xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/dbuf.h (revision 5bb0bdfe588c5df0f63ff8ac292cd608a5f4492a)
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, 2018 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
42 extern "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  */
76 typedef 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 
86 typedef enum dbuf_cached_state {
87 	DB_NO_CACHE = -1,
88 	DB_DBUF_CACHE,
89 	DB_DBUF_METADATA_CACHE,
90 	DB_CACHE_MAX
91 } dbuf_cached_state_t;
92 
93 struct dnode;
94 struct dmu_tx;
95 
96 /*
97  * level = 0 means the user data
98  * level = 1 means the single indirect block
99  * etc.
100  */
101 
102 struct dmu_buf_impl;
103 
104 typedef enum override_states {
105 	DR_NOT_OVERRIDDEN,
106 	DR_IN_DMU_SYNC,
107 	DR_OVERRIDDEN
108 } override_states_t;
109 
110 typedef struct dbuf_dirty_record {
111 	/* link on our parents dirty list */
112 	list_node_t dr_dirty_node;
113 
114 	/* transaction group this data will sync in */
115 	uint64_t dr_txg;
116 
117 	/* zio of outstanding write IO */
118 	zio_t *dr_zio;
119 
120 	/* pointer back to our dbuf */
121 	struct dmu_buf_impl *dr_dbuf;
122 
123 	/* pointer to next dirty record */
124 	struct dbuf_dirty_record *dr_next;
125 
126 	/* pointer to parent dirty record */
127 	struct dbuf_dirty_record *dr_parent;
128 
129 	/* How much space was changed to dsl_pool_dirty_space() for this? */
130 	unsigned int dr_accounted;
131 
132 	/* A copy of the bp that points to us */
133 	blkptr_t dr_bp_copy;
134 
135 	union dirty_types {
136 		struct dirty_indirect {
137 
138 			/* protect access to list */
139 			kmutex_t dr_mtx;
140 
141 			/* Our list of dirty children */
142 			list_t dr_children;
143 		} di;
144 		struct dirty_leaf {
145 
146 			/*
147 			 * dr_data is set when we dirty the buffer
148 			 * so that we can retain the pointer even if it
149 			 * gets COW'd in a subsequent transaction group.
150 			 */
151 			arc_buf_t *dr_data;
152 			blkptr_t dr_overridden_by;
153 			override_states_t dr_override_state;
154 			uint8_t dr_copies;
155 			boolean_t dr_nopwrite;
156 		} dl;
157 	} dt;
158 } dbuf_dirty_record_t;
159 
160 typedef struct dmu_buf_impl {
161 	/*
162 	 * The following members are immutable, with the exception of
163 	 * db.db_data, which is protected by db_mtx.
164 	 */
165 
166 	/* the publicly visible structure */
167 	dmu_buf_t db;
168 
169 	/* the objset we belong to */
170 	struct objset *db_objset;
171 
172 	/*
173 	 * handle to safely access the dnode we belong to (NULL when evicted)
174 	 */
175 	struct dnode_handle *db_dnode_handle;
176 
177 	/*
178 	 * our parent buffer; if the dnode points to us directly,
179 	 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
180 	 * only accessed by sync thread ???
181 	 * (NULL when evicted)
182 	 * May change from NULL to non-NULL under the protection of db_mtx
183 	 * (see dbuf_check_blkptr())
184 	 */
185 	struct dmu_buf_impl *db_parent;
186 
187 	/*
188 	 * link for hash table of all dmu_buf_impl_t's
189 	 */
190 	struct dmu_buf_impl *db_hash_next;
191 
192 	/* our block number */
193 	uint64_t db_blkid;
194 
195 	/*
196 	 * Pointer to the blkptr_t which points to us. May be NULL if we
197 	 * don't have one yet. (NULL when evicted)
198 	 */
199 	blkptr_t *db_blkptr;
200 
201 	/*
202 	 * Our indirection level.  Data buffers have db_level==0.
203 	 * Indirect buffers which point to data buffers have
204 	 * db_level==1. etc.  Buffers which contain dnodes have
205 	 * db_level==0, since the dnodes are stored in a file.
206 	 */
207 	uint8_t db_level;
208 
209 	/* db_mtx protects the members below */
210 	kmutex_t db_mtx;
211 
212 	/*
213 	 * Current state of the buffer
214 	 */
215 	dbuf_states_t db_state;
216 
217 	/*
218 	 * Refcount accessed by dmu_buf_{hold,rele}.
219 	 * If nonzero, the buffer can't be destroyed.
220 	 * Protected by db_mtx.
221 	 */
222 	refcount_t db_holds;
223 
224 	/* buffer holding our data */
225 	arc_buf_t *db_buf;
226 
227 	kcondvar_t db_changed;
228 	dbuf_dirty_record_t *db_data_pending;
229 
230 	/* pointer to most recent dirty record for this buffer */
231 	dbuf_dirty_record_t *db_last_dirty;
232 
233 	/*
234 	 * Our link on the owner dnodes's dn_dbufs list.
235 	 * Protected by its dn_dbufs_mtx.
236 	 */
237 	avl_node_t db_link;
238 
239 	/* Link in dbuf_cache or dbuf_metadata_cache */
240 	multilist_node_t db_cache_link;
241 
242 	/* Tells us which dbuf cache this dbuf is in, if any */
243 	dbuf_cached_state_t db_caching_status;
244 
245 	/* Data which is unique to data (leaf) blocks: */
246 
247 	/* User callback information. */
248 	dmu_buf_user_t *db_user;
249 
250 	/*
251 	 * Evict user data as soon as the dirty and reference
252 	 * counts are equal.
253 	 */
254 	uint8_t db_user_immediate_evict;
255 
256 	/*
257 	 * This block was freed while a read or write was
258 	 * active.
259 	 */
260 	uint8_t db_freed_in_flight;
261 
262 	/*
263 	 * dnode_evict_dbufs() or dnode_evict_bonus() tried to
264 	 * evict this dbuf, but couldn't due to outstanding
265 	 * references.  Evict once the refcount drops to 0.
266 	 */
267 	uint8_t db_pending_evict;
268 
269 	uint8_t db_dirtycnt;
270 } dmu_buf_impl_t;
271 
272 /* Note: the dbuf hash table is exposed only for the mdb module */
273 #define	DBUF_MUTEXES 256
274 #define	DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
275 typedef struct dbuf_hash_table {
276 	uint64_t hash_table_mask;
277 	dmu_buf_impl_t **hash_table;
278 	kmutex_t hash_mutexes[DBUF_MUTEXES];
279 } dbuf_hash_table_t;
280 
281 uint64_t dbuf_whichblock(struct dnode *di, int64_t level, uint64_t offset);
282 
283 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
284 void dbuf_create_bonus(struct dnode *dn);
285 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
286 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
287 
288 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
289 
290 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
291 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
292     void *tag);
293 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid,
294     boolean_t fail_sparse, boolean_t fail_uncached,
295     void *tag, dmu_buf_impl_t **dbp);
296 
297 void dbuf_prefetch(struct dnode *dn, int64_t level, uint64_t blkid,
298     zio_priority_t prio, arc_flags_t aflags);
299 
300 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
301 boolean_t dbuf_try_add_ref(dmu_buf_t *db, objset_t *os, uint64_t obj,
302     uint64_t blkid, void *tag);
303 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
304 
305 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
306 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag, boolean_t evicting);
307 
308 dmu_buf_impl_t *dbuf_find(struct objset *os, uint64_t object, uint8_t level,
309     uint64_t blkid);
310 
311 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
312 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
313 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
314 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
315 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
316 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
317 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
318 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
319     bp_embedded_type_t etype, enum zio_compress comp,
320     int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
321 
322 void dbuf_destroy(dmu_buf_impl_t *db);
323 
324 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
325 void dbuf_unoverride(dbuf_dirty_record_t *dr);
326 void dbuf_sync_list(list_t *list, int level, dmu_tx_t *tx);
327 void dbuf_release_bp(dmu_buf_impl_t *db);
328 
329 boolean_t dbuf_can_remap(const dmu_buf_impl_t *buf);
330 
331 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
332     struct dmu_tx *);
333 
334 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
335 
336 #define	DB_DNODE(_db)		((_db)->db_dnode_handle->dnh_dnode)
337 #define	DB_DNODE_LOCK(_db)	((_db)->db_dnode_handle->dnh_zrlock)
338 #define	DB_DNODE_ENTER(_db)	(zrl_add(&DB_DNODE_LOCK(_db)))
339 #define	DB_DNODE_EXIT(_db)	(zrl_remove(&DB_DNODE_LOCK(_db)))
340 #define	DB_DNODE_HELD(_db)	(!zrl_is_zero(&DB_DNODE_LOCK(_db)))
341 
342 void dbuf_init(void);
343 void dbuf_fini(void);
344 
345 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
346 
347 #define	DBUF_GET_BUFC_TYPE(_db)	\
348 	(dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
349 
350 #define	DBUF_IS_CACHEABLE(_db)						\
351 	((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL ||		\
352 	(dbuf_is_metadata(_db) &&					\
353 	((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
354 
355 #define	DBUF_IS_L2CACHEABLE(_db)					\
356 	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
357 	(dbuf_is_metadata(_db) &&					\
358 	((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
359 
360 #define	DNODE_LEVEL_IS_L2CACHEABLE(_dn, _level)				\
361 	((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_ALL ||	\
362 	(((_level) > 0 ||						\
363 	DMU_OT_IS_METADATA((_dn)->dn_handle->dnh_dnode->dn_type)) &&	\
364 	((_dn)->dn_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
365 
366 #ifdef ZFS_DEBUG
367 
368 /*
369  * There should be a ## between the string literal and fmt, to make it
370  * clear that we're joining two strings together, but gcc does not
371  * support that preprocessor token.
372  */
373 #define	dprintf_dbuf(dbuf, fmt, ...) do { \
374 	if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
375 	char __db_buf[32]; \
376 	uint64_t __db_obj = (dbuf)->db.db_object; \
377 	if (__db_obj == DMU_META_DNODE_OBJECT) \
378 		(void) strcpy(__db_buf, "mdn"); \
379 	else \
380 		(void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
381 		    (u_longlong_t)__db_obj); \
382 	dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
383 	    "obj=%s lvl=%u blkid=%lld " fmt, \
384 	    __db_buf, (dbuf)->db_level, \
385 	    (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
386 	} \
387 _NOTE(CONSTCOND) } while (0)
388 
389 #define	dprintf_dbuf_bp(db, bp, fmt, ...) do {			\
390 	if (zfs_flags & ZFS_DEBUG_DPRINTF) {			\
391 	char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);	\
392 	snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp);		\
393 	dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf);	\
394 	kmem_free(__blkbuf, BP_SPRINTF_LEN);			\
395 	}							\
396 _NOTE(CONSTCOND) } while (0)
397 
398 #define	DBUF_VERIFY(db)	dbuf_verify(db)
399 
400 #else
401 
402 #define	dprintf_dbuf(db, fmt, ...)
403 #define	dprintf_dbuf_bp(db, bp, fmt, ...)
404 #define	DBUF_VERIFY(db)
405 
406 #endif
407 
408 
409 #ifdef	__cplusplus
410 }
411 #endif
412 
413 #endif /* _SYS_DBUF_H */
414