dbuf.h revision 277585
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, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 */
26
27#ifndef	_SYS_DBUF_H
28#define	_SYS_DBUF_H
29
30#include <sys/dmu.h>
31#include <sys/spa.h>
32#include <sys/txg.h>
33#include <sys/zio.h>
34#include <sys/arc.h>
35#include <sys/zfs_context.h>
36#include <sys/refcount.h>
37#include <sys/zrlock.h>
38
39#ifdef	__cplusplus
40extern "C" {
41#endif
42
43#define	IN_DMU_SYNC 2
44
45/*
46 * define flags for dbuf_read
47 */
48
49#define	DB_RF_MUST_SUCCEED	(1 << 0)
50#define	DB_RF_CANFAIL		(1 << 1)
51#define	DB_RF_HAVESTRUCT	(1 << 2)
52#define	DB_RF_NOPREFETCH	(1 << 3)
53#define	DB_RF_NEVERWAIT		(1 << 4)
54#define	DB_RF_CACHED		(1 << 5)
55
56/*
57 * The simplified state transition diagram for dbufs looks like:
58 *
59 *		+----> READ ----+
60 *		|		|
61 *		|		V
62 *  (alloc)-->UNCACHED	     CACHED-->EVICTING-->(free)
63 *		|		^	 ^
64 *		|		|	 |
65 *		+----> FILL ----+	 |
66 *		|			 |
67 *		|			 |
68 *		+--------> NOFILL -------+
69 *
70 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
71 * to find all dbufs in a range of a dnode and must be less than any other
72 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
73 */
74typedef enum dbuf_states {
75	DB_SEARCH = -1,
76	DB_UNCACHED,
77	DB_FILL,
78	DB_NOFILL,
79	DB_READ,
80	DB_CACHED,
81	DB_EVICTING
82} dbuf_states_t;
83
84struct dnode;
85struct dmu_tx;
86
87/*
88 * level = 0 means the user data
89 * level = 1 means the single indirect block
90 * etc.
91 */
92
93struct dmu_buf_impl;
94
95typedef enum override_states {
96	DR_NOT_OVERRIDDEN,
97	DR_IN_DMU_SYNC,
98	DR_OVERRIDDEN
99} override_states_t;
100
101typedef struct dbuf_dirty_record {
102	/* link on our parents dirty list */
103	list_node_t dr_dirty_node;
104
105	/* transaction group this data will sync in */
106	uint64_t dr_txg;
107
108	/* zio of outstanding write IO */
109	zio_t *dr_zio;
110
111	/* pointer back to our dbuf */
112	struct dmu_buf_impl *dr_dbuf;
113
114	/* pointer to next dirty record */
115	struct dbuf_dirty_record *dr_next;
116
117	/* pointer to parent dirty record */
118	struct dbuf_dirty_record *dr_parent;
119
120	/* How much space was changed to dsl_pool_dirty_space() for this? */
121	unsigned int dr_accounted;
122
123	union dirty_types {
124		struct dirty_indirect {
125
126			/* protect access to list */
127			kmutex_t dr_mtx;
128
129			/* Our list of dirty children */
130			list_t dr_children;
131		} di;
132		struct dirty_leaf {
133
134			/*
135			 * dr_data is set when we dirty the buffer
136			 * so that we can retain the pointer even if it
137			 * gets COW'd in a subsequent transaction group.
138			 */
139			arc_buf_t *dr_data;
140			blkptr_t dr_overridden_by;
141			override_states_t dr_override_state;
142			uint8_t dr_copies;
143			boolean_t dr_nopwrite;
144		} dl;
145	} dt;
146} dbuf_dirty_record_t;
147
148typedef struct dmu_buf_impl {
149	/*
150	 * The following members are immutable, with the exception of
151	 * db.db_data, which is protected by db_mtx.
152	 */
153
154	/* the publicly visible structure */
155	dmu_buf_t db;
156
157	/* the objset we belong to */
158	struct objset *db_objset;
159
160	/*
161	 * handle to safely access the dnode we belong to (NULL when evicted)
162	 */
163	struct dnode_handle *db_dnode_handle;
164
165	/*
166	 * our parent buffer; if the dnode points to us directly,
167	 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
168	 * only accessed by sync thread ???
169	 * (NULL when evicted)
170	 * May change from NULL to non-NULL under the protection of db_mtx
171	 * (see dbuf_check_blkptr())
172	 */
173	struct dmu_buf_impl *db_parent;
174
175	/*
176	 * link for hash table of all dmu_buf_impl_t's
177	 */
178	struct dmu_buf_impl *db_hash_next;
179
180	/* our block number */
181	uint64_t db_blkid;
182
183	/*
184	 * Pointer to the blkptr_t which points to us. May be NULL if we
185	 * don't have one yet. (NULL when evicted)
186	 */
187	blkptr_t *db_blkptr;
188
189	/*
190	 * Our indirection level.  Data buffers have db_level==0.
191	 * Indirect buffers which point to data buffers have
192	 * db_level==1. etc.  Buffers which contain dnodes have
193	 * db_level==0, since the dnodes are stored in a file.
194	 */
195	uint8_t db_level;
196
197	/* db_mtx protects the members below */
198	kmutex_t db_mtx;
199
200	/*
201	 * Current state of the buffer
202	 */
203	dbuf_states_t db_state;
204
205	/*
206	 * Refcount accessed by dmu_buf_{hold,rele}.
207	 * If nonzero, the buffer can't be destroyed.
208	 * Protected by db_mtx.
209	 */
210	refcount_t db_holds;
211
212	/* buffer holding our data */
213	arc_buf_t *db_buf;
214
215	kcondvar_t db_changed;
216	dbuf_dirty_record_t *db_data_pending;
217
218	/* pointer to most recent dirty record for this buffer */
219	dbuf_dirty_record_t *db_last_dirty;
220
221	/*
222	 * Our link on the owner dnodes's dn_dbufs list.
223	 * Protected by its dn_dbufs_mtx.
224	 */
225	avl_node_t db_link;
226
227	/* Data which is unique to data (leaf) blocks: */
228
229	/* stuff we store for the user (see dmu_buf_set_user) */
230	void *db_user_ptr;
231	dmu_buf_evict_func_t *db_evict_func;
232
233	uint8_t db_immediate_evict;
234	uint8_t db_freed_in_flight;
235
236	uint8_t db_dirtycnt;
237} dmu_buf_impl_t;
238
239/* Note: the dbuf hash table is exposed only for the mdb module */
240#define	DBUF_MUTEXES 256
241#define	DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
242typedef struct dbuf_hash_table {
243	uint64_t hash_table_mask;
244	dmu_buf_impl_t **hash_table;
245	kmutex_t hash_mutexes[DBUF_MUTEXES];
246} dbuf_hash_table_t;
247
248
249uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
250
251dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
252void dbuf_create_bonus(struct dnode *dn);
253int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
254void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
255
256void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
257
258dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
259dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
260    void *tag);
261int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
262    void *tag, dmu_buf_impl_t **dbp);
263
264void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
265
266void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
267uint64_t dbuf_refcount(dmu_buf_impl_t *db);
268
269void dbuf_rele(dmu_buf_impl_t *db, void *tag);
270void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
271
272dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
273
274int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
275void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
276void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
277void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
278void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
279dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
280arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
281void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
282    bp_embedded_type_t etype, enum zio_compress comp,
283    int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
284
285void dbuf_clear(dmu_buf_impl_t *db);
286void dbuf_evict(dmu_buf_impl_t *db);
287
288void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
289void dbuf_unoverride(dbuf_dirty_record_t *dr);
290void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
291void dbuf_release_bp(dmu_buf_impl_t *db);
292
293void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
294    struct dmu_tx *);
295
296void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
297
298#define	DB_DNODE(_db)		((_db)->db_dnode_handle->dnh_dnode)
299#define	DB_DNODE_LOCK(_db)	((_db)->db_dnode_handle->dnh_zrlock)
300#define	DB_DNODE_ENTER(_db)	(zrl_add(&DB_DNODE_LOCK(_db)))
301#define	DB_DNODE_EXIT(_db)	(zrl_remove(&DB_DNODE_LOCK(_db)))
302#define	DB_DNODE_HELD(_db)	(!zrl_is_zero(&DB_DNODE_LOCK(_db)))
303
304void dbuf_init(void);
305void dbuf_fini(void);
306
307boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
308
309#define	DBUF_GET_BUFC_TYPE(_db)	\
310	(dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
311
312#define	DBUF_IS_CACHEABLE(_db)						\
313	((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
314	(dbuf_is_metadata(_db) &&					\
315	((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
316
317#define	DBUF_IS_L2CACHEABLE(_db)					\
318	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
319	(dbuf_is_metadata(_db) &&					\
320	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
321
322#define	DBUF_IS_L2COMPRESSIBLE(_db)					\
323	((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF ||		\
324	(dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
325
326#ifdef ZFS_DEBUG
327
328/*
329 * There should be a ## between the string literal and fmt, to make it
330 * clear that we're joining two strings together, but gcc does not
331 * support that preprocessor token.
332 */
333#define	dprintf_dbuf(dbuf, fmt, ...) do { \
334	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
335	char __db_buf[32]; \
336	uint64_t __db_obj = (dbuf)->db.db_object; \
337	if (__db_obj == DMU_META_DNODE_OBJECT) \
338		(void) strcpy(__db_buf, "mdn"); \
339	else \
340		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
341		    (u_longlong_t)__db_obj); \
342	dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
343	    "obj=%s lvl=%u blkid=%lld " fmt, \
344	    __db_buf, (dbuf)->db_level, \
345	    (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
346	} \
347_NOTE(CONSTCOND) } while (0)
348
349#define	dprintf_dbuf_bp(db, bp, fmt, ...) do {			\
350	if (zfs_flags & ZFS_DEBUG_DPRINTF) {			\
351	char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);	\
352	snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp);		\
353	dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);	\
354	kmem_free(__blkbuf, BP_SPRINTF_LEN);			\
355	}							\
356_NOTE(CONSTCOND) } while (0)
357
358#define	DBUF_VERIFY(db)	dbuf_verify(db)
359
360#else
361
362#define	dprintf_dbuf(db, fmt, ...)
363#define	dprintf_dbuf_bp(db, bp, fmt, ...)
364#define	DBUF_VERIFY(db)
365
366#endif
367
368
369#ifdef	__cplusplus
370}
371#endif
372
373#endif /* _SYS_DBUF_H */
374