsigned int dr_accounted; 123 124 /* A copy of the bp that points to us */ 125 blkptr_t dr_bp_copy; 126 127 union dirty_types { 128 struct dirty_indirect { 129 130 /* protect access to list */ 131 kmutex_t dr_mtx; 132 133 /* Our list of dirty children */ 134 list_t dr_children; 135 } di; 136 struct dirty_leaf { 137 138 /* 139 * dr_data is set when we dirty the buffer 140 * so that we can retain the pointer even if it 141 * gets COW'd in a subsequent transaction group. 142 */ 143 arc_buf_t *dr_data; 144 blkptr_t dr_overridden_by; 145 override_states_t dr_override_state; 146 uint8_t dr_copies; 147 boolean_t dr_nopwrite; 148 } dl; 149 } dt; 150 } dbuf_dirty_record_t; 151 152 typedef struct dmu_buf_impl { 153 /* 154 * The following members are immutable, with the exception of 155 * db.db_data, which is protected by db_mtx. 156 */ 157 158 /* the publicly visible structure */ 159 dmu_buf_t db; 160 161 /* the objset we belong to */ 162 struct objset *db_objset; 163 164 /* 165 * handle to safely access the dnode we belong to (NULL when evicted) 166 */ 167 struct dnode_handle *db_dnode_handle; 168 169 /* 170 * our parent buffer; if the dnode points to us directly, 171 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf 172 * only accessed by sync thread ??? 173 * (NULL when evicted) 174 * May change from NULL to non-NULL under the protection of db_mtx 175 * (see dbuf_check_blkptr()) 176 */ 177 struct dmu_buf_impl *db_parent; 178 179 /* 180 * link for hash table of all dmu_buf_impl_t's 181 */ 182 struct dmu_buf_impl *db_hash_next; 183 184 /* our block number */ 185 uint64_t db_blkid; 186 187 /* 188 * Pointer to the blkptr_t which points to us. May be NULL if we 189 * don't have one yet. (NULL when evicted) 190 */ 191 blkptr_t *db_blkptr; 192 193 /* 194 * Our indirection level. Data buffers have db_level==0. 195 * Indirect buffers which point to data buffers have 196 * db_level==1. etc. Buffers which contain dnodes have 197 * db_level==0, since the dnodes are stored in a file. 198 */ 199 uint8_t db_level; 200 201 /* db_mtx protects the members below */ 202 kmutex_t db_mtx; 203 204 /* 205 * Current state of the buffer 206 */ 207 dbuf_states_t db_state; 208 209 /* 210 * Refcount accessed by dmu_buf_{hold,rele}. 211 * If nonzero, the buffer can't be destroyed. 212 * Protected by db_mtx. 213 */ 214 refcount_t db_holds; 215 216 /* buffer holding our data */ 217 arc_buf_t *db_buf; 218 219 kcondvar_t db_changed; 220 dbuf_dirty_record_t *db_data_pending; 221 222 /* pointer to most recent dirty record for this buffer */ 223 dbuf_dirty_record_t *db_last_dirty; 224 225 /* 226 * Our link on the owner dnodes's dn_dbufs list. 227 * Protected by its dn_dbufs_mtx. 228 */ 229 avl_node_t db_link; 230 231 /* Data which is unique to data (leaf) blocks: */ 232 233 /* User callback information. */ 234 dmu_buf_user_t *db_user; 235 236 /* 237 * Evict user data as soon as the dirty and reference 238 * counts are equal. 239 */ 240 uint8_t db_user_immediate_evict; 241 242 /* 243 * This block was freed while a read or write was 244 * active. 245 */ 246 uint8_t db_freed_in_flight; 247 248 /* 249 * dnode_evict_dbufs() or dnode_evict_bonus() tried to 250 * evict this dbuf, but couldn't due to outstanding 251 * references. Evict once the refcount drops to 0. 252 */ 253 uint8_t db_pending_evict; 254 255 uint8_t db_dirtycnt; 256 } dmu_buf_impl_t; 257 258 /* Note: the dbuf hash table is exposed only for the mdb module */ 259 #define DBUF_MUTEXES 256 260 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) 261 typedef struct dbuf_hash_table { 262 uint64_t hash_table_mask; 263 dmu_buf_impl_t **hash_table; 264 kmutex_t hash_mutexes[DBUF_MUTEXES]; 265 } dbuf_hash_table_t; 266 267 uint64_t dbuf_whichblock(struct dnode *di, int64_t level, uint64_t offset); 268 269 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data); 270 void dbuf_create_bonus(struct dnode *dn); 271 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx); 272 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag); 273 274 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx); 275 276 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag); 277 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, 278 void *tag); 279 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, 280 boolean_t fail_sparse, boolean_t fail_uncached, 281 void *tag, dmu_buf_impl_t **dbp); 282 283 void dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid, 284 zio_priority_t prio, arc_flags_t aflags); 285 286 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag); 287 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj, 288 uint64_t blkid, void *tag); 289 uint64_t dbuf_refcount(dmu_buf_impl_t *db); 290 291 void dbuf_rele(dmu_buf_impl_t *db, void *tag); 292 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag); 293 294 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level, 295 uint64_t blkid); 296 297 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); 298 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx); 299 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 300 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); 301 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx); 302 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 303 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db); 304 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, 305 bp_embedded_type_t etype, enum zio_compress comp, 306 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx); 307 308 void dbuf_clear(dmu_buf_impl_t *db); 309 void dbuf_evict(dmu_buf_impl_t *db); 310 311 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 312 void dbuf_unoverride(dbuf_dirty_record_t *dr); 313 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx); 314 void dbuf_release_bp(dmu_buf_impl_t *db); 315 316 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, 317 struct dmu_tx *); 318 319 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); 320 321 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode) 322 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock) 323 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db))) 324 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db))) 325 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db))) 326 327 void dbuf_init(void); 328 void dbuf_fini(void); 329 330 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db); 331 332 #define DBUF_GET_BUFC_TYPE(_db) \ 333 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) 334 335 #define DBUF_IS_CACHEABLE(_db) \ 336 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ 337 (dbuf_is_metadata(_db) && \ 338 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) 339 340 #define DBUF_IS_L2CACHEABLE(_db) \ 341 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \ 342 (dbuf_is_metadata(_db) && \ 343 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA))) 344 345 #define DBUF_IS_L2COMPRESSIBLE(_db) \ 346 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF || \ 347 (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE)) 348 349 #ifdef ZFS_DEBUG 350 351 /* 352 * There should be a ## between the string literal and fmt, to make it 353 * clear that we're joining two strings together, but gcc does not 354 * support that preprocessor token. 355 */ 356 #define dprintf_dbuf(dbuf, fmt, ...) do { \ 357 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 358 char __db_buf[32]; \ 359 uint64_t __db_obj = (dbuf)->db.db_object; \ 360 if (__db_obj == DMU_META_DNODE_OBJECT) \ 361 (void) strcpy(__db_buf, "mdn"); \ 362 else \ 363 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 364 (u_longlong_t)__db_obj); \ 365 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ 366 "obj=%s lvl=%u blkid=%lld " fmt, \ 367 __db_buf, (dbuf)->db_level, \ 368 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ 369 } \ 370 _NOTE(CONSTCOND) } while (0) 371 372 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ 373 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 374 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ 375 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ 376 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ 377 kmem_free(__blkbuf, BP_SPRINTF_LEN); \ 378 } \ 379 _NOTE(CONSTCOND) } while (0) 380 381 #define DBUF_VERIFY(db) dbuf_verify(db) 382 383 #else 384 385 #define dprintf_dbuf(db, fmt, ...) 386 #define dprintf_dbuf_bp(db, bp, fmt, ...) 387 #define DBUF_VERIFY(db) 388 389 #endif 390 391 392 #ifdef __cplusplus 393 } 394 #endif 395 396 #endif /* _SYS_DBUF_H */ 397