Print this page
5056 ZFS deadlock on db_mtx and dn_holds
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.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.
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
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 + * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 26 */
26 27
27 28 #ifndef _SYS_DBUF_H
28 29 #define _SYS_DBUF_H
29 30
30 31 #include <sys/dmu.h>
31 32 #include <sys/spa.h>
32 33 #include <sys/txg.h>
33 34 #include <sys/zio.h>
34 35 #include <sys/arc.h>
35 36 #include <sys/zfs_context.h>
36 37 #include <sys/refcount.h>
37 38 #include <sys/zrlock.h>
38 39
39 40 #ifdef __cplusplus
40 41 extern "C" {
41 42 #endif
42 43
43 44 #define IN_DMU_SYNC 2
44 45
45 46 /*
46 47 * define flags for dbuf_read
47 48 */
48 49
49 50 #define DB_RF_MUST_SUCCEED (1 << 0)
50 51 #define DB_RF_CANFAIL (1 << 1)
51 52 #define DB_RF_HAVESTRUCT (1 << 2)
52 53 #define DB_RF_NOPREFETCH (1 << 3)
53 54 #define DB_RF_NEVERWAIT (1 << 4)
54 55 #define DB_RF_CACHED (1 << 5)
55 56
56 57 /*
57 58 * The simplified state transition diagram for dbufs looks like:
58 59 *
59 60 * +----> READ ----+
60 61 * | |
61 62 * | V
62 63 * (alloc)-->UNCACHED CACHED-->EVICTING-->(free)
63 64 * | ^ ^
64 65 * | | |
65 66 * +----> FILL ----+ |
66 67 * | |
67 68 * | |
68 69 * +--------> NOFILL -------+
69 70 *
70 71 * DB_SEARCH is an invalid state for a dbuf. It is used by dbuf_free_range
71 72 * to find all dbufs in a range of a dnode and must be less than any other
72 73 * dbuf_states_t (see comment on dn_dbufs in dnode.h).
73 74 */
74 75 typedef enum dbuf_states {
75 76 DB_SEARCH = -1,
76 77 DB_UNCACHED,
77 78 DB_FILL,
78 79 DB_NOFILL,
79 80 DB_READ,
80 81 DB_CACHED,
81 82 DB_EVICTING
82 83 } dbuf_states_t;
83 84
84 85 struct dnode;
85 86 struct dmu_tx;
86 87
87 88 /*
88 89 * level = 0 means the user data
89 90 * level = 1 means the single indirect block
90 91 * etc.
91 92 */
92 93
93 94 struct dmu_buf_impl;
94 95
95 96 typedef enum override_states {
96 97 DR_NOT_OVERRIDDEN,
97 98 DR_IN_DMU_SYNC,
98 99 DR_OVERRIDDEN
99 100 } override_states_t;
100 101
101 102 typedef struct dbuf_dirty_record {
102 103 /* link on our parents dirty list */
103 104 list_node_t dr_dirty_node;
104 105
105 106 /* transaction group this data will sync in */
106 107 uint64_t dr_txg;
107 108
108 109 /* zio of outstanding write IO */
109 110 zio_t *dr_zio;
110 111
111 112 /* pointer back to our dbuf */
112 113 struct dmu_buf_impl *dr_dbuf;
113 114
114 115 /* pointer to next dirty record */
115 116 struct dbuf_dirty_record *dr_next;
116 117
117 118 /* pointer to parent dirty record */
118 119 struct dbuf_dirty_record *dr_parent;
119 120
120 121 /* How much space was changed to dsl_pool_dirty_space() for this? */
121 122 unsigned int dr_accounted;
122 123
123 124 union dirty_types {
124 125 struct dirty_indirect {
125 126
126 127 /* protect access to list */
127 128 kmutex_t dr_mtx;
128 129
129 130 /* Our list of dirty children */
130 131 list_t dr_children;
131 132 } di;
132 133 struct dirty_leaf {
133 134
134 135 /*
135 136 * dr_data is set when we dirty the buffer
136 137 * so that we can retain the pointer even if it
137 138 * gets COW'd in a subsequent transaction group.
138 139 */
139 140 arc_buf_t *dr_data;
140 141 blkptr_t dr_overridden_by;
141 142 override_states_t dr_override_state;
142 143 uint8_t dr_copies;
143 144 boolean_t dr_nopwrite;
144 145 } dl;
145 146 } dt;
146 147 } dbuf_dirty_record_t;
147 148
148 149 typedef struct dmu_buf_impl {
149 150 /*
150 151 * The following members are immutable, with the exception of
151 152 * db.db_data, which is protected by db_mtx.
152 153 */
153 154
154 155 /* the publicly visible structure */
155 156 dmu_buf_t db;
156 157
157 158 /* the objset we belong to */
158 159 struct objset *db_objset;
159 160
160 161 /*
161 162 * handle to safely access the dnode we belong to (NULL when evicted)
162 163 */
163 164 struct dnode_handle *db_dnode_handle;
164 165
165 166 /*
166 167 * our parent buffer; if the dnode points to us directly,
167 168 * db_parent == db_dnode_handle->dnh_dnode->dn_dbuf
168 169 * only accessed by sync thread ???
169 170 * (NULL when evicted)
170 171 * May change from NULL to non-NULL under the protection of db_mtx
171 172 * (see dbuf_check_blkptr())
172 173 */
173 174 struct dmu_buf_impl *db_parent;
174 175
175 176 /*
176 177 * link for hash table of all dmu_buf_impl_t's
177 178 */
178 179 struct dmu_buf_impl *db_hash_next;
179 180
180 181 /* our block number */
181 182 uint64_t db_blkid;
182 183
183 184 /*
184 185 * Pointer to the blkptr_t which points to us. May be NULL if we
185 186 * don't have one yet. (NULL when evicted)
186 187 */
187 188 blkptr_t *db_blkptr;
188 189
189 190 /*
190 191 * Our indirection level. Data buffers have db_level==0.
191 192 * Indirect buffers which point to data buffers have
192 193 * db_level==1. etc. Buffers which contain dnodes have
193 194 * db_level==0, since the dnodes are stored in a file.
194 195 */
195 196 uint8_t db_level;
196 197
197 198 /* db_mtx protects the members below */
198 199 kmutex_t db_mtx;
199 200
200 201 /*
201 202 * Current state of the buffer
202 203 */
203 204 dbuf_states_t db_state;
204 205
205 206 /*
206 207 * Refcount accessed by dmu_buf_{hold,rele}.
207 208 * If nonzero, the buffer can't be destroyed.
208 209 * Protected by db_mtx.
209 210 */
210 211 refcount_t db_holds;
211 212
212 213 /* buffer holding our data */
213 214 arc_buf_t *db_buf;
214 215
215 216 kcondvar_t db_changed;
216 217 dbuf_dirty_record_t *db_data_pending;
217 218
218 219 /* pointer to most recent dirty record for this buffer */
|
↓ open down ↓ |
184 lines elided |
↑ open up ↑ |
219 220 dbuf_dirty_record_t *db_last_dirty;
220 221
221 222 /*
222 223 * Our link on the owner dnodes's dn_dbufs list.
223 224 * Protected by its dn_dbufs_mtx.
224 225 */
225 226 avl_node_t db_link;
226 227
227 228 /* Data which is unique to data (leaf) blocks: */
228 229
229 - /* stuff we store for the user (see dmu_buf_set_user) */
230 - void *db_user_ptr;
231 - dmu_buf_evict_func_t *db_evict_func;
230 + /* User callback information. */
231 + dmu_buf_user_t *db_user;
232 232
233 233 uint8_t db_immediate_evict;
234 234 uint8_t db_freed_in_flight;
235 235
236 236 uint8_t db_dirtycnt;
237 237 } dmu_buf_impl_t;
238 238
239 239 /* Note: the dbuf hash table is exposed only for the mdb module */
240 240 #define DBUF_MUTEXES 256
241 241 #define DBUF_HASH_MUTEX(h, idx) (&(h)->hash_mutexes[(idx) & (DBUF_MUTEXES-1)])
242 242 typedef struct dbuf_hash_table {
243 243 uint64_t hash_table_mask;
244 244 dmu_buf_impl_t **hash_table;
245 245 kmutex_t hash_mutexes[DBUF_MUTEXES];
246 246 } dbuf_hash_table_t;
247 247
248 248
249 249 uint64_t dbuf_whichblock(struct dnode *di, uint64_t offset);
250 250
251 251 dmu_buf_impl_t *dbuf_create_tlib(struct dnode *dn, char *data);
252 252 void dbuf_create_bonus(struct dnode *dn);
253 253 int dbuf_spill_set_blksz(dmu_buf_t *db, uint64_t blksz, dmu_tx_t *tx);
254 254 void dbuf_spill_hold(struct dnode *dn, dmu_buf_impl_t **dbp, void *tag);
255 255
256 256 void dbuf_rm_spill(struct dnode *dn, dmu_tx_t *tx);
257 257
258 258 dmu_buf_impl_t *dbuf_hold(struct dnode *dn, uint64_t blkid, void *tag);
259 259 dmu_buf_impl_t *dbuf_hold_level(struct dnode *dn, int level, uint64_t blkid,
260 260 void *tag);
261 261 int dbuf_hold_impl(struct dnode *dn, uint8_t level, uint64_t blkid, int create,
262 262 void *tag, dmu_buf_impl_t **dbp);
263 263
264 264 void dbuf_prefetch(struct dnode *dn, uint64_t blkid, zio_priority_t prio);
265 265
266 266 void dbuf_add_ref(dmu_buf_impl_t *db, void *tag);
267 267 uint64_t dbuf_refcount(dmu_buf_impl_t *db);
268 268
269 269 void dbuf_rele(dmu_buf_impl_t *db, void *tag);
270 270 void dbuf_rele_and_unlock(dmu_buf_impl_t *db, void *tag);
271 271
272 272 dmu_buf_impl_t *dbuf_find(struct dnode *dn, uint8_t level, uint64_t blkid);
273 273
274 274 int dbuf_read(dmu_buf_impl_t *db, zio_t *zio, uint32_t flags);
275 275 void dmu_buf_will_not_fill(dmu_buf_t *db, dmu_tx_t *tx);
276 276 void dmu_buf_will_fill(dmu_buf_t *db, dmu_tx_t *tx);
277 277 void dmu_buf_fill_done(dmu_buf_t *db, dmu_tx_t *tx);
278 278 void dbuf_assign_arcbuf(dmu_buf_impl_t *db, arc_buf_t *buf, dmu_tx_t *tx);
279 279 dbuf_dirty_record_t *dbuf_dirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
280 280 arc_buf_t *dbuf_loan_arcbuf(dmu_buf_impl_t *db);
281 281 void dmu_buf_write_embedded(dmu_buf_t *dbuf, void *data,
282 282 bp_embedded_type_t etype, enum zio_compress comp,
283 283 int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx);
284 284
285 285 void dbuf_clear(dmu_buf_impl_t *db);
286 286 void dbuf_evict(dmu_buf_impl_t *db);
287 287
288 288 void dbuf_setdirty(dmu_buf_impl_t *db, dmu_tx_t *tx);
289 289 void dbuf_unoverride(dbuf_dirty_record_t *dr);
290 290 void dbuf_sync_list(list_t *list, dmu_tx_t *tx);
291 291 void dbuf_release_bp(dmu_buf_impl_t *db);
292 292
293 293 void dbuf_free_range(struct dnode *dn, uint64_t start, uint64_t end,
