Print this page
Possibility to physically reserve space without writing leaf blocks
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/sys/dbuf.h
+++ new/usr/src/uts/common/fs/zfs/sys/dbuf.h
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 25 */
26 26
27 27 #ifndef _SYS_DBUF_H
28 28 #define _SYS_DBUF_H
29 29
30 30 #include <sys/dmu.h>
31 31 #include <sys/spa.h>
32 32 #include <sys/txg.h>
33 33 #include <sys/zio.h>
34 34 #include <sys/arc.h>
35 35 #include <sys/zfs_context.h>
36 36 #include <sys/refcount.h>
37 37 #include <sys/zrlock.h>
38 38
39 39 #ifdef __cplusplus
40 40 extern "C" {
41 41 #endif
42 42
43 43 #define IN_DMU_SYNC 2
44 44
45 45 /*
46 46 * define flags for dbuf_read
47 47 */
48 48
49 49 #define DB_RF_MUST_SUCCEED (1 << 0)
50 50 #define DB_RF_CANFAIL (1 << 1)
51 51 #define DB_RF_HAVESTRUCT (1 << 2)
52 52 #define DB_RF_NOPREFETCH (1 << 3)
53 53 #define DB_RF_NEVERWAIT (1 << 4)
54 54 #define DB_RF_CACHED (1 << 5)
55 55
56 56 /*
57 57 * The simplified state transition diagram for dbufs looks like:
58 58 *
59 59 * +----> READ ----+
60 60 * | |
61 61 * | V
62 62 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
63 63 * | ^ ^
64 64 * | | |
65 65 * +----> FILL ----+ |
66 66 * | |
67 67 * | |
68 68 * +--------> NOFILL -------+
69 69 *
70 70 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
71 71 * to find all dbufs in a range of a dnode and must be less than any other
72 72 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
73 73 */
74 74 typedef enum dbuf_states {
75 75 DB_SEARCH = -1,
76 76 DB_UNCACHED,
77 77 DB_FILL,
78 78 DB_NOFILL,
79 79 DB_READ,
80 80 DB_CACHED,
81 81 DB_EVICTING
82 82 } dbuf_states_t;
83 83
84 84 struct dnode;
85 85 struct dmu_tx;
86 86
87 87 /*
88 88 * level = 0 means the user data
89 89 * level = 1 means the single indirect block
90 90 * etc.
91 91 */
92 92
93 93 struct dmu_buf_impl;
94 94
95 95 typedef enum override_states {
96 96 DR_NOT_OVERRIDDEN,
97 97 DR_IN_DMU_SYNC,
98 98 DR_OVERRIDDEN
99 99 } override_states_t;
100 100
101 101 typedef struct dbuf_dirty_record {
102 102 /* link on our parents dirty list */
103 103 list_node_t dr_dirty_node;
104 104
105 105 /* transaction group this data will sync in */
106 106 uint64_t dr_txg;
107 107
108 108 /* zio of outstanding write IO */
109 109 zio_t *dr_zio;
110 110
111 111 /* pointer back to our dbuf */
112 112 struct dmu_buf_impl *dr_dbuf;
113 113
114 114 /* pointer to next dirty record */
115 115 struct dbuf_dirty_record *dr_next;
116 116
117 117 /* pointer to parent dirty record */
118 118 struct dbuf_dirty_record *dr_parent;
119 119
120 120 /* How much space was changed to dsl_pool_dirty_space() for this? */
121 121 unsigned int dr_accounted;
122 122
123 123 union dirty_types {
124 124 struct dirty_indirect {
125 125
126 126 /* protect access to list */
127 127 kmutex_t dr_mtx;
128 128
129 129 /* Our list of dirty children */
130 130 list_t dr_children;
131 131 } di;
132 132 struct dirty_leaf {
133 133
134 134 /*
135 135 * dr_data is set when we dirty the buffer
|
↓ open down ↓ |
135 lines elided |
↑ open up ↑ |
136 136 * so that we can retain the pointer even if it
137 137 * gets COW'd in a subsequent transaction group.
138 138 */
139 139 arc_buf_t *dr_data;
140 140 blkptr_t dr_overridden_by;
141 141 override_states_t dr_override_state;
142 142 uint8_t dr_copies;
143 143 boolean_t dr_nopwrite;
144 144 } dl;
145 145 } dt;
146 +
147 + boolean_t dr_zero_write;
146 148 } dbuf_dirty_record_t;
147 149
148 150 typedef struct dmu_buf_impl {
149 151 /*
150 152 * The following members are immutable, with the exception of
151 153 * db.db_data, which is protected by db_mtx.
152 154 */
153 155
154 156 /* the publicly visible structure */
155 157 dmu_buf_t db;
156 158
157 159 /* the objset we belong to */
158 160 struct objset *db_objset;
159 161
160 162 /*
161 163 * handle to safely access the dnode we belong to (NULL when evicted)
162 164 */
163 165 struct dnode_handle *db_dnode_handle;
164 166
165 167 /*
166 168 * our parent buffer; if the dnode points to us directly,
167 169 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
168 170 * only accessed by sync thread ???
169 171 * (NULL when evicted)
170 172 * May change from NULL to non-NULL under the protection of db_mtx
171 173 * (see dbuf_check_blkptr())
172 174 */
173 175 struct dmu_buf_impl *db_parent;
174 176
175 177 /*
176 178 * link for hash table of all dmu_buf_impl_t's
177 179 */
178 180 struct dmu_buf_impl *db_hash_next;
179 181
180 182 /* our block number */
181 183 uint64_t db_blkid;
182 184
183 185 /*
184 186 * Pointer to the blkptr_t which points to us. May be NULL if we
185 187 * don't have one yet. (NULL when evicted)
186 188 */
187 189 blkptr_t *db_blkptr;
188 190
189 191 /*
190 192 * Our indirection level. Data buffers have db_level==0.
191 193 * Indirect buffers which point to data buffers have
192 194 * db_level==1. etc. Buffers which contain dnodes have
193 195 * db_level==0, since the dnodes are stored in a file.
194 196 */
195 197 uint8_t db_level;
196 198
197 199 /* db_mtx protects the members below */
198 200 kmutex_t db_mtx;
199 201
200 202 /*
201 203 * Current state of the buffer
202 204 */
203 205 dbuf_states_t db_state;
204 206
205 207 /*
206 208 * Refcount accessed by dmu_buf_{hold,rele}.
207 209 * If nonzero, the buffer can't be destroyed.
208 210 * Protected by db_mtx.
209 211 */
210 212 refcount_t db_holds;
211 213
212 214 /* buffer holding our data */
213 215 arc_buf_t *db_buf;
214 216
215 217 kcondvar_t db_changed;
216 218 dbuf_dirty_record_t *db_data_pending;
217 219
218 220 /* pointer to most recent dirty record for this buffer */
219 221 dbuf_dirty_record_t *db_last_dirty;
220 222
221 223 /*
222 224 * Our link on the owner dnodes's dn_dbufs list.
223 225 * Protected by its dn_dbufs_mtx.
224 226 */
225 227 avl_node_t db_link;
226 228
227 229 /* Data which is unique to data (leaf) blocks: */
228 230
229 231 /* stuff we store for the user (see dmu_buf_set_user) */
230 232 void *db_user_ptr;
231 233 void **db_user_data_ptr_ptr;
232 234 dmu_buf_evict_func_t *db_evict_func;
233 235
234 236 uint8_t db_immediate_evict;
235 237 uint8_t db_freed_in_flight;
236 238
237 239 uint8_t db_dirtycnt;
238 240 } dmu_buf_impl_t;
239 241
240 242 /* Note: the dbuf hash table is exposed only for the mdb module */
241 243 #define DBUF_MUTEXES 256
242 244 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
243 245 typedef struct dbuf_hash_table {
244 246 uint64_t hash_table_mask;
245 247 dmu_buf_impl_t **hash_table;
246 248 kmutex_t hash_mutexes[DBUF_MUTEXES];
247 249 } dbuf_hash_table_t;
248 250
249 251
250 252 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
251 253
252 254 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
253 255 void dbuf_create_bonus(struct dnode *dn);
254 256 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
255 257 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
256 258
257 259 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
258 260
259 261 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
260 262 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
261 263 void *tag);
262 264 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
263 265 void *tag, dmu_buf_impl_t **dbp);
264 266
265 267 void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
266 268
267 269 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
|
↓ open down ↓ |
112 lines elided |
↑ open up ↑ |
268 270 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
269 271
270 272 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
271 273 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
272 274
273 275 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
274 276
275 277 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
276 278 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
277 279 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
280 +void dmu_buf_will_zero_fill(dmu_buf_t *db, dmu_tx_t *tx);
278 281 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
279 282 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
280 -dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
283 +dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx, boolean_t zero_write);
284 +dbuf_dirty_record_t *dbuf_zero_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
281 285 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
282 286 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
283 287 bp_embedded_type_t etype, enum zio_compress comp,
284 288 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
285 289
286 290 void dbuf_clear(dmu_buf_impl_t *db);
287 291 void dbuf_evict(dmu_buf_impl_t *db);
288 292
289 293 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
290 294 void dbuf_unoverride(dbuf_dirty_record_t *dr);
291 295 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
292 296 void dbuf_release_bp(dmu_buf_impl_t *db);
293 297
294 298 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
295 299 struct dmu_tx *);
296 300
297 301 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
298 302
299 303 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
300 304 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
301 305 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
302 306 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
303 307 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
304 308
305 309 void dbuf_init(void);
306 310 void dbuf_fini(void);
307 311
308 312 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
309 313
310 314 #define DBUF_GET_BUFC_TYPE(_db) \
311 315 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
312 316
313 317 #define DBUF_IS_CACHEABLE(_db) \
314 318 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
315 319 (dbuf_is_metadata(_db) && \
316 320 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
317 321
318 322 #define DBUF_IS_L2CACHEABLE(_db) \
319 323 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
320 324 (dbuf_is_metadata(_db) && \
321 325 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
322 326
323 327 #define DBUF_IS_L2COMPRESSIBLE(_db) \
324 328 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF || \
325 329 (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
326 330
327 331 #ifdef ZFS_DEBUG
328 332
329 333 /*
330 334 * There should be a ## between the string literal and fmt, to make it
331 335 * clear that we're joining two strings together, but gcc does not
332 336 * support that preprocessor token.
333 337 */
334 338 #define dprintf_dbuf(dbuf, fmt, ...) do { \
335 339 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
336 340 char __db_buf[32]; \
337 341 uint64_t __db_obj = (dbuf)->db.db_object; \
338 342 if (__db_obj == DMU_META_DNODE_OBJECT) \
339 343 (void) strcpy(__db_buf, "mdn"); \
340 344 else \
341 345 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
342 346 (u_longlong_t)__db_obj); \
343 347 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
344 348 "obj=%s lvl=%u blkid=%lld " fmt, \
345 349 __db_buf, (dbuf)->db_level, \
346 350 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
347 351 } \
348 352 _NOTE(CONSTCOND) } while (0)
349 353
350 354 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
351 355 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
352 356 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
353 357 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
354 358 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
355 359 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
356 360 } \
357 361 _NOTE(CONSTCOND) } while (0)
358 362
359 363 #define DBUF_VERIFY(db) dbuf_verify(db)
360 364
361 365 #else
362 366
363 367 #define dprintf_dbuf(db, fmt, ...)
364 368 #define dprintf_dbuf_bp(db, bp, fmt, ...)
365 369 #define DBUF_VERIFY(db)
366 370
367 371 #endif
368 372
369 373
370 374 #ifdef __cplusplus
371 375 }
372 376 #endif
373 377
374 378 #endif /* _SYS_DBUF_H */
|
↓ open down ↓ |
84 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX