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 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
40 extern "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 typedef enum dbuf_states {
71 DB_UNCACHED,
72 DB_FILL,
73 DB_NOFILL,
74 DB_READ,
75 DB_CACHED,
76 DB_EVICTING
77 } dbuf_states_t;
78
79 struct dnode;
80 struct dmu_tx;
81
82 /*
83 * level = 0 means the user data
84 * level = 1 means the single indirect block
85 * etc.
86 */
87
88 struct dmu_buf_impl;
89
90 typedef enum override_states {
91 DR_NOT_OVERRIDDEN,
92 DR_IN_DMU_SYNC,
93 DR_OVERRIDDEN
94 } override_states_t;
95
96 typedef struct dbuf_dirty_record {
97 /* link on our parents dirty list */
98 list_node_t dr_dirty_node;
99
100 /* transaction group this data will sync in */
101 uint64_t dr_txg;
102
103 /* zio of outstanding write IO */
104 zio_t *dr_zio;
105
106 /* pointer back to our dbuf */
107 struct dmu_buf_impl *dr_dbuf;
108
109 /* pointer to next dirty record */
110 struct dbuf_dirty_record *dr_next;
111
112 /* pointer to parent dirty record */
113 struct dbuf_dirty_record *dr_parent;
114
115 union dirty_types {
116 struct dirty_indirect {
117
118 /* protect access to list */
119 kmutex_t dr_mtx;
120
121 /* Our list of dirty children */
122 list_t dr_children;
123 } di;
124 struct dirty_leaf {
125
126 /*
127 * dr_data is set when we dirty the buffer
128 * so that we can retain the pointer even if it
129 * gets COW'd in a subsequent transaction group.
130 */
131 arc_buf_t *dr_data;
132 blkptr_t dr_overridden_by;
133 override_states_t dr_override_state;
134 uint8_t dr_copies;
135 boolean_t dr_nopwrite;
136 } dl;
137 } dt;
138 } dbuf_dirty_record_t;
139
140 typedef struct dmu_buf_impl {
141 /*
142 * The following members are immutable, with the exception of
143 * db.db_data, which is protected by db_mtx.
144 */
145
146 /* the publicly visible structure */
147 dmu_buf_t db;
148
149 /* the objset we belong to */
150 struct objset *db_objset;
151
152 /*
153 * handle to safely access the dnode we belong to (NULL when evicted)
154 */
155 struct dnode_handle *db_dnode_handle;
156
157 /*
158 * our parent buffer; if the dnode points to us directly,
159 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
160 * only accessed by sync thread ???
161 * (NULL when evicted)
162 * May change from NULL to non-NULL under the protection of db_mtx
163 * (see dbuf_check_blkptr())
164 */
165 struct dmu_buf_impl *db_parent;
166
167 /*
168 * link for hash table of all dmu_buf_impl_t's
169 */
170 struct dmu_buf_impl *db_hash_next;
171
172 /* our block number */
173 uint64_t db_blkid;
174
175 /*
176 * Pointer to the blkptr_t which points to us. May be NULL if we
177 * don't have one yet. (NULL when evicted)
178 */
179 blkptr_t *db_blkptr;
180
181 /*
182 * Our indirection level. Data buffers have db_level==0.
183 * Indirect buffers which point to data buffers have
184 * db_level==1. etc. Buffers which contain dnodes have
185 * db_level==0, since the dnodes are stored in a file.
186 */
187 uint8_t db_level;
188
189 /* db_mtx protects the members below */
190 kmutex_t db_mtx;
191
192 /*
193 * Current state of the buffer
194 */
195 dbuf_states_t db_state;
196
197 /*
198 * Refcount accessed by dmu_buf_{hold,rele}.
199 * If nonzero, the buffer can't be destroyed.
200 * Protected by db_mtx.
201 */
202 refcount_t db_holds;
203
204 /* buffer holding our data */
205 arc_buf_t *db_buf;
206
207 kcondvar_t db_changed;
208 dbuf_dirty_record_t *db_data_pending;
209
210 /* pointer to most recent dirty record for this buffer */
211 dbuf_dirty_record_t *db_last_dirty;
212
213 /*
214 * Our link on the owner dnodes's dn_dbufs list.
215 * Protected by its dn_dbufs_mtx.
216 */
217 list_node_t db_link;
218
219 /* Data which is unique to data (leaf) blocks: */
220
221 /* stuff we store for the user (see dmu_buf_set_user) */
222 void *db_user_ptr;
223 void **db_user_data_ptr_ptr;
224 dmu_buf_evict_func_t *db_evict_func;
225
226 uint8_t db_immediate_evict;
227 uint8_t db_freed_in_flight;
228
229 uint8_t db_dirtycnt;
230 } dmu_buf_impl_t;
231
232 /* Note: the dbuf hash table is exposed only for the mdb module */
233 #define DBUF_MUTEXES 256
234 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
235 typedef struct dbuf_hash_table {
236 uint64_t hash_table_mask;
237 dmu_buf_impl_t **hash_table;
238 kmutex_t hash_mutexes[DBUF_MUTEXES];
239 } dbuf_hash_table_t;
240
241
242 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
243
244 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
245 void dbuf_create_bonus(struct dnode *dn);
246 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
247 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
248
249 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
250
251 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
252 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
253 void *tag);
254 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
255 void *tag, dmu_buf_impl_t **dbp);
256
257 void dbuf_prefetch(struct dnode *dn, uint64_t blkid);
258
259 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
260 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
261
262 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
263 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
264
265 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
266
267 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
268 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
269 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
270 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
271 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
272 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
273 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
274 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
275 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
276
277 void dbuf_clear(dmu_buf_impl_t *db);
278 void dbuf_evict(dmu_buf_impl_t *db);
279
280 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
281 void dbuf_unoverride(dbuf_dirty_record_t *dr);
282 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
283 void dbuf_release_bp(dmu_buf_impl_t *db);
284
285 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
286 struct dmu_tx *);
287
288 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
289
290 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
291 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
292 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
293 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
294 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
295 #define DB_GET_SPA(_spa_p, _db) { \
296 dnode_t *__dn; \
297 DB_DNODE_ENTER(_db); \
298 __dn = DB_DNODE(_db); \
299 *(_spa_p) = __dn->dn_objset->os_spa; \
300 DB_DNODE_EXIT(_db); \
301 }
302 #define DB_GET_OBJSET(_os_p, _db) { \
303 dnode_t *__dn; \
304 DB_DNODE_ENTER(_db); \
305 __dn = DB_DNODE(_db); \
306 *(_os_p) = __dn->dn_objset; \
307 DB_DNODE_EXIT(_db); \
308 }
309
310 void dbuf_init(void);
311 void dbuf_fini(void);
312
313 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
314
315 #define DBUF_GET_BUFC_TYPE(_db) \
316 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
317
318 #define DBUF_IS_CACHEABLE(_db) \
319 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
320 (dbuf_is_metadata(_db) && \
321 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
322
323 #define DBUF_IS_L2CACHEABLE(_db) \
324 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
325 (dbuf_is_metadata(_db) && \
326 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
327
328 #define DBUF_IS_L2COMPRESSIBLE(_db) \
329 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF)
330
331 #ifdef ZFS_DEBUG
332
333 /*
334 * There should be a ## between the string literal and fmt, to make it
335 * clear that we're joining two strings together, but gcc does not
336 * support that preprocessor token.
337 */
338 #define dprintf_dbuf(dbuf, fmt, ...) do { \
339 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
340 char __db_buf[32]; \
341 uint64_t __db_obj = (dbuf)->db.db_object; \
342 if (__db_obj == DMU_META_DNODE_OBJECT) \
343 (void) strcpy(__db_buf, "mdn"); \
344 else \
345 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
346 (u_longlong_t)__db_obj); \
347 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
348 "obj=%s lvl=%u blkid=%lld " fmt, \
349 __db_buf, (dbuf)->db_level, \
350 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
351 } \
352 _NOTE(CONSTCOND) } while (0)
353
354 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
355 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
356 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
357 sprintf_blkptr(__blkbuf, bp); \
358 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
359 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
360 } \
361 _NOTE(CONSTCOND) } while (0)
362
363 #define DBUF_VERIFY(db) dbuf_verify(db)
364
365 #else
366
367 #define dprintf_dbuf(db, fmt, ...)
368 #define dprintf_dbuf_bp(db, bp, fmt, ...)
369 #define DBUF_VERIFY(db)
370
371 #endif
372
373
374 #ifdef __cplusplus
375 }
376 #endif
377
378 #endif /* _SYS_DBUF_H */