dbuf.h revision 307266
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, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 */
27
28#ifndef	_SYS_DBUF_H
29#define	_SYS_DBUF_H
30
31#include <sys/dmu.h>
32#include <sys/spa.h>
33#include <sys/txg.h>
34#include <sys/zio.h>
35#include <sys/arc.h>
36#include <sys/zfs_context.h>
37#include <sys/refcount.h>
38#include <sys/zrlock.h>
39#include <sys/multilist.h>
40
41#ifdef	__cplusplus
42extern "C" {
43#endif
44
45#define	IN_DMU_SYNC 2
46
47/*
48 * define flags for dbuf_read
49 */
50
51#define	DB_RF_MUST_SUCCEED	(1 << 0)
52#define	DB_RF_CANFAIL		(1 << 1)
53#define	DB_RF_HAVESTRUCT	(1 << 2)
54#define	DB_RF_NOPREFETCH	(1 << 3)
55#define	DB_RF_NEVERWAIT		(1 << 4)
56#define	DB_RF_CACHED		(1 << 5)
57
58/*
59 * The simplified state transition diagram for dbufs looks like:
60 *
61 *		+----> READ ----+
62 *		|		|
63 *		|		V
64 *  (alloc)-->UNCACHED	     CACHED-->EVICTING-->(free)
65 *		|		^	 ^
66 *		|		|	 |
67 *		+----> FILL ----+	 |
68 *		|			 |
69 *		|			 |
70 *		+--------> NOFILL -------+
71 *
72 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
73 * to find all dbufs in a range of a dnode and must be less than any other
74 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
75 */
76typedef enum dbuf_states {
77	DB_SEARCH = -1,
78	DB_UNCACHED,
79	DB_FILL,
80	DB_NOFILL,
81	DB_READ,
82	DB_CACHED,
83	DB_EVICTING
84} dbuf_states_t;
85
86struct dnode;
87struct dmu_tx;
88
89/*
90 * level = 0 means the user data
91 * level = 1 means the single indirect block
92 * etc.
93 */
94
95struct dmu_buf_impl;
96
97typedef enum override_states {
98	DR_NOT_OVERRIDDEN,
99	DR_IN_DMU_SYNC,
100	DR_OVERRIDDEN
101} override_states_t;
102
103typedef struct dbuf_dirty_record {
104	/* link on our parents dirty list */
105	list_node_t dr_dirty_node;
106
107	/* transaction group this data will sync in */
108	uint64_t dr_txg;
109
110	/* zio of outstanding write IO */
111	zio_t *dr_zio;
112
113	/* pointer back to our dbuf */
114	struct dmu_buf_impl *dr_dbuf;
115
116	/* pointer to next dirty record */
117	struct dbuf_dirty_record *dr_next;
118
119	/* pointer to parent dirty record */
120	struct dbuf_dirty_record *dr_parent;
121
122	/* How much space was changed to dsl_pool_dirty_space() for this? */
123	unsigned int dr_accounted;
124
125	/* A copy of the bp that points to us */
126	blkptr_t dr_bp_copy;
127
128	union dirty_types {
129		struct dirty_indirect {
130
131			/* protect access to list */
132			kmutex_t dr_mtx;
133
134			/* Our list of dirty children */
135			list_t dr_children;
136		} di;
137		struct dirty_leaf {
138
139			/*
140			 * dr_data is set when we dirty the buffer
141			 * so that we can retain the pointer even if it
142			 * gets COW'd in a subsequent transaction group.
143			 */
144			arc_buf_t *dr_data;
145			blkptr_t dr_overridden_by;
146			override_states_t dr_override_state;
147			uint8_t dr_copies;
148			boolean_t dr_nopwrite;
149		} dl;
150	} dt;
151} dbuf_dirty_record_t;
152
153typedef struct dmu_buf_impl {
154	/*
155	 * The following members are immutable, with the exception of
156	 * db.db_data, which is protected by db_mtx.
157	 */
158
159	/* the publicly visible structure */
160	dmu_buf_t db;
161
162	/* the objset we belong to */
163	struct objset *db_objset;
164
165	/*
166	 * handle to safely access the dnode we belong to (NULL when evicted)
167	 */
168	struct dnode_handle *db_dnode_handle;
169
170	/*
171	 * our parent buffer; if the dnode points to us directly,
172	 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
173	 * only accessed by sync thread ???
174	 * (NULL when evicted)
175	 * May change from NULL to non-NULL under the protection of db_mtx
176	 * (see dbuf_check_blkptr())
177	 */
178	struct dmu_buf_impl *db_parent;
179
180	/*
181	 * link for hash table of all dmu_buf_impl_t's
182	 */
183	struct dmu_buf_impl *db_hash_next;
184
185	/* our block number */
186	uint64_t db_blkid;
187
188	/*
189	 * Pointer to the blkptr_t which points to us. May be NULL if we
190	 * don't have one yet. (NULL when evicted)
191	 */
192	blkptr_t *db_blkptr;
193
194	/*
195	 * Our indirection level.  Data buffers have db_level==0.
196	 * Indirect buffers which point to data buffers have
197	 * db_level==1. etc.  Buffers which contain dnodes have
198	 * db_level==0, since the dnodes are stored in a file.
199	 */
200	uint8_t db_level;
201
202	/* db_mtx protects the members below */
203	kmutex_t db_mtx;
204
205	/*
206	 * Current state of the buffer
207	 */
208	dbuf_states_t db_state;
209
210	/*
211	 * Refcount accessed by dmu_buf_{hold,rele}.
212	 * If nonzero, the buffer can't be destroyed.
213	 * Protected by db_mtx.
