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/dsl_dataset.h
+++ new/usr/src/uts/common/fs/zfs/sys/dsl_dataset.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
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
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) 2013 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
25 25 * Copyright (c) 2013 Steven Hartland. All rights reserved.
26 + * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 27 */
27 28
28 29 #ifndef _SYS_DSL_DATASET_H
29 30 #define _SYS_DSL_DATASET_H
30 31
31 32 #include <sys/dmu.h>
32 33 #include <sys/spa.h>
33 34 #include <sys/txg.h>
34 35 #include <sys/zio.h>
35 36 #include <sys/bplist.h>
36 37 #include <sys/dsl_synctask.h>
37 38 #include <sys/zfs_context.h>
38 39 #include <sys/dsl_deadlist.h>
39 40 #include <sys/refcount.h>
40 41
41 42 #ifdef __cplusplus
42 43 extern "C" {
43 44 #endif
44 45
45 46 struct dsl_dataset;
46 47 struct dsl_dir;
47 48 struct dsl_pool;
48 49
49 50 #define DS_FLAG_INCONSISTENT (1ULL<<0)
50 51 #define DS_IS_INCONSISTENT(ds) \
51 52 (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT)
52 53
53 54 /*
54 55 * Do not allow this dataset to be promoted.
55 56 */
56 57 #define DS_FLAG_NOPROMOTE (1ULL<<1)
57 58
58 59 /*
59 60 * DS_FLAG_UNIQUE_ACCURATE is set if ds_unique_bytes has been correctly
60 61 * calculated for head datasets (starting with SPA_VERSION_UNIQUE_ACCURATE,
61 62 * refquota/refreservations).
62 63 */
63 64 #define DS_FLAG_UNIQUE_ACCURATE (1ULL<<2)
64 65
65 66 /*
66 67 * DS_FLAG_DEFER_DESTROY is set after 'zfs destroy -d' has been called
67 68 * on a dataset. This allows the dataset to be destroyed using 'zfs release'.
68 69 */
69 70 #define DS_FLAG_DEFER_DESTROY (1ULL<<3)
70 71 #define DS_IS_DEFER_DESTROY(ds) \
71 72 (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_DEFER_DESTROY)
72 73
73 74 /*
74 75 * DS_FIELD_* are strings that are used in the "extensified" dataset zap object.
75 76 * They should be of the format <reverse-dns>:<field>.
76 77 */
77 78
78 79 /*
79 80 * This field's value is the object ID of a zap object which contains the
80 81 * bookmarks of this dataset. If it is present, then this dataset is counted
81 82 * in the refcount of the SPA_FEATURES_BOOKMARKS feature.
82 83 */
83 84 #define DS_FIELD_BOOKMARK_NAMES "com.delphix:bookmarks"
84 85
85 86 /*
86 87 * This field is present (with value=0) if this dataset may contain large
87 88 * blocks (>128KB). If it is present, then this dataset
88 89 * is counted in the refcount of the SPA_FEATURE_LARGE_BLOCKS feature.
89 90 */
90 91 #define DS_FIELD_LARGE_BLOCKS "org.open-zfs:large_blocks"
91 92
92 93 /*
93 94 * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose
94 95 * name lookups should be performed case-insensitively.
95 96 */
96 97 #define DS_FLAG_CI_DATASET (1ULL<<16)
97 98
98 99 #define DS_CREATE_FLAG_NODIRTY (1ULL<<24)
99 100
100 101 typedef struct dsl_dataset_phys {
101 102 uint64_t ds_dir_obj; /* DMU_OT_DSL_DIR */
102 103 uint64_t ds_prev_snap_obj; /* DMU_OT_DSL_DATASET */
103 104 uint64_t ds_prev_snap_txg;
104 105 uint64_t ds_next_snap_obj; /* DMU_OT_DSL_DATASET */
105 106 uint64_t ds_snapnames_zapobj; /* DMU_OT_DSL_DS_SNAP_MAP 0 for snaps */
106 107 uint64_t ds_num_children; /* clone/snap children; ==0 for head */
107 108 uint64_t ds_creation_time; /* seconds since 1970 */
108 109 uint64_t ds_creation_txg;
109 110 uint64_t ds_deadlist_obj; /* DMU_OT_DEADLIST */
110 111 /*
111 112 * ds_referenced_bytes, ds_compressed_bytes, and ds_uncompressed_bytes
112 113 * include all blocks referenced by this dataset, including those
113 114 * shared with any other datasets.
114 115 */
115 116 uint64_t ds_referenced_bytes;
116 117 uint64_t ds_compressed_bytes;
117 118 uint64_t ds_uncompressed_bytes;
118 119 uint64_t ds_unique_bytes; /* only relevant to snapshots */
119 120 /*
120 121 * The ds_fsid_guid is a 56-bit ID that can change to avoid
121 122 * collisions. The ds_guid is a 64-bit ID that will never
122 123 * change, so there is a small probability that it will collide.
123 124 */
124 125 uint64_t ds_fsid_guid;
|
↓ open down ↓ |
89 lines elided |
↑ open up ↑ |
125 126 uint64_t ds_guid;
126 127 uint64_t ds_flags; /* DS_FLAG_* */
127 128 blkptr_t ds_bp;
128 129 uint64_t ds_next_clones_obj; /* DMU_OT_DSL_CLONES */
129 130 uint64_t ds_props_obj; /* DMU_OT_DSL_PROPS for snaps */
130 131 uint64_t ds_userrefs_obj; /* DMU_OT_USERREFS */
131 132 uint64_t ds_pad[5]; /* pad out to 320 bytes for good measure */
132 133 } dsl_dataset_phys_t;
133 134
134 135 typedef struct dsl_dataset {
136 + dmu_buf_user_t ds_dbu;
137 +
135 138 /* Immutable: */
136 139 struct dsl_dir *ds_dir;
137 140 dmu_buf_t *ds_dbuf;
138 141 uint64_t ds_object;
139 142 uint64_t ds_fsid_guid;
143 + boolean_t ds_is_snapshot;
140 144
141 145 /* only used in syncing context, only valid for non-snapshots: */
142 146 struct dsl_dataset *ds_prev;
143 147 uint64_t ds_bookmarks; /* DMU_OTN_ZAP_METADATA */
144 148 boolean_t ds_large_blocks;
145 149 boolean_t ds_need_large_blocks;
146 150
147 151 /* has internal locking: */
148 152 dsl_deadlist_t ds_deadlist;
149 153 bplist_t ds_pending_deadlist;
150 154
151 155 /* protected by lock on pool's dp_dirty_datasets list */
152 156 txg_node_t ds_dirty_link;
153 157 list_node_t ds_synced_link;
154 158
155 159 /*
156 160 * ds_phys->ds_<accounting> is also protected by ds_lock.
157 161 * Protected by ds_lock:
158 162 */
159 163 kmutex_t ds_lock;
160 164 objset_t *ds_objset;
161 165 uint64_t ds_userrefs;
162 166 void *ds_owner;
163 167
164 168 /*
165 169 * Long holds prevent the ds from being destroyed; they allow the
166 170 * ds to remain held even after dropping the dp_config_rwlock.
167 171 * Owning counts as a long hold. See the comments above
168 172 * dsl_pool_hold() for details.
169 173 */
170 174 refcount_t ds_longholds;
171 175
172 176 /* no locking; only for making guesses */
173 177 uint64_t ds_trysnap_txg;
174 178
175 179 /* for objset_open() */
176 180 kmutex_t ds_opening_lock;
177 181
178 182 uint64_t ds_reserved; /* cached refreservation */
179 183 uint64_t ds_quota; /* cached refquota */
180 184
181 185 kmutex_t ds_sendstream_lock;
182 186 list_t ds_sendstreams;
183 187
184 188 /* Protected by ds_lock; keep at end of struct for better locality */
185 189 char ds_snapname[MAXNAMELEN];
186 190 } dsl_dataset_t;
187 191
188 192 inline dsl_dataset_phys_t *
189 193 dsl_dataset_phys(dsl_dataset_t *ds)
|
↓ open down ↓ |
40 lines elided |
↑ open up ↑ |
190 194 {
191 195 return (ds->ds_dbuf->db_data);
192 196 }
193 197
194 198 /*
195 199 * The max length of a temporary tag prefix is the number of hex digits
196 200 * required to express UINT64_MAX plus one for the hyphen.
