Print this page
3740 Poor ZFS send / receive performance due to snapshot hold / release processing
Submitted by: Steven Hartland <steven.hartland@multiplay.co.uk>
| 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.
|
↓ 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 by Delphix. All rights reserved.
24 24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 + * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 26 */
26 27
27 28 #ifndef _SYS_DSL_DATASET_H
28 29 #define _SYS_DSL_DATASET_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/bplist.h>
35 36 #include <sys/dsl_synctask.h>
36 37 #include <sys/zfs_context.h>
37 38 #include <sys/dsl_deadlist.h>
38 39 #include <sys/refcount.h>
39 40
40 41 #ifdef __cplusplus
41 42 extern "C" {
42 43 #endif
43 44
44 45 struct dsl_dataset;
45 46 struct dsl_dir;
46 47 struct dsl_pool;
47 48
48 49 #define DS_FLAG_INCONSISTENT (1ULL<<0)
49 50 #define DS_IS_INCONSISTENT(ds) \
50 51 ((ds)->ds_phys->ds_flags & DS_FLAG_INCONSISTENT)
51 52 /*
52 53 * Note: nopromote can not yet be set, but we want support for it in this
53 54 * on-disk version, so that we don't need to upgrade for it later.
54 55 */
55 56 #define DS_FLAG_NOPROMOTE (1ULL<<1)
56 57
57 58 /*
58 59 * DS_FLAG_UNIQUE_ACCURATE is set if ds_unique_bytes has been correctly
59 60 * calculated for head datasets (starting with SPA_VERSION_UNIQUE_ACCURATE,
60 61 * refquota/refreservations).
61 62 */
62 63 #define DS_FLAG_UNIQUE_ACCURATE (1ULL<<2)
63 64
64 65 /*
65 66 * DS_FLAG_DEFER_DESTROY is set after 'zfs destroy -d' has been called
66 67 * on a dataset. This allows the dataset to be destroyed using 'zfs release'.
67 68 */
68 69 #define DS_FLAG_DEFER_DESTROY (1ULL<<3)
69 70 #define DS_IS_DEFER_DESTROY(ds) \
70 71 ((ds)->ds_phys->ds_flags & DS_FLAG_DEFER_DESTROY)
71 72
72 73 /*
73 74 * DS_FLAG_CI_DATASET is set if the dataset contains a file system whose
74 75 * name lookups should be performed case-insensitively.
75 76 */
76 77 #define DS_FLAG_CI_DATASET (1ULL<<16)
77 78
78 79 #define DS_CREATE_FLAG_NODIRTY (1ULL<<24)
79 80
80 81 typedef struct dsl_dataset_phys {
81 82 uint64_t ds_dir_obj; /* DMU_OT_DSL_DIR */
82 83 uint64_t ds_prev_snap_obj; /* DMU_OT_DSL_DATASET */
83 84 uint64_t ds_prev_snap_txg;
84 85 uint64_t ds_next_snap_obj; /* DMU_OT_DSL_DATASET */
85 86 uint64_t ds_snapnames_zapobj; /* DMU_OT_DSL_DS_SNAP_MAP 0 for snaps */
86 87 uint64_t ds_num_children; /* clone/snap children; ==0 for head */
87 88 uint64_t ds_creation_time; /* seconds since 1970 */
88 89 uint64_t ds_creation_txg;
89 90 uint64_t ds_deadlist_obj; /* DMU_OT_DEADLIST */
90 91 /*
91 92 * ds_referenced_bytes, ds_compressed_bytes, and ds_uncompressed_bytes
92 93 * include all blocks referenced by this dataset, including those
93 94 * shared with any other datasets.
94 95 */
95 96 uint64_t ds_referenced_bytes;
96 97 uint64_t ds_compressed_bytes;
97 98 uint64_t ds_uncompressed_bytes;
98 99 uint64_t ds_unique_bytes; /* only relevant to snapshots */
99 100 /*
100 101 * The ds_fsid_guid is a 56-bit ID that can change to avoid
101 102 * collisions. The ds_guid is a 64-bit ID that will never
102 103 * change, so there is a small probability that it will collide.
103 104 */
104 105 uint64_t ds_fsid_guid;
105 106 uint64_t ds_guid;
106 107 uint64_t ds_flags; /* DS_FLAG_* */
107 108 blkptr_t ds_bp;
108 109 uint64_t ds_next_clones_obj; /* DMU_OT_DSL_CLONES */
109 110 uint64_t ds_props_obj; /* DMU_OT_DSL_PROPS for snaps */
110 111 uint64_t ds_userrefs_obj; /* DMU_OT_USERREFS */
111 112 uint64_t ds_pad[5]; /* pad out to 320 bytes for good measure */
112 113 } dsl_dataset_phys_t;
113 114
114 115 typedef struct dsl_dataset {
115 116 /* Immutable: */
116 117 struct dsl_dir *ds_dir;
117 118 dsl_dataset_phys_t *ds_phys;
118 119 dmu_buf_t *ds_dbuf;
119 120 uint64_t ds_object;
120 121 uint64_t ds_fsid_guid;
121 122
122 123 /* only used in syncing context, only valid for non-snapshots: */
123 124 struct dsl_dataset *ds_prev;
124 125
125 126 /* has internal locking: */
126 127 dsl_deadlist_t ds_deadlist;
127 128 bplist_t ds_pending_deadlist;
128 129
129 130 /* protected by lock on pool's dp_dirty_datasets list */
130 131 txg_node_t ds_dirty_link;
131 132 list_node_t ds_synced_link;
132 133
133 134 /*
134 135 * ds_phys->ds_<accounting> is also protected by ds_lock.
135 136 * Protected by ds_lock:
136 137 */
137 138 kmutex_t ds_lock;
138 139 objset_t *ds_objset;
139 140 uint64_t ds_userrefs;
140 141 void *ds_owner;
141 142
142 143 /*
143 144 * Long holds prevent the ds from being destroyed; they allow the
144 145 * ds to remain held even after dropping the dp_config_rwlock.
145 146 * Owning counts as a long hold. See the comments above
146 147 * dsl_pool_hold() for details.
147 148 */
148 149 refcount_t ds_longholds;
149 150
150 151 /* no locking; only for making guesses */
151 152 uint64_t ds_trysnap_txg;
152 153
153 154 /* for objset_open() */
154 155 kmutex_t ds_opening_lock;
155 156
156 157 uint64_t ds_reserved; /* cached refreservation */
157 158 uint64_t ds_quota; /* cached refquota */
158 159
159 160 kmutex_t ds_sendstream_lock;
160 161 list_t ds_sendstreams;
161 162
162 163 /* Protected by ds_lock; keep at end of struct for better locality */
163 164 char ds_snapname[MAXNAMELEN];
164 165 } dsl_dataset_t;
165 166
166 167 /*
167 168 * The max length of a temporary tag prefix is the number of hex digits
168 169 * required to express UINT64_MAX plus one for the hyphen.
169 170 */
170 171 #define MAX_TAG_PREFIX_LEN 17
171 172
172 173 #define dsl_dataset_is_snapshot(ds) \
173 174 ((ds)->ds_phys->ds_num_children != 0)
174 175
175 176 #define DS_UNIQUE_IS_ACCURATE(ds) \
176 177 (((ds)->ds_phys->ds_flags & DS_FLAG_UNIQUE_ACCURATE) != 0)
177 178
178 179 int dsl_dataset_hold(struct dsl_pool *dp, const char *name, void *tag,
179 180 dsl_dataset_t **dsp);
|
↓ open down ↓ |
145 lines elided |
↑ open up ↑ |
180 181 int dsl_dataset_hold_obj(struct dsl_pool *dp, uint64_t dsobj, void *tag,
181 182 dsl_dataset_t **);
182 183 void dsl_dataset_rele(dsl_dataset_t *ds, void *tag);
183 184 int dsl_dataset_own(struct dsl_pool *dp, const char *name,
184 185 void *tag, dsl_dataset_t **dsp);
185 186 int dsl_dataset_own_obj(struct dsl_pool *dp, uint64_t dsobj,
186 187 void *tag, dsl_dataset_t **dsp);
187 188 void dsl_dataset_disown(dsl_dataset_t *ds, void *tag);
188 189 void dsl_dataset_name(dsl_dataset_t *ds, char *name);
189 190 boolean_t dsl_dataset_tryown(dsl_dataset_t *ds, void *tag);
190 -void dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
191 - minor_t minor);
192 191 uint64_t dsl_dataset_create_sync(dsl_dir_t *pds, const char *lastname,
193 192 dsl_dataset_t *origin, uint64_t flags, cred_t *, dmu_tx_t *);
194 193 uint64_t dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
195 194 uint64_t flags, dmu_tx_t *tx);
196 195 int dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors);
197 196 int dsl_dataset_promote(const char *name, char *conflsnap);
198 197 int dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
199 198 boolean_t force);
200 199 int dsl_dataset_rename_snapshot(const char *fsname,
201 200 const char *oldsnapname, const char *newsnapname, boolean_t recursive);
202 201 int dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
203 202 minor_t cleanup_minor, const char *htag);
204 203
205 204 blkptr_t *dsl_dataset_get_blkptr(dsl_dataset_t *ds);
206 205 void dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx);
207 206
208 207 spa_t *dsl_dataset_get_spa(dsl_dataset_t *ds);
209 208
210 209 boolean_t dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds);
211 210
212 211 void dsl_dataset_sync(dsl_dataset_t *os, zio_t *zio, dmu_tx_t *tx);
213 212
214 213 void dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp,
215 214 dmu_tx_t *tx);
216 215 int dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp,
217 216 dmu_tx_t *tx, boolean_t async);
218 217 boolean_t dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
219 218 uint64_t blk_birth);
220 219 uint64_t dsl_dataset_prev_snap_txg(dsl_dataset_t *ds);
221 220
222 221 void dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx);
223 222 void dsl_dataset_stats(dsl_dataset_t *os, nvlist_t *nv);
224 223 void dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat);
225 224 void dsl_dataset_space(dsl_dataset_t *ds,
226 225 uint64_t *refdbytesp, uint64_t *availbytesp,
227 226 uint64_t *usedobjsp, uint64_t *availobjsp);
228 227 uint64_t dsl_dataset_fsid_guid(dsl_dataset_t *ds);
229 228 int dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
230 229 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp);
231 230 int dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap, dsl_dataset_t *last,
232 231 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp);
233 232 boolean_t dsl_dataset_is_dirty(dsl_dataset_t *ds);
234 233
235 234 int dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf);
236 235
237 236 int dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
238 237 uint64_t asize, uint64_t inflight, uint64_t *used,
239 238 uint64_t *ref_rsrv);
240 239 int dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
241 240 uint64_t quota);
242 241 int dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
243 242 uint64_t reservation);
244 243
245 244 boolean_t dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier);
246 245 void dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag);
247 246 void dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag);
248 247 boolean_t dsl_dataset_long_held(dsl_dataset_t *ds);
249 248
250 249 int dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
251 250 dsl_dataset_t *origin_head, boolean_t force);
252 251 void dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
253 252 dsl_dataset_t *origin_head, dmu_tx_t *tx);
254 253 int dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
255 254 dmu_tx_t *tx);
256 255 void dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
257 256 dmu_tx_t *tx);
258 257
259 258 void dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
260 259 dmu_tx_t *tx);
261 260 void dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds);
262 261 int dsl_dataset_get_snapname(dsl_dataset_t *ds);
263 262 int dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name,
264 263 uint64_t *value);
265 264 int dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx);
266 265 void dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
267 266 zprop_source_t source, uint64_t value, dmu_tx_t *tx);
268 267 int dsl_dataset_rollback(const char *fsname);
269 268
270 269 #ifdef ZFS_DEBUG
271 270 #define dprintf_ds(ds, fmt, ...) do { \
272 271 if (zfs_flags & ZFS_DEBUG_DPRINTF) { \
273 272 char *__ds_name = kmem_alloc(MAXNAMELEN, KM_SLEEP); \
274 273 dsl_dataset_name(ds, __ds_name); \
275 274 dprintf("ds=%s " fmt, __ds_name, __VA_ARGS__); \
276 275 kmem_free(__ds_name, MAXNAMELEN); \
277 276 } \
278 277 _NOTE(CONSTCOND) } while (0)
279 278 #else
280 279 #define dprintf_ds(dd, fmt, ...)
281 280 #endif
282 281
283 282 #ifdef __cplusplus
284 283 }
285 284 #endif
286 285
287 286 #endif /* _SYS_DSL_DATASET_H */
|
↓ open down ↓ |
86 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX