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 2010 Sun Microsystems, Inc.  All rights reserved.
23 * Use is subject to license terms.
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
37#ifdef	__cplusplus
38extern "C" {
39#endif
40
41#define	DB_BONUS_BLKID (-1ULL)
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 */
69typedef 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
78struct dnode;
79struct dmu_tx;
80
81/*
82 * level = 0 means the user data
83 * level = 1 means the single indirect block
84 * etc.
85 */
86
87#define	LIST_LINK_INACTIVE(link) \
88	((link)->list_next == NULL && (link)->list_prev == NULL)
89
90struct dmu_buf_impl;
91
92typedef enum override_states {
93	DR_NOT_OVERRIDDEN,
94	DR_IN_DMU_SYNC,
95	DR_OVERRIDDEN
96} override_states_t;
97
98typedef struct dbuf_dirty_record {
99	/* link on our parents dirty list */
100	list_node_t dr_dirty_node;
101
102	/* transaction group this data will sync in */
103	uint64_t dr_txg;
104
105	/* zio of outstanding write IO */
106	zio_t *dr_zio;
107
108	/* pointer back to our dbuf */
109	struct dmu_buf_impl *dr_dbuf;
110
111	/* pointer to next dirty record */
112	struct dbuf_dirty_record *dr_next;
113
114	/* pointer to parent dirty record */
115	struct dbuf_dirty_record *dr_parent;
116
117	union dirty_types {
118		struct dirty_indirect {
119
120			/* protect access to list */
121			kmutex_t dr_mtx;
122
123			/* Our list of dirty children */
124			list_t dr_children;
125		} di;
126		struct dirty_leaf {
127
128			/*
129			 * dr_data is set when we dirty the buffer
130			 * so that we can retain the pointer even if it
131			 * gets COW'd in a subsequent transaction group.
132			 */
133			arc_buf_t *dr_data;
134			blkptr_t dr_overridden_by;
135			override_states_t dr_override_state;
136			uint8_t dr_copies;
137		} dl;
138	} dt;
139} dbuf_dirty_record_t;
140
141typedef struct dmu_buf_impl {
142	/*
143	 * The following members are immutable, with the exception of
144	 * db.db_data, which is protected by db_mtx.
145	 */
146
147	/* the publicly visible structure */
148	dmu_buf_t db;
149
150	/* the objset we belong to */
151	struct objset *db_objset;
152
153	/*
154	 * the dnode we belong to (NULL when evicted)
155	 */
156	struct dnode *db_dnode;
157
158	/*
159	 * our parent buffer; if the dnode points to us directly,
160	 * db_parent == db_dnode->dn_dbuf
161	 * only accessed by sync thread ???
162	 * (NULL when evicted)
163	 */
164	struct dmu_buf_impl *db_parent;
165
166	/*
167	 * link for hash table of all dmu_buf_impl_t's
168	 */
169	struct dmu_buf_impl *db_hash_next;
170
171	/* our block number */
172	uint64_t db_blkid;
173
174	/*
175	 * Pointer to the blkptr_t which points to us. May be NULL if we
176	 * don't have one yet. (NULL when evicted)
177	 */
178	blkptr_t *db_blkptr;
179
180	/*
181	 * Our indirection level.  Data buffers have db_level==0.
182	 * Indirect buffers which point to data buffers have
183	 * db_level==1. etc.  Buffers which contain dnodes have
184	 * db_level==0, since the dnodes are stored in a file.
185	 */
186	uint8_t db_level;
187
188	/* db_mtx protects the members below */
189	kmutex_t db_mtx;
190
191	/*
192	 * Current state of the buffer
193	 */
194	dbuf_states_t db_state;
195
196	/*
197	 * Refcount accessed by dmu_buf_{hold,rele}.
198	 * If nonzero, the buffer can't be destroyed.
199	 * Protected by db_mtx.
200	 */
201	refcount_t db_holds;
202
203	/* buffer holding our data */
204	arc_buf_t *db_buf;
205
206	kcondvar_t db_changed;
207	dbuf_dirty_record_t *db_data_pending;
208
209	/* pointer to most recent dirty record for this buffer */
210	dbuf_dirty_record_t *db_last_dirty;
211
212	/*
213	 * Our link on the owner dnodes's dn_dbufs list.
214	 * Protected by its dn_dbufs_mtx.
215	 */
216	list_node_t db_link;
217
218	/* Data which is unique to data (leaf) blocks: */
219
220	/* stuff we store for the user (see dmu_buf_set_user) */
221	void *db_user_ptr;
222	void **db_user_data_ptr_ptr;
223	dmu_buf_evict_func_t *db_evict_func;
224
225	uint8_t db_immediate_evict;
226	uint8_t db_freed_in_flight;
227
228	uint8_t db_dirtycnt;
229} dmu_buf_impl_t;
230
231/* Note: the dbuf hash table is exposed only for the mdb module */
232#define	DBUF_MUTEXES 256
233#define	DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
234typedef struct dbuf_hash_table {
235	uint64_t hash_table_mask;
236	dmu_buf_impl_t **hash_table;
237	kmutex_t hash_mutexes[DBUF_MUTEXES];
238} dbuf_hash_table_t;
239
240
241uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
242
243dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
244void dbuf_create_bonus(struct dnode *dn);
245
246dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
247dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
248    void *tag);
249int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
250    void *tag, dmu_buf_impl_t **dbp);
251
252void dbuf_prefetch(struct dnode *dn, uint64_t blkid);
253
254void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
255uint64_t dbuf_refcount(dmu_buf_impl_t *db);
256
257void dbuf_rele(dmu_buf_impl_t *db, void *tag);
258void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
259
260dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
261
262int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
263void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
264void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
265void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
266void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
267void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
268void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
269dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
270arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
271
272void dbuf_clear(dmu_buf_impl_t *db);
273void dbuf_evict(dmu_buf_impl_t *db);
274
275void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
276void dbuf_unoverride(dbuf_dirty_record_t *dr);
277void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
278
279void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
280    struct dmu_tx *);
281
282void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
283
284void dbuf_init(void);
285void dbuf_fini(void);
286
287#define	DBUF_IS_METADATA(db)	\
288	((db)->db_level > 0 || dmu_ot[(db)->db_dnode->dn_type].ot_metadata)
289
290#define	DBUF_GET_BUFC_TYPE(db)	\
291	(DBUF_IS_METADATA(db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
292
293#define	DBUF_IS_CACHEABLE(db)						\
294	((db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
295	(DBUF_IS_METADATA(db) &&					\
296	((db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
297
298#define	DBUF_IS_L2CACHEABLE(db)						\
299	((db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
300	(DBUF_IS_METADATA(db) &&					\
301	((db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
302
303#ifdef ZFS_DEBUG
304
305/*
306 * There should be a ## between the string literal and fmt, to make it
307 * clear that we're joining two strings together, but gcc does not
308 * support that preprocessor token.
309 */
310#define	dprintf_dbuf(dbuf, fmt, ...) do { \
311	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
312	char __db_buf[32]; \
313	uint64_t __db_obj = (dbuf)->db.db_object; \
314	if (__db_obj == DMU_META_DNODE_OBJECT) \
315		(void) strcpy(__db_buf, "mdn"); \
316	else \
317		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
318		    (u_longlong_t)__db_obj); \
319	dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
320	    "obj=%s lvl=%u blkid=%lld " fmt, \
321	    __db_buf, (dbuf)->db_level, \
322	    (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
323	} \
324_NOTE(CONSTCOND) } while (0)
325
326#define	dprintf_dbuf_bp(db, bp, fmt, ...) do {			\
327	if (zfs_flags & ZFS_DEBUG_DPRINTF) {			\
328	char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);	\
329	sprintf_blkptr(__blkbuf, bp);				\
330	dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);	\
331	kmem_free(__blkbuf, BP_SPRINTF_LEN);			\
332	} 							\
333_NOTE(CONSTCOND) } while (0)
334
335#define	DBUF_VERIFY(db)	dbuf_verify(db)
336
337#else
338
339#define	dprintf_dbuf(db, fmt, ...)
340#define	dprintf_dbuf_bp(db, bp, fmt, ...)
341#define	DBUF_VERIFY(db)
342
343#endif
344
345
346#ifdef	__cplusplus
347}
348#endif
349
350#endif /* _SYS_DBUF_H */
351