]>
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. |
cc9bb3e5 | 23 | * Copyright (c) 2011, 2015 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> |
34dc7c2f | 51 | |
e8b96c60 MA |
52 | /* |
53 | * ZFS Write Throttle | |
54 | * ------------------ | |
55 | * | |
56 | * ZFS must limit the rate of incoming writes to the rate at which it is able | |
57 | * to sync data modifications to the backend storage. Throttling by too much | |
58 | * creates an artificial limit; throttling by too little can only be sustained | |
59 | * for short periods and would lead to highly lumpy performance. On a per-pool | |
60 | * basis, ZFS tracks the amount of modified (dirty) data. As operations change | |
61 | * data, the amount of dirty data increases; as ZFS syncs out data, the amount | |
62 | * of dirty data decreases. When the amount of dirty data exceeds a | |
63 | * predetermined threshold further modifications are blocked until the amount | |
64 | * of dirty data decreases (as data is synced out). | |
65 | * | |
66 | * The limit on dirty data is tunable, and should be adjusted according to | |
67 | * both the IO capacity and available memory of the system. The larger the | |
68 | * window, the more ZFS is able to aggregate and amortize metadata (and data) | |
69 | * changes. However, memory is a limited resource, and allowing for more dirty | |
70 | * data comes at the cost of keeping other useful data in memory (for example | |
71 | * ZFS data cached by the ARC). | |
72 | * | |
73 | * Implementation | |
74 | * | |
75 | * As buffers are modified dsl_pool_willuse_space() increments both the per- | |
76 | * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of | |
77 | * dirty space used; dsl_pool_dirty_space() decrements those values as data | |
78 | * is synced out from dsl_pool_sync(). While only the poolwide value is | |
79 | * relevant, the per-txg value is useful for debugging. The tunable | |
80 | * zfs_dirty_data_max determines the dirty space limit. Once that value is | |
81 | * exceeded, new writes are halted until space frees up. | |
82 | * | |
83 | * The zfs_dirty_data_sync tunable dictates the threshold at which we | |
84 | * ensure that there is a txg syncing (see the comment in txg.c for a full | |
85 | * description of transaction group stages). | |
86 | * | |
87 | * The IO scheduler uses both the dirty space limit and current amount of | |
88 | * dirty data as inputs. Those values affect the number of concurrent IOs ZFS | |
89 | * issues. See the comment in vdev_queue.c for details of the IO scheduler. | |
90 | * | |
91 | * The delay is also calculated based on the amount of dirty data. See the | |
92 | * comment above dmu_tx_delay() for details. | |
93 | */ | |
94 | ||
95 | /* | |
96 | * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory, | |
97 | * capped at zfs_dirty_data_max_max. It can also be overridden with a module | |
98 | * parameter. | |
99 | */ | |
100 | unsigned long zfs_dirty_data_max = 0; | |
101 | unsigned long zfs_dirty_data_max_max = 0; | |
102 | int zfs_dirty_data_max_percent = 10; | |
103 | int zfs_dirty_data_max_max_percent = 25; | |
b128c09f | 104 | |
e8b96c60 MA |
105 | /* |
106 | * If there is at least this much dirty data, push out a txg. | |
107 | */ | |
108 | unsigned long zfs_dirty_data_sync = 64 * 1024 * 1024; | |
34dc7c2f | 109 | |
e8b96c60 MA |
110 | /* |
111 | * Once there is this amount of dirty data, the dmu_tx_delay() will kick in | |
112 | * and delay each transaction. | |
113 | * This value should be >= zfs_vdev_async_write_active_max_dirty_percent. | |
114 | */ | |
115 | int zfs_delay_min_dirty_percent = 60; | |
b128c09f | 116 | |
e8b96c60 MA |
117 | /* |
118 | * This controls how quickly the delay approaches infinity. | |
119 | * Larger values cause it to delay more for a given amount of dirty data. | |
120 | * Therefore larger values will cause there to be less dirty data for a | |
121 | * given throughput. | |
122 | * | |
123 | * For the smoothest delay, this value should be about 1 billion divided | |
124 | * by the maximum number of operations per second. This will smoothly | |
125 | * handle between 10x and 1/10th this number. | |
126 | * | |
127 | * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the | |
128 | * multiply in dmu_tx_delay(). | |
129 | */ | |
130 | unsigned long zfs_delay_scale = 1000 * 1000 * 1000 / 2000; | |
b128c09f | 131 | |
63fd3c6c AL |
132 | hrtime_t zfs_throttle_delay = MSEC2NSEC(10); |
133 | hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); | |
134 | ||
428870ff | 135 | int |
b128c09f | 136 | dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) |
34dc7c2f BB |
137 | { |
138 | uint64_t obj; | |
139 | int err; | |
140 | ||
141 | err = zap_lookup(dp->dp_meta_objset, | |
d683ddbb | 142 | dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj, |
b128c09f | 143 | name, sizeof (obj), 1, &obj); |
34dc7c2f BB |
144 | if (err) |
145 | return (err); | |
146 | ||
13fe0198 | 147 | return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); |
34dc7c2f BB |
148 | } |
149 | ||
150 | static dsl_pool_t * | |
151 | dsl_pool_open_impl(spa_t *spa, uint64_t txg) | |
152 | { | |
153 | dsl_pool_t *dp; | |
154 | blkptr_t *bp = spa_get_rootblkptr(spa); | |
34dc7c2f BB |
155 | |
156 | dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); | |
157 | dp->dp_spa = spa; | |
158 | dp->dp_meta_rootbp = *bp; | |
13fe0198 | 159 | rrw_init(&dp->dp_config_rwlock, B_TRUE); |
34dc7c2f BB |
160 | txg_init(dp, txg); |
161 | ||
162 | txg_list_create(&dp->dp_dirty_datasets, | |
163 | offsetof(dsl_dataset_t, ds_dirty_link)); | |
29809a6c MA |
164 | txg_list_create(&dp->dp_dirty_zilogs, |
165 | offsetof(zilog_t, zl_dirty_link)); | |
34dc7c2f BB |
166 | txg_list_create(&dp->dp_dirty_dirs, |
167 | offsetof(dsl_dir_t, dd_dirty_link)); | |
168 | txg_list_create(&dp->dp_sync_tasks, | |
13fe0198 | 169 | offsetof(dsl_sync_task_t, dst_node)); |
34dc7c2f BB |
170 | |
171 | mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); | |
e8b96c60 | 172 | cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL); |
34dc7c2f | 173 | |
1229323d | 174 | dp->dp_iput_taskq = taskq_create("z_iput", max_ncpus, defclsyspri, |
aa9af22c | 175 | max_ncpus * 8, INT_MAX, TASKQ_PREPOPULATE | TASKQ_DYNAMIC); |
9babb374 | 176 | |
34dc7c2f BB |
177 | return (dp); |
178 | } | |
179 | ||
180 | int | |
9ae529ec | 181 | dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) |
34dc7c2f BB |
182 | { |
183 | int err; | |
184 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
9ae529ec | 185 | |
b7faa7aa G |
186 | /* |
187 | * Initialize the caller's dsl_pool_t structure before we actually open | |
188 | * the meta objset. This is done because a self-healing write zio may | |
189 | * be issued as part of dmu_objset_open_impl() and the spa needs its | |
190 | * dsl_pool_t initialized in order to handle the write. | |
191 | */ | |
192 | *dpp = dp; | |
193 | ||
9ae529ec CS |
194 | err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, |
195 | &dp->dp_meta_objset); | |
b7faa7aa | 196 | if (err != 0) { |
9ae529ec | 197 | dsl_pool_close(dp); |
b7faa7aa G |
198 | *dpp = NULL; |
199 | } | |
9ae529ec CS |
200 | |
201 | return (err); | |
202 | } | |
203 | ||
204 | int | |
205 | dsl_pool_open(dsl_pool_t *dp) | |
206 | { | |
207 | int err; | |
b128c09f BB |
208 | dsl_dir_t *dd; |
209 | dsl_dataset_t *ds; | |
428870ff | 210 | uint64_t obj; |
34dc7c2f | 211 | |
13fe0198 | 212 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
34dc7c2f BB |
213 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
214 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, | |
215 | &dp->dp_root_dir_obj); | |
216 | if (err) | |
217 | goto out; | |
218 | ||
13fe0198 | 219 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
220 | NULL, dp, &dp->dp_root_dir); |
221 | if (err) | |
222 | goto out; | |
223 | ||
b128c09f | 224 | err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); |
34dc7c2f BB |
225 | if (err) |
226 | goto out; | |
227 | ||
9ae529ec | 228 | if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { |
b128c09f BB |
229 | err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); |
230 | if (err) | |
231 | goto out; | |
d683ddbb JG |
232 | err = dsl_dataset_hold_obj(dp, |
233 | dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds); | |
9babb374 BB |
234 | if (err == 0) { |
235 | err = dsl_dataset_hold_obj(dp, | |
d683ddbb | 236 | dsl_dataset_phys(ds)->ds_prev_snap_obj, dp, |
9babb374 BB |
237 | &dp->dp_origin_snap); |
238 | dsl_dataset_rele(ds, FTAG); | |
239 | } | |
13fe0198 | 240 | dsl_dir_rele(dd, dp); |
b128c09f BB |
241 | if (err) |
242 | goto out; | |
b128c09f BB |
243 | } |
244 | ||
9ae529ec | 245 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { |
428870ff BB |
246 | err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, |
247 | &dp->dp_free_dir); | |
b128c09f BB |
248 | if (err) |
249 | goto out; | |
428870ff | 250 | |
b128c09f | 251 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 252 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); |
b128c09f BB |
253 | if (err) |
254 | goto out; | |
13fe0198 | 255 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff | 256 | dp->dp_meta_objset, obj)); |
b128c09f BB |
257 | } |
258 | ||
fbeddd60 MA |
259 | /* |
260 | * Note: errors ignored, because the leak dir will not exist if we | |
261 | * have not encountered a leak yet. | |
262 | */ | |
263 | (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME, | |
264 | &dp->dp_leak_dir); | |
265 | ||
fa86b5db | 266 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) { |
9ae529ec CS |
267 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
268 | DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, | |
269 | &dp->dp_bptree_obj); | |
270 | if (err != 0) | |
271 | goto out; | |
272 | } | |
273 | ||
fa86b5db | 274 | if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) { |
753c3839 MA |
275 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
276 | DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, | |
277 | &dp->dp_empty_bpobj); | |
278 | if (err != 0) | |
279 | goto out; | |
280 | } | |
281 | ||
428870ff BB |
282 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
283 | DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, | |
284 | &dp->dp_tmp_userrefs_obj); | |
285 | if (err == ENOENT) | |
286 | err = 0; | |
287 | if (err) | |
288 | goto out; | |
289 | ||
9ae529ec | 290 | err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); |
428870ff | 291 | |
34dc7c2f | 292 | out: |
13fe0198 | 293 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
34dc7c2f BB |
294 | return (err); |
295 | } | |
296 | ||
297 | void | |
298 | dsl_pool_close(dsl_pool_t *dp) | |
299 | { | |
b128c09f | 300 | /* |
e8b96c60 MA |
301 | * Drop our references from dsl_pool_open(). |
302 | * | |
b128c09f BB |
303 | * Since we held the origin_snap from "syncing" context (which |
304 | * includes pool-opening context), it actually only got a "ref" | |
305 | * and not a hold, so just drop that here. | |
306 | */ | |
307 | if (dp->dp_origin_snap) | |
13fe0198 | 308 | dsl_dataset_rele(dp->dp_origin_snap, dp); |
34dc7c2f | 309 | if (dp->dp_mos_dir) |
13fe0198 | 310 | dsl_dir_rele(dp->dp_mos_dir, dp); |
428870ff | 311 | if (dp->dp_free_dir) |
13fe0198 | 312 | dsl_dir_rele(dp->dp_free_dir, dp); |
fbeddd60 MA |
313 | if (dp->dp_leak_dir) |
314 | dsl_dir_rele(dp->dp_leak_dir, dp); | |
34dc7c2f | 315 | if (dp->dp_root_dir) |
13fe0198 | 316 | dsl_dir_rele(dp->dp_root_dir, dp); |
34dc7c2f | 317 | |
428870ff BB |
318 | bpobj_close(&dp->dp_free_bpobj); |
319 | ||
34dc7c2f BB |
320 | /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ |
321 | if (dp->dp_meta_objset) | |
428870ff | 322 | dmu_objset_evict(dp->dp_meta_objset); |
34dc7c2f BB |
323 | |
324 | txg_list_destroy(&dp->dp_dirty_datasets); | |
29809a6c | 325 | txg_list_destroy(&dp->dp_dirty_zilogs); |
428870ff | 326 | txg_list_destroy(&dp->dp_sync_tasks); |
34dc7c2f | 327 | txg_list_destroy(&dp->dp_dirty_dirs); |
34dc7c2f | 328 | |
ca0bf58d PS |
329 | /* |
330 | * We can't set retry to TRUE since we're explicitly specifying | |
331 | * a spa to flush. This is good enough; any missed buffers for | |
332 | * this spa won't cause trouble, and they'll eventually fall | |
333 | * out of the ARC just like any other unused buffer. | |
334 | */ | |
335 | arc_flush(dp->dp_spa, FALSE); | |
336 | ||
34dc7c2f | 337 | txg_fini(dp); |
428870ff | 338 | dsl_scan_fini(dp); |
0c66c32d JG |
339 | dmu_buf_user_evict_wait(); |
340 | ||
13fe0198 | 341 | rrw_destroy(&dp->dp_config_rwlock); |
34dc7c2f | 342 | mutex_destroy(&dp->dp_lock); |
3558fd73 | 343 | taskq_destroy(dp->dp_iput_taskq); |
b128c09f | 344 | if (dp->dp_blkstats) |
79c76d5b | 345 | vmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); |
34dc7c2f BB |
346 | kmem_free(dp, sizeof (dsl_pool_t)); |
347 | } | |
348 | ||
349 | dsl_pool_t * | |
b128c09f | 350 | dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) |
34dc7c2f BB |
351 | { |
352 | int err; | |
353 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
354 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
428870ff | 355 | objset_t *os; |
b128c09f | 356 | dsl_dataset_t *ds; |
428870ff | 357 | uint64_t obj; |
b128c09f | 358 | |
13fe0198 MA |
359 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
360 | ||
b128c09f | 361 | /* create and open the MOS (meta-objset) */ |
428870ff BB |
362 | dp->dp_meta_objset = dmu_objset_create_impl(spa, |
363 | NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); | |
34dc7c2f BB |
364 | |
365 | /* create the pool directory */ | |
366 | err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
367 | DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); | |
c99c9001 | 368 | ASSERT0(err); |
34dc7c2f | 369 | |
428870ff | 370 | /* Initialize scan structures */ |
13fe0198 | 371 | VERIFY0(dsl_scan_init(dp, txg)); |
428870ff | 372 | |
34dc7c2f | 373 | /* create and open the root dir */ |
b128c09f | 374 | dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); |
13fe0198 | 375 | VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
376 | NULL, dp, &dp->dp_root_dir)); |
377 | ||
378 | /* create and open the meta-objset dir */ | |
b128c09f | 379 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); |
13fe0198 | 380 | VERIFY0(dsl_pool_open_special_dir(dp, |
b128c09f BB |
381 | MOS_DIR_NAME, &dp->dp_mos_dir)); |
382 | ||
428870ff BB |
383 | if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { |
384 | /* create and open the free dir */ | |
385 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, | |
386 | FREE_DIR_NAME, tx); | |
13fe0198 | 387 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
388 | FREE_DIR_NAME, &dp->dp_free_dir)); |
389 | ||
390 | /* create and open the free_bplist */ | |
f1512ee6 | 391 | obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx); |
428870ff BB |
392 | VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
393 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); | |
13fe0198 | 394 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff BB |
395 | dp->dp_meta_objset, obj)); |
396 | } | |
397 | ||
b128c09f BB |
398 | if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) |
399 | dsl_pool_create_origin(dp, tx); | |
400 | ||
401 | /* create the root dataset */ | |
428870ff | 402 | obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); |
b128c09f BB |
403 | |
404 | /* create the root objset */ | |
13fe0198 | 405 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); |
cc9bb3e5 | 406 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
0fe3d820 BB |
407 | VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds, |
408 | dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx))); | |
cc9bb3e5 | 409 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
b128c09f | 410 | #ifdef _KERNEL |
428870ff | 411 | zfs_create_fs(os, kcred, zplprops, tx); |
b128c09f BB |
412 | #endif |
413 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
414 | |
415 | dmu_tx_commit(tx); | |
416 | ||
13fe0198 MA |
417 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
418 | ||
34dc7c2f BB |
419 | return (dp); |
420 | } | |
421 | ||
29809a6c MA |
422 | /* |
423 | * Account for the meta-objset space in its placeholder dsl_dir. | |
424 | */ | |
425 | void | |
426 | dsl_pool_mos_diduse_space(dsl_pool_t *dp, | |
427 | int64_t used, int64_t comp, int64_t uncomp) | |
428 | { | |
429 | ASSERT3U(comp, ==, uncomp); /* it's all metadata */ | |
430 | mutex_enter(&dp->dp_lock); | |
431 | dp->dp_mos_used_delta += used; | |
432 | dp->dp_mos_compressed_delta += comp; | |
433 | dp->dp_mos_uncompressed_delta += uncomp; | |
434 | mutex_exit(&dp->dp_lock); | |
435 | } | |
436 | ||
e8b96c60 MA |
437 | static void |
438 | dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx) | |
439 | { | |
440 | zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
441 | dmu_objset_sync(dp->dp_meta_objset, zio, tx); | |
442 | VERIFY0(zio_wait(zio)); | |
443 | dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); | |
444 | spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); | |
445 | } | |
446 | ||
447 | static void | |
448 | dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta) | |
449 | { | |
450 | ASSERT(MUTEX_HELD(&dp->dp_lock)); | |
451 | ||
452 | if (delta < 0) | |
453 | ASSERT3U(-delta, <=, dp->dp_dirty_total); | |
454 | ||
455 | dp->dp_dirty_total += delta; | |
456 | ||
457 | /* | |
458 | * Note: we signal even when increasing dp_dirty_total. | |
459 | * This ensures forward progress -- each thread wakes the next waiter. | |
460 | */ | |
461 | if (dp->dp_dirty_total <= zfs_dirty_data_max) | |
462 | cv_signal(&dp->dp_spaceavail_cv); | |
463 | } | |
464 | ||
34dc7c2f BB |
465 | void |
466 | dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) | |
467 | { | |
468 | zio_t *zio; | |
469 | dmu_tx_t *tx; | |
470 | dsl_dir_t *dd; | |
471 | dsl_dataset_t *ds; | |
428870ff | 472 | objset_t *mos = dp->dp_meta_objset; |
29809a6c MA |
473 | list_t synced_datasets; |
474 | ||
475 | list_create(&synced_datasets, sizeof (dsl_dataset_t), | |
476 | offsetof(dsl_dataset_t, ds_synced_link)); | |
34dc7c2f BB |
477 | |
478 | tx = dmu_tx_create_assigned(dp, txg); | |
479 | ||
e8b96c60 MA |
480 | /* |
481 | * Write out all dirty blocks of dirty datasets. | |
482 | */ | |
34dc7c2f | 483 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
e8b96c60 | 484 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { |
9babb374 BB |
485 | /* |
486 | * We must not sync any non-MOS datasets twice, because | |
487 | * we may have taken a snapshot of them. However, we | |
488 | * may sync newly-created datasets on pass 2. | |
489 | */ | |
490 | ASSERT(!list_link_active(&ds->ds_synced_link)); | |
29809a6c | 491 | list_insert_tail(&synced_datasets, ds); |
34dc7c2f BB |
492 | dsl_dataset_sync(ds, zio, tx); |
493 | } | |
e8b96c60 | 494 | VERIFY0(zio_wait(zio)); |
9babb374 | 495 | |
e8b96c60 MA |
496 | /* |
497 | * We have written all of the accounted dirty data, so our | |
498 | * dp_space_towrite should now be zero. However, some seldom-used | |
499 | * code paths do not adhere to this (e.g. dbuf_undirty(), also | |
500 | * rounding error in dbuf_write_physdone). | |
501 | * Shore up the accounting of any dirtied space now. | |
502 | */ | |
503 | dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg); | |
34dc7c2f | 504 | |
539d33c7 GM |
505 | /* |
506 | * Update the long range free counter after | |
507 | * we're done syncing user data | |
508 | */ | |
509 | mutex_enter(&dp->dp_lock); | |
510 | ASSERT(spa_sync_pass(dp->dp_spa) == 1 || | |
511 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] == 0); | |
512 | dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] = 0; | |
513 | mutex_exit(&dp->dp_lock); | |
514 | ||
29809a6c MA |
515 | /* |
516 | * After the data blocks have been written (ensured by the zio_wait() | |
517 | * above), update the user/group space accounting. | |
518 | */ | |
e8b96c60 MA |
519 | for (ds = list_head(&synced_datasets); ds != NULL; |
520 | ds = list_next(&synced_datasets, ds)) { | |
428870ff | 521 | dmu_objset_do_userquota_updates(ds->ds_objset, tx); |
e8b96c60 | 522 | } |
9babb374 BB |
523 | |
524 | /* | |
525 | * Sync the datasets again to push out the changes due to | |
428870ff | 526 | * userspace updates. This must be done before we process the |
29809a6c MA |
527 | * sync tasks, so that any snapshots will have the correct |
528 | * user accounting information (and we won't get confused | |
529 | * about which blocks are part of the snapshot). | |
9babb374 BB |
530 | */ |
531 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
e8b96c60 | 532 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) { |
9babb374 BB |
533 | ASSERT(list_link_active(&ds->ds_synced_link)); |
534 | dmu_buf_rele(ds->ds_dbuf, ds); | |
535 | dsl_dataset_sync(ds, zio, tx); | |
536 | } | |
e8b96c60 | 537 | VERIFY0(zio_wait(zio)); |
9babb374 | 538 | |
428870ff | 539 | /* |
29809a6c MA |
540 | * Now that the datasets have been completely synced, we can |
541 | * clean up our in-memory structures accumulated while syncing: | |
542 | * | |
543 | * - move dead blocks from the pending deadlist to the on-disk deadlist | |
29809a6c | 544 | * - release hold from dsl_dataset_dirty() |
428870ff | 545 | */ |
e8b96c60 | 546 | while ((ds = list_remove_head(&synced_datasets)) != NULL) { |
0efd9791 | 547 | dsl_dataset_sync_done(ds, tx); |
428870ff BB |
548 | } |
549 | ||
e8b96c60 | 550 | while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) { |
34dc7c2f | 551 | dsl_dir_sync(dd, tx); |
e8b96c60 | 552 | } |
b128c09f | 553 | |
29809a6c MA |
554 | /* |
555 | * The MOS's space is accounted for in the pool/$MOS | |
556 | * (dp_mos_dir). We can't modify the mos while we're syncing | |
557 | * it, so we remember the deltas and apply them here. | |
558 | */ | |
559 | if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || | |
560 | dp->dp_mos_uncompressed_delta != 0) { | |
561 | dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, | |
562 | dp->dp_mos_used_delta, | |
563 | dp->dp_mos_compressed_delta, | |
564 | dp->dp_mos_uncompressed_delta, tx); | |
565 | dp->dp_mos_used_delta = 0; | |
566 | dp->dp_mos_compressed_delta = 0; | |
567 | dp->dp_mos_uncompressed_delta = 0; | |
568 | } | |
569 | ||
428870ff BB |
570 | if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || |
571 | list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { | |
e8b96c60 | 572 | dsl_pool_sync_mos(dp, tx); |
34dc7c2f BB |
573 | } |
574 | ||
29809a6c MA |
575 | /* |
576 | * If we modify a dataset in the same txg that we want to destroy it, | |
577 | * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. | |
578 | * dsl_dir_destroy_check() will fail if there are unexpected holds. | |
579 | * Therefore, we want to sync the MOS (thus syncing the dd_dbuf | |
580 | * and clearing the hold on it) before we process the sync_tasks. | |
581 | * The MOS data dirtied by the sync_tasks will be synced on the next | |
582 | * pass. | |
583 | */ | |
29809a6c | 584 | if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { |
13fe0198 | 585 | dsl_sync_task_t *dst; |