214	 */
215	refcount_t db_holds;
216
217	/* buffer holding our data */
218	arc_buf_t *db_buf;
219
220	kcondvar_t db_changed;
221	dbuf_dirty_record_t *db_data_pending;
222
223	/* pointer to most recent dirty record for this buffer */
224	dbuf_dirty_record_t *db_last_dirty;
225
226	/*
227	 * Our link on the owner dnodes's dn_dbufs list.
228	 * Protected by its dn_dbufs_mtx.
229	 */
230	avl_node_t db_link;
231
232	/*
233	 * Link in dbuf_cache.
234	 */
235	multilist_node_t db_cache_link;
236
237	/* Data which is unique to data (leaf) blocks: */
238
239	/* User callback information. */
240	dmu_buf_user_t *db_user;
241
242	/*
243	 * Evict user data as soon as the dirty and reference
244	 * counts are equal.
245	 */
246	uint8_t db_user_immediate_evict;
247
248	/*
249	 * This block was freed while a read or write was
250	 * active.
251	 */
252	uint8_t db_freed_in_flight;
253
254	/*
255	 * dnode_evict_dbufs() or dnode_evict_bonus() tried to
256	 * evict this dbuf, but couldn't due to outstanding
257	 * references.  Evict once the refcount drops to 0.
258	 */
259	uint8_t db_pending_evict;
260
261	uint8_t db_dirtycnt;
262} dmu_buf_impl_t;
263
264/* Note: the dbuf hash table is exposed only for the mdb module */
265#define	DBUF_MUTEXES 256
266#define	DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
267typedef struct dbuf_hash_table {
268	uint64_t hash_table_mask;
269	dmu_buf_impl_t **hash_table;
270	kmutex_t hash_mutexes[DBUF_MUTEXES];
271} dbuf_hash_table_t;
272
273uint64_t dbuf_whichblock(struct dnode *di, int64_t level, uint64_t offset);
274
275dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
276void dbuf_create_bonus(struct dnode *dn);
277int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
278void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
279
280void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
281
282dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
283dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
284    void *tag);
285int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
286    boolean_t fail_sparse, boolean_t fail_uncached,
287    void *tag, dmu_buf_impl_t **dbp);
288
289void dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
290    zio_priority_t prio, arc_flags_t aflags);
291
292void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
293boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
294    uint64_t blkid, void *tag);
295uint64_t dbuf_refcount(dmu_buf_impl_t *db);
296
297void dbuf_rele(dmu_buf_impl_t *db, void *tag);
298void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
299
300dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
301    uint64_t blkid);
302
303int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
304void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
305void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
306void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
307void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
308dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
309arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
310void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
311    bp_embedded_type_t etype, enum zio_compress comp,
312    int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
313
314void dbuf_destroy(dmu_buf_impl_t *db);
315
316void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
317void dbuf_unoverride(dbuf_dirty_record_t *dr);
318void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
319void dbuf_release_bp(dmu_buf_impl_t *db);
320
321void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
322    struct dmu_tx *);
323
324void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
325
326#define	DB_DNODE(_db)		((_db)->db_dnode_handle->dnh_dnode)
327#define	DB_DNODE_LOCK(_db)	((_db)->db_dnode_handle->dnh_zrlock)
328#define	DB_DNODE_ENTER(_db)	(zrl_add(&DB_DNODE_LOCK(_db)))
329#define	DB_DNODE_EXIT(_db)	(zrl_remove(&DB_DNODE_LOCK(_db)))
330#define	DB_DNODE_HELD(_db)	(!zrl_is_zero(&DB_DNODE_LOCK(_db)))
331
332void dbuf_init(void);
333void dbuf_fini(void);
334
335boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
336
337#define	DBUF_GET_BUFC_TYPE(_db)	\
338	(dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
339
340#define	DBUF_IS_CACHEABLE(_db)						\
341	((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
342	(dbuf_is_metadata(_db) &&					\
343	((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
344
345#define	DBUF_IS_L2CACHEABLE(_db)					\
346	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
347	(dbuf_is_metadata(_db) &&					\
348	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
349
350#ifdef ZFS_DEBUG
351
352/*
353 * There should be a ## between the string literal and fmt, to make it
354 * clear that we're joining two strings together, but gcc does not
355 * support that preprocessor token.
356 */
357#define	dprintf_dbuf(dbuf, fmt, ...) do { \
358	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
359	char __db_buf[32]; \
360	uint64_t __db_obj = (dbuf)->db.db_object; \
361	if (__db_obj == DMU_META_DNODE_OBJECT) \
362		(void) strcpy(__db_buf, "mdn"); \
363	else \
364		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
365		    (u_longlong_t)__db_obj); \
366	dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
367	    "obj=%s lvl=%u blkid=%lld " fmt, \
368	    __db_buf, (dbuf)->db_level, \
369	    (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
370	} \
371_NOTE(CONSTCOND) } while (0)
372
373#define	dprintf_dbuf_bp(db, bp, fmt, ...) do {			\
374	if (zfs_flags & ZFS_DEBUG_DPRINTF) {			\
375	char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);	\
376	snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp);		\
377	dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);	\
378	kmem_free(__blkbuf, BP_SPRINTF_LEN);			\
379	}							\
380_NOTE(CONSTCOND) } while (0)
381
382#define	DBUF_VERIFY(db)	dbuf_verify(db)
383
384#else
385
386#define	dprintf_dbuf(db, fmt, ...)
387#define	dprintf_dbuf_bp(db, bp, fmt, ...)
388#define	DBUF_VERIFY(db)
389
390#endif
391
392
393#ifdef	__cplusplus
394}
395#endif
396
397#endif /* _SYS_DBUF_H */
398