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, 2020 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/zfs_refcount.h> 38 #include <sys/zrlock.h> 39 #include <sys/multilist.h> 40 41 #ifdef __cplusplus 42 extern "C" { 43 #endif 44 45 #define IN_DMU_SYNC 2 46 47 /* 48 * define flags for dbuf_read 49 */ 50 51 #define DB_RF_MUST_SUCCEED (1 << 0) 52 #define DB_RF_CANFAIL (1 << 1) 53 #define DB_RF_HAVESTRUCT (1 << 2) 54 #define DB_RF_NOPREFETCH (1 << 3) 55 #define DB_RF_NEVERWAIT (1 << 4) 56 #define DB_RF_CACHED (1 << 5) 57 #define DB_RF_NO_DECRYPT (1 << 6) 58 59 /* 60 * The simplified state transition diagram for dbufs looks like: 61 * 62 * +----> READ ----+ 63 * | | 64 * | V 65 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free) 66 * | ^ ^ 67 * | | | 68 * +----> FILL ----+ | 69 * | | 70 * | | 71 * +--------> NOFILL -------+ 72 * 73 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range 74 * to find all dbufs in a range of a dnode and must be less than any other 75 * dbuf_states_t (see comment on dn_dbufs in dnode.h). 76 */ 77 typedef enum dbuf_states { 78 DB_SEARCH = -1, 79 DB_UNCACHED, 80 DB_FILL, 81 DB_NOFILL, 82 DB_READ, 83 DB_CACHED, 84 DB_EVICTING 85 } dbuf_states_t; 86 87 typedef enum dbuf_cached_state { 88 DB_NO_CACHE = -1, 89 DB_DBUF_CACHE, 90 DB_DBUF_METADATA_CACHE, 91 DB_CACHE_MAX 92 } dbuf_cached_state_t; 93 94 struct dnode; 95 struct dmu_tx; 96 97 /* 98 * level = 0 means the user data 99 * level = 1 means the single indirect block 100 * etc. 101 */ 102 103 struct dmu_buf_impl; 104 105 typedef enum override_states { 106 DR_NOT_OVERRIDDEN, 107 DR_IN_DMU_SYNC, 108 DR_OVERRIDDEN 109 } override_states_t; 110 111 typedef enum db_lock_type { 112 DLT_NONE, 113 DLT_PARENT, 114 DLT_OBJSET 115 } db_lock_type_t; 116 117 typedef struct dbuf_dirty_record { 118 /* link on our parents dirty list */ 119 list_node_t dr_dirty_node; 120 121 /* transaction group this data will sync in */ 122 uint64_t dr_txg; 123 124 /* zio of outstanding write IO */ 125 zio_t *dr_zio; 126 127 /* pointer back to our dbuf */ 128 struct dmu_buf_impl *dr_dbuf; 129 130 /* list link for dbuf dirty records */ 131 list_node_t dr_dbuf_node; 132 133 /* 134 * The dnode we are part of. Note that the dnode can not be moved or 135 * evicted due to the hold that's added by dnode_setdirty() or 136 * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or 137 * userquota_updates_task(). This hold is necessary for 138 * dirty_lightweight_leaf-type dirty records, which don't have a hold 139 * on a dbuf. 140 */ 141 dnode_t *dr_dnode; 142 143 /* pointer to parent dirty record */ 144 struct dbuf_dirty_record *dr_parent; 145 146 /* How much space was changed to dsl_pool_dirty_space() for this? */ 147 unsigned int dr_accounted; 148 149 /* A copy of the bp that points to us */ 150 blkptr_t dr_bp_copy; 151 152 union dirty_types { 153 struct dirty_indirect { 154 155 /* protect access to list */ 156 kmutex_t dr_mtx; 157 158 /* Our list of dirty children */ 159 list_t dr_children; 160 } di; 161 struct dirty_leaf { 162 163 /* 164 * dr_data is set when we dirty the buffer 165 * so that we can retain the pointer even if it 166 * gets COW'd in a subsequent transaction group. 167 */ 168 arc_buf_t *dr_data; 169 blkptr_t dr_overridden_by; 170 override_states_t dr_override_state; 171 uint8_t dr_copies; 172 boolean_t dr_nopwrite; 173 boolean_t dr_has_raw_params; 174 175 /* 176 * If dr_has_raw_params is set, the following crypt 177 * params will be set on the BP that's written. 178 */ 179 boolean_t dr_byteorder; 180 uint8_t dr_salt[ZIO_DATA_SALT_LEN]; 181 uint8_t dr_iv[ZIO_DATA_IV_LEN]; 182 uint8_t dr_mac[ZIO_DATA_MAC_LEN]; 183 } dl; 184 struct dirty_lightweight_leaf { 185 /* 186 * This dirty record refers to a leaf (level=0) 187 * block, whose dbuf has not been instantiated for 188 * performance reasons. 189 */ 190 uint64_t dr_blkid; 191 abd_t *dr_abd; 192 zio_prop_t dr_props; 193 enum zio_flag dr_flags; 194 } dll; 195 } dt; 196 } dbuf_dirty_record_t; 197 198 typedef struct dmu_buf_impl { 199 /* 200 * The following members are immutable, with the exception of 201 * db.db_data, which is protected by db_mtx. 202 */ 203 204 /* the publicly visible structure */ 205 dmu_buf_t db; 206 207 /* the objset we belong to */ 208 struct objset *db_objset; 209 210 /* 211 * handle to safely access the dnode we belong to (NULL when evicted) 212 */ 213 struct dnode_handle *db_dnode_handle; 214 215 /* 216 * our parent buffer; if the dnode points to us directly, 217 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf 218 * only accessed by sync thread ??? 219 * (NULL when evicted) 220 * May change from NULL to non-NULL under the protection of db_mtx 221 * (see dbuf_check_blkptr()) 222 */ 223 struct dmu_buf_impl *db_parent; 224 225 /* 226 * link for hash table of all dmu_buf_impl_t's 227 */ 228 struct dmu_buf_impl *db_hash_next; 229 230 /* 231 * Our link on the owner dnodes's dn_dbufs list. 232 * Protected by its dn_dbufs_mtx. Should be on the same cache line 233 * as db_level and db_blkid for the best avl_add() performance. 234 */ 235 avl_node_t db_link; 236 237 /* our block number */ 238 uint64_t db_blkid; 239 240 /* 241 * Pointer to the blkptr_t which points to us. May be NULL if we 242 * don't have one yet. (NULL when evicted) 243 */ 244 blkptr_t *db_blkptr; 245 246 /* 247 * Our indirection level. Data buffers have db_level==0. 248 * Indirect buffers which point to data buffers have 249 * db_level==1. etc. Buffers which contain dnodes have 250 * db_level==0, since the dnodes are stored in a file. 251 */ 252 uint8_t db_level; 253 254 /* 255 * Protects db_buf's contents if they contain an indirect block or data 256 * block of the meta-dnode. We use this lock to protect the structure of 257 * the block tree. This means that when modifying this dbuf's data, we 258 * grab its rwlock. When modifying its parent's data (including the 259 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering 260 * for this lock is: 261 * 1) dn_struct_rwlock 262 * 2) db_rwlock 263 * We don't currently grab multiple dbufs' db_rwlocks at once. 264 */ 265 krwlock_t db_rwlock; 266 267 /* buffer holding our data */ 268 arc_buf_t *db_buf; 269 270 /* db_mtx protects the members below */ 271 kmutex_t db_mtx; 272 273 /* 274 * Current state of the buffer 275 */ 276 dbuf_states_t db_state; 277 278 /* 279 * Refcount accessed by dmu_buf_{hold,rele}. 280 * If nonzero, the buffer can't be destroyed. 281 * Protected by db_mtx. 282 */ 283 zfs_refcount_t db_holds; 284 285 kcondvar_t db_changed; 286 dbuf_dirty_record_t *db_data_pending; 287 288 /* List of dirty records for the buffer sorted newest to oldest. */ 289 list_t db_dirty_records; 290 291 /* Link in dbuf_cache or dbuf_metadata_cache */ 292 multilist_node_t db_cache_link; 293 294 /* Tells us which dbuf cache this dbuf is in, if any */ 295 dbuf_cached_state_t db_caching_status; 296 297 /* Data which is unique to data (leaf) blocks: */ 298 299 /* User callback information. */ 300 dmu_buf_user_t *db_user; 301 302 /* 303 * Evict user data as soon as the dirty and reference 304 * counts are equal. 305 */ 306 uint8_t db_user_immediate_evict; 307 308 /* 309 * This block was freed while a read or write was 310 * active. 311 */ 312 uint8_t db_freed_in_flight; 313 314 /* 315 * dnode_evict_dbufs() or dnode_evict_bonus() tried to 316 * evict this dbuf, but couldn't due to outstanding 317 * references. Evict once the refcount drops to 0. 318 */ 319 uint8_t db_pending_evict; 320 321 uint8_t db_dirtycnt; 322 } dmu_buf_impl_t; 323 324 /* Note: the dbuf hash table is exposed only for the mdb module */ 325 #define DBUF_MUTEXES 2048 326 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) 327 typedef struct dbuf_hash_table { 328 uint64_t hash_table_mask; 329 dmu_buf_impl_t **hash_table; 330 kmutex_t hash_mutexes[DBUF_MUTEXES] ____cacheline_aligned; 331 } dbuf_hash_table_t; 332 333 typedef void (*dbuf_prefetch_fn)(void *, boolean_t); 334 335 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level, 336 const uint64_t offset); 337 338 void dbuf_create_bonus(struct dnode *dn); 339 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx); 340 341 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx); 342 343 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag); 344 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, 345 void *tag); 346 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, 347 boolean_t fail_sparse, boolean_t fail_uncached, 348 void *tag, dmu_buf_impl_t **dbp); 349 350 int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid, 351 zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb, 352 void *arg); 353 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid, 354 zio_priority_t prio, arc_flags_t aflags); 355 356 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag); 357 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj, 358 uint64_t blkid, void *tag); 359 uint64_t dbuf_refcount(dmu_buf_impl_t *db); 360 361 void dbuf_rele(dmu_buf_impl_t *db, void *tag); 362 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting); 363 364 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level, 365 uint64_t blkid); 366 367 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); 368 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx); 369 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 370 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); 371 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx); 372 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 373 dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid, 374 dmu_tx_t *tx); 375 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db); 376 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, 377 bp_embedded_type_t etype, enum zio_compress comp, 378 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx); 379 380 int dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, 381 const struct zio_prop *zp, enum zio_flag flags, dmu_tx_t *tx); 382 383 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx); 384 void dbuf_destroy(dmu_buf_impl_t *db); 385 386 void dbuf_unoverride(dbuf_dirty_record_t *dr); 387 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx); 388 void dbuf_release_bp(dmu_buf_impl_t *db); 389 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, void *tag); 390 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, void *tag); 391 392 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, 393 struct dmu_tx *); 394 395 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); 396 397 void dbuf_stats_init(dbuf_hash_table_t *hash); 398 void dbuf_stats_destroy(void); 399 400 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid, 401 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift); 402 403 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode) 404 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock) 405 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db))) 406 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db))) 407 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db))) 408 409 void dbuf_init(void); 410 void dbuf_fini(void); 411 412 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db); 413 414 static inline dbuf_dirty_record_t * 415 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg) 416 { 417 dbuf_dirty_record_t *dr; 418 419 for (dr = list_head(&db->db_dirty_records); 420 dr != NULL && dr->dr_txg > txg; 421 dr = list_next(&db->db_dirty_records, dr)) 422 continue; 423 return (dr); 424 } 425 426 static inline dbuf_dirty_record_t * 427 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg) 428 { 429 dbuf_dirty_record_t *dr; 430 431 dr = dbuf_find_dirty_lte(db, txg); 432 if (dr && dr->dr_txg == txg) 433 return (dr); 434 return (NULL); 435 } 436 437 #define DBUF_GET_BUFC_TYPE(_db) \ 438 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) 439 440 #define DBUF_IS_CACHEABLE(_db) \ 441 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ 442 (dbuf_is_metadata(_db) && \ 443 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) 444 445 #define DBUF_IS_L2CACHEABLE(_db) \ 446 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \ 447 (dbuf_is_metadata(_db) && \ 448 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA))) 449 450 #define DNODE_LEVEL_IS_L2CACHEABLE(_dn, _level) \ 451 ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_ALL || \ 452 (((_level) > 0 || \ 453 DMU_OT_IS_METADATA((_dn)->dn_handle->dnh_dnode->dn_type)) && \ 454 ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_METADATA))) 455 456 #ifdef ZFS_DEBUG 457 458 /* 459 * There should be a ## between the string literal and fmt, to make it 460 * clear that we're joining two strings together, but gcc does not 461 * support that preprocessor token. 462 */ 463 #define dprintf_dbuf(dbuf, fmt, ...) do { \ 464 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 465 char __db_buf[32]; \ 466 uint64_t __db_obj = (dbuf)->db.db_object; \ 467 if (__db_obj == DMU_META_DNODE_OBJECT) \ 468 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \ 469 else \ 470 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 471 (u_longlong_t)__db_obj); \ 472 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ 473 "obj=%s lvl=%u blkid=%lld " fmt, \ 474 __db_buf, (dbuf)->db_level, \ 475 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ 476 } \ 477 } while (0) 478 479 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ 480 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 481 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ 482 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ 483 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ 484 kmem_free(__blkbuf, BP_SPRINTF_LEN); \ 485 } \ 486 } while (0) 487 488 #define DBUF_VERIFY(db) dbuf_verify(db) 489 490 #else 491 492 #define dprintf_dbuf(db, fmt, ...) 493 #define dprintf_dbuf_bp(db, bp, fmt, ...) 494 #define DBUF_VERIFY(db) 495 496 #endif 497 498 499 #ifdef __cplusplus 500 } 501 #endif 502 503 #endif /* _SYS_DBUF_H */ 504