Print this page
4045 zfs write throttle & i/o scheduler performance work
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.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 *
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
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 - * Copyright (c) 2012 by Delphix. All rights reserved.
23 + * Copyright (c) 2013 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 typedef enum dbuf_states {
71 71 DB_UNCACHED,
72 72 DB_FILL,
73 73 DB_NOFILL,
74 74 DB_READ,
75 75 DB_CACHED,
76 76 DB_EVICTING
77 77 } dbuf_states_t;
78 78
79 79 struct dnode;
80 80 struct dmu_tx;
81 81
82 82 /*
83 83 * level = 0 means the user data
84 84 * level = 1 means the single indirect block
85 85 * etc.
86 86 */
87 87
88 88 struct dmu_buf_impl;
89 89
90 90 typedef enum override_states {
91 91 DR_NOT_OVERRIDDEN,
92 92 DR_IN_DMU_SYNC,
93 93 DR_OVERRIDDEN
94 94 } override_states_t;
95 95
96 96 typedef struct dbuf_dirty_record {
97 97 /* link on our parents dirty list */
98 98 list_node_t dr_dirty_node;
99 99
100 100 /* transaction group this data will sync in */
101 101 uint64_t dr_txg;
102 102
103 103 /* zio of outstanding write IO */
104 104 zio_t *dr_zio;
|
↓ open down ↓ |
71 lines elided |
↑ open up ↑ |
105 105
106 106 /* pointer back to our dbuf */
107 107 struct dmu_buf_impl *dr_dbuf;
108 108
109 109 /* pointer to next dirty record */
110 110 struct dbuf_dirty_record *dr_next;
111 111
112 112 /* pointer to parent dirty record */
113 113 struct dbuf_dirty_record *dr_parent;
114 114
115 + /* How much space was changed to dsl_pool_dirty_space() for this? */
116 + unsigned int dr_accounted;
117 +
115 118 union dirty_types {
116 119 struct dirty_indirect {
117 120
118 121 /* protect access to list */
119 122 kmutex_t dr_mtx;
120 123
121 124 /* Our list of dirty children */
122 125 list_t dr_children;
123 126 } di;
124 127 struct dirty_leaf {
125 128
126 129 /*
127 130 * dr_data is set when we dirty the buffer
128 131 * so that we can retain the pointer even if it
129 132 * gets COW'd in a subsequent transaction group.
130 133 */
131 134 arc_buf_t *dr_data;
132 135 blkptr_t dr_overridden_by;
133 136 override_states_t dr_override_state;
134 137 uint8_t dr_copies;
135 138 boolean_t dr_nopwrite;
136 139 } dl;
137 140 } dt;
138 141 } dbuf_dirty_record_t;
139 142
140 143 typedef struct dmu_buf_impl {
141 144 /*
142 145 * The following members are immutable, with the exception of
143 146 * db.db_data, which is protected by db_mtx.
144 147 */
145 148
146 149 /* the publicly visible structure */
147 150 dmu_buf_t db;
148 151
149 152 /* the objset we belong to */
150 153 struct objset *db_objset;
151 154
152 155 /*
153 156 * handle to safely access the dnode we belong to (NULL when evicted)
154 157 */
155 158 struct dnode_handle *db_dnode_handle;
156 159
157 160 /*
158 161 * our parent buffer; if the dnode points to us directly,
159 162 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
160 163 * only accessed by sync thread ???
161 164 * (NULL when evicted)
162 165 * May change from NULL to non-NULL under the protection of db_mtx
163 166 * (see dbuf_check_blkptr())
164 167 */
165 168 struct dmu_buf_impl *db_parent;
166 169
167 170 /*
168 171 * link for hash table of all dmu_buf_impl_t's
169 172 */
170 173 struct dmu_buf_impl *db_hash_next;
171 174
172 175 /* our block number */
173 176 uint64_t db_blkid;
174 177
175 178 /*
176 179 * Pointer to the blkptr_t which points to us. May be NULL if we
177 180 * don't have one yet. (NULL when evicted)
178 181 */
179 182 blkptr_t *db_blkptr;
180 183
181 184 /*
182 185 * Our indirection level. Data buffers have db_level==0.
183 186 * Indirect buffers which point to data buffers have
184 187 * db_level==1. etc. Buffers which contain dnodes have
185 188 * db_level==0, since the dnodes are stored in a file.
186 189 */
187 190 uint8_t db_level;
188 191
189 192 /* db_mtx protects the members below */
190 193 kmutex_t db_mtx;
191 194
192 195 /*
193 196 * Current state of the buffer
194 197 */
195 198 dbuf_states_t db_state;
196 199
197 200 /*
198 201 * Refcount accessed by dmu_buf_{hold,rele}.
199 202 * If nonzero, the buffer can't be destroyed.
200 203 * Protected by db_mtx.
201 204 */
202 205 refcount_t db_holds;
203 206
204 207 /* buffer holding our data */
205 208 arc_buf_t *db_buf;
206 209
207 210 kcondvar_t db_changed;
208 211 dbuf_dirty_record_t *db_data_pending;
209 212
210 213 /* pointer to most recent dirty record for this buffer */
211 214 dbuf_dirty_record_t *db_last_dirty;
212 215
213 216 /*
214 217 * Our link on the owner dnodes's dn_dbufs list.
215 218 * Protected by its dn_dbufs_mtx.
216 219 */
217 220 list_node_t db_link;
218 221
219 222 /* Data which is unique to data (leaf) blocks: */
220 223
221 224 /* stuff we store for the user (see dmu_buf_set_user) */
222 225 void *db_user_ptr;
223 226 void **db_user_data_ptr_ptr;
224 227 dmu_buf_evict_func_t *db_evict_func;
225 228
226 229 uint8_t db_immediate_evict;
227 230 uint8_t db_freed_in_flight;
228 231
229 232 uint8_t db_dirtycnt;
230 233 } dmu_buf_impl_t;
231 234
232 235 /* Note: the dbuf hash table is exposed only for the mdb module */
233 236 #define DBUF_MUTEXES 256
234 237 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
235 238 typedef struct dbuf_hash_table {
236 239 uint64_t hash_table_mask;
237 240 dmu_buf_impl_t **hash_table;
238 241 kmutex_t hash_mutexes[DBUF_MUTEXES];
239 242 } dbuf_hash_table_t;
240 243
241 244
242 245 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
243 246
244 247 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
245 248 void dbuf_create_bonus(struct dnode *dn);
246 249 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
|
↓ open down ↓ |
122 lines elided |
↑ open up ↑ |
247 250 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
248 251
249 252 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
250 253
251 254 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
252 255 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
253 256 void *tag);
254 257 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
255 258 void *tag, dmu_buf_impl_t **dbp);
256 259
257 -void dbuf_prefetch(struct dnode *dn, uint64_t blkid);
260 +void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
258 261
259 262 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
260 263 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
261 264
262 265 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
263 266 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
264 267
265 268 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
266 269
267 270 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
268 271 void dbuf_will_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
269 272 void dbuf_fill_done(dmu_buf_impl_t *db, dmu_tx_t *tx);
270 273 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
271 274 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
272 275 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
273 276 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
274 277 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
275 278 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
276 279
277 280 void dbuf_clear(dmu_buf_impl_t *db);
278 281 void dbuf_evict(dmu_buf_impl_t *db);
279 282
280 283 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
281 284 void dbuf_unoverride(dbuf_dirty_record_t *dr);
282 285 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
283 286 void dbuf_release_bp(dmu_buf_impl_t *db);
284 287
285 288 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
286 289 struct dmu_tx *);
287 290
288 291 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
289 292
290 293 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
291 294 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
292 295 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
293 296 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
294 297 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
295 298 #define DB_GET_SPA(_spa_p, _db) { \
296 299 dnode_t *__dn; \
297 300 DB_DNODE_ENTER(_db); \
298 301 __dn = DB_DNODE(_db); \
299 302 *(_spa_p) = __dn->dn_objset->os_spa; \
300 303 DB_DNODE_EXIT(_db); \
301 304 }
302 305 #define DB_GET_OBJSET(_os_p, _db) { \
303 306 dnode_t *__dn; \
304 307 DB_DNODE_ENTER(_db); \
305 308 __dn = DB_DNODE(_db); \
306 309 *(_os_p) = __dn->dn_objset; \
307 310 DB_DNODE_EXIT(_db); \
308 311 }
309 312
310 313 void dbuf_init(void);
311 314 void dbuf_fini(void);
312 315
313 316 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
314 317
315 318 #define DBUF_GET_BUFC_TYPE(_db) \
316 319 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
317 320
318 321 #define DBUF_IS_CACHEABLE(_db) \
319 322 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
320 323 (dbuf_is_metadata(_db) && \
321 324 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
322 325
323 326 #define DBUF_IS_L2CACHEABLE(_db) \
324 327 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
325 328 (dbuf_is_metadata(_db) && \
326 329 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
327 330
328 331 #define DBUF_IS_L2COMPRESSIBLE(_db) \
329 332 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF)
330 333
331 334 #ifdef ZFS_DEBUG
332 335
333 336 /*
334 337 * There should be a ## between the string literal and fmt, to make it
335 338 * clear that we're joining two strings together, but gcc does not
336 339 * support that preprocessor token.
337 340 */
338 341 #define dprintf_dbuf(dbuf, fmt, ...) do { \
339 342 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
340 343 char __db_buf[32]; \
341 344 uint64_t __db_obj = (dbuf)->db.db_object; \
342 345 if (__db_obj == DMU_META_DNODE_OBJECT) \
343 346 (void) strcpy(__db_buf, "mdn"); \
344 347 else \
345 348 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
346 349 (u_longlong_t)__db_obj); \
347 350 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
348 351 "obj=%s lvl=%u blkid=%lld " fmt, \
349 352 __db_buf, (dbuf)->db_level, \
350 353 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
351 354 } \
352 355 _NOTE(CONSTCOND) } while (0)
353 356
354 357 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
355 358 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
356 359 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
357 360 sprintf_blkptr(__blkbuf, bp); \
358 361 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
359 362 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
360 363 } \
361 364 _NOTE(CONSTCOND) } while (0)
362 365
363 366 #define DBUF_VERIFY(db) dbuf_verify(db)
364 367
365 368 #else
366 369
367 370 #define dprintf_dbuf(db, fmt, ...)
368 371 #define dprintf_dbuf_bp(db, bp, fmt, ...)
369 372 #define DBUF_VERIFY(db)
370 373
371 374 #endif
372 375
373 376
374 377 #ifdef __cplusplus
375 378 }
376 379 #endif
377 380
378 381 #endif /* _SYS_DBUF_H */
|
↓ open down ↓ |
111 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX