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.
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.
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]
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
29 #pragma ident "%Z%%M% %I% %E% SMI"
36 #include <sys/zfs_context.h>
37 #include <sys/refcount.h>
43 #define DB_BONUS_BLKID (-1ULL)
47 * define flags for dbuf_read
50 #define DB_RF_MUST_SUCCEED (1 << 0)
51 #define DB_RF_CANFAIL (1 << 1)
52 #define DB_RF_HAVESTRUCT (1 << 2)
53 #define DB_RF_NOPREFETCH (1 << 3)
54 #define DB_RF_NEVERWAIT (1 << 4)
55 #define DB_RF_CACHED (1 << 5)
58 * The state transition diagram for dbufs looks like:
63 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
68 typedef enum dbuf_states {
81 * level = 0 means the user data
82 * level = 1 means the single indirect block
86 #define LIST_LINK_INACTIVE(link) \
87 ((link)->list_next == NULL && (link)->list_prev == NULL)
91 typedef enum override_states {
97 typedef struct dbuf_dirty_record {
98 /* link on our parents dirty list */
99 list_node_t dr_dirty_node;
101 /* transaction group this data will sync in */
104 /* zio of outstanding write IO */
107 /* pointer back to our dbuf */
108 struct dmu_buf_impl *dr_dbuf;
110 /* pointer to next dirty record */
111 struct dbuf_dirty_record *dr_next;
113 /* pointer to parent dirty record */
114 struct dbuf_dirty_record *dr_parent;
117 struct dirty_indirect {
119 /* protect access to list */
122 /* Our list of dirty children */
128 * dr_data is set when we dirty the buffer
129 * so that we can retain the pointer even if it
130 * gets COW'd in a subsequent transaction group.
133 blkptr_t dr_overridden_by;
134 override_states_t dr_override_state;
137 } dbuf_dirty_record_t;
139 typedef struct dmu_buf_impl {
141 * The following members are immutable, with the exception of
142 * db.db_data, which is protected by db_mtx.
145 /* the publicly visible structure */
148 /* the objset we belong to */
149 struct objset_impl *db_objset;
152 * the dnode we belong to (NULL when evicted)
154 struct dnode *db_dnode;
157 * our parent buffer; if the dnode points to us directly,
158 * db_parent == db_dnode->dn_dbuf
159 * only accessed by sync thread ???
160 * (NULL when evicted)
162 struct dmu_buf_impl *db_parent;
165 * link for hash table of all dmu_buf_impl_t's
167 struct dmu_buf_impl *db_hash_next;
169 /* our block number */
173 * Pointer to the blkptr_t which points to us. May be NULL if we
174 * don't have one yet. (NULL when evicted)
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.
186 /* db_mtx protects the members below */
190 * Current state of the buffer
192 dbuf_states_t db_state;
195 * Refcount accessed by dmu_buf_{hold,rele}.
196 * If nonzero, the buffer can't be destroyed.
197 * Protected by db_mtx.
201 /* buffer holding our data */
204 kcondvar_t db_changed;
205 dbuf_dirty_record_t *db_data_pending;
207 /* pointer to most recent dirty record for this buffer */
208 dbuf_dirty_record_t *db_last_dirty;
211 * Our link on the owner dnodes's dn_dbufs list.
212 * Protected by its dn_dbufs_mtx.
216 /* Data which is unique to data (leaf) blocks: */
218 /* stuff we store for the user (see dmu_buf_set_user) */
220 void **db_user_data_ptr_ptr;
221 dmu_buf_evict_func_t *db_evict_func;
223 uint8_t db_immediate_evict;
224 uint8_t db_freed_in_flight;
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];
239 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
241 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
242 void dbuf_create_bonus(struct dnode *dn);
244 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
245 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
247 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
248 void *tag, dmu_buf_impl_t **dbp);
250 void dbuf_prefetch(struct dnode *dn, uint64_t blkid);
252 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
253 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
255 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
257 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
259 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
260 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
261 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
262 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
263 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
264 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
265 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
266 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
268 void dbuf_clear(dmu_buf_impl_t *db);
269 void dbuf_evict(dmu_buf_impl_t *db);
271 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
272 void dbuf_unoverride(dbuf_dirty_record_t *dr);
273 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
275 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
278 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
280 void dbuf_init(void);
281 void dbuf_fini(void);
283 #define DBUF_IS_METADATA(db) \
284 ((db)->db_level > 0 || dmu_ot[(db)->db_dnode->dn_type].ot_metadata)
286 #define DBUF_GET_BUFC_TYPE(db) \
287 (DBUF_IS_METADATA(db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
289 #define DBUF_IS_CACHEABLE(db) \
290 ((db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
291 (DBUF_IS_METADATA(db) && \
292 ((db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
294 #define DBUF_IS_L2CACHEABLE(db) \
295 ((db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
296 (DBUF_IS_METADATA(db) && \
297 ((db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
302 * There should be a ## between the string literal and fmt, to make it
303 * clear that we're joining two strings together, but gcc does not
304 * support that preprocessor token.
306 #define dprintf_dbuf(dbuf, fmt, ...) do { \
307 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
309 uint64_t __db_obj = (dbuf)->db.db_object; \
310 if (__db_obj == DMU_META_DNODE_OBJECT) \
311 (void) strcpy(__db_buf, "mdn"); \
313 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
314 (u_longlong_t)__db_obj); \
315 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
316 "obj=%s lvl=%u blkid=%lld " fmt, \
317 __db_buf, (dbuf)->db_level, \
318 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
320 _NOTE(CONSTCOND) } while (0)
322 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
323 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
324 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
325 sprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
326 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
327 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
329 _NOTE(CONSTCOND) } while (0)
331 #define DBUF_VERIFY(db) dbuf_verify(db)
335 #define dprintf_dbuf(db, fmt, ...)
336 #define dprintf_dbuf_bp(db, bp, fmt, ...)
337 #define DBUF_VERIFY(db)
346 #endif /* _SYS_DBUF_H */