197 201 */
198 202 #define MAX_TAG_PREFIX_LEN 17
199 203
200 -inline boolean_t
201 -dsl_dataset_is_snapshot(dsl_dataset_t *ds)
202 -{
203 - return (dsl_dataset_phys(ds)->ds_num_children != 0);
204 -}
205 -
206 204 #define DS_UNIQUE_IS_ACCURATE(ds) \
207 205 ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0)
208 206
209 207 int dsl_dataset_hold(struct dsl_pool *dp, const char *name, void *tag,
210 208 dsl_dataset_t **dsp);
211 209 int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, void *tag,
212 210 dsl_dataset_t **);
213 211 void dsl_dataset_rele(dsl_dataset_t *ds, void *tag);
214 212 int dsl_dataset_own(struct dsl_pool *dp, const char *name,
215 213 void *tag, dsl_dataset_t **dsp);
216 214 int dsl_dataset_own_obj(struct dsl_pool *dp, uint64_t dsobj,
217 215 void *tag, dsl_dataset_t **dsp);
218 216 void dsl_dataset_disown(dsl_dataset_t *ds, void *tag);
219 217 void dsl_dataset_name(dsl_dataset_t *ds, char *name);
220 218 boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag);
221 219 uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname,
222 220 dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *);
223 221 uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
224 222 uint64_t flags, dmu_tx_t *tx);
225 223 int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors);
226 224 int dsl_dataset_promote(const char *name, char *conflsnap);
227 225 int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
228 226 boolean_t force);
229 227 int dsl_dataset_rename_snapshot(const char *fsname,
230 228 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
231 229 int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
232 230 minor_t cleanup_minor, const char *htag);
233 231
234 232 blkptr_t *dsl_dataset_get_blkptr(dsl_dataset_t *ds);
235 233 void dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx);
236 234
237 235 spa_t *dsl_dataset_get_spa(dsl_dataset_t *ds);
238 236
239 237 boolean_t dsl_dataset_modified_since_snap(dsl_dataset_t *ds,
240 238 dsl_dataset_t *snap);
241 239
242 240 void dsl_dataset_sync(dsl_dataset_t *os, zio_t *zio, dmu_tx_t *tx);
243 241
244 242 void dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp,
245 243 dmu_tx_t *tx);
246 244 int dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp,
247 245 dmu_tx_t *tx, boolean_t async);
248 246 boolean_t dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
249 247 uint64_t blk_birth);
250 248 uint64_t dsl_dataset_prev_snap_txg(dsl_dataset_t *ds);
251 249
252 250 void dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx);
253 251 void dsl_dataset_stats(dsl_dataset_t *os, nvlist_t *nv);
254 252 void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat);
255 253 void dsl_dataset_space(dsl_dataset_t *ds,
256 254 uint64_t *refdbytesp, uint64_t *availbytesp,
257 255 uint64_t *usedobjsp, uint64_t *availobjsp);
258 256 uint64_t dsl_dataset_fsid_guid(dsl_dataset_t *ds);
259 257 int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
260 258 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp);
261 259 int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *last,
262 260 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp);
263 261 boolean_t dsl_dataset_is_dirty(dsl_dataset_t *ds);
264 262 int dsl_dataset_activate_large_blocks(const char *dsname);
265 263 void dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx);
266 264
267 265 int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf);
268 266
269 267 int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
270 268 uint64_t asize, uint64_t inflight, uint64_t *used,
271 269 uint64_t *ref_rsrv);
272 270 int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
273 271 uint64_t quota);
274 272 int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
275 273 uint64_t reservation);
276 274
277 275 boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
278 276 uint64_t earlier_txg);
279 277 void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag);
280 278 void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag);
281 279 boolean_t dsl_dataset_long_held(dsl_dataset_t *ds);
282 280
283 281 int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
284 282 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx);
285 283 void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
286 284 dsl_dataset_t *origin_head, dmu_tx_t *tx);
287 285 int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
288 286 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr);
289 287 void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
290 288 dmu_tx_t *tx);
291 289
292 290 void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
293 291 dmu_tx_t *tx);
294 292 void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds);
295 293 int dsl_dataset_get_snapname(dsl_dataset_t *ds);
296 294 int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name,
297 295 uint64_t *value);
298 296 int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
299 297 boolean_t adj_cnt);
300 298 void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
301 299 zprop_source_t source, uint64_t value, dmu_tx_t *tx);
302 300 void dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx);
303 301 int dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result);
304 302
305 303 #ifdef ZFS_DEBUG
306 304 #define dprintf_ds(ds, fmt, ...) do { \
307 305 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
308 306 char *__ds_name = kmem_alloc(MAXNAMELEN, KM_SLEEP); \
309 307 dsl_dataset_name(ds, __ds_name); \
310 308 dprintf("ds=%s " fmt, __ds_name, __VA_ARGS__); \
311 309 kmem_free(__ds_name, MAXNAMELEN); \
312 310 } \
313 311 _NOTE(CONSTCOND) } while (0)
314 312 #else
315 313 #define dprintf_ds(dd, fmt, ...)
316 314 #endif
317 315
318 316 #ifdef __cplusplus
319 317 }
320 318 #endif
321 319
322 320 #endif /* _SYS_DSL_DATASET_H */
|
↓ open down ↓ |
107 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX