]>
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. |
64fc7762 | 23 | * Copyright (c) 2011, 2017 by Delphix. All rights reserved. |
95fd54a1 | 24 | * Copyright (c) 2013 Steven Hartland. All rights reserved. |
0c66c32d | 25 | * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved. |
539d33c7 | 26 | * Copyright 2016 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
27 | */ |
28 | ||
34dc7c2f BB |
29 | #include <sys/dsl_pool.h> |
30 | #include <sys/dsl_dataset.h> | |
428870ff | 31 | #include <sys/dsl_prop.h> |
34dc7c2f BB |
32 | #include <sys/dsl_dir.h> |
33 | #include <sys/dsl_synctask.h> | |
428870ff BB |
34 | #include <sys/dsl_scan.h> |
35 | #include <sys/dnode.h> | |
34dc7c2f BB |
36 | #include <sys/dmu_tx.h> |
37 | #include <sys/dmu_objset.h> | |
38 | #include <sys/arc.h> | |
39 | #include <sys/zap.h> | |
40 | #include <sys/zio.h> | |
41 | #include <sys/zfs_context.h> | |
42 | #include <sys/fs/zfs.h> | |
b128c09f BB |
43 | #include <sys/zfs_znode.h> |
44 | #include <sys/spa_impl.h> | |
428870ff | 45 | #include <sys/dsl_deadlist.h> |
9ae529ec CS |
46 | #include <sys/bptree.h> |
47 | #include <sys/zfeature.h> | |
29809a6c | 48 | #include <sys/zil_impl.h> |
13fe0198 | 49 | #include <sys/dsl_userhold.h> |
49ee64e5 | 50 | #include <sys/trace_txg.h> |
379ca9cf | 51 | #include <sys/mmp.h> |
34dc7c2f | 52 | |
e8b96c60 MA |
53 | /* |
54 | * ZFS Write Throttle | |
55 | * ------------------ | |
56 | * | |
57 | * ZFS must limit the rate of incoming writes to the rate at which it is able | |
58 | * to sync data modifications to the backend storage. Throttling by too much | |
59 | * creates an artificial limit; throttling by too little can only be sustained | |
60 | * for short periods and would lead to highly lumpy performance. On a per-pool | |
61 | * basis, ZFS tracks the amount of modified (dirty) data. As operations change | |
62 | * data, the amount of dirty data increases; as ZFS syncs out data, the amount | |
63 | * of dirty data decreases. When the amount of dirty data exceeds a | |
64 | * predetermined threshold further modifications are blocked until the amount | |
65 | * of dirty data decreases (as data is synced out). | |
66 | * | |
67 | * The limit on dirty data is tunable, and should be adjusted according to | |
68 | * both the IO capacity and available memory of the system. The larger the | |
69 | * window, the more ZFS is able to aggregate and amortize metadata (and data) | |
70 | * changes. However, memory is a limited resource, and allowing for more dirty | |
71 | * data comes at the cost of keeping other useful data in memory (for example | |
72 | * ZFS data cached by the ARC). | |
73 | * | |
74 | * Implementation | |
75 | * | |
76 | * As buffers are modified dsl_pool_willuse_space() increments both the per- | |
77 | * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of | |
78 | * dirty space used; dsl_pool_dirty_space() decrements those values as data | |
79 | * is synced out from dsl_pool_sync(). While only the poolwide value is | |
80 | * relevant, the per-txg value is useful for debugging. The tunable | |
81 | * zfs_dirty_data_max determines the dirty space limit. Once that value is | |
82 | * exceeded, new writes are halted until space frees up. | |
83 | * | |
84 | * The zfs_dirty_data_sync tunable dictates the threshold at which we | |
85 | * ensure that there is a txg syncing (see the comment in txg.c for a full | |
86 | * description of transaction group stages). | |
87 | * | |
88 | * The IO scheduler uses both the dirty space limit and current amount of | |
89 | * dirty data as inputs. Those values affect the number of concurrent IOs ZFS | |
90 | * issues. See the comment in vdev_queue.c for details of the IO scheduler. | |
91 | * | |
92 | * The delay is also calculated based on the amount of dirty data. See the | |
93 | * comment above dmu_tx_delay() for details. | |
94 | */ | |
95 | ||
96 | /* | |
97 | * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, | |
98 | * capped at zfs_dirty_data_max_max. It can also be overridden with a module | |
99 | * parameter. | |
100 | */ | |
101 | unsigned long zfs_dirty_data_max = 0; | |
102 | unsigned long zfs_dirty_data_max_max = 0; | |
103 | int zfs_dirty_data_max_percent = 10; | |
104 | int zfs_dirty_data_max_max_percent = 25; | |
b128c09f | 105 | |
e8b96c60 MA |
106 | /* |
107 | * If there is at least this much dirty data, push out a txg. | |
108 | */ | |
109 | unsigned long zfs_dirty_data_sync = 64 * 1024 * 1024; | |
34dc7c2f | 110 | |
e8b96c60 MA |
111 | /* |
112 | * Once there is this amount of dirty data, the dmu_tx_delay() will kick in | |
113 | * and delay each transaction. | |
114 | * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. | |
115 | */ | |
116 | int zfs_delay_min_dirty_percent = 60; | |
b128c09f | 117 | |
e8b96c60 MA |
118 | /* |
119 | * This controls how quickly the delay approaches infinity. | |
120 | * Larger values cause it to delay more for a given amount of dirty data. | |
121 | * Therefore larger values will cause there to be less dirty data for a | |
122 | * given throughput. | |
123 | * | |
124 | * For the smoothest delay, this value should be about 1 billion divided | |
125 | * by the maximum number of operations per second. This will smoothly | |
126 | * handle between 10x and 1/10th this number. | |
127 | * | |
128 | * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the | |
129 | * multiply in dmu_tx_delay(). | |
130 | */ | |
131 | unsigned long zfs_delay_scale = 1000 * 1000 * 1000 / 2000; | |
b128c09f | 132 | |
64fc7762 MA |
133 | /* |
134 | * This determines the number of threads used by the dp_sync_taskq. | |
135 | */ | |
136 | int zfs_sync_taskq_batch_pct = 75; | |
137 | ||
a032ac4b BB |
138 | /* |
139 | * These tunables determine the behavior of how zil_itxg_clean() is | |
140 | * called via zil_clean() in the context of spa_sync(). When an itxg | |
141 | * list needs to be cleaned, TQ_NOSLEEP will be used when dispatching. | |
142 | * If the dispatch fails, the call to zil_itxg_clean() will occur | |
143 | * synchronously in the context of spa_sync(), which can negatively | |
144 | * impact the performance of spa_sync() (e.g. in the case of the itxg | |
145 | * list having a large number of itxs that needs to be cleaned). | |
146 | * | |
147 | * Thus, these tunables can be used to manipulate the behavior of the | |
148 | * taskq used by zil_clean(); they determine the number of taskq entries | |
149 | * that are pre-populated when the taskq is first created (via the | |
150 | * "zfs_zil_clean_taskq_minalloc" tunable) and the maximum number of | |
151 | * taskq entries that are cached after an on-demand allocation (via the | |
152 | * "zfs_zil_clean_taskq_maxalloc"). | |
153 | * | |
154 | * The idea being, we want to try reasonably hard to ensure there will | |
155 | * already be a taskq entry pre-allocated by the time that it is needed | |
156 | * by zil_clean(). This way, we can avoid the possibility of an | |
157 | * on-demand allocation of a new taskq entry from failing, which would | |
158 | * result in zil_itxg_clean() being called synchronously from zil_clean() | |
159 | * (which can adversely affect performance of spa_sync()). | |
160 | * | |
161 | * Additionally, the number of threads used by the taskq can be | |
162 | * configured via the "zfs_zil_clean_taskq_nthr_pct" tunable. | |
163 | */ | |
164 | int zfs_zil_clean_taskq_nthr_pct = 100; | |
165 | int zfs_zil_clean_taskq_minalloc = 1024; | |
166 | int zfs_zil_clean_taskq_maxalloc = 1024 * 1024; | |
167 | ||
428870ff | 168 | int |
b128c09f | 169 | dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) |
34dc7c2f BB |
170 | { |
171 | uint64_t obj; | |
172 | int err; | |
173 | ||
174 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 175 | dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, |
b128c09f | 176 | name, sizeof (obj), 1, &obj); |
34dc7c2f BB |
177 | if (err) |
178 | return (err); | |
179 | ||
13fe0198 | 180 | return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); |
34dc7c2f BB |
181 | } |
182 | ||
183 | static dsl_pool_t * | |
184 | dsl_pool_open_impl(spa_t *spa, uint64_t txg) | |
185 | { | |
186 | dsl_pool_t *dp; | |
187 | blkptr_t *bp = spa_get_rootblkptr(spa); | |
34dc7c2f BB |
188 | |
189 | dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); | |
190 | dp->dp_spa = spa; | |
191 | dp->dp_meta_rootbp = *bp; | |
13fe0198 | 192 | rrw_init(&dp->dp_config_rwlock, B_TRUE); |
34dc7c2f | 193 | txg_init(dp, txg); |
379ca9cf | 194 | mmp_init(spa); |
34dc7c2f | 195 | |
4747a7d3 | 196 | txg_list_create(&dp->dp_dirty_datasets, spa, |
34dc7c2f | 197 | offsetof(dsl_dataset_t, ds_dirty_link)); |
4747a7d3 | 198 | txg_list_create(&dp->dp_dirty_zilogs, spa, |
29809a6c | 199 | offsetof(zilog_t, zl_dirty_link)); |
4747a7d3 | 200 | txg_list_create(&dp->dp_dirty_dirs, spa, |
34dc7c2f | 201 | offsetof(dsl_dir_t, dd_dirty_link)); |
4747a7d3 | 202 | txg_list_create(&dp->dp_sync_tasks, spa, |
13fe0198 | 203 | offsetof(dsl_sync_task_t, dst_node)); |
34dc7c2f | 204 | |
64fc7762 MA |
205 | dp->dp_sync_taskq = taskq_create("dp_sync_taskq", |
206 | zfs_sync_taskq_batch_pct, minclsyspri, 1, INT_MAX, | |
207 | TASKQ_THREADS_CPU_PCT); | |
208 | ||
a032ac4b BB |
209 | dp->dp_zil_clean_taskq = taskq_create("dp_zil_clean_taskq", |
210 | zfs_zil_clean_taskq_nthr_pct, minclsyspri, | |
211 | zfs_zil_clean_taskq_minalloc, | |
212 | zfs_zil_clean_taskq_maxalloc, | |
213 | TASKQ_PREPOPULATE | TASKQ_THREADS_CPU_PCT); | |
214 | ||
34dc7c2f | 215 | mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); |
e8b96c60 | 216 | cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); |
34dc7c2f | 217 | |
1229323d | 218 | dp->dp_iput_taskq = taskq_create("z_iput", max_ncpus, defclsyspri, |
aa9af22c | 219 | max_ncpus * 8, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); |
9babb374 | 220 | |
34dc7c2f BB |
221 | return (dp); |
222 | } | |
223 | ||
224 | int | |
9ae529ec | 225 | dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) |
34dc7c2f BB |
226 | { |
227 | int err; | |
228 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
9ae529ec | 229 | |
b7faa7aa G |
230 | /* |
231 | * Initialize the caller's dsl_pool_t structure before we actually open | |
232 | * the meta objset. This is done because a self-healing write zio may | |
233 | * be issued as part of dmu_objset_open_impl() and the spa needs its | |
234 | * dsl_pool_t initialized in order to handle the write. | |
235 | */ | |
236 | *dpp = dp; | |
237 | ||
9ae529ec CS |
238 | err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, |
239 | &dp->dp_meta_objset); | |
b7faa7aa | 240 | if (err != 0) { |
9ae529ec | 241 | dsl_pool_close(dp); |
b7faa7aa G |
242 | *dpp = NULL; |
243 | } | |
9ae529ec CS |
244 | |
245 | return (err); | |
246 | } | |
247 | ||
248 | int | |
249 | dsl_pool_open(dsl_pool_t *dp) | |
250 | { | |
251 | int err; | |
b128c09f BB |
252 | dsl_dir_t *dd; |
253 | dsl_dataset_t *ds; | |
428870ff | 254 | uint64_t obj; |
34dc7c2f | 255 | |
13fe0198 | 256 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
34dc7c2f BB |
257 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
258 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, | |
259 | &dp->dp_root_dir_obj); | |
260 | if (err) | |
261 | goto out; | |
262 | ||
13fe0198 | 263 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
264 | NULL, dp, &dp->dp_root_dir); |
265 | if (err) | |
266 | goto out; | |
267 | ||
b128c09f | 268 | err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); |
34dc7c2f BB |
269 | if (err) |
270 | goto out; | |
271 | ||
9ae529ec | 272 | if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { |
b128c09f BB |
273 | err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); |
274 | if (err) | |
275 | goto out; | |
d683ddbb JG |
276 | err = dsl_dataset_hold_obj(dp, |
277 | dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); | |
9babb374 BB |
278 | if (err == 0) { |
279 | err = dsl_dataset_hold_obj(dp, | |
d683ddbb | 280 | dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, |
9babb374 BB |
281 | &dp->dp_origin_snap); |
282 | dsl_dataset_rele(ds, FTAG); | |
283 | } | |
13fe0198 | 284 | dsl_dir_rele(dd, dp); |
b128c09f BB |
285 | if (err) |
286 | goto out; | |
b128c09f BB |
287 | } |
288 | ||
9ae529ec | 289 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { |
428870ff BB |
290 | err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, |
291 | &dp->dp_free_dir); | |
b128c09f BB |
292 | if (err) |
293 | goto out; | |
428870ff | 294 | |
b128c09f | 295 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 296 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); |
b128c09f BB |
297 | if (err) |
298 | goto out; | |
13fe0198 | 299 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff | 300 | dp->dp_meta_objset, obj)); |
b128c09f BB |
301 | } |
302 | ||
a1d477c2 MA |
303 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS)) { |
304 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
305 | DMU_POOL_OBSOLETE_BPOBJ, sizeof (uint64_t), 1, &obj); | |
306 | if (err == 0) { | |
307 | VERIFY0(bpobj_open(&dp->dp_obsolete_bpobj, | |
308 | dp->dp_meta_objset, obj)); | |
309 | } else if (err == ENOENT) { | |
310 | /* | |
311 | * We might not have created the remap bpobj yet. | |
312 | */ | |
313 | err = 0; | |
314 | } else { | |
315 | goto out; | |
316 | } | |
317 | } | |
318 | ||
fbeddd60 | 319 | /* |
a1d477c2 MA |
320 | * Note: errors ignored, because the these special dirs, used for |
321 | * space accounting, are only created on demand. | |
fbeddd60 MA |
322 | */ |
323 | (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, | |
324 | &dp->dp_leak_dir); | |
325 | ||
fa86b5db | 326 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { |
9ae529ec CS |
327 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
328 | DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, | |
329 | &dp->dp_bptree_obj); | |
330 | if (err != 0) | |
331 | goto out; | |
332 | } | |
333 | ||
fa86b5db | 334 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { |
753c3839 MA |
335 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
336 | DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, | |
337 | &dp->dp_empty_bpobj); | |
338 | if (err != 0) | |
339 | goto out; | |
340 | } | |
341 | ||
428870ff BB |
342 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
343 | DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, | |
344 | &dp->dp_tmp_userrefs_obj); | |
345 | if (err == ENOENT) | |
346 | err = 0; | |
347 | if (err) | |
348 | goto out; | |
349 | ||
9ae529ec | 350 | err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); |
428870ff | 351 | |
34dc7c2f | 352 | out: |
13fe0198 | 353 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
34dc7c2f BB |
354 | return (err); |
355 | } | |
356 | ||
357 | void | |
358 | dsl_pool_close(dsl_pool_t *dp) | |
359 | { | |
b128c09f | 360 | /* |
e8b96c60 MA |
361 | * Drop our references from dsl_pool_open(). |
362 | * | |
b128c09f BB |
363 | * Since we held the origin_snap from "syncing" context (which |
364 | * includes pool-opening context), it actually only got a "ref" | |
365 | * and not a hold, so just drop that here. | |
366 | */ | |
a1d477c2 | 367 | if (dp->dp_origin_snap != NULL) |
13fe0198 | 368 | dsl_dataset_rele(dp->dp_origin_snap, dp); |
a1d477c2 | 369 | if (dp->dp_mos_dir != NULL) |
13fe0198 | 370 | dsl_dir_rele(dp->dp_mos_dir, dp); |
a1d477c2 | 371 | if (dp->dp_free_dir != NULL) |
13fe0198 | 372 | dsl_dir_rele(dp->dp_free_dir, dp); |
a1d477c2 | 373 | if (dp->dp_leak_dir != NULL) |
fbeddd60 | 374 | dsl_dir_rele(dp->dp_leak_dir, dp); |
a1d477c2 | 375 | if (dp->dp_root_dir != NULL) |
13fe0198 | 376 | dsl_dir_rele(dp->dp_root_dir, dp); |
34dc7c2f | 377 | |
428870ff | 378 | bpobj_close(&dp->dp_free_bpobj); |
a1d477c2 | 379 | bpobj_close(&dp->dp_obsolete_bpobj); |
428870ff | 380 | |
34dc7c2f | 381 | /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ |
a1d477c2 | 382 | if (dp->dp_meta_objset != NULL) |
428870ff | 383 | dmu_objset_evict(dp->dp_meta_objset); |
34dc7c2f BB |
384 | |
385 | txg_list_destroy(&dp->dp_dirty_datasets); | |
29809a6c | 386 | txg_list_destroy(&dp->dp_dirty_zilogs); |
428870ff | 387 | txg_list_destroy(&dp->dp_sync_tasks); |
34dc7c2f | 388 | txg_list_destroy(&dp->dp_dirty_dirs); |
34dc7c2f | 389 | |
a032ac4b | 390 | taskq_destroy(dp->dp_zil_clean_taskq); |
64fc7762 MA |
391 | taskq_destroy(dp->dp_sync_taskq); |
392 | ||
ca0bf58d PS |
393 | /* |
394 | * We can't set retry to TRUE since we're explicitly specifying | |
395 | * a spa to flush. This is good enough; any missed buffers for | |
396 | * this spa won't cause trouble, and they'll eventually fall | |
397 | * out of the ARC just like any other unused buffer. | |
398 | */ | |
399 | arc_flush(dp->dp_spa, FALSE); | |
400 | ||
379ca9cf | 401 | mmp_fini(dp->dp_spa); |
34dc7c2f | 402 | txg_fini(dp); |
428870ff | 403 | dsl_scan_fini(dp); |
0c66c32d JG |
404 | dmu_buf_user_evict_wait(); |
405 | ||
13fe0198 | 406 | rrw_destroy(&dp->dp_config_rwlock); |
34dc7c2f | 407 | mutex_destroy(&dp->dp_lock); |
c17486b2 | 408 | cv_destroy(&dp->dp_spaceavail_cv); |
3558fd73 | 409 | taskq_destroy(dp->dp_iput_taskq); |
a1d477c2 | 410 | if (dp->dp_blkstats != NULL) { |
d4a72f23 | 411 | mutex_destroy(&dp->dp_blkstats->zab_lock); |
79c76d5b | 412 | vmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); |
d4a72f23 | 413 | } |
34dc7c2f BB |
414 | kmem_free(dp, sizeof (dsl_pool_t)); |
415 | } | |
416 | ||
a1d477c2 MA |
417 | void |
418 | dsl_pool_create_obsolete_bpobj(dsl_pool_t *dp, dmu_tx_t *tx) | |
419 | { | |
420 | uint64_t obj; | |
421 | /* | |
422 | * Currently, we only create the obsolete_bpobj where there are | |
423 | * indirect vdevs with referenced mappings. | |
424 | */ | |
425 | ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_DEVICE_REMOVAL)); | |
426 | /* create and open the obsolete_bpobj */ | |
427 | obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); | |
428 | VERIFY0(bpobj_open(&dp->dp_obsolete_bpobj, dp->dp_meta_objset, obj)); | |
429 | VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
430 | DMU_POOL_OBSOLETE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); | |
431 | spa_feature_incr(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); | |
432 | } | |
433 | ||
434 | void | |
435 | dsl_pool_destroy_obsolete_bpobj(dsl_pool_t *dp, dmu_tx_t *tx) | |
436 | { | |
437 | spa_feature_decr(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx); | |
438 | VERIFY0(zap_remove(dp->dp_meta_objset, | |
439 | DMU_POOL_DIRECTORY_OBJECT, | |
440 | DMU_POOL_OBSOLETE_BPOBJ, tx)); | |
441 | bpobj_free(dp->dp_meta_objset, | |
442 | dp->dp_obsolete_bpobj.bpo_object, tx); | |
443 | bpobj_close(&dp->dp_obsolete_bpobj); | |
444 | } | |
445 | ||
34dc7c2f | 446 | dsl_pool_t * |
b5256303 TC |
447 | dsl_pool_create(spa_t *spa, nvlist_t *zplprops, dsl_crypto_params_t *dcp, |
448 | uint64_t txg) | |
34dc7c2f BB |
449 | { |
450 | int err; | |
451 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
452 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
b128c09f | 453 | dsl_dataset_t *ds; |
428870ff | 454 | uint64_t obj; |
b128c09f | 455 | |
13fe0198 MA |
456 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
457 | ||
b128c09f | 458 | /* create and open the MOS (meta-objset) */ |
428870ff BB |
459 | dp->dp_meta_objset = dmu_objset_create_impl(spa, |
460 | NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); | |
b5256303 | 461 | spa->spa_meta_objset = dp->dp_meta_objset; |
34dc7c2f BB |
462 | |
463 | /* create the pool directory */ | |
464 | err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
465 | DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); | |
c99c9001 | 466 | ASSERT0(err); |
34dc7c2f | 467 | |
428870ff | 468 | /* Initialize scan structures */ |
13fe0198 | 469 | VERIFY0(dsl_scan_init(dp, txg)); |
428870ff | 470 | |
34dc7c2f | 471 | /* create and open the root dir */ |
b128c09f | 472 | dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); |
13fe0198 | 473 | VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
474 | NULL, dp, &dp->dp_root_dir)); |
475 | ||
476 | /* create and open the meta-objset dir */ | |
b128c09f | 477 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); |
13fe0198 | 478 | VERIFY0(dsl_pool_open_special_dir(dp, |
b128c09f BB |
479 | MOS_DIR_NAME, &dp->dp_mos_dir)); |
480 | ||
428870ff BB |
481 | if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { |
482 | /* create and open the free dir */ | |
483 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, | |
484 | FREE_DIR_NAME, tx); | |
13fe0198 | 485 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
486 | FREE_DIR_NAME, &dp->dp_free_dir)); |
487 | ||
488 | /* create and open the free_bplist */ | |
f1512ee6 | 489 | obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); |
428870ff BB |
490 | VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
491 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); | |
13fe0198 | 492 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff BB |
493 | dp->dp_meta_objset, obj)); |
494 | } | |
495 | ||
b128c09f BB |
496 | if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) |
497 | dsl_pool_create_origin(dp, tx); | |
498 | ||
b5256303 TC |
499 | /* |
500 | * Some features may be needed when creating the root dataset, so we | |
501 | * create the feature objects here. | |
502 | */ | |
503 | if (spa_version(spa) >= SPA_VERSION_FEATURES) | |
504 | spa_feature_create_zap_objects(spa, tx); | |
505 | ||
506 | if (dcp != NULL && dcp->cp_crypt != ZIO_CRYPT_OFF && | |
507 | dcp->cp_crypt != ZIO_CRYPT_INHERIT) | |
508 | spa_feature_enable(spa, SPA_FEATURE_ENCRYPTION, tx); | |
509 | ||
b128c09f | 510 | /* create the root dataset */ |
b5256303 | 511 | obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, dcp, 0, tx); |
b128c09f BB |
512 | |
513 | /* create the root objset */ | |
13fe0198 | 514 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); |
b128c09f | 515 | #ifdef _KERNEL |
d8fdfc2d BB |
516 | { |
517 | objset_t *os; | |
518 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); | |
519 | os = dmu_objset_create_impl(dp->dp_spa, ds, | |
520 | dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx); | |
521 | rrw_exit(&ds->ds_bp_rwlock, FTAG); | |
522 | zfs_create_fs(os, kcred, zplprops, tx); | |
523 | } | |
b128c09f BB |
524 | #endif |
525 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
526 | |
527 | dmu_tx_commit(tx); | |
528 | ||
13fe0198 MA |
529 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
530 | ||
34dc7c2f BB |
531 | return (dp); |
532 | } | |
533 | ||
29809a6c MA |
534 | /* |
535 | * Account for the meta-objset space in its placeholder dsl_dir. | |
536 | */ | |
537 | void | |
538 | dsl_pool_mos_diduse_space(dsl_pool_t *dp, | |
539 | int64_t used, int64_t comp, int64_t uncomp) | |
540 | { | |
541 | ASSERT3U(comp, ==, uncomp); /* it's all metadata */ | |
542 | mutex_enter(&dp->dp_lock); | |
543 | dp->dp_mos_used_delta += used; | |
544 | dp->dp_mos_compressed_delta += comp; | |
545 | dp->dp_mos_uncompressed_delta += uncomp; | |
546 | mutex_exit(&dp->dp_lock); | |
547 | } | |
548 | ||
e8b96c60 MA |
549 | static void |
550 | dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) | |
551 | { | |
552 | zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
553 | dmu_objset_sync(dp->dp_meta_objset, zio, tx); | |
554 | VERIFY0(zio_wait(zio)); | |
555 | dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); | |
556 | spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); | |
557 | } | |
558 | ||
559 | static void | |
560 | dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) | |
561 | { | |
562 | ASSERT(MUTEX_HELD(&dp->dp_lock)); | |
563 | ||
564 | if (delta < 0) | |
565 | ASSERT3U(-delta, <=, dp->dp_dirty_total); | |
566 | ||
567 | dp->dp_dirty_total += delta; | |
568 | ||
569 | /* | |
570 | * Note: we signal even when increasing dp_dirty_total. | |
571 | * This ensures forward progress -- each thread wakes the next waiter. | |
572 | */ | |
c0c8cc7b | 573 | if (dp->dp_dirty_total < zfs_dirty_data_max) |
e8b96c60 MA |
574 | cv_signal(&dp->dp_spaceavail_cv); |
575 | } | |
576 | ||
34dc7c2f BB |
577 | void |
578 | dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) | |
579 | { | |
580 | zio_t *zio; | |
581 | dmu_tx_t *tx; | |
582 | dsl_dir_t *dd; | |
583 | dsl_dataset_t *ds; | |
428870ff | 584 | objset_t *mos = dp->dp_meta_objset; |
29809a6c MA |
585 | list_t synced_datasets; |
586 | ||
587 | list_create(&synced_datasets, sizeof (dsl_dataset_t), | |
588 | offsetof(dsl_dataset_t, ds_synced_link)); | |
34dc7c2f BB |
589 | |
590 | tx = dmu_tx_create_assigned(dp, txg); | |
591 | ||
e8b96c60 MA |
592 | /* |
593 | * Write out all dirty blocks of dirty datasets. | |
594 | */ | |
34dc7c2f | 595 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
e8b96c60 | 596 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { |
9babb374 BB |
597 | /* |
598 | * We must not sync any non-MOS datasets twice, because | |
599 | * we may have taken a snapshot of them. However, we | |
600 | * may sync newly-created datasets on pass 2. | |
601 | */ | |
602 | ASSERT(!list_link_active(&ds->ds_synced_link)); | |
29809a6c | 603 | list_insert_tail(&synced_datasets, ds); |
34dc7c2f BB |
604 | dsl_dataset_sync(ds, zio, tx); |
605 | } | |
e8b96c60 | 606 | VERIFY0(zio_wait(zio)); |
9babb374 | 607 | |
e8b96c60 MA |
608 | /* |
609 | * We have written all of the accounted dirty data, so our | |
610 | * dp_space_towrite should now be zero. However, some seldom-used | |
611 | * code paths do not adhere to this (e.g. dbuf_undirty(), also | |
612 | * rounding error in dbuf_write_physdone). | |
613 | * Shore up the accounting of any dirtied space now. | |
614 | */ | |
615 | dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); | |
34dc7c2f | 616 | |
539d33c7 GM |
617 | /* |
618 | * Update the long range free counter after | |
619 | * we're done syncing user data | |
620 | */ | |
621 | mutex_enter(&dp->dp_lock); | |
622 | ASSERT(spa_sync_pass(dp->dp_spa) == 1 || | |
623 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] == 0); | |
624 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] = 0; | |
625 | mutex_exit(&dp->dp_lock); | |
626 | ||
29809a6c MA |
627 | /* |
628 | * After the data blocks have been written (ensured by the zio_wait() | |
9c5167d1 | 629 | * above), update the user/group/project space accounting. This happens |
64fc7762 MA |
630 | * in tasks dispatched to dp_sync_taskq, so wait for them before |
631 | * continuing. | |
29809a6c | 632 | */ |
e8b96c60 MA |
633 | for (ds = list_head(&synced_datasets); ds != NULL; |
634 | ds = list_next(&synced_datasets, ds)) { | |
428870ff | 635 | dmu_objset_do_userquota_updates(ds->ds_objset, tx); |
e8b96c60 | 636 | } |
64fc7762 | 637 | taskq_wait(dp->dp_sync_taskq); |
9babb374 BB |
638 | |
639 | /* | |
640 | * Sync the datasets again to push out the changes due to | |
428870ff | 641 | * userspace updates. This must be done before we process the |
29809a6c MA |
642 | * sync tasks, so that any snapshots will have the correct |
643 | * user accounting information (and we won't get confused | |
644 | * about which blocks are part of the snapshot). | |
9babb374 BB |
645 | */ |
646 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
e8b96c60 | 647 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { |
9babb374 BB |
648 | ASSERT(list_link_active(&ds->ds_synced_link)); |
649 | dmu_buf_rele(ds->ds_dbuf, ds); | |
650 | dsl_dataset_sync(ds, zio, tx); | |
651 | } | |
e8b96c60 | 652 | VERIFY0(zio_wait(zio)); |
9babb374 | 653 | |
428870ff | 654 | /* |
29809a6c MA |
655 | * Now that the datasets have been completely synced, we can |
656 | * clean up our in-memory structures accumulated while syncing: | |
657 | * | |
658 | * - move dead blocks from the pending deadlist to the on-disk deadlist | |
29809a6c | 659 | * - release hold from dsl_dataset_dirty() |
428870ff | 660 | */ |
e8b96c60 | 661 | while ((ds = list_remove_head(&synced_datasets)) != NULL) { |
0efd9791 | 662 | dsl_dataset_sync_done(ds, tx); |
428870ff BB |
663 | } |
664 | ||
e8b96c60 | 665 | while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { |
34dc7c2f | 666 | dsl_dir_sync(dd, tx); |
e8b96c60 | 667 | } |
b128c09f | 668 | |
29809a6c MA |
669 | /* |
670 | * The MOS's space is accounted for in the pool/$MOS | |
671 | * (dp_mos_dir). We can't modify the mos while we're syncing | |
672 | * it, so we remember the deltas and apply them here. | |
673 | */ | |
674 | if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || | |
675 | dp->dp_mos_uncompressed_delta != 0) { | |
676 | dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, | |
677 | dp->dp_mos_used_delta, | |
678 | dp->dp_mos_compressed_delta, | |
679 | dp->dp_mos_uncompressed_delta, tx); | |
680 | dp->dp_mos_used_delta = 0; | |
681 | dp->dp_mos_compressed_delta = 0; | |
682 | dp->dp_mos_uncompressed_delta = 0; | |
683 | } | |
684 | ||
64fc7762 | 685 | if (!multilist_is_empty(mos->os_dirty_dnodes[txg & TXG_MASK])) { |
e8b96c60 | 686 | dsl_pool_sync_mos(dp, tx); |
34dc7c2f BB |
687 | } |
688 | ||
29809a6c MA |
689 | /* |
690 | * If we modify a dataset in the same txg that we want to destroy it, | |
691 | * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. | |
692 | * dsl_dir_destroy_check() will fail if there are unexpected holds. | |
693 | * Therefore, we want to sync the MOS (thus syncing the dd_dbuf | |
694 | * and clearing the hold on it) before we process the sync_tasks. | |
695 | * The MOS data dirtied by the sync_tasks will be synced on the next | |
696 | * pass. | |
697 | */ | |
29809a6c | 698 | if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { |
13fe0198 | 699 | dsl_sync_task_t *dst; |
29809a6c MA |
700 | /* |
701 | * No more sync tasks should have been added while we | |
702 | * were syncing. | |
703 | */ | |
e8b96c60 MA |
704 | ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); |
705 | while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) | |
13fe0198 | 706 | dsl_sync_task_sync(dst, tx); |
29809a6c MA |
707 | } |
708 | ||
34dc7c2f | 709 | dmu_tx_commit(tx); |
b128c09f | 710 | |
e8b96c60 | 711 | DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); |
34dc7c2f BB |
712 | } |
713 | ||
714 | void | |
428870ff | 715 | dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) |
34dc7c2f | 716 | { |
29809a6c | 717 | zilog_t *zilog; |
34dc7c2f | 718 | |
55922e73 | 719 | while ((zilog = txg_list_head(&dp->dp_dirty_zilogs, txg))) { |
e8b96c60 | 720 | dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); |
55922e73 GW |
721 | /* |
722 | * We don't remove the zilog from the dp_dirty_zilogs | |
723 | * list until after we've cleaned it. This ensures that | |
724 | * callers of zilog_is_dirty() receive an accurate | |
725 | * answer when they are racing with the spa sync thread. | |
726 | */ | |
29809a6c | 727 | zil_clean(zilog, txg); |
55922e73 | 728 | (void) txg_list_remove_this(&dp->dp_dirty_zilogs, zilog, txg); |
29809a6c MA |
729 | ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); |
730 | dmu_buf_rele(ds->ds_dbuf, zilog); | |
34dc7c2f | 731 | } |
428870ff | 732 | ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); |
34dc7c2f BB |
733 | } |
734 | ||
735 | /* | |
736 | * TRUE if the current thread is the tx_sync_thread or if we | |
737 | * are being called from SPA context during pool initialization. | |
738 | */ | |
739 | int | |
740 | dsl_pool_sync_context(dsl_pool_t *dp) | |
741 | { | |
742 | return (curthread == dp->dp_tx.tx_sync_thread || | |
64fc7762 MA |
743 | spa_is_initializing(dp->dp_spa) || |
744 | taskq_member(dp->dp_sync_taskq, curthread)); | |
34dc7c2f BB |
745 | } |
746 | ||
747 | uint64_t | |
748 | dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) | |
749 | { | |
750 | uint64_t space, resv; | |
751 | ||
752 | /* | |
34dc7c2f BB |
753 | * If we're trying to assess whether it's OK to do a free, |
754 | * cut the reservation in half to allow forward progress | |
755 | * (e.g. make it possible to rm(1) files from a full pool). | |
756 | */ | |
757 | space = spa_get_dspace(dp->dp_spa); | |
0c60cc32 | 758 | resv = spa_get_slop_space(dp->dp_spa); |
34dc7c2f BB |
759 | if (netfree) |
760 | resv >>= 1; | |
761 | ||
762 | return (space - resv); | |
763 | } | |
764 | ||
e8b96c60 MA |
765 | boolean_t |
766 | dsl_pool_need_dirty_delay(dsl_pool_t *dp) | |
34dc7c2f | 767 | { |
e8b96c60 MA |
768 | uint64_t delay_min_bytes = |
769 | zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; | |
770 | boolean_t rv; | |
34dc7c2f | 771 | |
e8b96c60 MA |
772 | mutex_enter(&dp->dp_lock); |
773 | if (dp->dp_dirty_total > zfs_dirty_data_sync) | |
774 | txg_kick(dp); | |
775 | rv = (dp->dp_dirty_total > delay_min_bytes); | |
776 | mutex_exit(&dp->dp_lock); | |
777 | return (rv); | |
34dc7c2f BB |
778 | } |
779 | ||
780 | void | |
e8b96c60 | 781 | dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) |
34dc7c2f | 782 | { |
e8b96c60 MA |
783 | if (space > 0) { |
784 | mutex_enter(&dp->dp_lock); | |
785 | dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; | |
786 | dsl_pool_dirty_delta(dp, space); | |
787 | mutex_exit(&dp->dp_lock); | |
788 | } | |
34dc7c2f BB |
789 | } |
790 | ||
791 | void | |
e8b96c60 | 792 | dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) |
34dc7c2f | 793 | { |
e8b96c60 MA |
794 | ASSERT3S(space, >=, 0); |
795 | if (space == 0) | |
34dc7c2f BB |
796 | return; |
797 | ||
e8b96c60 MA |
798 | mutex_enter(&dp->dp_lock); |
799 | if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { | |
800 | /* XXX writing something we didn't dirty? */ | |
801 | space = dp->dp_dirty_pertxg[txg & TXG_MASK]; | |
34dc7c2f | 802 | } |
e8b96c60 MA |
803 | ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); |
804 | dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; | |
805 | ASSERT3U(dp->dp_dirty_total, >=, space); | |
806 | dsl_pool_dirty_delta(dp, -space); | |
807 | mutex_exit(&dp->dp_lock); | |
34dc7c2f | 808 | } |
b128c09f BB |
809 | |
810 | /* ARGSUSED */ | |
811 | static int | |
13fe0198 | 812 | upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) |
b128c09f BB |
813 | { |
814 | dmu_tx_t *tx = arg; | |
815 | dsl_dataset_t *ds, *prev = NULL; | |
816 | int err; | |
b128c09f | 817 | |
13fe0198 | 818 | err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); |
b128c09f BB |
819 | if (err) |
820 | return (err); | |
821 | ||
d683ddbb JG |
822 | while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { |
823 | err = dsl_dataset_hold_obj(dp, | |
824 | dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); | |
b128c09f BB |
825 | if (err) { |
826 | dsl_dataset_rele(ds, FTAG); | |
827 | return (err); | |
828 | } | |
829 | ||
d683ddbb | 830 | if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) |
b128c09f BB |
831 | break; |
832 | dsl_dataset_rele(ds, FTAG); | |
833 | ds = prev; | |
834 | prev = NULL; | |
835 | } | |
836 | ||
837 | if (prev == NULL) { | |
838 | prev = dp->dp_origin_snap; | |
839 | ||
840 | /* | |
841 | * The $ORIGIN can't have any data, or the accounting | |
842 | * will be wrong. | |
843 | */ | |
cc9bb3e5 | 844 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
d683ddbb | 845 | ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); |
cc9bb3e5 | 846 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
b128c09f BB |
847 | |
848 | /* The origin doesn't get attached to itself */ | |
849 | if (ds->ds_object == prev->ds_object) { | |
850 | dsl_dataset_rele(ds, FTAG); | |
851 | return (0); | |
852 | } | |
853 | ||
854 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
d683ddbb JG |
855 | dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; |
856 | dsl_dataset_phys(ds)->ds_prev_snap_txg = | |
857 | dsl_dataset_phys(prev)->ds_creation_txg; | |
b128c09f BB |
858 | |
859 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); | |
d683ddbb | 860 | dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; |
b128c09f BB |
861 | |
862 | dmu_buf_will_dirty(prev->ds_dbuf, tx); | |
d683ddbb | 863 | dsl_dataset_phys(prev)->ds_num_children++; |
b128c09f | 864 | |
d683ddbb | 865 | if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { |
b128c09f | 866 | ASSERT(ds->ds_prev == NULL); |
13fe0198 | 867 | VERIFY0(dsl_dataset_hold_obj(dp, |
d683ddbb JG |
868 | dsl_dataset_phys(ds)->ds_prev_snap_obj, |
869 | ds, &ds->ds_prev)); | |
b128c09f BB |
870 | } |
871 | } | |
872 | ||
d683ddbb JG |
873 | ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); |
874 | ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); | |
b128c09f | 875 | |
d683ddbb | 876 | if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { |
428870ff | 877 | dmu_buf_will_dirty(prev->ds_dbuf, tx); |
d683ddbb | 878 | dsl_dataset_phys(prev)->ds_next_clones_obj = |
b128c09f BB |
879 | zap_create(dp->dp_meta_objset, |
880 | DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); | |
881 | } | |
13fe0198 | 882 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
d683ddbb | 883 | dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); |
b128c09f BB |
884 | |
885 | dsl_dataset_rele(ds, FTAG); | |
886 | if (prev != dp->dp_origin_snap) | |
887 | dsl_dataset_rele(prev, FTAG); | |
888 | return (0); | |
889 | } | |
890 | ||
891 | void | |
892 | dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
893 | { | |
894 | ASSERT(dmu_tx_is_syncing(tx)); | |
895 | ASSERT(dp->dp_origin_snap != NULL); | |
896 | ||
13fe0198 | 897 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, |
9c43027b | 898 | tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); |
428870ff BB |
899 | } |
900 | ||
901 | /* ARGSUSED */ | |
902 | static int | |
13fe0198 | 903 | upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
428870ff BB |
904 | { |
905 | dmu_tx_t *tx = arg; | |
428870ff BB |
906 | objset_t *mos = dp->dp_meta_objset; |
907 | ||
d683ddbb | 908 | if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { |
428870ff BB |
909 | dsl_dataset_t *origin; |
910 | ||
13fe0198 | 911 | VERIFY0(dsl_dataset_hold_obj(dp, |
d683ddbb | 912 | dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); |
428870ff | 913 | |
d683ddbb | 914 | if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { |
428870ff | 915 | dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); |
d683ddbb JG |
916 | dsl_dir_phys(origin->ds_dir)->dd_clones = |
917 | zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, | |
918 | 0, tx); | |
428870ff BB |
919 | } |
920 | ||
13fe0198 | 921 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
d683ddbb JG |
922 | dsl_dir_phys(origin->ds_dir)->dd_clones, |
923 | ds->ds_object, tx)); | |
428870ff BB |
924 | |
925 | dsl_dataset_rele(origin, FTAG); | |
926 | } | |
428870ff BB |
927 | return (0); |
928 | } | |
929 | ||
930 | void | |
931 | dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
932 | { | |
428870ff BB |
933 | uint64_t obj; |
934 | ||
d6320ddb BB |
935 | ASSERT(dmu_tx_is_syncing(tx)); |
936 | ||
428870ff | 937 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); |
13fe0198 | 938 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
939 | FREE_DIR_NAME, &dp->dp_free_dir)); |
940 | ||
941 | /* | |
942 | * We can't use bpobj_alloc(), because spa_version() still | |
943 | * returns the old version, and we need a new-version bpobj with | |
944 | * subobj support. So call dmu_object_alloc() directly. | |
945 | */ | |
946 | obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, | |
f1512ee6 | 947 | SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); |
13fe0198 | 948 | VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 949 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); |
13fe0198 | 950 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); |
428870ff | 951 | |
13fe0198 | 952 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
9c43027b | 953 | upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); |
b128c09f BB |
954 | } |
955 | ||
956 | void | |
957 | dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) | |
958 | { | |
959 | uint64_t dsobj; | |
960 | dsl_dataset_t *ds; | |
961 | ||
962 | ASSERT(dmu_tx_is_syncing(tx)); | |
963 | ASSERT(dp->dp_origin_snap == NULL); | |
13fe0198 | 964 | ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); |
b128c09f BB |
965 | |
966 | /* create the origin dir, ds, & snap-ds */ | |
b128c09f | 967 | dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, |
b5256303 | 968 | NULL, 0, kcred, NULL, tx); |
13fe0198 MA |
969 | VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); |
970 | dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); | |
d683ddbb | 971 | VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, |
b128c09f BB |
972 | dp, &dp->dp_origin_snap)); |
973 | dsl_dataset_rele(ds, FTAG); | |
b128c09f | 974 | } |
9babb374 BB |
975 | |
976 | taskq_t * | |
3558fd73 | 977 | dsl_pool_iput_taskq(dsl_pool_t *dp) |
9babb374 | 978 | { |
3558fd73 | 979 | return (dp->dp_iput_taskq); |
9babb374 | 980 | } |
428870ff BB |
981 | |
982 | /* | |
983 | * Walk through the pool-wide zap object of temporary snapshot user holds | |
984 | * and release them. | |
985 | */ | |
986 | void | |
987 | dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) | |
988 | { | |
989 | zap_attribute_t za; | |
990 | zap_cursor_t zc; | |
991 | objset_t *mos = dp->dp_meta_objset; | |
992 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
95fd54a1 | 993 | nvlist_t *holds; |
428870ff BB |
994 | |
995 | if (zapobj == 0) | |
996 | return; | |
997 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
998 | ||
95fd54a1 SH |
999 | holds = fnvlist_alloc(); |
1000 | ||
428870ff BB |
1001 | for (zap_cursor_init(&zc, mos, zapobj); |
1002 | zap_cursor_retrieve(&zc, &za) == 0; | |
1003 | zap_cursor_advance(&zc)) { | |
1004 | char *htag; | |
95fd54a1 | 1005 | nvlist_t *tags; |
428870ff BB |
1006 | |
1007 | htag = strchr(za.za_name, '-'); | |
1008 | *htag = '\0'; | |
1009 | ++htag; | |
95fd54a1 SH |
1010 | if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { |
1011 | tags = fnvlist_alloc(); | |
1012 | fnvlist_add_boolean(tags, htag); | |
1013 | fnvlist_add_nvlist(holds, za.za_name, tags); | |
1014 | fnvlist_free(tags); | |
1015 | } else { | |
1016 | fnvlist_add_boolean(tags, htag); | |
1017 | } | |
428870ff | 1018 | } |
95fd54a1 SH |
1019 | dsl_dataset_user_release_tmp(dp, holds); |
1020 | fnvlist_free(holds); | |
428870ff BB |
1021 | zap_cursor_fini(&zc); |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * Create the pool-wide zap object for storing temporary snapshot holds. | |
1026 | */ | |
1027 | void | |
1028 | dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) | |
1029 | { | |
1030 | objset_t *mos = dp->dp_meta_objset; | |
1031 | ||
1032 | ASSERT(dp->dp_tmp_userrefs_obj == 0); | |
1033 | ASSERT(dmu_tx_is_syncing(tx)); | |
1034 | ||
9ae529ec CS |
1035 | dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, |
1036 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); | |
428870ff BB |
1037 | } |
1038 | ||
1039 | static int | |
1040 | dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, | |
13fe0198 | 1041 | const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) |
428870ff BB |
1042 | { |
1043 | objset_t *mos = dp->dp_meta_objset; | |
1044 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
1045 | char *name; | |
1046 | int error; | |
1047 | ||
1048 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
1049 | ASSERT(dmu_tx_is_syncing(tx)); | |
1050 | ||
1051 | /* | |
1052 | * If the pool was created prior to SPA_VERSION_USERREFS, the | |
1053 | * zap object for temporary holds might not exist yet. | |
1054 | */ | |
1055 | if (zapobj == 0) { | |
1056 | if (holding) { | |
1057 | dsl_pool_user_hold_create_obj(dp, tx); | |
1058 | zapobj = dp->dp_tmp_userrefs_obj; | |
1059 | } else { | |
2e528b49 | 1060 | return (SET_ERROR(ENOENT)); |
428870ff BB |
1061 | } |
1062 | } | |
1063 | ||
1064 | name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); | |
1065 | if (holding) | |
13fe0198 | 1066 | error = zap_add(mos, zapobj, name, 8, 1, &now, tx); |
428870ff BB |
1067 | else |
1068 | error = zap_remove(mos, zapobj, name, tx); | |
1069 | strfree(name); | |
1070 | ||
1071 | return (error); | |
1072 | } | |
1073 | ||
1074 | /* | |
1075 | * Add a temporary hold for the given dataset object and tag. | |
1076 | */ | |
1077 | int | |
1078 | dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
13fe0198 | 1079 | uint64_t now, dmu_tx_t *tx) |
428870ff BB |
1080 | { |
1081 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); | |
1082 | } | |
1083 | ||
1084 | /* | |
1085 | * Release a temporary hold for the given dataset object and tag. | |
1086 | */ | |
1087 | int | |
1088 | dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
1089 | dmu_tx_t *tx) | |
1090 | { | |
13fe0198 | 1091 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, |
428870ff BB |
1092 | tx, B_FALSE)); |
1093 | } | |
c409e464 | 1094 | |
13fe0198 MA |
1095 | /* |
1096 | * DSL Pool Configuration Lock | |
1097 | * | |
1098 | * The dp_config_rwlock protects against changes to DSL state (e.g. dataset | |
1099 | * creation / destruction / rename / property setting). It must be held for | |
1100 | * read to hold a dataset or dsl_dir. I.e. you must call | |
1101 | * dsl_pool_config_enter() or dsl_pool_hold() before calling | |
1102 | * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock | |
1103 | * must be held continuously until all datasets and dsl_dirs are released. | |
1104 | * | |
1105 | * The only exception to this rule is that if a "long hold" is placed on | |
1106 | * a dataset, then the dp_config_rwlock may be dropped while the dataset | |
1107 | * is still held. The long hold will prevent the dataset from being | |
1108 | * destroyed -- the destroy will fail with EBUSY. A long hold can be | |
1109 | * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset | |
1110 | * (by calling dsl_{dataset,objset}_{try}own{_obj}). | |
1111 | * | |
1112 | * Legitimate long-holders (including owners) should be long-running, cancelable | |
1113 | * tasks that should cause "zfs destroy" to fail. This includes DMU | |
1114 | * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), | |
1115 | * "zfs send", and "zfs diff". There are several other long-holders whose | |
1116 | * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). | |
1117 | * | |
1118 | * The usual formula for long-holding would be: | |
1119 | * dsl_pool_hold() | |
1120 | * dsl_dataset_hold() | |
1121 | * ... perform checks ... | |
1122 | * dsl_dataset_long_hold() | |
1123 | * dsl_pool_rele() | |
1124 | * ... perform long-running task ... | |
1125 | * dsl_dataset_long_rele() | |
1126 | * dsl_dataset_rele() | |
1127 | * | |
1128 | * Note that when the long hold is released, the dataset is still held but | |
1129 | * the pool is not held. The dataset may change arbitrarily during this time | |
1130 | * (e.g. it could be destroyed). Therefore you shouldn't do anything to the | |
1131 | * dataset except release it. | |
1132 | * | |
1133 | * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only | |
1134 | * or modifying operations. | |
1135 | * | |
1136 | * Modifying operations should generally use dsl_sync_task(). The synctask | |
1137 | * infrastructure enforces proper locking strategy with respect to the | |
1138 | * dp_config_rwlock. See the comment above dsl_sync_task() for details. | |
1139 | * | |
1140 | * Read-only operations will manually hold the pool, then the dataset, obtain | |
1141 | * information from the dataset, then release the pool and dataset. | |
1142 | * dmu_objset_{hold,rele}() are convenience routines that also do the pool | |
1143 | * hold/rele. | |
1144 | */ | |
1145 | ||
1146 | int | |
1147 | dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) | |
1148 | { | |
1149 | spa_t *spa; | |
1150 | int error; | |
1151 | ||
1152 | error = spa_open(name, &spa, tag); | |
1153 | if (error == 0) { | |
1154 | *dp = spa_get_dsl(spa); | |
1155 | dsl_pool_config_enter(*dp, tag); | |
1156 | } | |
1157 | return (error); | |
1158 | } | |
1159 | ||
1160 | void | |
1161 | dsl_pool_rele(dsl_pool_t *dp, void *tag) | |
1162 | { | |
1163 | dsl_pool_config_exit(dp, tag); | |
1164 | spa_close(dp->dp_spa, tag); | |
1165 | } | |
1166 | ||
1167 | void | |
1168 | dsl_pool_config_enter(dsl_pool_t *dp, void *tag) | |
1169 | { | |
1170 | /* | |
1171 | * We use a "reentrant" reader-writer lock, but not reentrantly. | |
1172 | * | |
1173 | * The rrwlock can (with the track_all flag) track all reading threads, | |
1174 | * which is very useful for debugging which code path failed to release | |
1175 | * the lock, and for verifying that the *current* thread does hold | |
1176 | * the lock. | |
1177 | * | |
1178 | * (Unlike a rwlock, which knows that N threads hold it for | |
1179 | * read, but not *which* threads, so rw_held(RW_READER) returns TRUE | |
1180 | * if any thread holds it for read, even if this thread doesn't). | |
1181 | */ | |
1182 | ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); | |
1183 | rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); | |
1184 | } | |
1185 | ||
5e8cd5d1 AJ |
1186 | void |
1187 | dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) | |
1188 | { | |
1189 | ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); | |
1190 | rrw_enter_read_prio(&dp->dp_config_rwlock, tag); | |
1191 | } | |
1192 | ||
13fe0198 MA |
1193 | void |
1194 | dsl_pool_config_exit(dsl_pool_t *dp, void *tag) | |
1195 | { | |
1196 | rrw_exit(&dp->dp_config_rwlock, tag); | |
1197 | } | |
1198 | ||
1199 | boolean_t | |
1200 | dsl_pool_config_held(dsl_pool_t *dp) | |
1201 | { | |
1202 | return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); | |
1203 | } | |
1204 | ||
9c43027b AJ |
1205 | boolean_t |
1206 | dsl_pool_config_held_writer(dsl_pool_t *dp) | |
1207 | { | |
1208 | return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); | |
1209 | } | |
1210 | ||
c409e464 | 1211 | #if defined(_KERNEL) && defined(HAVE_SPL) |
40a806df NB |
1212 | EXPORT_SYMBOL(dsl_pool_config_enter); |
1213 | EXPORT_SYMBOL(dsl_pool_config_exit); | |
1214 | ||
02730c33 | 1215 | /* BEGIN CSTYLED */ |
d1d7e268 | 1216 | /* zfs_dirty_data_max_percent only applied at module load in arc_init(). */ |
e8b96c60 MA |
1217 | module_param(zfs_dirty_data_max_percent, int, 0444); |
1218 | MODULE_PARM_DESC(zfs_dirty_data_max_percent, "percent of ram can be dirty"); | |
c409e464 | 1219 | |
d1d7e268 | 1220 | /* zfs_dirty_data_max_max_percent only applied at module load in arc_init(). */ |
e8b96c60 MA |
1221 | module_param(zfs_dirty_data_max_max_percent, int, 0444); |
1222 | MODULE_PARM_DESC(zfs_dirty_data_max_max_percent, | |
d1d7e268 | 1223 | "zfs_dirty_data_max upper bound as % of RAM"); |
c409e464 | 1224 | |
e8b96c60 MA |
1225 | module_param(zfs_delay_min_dirty_percent, int, 0644); |
1226 | MODULE_PARM_DESC(zfs_delay_min_dirty_percent, "transaction delay threshold"); | |
c409e464 | 1227 | |
e8b96c60 MA |
1228 | module_param(zfs_dirty_data_max, ulong, 0644); |
1229 | MODULE_PARM_DESC(zfs_dirty_data_max, "determines the dirty space limit"); | |
c409e464 | 1230 | |
d1d7e268 | 1231 | /* zfs_dirty_data_max_max only applied at module load in arc_init(). */ |
e8b96c60 MA |
1232 | module_param(zfs_dirty_data_max_max, ulong, 0444); |
1233 | MODULE_PARM_DESC(zfs_dirty_data_max_max, | |
d1d7e268 | 1234 | "zfs_dirty_data_max upper bound in bytes"); |
c409e464 | 1235 | |
e8b96c60 MA |
1236 | module_param(zfs_dirty_data_sync, ulong, 0644); |
1237 | MODULE_PARM_DESC(zfs_dirty_data_sync, "sync txg when this much dirty data"); | |
c409e464 | 1238 | |
e8b96c60 MA |
1239 | module_param(zfs_delay_scale, ulong, 0644); |
1240 | MODULE_PARM_DESC(zfs_delay_scale, "how quickly delay approaches infinity"); | |
64fc7762 MA |
1241 | |
1242 | module_param(zfs_sync_taskq_batch_pct, int, 0644); | |
1243 | MODULE_PARM_DESC(zfs_sync_taskq_batch_pct, | |
1244 | "max percent of CPUs that are used to sync dirty data"); | |
a032ac4b BB |
1245 | |
1246 | module_param(zfs_zil_clean_taskq_nthr_pct, int, 0644); | |
1247 | MODULE_PARM_DESC(zfs_zil_clean_taskq_nthr_pct, | |
1248 | "max percent of CPUs that are used per dp_sync_taskq"); | |
1249 | ||
1250 | module_param(zfs_zil_clean_taskq_minalloc, int, 0644); | |
1251 | MODULE_PARM_DESC(zfs_zil_clean_taskq_minalloc, | |
1252 | "number of taskq entries that are pre-populated"); | |
1253 | ||
1254 | module_param(zfs_zil_clean_taskq_maxalloc, int, 0644); | |
1255 | MODULE_PARM_DESC(zfs_zil_clean_taskq_maxalloc, | |
1256 | "max number of taskq entries that are cached"); | |
1257 | ||
02730c33 | 1258 | /* END CSTYLED */ |
c409e464 | 1259 | #endif |