Print this page
patch nuke-the-dbuf-hash
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 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
26 26 */
27 27
28 28 #ifndef _SYS_DBUF_H
29 29 #define _SYS_DBUF_H
30 30
31 31 #include <sys/dmu.h>
32 32 #include <sys/spa.h>
33 33 #include <sys/txg.h>
34 34 #include <sys/zio.h>
35 35 #include <sys/arc.h>
36 36 #include <sys/zfs_context.h>
37 37 #include <sys/refcount.h>
38 38 #include <sys/zrlock.h>
39 39
40 40 #ifdef __cplusplus
41 41 extern "C" {
42 42 #endif
43 43
44 44 #define IN_DMU_SYNC 2
45 45
46 46 /*
47 47 * define flags for dbuf_read
48 48 */
49 49
50 50 #define DB_RF_MUST_SUCCEED (1 << 0)
51 51 #define DB_RF_CANFAIL (1 << 1)
52 52 #define DB_RF_HAVESTRUCT (1 << 2)
53 53 #define DB_RF_NOPREFETCH (1 << 3)
54 54 #define DB_RF_NEVERWAIT (1 << 4)
55 55 #define DB_RF_CACHED (1 << 5)
56 56
57 57 #define DBUF_EVICT_ALL -1
58 58
59 59 /*
60 60 * The simplified state transition diagram for dbufs looks like:
61 61 *
62 62 * +----> READ ----+
63 63 * | |
64 64 * | V
65 65 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
66 66 * | ^ ^
67 67 * | | |
68 68 * +----> FILL ----+ |
69 69 * | |
70 70 * | |
71 71 * +--------> NOFILL -------+
72 72 *
73 73 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
74 74 * to find all dbufs in a range of a dnode and must be less than any other
75 75 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
76 76 */
77 77 typedef enum dbuf_states {
78 78 DB_SEARCH = -1,
79 79 DB_UNCACHED,
80 80 DB_FILL,
81 81 DB_NOFILL,
82 82 DB_READ,
83 83 DB_CACHED,
84 84 DB_EVICTING
85 85 } dbuf_states_t;
86 86
87 87 struct dnode;
88 88 struct dmu_tx;
89 89
90 90 /*
91 91 * level = 0 means the user data
92 92 * level = 1 means the single indirect block
93 93 * etc.
94 94 */
95 95
96 96 struct dmu_buf_impl;
97 97
98 98 typedef enum override_states {
99 99 DR_NOT_OVERRIDDEN,
100 100 DR_IN_DMU_SYNC,
101 101 DR_OVERRIDDEN
102 102 } override_states_t;
103 103
104 104 typedef struct dbuf_dirty_record {
105 105 /* link on our parents dirty list */
106 106 list_node_t dr_dirty_node;
107 107
108 108 /* transaction group this data will sync in */
109 109 uint64_t dr_txg;
110 110
111 111 /* zio of outstanding write IO */
112 112 zio_t *dr_zio;
113 113
114 114 /* pointer back to our dbuf */
115 115 struct dmu_buf_impl *dr_dbuf;
116 116
117 117 /* pointer to next dirty record */
118 118 struct dbuf_dirty_record *dr_next;
119 119
120 120 /* pointer to parent dirty record */
121 121 struct dbuf_dirty_record *dr_parent;
122 122
123 123 /* How much space was changed to dsl_pool_dirty_space() for this? */
124 124 unsigned int dr_accounted;
125 125
126 126 /* use special class of dirty entry */
127 127 boolean_t dr_usesc;
128 128
129 129 union dirty_types {
130 130 struct dirty_indirect {
131 131
132 132 /* protect access to list */
133 133 kmutex_t dr_mtx;
134 134
135 135 /* Our list of dirty children */
136 136 list_t dr_children;
137 137 } di;
138 138 struct dirty_leaf {
139 139
140 140 /*
141 141 * dr_data is set when we dirty the buffer
142 142 * so that we can retain the pointer even if it
143 143 * gets COW'd in a subsequent transaction group.
144 144 */
145 145 arc_buf_t *dr_data;
146 146 blkptr_t dr_overridden_by;
147 147 override_states_t dr_override_state;
148 148 uint8_t dr_copies;
149 149 boolean_t dr_nopwrite;
150 150 } dl;
151 151 } dt;
152 152 } dbuf_dirty_record_t;
153 153
154 154 typedef struct dmu_buf_impl {
155 155 /*
156 156 * The following members are immutable, with the exception of
157 157 * db.db_data, which is protected by db_mtx.
158 158 */
159 159
160 160 /* the publicly visible structure */
161 161 dmu_buf_t db;
162 162
163 163 /* the objset we belong to */
164 164 struct objset *db_objset;
165 165
166 166 /*
167 167 * handle to safely access the dnode we belong to (NULL when evicted)
168 168 */
169 169 struct dnode_handle *db_dnode_handle;
170 170
↓ open down ↓ |
170 lines elided |
↑ open up ↑ |
171 171 /*
172 172 * our parent buffer; if the dnode points to us directly,
173 173 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
174 174 * only accessed by sync thread ???
175 175 * (NULL when evicted)
176 176 * May change from NULL to non-NULL under the protection of db_mtx
177 177 * (see dbuf_check_blkptr())
178 178 */
179 179 struct dmu_buf_impl *db_parent;
180 180
181 - /*
182 - * link for hash table of all dmu_buf_impl_t's
183 - */
184 - struct dmu_buf_impl *db_hash_next;
185 -
186 181 /* our block number */
187 182 uint64_t db_blkid;
188 183
189 184 /*
190 185 * Pointer to the blkptr_t which points to us. May be NULL if we
191 186 * don't have one yet. (NULL when evicted)
192 187 */
193 188 blkptr_t *db_blkptr;
194 189
195 190 /*
196 191 * Our indirection level. Data buffers have db_level==0.
197 192 * Indirect buffers which point to data buffers have
198 193 * db_level==1. etc. Buffers which contain dnodes have
199 194 * db_level==0, since the dnodes are stored in a file.
200 195 */
201 196 uint8_t db_level;
202 197
203 198 /* db_mtx protects the members below */
204 199 kmutex_t db_mtx;
205 200
206 201 /*
207 202 * Current state of the buffer
208 203 */
209 204 dbuf_states_t db_state;
210 205
211 206 /*
212 207 * Refcount accessed by dmu_buf_{hold,rele}.
213 208 * If nonzero, the buffer can't be destroyed.
214 209 * Protected by db_mtx.
215 210 */
216 211 refcount_t db_holds;
217 212
218 213 /* buffer holding our data */
219 214 arc_buf_t *db_buf;
220 215
221 216 kcondvar_t db_changed;
222 217 dbuf_dirty_record_t *db_data_pending;
223 218
224 219 /* pointer to most recent dirty record for this buffer */
225 220 dbuf_dirty_record_t *db_last_dirty;
226 221
227 222 /*
228 223 * Our link on the owner dnodes's dn_dbufs list.
229 224 * Protected by its dn_dbufs_mtx.
230 225 */
231 226 avl_node_t db_link;
232 227
233 228 /* Data which is unique to data (leaf) blocks: */
234 229
235 230 /* stuff we store for the user (see dmu_buf_set_user) */
236 231 void *db_user_ptr;
237 232 void **db_user_data_ptr_ptr;
238 233 dmu_buf_evict_func_t *db_evict_func;
239 234
240 235 uint8_t db_immediate_evict;
241 236 uint8_t db_freed_in_flight;
242 237
243 238 uint8_t db_dirtycnt;
244 239 } dmu_buf_impl_t;
245 240
246 241 /* Note: the dbuf hash table is exposed only for the mdb module */
247 242 #define DBUF_MUTEXES 256
248 243 #define DBUF_LOCK_PAD 64
249 244 typedef struct {
250 245 kmutex_t mtx;
251 246 #ifdef _KERNEL
252 247 unsigned char pad[(DBUF_LOCK_PAD - sizeof (kmutex_t))];
253 248 #endif
254 249 } dbuf_mutex_t;
255 250 #define DBUF_HASH_MUTEX(h, idx) \
256 251 (&((h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)].mtx))
257 252 typedef struct dbuf_hash_table {
258 253 uint64_t hash_table_mask;
259 254 dmu_buf_impl_t **hash_table;
260 255 dbuf_mutex_t hash_mutexes[DBUF_MUTEXES];
261 256 } dbuf_hash_table_t;
262 257
263 258
264 259 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
265 260
266 261 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
267 262 void dbuf_create_bonus(struct dnode *dn);
268 263 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
269 264 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
270 265
271 266 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
272 267
273 268 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
274 269 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
275 270 void *tag);
276 271 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
277 272 void *tag, dmu_buf_impl_t **dbp);
278 273
279 274 void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
280 275
281 276 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
282 277 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
283 278
284 279 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
285 280 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
286 281
287 282 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
288 283
289 284 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
290 285 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
291 286 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
292 287 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
293 288 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
294 289 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
295 290 dbuf_dirty_record_t *dbuf_dirty_sc(dmu_buf_impl_t *db, dmu_tx_t *tx,
296 291 boolean_t usesc);
297 292 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
298 293 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
299 294 bp_embedded_type_t etype, enum zio_compress comp,
300 295 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
301 296
302 297 void dbuf_clear(dmu_buf_impl_t *db);
303 298 void dbuf_evict(dmu_buf_impl_t *db);
304 299
305 300 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
306 301 void dbuf_unoverride(dbuf_dirty_record_t *dr);
307 302 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
308 303 void dbuf_release_bp(dmu_buf_impl_t *db);
309 304
310 305 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
311 306 struct dmu_tx *);
312 307
313 308 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
314 309
315 310 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
316 311 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
317 312 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
318 313 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
319 314 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
320 315
321 316 void dbuf_init(void);
322 317 void dbuf_fini(void);
323 318
324 319 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
325 320 boolean_t dbuf_ddt_is_l2cacheable(dmu_buf_impl_t *db);
326 321 boolean_t dbuf_meta_is_l2cacheable(dmu_buf_impl_t *db);
327 322
328 323 #define DBUF_GET_BUFC_TYPE(_db) \
329 324 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
330 325
331 326 #define DBUF_IS_CACHEABLE(_db) \
332 327 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
333 328 (dbuf_is_metadata(_db) && \
334 329 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
335 330
336 331
337 332 /*
338 333 * Checks whether we need to cache dbuf in l2arc.
339 334 * Metadata is l2cacheable if it is not placed on special device
340 335 * or it is placed on special device in "dual" mode. We need to check
341 336 * for ddt in ZFS_CACHE_ALL and ZFS_CACHE_METADATA because it is in MOS.
342 337 * ZFS_CACHE_DATA mode actually means to cache both data and cacheable
343 338 * metadata.
344 339 */
345 340 #define DBUF_IS_L2CACHEABLE(_db) \
346 341 (((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL && \
347 342 (dbuf_ddt_is_l2cacheable(_db) == B_TRUE)) || \
348 343 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA && \
349 344 (dbuf_is_metadata(_db)) && \
350 345 (dbuf_ddt_is_l2cacheable(_db) == B_TRUE)) || \
351 346 ((dbuf_meta_is_l2cacheable(_db) == B_TRUE) && \
352 347 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_DATA)))
353 348
354 349 #define DBUF_IS_L2COMPRESSIBLE(_db) \
355 350 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF || \
356 351 (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
357 352
358 353 #ifdef ZFS_DEBUG
359 354
360 355 /*
361 356 * There should be a ## between the string literal and fmt, to make it
362 357 * clear that we're joining two strings together, but gcc does not
363 358 * support that preprocessor token.
364 359 */
365 360 #define dprintf_dbuf(dbuf, fmt, ...) do { \
366 361 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
367 362 char __db_buf[32]; \
368 363 uint64_t __db_obj = (dbuf)->db.db_object; \
369 364 if (__db_obj == DMU_META_DNODE_OBJECT) \
370 365 (void) strcpy(__db_buf, "mdn"); \
371 366 else \
372 367 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
373 368 (u_longlong_t)__db_obj); \
374 369 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
375 370 "obj=%s lvl=%u blkid=%lld " fmt, \
376 371 __db_buf, (dbuf)->db_level, \
377 372 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
378 373 } \
379 374 _NOTE(CONSTCOND) } while (0)
380 375
381 376 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
382 377 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
383 378 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
384 379 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
385 380 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
386 381 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
387 382 } \
388 383 _NOTE(CONSTCOND) } while (0)
389 384
390 385 #define DBUF_VERIFY(db) dbuf_verify(db)
391 386
392 387 #else
393 388
394 389 #define dprintf_dbuf(db, fmt, ...)
395 390 #define dprintf_dbuf_bp(db, bp, fmt, ...)
396 391 #define DBUF_VERIFY(db)
397 392
398 393 #endif
399 394
400 395
401 396 #ifdef __cplusplus
402 397 }
403 398 #endif
404 399
405 400 #endif /* _SYS_DBUF_H */
↓ open down ↓ |
210 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX