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/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 struct dnode;
  88 struct dmu_tx;
  89
  90 /*
  91  * level = 0 means the user data
  92  * level = 1 means the single indirect block
  93  * etc.
  94  */
  95
  96 struct dmu_buf_impl;
  97
  98 typedef enum override_states {
  99         DR_NOT_OVERRIDDEN,
 100         DR_IN_DMU_SYNC,
 101         DR_OVERRIDDEN
 102 } override_states_t;
 103
 104 typedef struct dbuf_dirty_record {
 105         /* link on our parents dirty list */
 106         list_node_t dr_dirty_node;
 107
 108         /* transaction group this data will sync in */
 109         uint64_t dr_txg;
 110
 111         /* zio of outstanding write IO */
 112         zio_t *dr_zio;
 113
 114         /* pointer back to our dbuf */
 115         struct dmu_buf_impl *dr_dbuf;
 116
 117         /* pointer to next dirty record */
 118         struct dbuf_dirty_record *dr_next;
 119
 120         /* pointer to parent dirty record */
 121         struct dbuf_dirty_record *dr_parent;
 122
 123         /* How much space was changed to dsl_pool_dirty_space() for this? */
 124         unsigned int dr_accounted;
 125
 126         /* A copy of the bp that points to us */
 127         blkptr_t dr_bp_copy;
 128
 129         union dirty_types {
 130                 struct dirty_indirect {
 131
 132                         /* protect access to list */
 133                         kmutex_t dr_mtx;
 134
 135                         /* Our list of dirty children */
 136                         list_t dr_children;
 137                 } di;
 138                 struct dirty_leaf {
 139
 140                         /*
 141                          * dr_data is set when we dirty the buffer
 142                          * so that we can retain the pointer even if it
 143                          * gets COW'd in a subsequent transaction group.
 144                          */
 145                         arc_buf_t *dr_data;
 146                         blkptr_t dr_overridden_by;
 147                         override_states_t dr_override_state;
 148                         uint8_t dr_copies;
 149                         boolean_t dr_nopwrite;
 150                         boolean_t dr_has_raw_params;
 151
 152                         /*
 153                          * If dr_has_raw_params is set, the following crypt
 154                          * params will be set on the BP that's written.
 155                          */
 156                         boolean_t dr_byteorder;
 157                         uint8_t dr_salt[ZIO_DATA_SALT_LEN];
 158                         uint8_t dr_iv[ZIO_DATA_IV_LEN];
 159                         uint8_t dr_mac[ZIO_DATA_MAC_LEN];
 160                 } dl;
 161         } dt;
 162 } dbuf_dirty_record_t;
 163
 164 typedef struct dmu_buf_impl {
 165         /*
 166          * The following members are immutable, with the exception of
 167          * db.db_data, which is protected by db_mtx.
 168          */
 169
 170         /* the publicly visible structure */
 171         dmu_buf_t db;
 172
 173         /* the objset we belong to */
 174         struct objset *db_objset;
 175
 176         /*
 177          * handle to safely access the dnode we belong to (NULL when evicted)
 178          */
 179         struct dnode_handle *db_dnode_handle;
 180
 181         /*
 182          * our parent buffer; if the dnode points to us directly,
 183          * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
 184          * only accessed by sync thread ???
 185          * (NULL when evicted)
 186          * May change from NULL to non-NULL under the protection of db_mtx
 187          * (see dbuf_check_blkptr())
 188          */
 189         struct dmu_buf_impl *db_parent;
 190
 191         /*
 192          * link for hash table of all dmu_buf_impl_t's
 193          */
 194         struct dmu_buf_impl *db_hash_next;
 195
 196         /* our block number */
 197         uint64_t db_blkid;
 198
 199         /*
 200          * Pointer to the blkptr_t which points to us. May be NULL if we
 201          * don't have one yet. (NULL when evicted)
 202          */
 203         blkptr_t *db_blkptr;
 204
 205         /*
 206          * Our indirection level.  Data buffers have db_level==0.
 207          * Indirect buffers which point to data buffers have
 208          * db_level==1. etc.  Buffers which contain dnodes have
 209          * db_level==0, since the dnodes are stored in a file.
 210          */
 211         uint8_t db_level;
 212
 213         /* db_mtx protects the members below */
 214         kmutex_t db_mtx;
 215
 216         /*
 217          * Current state of the buffer
 218          */
 219         dbuf_states_t db_state;
 220
 221         /*
 222          * Refcount accessed by dmu_buf_{hold,rele}.
 223          * If nonzero, the buffer can't be destroyed.
 224          * Protected by db_mtx.
 225          */
 226         refcount_t db_holds;
 227
 228         /* buffer holding our data */
 229         arc_buf_t *db_buf;
 230
 231         kcondvar_t db_changed;
 232         dbuf_dirty_record_t *db_data_pending;
 233
 234         /* pointer to most recent dirty record for this buffer */
 235         dbuf_dirty_record_t *db_last_dirty;
 236
 237         /*
 238          * Our link on the owner dnodes's dn_dbufs list.
 239          * Protected by its dn_dbufs_mtx.
 240          */
 241         avl_node_t db_link;
 242
 243         /*
 244          * Link in dbuf_cache.
 245          */
 246         multilist_node_t db_cache_link;
 247
 248         /* Data which is unique to data (leaf) blocks: */
 249
 250         /* User callback information. */
 251         dmu_buf_user_t *db_user;
 252
 253         /*
 254          * Evict user data as soon as the dirty and reference
 255          * counts are equal.
 256          */
 257         uint8_t db_user_immediate_evict;
 258
 259         /*
 260          * This block was freed while a read or write was
 261          * active.
 262          */
 263         uint8_t db_freed_in_flight;
 264
 265         /*
 266          * dnode_evict_dbufs() or dnode_evict_bonus() tried to
 267          * evict this dbuf, but couldn't due to outstanding
 268          * references.  Evict once the refcount drops to 0.
 269          */
 270         uint8_t db_pending_evict;
 271
 272         uint8_t db_dirtycnt;
 273 } dmu_buf_impl_t;
 274
 275 /* Note: the dbuf hash table is exposed only for the mdb module */
 276 #define DBUF_MUTEXES 8192
 277 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
 278 typedef struct dbuf_hash_table {
 279         uint64_t hash_table_mask;
 280         dmu_buf_impl_t **hash_table;
 281         kmutex_t hash_mutexes[DBUF_MUTEXES];
 282 } dbuf_hash_table_t;
 283
 284 uint64_t dbuf_whichblock(const struct dnode *di, const int64_t level,
 285     const uint64_t offset);
 286
 287 void dbuf_create_bonus(struct dnode *dn);
 288 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
 289
 290 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
 291
 292 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
 293 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
 294     void *tag);
 295 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
 296     boolean_t fail_sparse, boolean_t fail_uncached,
 297     void *tag, dmu_buf_impl_t **dbp);
 298
 299 void dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
 300     zio_priority_t prio, arc_flags_t aflags);
 301
 302 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
 303 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
 304     uint64_t blkid, void *tag);
 305 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
 306
 307 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
 308 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting);
 309
 310 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
 311     uint64_t blkid);
 312
 313 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
 314 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
 315 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
 316 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
 317 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
 318 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 319 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
 320 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
 321     bp_embedded_type_t etype, enum zio_compress comp,
 322     int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
 323
 324 void dbuf_destroy(dmu_buf_impl_t *db);
 325
 326 void dbuf_unoverride(dbuf_dirty_record_t *dr);
 327 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
 328 void dbuf_release_bp(dmu_buf_impl_t *db);
 329
 330 boolean_t dbuf_can_remap(const dmu_buf_impl_t *buf);
 331
 332 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
 333     struct dmu_tx *);
 334
 335 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
 336
 337 void dbuf_stats_init(dbuf_hash_table_t *hash);
 338 void dbuf_stats_destroy(void);
 339
 340 #define DB_DNODE(_db)           ((_db)->db_dnode_handle->dnh_dnode)
 341 #define DB_DNODE_LOCK(_db)      ((_db)->db_dnode_handle->dnh_zrlock)
 342 #define DB_DNODE_ENTER(_db)     (zrl_add(&DB_DNODE_LOCK(_db)))
 343 #define DB_DNODE_EXIT(_db)      (zrl_remove(&DB_DNODE_LOCK(_db)))
 344 #define DB_DNODE_HELD(_db)      (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
 345
 346 void dbuf_init(void);
 347 void dbuf_fini(void);
 348
 349 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
 350
 351 #define DBUF_GET_BUFC_TYPE(_db) \
 352         (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
 353
 354 #define DBUF_IS_CACHEABLE(_db)                                          \
 355         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 356         (dbuf_is_metadata(_db) &&                                       \
 357         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
 358
 359 #define DBUF_IS_L2CACHEABLE(_db)                                        \
 360         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||       \
 361         (dbuf_is_metadata(_db) &&                                       \
 362         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 363
 364 #define DNODE_LEVEL_IS_L2CACHEABLE(_dn, _level)                         \
 365         ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_ALL ||       \
 366         (((_level) > 0 ||                                               \
 367         DMU_OT_IS_METADATA((_dn)->dn_handle->dnh_dnode->dn_type)) &&    \
 368         ((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 369
 370 #ifdef ZFS_DEBUG
 371
 372 /*
 373  * There should be a ## between the string literal and fmt, to make it
 374  * clear that we're joining two strings together, but gcc does not
 375  * support that preprocessor token.
 376  */
 377 #define dprintf_dbuf(dbuf, fmt, ...) do { \
 378         if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
 379         char __db_buf[32]; \
 380         uint64_t __db_obj = (dbuf)->db.db_object; \
 381         if (__db_obj == DMU_META_DNODE_OBJECT) \
 382                 (void) strcpy(__db_buf, "mdn"); \
 383         else \
 384                 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
 385                     (u_longlong_t)__db_obj); \
 386         dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
 387             "obj=%s lvl=%u blkid=%lld " fmt, \
 388             __db_buf, (dbuf)->db_level, \
 389             (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
 390         } \
 391 _NOTE(CONSTCOND) } while (0)
 392
 393 #define dprintf_dbuf_bp(db, bp, fmt, ...) do {                  \
 394         if (zfs_flags & ZFS_DEBUG_DPRINTF) {                    \
 395         char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);  \
 396         snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp);          \
 397         dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);   \
 398         kmem_free(__blkbuf, BP_SPRINTF_LEN);                    \
 399         }                                                       \
 400 _NOTE(CONSTCOND) } while (0)
 401
 402 #define DBUF_VERIFY(db) dbuf_verify(db)
 403
 404 #else
 405
 406 #define dprintf_dbuf(db, fmt, ...)
 407 #define dprintf_dbuf_bp(db, bp, fmt, ...)
 408 #define DBUF_VERIFY(db)
 409
 410 #endif
 411
 412
 413 #ifdef  __cplusplus
 414 }
 415 #endif
 416
 417 #endif /* _SYS_DBUF_H */