include/sys/dbuf.h

   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, 2018 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         /* pointer to parent dirty record */
 134         struct dbuf_dirty_record *dr_parent;
 135
 136         /* How much space was changed to dsl_pool_dirty_space() for this? */
 137         unsigned int dr_accounted;
 138
 139         /* A copy of the bp that points to us */
 140         blkptr_t dr_bp_copy;
 141
 142         union dirty_types {
 143                 struct dirty_indirect {
 144
 145                         /* protect access to list */
 146                         kmutex_t dr_mtx;
 147
 148                         /* Our list of dirty children */
 149                         list_t dr_children;
 150                 } di;
 151                 struct dirty_leaf {
 152
 153                         /*
 154                          * dr_data is set when we dirty the buffer
 155                          * so that we can retain the pointer even if it
 156                          * gets COW'd in a subsequent transaction group.
 157                          */
 158                         arc_buf_t *dr_data;
 159                         blkptr_t dr_overridden_by;
 160                         override_states_t dr_override_state;
 161                         uint8_t dr_copies;
 162                         boolean_t dr_nopwrite;
 163                         boolean_t dr_has_raw_params;
 164
 165                         /*
 166                          * If dr_has_raw_params is set, the following crypt
 167                          * params will be set on the BP that's written.
 168                          */
 169                         boolean_t dr_byteorder;
 170                         uint8_t dr_salt[ZIO_DATA_SALT_LEN];
 171                         uint8_t dr_iv[ZIO_DATA_IV_LEN];
 172                         uint8_t dr_mac[ZIO_DATA_MAC_LEN];
 173                 } dl;
 174         } dt;
 175 } dbuf_dirty_record_t;
 176
 177 typedef struct dmu_buf_impl {
 178         /*
 179          * The following members are immutable, with the exception of
 180          * db.db_data, which is protected by db_mtx.
 181          */
 182
 183         /* the publicly visible structure */
 184         dmu_buf_t db;
 185
 186         /* the objset we belong to */
 187         struct objset *db_objset;
 188
 189         /*
 190          * handle to safely access the dnode we belong to (NULL when evicted)
 191          */
 192         struct dnode_handle *db_dnode_handle;
 193
 194         /*
 195          * our parent buffer; if the dnode points to us directly,
 196          * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
 197          * only accessed by sync thread ???
 198          * (NULL when evicted)
 199          * May change from NULL to non-NULL under the protection of db_mtx
 200          * (see dbuf_check_blkptr())
 201          */
 202         struct dmu_buf_impl *db_parent;
 203
 204         /*
 205          * link for hash table of all dmu_buf_impl_t's
 206          */
 207         struct dmu_buf_impl *db_hash_next;
 208
 209         /*
 210          * Our link on the owner dnodes's dn_dbufs list.
 211          * Protected by its dn_dbufs_mtx.  Should be on the same cache line
 212          * as db_level and db_blkid for the best avl_add() performance.
 213          */
 214         avl_node_t db_link;
 215
 216         /* our block number */
 217         uint64_t db_blkid;
 218
 219         /*
 220          * Pointer to the blkptr_t which points to us. May be NULL if we
 221          * don't have one yet. (NULL when evicted)
 222          */
 223         blkptr_t *db_blkptr;
 224
 225         /*
 226          * Our indirection level.  Data buffers have db_level==0.
 227          * Indirect buffers which point to data buffers have
 228          * db_level==1. etc.  Buffers which contain dnodes have
 229          * db_level==0, since the dnodes are stored in a file.
 230          */
 231         uint8_t db_level;
 232
 233         /*
 234          * Protects db_buf's contents if they contain an indirect block or data
 235          * block of the meta-dnode. We use this lock to protect the structure of
 236          * the block tree. This means that when modifying this dbuf's data, we
 237          * grab its rwlock. When modifying its parent's data (including the
 238          * blkptr to this dbuf), we grab the parent's rwlock. The lock ordering
 239          * for this lock is:
 240          * 1) dn_struct_rwlock
 241          * 2) db_rwlock
 242          * We don't currently grab multiple dbufs' db_rwlocks at once.
 243          */
 244         krwlock_t db_rwlock;
 245
 246         /* buffer holding our data */
 247         arc_buf_t *db_buf;
 248
 249         /* db_mtx protects the members below */
 250         kmutex_t db_mtx;
 251
 252         /*
 253          * Current state of the buffer
 254          */
 255         dbuf_states_t db_state;
 256
 257         /*
 258          * Refcount accessed by dmu_buf_{hold,rele}.
 259          * If nonzero, the buffer can't be destroyed.
 260          * Protected by db_mtx.
 261          */
 262         zfs_refcount_t db_holds;
 263
 264         kcondvar_t db_changed;
 265         dbuf_dirty_record_t *db_data_pending;
 266
 267         /* List of dirty records for the buffer sorted newest to oldest. */
 268         list_t db_dirty_records;
 269
 270         /* Link in dbuf_cache or dbuf_metadata_cache */
 271         multilist_node_t db_cache_link;
 272
 273         /* Tells us which dbuf cache this dbuf is in, if any */
 274         dbuf_cached_state_t db_caching_status;
 275
 276         /* Data which is unique to data (leaf) blocks: */
 277
 278         /* User callback information. */
 279         dmu_buf_user_t *db_user;
 280
 281         /*
 282          * Evict user data as soon as the dirty and reference
 283          * counts are equal.
 284          */
 285         uint8_t db_user_immediate_evict;
 286
 287         /*
 288          * This block was freed while a read or write was
 289          * active.
 290          */
 291         uint8_t db_freed_in_flight;
 292
 293         /*
 294          * dnode_evict_dbufs() or dnode_evict_bonus() tried to
 295          * evict this dbuf, but couldn't due to outstanding
 296          * references.  Evict once the refcount drops to 0.
 297          */
 298         uint8_t db_pending_evict;
 299
 300         uint8_t db_dirtycnt;
 301 } dmu_buf_impl_t;
 302
 303 /* Note: the dbuf hash table is exposed only for the mdb module */
 304 #define DBUF_MUTEXES 8192
 305 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
 306 typedef struct dbuf_hash_table {
 307         uint64_t hash_table_mask;
 308         dmu_buf_impl_t **hash_table;
 309         kmutex_t hash_mutexes[DBUF_MUTEXES];
 310 } dbuf_hash_table_t;
 311
 312 typedef void (*dbuf_prefetch_fn)(void *, boolean_t);
 313
 314 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
 315     const uint64_t offset);
 316
 317 void dbuf_create_bonus(struct dnode *dn);
 318 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
 319
 320 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
 321
 322 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
 323 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
 324     void *tag);
 325 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
 326     boolean_t fail_sparse, boolean_t fail_uncached,
 327     void *tag, dmu_buf_impl_t **dbp);
 328
 329 int dbuf_prefetch_impl(struct dnode *dn, int64_t level, uint64_t blkid,
 330     zio_priority_t prio, arc_flags_t aflags, dbuf_prefetch_fn cb,
 331     void *arg);
 332 int dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
 333     zio_priority_t prio, arc_flags_t aflags);
 334
 335 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
 336 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
 337     uint64_t blkid, void *tag);
 338 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
 339
 340 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
 341 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting);
 342
 343 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
 344     uint64_t blkid);
 345
 346 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
 347 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
 348 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
 349 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
 350 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
 351 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 352 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
 353 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
 354     bp_embedded_type_t etype, enum zio_compress comp,
 355     int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
 356
 357 void dmu_buf_redact(dmu_buf_t *dbuf, dmu_tx_t *tx);
 358 void dbuf_destroy(dmu_buf_impl_t *db);
 359
 360 void dbuf_unoverride(dbuf_dirty_record_t *dr);
 361 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
 362 void dbuf_release_bp(dmu_buf_impl_t *db);
 363 db_lock_type_t dmu_buf_lock_parent(dmu_buf_impl_t *db, krw_t rw, void *tag);
 364 void dmu_buf_unlock_parent(dmu_buf_impl_t *db, db_lock_type_t type, void *tag);
 365
 366 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
 367     struct dmu_tx *);
 368
 369 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
 370
 371 void dbuf_stats_init(dbuf_hash_table_t *hash);
 372 void dbuf_stats_destroy(void);
 373
 374 int dbuf_dnode_findbp(dnode_t *dn, uint64_t level, uint64_t blkid,
 375     blkptr_t *bp, uint16_t *datablkszsec, uint8_t *indblkshift);
 376
 377 #define DB_DNODE(_db)           ((_db)->db_dnode_handle->dnh_dnode)
 378 #define DB_DNODE_LOCK(_db)      ((_db)->db_dnode_handle->dnh_zrlock)
 379 #define DB_DNODE_ENTER(_db)     (zrl_add(&DB_DNODE_LOCK(_db)))
 380 #define DB_DNODE_EXIT(_db)      (zrl_remove(&DB_DNODE_LOCK(_db)))
 381 #define DB_DNODE_HELD(_db)      (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
 382
 383 void dbuf_init(void);
 384 void dbuf_fini(void);
 385
 386 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
 387
 388 static inline dbuf_dirty_record_t *
 389 dbuf_find_dirty_lte(dmu_buf_impl_t *db, uint64_t txg)
 390 {
 391         dbuf_dirty_record_t *dr;
 392
 393         for (dr = list_head(&db->db_dirty_records);
 394             dr != NULL && dr->dr_txg > txg;
 395             dr = list_next(&db->db_dirty_records, dr))
 396                 continue;
 397         return (dr);
 398 }
 399
 400 static inline dbuf_dirty_record_t *
 401 dbuf_find_dirty_eq(dmu_buf_impl_t *db, uint64_t txg)
 402 {
 403         dbuf_dirty_record_t *dr;
 404
 405         dr = dbuf_find_dirty_lte(db, txg);
 406         if (dr && dr->dr_txg == txg)
 407                 return (dr);
 408         return (NULL);
 409 }
 410
 411 #define DBUF_GET_BUFC_TYPE(_db) \
 412         (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
 413
 414 #define DBUF_IS_CACHEABLE(_db)                                          \
 415         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 416         (dbuf_is_metadata(_db) &&                                       \
 417         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
 418
 419 #define DBUF_IS_L2CACHEABLE(_db)                                        \
 420         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||       \
 421         (dbuf_is_metadata(_db) &&                                       \
 422         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 423
 424 #define DNODE_LEVEL_IS_L2CACHEABLE(_dn, _level)                         \
 425         ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_ALL ||       \
 426         (((_level) > 0 ||                                               \
 427         DMU_OT_IS_METADATA((_dn)->dn_handle->dnh_dnode->dn_type)) &&    \
 428         ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 429
 430 #ifdef ZFS_DEBUG
 431
 432 /*
 433  * There should be a ## between the string literal and fmt, to make it
 434  * clear that we're joining two strings together, but gcc does not
 435  * support that preprocessor token.
 436  */
 437 #define dprintf_dbuf(dbuf, fmt, ...) do { \
 438         if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
 439         char __db_buf[32]; \
 440         uint64_t __db_obj = (dbuf)->db.db_object; \
 441         if (__db_obj == DMU_META_DNODE_OBJECT) \
 442                 (void) strcpy(__db_buf, "mdn"); \
 443         else \
 444                 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
 445                     (u_longlong_t)__db_obj); \
 446         dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
 447             "obj=%s lvl=%u blkid=%lld " fmt, \
 448             __db_buf, (dbuf)->db_level, \
 449             (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
 450         } \
 451 _NOTE(CONSTCOND) } while (0)
 452
 453 #define dprintf_dbuf_bp(db, bp, fmt, ...) do {                  \
 454         if (zfs_flags & ZFS_DEBUG_DPRINTF) {                    \
 455         char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);  \
 456         snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp);          \
 457         dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);   \
 458         kmem_free(__blkbuf, BP_SPRINTF_LEN);                    \
 459         }                                                       \
 460 _NOTE(CONSTCOND) } while (0)
 461
 462 #define DBUF_VERIFY(db) dbuf_verify(db)
 463
 464 #else
 465
 466 #define dprintf_dbuf(db, fmt, ...)
 467 #define dprintf_dbuf_bp(db, bp, fmt, ...)
 468 #define DBUF_VERIFY(db)
 469
 470 #endif
 471
 472
 473 #ifdef  __cplusplus
 474 }
 475 #endif
 476
 477 #endif /* _SYS_DBUF_H */