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 * Copyright 2014 Nexenta Systems, Inc. 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 #define DBUF_EVICT_ALL -1
58
59 /*
60 * The simplified state transition diagram for dbufs looks like:
61 *
62 * +----> READ ----+
63 * | |
64 * | V
65 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
66 * | ^ ^
67 * | | |
68 * +----> FILL ----+ |
69 * | |
70 * | |
71 * +--------> NOFILL -------+
72 *
73 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
74 * to find all dbufs in a range of a dnode and must be less than any other
75 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
76 */
77 typedef enum dbuf_states {
78 DB_SEARCH = -1,
79 DB_UNCACHED,
80 DB_FILL,
81 DB_NOFILL,
82 DB_READ,
83 DB_CACHED,
84 DB_EVICTING
85 } dbuf_states_t;
86
87 struct dnode;
88 struct dmu_tx;
89
90 /*
91 * level = 0 means the user data
92 * level = 1 means the single indirect block
93 * etc.
94 */
95
96 struct dmu_buf_impl;
97
98 typedef enum override_states {
99 DR_NOT_OVERRIDDEN,
100 DR_IN_DMU_SYNC,
101 DR_OVERRIDDEN
102 } override_states_t;
103
104 typedef struct dbuf_dirty_record {
105 /* link on our parents dirty list */
106 list_node_t dr_dirty_node;
107
108 /* transaction group this data will sync in */
109 uint64_t dr_txg;
110
111 /* zio of outstanding write IO */
112 zio_t *dr_zio;
113
114 /* pointer back to our dbuf */
115 struct dmu_buf_impl *dr_dbuf;
116
117 /* pointer to next dirty record */
118 struct dbuf_dirty_record *dr_next;
119
120 /* pointer to parent dirty record */
121 struct dbuf_dirty_record *dr_parent;
122
123 /* How much space was changed to dsl_pool_dirty_space() for this? */
124 unsigned int dr_accounted;
125
126 /* use special class of dirty entry */
127 boolean_t dr_usesc;
128
129 union dirty_types {
130 struct dirty_indirect {
131
132 /* protect access to list */
133 kmutex_t dr_mtx;
134
135 /* Our list of dirty children */
136 list_t dr_children;
137 } di;
138 struct dirty_leaf {
139
140 /*
141 * dr_data is set when we dirty the buffer
142 * so that we can retain the pointer even if it
143 * gets COW'd in a subsequent transaction group.
144 */
145 arc_buf_t *dr_data;
146 blkptr_t dr_overridden_by;
147 override_states_t dr_override_state;
148 uint8_t dr_copies;
149 boolean_t dr_nopwrite;
150 } dl;
151 } dt;
152 } dbuf_dirty_record_t;
153
154 typedef struct dmu_buf_impl {
155 /*
156 * The following members are immutable, with the exception of
157 * db.db_data, which is protected by db_mtx.
158 */
159
160 /* the publicly visible structure */
161 dmu_buf_t db;
162
163 /* the objset we belong to */
164 struct objset *db_objset;
165
166 /*
167 * handle to safely access the dnode we belong to (NULL when evicted)
168 */
169 struct dnode_handle *db_dnode_handle;
170
171 /*
172 * our parent buffer; if the dnode points to us directly,
173 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
174 * only accessed by sync thread ???
175 * (NULL when evicted)
176 * May change from NULL to non-NULL under the protection of db_mtx
177 * (see dbuf_check_blkptr())
178 */
179 struct dmu_buf_impl *db_parent;
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 /* stuff we store for the user (see dmu_buf_set_user) */
231 void *db_user_ptr;
232 void **db_user_data_ptr_ptr;
233 dmu_buf_evict_func_t *db_evict_func;
234
235 uint8_t db_immediate_evict;
236 uint8_t db_freed_in_flight;
237
238 uint8_t db_dirtycnt;
239 } dmu_buf_impl_t;
240
241 /* Note: the dbuf hash table is exposed only for the mdb module */
242 #define DBUF_MUTEXES 256
243 #define DBUF_LOCK_PAD 64
244 typedef struct {
245 kmutex_t mtx;
246 #ifdef _KERNEL
247 unsigned char pad[(DBUF_LOCK_PAD - sizeof (kmutex_t))];
248 #endif
249 } dbuf_mutex_t;
250 #define DBUF_HASH_MUTEX(h, idx) \
251 (&((h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)].mtx))
252 typedef struct dbuf_hash_table {
253 uint64_t hash_table_mask;
254 dmu_buf_impl_t **hash_table;
255 dbuf_mutex_t hash_mutexes[DBUF_MUTEXES];
256 } dbuf_hash_table_t;
257
258
259 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
260
261 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
262 void dbuf_create_bonus(struct dnode *dn);
263 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
264 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
265
266 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
267
268 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
269 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
270 void *tag);
271 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
272 void *tag, dmu_buf_impl_t **dbp);
273
274 void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
275
276 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
277 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
278
279 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
280 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
281
282 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
283
284 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
285 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
286 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
287 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
288 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
289 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
290 dbuf_dirty_record_t *dbuf_dirty_sc(dmu_buf_impl_t *db, dmu_tx_t *tx,
291 boolean_t usesc);
292 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
293 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
294 bp_embedded_type_t etype, enum zio_compress comp,
295 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
296
297 void dbuf_clear(dmu_buf_impl_t *db);
298 void dbuf_evict(dmu_buf_impl_t *db);
299
300 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
301 void dbuf_unoverride(dbuf_dirty_record_t *dr);
302 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
303 void dbuf_release_bp(dmu_buf_impl_t *db);
304
305 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
306 struct dmu_tx *);
307
308 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
309
310 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
311 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
312 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
313 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
314 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
315
316 void dbuf_init(void);
317 void dbuf_fini(void);
318
319 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
320 boolean_t dbuf_ddt_is_l2cacheable(dmu_buf_impl_t *db);
321 boolean_t dbuf_meta_is_l2cacheable(dmu_buf_impl_t *db);
322
323 #define DBUF_GET_BUFC_TYPE(_db) \
324 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
325
326 #define DBUF_IS_CACHEABLE(_db) \
327 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
328 (dbuf_is_metadata(_db) && \
329 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
330
331
332 /*
333 * Checks whether we need to cache dbuf in l2arc.
334 * Metadata is l2cacheable if it is not placed on special device
335 * or it is placed on special device in "dual" mode. We need to check
336 * for ddt in ZFS_CACHE_ALL and ZFS_CACHE_METADATA because it is in MOS.
337 * ZFS_CACHE_DATA mode actually means to cache both data and cacheable
338 * metadata.
339 */
340 #define DBUF_IS_L2CACHEABLE(_db) \
341 (((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL && \
342 (dbuf_ddt_is_l2cacheable(_db) == B_TRUE)) || \
343 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA && \
344 (dbuf_is_metadata(_db)) && \
345 (dbuf_ddt_is_l2cacheable(_db) == B_TRUE)) || \
346 ((dbuf_meta_is_l2cacheable(_db) == B_TRUE) && \
347 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_DATA)))
348
349 #define DBUF_IS_L2COMPRESSIBLE(_db) \
350 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF || \
351 (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
352
353 #ifdef ZFS_DEBUG
354
355 /*
356 * There should be a ## between the string literal and fmt, to make it
357 * clear that we're joining two strings together, but gcc does not
358 * support that preprocessor token.
359 */
360 #define dprintf_dbuf(dbuf, fmt, ...) do { \
361 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
362 char __db_buf[32]; \
363 uint64_t __db_obj = (dbuf)->db.db_object; \
364 if (__db_obj == DMU_META_DNODE_OBJECT) \
365 (void) strcpy(__db_buf, "mdn"); \
366 else \
367 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
368 (u_longlong_t)__db_obj); \
369 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
370 "obj=%s lvl=%u blkid=%lld " fmt, \
371 __db_buf, (dbuf)->db_level, \
372 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
373 } \
374 _NOTE(CONSTCOND) } while (0)
375
376 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
377 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
378 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
379 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
380 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
381 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
382 } \
383 _NOTE(CONSTCOND) } while (0)
384
385 #define DBUF_VERIFY(db) dbuf_verify(db)
386
387 #else
388
389 #define dprintf_dbuf(db, fmt, ...)
390 #define dprintf_dbuf_bp(db, bp, fmt, ...)
391 #define DBUF_VERIFY(db)
392
393 #endif
394
395
396 #ifdef __cplusplus
397 }
398 #endif
399
400 #endif /* _SYS_DBUF_H */