]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
2e528b49 | 23 | * Copyright (c) 2013 by Delphix. All rights reserved. |
95fd54a1 | 24 | * Copyright (c) 2013 Steven Hartland. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/dsl_pool.h> |
28 | #include <sys/dsl_dataset.h> | |
428870ff | 29 | #include <sys/dsl_prop.h> |
34dc7c2f BB |
30 | #include <sys/dsl_dir.h> |
31 | #include <sys/dsl_synctask.h> | |
428870ff BB |
32 | #include <sys/dsl_scan.h> |
33 | #include <sys/dnode.h> | |
34dc7c2f BB |
34 | #include <sys/dmu_tx.h> |
35 | #include <sys/dmu_objset.h> | |
36 | #include <sys/arc.h> | |
37 | #include <sys/zap.h> | |
38 | #include <sys/zio.h> | |
39 | #include <sys/zfs_context.h> | |
40 | #include <sys/fs/zfs.h> | |
b128c09f BB |
41 | #include <sys/zfs_znode.h> |
42 | #include <sys/spa_impl.h> | |
428870ff | 43 | #include <sys/dsl_deadlist.h> |
9ae529ec CS |
44 | #include <sys/bptree.h> |
45 | #include <sys/zfeature.h> | |
29809a6c | 46 | #include <sys/zil_impl.h> |
13fe0198 | 47 | #include <sys/dsl_userhold.h> |
34dc7c2f | 48 | |
e8b96c60 MA |
49 | /* |
50 | * ZFS Write Throttle | |
51 | * ------------------ | |
52 | * | |
53 | * ZFS must limit the rate of incoming writes to the rate at which it is able | |
54 | * to sync data modifications to the backend storage. Throttling by too much | |
55 | * creates an artificial limit; throttling by too little can only be sustained | |
56 | * for short periods and would lead to highly lumpy performance. On a per-pool | |
57 | * basis, ZFS tracks the amount of modified (dirty) data. As operations change | |
58 | * data, the amount of dirty data increases; as ZFS syncs out data, the amount | |
59 | * of dirty data decreases. When the amount of dirty data exceeds a | |
60 | * predetermined threshold further modifications are blocked until the amount | |
61 | * of dirty data decreases (as data is synced out). | |
62 | * | |
63 | * The limit on dirty data is tunable, and should be adjusted according to | |
64 | * both the IO capacity and available memory of the system. The larger the | |
65 | * window, the more ZFS is able to aggregate and amortize metadata (and data) | |
66 | * changes. However, memory is a limited resource, and allowing for more dirty | |
67 | * data comes at the cost of keeping other useful data in memory (for example | |
68 | * ZFS data cached by the ARC). | |
69 | * | |
70 | * Implementation | |
71 | * | |
72 | * As buffers are modified dsl_pool_willuse_space() increments both the per- | |
73 | * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of | |
74 | * dirty space used; dsl_pool_dirty_space() decrements those values as data | |
75 | * is synced out from dsl_pool_sync(). While only the poolwide value is | |
76 | * relevant, the per-txg value is useful for debugging. The tunable | |
77 | * zfs_dirty_data_max determines the dirty space limit. Once that value is | |
78 | * exceeded, new writes are halted until space frees up. | |
79 | * | |
80 | * The zfs_dirty_data_sync tunable dictates the threshold at which we | |
81 | * ensure that there is a txg syncing (see the comment in txg.c for a full | |
82 | * description of transaction group stages). | |
83 | * | |
84 | * The IO scheduler uses both the dirty space limit and current amount of | |
85 | * dirty data as inputs. Those values affect the number of concurrent IOs ZFS | |
86 | * issues. See the comment in vdev_queue.c for details of the IO scheduler. | |
87 | * | |
88 | * The delay is also calculated based on the amount of dirty data. See the | |
89 | * comment above dmu_tx_delay() for details. | |
90 | */ | |
91 | ||
92 | /* | |
93 | * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, | |
94 | * capped at zfs_dirty_data_max_max. It can also be overridden with a module | |
95 | * parameter. | |
96 | */ | |
97 | unsigned long zfs_dirty_data_max = 0; | |
98 | unsigned long zfs_dirty_data_max_max = 0; | |
99 | int zfs_dirty_data_max_percent = 10; | |
100 | int zfs_dirty_data_max_max_percent = 25; | |
b128c09f | 101 | |
e8b96c60 MA |
102 | /* |
103 | * If there is at least this much dirty data, push out a txg. | |
104 | */ | |
105 | unsigned long zfs_dirty_data_sync = 64 * 1024 * 1024; | |
34dc7c2f | 106 | |
e8b96c60 MA |
107 | /* |
108 | * Once there is this amount of dirty data, the dmu_tx_delay() will kick in | |
109 | * and delay each transaction. | |
110 | * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. | |
111 | */ | |
112 | int zfs_delay_min_dirty_percent = 60; | |
b128c09f | 113 | |
e8b96c60 MA |
114 | /* |
115 | * This controls how quickly the delay approaches infinity. | |
116 | * Larger values cause it to delay more for a given amount of dirty data. | |
117 | * Therefore larger values will cause there to be less dirty data for a | |
118 | * given throughput. | |
119 | * | |
120 | * For the smoothest delay, this value should be about 1 billion divided | |
121 | * by the maximum number of operations per second. This will smoothly | |
122 | * handle between 10x and 1/10th this number. | |
123 | * | |
124 | * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the | |
125 | * multiply in dmu_tx_delay(). | |
126 | */ | |
127 | unsigned long zfs_delay_scale = 1000 * 1000 * 1000 / 2000; | |
b128c09f | 128 | |
63fd3c6c AL |
129 | hrtime_t zfs_throttle_delay = MSEC2NSEC(10); |
130 | hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); | |
131 | ||
428870ff | 132 | int |
b128c09f | 133 | dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) |
34dc7c2f BB |
134 | { |
135 | uint64_t obj; | |
136 | int err; | |
137 | ||
138 | err = zap_lookup(dp->dp_meta_objset, | |
139 | dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, | |
b128c09f | 140 | name, sizeof (obj), 1, &obj); |
34dc7c2f BB |
141 | if (err) |
142 | return (err); | |
143 | ||
13fe0198 | 144 | return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); |
34dc7c2f BB |
145 | } |
146 | ||
147 | static dsl_pool_t * | |
148 | dsl_pool_open_impl(spa_t *spa, uint64_t txg) | |
149 | { | |
150 | dsl_pool_t *dp; | |
151 | blkptr_t *bp = spa_get_rootblkptr(spa); | |
34dc7c2f BB |
152 | |
153 | dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); | |
154 | dp->dp_spa = spa; | |
155 | dp->dp_meta_rootbp = *bp; | |
13fe0198 | 156 | rrw_init(&dp->dp_config_rwlock, B_TRUE); |
34dc7c2f BB |
157 | txg_init(dp, txg); |
158 | ||
159 | txg_list_create(&dp->dp_dirty_datasets, | |
160 | offsetof(dsl_dataset_t, ds_dirty_link)); | |
29809a6c MA |
161 | txg_list_create(&dp->dp_dirty_zilogs, |
162 | offsetof(zilog_t, zl_dirty_link)); | |
34dc7c2f BB |
163 | txg_list_create(&dp->dp_dirty_dirs, |
164 | offsetof(dsl_dir_t, dd_dirty_link)); | |
165 | txg_list_create(&dp->dp_sync_tasks, | |
13fe0198 | 166 | offsetof(dsl_sync_task_t, dst_node)); |
34dc7c2f BB |
167 | |
168 | mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); | |
e8b96c60 | 169 | cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); |
34dc7c2f | 170 | |
3558fd73 | 171 | dp->dp_iput_taskq = taskq_create("zfs_iput_taskq", 1, minclsyspri, |
9babb374 BB |
172 | 1, 4, 0); |
173 | ||
34dc7c2f BB |
174 | return (dp); |
175 | } | |
176 | ||
177 | int | |
9ae529ec | 178 | dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) |
34dc7c2f BB |
179 | { |
180 | int err; | |
181 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
9ae529ec CS |
182 | |
183 | err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, | |
184 | &dp->dp_meta_objset); | |
185 | if (err != 0) | |
186 | dsl_pool_close(dp); | |
187 | else | |
188 | *dpp = dp; | |
189 | ||
190 | return (err); | |
191 | } | |
192 | ||
193 | int | |
194 | dsl_pool_open(dsl_pool_t *dp) | |
195 | { | |
196 | int err; | |
b128c09f BB |
197 | dsl_dir_t *dd; |
198 | dsl_dataset_t *ds; | |
428870ff | 199 | uint64_t obj; |
34dc7c2f | 200 | |
13fe0198 | 201 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
34dc7c2f BB |
202 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
203 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, | |
204 | &dp->dp_root_dir_obj); | |
205 | if (err) | |
206 | goto out; | |
207 | ||
13fe0198 | 208 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
209 | NULL, dp, &dp->dp_root_dir); |
210 | if (err) | |
211 | goto out; | |
212 | ||
b128c09f | 213 | err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); |
34dc7c2f BB |
214 | if (err) |
215 | goto out; | |
216 | ||
9ae529ec | 217 | if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { |
b128c09f BB |
218 | err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); |
219 | if (err) | |
220 | goto out; | |
221 | err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, | |
222 | FTAG, &ds); | |
9babb374 BB |
223 | if (err == 0) { |
224 | err = dsl_dataset_hold_obj(dp, | |
225 | ds->ds_phys->ds_prev_snap_obj, dp, | |
226 | &dp->dp_origin_snap); | |
227 | dsl_dataset_rele(ds, FTAG); | |
228 | } | |
13fe0198 | 229 | dsl_dir_rele(dd, dp); |
b128c09f BB |
230 | if (err) |
231 | goto out; | |
b128c09f BB |
232 | } |
233 | ||
9ae529ec | 234 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { |
428870ff BB |
235 | err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, |
236 | &dp->dp_free_dir); | |
b128c09f BB |
237 | if (err) |
238 | goto out; | |
428870ff | 239 | |
b128c09f | 240 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 241 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); |
b128c09f BB |
242 | if (err) |
243 | goto out; | |
13fe0198 | 244 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff | 245 | dp->dp_meta_objset, obj)); |
b128c09f BB |
246 | } |
247 | ||
fa86b5db | 248 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { |
9ae529ec CS |
249 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
250 | DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, | |
251 | &dp->dp_bptree_obj); | |
252 | if (err != 0) | |
253 | goto out; | |
254 | } | |
255 | ||
fa86b5db | 256 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { |
753c3839 MA |
257 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
258 | DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, | |
259 | &dp->dp_empty_bpobj); | |
260 | if (err != 0) | |
261 | goto out; | |
262 | } | |
263 | ||
428870ff BB |
264 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
265 | DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, | |
266 | &dp->dp_tmp_userrefs_obj); | |
267 | if (err == ENOENT) | |
268 | err = 0; | |
269 | if (err) | |
270 | goto out; | |
271 | ||
9ae529ec | 272 | err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); |
428870ff | 273 | |
34dc7c2f | 274 | out: |
13fe0198 | 275 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
34dc7c2f BB |
276 | return (err); |
277 | } | |
278 | ||
279 | void | |
280 | dsl_pool_close(dsl_pool_t *dp) | |
281 | { | |
b128c09f | 282 | /* |
e8b96c60 MA |
283 | * Drop our references from dsl_pool_open(). |
284 | * | |
b128c09f BB |
285 | * Since we held the origin_snap from "syncing" context (which |
286 | * includes pool-opening context), it actually only got a "ref" | |
287 | * and not a hold, so just drop that here. | |
288 | */ | |
289 | if (dp->dp_origin_snap) | |
13fe0198 | 290 | dsl_dataset_rele(dp->dp_origin_snap, dp); |
34dc7c2f | 291 | if (dp->dp_mos_dir) |
13fe0198 | 292 | dsl_dir_rele(dp->dp_mos_dir, dp); |
428870ff | 293 | if (dp->dp_free_dir) |
13fe0198 | 294 | dsl_dir_rele(dp->dp_free_dir, dp); |
34dc7c2f | 295 | if (dp->dp_root_dir) |
13fe0198 | 296 | dsl_dir_rele(dp->dp_root_dir, dp); |
34dc7c2f | 297 | |
428870ff BB |
298 | bpobj_close(&dp->dp_free_bpobj); |
299 | ||
34dc7c2f BB |
300 | /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ |
301 | if (dp->dp_meta_objset) | |
428870ff | 302 | dmu_objset_evict(dp->dp_meta_objset); |
34dc7c2f BB |
303 | |
304 | txg_list_destroy(&dp->dp_dirty_datasets); | |
29809a6c | 305 | txg_list_destroy(&dp->dp_dirty_zilogs); |
428870ff | 306 | txg_list_destroy(&dp->dp_sync_tasks); |
34dc7c2f | 307 | txg_list_destroy(&dp->dp_dirty_dirs); |
34dc7c2f BB |
308 | |
309 | arc_flush(dp->dp_spa); | |
310 | txg_fini(dp); | |
428870ff | 311 | dsl_scan_fini(dp); |
13fe0198 | 312 | rrw_destroy(&dp->dp_config_rwlock); |
34dc7c2f | 313 | mutex_destroy(&dp->dp_lock); |
3558fd73 | 314 | taskq_destroy(dp->dp_iput_taskq); |
b128c09f BB |
315 | if (dp->dp_blkstats) |
316 | kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); | |
34dc7c2f BB |
317 | kmem_free(dp, sizeof (dsl_pool_t)); |
318 | } | |
319 | ||
320 | dsl_pool_t * | |
b128c09f | 321 | dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) |
34dc7c2f BB |
322 | { |
323 | int err; | |
324 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
325 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
428870ff | 326 | objset_t *os; |
b128c09f | 327 | dsl_dataset_t *ds; |
428870ff | 328 | uint64_t obj; |
b128c09f | 329 | |
13fe0198 MA |
330 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
331 | ||
b128c09f | 332 | /* create and open the MOS (meta-objset) */ |
428870ff BB |
333 | dp->dp_meta_objset = dmu_objset_create_impl(spa, |
334 | NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); | |
34dc7c2f BB |
335 | |
336 | /* create the pool directory */ | |
337 | err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
338 | DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); | |
c99c9001 | 339 | ASSERT0(err); |
34dc7c2f | 340 | |
428870ff | 341 | /* Initialize scan structures */ |
13fe0198 | 342 | VERIFY0(dsl_scan_init(dp, txg)); |
428870ff | 343 | |
34dc7c2f | 344 | /* create and open the root dir */ |
b128c09f | 345 | dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); |
13fe0198 | 346 | VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
347 | NULL, dp, &dp->dp_root_dir)); |
348 | ||
349 | /* create and open the meta-objset dir */ | |
b128c09f | 350 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); |
13fe0198 | 351 | VERIFY0(dsl_pool_open_special_dir(dp, |
b128c09f BB |
352 | MOS_DIR_NAME, &dp->dp_mos_dir)); |
353 | ||
428870ff BB |
354 | if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { |
355 | /* create and open the free dir */ | |
356 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, | |
357 | FREE_DIR_NAME, tx); | |
13fe0198 | 358 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
359 | FREE_DIR_NAME, &dp->dp_free_dir)); |
360 | ||
361 | /* create and open the free_bplist */ | |
362 | obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); | |
363 | VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
364 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); | |
13fe0198 | 365 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff BB |
366 | dp->dp_meta_objset, obj)); |
367 | } | |
368 | ||
b128c09f BB |
369 | if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) |
370 | dsl_pool_create_origin(dp, tx); | |
371 | ||
372 | /* create the root dataset */ | |
428870ff | 373 | obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); |
b128c09f BB |
374 | |
375 | /* create the root objset */ | |
13fe0198 | 376 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); |
0fe3d820 BB |
377 | VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds, |
378 | dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx))); | |
b128c09f | 379 | #ifdef _KERNEL |
428870ff | 380 | zfs_create_fs(os, kcred, zplprops, tx); |
b128c09f BB |
381 | #endif |
382 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
383 | |
384 | dmu_tx_commit(tx); | |
385 | ||
13fe0198 MA |
386 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
387 | ||
34dc7c2f BB |
388 | return (dp); |
389 | } | |
390 | ||
29809a6c MA |
391 | /* |
392 | * Account for the meta-objset space in its placeholder dsl_dir. | |
393 | */ | |
394 | void | |
395 | dsl_pool_mos_diduse_space(dsl_pool_t *dp, | |
396 | int64_t used, int64_t comp, int64_t uncomp) | |
397 | { | |
398 | ASSERT3U(comp, ==, uncomp); /* it's all metadata */ | |
399 | mutex_enter(&dp->dp_lock); | |
400 | dp->dp_mos_used_delta += used; | |
401 | dp->dp_mos_compressed_delta += comp; | |
402 | dp->dp_mos_uncompressed_delta += uncomp; | |
403 | mutex_exit(&dp->dp_lock); | |
404 | } | |
405 | ||
428870ff BB |
406 | static int |
407 | deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
408 | { | |
409 | dsl_deadlist_t *dl = arg; | |
410 | dsl_deadlist_insert(dl, bp, tx); | |
411 | return (0); | |
412 | } | |
413 | ||
e8b96c60 MA |
414 | static void |
415 | dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) | |
416 | { | |
417 | zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
418 | dmu_objset_sync(dp->dp_meta_objset, zio, tx); | |
419 | VERIFY0(zio_wait(zio)); | |
420 | dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); | |
421 | spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); | |
422 | } | |
423 | ||
424 | static void | |
425 | dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) | |
426 | { | |
427 | ASSERT(MUTEX_HELD(&dp->dp_lock)); | |
428 | ||
429 | if (delta < 0) | |
430 | ASSERT3U(-delta, <=, dp->dp_dirty_total); | |
431 | ||
432 | dp->dp_dirty_total += delta; | |
433 | ||
434 | /* | |
435 | * Note: we signal even when increasing dp_dirty_total. | |
436 | * This ensures forward progress -- each thread wakes the next waiter. | |
437 | */ | |
438 | if (dp->dp_dirty_total <= zfs_dirty_data_max) | |
439 | cv_signal(&dp->dp_spaceavail_cv); | |
440 | } | |
441 | ||
34dc7c2f BB |
442 | void |
443 | dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) | |
444 | { | |
445 | zio_t *zio; | |
446 | dmu_tx_t *tx; | |
447 | dsl_dir_t *dd; | |
448 | dsl_dataset_t *ds; | |
428870ff | 449 | objset_t *mos = dp->dp_meta_objset; |
29809a6c MA |
450 | list_t synced_datasets; |
451 | ||
452 | list_create(&synced_datasets, sizeof (dsl_dataset_t), | |
453 | offsetof(dsl_dataset_t, ds_synced_link)); | |
34dc7c2f BB |
454 | |
455 | tx = dmu_tx_create_assigned(dp, txg); | |
456 | ||
e8b96c60 MA |
457 | /* |
458 | * Write out all dirty blocks of dirty datasets. | |
459 | */ | |
34dc7c2f | 460 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
e8b96c60 | 461 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { |
9babb374 BB |
462 | /* |
463 | * We must not sync any non-MOS datasets twice, because | |
464 | * we may have taken a snapshot of them. However, we | |
465 | * may sync newly-created datasets on pass 2. | |
466 | */ | |
467 | ASSERT(!list_link_active(&ds->ds_synced_link)); | |
29809a6c | 468 | list_insert_tail(&synced_datasets, ds); |
34dc7c2f BB |
469 | dsl_dataset_sync(ds, zio, tx); |
470 | } | |
e8b96c60 | 471 | VERIFY0(zio_wait(zio)); |
9babb374 | 472 | |
e8b96c60 MA |
473 | /* |
474 | * We have written all of the accounted dirty data, so our | |
475 | * dp_space_towrite should now be zero. However, some seldom-used | |
476 | * code paths do not adhere to this (e.g. dbuf_undirty(), also | |
477 | * rounding error in dbuf_write_physdone). | |
478 | * Shore up the accounting of any dirtied space now. | |
479 | */ | |
480 | dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); | |
34dc7c2f | 481 | |
29809a6c MA |
482 | /* |
483 | * After the data blocks have been written (ensured by the zio_wait() | |
484 | * above), update the user/group space accounting. | |
485 | */ | |
e8b96c60 MA |
486 | for (ds = list_head(&synced_datasets); ds != NULL; |
487 | ds = list_next(&synced_datasets, ds)) { | |
428870ff | 488 | dmu_objset_do_userquota_updates(ds->ds_objset, tx); |
e8b96c60 | 489 | } |
9babb374 BB |
490 | |
491 | /* | |
492 | * Sync the datasets again to push out the changes due to | |
428870ff | 493 | * userspace updates. This must be done before we process the |
29809a6c MA |
494 | * sync tasks, so that any snapshots will have the correct |
495 | * user accounting information (and we won't get confused | |
496 | * about which blocks are part of the snapshot). | |
9babb374 BB |
497 | */ |
498 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
e8b96c60 | 499 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { |
9babb374 BB |
500 | ASSERT(list_link_active(&ds->ds_synced_link)); |
501 | dmu_buf_rele(ds->ds_dbuf, ds); | |
502 | dsl_dataset_sync(ds, zio, tx); | |
503 | } | |
e8b96c60 | 504 | VERIFY0(zio_wait(zio)); |
9babb374 | 505 | |
428870ff | 506 | /* |
29809a6c MA |
507 | * Now that the datasets have been completely synced, we can |
508 | * clean up our in-memory structures accumulated while syncing: | |
509 | * | |
510 | * - move dead blocks from the pending deadlist to the on-disk deadlist | |
29809a6c | 511 | * - release hold from dsl_dataset_dirty() |
428870ff | 512 | */ |
e8b96c60 | 513 | while ((ds = list_remove_head(&synced_datasets)) != NULL) { |
29809a6c | 514 | ASSERTV(objset_t *os = ds->ds_objset); |
428870ff BB |
515 | bplist_iterate(&ds->ds_pending_deadlist, |
516 | deadlist_enqueue_cb, &ds->ds_deadlist, tx); | |
29809a6c MA |
517 | ASSERT(!dmu_objset_is_dirty(os, txg)); |
518 | dmu_buf_rele(ds->ds_dbuf, ds); | |
428870ff BB |
519 | } |
520 | ||
e8b96c60 | 521 | while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { |
34dc7c2f | 522 | dsl_dir_sync(dd, tx); |
e8b96c60 | 523 | } |
b128c09f | 524 | |
29809a6c MA |
525 | /* |
526 | * The MOS's space is accounted for in the pool/$MOS | |
527 | * (dp_mos_dir). We can't modify the mos while we're syncing | |
528 | * it, so we remember the deltas and apply them here. | |
529 | */ | |
530 | if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || | |
531 | dp->dp_mos_uncompressed_delta != 0) { | |
532 | dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, | |
533 | dp->dp_mos_used_delta, | |
534 | dp->dp_mos_compressed_delta, | |
535 | dp->dp_mos_uncompressed_delta, tx); | |
536 | dp->dp_mos_used_delta = 0; | |
537 | dp->dp_mos_compressed_delta = 0; | |
538 | dp->dp_mos_uncompressed_delta = 0; | |
539 | } | |
540 | ||
428870ff BB |
541 | if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || |
542 | list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { | |
e8b96c60 | 543 | dsl_pool_sync_mos(dp, tx); |
34dc7c2f BB |
544 | } |
545 | ||
29809a6c MA |
546 | /* |
547 | * If we modify a dataset in the same txg that we want to destroy it, | |
548 | * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. | |
549 | * dsl_dir_destroy_check() will fail if there are unexpected holds. | |
550 | * Therefore, we want to sync the MOS (thus syncing the dd_dbuf | |
551 | * and clearing the hold on it) before we process the sync_tasks. | |
552 | * The MOS data dirtied by the sync_tasks will be synced on the next | |
553 | * pass. | |
554 | */ | |
29809a6c | 555 | if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { |
13fe0198 | 556 | dsl_sync_task_t *dst; |
29809a6c MA |
557 | /* |
558 | * No more sync tasks should have been added while we | |
559 | * were syncing. | |
560 | */ | |
e8b96c60 MA |
561 | ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); |
562 | while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) | |
13fe0198 | 563 | dsl_sync_task_sync(dst, tx); |
29809a6c MA |
564 | } |
565 | ||
34dc7c2f | 566 | dmu_tx_commit(tx); |
b128c09f | 567 | |
e8b96c60 | 568 | DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); |
34dc7c2f BB |
569 | } |
570 | ||
571 | void | |
428870ff | 572 | dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) |
34dc7c2f | 573 | { |
29809a6c | 574 | zilog_t *zilog; |
34dc7c2f | 575 | |
29809a6c | 576 | while ((zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg))) { |
e8b96c60 | 577 | dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); |
29809a6c MA |
578 | zil_clean(zilog, txg); |
579 | ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); | |
580 | dmu_buf_rele(ds->ds_dbuf, zilog); | |
34dc7c2f | 581 | } |
428870ff | 582 | ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); |
34dc7c2f BB |
583 | } |
584 | ||
585 | /* | |
586 | * TRUE if the current thread is the tx_sync_thread or if we | |
587 | * are being called from SPA context during pool initialization. | |
588 | */ | |
589 | int | |
590 | dsl_pool_sync_context(dsl_pool_t *dp) | |
591 | { | |
592 | return (curthread == dp->dp_tx.tx_sync_thread || | |
9ae529ec | 593 | spa_is_initializing(dp->dp_spa)); |
34dc7c2f BB |
594 | } |
595 | ||
596 | uint64_t | |
597 | dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) | |
598 | { | |
599 | uint64_t space, resv; | |
600 | ||
601 | /* | |
602 | * Reserve about 1.6% (1/64), or at least 32MB, for allocation | |
603 | * efficiency. | |
604 | * XXX The intent log is not accounted for, so it must fit | |
605 | * within this slop. | |
606 | * | |
607 | * If we're trying to assess whether it's OK to do a free, | |
608 | * cut the reservation in half to allow forward progress | |
609 | * (e.g. make it possible to rm(1) files from a full pool). | |
610 | */ | |
611 | space = spa_get_dspace(dp->dp_spa); | |
612 | resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1); | |
613 | if (netfree) | |
614 | resv >>= 1; | |
615 | ||
616 | return (space - resv); | |
617 | } | |
618 | ||
e8b96c60 MA |
619 | boolean_t |
620 | dsl_pool_need_dirty_delay(dsl_pool_t *dp) | |
34dc7c2f | 621 | { |
e8b96c60 MA |
622 | uint64_t delay_min_bytes = |
623 | zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; | |
624 | boolean_t rv; | |
34dc7c2f | 625 | |
e8b96c60 MA |
626 | mutex_enter(&dp->dp_lock); |
627 | if (dp->dp_dirty_total > zfs_dirty_data_sync) | |
628 | txg_kick(dp); | |
629 | rv = (dp->dp_dirty_total > delay_min_bytes); | |
630 | mutex_exit(&dp->dp_lock); | |
631 | return (rv); | |
34dc7c2f BB |
632 | } |
633 | ||
634 | void | |
e8b96c60 | 635 | dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) |
34dc7c2f | 636 | { |
e8b96c60 MA |
637 | if (space > 0) { |
638 | mutex_enter(&dp->dp_lock); | |
639 | dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; | |
640 | dsl_pool_dirty_delta(dp, space); | |
641 | mutex_exit(&dp->dp_lock); | |
642 | } | |
34dc7c2f BB |
643 | } |
644 | ||
645 | void | |
e8b96c60 | 646 | dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) |
34dc7c2f | 647 | { |
e8b96c60 MA |
648 | ASSERT3S(space, >=, 0); |
649 | if (space == 0) | |
34dc7c2f BB |
650 | return; |
651 | ||
e8b96c60 MA |
652 | mutex_enter(&dp->dp_lock); |
653 | if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { | |
654 | /* XXX writing something we didn't dirty? */ | |
655 | space = dp->dp_dirty_pertxg[txg & TXG_MASK]; | |
34dc7c2f | 656 | } |
e8b96c60 MA |
657 | ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); |
658 | dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; | |
659 | ASSERT3U(dp->dp_dirty_total, >=, space); | |
660 | dsl_pool_dirty_delta(dp, -space); | |
661 | mutex_exit(&dp->dp_lock); | |
34dc7c2f | 662 | } |
b128c09f BB |
663 | |
664 | /* ARGSUSED */ | |
665 | static int | |
13fe0198 | 666 | upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) |
b128c09f BB |
667 | { |
668 | dmu_tx_t *tx = arg; | |
669 | dsl_dataset_t *ds, *prev = NULL; | |
670 | int err; | |
b128c09f | 671 | |
13fe0198 | 672 | err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); |
b128c09f BB |
673 | if (err) |
674 | return (err); | |
675 | ||
676 | while (ds->ds_phys->ds_prev_snap_obj != 0) { | |
677 | err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, | |
678 | FTAG, &prev); | |
679 | if (err) { | |
680 | dsl_dataset_rele(ds, FTAG); | |
681 | return (err); | |
682 | } | |
683 | ||
684 | if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) | |
685 | break; | |
686 | dsl_dataset_rele(ds, FTAG); | |
687 | ds = prev; | |
688 | prev = NULL; | |
689 | } | |
690 | ||
691 | if (prev == NULL) { | |
692 | prev = dp->dp_origin_snap; | |
693 | ||
694 | /* | |
695 | * The $ORIGIN can't have any data, or the accounting | |
696 | * will be wrong. | |
697 | */ | |
13fe0198 | 698 | ASSERT0(prev->ds_phys->ds_bp.blk_birth); |
b128c09f BB |
699 | |
700 | /* The origin doesn't get attached to itself */ | |
701 | if (ds->ds_object == prev->ds_object) { | |
702 | dsl_dataset_rele(ds, FTAG); | |
703 | return (0); | |
704 | } | |
705 | ||
706 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
707 | ds->ds_phys->ds_prev_snap_obj = prev->ds_object; | |
708 | ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; | |
709 | ||
710 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); | |
711 | ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; | |
712 | ||
713 | dmu_buf_will_dirty(prev->ds_dbuf, tx); | |
714 | prev->ds_phys->ds_num_children++; | |
715 | ||
716 | if (ds->ds_phys->ds_next_snap_obj == 0) { | |
717 | ASSERT(ds->ds_prev == NULL); | |
13fe0198 | 718 | VERIFY0(dsl_dataset_hold_obj(dp, |
b128c09f BB |
719 | ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); |
720 | } | |
721 | } | |
722 | ||
13fe0198 MA |
723 | ASSERT3U(ds->ds_dir->dd_phys->dd_origin_obj, ==, prev->ds_object); |
724 | ASSERT3U(ds->ds_phys->ds_prev_snap_obj, ==, prev->ds_object); | |
b128c09f BB |
725 | |
726 | if (prev->ds_phys->ds_next_clones_obj == 0) { | |
428870ff | 727 | dmu_buf_will_dirty(prev->ds_dbuf, tx); |
b128c09f BB |
728 | prev->ds_phys->ds_next_clones_obj = |
729 | zap_create(dp->dp_meta_objset, | |
730 | DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); | |
731 | } | |
13fe0198 | 732 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
b128c09f BB |
733 | prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); |
734 | ||
735 | dsl_dataset_rele(ds, FTAG); | |
736 | if (prev != dp->dp_origin_snap) | |
737 | dsl_dataset_rele(prev, FTAG); | |
738 | return (0); | |
739 | } | |
740 | ||
741 | void | |
742 | dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
743 | { | |
744 | ASSERT(dmu_tx_is_syncing(tx)); | |
745 | ASSERT(dp->dp_origin_snap != NULL); | |
746 | ||
13fe0198 | 747 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, |
428870ff BB |
748 | tx, DS_FIND_CHILDREN)); |
749 | } | |
750 | ||
751 | /* ARGSUSED */ | |
752 | static int | |
13fe0198 | 753 | upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
428870ff BB |
754 | { |
755 | dmu_tx_t *tx = arg; | |
428870ff BB |
756 | objset_t *mos = dp->dp_meta_objset; |
757 | ||
13fe0198 | 758 | if (ds->ds_dir->dd_phys->dd_origin_obj != 0) { |
428870ff BB |
759 | dsl_dataset_t *origin; |
760 | ||
13fe0198 | 761 | VERIFY0(dsl_dataset_hold_obj(dp, |
428870ff BB |
762 | ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin)); |
763 | ||
764 | if (origin->ds_dir->dd_phys->dd_clones == 0) { | |
765 | dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); | |
766 | origin->ds_dir->dd_phys->dd_clones = zap_create(mos, | |
767 | DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); | |
768 | } | |
769 | ||
13fe0198 MA |
770 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
771 | origin->ds_dir->dd_phys->dd_clones, ds->ds_object, tx)); | |
428870ff BB |
772 | |
773 | dsl_dataset_rele(origin, FTAG); | |
774 | } | |
428870ff BB |
775 | return (0); |
776 | } | |
777 | ||
778 | void | |
779 | dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
780 | { | |
428870ff BB |
781 | uint64_t obj; |
782 | ||
d6320ddb BB |
783 | ASSERT(dmu_tx_is_syncing(tx)); |
784 | ||
428870ff | 785 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); |
13fe0198 | 786 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
787 | FREE_DIR_NAME, &dp->dp_free_dir)); |
788 | ||
789 | /* | |
790 | * We can't use bpobj_alloc(), because spa_version() still | |
791 | * returns the old version, and we need a new-version bpobj with | |
792 | * subobj support. So call dmu_object_alloc() directly. | |
793 | */ | |
794 | obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, | |
795 | SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); | |
13fe0198 | 796 | VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 797 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); |
13fe0198 | 798 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); |
428870ff | 799 | |
13fe0198 | 800 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
428870ff | 801 | upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); |
b128c09f BB |
802 | } |
803 | ||
804 | void | |
805 | dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) | |
806 | { | |
807 | uint64_t dsobj; | |
808 | dsl_dataset_t *ds; | |
809 | ||
810 | ASSERT(dmu_tx_is_syncing(tx)); | |
811 | ASSERT(dp->dp_origin_snap == NULL); | |
13fe0198 | 812 | ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); |
b128c09f BB |
813 | |
814 | /* create the origin dir, ds, & snap-ds */ | |
b128c09f BB |
815 | dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, |
816 | NULL, 0, kcred, tx); | |
13fe0198 MA |
817 | VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); |
818 | dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); | |
819 | VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, | |
b128c09f BB |
820 | dp, &dp->dp_origin_snap)); |
821 | dsl_dataset_rele(ds, FTAG); | |
b128c09f | 822 | } |
9babb374 BB |
823 | |
824 | taskq_t * | |
3558fd73 | 825 | dsl_pool_iput_taskq(dsl_pool_t *dp) |
9babb374 | 826 | { |
3558fd73 | 827 | return (dp->dp_iput_taskq); |
9babb374 | 828 | } |
428870ff BB |
829 | |
830 | /* | |
831 | * Walk through the pool-wide zap object of temporary snapshot user holds | |
832 | * and release them. | |
833 | */ | |
834 | void | |
835 | dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) | |
836 | { | |
837 | zap_attribute_t za; | |
838 | zap_cursor_t zc; | |
839 | objset_t *mos = dp->dp_meta_objset; | |
840 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
95fd54a1 | 841 | nvlist_t *holds; |
428870ff BB |
842 | |
843 | if (zapobj == 0) | |
844 | return; | |
845 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
846 | ||
95fd54a1 SH |
847 | holds = fnvlist_alloc(); |
848 | ||
428870ff BB |
849 | for (zap_cursor_init(&zc, mos, zapobj); |
850 | zap_cursor_retrieve(&zc, &za) == 0; | |
851 | zap_cursor_advance(&zc)) { | |
852 | char *htag; | |
95fd54a1 | 853 | nvlist_t *tags; |
428870ff BB |
854 | |
855 | htag = strchr(za.za_name, '-'); | |
856 | *htag = '\0'; | |
857 | ++htag; | |
95fd54a1 SH |
858 | if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { |
859 | tags = fnvlist_alloc(); | |
860 | fnvlist_add_boolean(tags, htag); | |
861 | fnvlist_add_nvlist(holds, za.za_name, tags); | |
862 | fnvlist_free(tags); | |
863 | } else { | |
864 | fnvlist_add_boolean(tags, htag); | |
865 | } | |
428870ff | 866 | } |
95fd54a1 SH |
867 | dsl_dataset_user_release_tmp(dp, holds); |
868 | fnvlist_free(holds); | |
428870ff BB |
869 | zap_cursor_fini(&zc); |
870 | } | |
871 | ||
872 | /* | |
873 | * Create the pool-wide zap object for storing temporary snapshot holds. | |
874 | */ | |
875 | void | |
876 | dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) | |
877 | { | |
878 | objset_t *mos = dp->dp_meta_objset; | |
879 | ||
880 | ASSERT(dp->dp_tmp_userrefs_obj == 0); | |
881 | ASSERT(dmu_tx_is_syncing(tx)); | |
882 | ||
9ae529ec CS |
883 | dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, |
884 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); | |
428870ff BB |
885 | } |
886 | ||
887 | static int | |
888 | dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, | |
13fe0198 | 889 | const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) |
428870ff BB |
890 | { |
891 | objset_t *mos = dp->dp_meta_objset; | |
892 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
893 | char *name; | |
894 | int error; | |
895 | ||
896 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
897 | ASSERT(dmu_tx_is_syncing(tx)); | |
898 | ||
899 | /* | |
900 | * If the pool was created prior to SPA_VERSION_USERREFS, the | |
901 | * zap object for temporary holds might not exist yet. | |
902 | */ | |
903 | if (zapobj == 0) { | |
904 | if (holding) { | |
905 | dsl_pool_user_hold_create_obj(dp, tx); | |
906 | zapobj = dp->dp_tmp_userrefs_obj; | |
907 | } else { | |
2e528b49 | 908 | return (SET_ERROR(ENOENT)); |
428870ff BB |
909 | } |
910 | } | |
911 | ||
912 | name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); | |
913 | if (holding) | |
13fe0198 | 914 | error = zap_add(mos, zapobj, name, 8, 1, &now, tx); |
428870ff BB |
915 | else |
916 | error = zap_remove(mos, zapobj, name, tx); | |
917 | strfree(name); | |
918 | ||
919 | return (error); | |
920 | } | |
921 | ||
922 | /* | |
923 | * Add a temporary hold for the given dataset object and tag. | |
924 | */ | |
925 | int | |
926 | dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
13fe0198 | 927 | uint64_t now, dmu_tx_t *tx) |
428870ff BB |
928 | { |
929 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); | |
930 | } | |
931 | ||
932 | /* | |
933 | * Release a temporary hold for the given dataset object and tag. | |
934 | */ | |
935 | int | |
936 | dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
937 | dmu_tx_t *tx) | |
938 | { | |
13fe0198 | 939 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, |
428870ff BB |
940 | tx, B_FALSE)); |
941 | } | |
c409e464 | 942 | |
13fe0198 MA |
943 | /* |
944 | * DSL Pool Configuration Lock | |
945 | * | |
946 | * The dp_config_rwlock protects against changes to DSL state (e.g. dataset | |
947 | * creation / destruction / rename / property setting). It must be held for | |
948 | * read to hold a dataset or dsl_dir. I.e. you must call | |
949 | * dsl_pool_config_enter() or dsl_pool_hold() before calling | |
950 | * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock | |
951 | * must be held continuously until all datasets and dsl_dirs are released. | |
952 | * | |
953 | * The only exception to this rule is that if a "long hold" is placed on | |
954 | * a dataset, then the dp_config_rwlock may be dropped while the dataset | |
955 | * is still held. The long hold will prevent the dataset from being | |
956 | * destroyed -- the destroy will fail with EBUSY. A long hold can be | |
957 | * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset | |
958 | * (by calling dsl_{dataset,objset}_{try}own{_obj}). | |
959 | * | |
960 | * Legitimate long-holders (including owners) should be long-running, cancelable | |
961 | * tasks that should cause "zfs destroy" to fail. This includes DMU | |
962 | * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), | |
963 | * "zfs send", and "zfs diff". There are several other long-holders whose | |
964 | * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). | |
965 | * | |
966 | * The usual formula for long-holding would be: | |
967 | * dsl_pool_hold() | |
968 | * dsl_dataset_hold() | |
969 | * ... perform checks ... | |
970 | * dsl_dataset_long_hold() | |
971 | * dsl_pool_rele() | |
972 | * ... perform long-running task ... | |
973 | * dsl_dataset_long_rele() | |
974 | * dsl_dataset_rele() | |
975 | * | |
976 | * Note that when the long hold is released, the dataset is still held but | |
977 | * the pool is not held. The dataset may change arbitrarily during this time | |
978 | * (e.g. it could be destroyed). Therefore you shouldn't do anything to the | |
979 | * dataset except release it. | |
980 | * | |
981 | * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only | |
982 | * or modifying operations. | |
983 | * | |
984 | * Modifying operations should generally use dsl_sync_task(). The synctask | |
985 | * infrastructure enforces proper locking strategy with respect to the | |
986 | * dp_config_rwlock. See the comment above dsl_sync_task() for details. | |
987 | * | |
988 | * Read-only operations will manually hold the pool, then the dataset, obtain | |
989 | * information from the dataset, then release the pool and dataset. | |
990 | * dmu_objset_{hold,rele}() are convenience routines that also do the pool | |
991 | * hold/rele. | |
992 | */ | |
993 | ||
994 | int | |
995 | dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) | |
996 | { | |
997 | spa_t *spa; | |
998 | int error; | |
999 | ||
1000 | error = spa_open(name, &spa, tag); | |
1001 | if (error == 0) { | |
1002 | *dp = spa_get_dsl(spa); | |
1003 | dsl_pool_config_enter(*dp, tag); | |
1004 | } | |
1005 | return (error); | |
1006 | } | |
1007 | ||
1008 | void | |
1009 | dsl_pool_rele(dsl_pool_t *dp, void *tag) | |
1010 | { | |
1011 | dsl_pool_config_exit(dp, tag); | |
1012 | spa_close(dp->dp_spa, tag); | |
1013 | } | |
1014 | ||
1015 | void | |
1016 | dsl_pool_config_enter(dsl_pool_t *dp, void *tag) | |
1017 | { | |
1018 | /* | |
1019 | * We use a "reentrant" reader-writer lock, but not reentrantly. | |
1020 | * | |
1021 | * The rrwlock can (with the track_all flag) track all reading threads, | |
1022 | * which is very useful for debugging which code path failed to release | |
1023 | * the lock, and for verifying that the *current* thread does hold | |
1024 | * the lock. | |
1025 | * | |
1026 | * (Unlike a rwlock, which knows that N threads hold it for | |
1027 | * read, but not *which* threads, so rw_held(RW_READER) returns TRUE | |
1028 | * if any thread holds it for read, even if this thread doesn't). | |
1029 | */ | |
1030 | ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); | |
1031 | rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); | |
1032 | } | |
1033 | ||
1034 | void | |
1035 | dsl_pool_config_exit(dsl_pool_t *dp, void *tag) | |
1036 | { | |
1037 | rrw_exit(&dp->dp_config_rwlock, tag); | |
1038 | } | |
1039 | ||
1040 | boolean_t | |
1041 | dsl_pool_config_held(dsl_pool_t *dp) | |
1042 | { | |
1043 | return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); | |
1044 | } | |
1045 | ||
c409e464 | 1046 | #if defined(_KERNEL) && defined(HAVE_SPL) |
40a806df NB |
1047 | EXPORT_SYMBOL(dsl_pool_config_enter); |
1048 | EXPORT_SYMBOL(dsl_pool_config_exit); | |
1049 | ||
d1d7e268 | 1050 | /* zfs_dirty_data_max_percent only applied at module load in arc_init(). */ |
e8b96c60 MA |
1051 | module_param(zfs_dirty_data_max_percent, int, 0444); |
1052 | MODULE_PARM_DESC(zfs_dirty_data_max_percent, "percent of ram can be dirty"); | |
c409e464 | 1053 | |
d1d7e268 | 1054 | /* zfs_dirty_data_max_max_percent only applied at module load in arc_init(). */ |
e8b96c60 MA |
1055 | module_param(zfs_dirty_data_max_max_percent, int, 0444); |
1056 | MODULE_PARM_DESC(zfs_dirty_data_max_max_percent, | |
d1d7e268 | 1057 | "zfs_dirty_data_max upper bound as % of RAM"); |
c409e464 | 1058 | |
e8b96c60 MA |
1059 | module_param(zfs_delay_min_dirty_percent, int, 0644); |
1060 | MODULE_PARM_DESC(zfs_delay_min_dirty_percent, "transaction delay threshold"); | |
c409e464 | 1061 | |
e8b96c60 MA |
1062 | module_param(zfs_dirty_data_max, ulong, 0644); |
1063 | MODULE_PARM_DESC(zfs_dirty_data_max, "determines the dirty space limit"); | |
c409e464 | 1064 | |
d1d7e268 | 1065 | /* zfs_dirty_data_max_max only applied at module load in arc_init(). */ |
e8b96c60 MA |
1066 | module_param(zfs_dirty_data_max_max, ulong, 0444); |
1067 | MODULE_PARM_DESC(zfs_dirty_data_max_max, | |
d1d7e268 | 1068 | "zfs_dirty_data_max upper bound in bytes"); |
c409e464 | 1069 | |
e8b96c60 MA |
1070 | module_param(zfs_dirty_data_sync, ulong, 0644); |
1071 | MODULE_PARM_DESC(zfs_dirty_data_sync, "sync txg when this much dirty data"); | |
c409e464 | 1072 | |
e8b96c60 MA |
1073 | module_param(zfs_delay_scale, ulong, 0644); |
1074 | MODULE_PARM_DESC(zfs_delay_scale, "how quickly delay approaches infinity"); | |
c409e464 | 1075 | #endif |