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