294 294 struct dmu_tx *);
295 295
296 296 void dbuf_new_size(dmu_buf_impl_t *db, int size, dmu_tx_t *tx);
297 297
298 298 #define DB_DNODE(_db) ((_db)->db_dnode_handle->dnh_dnode)
299 299 #define DB_DNODE_LOCK(_db) ((_db)->db_dnode_handle->dnh_zrlock)
300 300 #define DB_DNODE_ENTER(_db) (zrl_add(&DB_DNODE_LOCK(_db)))
301 301 #define DB_DNODE_EXIT(_db) (zrl_remove(&DB_DNODE_LOCK(_db)))
302 302 #define DB_DNODE_HELD(_db) (!zrl_is_zero(&DB_DNODE_LOCK(_db)))
303 303
304 304 void dbuf_init(void);
305 305 void dbuf_fini(void);
306 306
307 307 boolean_t dbuf_is_metadata(dmu_buf_impl_t *db);
308 308
309 309 #define DBUF_GET_BUFC_TYPE(_db) \
310 310 (dbuf_is_metadata(_db) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
311 311
312 312 #define DBUF_IS_CACHEABLE(_db) \
313 313 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_ALL || \
314 314 (dbuf_is_metadata(_db) && \
315 315 ((_db)->db_objset->os_primary_cache == ZFS_CACHE_METADATA)))
316 316
317 317 #define DBUF_IS_L2CACHEABLE(_db) \
318 318 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_ALL || \
319 319 (dbuf_is_metadata(_db) && \
320 320 ((_db)->db_objset->os_secondary_cache == ZFS_CACHE_METADATA)))
321 321
322 322 #define DBUF_IS_L2COMPRESSIBLE(_db) \
323 323 ((_db)->db_objset->os_compress != ZIO_COMPRESS_OFF || \
324 324 (dbuf_is_metadata(_db) && zfs_mdcomp_disable == B_FALSE))
325 325
326 326 #ifdef ZFS_DEBUG
327 327
328 328 /*
329 329 * There should be a ## between the string literal and fmt, to make it
330 330 * clear that we're joining two strings together, but gcc does not
331 331 * support that preprocessor token.
332 332 */
333 333 #define dprintf_dbuf(dbuf, fmt, ...) do { \
334 334 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
335 335 char __db_buf[32]; \
336 336 uint64_t __db_obj = (dbuf)->db.db_object; \
337 337 if (__db_obj == DMU_META_DNODE_OBJECT) \
338 338 (void) strcpy(__db_buf, "mdn"); \
339 339 else \
340 340 (void) snprintf(__db_buf, sizeof (__db_buf), "%lld", \
341 341 (u_longlong_t)__db_obj); \
342 342 dprintf_ds((dbuf)->db_objset->os_dsl_dataset, \
343 343 "obj=%s lvl=%u blkid=%lld " fmt, \
344 344 __db_buf, (dbuf)->db_level, \
345 345 (u_longlong_t)(dbuf)->db_blkid, __VA_ARGS__); \
346 346 } \
347 347 _NOTE(CONSTCOND) } while (0)
348 348
349 349 #define dprintf_dbuf_bp(db, bp, fmt, ...) do { \
350 350 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
351 351 char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP); \
352 352 snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, bp); \
353 353 dprintf_dbuf(db, fmt " %s\n", __VA_ARGS__, __blkbuf); \
354 354 kmem_free(__blkbuf, BP_SPRINTF_LEN); \
355 355 } \
356 356 _NOTE(CONSTCOND) } while (0)
357 357
358 358 #define DBUF_VERIFY(db) dbuf_verify(db)
359 359
360 360 #else
361 361
362 362 #define dprintf_dbuf(db, fmt, ...)
363 363 #define dprintf_dbuf_bp(db, bp, fmt, ...)
364 364 #define DBUF_VERIFY(db)
365 365
366 366 #endif
367 367
368 368
369 369 #ifdef __cplusplus
370 370 }
371 371 #endif
372 372
373 373 #endif /* _SYS_DBUF_H */
|
↓ open down ↓ |
132 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX