]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
8951cb8d | 23 | * Copyright (c) 2012, 2014 by Delphix. All rights reserved. |
3a17a7a9 | 24 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
27 | #ifndef _SYS_DBUF_H | |
28 | #define _SYS_DBUF_H | |
29 | ||
34dc7c2f BB |
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> | |
572e2857 | 37 | #include <sys/zrlock.h> |
34dc7c2f BB |
38 | |
39 | #ifdef __cplusplus | |
40 | extern "C" { | |
41 | #endif | |
42 | ||
34dc7c2f BB |
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 | /* | |
b128c09f | 57 | * The simplified state transition diagram for dbufs looks like: |
34dc7c2f BB |
58 | * |
59 | * +----> READ ----+ | |
60 | * | | | |
61 | * | V | |
62 | * (alloc)-->UNCACHED CACHED-->EVICTING-->(free) | |
b128c09f BB |
63 | * | ^ ^ |
64 | * | | | | |
65 | * +----> FILL ----+ | | |
66 | * | | | |
67 | * | | | |
68 | * +--------> NOFILL -------+ | |
9925c28c AR |
69 | * |
70 | * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range | |
71 | * to find all dbufs in a range of a dnode and must be less than any other | |
72 | * dbuf_states_t (see comment on dn_dbufs in dnode.h). | |
34dc7c2f BB |
73 | */ |
74 | typedef enum dbuf_states { | |
9925c28c | 75 | DB_SEARCH = -1, |
34dc7c2f BB |
76 | DB_UNCACHED, |
77 | DB_FILL, | |
b128c09f | 78 | DB_NOFILL, |
34dc7c2f BB |
79 | DB_READ, |
80 | DB_CACHED, | |
81 | DB_EVICTING | |
82 | } dbuf_states_t; | |
83 | ||
34dc7c2f BB |
84 | struct dnode; |
85 | struct dmu_tx; | |
86 | ||
87 | /* | |
88 | * level = 0 means the user data | |
89 | * level = 1 means the single indirect block | |
90 | * etc. | |
91 | */ | |
92 | ||
34dc7c2f BB |
93 | struct dmu_buf_impl; |
94 | ||
95 | typedef enum override_states { | |
96 | DR_NOT_OVERRIDDEN, | |
97 | DR_IN_DMU_SYNC, | |
98 | DR_OVERRIDDEN | |
99 | } override_states_t; | |
100 | ||
101 | typedef struct dbuf_dirty_record { | |
102 | /* link on our parents dirty list */ | |
103 | list_node_t dr_dirty_node; | |
104 | ||
105 | /* transaction group this data will sync in */ | |
106 | uint64_t dr_txg; | |
107 | ||
108 | /* zio of outstanding write IO */ | |
109 | zio_t *dr_zio; | |
110 | ||
111 | /* pointer back to our dbuf */ | |
112 | struct dmu_buf_impl *dr_dbuf; | |
113 | ||
114 | /* pointer to next dirty record */ | |
115 | struct dbuf_dirty_record *dr_next; | |
116 | ||
117 | /* pointer to parent dirty record */ | |
118 | struct dbuf_dirty_record *dr_parent; | |
119 | ||
e8b96c60 MA |
120 | /* How much space was changed to dsl_pool_dirty_space() for this? */ |
121 | unsigned int dr_accounted; | |
122 | ||
34dc7c2f BB |
123 | union dirty_types { |
124 | struct dirty_indirect { | |
125 | ||
126 | /* protect access to list */ | |
127 | kmutex_t dr_mtx; | |
128 | ||
129 | /* Our list of dirty children */ | |
130 | list_t dr_children; | |
131 | } di; | |
132 | struct dirty_leaf { | |
133 | ||
134 | /* | |
135 | * dr_data is set when we dirty the buffer | |
136 | * so that we can retain the pointer even if it | |
137 | * gets COW'd in a subsequent transaction group. | |
138 | */ | |
139 | arc_buf_t *dr_data; | |
140 | blkptr_t dr_overridden_by; | |
141 | override_states_t dr_override_state; | |
428870ff | 142 | uint8_t dr_copies; |
03c6040b | 143 | boolean_t dr_nopwrite; |
34dc7c2f BB |
144 | } dl; |
145 | } dt; | |
146 | } dbuf_dirty_record_t; | |
147 | ||
148 | typedef struct dmu_buf_impl { | |
149 | /* | |
150 | * The following members are immutable, with the exception of | |
151 | * db.db_data, which is protected by db_mtx. | |
152 | */ | |
153 | ||
154 | /* the publicly visible structure */ | |
155 | dmu_buf_t db; | |
156 | ||
157 | /* the objset we belong to */ | |
428870ff | 158 | struct objset *db_objset; |
34dc7c2f BB |
159 | |
160 | /* | |
572e2857 | 161 | * handle to safely access the dnode we belong to (NULL when evicted) |
34dc7c2f | 162 | */ |
572e2857 | 163 | struct dnode_handle *db_dnode_handle; |
34dc7c2f BB |
164 | |
165 | /* | |
166 | * our parent buffer; if the dnode points to us directly, | |
572e2857 | 167 | * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf |
34dc7c2f BB |
168 | * only accessed by sync thread ??? |
169 | * (NULL when evicted) | |
572e2857 BB |
170 | * May change from NULL to non-NULL under the protection of db_mtx |
171 | * (see dbuf_check_blkptr()) | |
34dc7c2f BB |
172 | */ |
173 | struct dmu_buf_impl *db_parent; | |
174 | ||
175 | /* | |
176 | * link for hash table of all dmu_buf_impl_t's | |
177 | */ | |
178 | struct dmu_buf_impl *db_hash_next; | |
179 | ||
180 | /* our block number */ | |
181 | uint64_t db_blkid; | |
182 | ||
183 | /* | |
184 | * Pointer to the blkptr_t which points to us. May be NULL if we | |
185 | * don't have one yet. (NULL when evicted) | |
186 | */ | |
187 | blkptr_t *db_blkptr; | |
188 | ||
189 | /* | |
190 | * Our indirection level. Data buffers have db_level==0. | |
191 | * Indirect buffers which point to data buffers have | |
192 | * db_level==1. etc. Buffers which contain dnodes have | |
193 | * db_level==0, since the dnodes are stored in a file. | |
194 | */ | |
195 | uint8_t db_level; | |
196 | ||
197 | /* db_mtx protects the members below */ | |
198 | kmutex_t db_mtx; | |
199 | ||
200 | /* | |
201 | * Current state of the buffer | |
202 | */ | |
203 | dbuf_states_t db_state; | |
204 | ||
205 | /* | |
206 | * Refcount accessed by dmu_buf_{hold,rele}. | |
207 | * If nonzero, the buffer can't be destroyed. | |
208 | * Protected by db_mtx. | |
209 | */ | |
210 | refcount_t db_holds; | |
211 | ||
212 | /* buffer holding our data */ | |
213 | arc_buf_t *db_buf; | |
214 | ||
215 | kcondvar_t db_changed; | |
216 | dbuf_dirty_record_t *db_data_pending; | |
217 | ||
218 | /* pointer to most recent dirty record for this buffer */ | |
219 | dbuf_dirty_record_t *db_last_dirty; | |
220 | ||
221 | /* | |
222 | * Our link on the owner dnodes's dn_dbufs list. | |
223 | * Protected by its dn_dbufs_mtx. | |
224 | */ | |
8951cb8d | 225 | avl_node_t db_link; |
34dc7c2f BB |
226 | |
227 | /* Data which is unique to data (leaf) blocks: */ | |
228 | ||
229 | /* stuff we store for the user (see dmu_buf_set_user) */ | |
230 | void *db_user_ptr; | |
231 | void **db_user_data_ptr_ptr; | |
232 | dmu_buf_evict_func_t *db_evict_func; | |
233 | ||
234 | uint8_t db_immediate_evict; | |
235 | uint8_t db_freed_in_flight; | |
236 | ||
237 | uint8_t db_dirtycnt; | |
238 | } dmu_buf_impl_t; | |
239 | ||
240 | /* Note: the dbuf hash table is exposed only for the mdb module */ | |
b31d8ea7 | 241 | #define DBUF_MUTEXES 8192 |
34dc7c2f BB |
242 | #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)]) |
243 | typedef struct dbuf_hash_table { | |
244 | uint64_t hash_table_mask; | |
245 | dmu_buf_impl_t **hash_table; | |
246 | kmutex_t hash_mutexes[DBUF_MUTEXES]; | |
247 | } dbuf_hash_table_t; | |
248 | ||
249 | ||
250 | uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset); | |
251 | ||
34dc7c2f | 252 | void dbuf_create_bonus(struct dnode *dn); |
428870ff | 253 | int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx); |
428870ff BB |
254 | |
255 | void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx); | |
34dc7c2f BB |
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, int create, | |
261 | void *tag, dmu_buf_impl_t **dbp); | |
262 | ||
e8b96c60 | 263 | void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio); |
34dc7c2f BB |
264 | |
265 | void dbuf_add_ref(dmu_buf_impl_t *db, void *tag); | |
266 | uint64_t dbuf_refcount(dmu_buf_impl_t *db); | |
267 | ||
268 | void dbuf_rele(dmu_buf_impl_t *db, void *tag); | |
428870ff | 269 | void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag); |
34dc7c2f BB |
270 | |
271 | dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid); | |
272 | ||
273 | int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags); | |
b128c09f | 274 | void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx); |
34dc7c2f BB |
275 | void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx); |
276 | void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx); | |
9babb374 | 277 | void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx); |
34dc7c2f | 278 | dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx); |
428870ff | 279 | arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db); |
9b67f605 MA |
280 | void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data, |
281 | bp_embedded_type_t etype, enum zio_compress comp, | |
282 | int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx); | |
34dc7c2f BB |
283 | |
284 | void dbuf_clear(dmu_buf_impl_t *db); | |
285 | void dbuf_evict(dmu_buf_impl_t *db); | |
286 | ||
34dc7c2f BB |
287 | void dbuf_unoverride(dbuf_dirty_record_t *dr); |
288 | void dbuf_sync_list(list_t *list, dmu_tx_t *tx); | |
428870ff | 289 | void dbuf_release_bp(dmu_buf_impl_t *db); |
34dc7c2f | 290 | |
b128c09f | 291 | void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end, |
34dc7c2f BB |
292 | struct dmu_tx *); |
293 | ||
294 | void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx); | |
295 | ||
e0b0ca98 BB |
296 | void dbuf_stats_init(dbuf_hash_table_t *hash); |
297 | void dbuf_stats_destroy(void); | |
298 | ||
572e2857 BB |
299 | #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode) |
300 | #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock) | |
301 | #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db))) | |
302 | #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db))) | |
303 | #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db))) | |
572e2857 | 304 | |
34dc7c2f BB |
305 | void dbuf_init(void); |
306 | void dbuf_fini(void); | |
307 | ||
572e2857 BB |
308 | boolean_t dbuf_is_metadata(dmu_buf_impl_t *db); |
309 | ||
572e2857 | 310 | #define DBUF_GET_BUFC_TYPE(_db) \ |
13fe0198 | 311 | (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA) |
b128c09f | 312 | |
572e2857 BB |
313 | #define DBUF_IS_CACHEABLE(_db) \ |
314 | ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \ | |
13fe0198 | 315 | (dbuf_is_metadata(_db) && \ |
572e2857 | 316 | ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA))) |
b128c09f | 317 | |
572e2857 BB |
318 | #define DBUF_IS_L2CACHEABLE(_db) \ |
319 | ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \ | |
13fe0198 | 320 | (dbuf_is_metadata(_db) && \ |
572e2857 | 321 | ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA))) |
34dc7c2f | 322 | |
3a17a7a9 | 323 | #define DBUF_IS_L2COMPRESSIBLE(_db) \ |
4e59f475 SK |
324 | ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF || \ |
325 | (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE)) | |
3a17a7a9 | 326 | |
34dc7c2f BB |
327 | #ifdef ZFS_DEBUG |
328 | ||
329 | /* | |
330 | * There should be a ## between the string literal and fmt, to make it | |
331 | * clear that we're joining two strings together, but gcc does not | |
332 | * support that preprocessor token. | |
333 | */ | |
334 | #define dprintf_dbuf(dbuf, fmt, ...) do { \ | |
335 | if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ | |
336 | char __db_buf[32]; \ | |
337 | uint64_t __db_obj = (dbuf)->db.db_object; \ | |
338 | if (__db_obj == DMU_META_DNODE_OBJECT) \ | |
339 | (void) strcpy(__db_buf, "mdn"); \ | |
340 | else \ | |
341 | (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \ | |
342 | (u_longlong_t)__db_obj); \ | |
343 | dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \ | |
344 | "obj=%s lvl=%u blkid=%lld " fmt, \ | |
345 | __db_buf, (dbuf)->db_level, \ | |
346 | (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \ | |
347 | } \ | |
348 | _NOTE(CONSTCOND) } while (0) | |
349 | ||
79c76d5b BB |
350 | #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \ |
351 | if (zfs_flags & ZFS_DEBUG_DPRINTF) { \ | |
352 | char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \ | |
b0bc7a84 | 353 | snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \ |
79c76d5b BB |
354 | dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \ |
355 | kmem_free(__blkbuf, BP_SPRINTF_LEN); \ | |
356 | } \ | |
34dc7c2f BB |
357 | _NOTE(CONSTCOND) } while (0) |
358 | ||
359 | #define DBUF_VERIFY(db) dbuf_verify(db) | |
360 | ||
361 | #else | |
362 | ||
363 | #define dprintf_dbuf(db, fmt, ...) | |
364 | #define dprintf_dbuf_bp(db, bp, fmt, ...) | |
365 | #define DBUF_VERIFY(db) | |
366 | ||
367 | #endif | |
368 | ||
369 | ||
370 | #ifdef __cplusplus | |
371 | } | |
372 | #endif | |
373 | ||
374 | #endif /* _SYS_DBUF_H */ |