xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/dbuf.h (revision fca4268092e9961ebb9b5e0098dcebc545023586)
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 
40 #ifdef	__cplusplus
41 extern "C" {
42 #endif
43 
44 #define	IN_DMU_SYNC 2
45 
46 /*
47  * define flags for dbuf_read
48  */
49 
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)
56 
57 /*
58  * The simplified state transition diagram for dbufs looks like:
59  *
60  *		+----> READ ----+
61  *		|		|
62  *		|		V
63  *  (alloc)-->UNCACHED	     CACHED-->EVICTING-->(free)
64  *		|		^	 ^
65  *		|		|	 |
66  *		+----> FILL ----+	 |
67  *		|			 |
68  *		|			 |
69  *		+--------> NOFILL -------+
70  *
71  * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
72  * to find all dbufs in a range of a dnode and must be less than any other
73  * dbuf_states_t (see comment on dn_dbufs in dnode.h).
74  */
75 typedef enum dbuf_states {
76 	DB_SEARCH = -1,
77 	DB_UNCACHED,
78 	DB_FILL,
79 	DB_NOFILL,
80 	DB_READ,
81 	DB_CACHED,
82 	DB_EVICTING
83 } dbuf_states_t;
84 
85 struct dnode;
86 struct dmu_tx;
87 
88 /*
89  * level = 0 means the user data
90  * level = 1 means the single indirect block
91  * etc.
92  */
93 
94 struct dmu_buf_impl;
95 
96 typedef enum override_states {
97 	DR_NOT_OVERRIDDEN,
98 	DR_IN_DMU_SYNC,
99 	DR_OVERRIDDEN
100 } override_states_t;
101 
102 typedef struct dbuf_dirty_record {
103 	/* link on our parents dirty list */
104 	list_node_t dr_dirty_node;
105 
106 	/* transaction group this data will sync in */
107 	uint64_t dr_txg;
108 
109 	/* zio of outstanding write IO */
110 	zio_t *dr_zio;
111 
112 	/* pointer back to our dbuf */
113 	struct dmu_buf_impl *dr_dbuf;
114 
115 	/* pointer to next dirty record */
116 	struct dbuf_dirty_record *dr_next;
117 
118 	/* pointer to parent dirty record */
119 	struct dbuf_dirty_record *dr_parent;
120 
121 	/* How much space was changed to dsl_pool_dirty_space() for this? */
122 	unsigned int dr_accounted;
123 
124 	union dirty_types {
125 		struct dirty_indirect {
126 
127 			/* protect access to list */
128 			kmutex_t dr_mtx;
129 
130 			/* Our list of dirty children */
131 			list_t dr_children;
132 		} di;
133 		struct dirty_leaf {
134 
135 			/*
136 			 * dr_data is set when we dirty the buffer
137 			 * so that we can retain the pointer even if it
138 			 * gets COW'd in a subsequent transaction group.
139 			 */
140 			arc_buf_t *dr_data;
141 			blkptr_t dr_overridden_by;
142 			override_states_t dr_override_state;
143 			uint8_t dr_copies;
144 			boolean_t dr_nopwrite;
145 		} dl;
146 	} dt;
147 } dbuf_dirty_record_t;
148 
149 typedef struct dmu_buf_impl {
150 	/*
151 	 * The following members are immutable, with the exception of
152 	 * db.db_data, which is protected by db_mtx.
153 	 */
154 
155 	/* the publicly visible structure */
156 	dmu_buf_t db;
157 
158 	/* the objset we belong to */
159 	struct objset *db_objset;
160 
161 	/*
162 	 * handle to safely access the dnode we belong to (NULL when evicted)
163 	 */
164 	struct dnode_handle *db_dnode_handle;
165 
166 	/*
167 	 * our parent buffer; if the dnode points to us directly,
168 	 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
169 	 * only accessed by sync thread ???
170 	 * (NULL when evicted)
171 	 * May change from NULL to non-NULL under the protection of db_mtx
172 	 * (see dbuf_check_blkptr())
173 	 */
174 	struct dmu_buf_impl *db_parent;
175 
176 	/*
177 	 * link for hash table of all dmu_buf_impl_t's
178 	 */
179 	struct dmu_buf_impl *db_hash_next;
180 
181 	/* our block number */
182 	uint64_t db_blkid;
183 
184 	/*
185 	 * Pointer to the blkptr_t which points to us. May be NULL if we
186 	 * don't have one yet. (NULL when evicted)
187 	 */
188 	blkptr_t *db_blkptr;
189 
190 	/*
191 	 * Our indirection level.  Data buffers have db_level==0.
192 	 * Indirect buffers which point to data buffers have
193 	 * db_level==1. etc.  Buffers which contain dnodes have
194 	 * db_level==0, since the dnodes are stored in a file.
195 	 */
196 	uint8_t db_level;
197 
198 	/* db_mtx protects the members below */
199 	kmutex_t db_mtx;
200 
201 	/*
202 	 * Current state of the buffer
203 	 */
204 	dbuf_states_t db_state;
205 
206 	/*
207 	 * Refcount accessed by dmu_buf_{hold,rele}.
208 	 * If nonzero, the buffer can't be destroyed.
209 	 * Protected by db_mtx.
210 	 */
211 	refcount_t db_holds;
212 
213 	/* buffer holding our data */
214 	arc_buf_t *db_buf;
215 
216 	kcondvar_t db_changed;
217 	dbuf_dirty_record_t *db_data_pending;
218 
219 	/* pointer to most recent dirty record for this buffer */
220 	dbuf_dirty_record_t *db_last_dirty;
221 
222 	/*
223 	 * Our link on the owner dnodes's dn_dbufs list.
224 	 * Protected by its dn_dbufs_mtx.
225 	 */
226 	avl_node_t db_link;
227 
228 	/* Data which is unique to data (leaf) blocks: */
229 
230 	/* User callback information. */
231 	dmu_buf_user_t *db_user;
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)])
242 typedef 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 uint64_t dbuf_whichblock(struct dnode *di, int64_t level, uint64_t offset);
249 
250 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
251 void dbuf_create_bonus(struct dnode *dn);
252 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
253 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
254 
255 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
256 
257 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
258 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
259     void *tag);
260 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
261     boolean_t fail_sparse, boolean_t fail_uncached,
262     void *tag, dmu_buf_impl_t **dbp);
263 
264 void dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
265     zio_priority_t prio, arc_flags_t aflags);
266 
267 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
268 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
269     uint64_t blkid, void *tag);
270 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
271 
272 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
273 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
274 
275 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
276     uint64_t blkid);
277 
278 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
279 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
280 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
281 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
282 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
283 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
284 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
285 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
286     bp_embedded_type_t etype, enum zio_compress comp,
287     int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
288 
289 void dbuf_clear(dmu_buf_impl_t *db);
290 void dbuf_evict(dmu_buf_impl_t *db);
291 
292 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
293 void dbuf_unoverride(dbuf_dirty_record_t *dr);
294 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
295 void dbuf_release_bp(dmu_buf_impl_t *db);
296 
297 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
298     struct dmu_tx *);
299 
300 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
301 
302 #define	DB_DNODE(_db)		((_db)->db_dnode_handle->dnh_dnode)
303 #define	DB_DNODE_LOCK(_db)	((_db)->db_dnode_handle->dnh_zrlock)
304 #define	DB_DNODE_ENTER(_db)	(zrl_add(&DB_DNODE_LOCK(_db)))
305 #define	DB_DNODE_EXIT(_db)	(zrl_remove(&DB_DNODE_LOCK(_db)))
306 #define	DB_DNODE_HELD(_db)	(!zrl_is_zero(&DB_DNODE_LOCK(_db)))
307 
308 void dbuf_init(void);
309 void dbuf_fini(void);
310 
311 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
312 
313 #define	DBUF_GET_BUFC_TYPE(_db)	\
314 	(dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
315 
316 #define	DBUF_IS_CACHEABLE(_db)						\
317 	((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
318 	(dbuf_is_metadata(_db) &&					\
319 	((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
320 
321 #define	DBUF_IS_L2CACHEABLE(_db)					\
322 	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
323 	(dbuf_is_metadata(_db) &&					\
324 	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
325 
326 #define	DBUF_IS_L2COMPRESSIBLE(_db)					\
327 	((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF ||		\
328 	(dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
329 
330 #ifdef ZFS_DEBUG
331 
332 /*
333  * There should be a ## between the string literal and fmt, to make it
334  * clear that we're joining two strings together, but gcc does not
335  * support that preprocessor token.
336  */
337 #define	dprintf_dbuf(dbuf, fmt, ...) do { \
338 	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
339 	char __db_buf[32]; \
340 	uint64_t __db_obj = (dbuf)->db.db_object; \
341 	if (__db_obj == DMU_META_DNODE_OBJECT) \
342 		(void) strcpy(__db_buf, "mdn"); \
343 	else \
344 		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
345 		    (u_longlong_t)__db_obj); \
346 	dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
347 	    "obj=%s lvl=%u blkid=%lld " fmt, \
348 	    __db_buf, (dbuf)->db_level, \
349 	    (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
350 	} \
351 _NOTE(CONSTCOND) } while (0)
352 
353 #define	dprintf_dbuf_bp(db, bp, fmt, ...) do {			\
354 	if (zfs_flags & ZFS_DEBUG_DPRINTF) {			\
355 	char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);	\
356 	snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp);		\
357 	dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);	\
358 	kmem_free(__blkbuf, BP_SPRINTF_LEN);			\
359 	}							\
360 _NOTE(CONSTCOND) } while (0)
361 
362 #define	DBUF_VERIFY(db)	dbuf_verify(db)
363 
364 #else
365 
366 #define	dprintf_dbuf(db, fmt, ...)
367 #define	dprintf_dbuf_bp(db, bp, fmt, ...)
368 #define	DBUF_VERIFY(db)
369 
370 #endif
371 
372 
373 #ifdef	__cplusplus
374 }
375 #endif
376 
377 #endif /* _SYS_DBUF_H */
378