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 https://opensource.org/licenses/CDDL-1.0. 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 #define DB_RF_PARTIAL_FIRST (1 << 7) 59 #define DB_RF_PARTIAL_MORE (1 << 8) 60 61 /* 62 * The simplified state transition diagram for dbufs looks like: 63 * 64 * +----> READ ----+ 65 * | | 66 * | V 67 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free) 68 * | ^ ^ 69 * | | | 70 * +----> FILL ----+ | 71 * | | 72 * | | 73 * +--------> NOFILL -------+ 74 * 75 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range 76 * to find all dbufs in a range of a dnode and must be less than any other 77 * dbuf_states_t (see comment on dn_dbufs in dnode.h). 78 */ 79 typedef enum dbuf_states { 80 DB_SEARCH = -1, 81 DB_UNCACHED, 82 DB_FILL, 83 DB_NOFILL, 84 DB_READ, 85 DB_CACHED, 86 DB_EVICTING 87 } dbuf_states_t; 88 89 typedef enum dbuf_cached_state { 90 DB_NO_CACHE = -1, 91 DB_DBUF_CACHE, 92 DB_DBUF_METADATA_CACHE, 93 DB_CACHE_MAX 94 } dbuf_cached_state_t; 95 96 struct dnode; 97 struct dmu_tx; 98 99 /* 100 * level = 0 means the user data 101 * level = 1 means the single indirect block 102 * etc. 103 */ 104 105 struct dmu_buf_impl; 106 107 typedef enum override_states { 108 DR_NOT_OVERRIDDEN, 109 DR_IN_DMU_SYNC, 110 DR_OVERRIDDEN 111 } override_states_t; 112 113 typedef enum db_lock_type { 114 DLT_NONE, 115 DLT_PARENT, 116 DLT_OBJSET 117 } db_lock_type_t; 118 119 typedef struct dbuf_dirty_record { 120 /* link on our parents dirty list */ 121 list_node_t dr_dirty_node; 122 123 /* transaction group this data will sync in */ 124 uint64_t dr_txg; 125 126 /* zio of outstanding write IO */ 127 zio_t *dr_zio; 128 129 /* pointer back to our dbuf */ 130 struct dmu_buf_impl *dr_dbuf; 131 132 /* list link for dbuf dirty records */ 133 list_node_t dr_dbuf_node; 134 135 /* 136 * The dnode we are part of. Note that the dnode can not be moved or 137 * evicted due to the hold that's added by dnode_setdirty() or 138 * dmu_objset_sync_dnodes(), and released by dnode_rele_task() or 139 * userquota_updates_task(). This hold is necessary for 140 * dirty_lightweight_leaf-type dirty records, which don't have a hold 141 * on a dbuf. 142 */ 143 dnode_t *dr_dnode; 144 145 /* pointer to parent dirty record */ 146 struct dbuf_dirty_record *dr_parent; 147 148 /* How much space was changed to dsl_pool_dirty_space() for this? */ 149 unsigned int dr_accounted; 150 151 /* A copy of the bp that points to us */ 152 blkptr_t dr_bp_copy; 153 154 union dirty_types { 155 struct dirty_indirect { 156 157 /* protect access to list */ 158 kmutex_t dr_mtx; 159 160 /* Our list of dirty children */ 161 list_t dr_children; 162 } di; 163 struct dirty_leaf { 164 165 /* 166 * dr_data is set when we dirty the buffer 167 * so that we can retain the pointer even if it 168 * gets COW'd in a subsequent transaction group. 169 */ 170 arc_buf_t *dr_data; 171 blkptr_t dr_overridden_by; 172 override_states_t dr_override_state; 173 uint8_t dr_copies; 174 boolean_t dr_nopwrite; 175 boolean_t dr_brtwrite; 176 boolean_t dr_has_raw_params; 177 178 /* 179 * If dr_has_raw_params is set, the following crypt 180 * params will be set on the BP that's written. 181 */ 182 boolean_t dr_byteorder; 183 uint8_t dr_salt[ZIO_DATA_SALT_LEN]; 184 uint8_t dr_iv[ZIO_DATA_IV_LEN]; 185 uint8_t dr_mac[ZIO_DATA_MAC_LEN]; 186 } dl; 187 struct dirty_lightweight_leaf { 188 /* 189 * This dirty record refers to a leaf (level=0) 190 * block, whose dbuf has not been instantiated for 191 * performance reasons. 192 */ 193 uint64_t dr_blkid; 194 abd_t *dr_abd; 195 zio_prop_t dr_props; 196 zio_flag_t dr_flags; 197 } dll; 198 } dt; 199 } dbuf_dirty_record_t; 200 201 typedef struct dmu_buf_impl { 202 /* 203 * The following members are immutable, with the exception of 204 * db.db_data, which is protected by db_mtx. 205 */ 206 207 /* the publicly visible structure */ 208 dmu_buf_t db; 209 210 /* the objset we belong to */ 211 struct objset *db_objset; 212 213 /* 214 * handle to safely access the dnode we belong to (NULL when evicted) 215 */ 216 struct dnode_handle *db_dnode_handle; 217 218 /* 219 * our parent buffer; if the dnode points to us directly, 220 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf 221 * only accessed by sync thread ??? 222 * (NULL when evicted) 223 * May change from NULL to non-NULL under the protection of db_mtx 224 * (see dbuf_check_blkptr()) 225 */ 226 struct dmu_buf_impl *db_parent; 227 228 /* 229 * link for hash table of all dmu_buf_impl_t's 230 */ 231 struct dmu_buf_impl *db_hash_next; 232 233 /* 234 * Our link on the owner dnodes's dn_dbufs list. 235 * Protected by its dn_dbufs_mtx. Should be on the same cache line 236 * as db_level and db_blkid for the best avl_add() performance. 237 */ 238 avl_node_t db_link; 239 240 /* our block number */ 241 uint64_t db_blkid; 242 243 /* 244 * Pointer to the blkptr_t which points to us. May be NULL if we 245 * don't have one yet. (NULL when evicted) 246 */ 247 blkptr_t *db_blkptr; 248 249 /* 250 * Our indirection level. Data buffers have db_level==0. 251 * Indirect buffers which point to data buffers have 252 * db_level==1. etc. Buffers which contain dnodes have 253 * db_level==0, since the dnodes are stored in a file. 254 */ 255 uint8_t db_level; 256 257 /* 258 * Protects db_buf's contents if they contain an indirect block or data 259 * block of the meta-dnode. We use this lock to protect the structure of 260 * the block tree. This means that when modifying this dbuf's data, we 261 * grab its rwlock. When modifying its parent's data (including the 262 * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering 263 * for this lock is: 264 * 1) dn_struct_rwlock 265 * 2) db_rwlock 266 * We don't currently grab multiple dbufs' db_rwlocks at once. 267 */ 268 krwlock_t db_rwlock; 269 270 /* buffer holding our data */ 271 arc_buf_t *db_buf; 272 273 /* db_mtx protects the members below */ 274 kmutex_t db_mtx; 275 276 /* 277 * Current state of the buffer 278 */ 279 dbuf_states_t db_state; 280 281 /* 282 * Refcount accessed by dmu_buf_{hold,rele}. 283 * If nonzero, the buffer can't be destroyed. 284 * Protected by db_mtx. 285 */ 286 zfs_refcount_t db_holds; 287 288 kcondvar_t db_changed; 289 dbuf_dirty_record_t *db_data_pending; 290 291 /* List of dirty records for the buffer sorted newest to oldest. */ 292 list_t db_dirty_records; 293 294 /* Link in dbuf_cache or dbuf_metadata_cache */ 295 multilist_node_t db_cache_link; 296 297 /* Tells us which dbuf cache this dbuf is in, if any */ 298 dbuf_cached_state_t db_caching_status; 299 300 uint64_t db_hash; 301 302 /* Data which is unique to data (leaf) blocks: */ 303 304 /* User callback information. */ 305 dmu_buf_user_t *db_user; 306 307 /* 308 * Evict user data as soon as the dirty and reference 309 * counts are equal. 310 */ 311 uint8_t db_user_immediate_evict; 312 313 /* 314 * This block was freed while a read or write was 315 * active. 316 */ 317 uint8_t db_freed_in_flight; 318 319 /* 320 * dnode_evict_dbufs() or dnode_evict_bonus() tried to 321 * evict this dbuf, but couldn't due to outstanding 322 * references. Evict once the refcount drops to 0. 323 */ 324 uint8_t db_pending_evict; 325 326 uint8_t db_dirtycnt; 327 328 /* The buffer was partially read. More reads may follow. */ 329 uint8_t db_partial_read; 330 } dmu_buf_impl_t; 331 332 #define DBUF_HASH_MUTEX(h, idx) \ 333 (&(h)->hash_mutexes[(idx) & ((h)->hash_mutex_mask)]) 334 335 typedef struct dbuf_hash_table { 336 uint64_t hash_table_mask; 337 uint64_t hash_mutex_mask; 338 dmu_buf_impl_t **hash_table; 339 kmutex_t *hash_mutexes; 340 } dbuf_hash_table_t; 341 342 typedef void (*dbuf_prefetch_fn)(void *, uint64_t, uint64_t, boolean_t); 343 344 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level, 345 const uint64_t offset); 346 347 void dbuf_create_bonus(struct dnode *dn); 348 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx); 349 350 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx); 351 352 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, const void *tag); 353 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid, 354 const void *tag); 355 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, 356 boolean_t fail_sparse, boolean_t fail_uncached, 357 const void *tag, dmu_buf_impl_t **dbp); 358 359 int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid, 360 zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb, 361 void *arg); 362 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid, 363 zio_priority_t prio, arc_flags_t aflags); 364 365 void dbuf_add_ref(dmu_buf_impl_t *db, const void *tag); 366 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj, 367 uint64_t blkid, const void *tag); 368 uint64_t dbuf_refcount(dmu_buf_impl_t *db); 369 370 void dbuf_rele(dmu_buf_impl_t *db, const void *tag); 371 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, const void *tag, 372 boolean_t evicting); 373 374 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level, 375 uint64_t blkid, uint64_t *hash_out); 376 377 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); 378 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx); 379 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); 380 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); 381 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx); 382 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 383 dbuf_dirty_record_t *dbuf_dirty_lightweight(dnode_t *dn, uint64_t blkid, 384 dmu_tx_t *tx); 385 boolean_t dbuf_undirty(dmu_buf_impl_t *db, dmu_tx_t *tx); 386 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db); 387 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, 388 bp_embedded_type_t etype, enum zio_compress comp, 389 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx); 390 391 int dmu_lightweight_write_by_dnode(dnode_t *dn, uint64_t offset, abd_t *abd, 392 const struct zio_prop *zp, zio_flag_t flags, dmu_tx_t *tx); 393 394 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx); 395 void dbuf_destroy(dmu_buf_impl_t *db); 396 397 void dbuf_unoverride(dbuf_dirty_record_t *dr); 398 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx); 399 void dbuf_release_bp(dmu_buf_impl_t *db); 400 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, 401 const void *tag); 402 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, 403 const void *tag); 404 405 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, 406 struct dmu_tx *); 407 408 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); 409 410 void dbuf_stats_init(dbuf_hash_table_t *hash); 411 void dbuf_stats_destroy(void); 412 413 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid, 414 blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift); 415 416 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode) 417 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock) 418 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db))) 419 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db))) 420 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db))) 421 422 void dbuf_init(void); 423 void dbuf_fini(void); 424 425 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db); 426 427 static inline dbuf_dirty_record_t * 428 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg) 429 { 430 dbuf_dirty_record_t *dr; 431 432 for (dr = list_head(&db->db_dirty_records); 433 dr != NULL && dr->dr_txg > txg; 434 dr = list_next(&db->db_dirty_records, dr)) 435 continue; 436 return (dr); 437 } 438 439 static inline dbuf_dirty_record_t * 440 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg) 441 { 442 dbuf_dirty_record_t *dr; 443 444 dr = dbuf_find_dirty_lte(db, txg); 445 if (dr && dr->dr_txg == txg) 446 return (dr); 447 return (NULL); 448 } 449 450 #define DBUF_GET_BUFC_TYPE(_db) \ 451 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) 452 453 #define DBUF_IS_CACHEABLE(_db) \ 454 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ 455 (dbuf_is_metadata(_db) && \ 456 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) 457 458 boolean_t dbuf_is_l2cacheable(dmu_buf_impl_t *db); 459 460 #ifdef ZFS_DEBUG 461 462 /* 463 * There should be a ## between the string literal and fmt, to make it 464 * clear that we're joining two strings together, but gcc does not 465 * support that preprocessor token. 466 */ 467 #define dprintf_dbuf(dbuf, fmt, ...) do { \ 468 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 469 char __db_buf[32]; \ 470 uint64_t __db_obj = (dbuf)->db.db_object; \ 471 if (__db_obj == DMU_META_DNODE_OBJECT) \ 472 (void) strlcpy(__db_buf, "mdn", sizeof (__db_buf)); \ 473 else \ 474 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ 475 (u_longlong_t)__db_obj); \ 476 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ 477 "obj=%s lvl=%u blkid=%lld " fmt, \ 478 __db_buf, (dbuf)->db_level, \ 479 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ 480 } \ 481 } while (0) 482 483 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ 484 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ 485 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ 486 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ 487 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ 488 kmem_free(__blkbuf, BP_SPRINTF_LEN); \ 489 } \ 490 } while (0) 491 492 #define DBUF_VERIFY(db) dbuf_verify(db) 493 494 #else 495 496 #define dprintf_dbuf(db, fmt, ...) 497 #define dprintf_dbuf_bp(db, bp, fmt, ...) 498 #define DBUF_VERIFY(db) 499 500 #endif 501 502 503 #ifdef __cplusplus 504 } 505 #endif 506 507 #endif /* _SYS_DBUF_H */ 508