Print this page
3752 want more verifiable dbuf user eviction
Submitted by: Justin Gibbs <justing@spectralogic.com>
Submitted by: Will Andrews <willa@spectralogic.com>
| 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 by Delphix. All rights reserved.
24 24 */
25 25
26 26 #ifndef _SYS_DBUF_H
27 27 #define _SYS_DBUF_H
28 28
29 29 #include <sys/dmu.h>
30 30 #include <sys/spa.h>
31 31 #include <sys/txg.h>
32 32 #include <sys/zio.h>
33 33 #include <sys/arc.h>
34 34 #include <sys/zfs_context.h>
35 35 #include <sys/refcount.h>
36 36 #include <sys/zrlock.h>
37 37
38 38 #ifdef __cplusplus
39 39 extern "C" {
40 40 #endif
41 41
42 42 #define IN_DMU_SYNC 2
43 43
44 44 /*
45 45 * define flags for dbuf_read
46 46 */
47 47
48 48 #define DB_RF_MUST_SUCCEED (1 << 0)
49 49 #define DB_RF_CANFAIL (1 << 1)
50 50 #define DB_RF_HAVESTRUCT (1 << 2)
51 51 #define DB_RF_NOPREFETCH (1 << 3)
52 52 #define DB_RF_NEVERWAIT (1 << 4)
53 53 #define DB_RF_CACHED (1 << 5)
54 54
55 55 /*
56 56 * The simplified state transition diagram for dbufs looks like:
57 57 *
58 58 * +----> READ ----+
59 59 * | |
60 60 * | V
61 61 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
62 62 * | ^ ^
63 63 * | | |
64 64 * +----> FILL ----+ |
65 65 * | |
66 66 * | |
67 67 * +--------> NOFILL -------+
68 68 */
69 69 typedef enum dbuf_states {
70 70 DB_UNCACHED,
71 71 DB_FILL,
72 72 DB_NOFILL,
73 73 DB_READ,
74 74 DB_CACHED,
75 75 DB_EVICTING
76 76 } dbuf_states_t;
77 77
78 78 struct dnode;
79 79 struct dmu_tx;
80 80
81 81 /*
82 82 * level = 0 means the user data
83 83 * level = 1 means the single indirect block
84 84 * etc.
85 85 */
86 86
87 87 struct dmu_buf_impl;
88 88
89 89 typedef enum override_states {
90 90 DR_NOT_OVERRIDDEN,
91 91 DR_IN_DMU_SYNC,
92 92 DR_OVERRIDDEN
93 93 } override_states_t;
94 94
95 95 typedef struct dbuf_dirty_record {
96 96 /* link on our parents dirty list */
97 97 list_node_t dr_dirty_node;
98 98
99 99 /* transaction group this data will sync in */
100 100 uint64_t dr_txg;
101 101
102 102 /* zio of outstanding write IO */
103 103 zio_t *dr_zio;
104 104
105 105 /* pointer back to our dbuf */
106 106 struct dmu_buf_impl *dr_dbuf;
107 107
108 108 /* pointer to next dirty record */
109 109 struct dbuf_dirty_record *dr_next;
110 110
111 111 /* pointer to parent dirty record */
112 112 struct dbuf_dirty_record *dr_parent;
113 113
114 114 union dirty_types {
115 115 struct dirty_indirect {
116 116
117 117 /* protect access to list */
118 118 kmutex_t dr_mtx;
119 119
120 120 /* Our list of dirty children */
121 121 list_t dr_children;
122 122 } di;
123 123 struct dirty_leaf {
124 124
125 125 /*
126 126 * dr_data is set when we dirty the buffer
127 127 * so that we can retain the pointer even if it
128 128 * gets COW'd in a subsequent transaction group.
129 129 */
130 130 arc_buf_t *dr_data;
131 131 blkptr_t dr_overridden_by;
132 132 override_states_t dr_override_state;
133 133 uint8_t dr_copies;
134 134 boolean_t dr_nopwrite;
135 135 } dl;
136 136 } dt;
137 137 } dbuf_dirty_record_t;
138 138
139 139 typedef struct dmu_buf_impl {
140 140 /*
141 141 * The following members are immutable, with the exception of
142 142 * db.db_data, which is protected by db_mtx.
143 143 */
144 144
145 145 /* the publicly visible structure */
146 146 dmu_buf_t db;
147 147
148 148 /* the objset we belong to */
149 149 struct objset *db_objset;
150 150
151 151 /*
152 152 * handle to safely access the dnode we belong to (NULL when evicted)
153 153 */
154 154 struct dnode_handle *db_dnode_handle;
155 155
156 156 /*
157 157 * our parent buffer; if the dnode points to us directly,
158 158 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
159 159 * only accessed by sync thread ???
160 160 * (NULL when evicted)
161 161 * May change from NULL to non-NULL under the protection of db_mtx
162 162 * (see dbuf_check_blkptr())
163 163 */
164 164 struct dmu_buf_impl *db_parent;
165 165
166 166 /*
167 167 * link for hash table of all dmu_buf_impl_t's
168 168 */
169 169 struct dmu_buf_impl *db_hash_next;
170 170
171 171 /* our block number */
172 172 uint64_t db_blkid;
173 173
174 174 /*
175 175 * Pointer to the blkptr_t which points to us. May be NULL if we
176 176 * don't have one yet. (NULL when evicted)
177 177 */
178 178 blkptr_t *db_blkptr;
179 179
180 180 /*
181 181 * Our indirection level. Data buffers have db_level==0.
182 182 * Indirect buffers which point to data buffers have
183 183 * db_level==1. etc. Buffers which contain dnodes have
184 184 * db_level==0, since the dnodes are stored in a file.
185 185 */
186 186 uint8_t db_level;
187 187
188 188 /* db_mtx protects the members below */
189 189 kmutex_t db_mtx;
190 190
191 191 /*
192 192 * Current state of the buffer
193 193 */
194 194 dbuf_states_t db_state;
195 195
196 196 /*
197 197 * Refcount accessed by dmu_buf_{hold,rele}.
198 198 * If nonzero, the buffer can't be destroyed.
199 199 * Protected by db_mtx.
200 200 */
201 201 refcount_t db_holds;
202 202
203 203 /* buffer holding our data */
204 204 arc_buf_t *db_buf;
205 205
206 206 kcondvar_t db_changed;
207 207 dbuf_dirty_record_t *db_data_pending;
208 208
209 209 /* pointer to most recent dirty record for this buffer */
|
↓ open down ↓ |
209 lines elided |
↑ open up ↑ |
210 210 dbuf_dirty_record_t *db_last_dirty;
211 211
212 212 /*
213 213 * Our link on the owner dnodes's dn_dbufs list.
214 214 * Protected by its dn_dbufs_mtx.
215 215 */
216 216 list_node_t db_link;
217 217
218 218 /* Data which is unique to data (leaf) blocks: */
219 219
220 - /* stuff we store for the user (see dmu_buf_set_user) */
221 - void *db_user_ptr;
222 - void **db_user_data_ptr_ptr;
223 - dmu_buf_evict_func_t *db_evict_func;
220 + /* User callback information. See dmu_buf_set_user(). */
221 + dmu_buf_user_t *db_user;
224 222
225 223 uint8_t db_immediate_evict;
226 224 uint8_t db_freed_in_flight;
227 225
228 226 uint8_t db_dirtycnt;
229 227 } dmu_buf_impl_t;
230 228
231 229 /* Note: the dbuf hash table is exposed only for the mdb module */
232 230 #define DBUF_MUTEXES 256
233 231 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
234 232 typedef struct dbuf_hash_table {
235 233 uint64_t hash_table_mask;
236 234 dmu_buf_impl_t **hash_table;
237 235 kmutex_t hash_mutexes[DBUF_MUTEXES];
238 236 } dbuf_hash_table_t;
239 237
240 238
241 239 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
242 240
243 241 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
244 242 void dbuf_create_bonus(struct dnode *dn);
245 243 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
246 244 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
247 245
248 246 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
249 247
250 248 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
251 249 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
252 250 void *tag);
253 251 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
254 252 void *tag, dmu_buf_impl_t **dbp);
255 253
256 254 void dbuf_prefetch(struct dnode *dn, uint64_t blkid);
257 255
258 256 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
259 257 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
260 258
261 259 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
262 260 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
263 261
264 262 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
265 263
|
↓ open down ↓ |
32 lines elided |
↑ open up ↑ |
266 264 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
267 265 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
268 266 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
269 267 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
270 268 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
271 269 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
272 270 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
273 271 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
274 272 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
275 273
276 -void dbuf_clear(dmu_buf_impl_t *db);
277 -void dbuf_evict(dmu_buf_impl_t *db);
274 +void dbuf_clear(dmu_buf_impl_t *db, list_t *evict_list_p);
275 +void dbuf_evict(dmu_buf_impl_t *db, list_t *evict_list_p);
278 276
279 277 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
280 278 void dbuf_unoverride(dbuf_dirty_record_t *dr);
281 279 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
282 280 void dbuf_release_bp(dmu_buf_impl_t *db);
283 281
284 282 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
285 283 struct dmu_tx *);
286 284
287 285 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
288 286
289 287 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
290 288 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
291 289 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
292 290 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
293 291 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
294 292 #define DB_GET_SPA(_spa_p, _db) { \
295 293 dnode_t *__dn; \
296 294 DB_DNODE_ENTER(_db); \
297 295 __dn = DB_DNODE(_db); \
298 296 *(_spa_p) = __dn->dn_objset->os_spa; \
299 297 DB_DNODE_EXIT(_db); \
300 298 }
301 299 #define DB_GET_OBJSET(_os_p, _db) { \
302 300 dnode_t *__dn; \
303 301 DB_DNODE_ENTER(_db); \
304 302 __dn = DB_DNODE(_db); \
305 303 *(_os_p) = __dn->dn_objset; \
306 304 DB_DNODE_EXIT(_db); \
307 305 }
308 306
309 307 void dbuf_init(void);
310 308 void dbuf_fini(void);
311 309
312 310 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
313 311
314 312 #define DBUF_GET_BUFC_TYPE(_db) \
315 313 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
316 314
317 315 #define DBUF_IS_CACHEABLE(_db) \
318 316 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
319 317 (dbuf_is_metadata(_db) && \
320 318 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
321 319
322 320 #define DBUF_IS_L2CACHEABLE(_db) \
323 321 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
324 322 (dbuf_is_metadata(_db) && \
325 323 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
326 324
327 325 #ifdef ZFS_DEBUG
328 326
329 327 /*
330 328 * There should be a ## between the string literal and fmt, to make it
331 329 * clear that we're joining two strings together, but gcc does not
332 330 * support that preprocessor token.
333 331 */
334 332 #define dprintf_dbuf(dbuf, fmt, ...) do { \
335 333 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
336 334 char __db_buf[32]; \
337 335 uint64_t __db_obj = (dbuf)->db.db_object; \
338 336 if (__db_obj == DMU_META_DNODE_OBJECT) \
339 337 (void) strcpy(__db_buf, "mdn"); \
340 338 else \
341 339 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
342 340 (u_longlong_t)__db_obj); \
343 341 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
344 342 "obj=%s lvl=%u blkid=%lld " fmt, \
345 343 __db_buf, (dbuf)->db_level, \
346 344 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
347 345 } \
348 346 _NOTE(CONSTCOND) } while (0)
349 347
350 348 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
351 349 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
352 350 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
353 351 sprintf_blkptr(__blkbuf, bp); \
354 352 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
355 353 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
356 354 } \
357 355 _NOTE(CONSTCOND) } while (0)
358 356
359 357 #define DBUF_VERIFY(db) dbuf_verify(db)
360 358
361 359 #else
362 360
363 361 #define dprintf_dbuf(db, fmt, ...)
364 362 #define dprintf_dbuf_bp(db, bp, fmt, ...)
365 363 #define DBUF_VERIFY(db)
366 364
367 365 #endif
368 366
369 367
370 368 #ifdef __cplusplus
371 369 }
372 370 #endif
373 371
374 372 #endif /* _SYS_DBUF_H */
|
↓ open down ↓ |
87 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX