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) 2013 by Saso Kiselkov. All rights reserved.
  24  */
  25
  26 #ifndef _SYS_DBUF_H
  27 #define _SYS_DBUF_H
  28
  29 #include <sys/dmu.h>
  30 #include <sys/spa.h>
  31 #include <sys/txg.h>
  32 #include <sys/zio.h>
  33 #include <sys/arc.h>
  34 #include <sys/zfs_context.h>
  35 #include <sys/refcount.h>
  36 #include <sys/zrlock.h>
  37
  38 #ifdef  __cplusplus
  39 extern "C" {
  40 #endif
  41
  42 #define IN_DMU_SYNC 2
  43
  44 /*
  45  * define flags for dbuf_read
  46  */
  47
  48 #define DB_RF_MUST_SUCCEED      (1 << 0)
  49 #define DB_RF_CANFAIL           (1 << 1)
  50 #define DB_RF_HAVESTRUCT        (1 << 2)
  51 #define DB_RF_NOPREFETCH        (1 << 3)
  52 #define DB_RF_NEVERWAIT         (1 << 4)
  53 #define DB_RF_CACHED            (1 << 5)
  54
  55 /*
  56  * The simplified state transition diagram for dbufs looks like:
  57  *
  58  *              +----> READ ----+
  59  *              |               |
  60  *              |               V
  61  *  (alloc)-->UNCACHED       CACHED-->EVICTING-->(free)
  62  *              |               ^        ^
  63  *              |               |        |
  64  *              +----> FILL ----+        |
  65  *              |                        |
  66  *              |                        |
  67  *              +--------> NOFILL -------+
  68  */
  69 typedef enum dbuf_states {
  70         DB_UNCACHED,
  71         DB_FILL,
  72         DB_NOFILL,
  73         DB_READ,
  74         DB_CACHED,
  75         DB_EVICTING
  76 } dbuf_states_t;
  77
  78 struct dnode;
  79 struct dmu_tx;
  80
  81 /*
  82  * level = 0 means the user data
  83  * level = 1 means the single indirect block
  84  * etc.
  85  */
  86
  87 struct dmu_buf_impl;
  88
  89 typedef enum override_states {
  90         DR_NOT_OVERRIDDEN,
  91         DR_IN_DMU_SYNC,
  92         DR_OVERRIDDEN
  93 } override_states_t;
  94
  95 typedef struct dbuf_dirty_record {
  96         /* link on our parents dirty list */
  97         list_node_t dr_dirty_node;
  98
  99         /* transaction group this data will sync in */
 100         uint64_t dr_txg;
 101
 102         /* zio of outstanding write IO */
 103         zio_t *dr_zio;
 104
 105         /* pointer back to our dbuf */
 106         struct dmu_buf_impl *dr_dbuf;
 107
 108         /* pointer to next dirty record */
 109         struct dbuf_dirty_record *dr_next;
 110
 111         /* pointer to parent dirty record */
 112         struct dbuf_dirty_record *dr_parent;
 113
 114         union dirty_types {
 115                 struct dirty_indirect {
 116
 117                         /* protect access to list */
 118                         kmutex_t dr_mtx;
 119
 120                         /* Our list of dirty children */
 121                         list_t dr_children;
 122                 } di;
 123                 struct dirty_leaf {
 124
 125                         /*
 126                          * dr_data is set when we dirty the buffer
 127                          * so that we can retain the pointer even if it
 128                          * gets COW'd in a subsequent transaction group.
 129                          */
 130                         arc_buf_t *dr_data;
 131                         blkptr_t dr_overridden_by;
 132                         override_states_t dr_override_state;
 133                         uint8_t dr_copies;
 134                 } dl;
 135         } dt;
 136 } dbuf_dirty_record_t;
 137
 138 typedef struct dmu_buf_impl {
 139         /*
 140          * The following members are immutable, with the exception of
 141          * db.db_data, which is protected by db_mtx.
 142          */
 143
 144         /* the publicly visible structure */
 145         dmu_buf_t db;
 146
 147         /* the objset we belong to */
 148         struct objset *db_objset;
 149
 150         /*
 151          * handle to safely access the dnode we belong to (NULL when evicted)
 152          */
 153         struct dnode_handle *db_dnode_handle;
 154
 155         /*
 156          * our parent buffer; if the dnode points to us directly,
 157          * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
 158          * only accessed by sync thread ???
 159          * (NULL when evicted)
 160          * May change from NULL to non-NULL under the protection of db_mtx
 161          * (see dbuf_check_blkptr())
 162          */
 163         struct dmu_buf_impl *db_parent;
 164
 165         /*
 166          * link for hash table of all dmu_buf_impl_t's
 167          */
 168         struct dmu_buf_impl *db_hash_next;
 169
 170         /* our block number */
 171         uint64_t db_blkid;
 172
 173         /*
 174          * Pointer to the blkptr_t which points to us. May be NULL if we
 175          * don't have one yet. (NULL when evicted)
 176          */
 177         blkptr_t *db_blkptr;
 178
 179         /*
 180          * Our indirection level.  Data buffers have db_level==0.
 181          * Indirect buffers which point to data buffers have
 182          * db_level==1. etc.  Buffers which contain dnodes have
 183          * db_level==0, since the dnodes are stored in a file.
 184          */
 185         uint8_t db_level;
 186
 187         /* db_mtx protects the members below */
 188         kmutex_t db_mtx;
 189
 190         /*
 191          * Current state of the buffer
 192          */
 193         dbuf_states_t db_state;
 194
 195         /*
 196          * Refcount accessed by dmu_buf_{hold,rele}.
 197          * If nonzero, the buffer can't be destroyed.
 198          * Protected by db_mtx.
 199          */
 200         refcount_t db_holds;
 201
 202         /* buffer holding our data */
 203         arc_buf_t *db_buf;
 204
 205         kcondvar_t db_changed;
 206         dbuf_dirty_record_t *db_data_pending;
 207
 208         /* pointer to most recent dirty record for this buffer */
 209         dbuf_dirty_record_t *db_last_dirty;
 210
 211         /*
 212          * Our link on the owner dnodes's dn_dbufs list.
 213          * Protected by its dn_dbufs_mtx.
 214          */
 215         list_node_t db_link;
 216
 217         /* Data which is unique to data (leaf) blocks: */
 218
 219         /* stuff we store for the user (see dmu_buf_set_user) */
 220         void *db_user_ptr;
 221         void **db_user_data_ptr_ptr;
 222         dmu_buf_evict_func_t *db_evict_func;
 223
 224         uint8_t db_immediate_evict;
 225         uint8_t db_freed_in_flight;
 226
 227         uint8_t db_dirtycnt;
 228 } dmu_buf_impl_t;
 229
 230 /* Note: the dbuf hash table is exposed only for the mdb module */
 231 #define DBUF_MUTEXES 256
 232 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
 233 typedef struct dbuf_hash_table {
 234         uint64_t hash_table_mask;
 235         dmu_buf_impl_t **hash_table;
 236         kmutex_t hash_mutexes[DBUF_MUTEXES];
 237 } dbuf_hash_table_t;
 238
 239
 240 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
 241
 242 void dbuf_create_bonus(struct dnode *dn);
 243 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
 244
 245 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
 246
 247 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
 248 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
 249     void *tag);
 250 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
 251     void *tag, dmu_buf_impl_t **dbp);
 252
 253 void dbuf_prefetch(struct dnode *dn, uint64_t blkid);
 254
 255 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
 256 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
 257
 258 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
 259 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
 260
 261 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
 262
 263 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
 264 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 265 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
 266 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
 267 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
 268 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
 269 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
 270 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
 271 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
 272
 273 void dbuf_clear(dmu_buf_impl_t *db);
 274 void dbuf_evict(dmu_buf_impl_t *db);
 275
 276 void dbuf_unoverride(dbuf_dirty_record_t *dr);
 277 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
 278 void dbuf_release_bp(dmu_buf_impl_t *db);
 279
 280 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
 281     struct dmu_tx *);
 282
 283 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
 284
 285 #define DB_DNODE(_db)           ((_db)->db_dnode_handle->dnh_dnode)
 286 #define DB_DNODE_LOCK(_db)      ((_db)->db_dnode_handle->dnh_zrlock)
 287 #define DB_DNODE_ENTER(_db)     (zrl_add(&DB_DNODE_LOCK(_db)))
 288 #define DB_DNODE_EXIT(_db)      (zrl_remove(&DB_DNODE_LOCK(_db)))
 289 #define DB_DNODE_HELD(_db)      (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
 290 #define DB_GET_SPA(_spa_p, _db) {               \
 291         dnode_t *__dn;                          \
 292         DB_DNODE_ENTER(_db);                    \
 293         __dn = DB_DNODE(_db);                   \
 294         *(_spa_p) = __dn->dn_objset->os_spa;    \
 295         DB_DNODE_EXIT(_db);                     \
 296 }
 297 #define DB_GET_OBJSET(_os_p, _db) {             \
 298         dnode_t *__dn;                          \
 299         DB_DNODE_ENTER(_db);                    \
 300         __dn = DB_DNODE(_db);                   \
 301         *(_os_p) = __dn->dn_objset;             \
 302         DB_DNODE_EXIT(_db);                     \
 303 }
 304
 305 void dbuf_init(void);
 306 void dbuf_fini(void);
 307
 308 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
 309
 310 #define DBUF_GET_BUFC_TYPE(_db) \
 311         (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
 312
 313 #define DBUF_IS_CACHEABLE(_db)                                          \
 314         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||         \
 315         (dbuf_is_metadata(_db) &&                                       \
 316         ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
 317
 318 #define DBUF_IS_L2CACHEABLE(_db)                                        \
 319         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||       \
 320         (dbuf_is_metadata(_db) &&                                       \
 321         ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
 322
 323 #define DBUF_IS_L2COMPRESSIBLE(_db)                                     \
 324         ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF ||           \
 325         (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
 326
 327 #ifdef ZFS_DEBUG
 328
 329 /*
 330  * There should be a ## between the string literal and fmt, to make it
 331  * clear that we're joining two strings together, but gcc does not
 332  * support that preprocessor token.
 333  */
 334 #define dprintf_dbuf(dbuf, fmt, ...) do { \
 335         if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
 336         char __db_buf[32]; \
 337         uint64_t __db_obj = (dbuf)->db.db_object; \
 338         if (__db_obj == DMU_META_DNODE_OBJECT) \
 339                 (void) strcpy(__db_buf, "mdn"); \
 340         else \
 341                 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
 342                     (u_longlong_t)__db_obj); \
 343         dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
 344             "obj=%s lvl=%u blkid=%lld " fmt, \
 345             __db_buf, (dbuf)->db_level, \
 346             (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
 347         } \
 348 _NOTE(CONSTCOND) } while (0)
 349
 350 #define dprintf_dbuf_bp(db, bp, fmt, ...) do {                          \
 351         if (zfs_flags & ZFS_DEBUG_DPRINTF) {                            \
 352         char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_PUSHPAGE);       \
 353         sprintf_blkptr(__blkbuf, bp);                                   \
 354         dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);           \
 355         kmem_free(__blkbuf, BP_SPRINTF_LEN);                            \
 356         }                                                               \
 357 _NOTE(CONSTCOND) } while (0)
 358
 359 #define DBUF_VERIFY(db) dbuf_verify(db)
 360
 361 #else
 362
 363 #define dprintf_dbuf(db, fmt, ...)
 364 #define dprintf_dbuf_bp(db, bp, fmt, ...)
 365 #define DBUF_VERIFY(db)
 366
 367 #endif
 368
 369
 370 #ifdef  __cplusplus
 371 }
 372 #endif
 373
 374 #endif /* _SYS_DBUF_H */