29809a6c MA |
586 | /* |
587 | * No more sync tasks should have been added while we | |
588 | * were syncing. | |
589 | */ | |
e8b96c60 MA |
590 | ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1); |
591 | while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL) | |
13fe0198 | 592 | dsl_sync_task_sync(dst, tx); |
29809a6c MA |
593 | } |
594 | ||
34dc7c2f | 595 | dmu_tx_commit(tx); |
b128c09f | 596 | |
e8b96c60 | 597 | DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg); |
34dc7c2f BB |
598 | } |
599 | ||
600 | void | |
428870ff | 601 | dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) |
34dc7c2f | 602 | { |
29809a6c | 603 | zilog_t *zilog; |
34dc7c2f | 604 | |
29809a6c | 605 | while ((zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg))) { |
e8b96c60 | 606 | dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os); |
29809a6c MA |
607 | zil_clean(zilog, txg); |
608 | ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); | |
609 | dmu_buf_rele(ds->ds_dbuf, zilog); | |
34dc7c2f | 610 | } |
428870ff | 611 | ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); |
34dc7c2f BB |
612 | } |
613 | ||
614 | /* | |
615 | * TRUE if the current thread is the tx_sync_thread or if we | |
616 | * are being called from SPA context during pool initialization. | |
617 | */ | |
618 | int | |
619 | dsl_pool_sync_context(dsl_pool_t *dp) | |
620 | { | |
621 | return (curthread == dp->dp_tx.tx_sync_thread || | |
9ae529ec | 622 | spa_is_initializing(dp->dp_spa)); |
34dc7c2f BB |
623 | } |
624 | ||
625 | uint64_t | |
626 | dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) | |
627 | { | |
628 | uint64_t space, resv; | |
629 | ||
630 | /* | |
34dc7c2f BB |
631 | * If we're trying to assess whether it's OK to do a free, |
632 | * cut the reservation in half to allow forward progress | |
633 | * (e.g. make it possible to rm(1) files from a full pool). | |
634 | */ | |
635 | space = spa_get_dspace(dp->dp_spa); | |
0c60cc32 | 636 | resv = spa_get_slop_space(dp->dp_spa); |
34dc7c2f BB |
637 | if (netfree) |
638 | resv >>= 1; | |
639 | ||
640 | return (space - resv); | |
641 | } | |
642 | ||
e8b96c60 MA |
643 | boolean_t |
644 | dsl_pool_need_dirty_delay(dsl_pool_t *dp) | |
34dc7c2f | 645 | { |
e8b96c60 MA |
646 | uint64_t delay_min_bytes = |
647 | zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100; | |
648 | boolean_t rv; | |
34dc7c2f | 649 | |
e8b96c60 MA |
650 | mutex_enter(&dp->dp_lock); |
651 | if (dp->dp_dirty_total > zfs_dirty_data_sync) | |
652 | txg_kick(dp); | |
653 | rv = (dp->dp_dirty_total > delay_min_bytes); | |
654 | mutex_exit(&dp->dp_lock); | |
655 | return (rv); | |
34dc7c2f BB |
656 | } |
657 | ||
658 | void | |
e8b96c60 | 659 | dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) |
34dc7c2f | 660 | { |
e8b96c60 MA |
661 | if (space > 0) { |
662 | mutex_enter(&dp->dp_lock); | |
663 | dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space; | |
664 | dsl_pool_dirty_delta(dp, space); | |
665 | mutex_exit(&dp->dp_lock); | |
666 | } | |
34dc7c2f BB |
667 | } |
668 | ||
669 | void | |
e8b96c60 | 670 | dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg) |
34dc7c2f | 671 | { |
e8b96c60 MA |
672 | ASSERT3S(space, >=, 0); |
673 | if (space == 0) | |
34dc7c2f BB |
674 | return; |
675 | ||
e8b96c60 MA |
676 | mutex_enter(&dp->dp_lock); |
677 | if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) { | |
678 | /* XXX writing something we didn't dirty? */ | |
679 | space = dp->dp_dirty_pertxg[txg & TXG_MASK]; | |
34dc7c2f | 680 | } |
e8b96c60 MA |
681 | ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space); |
682 | dp->dp_dirty_pertxg[txg & TXG_MASK] -= space; | |
683 | ASSERT3U(dp->dp_dirty_total, >=, space); | |
684 | dsl_pool_dirty_delta(dp, -space); | |
685 | mutex_exit(&dp->dp_lock); | |
34dc7c2f | 686 | } |
b128c09f BB |
687 | |
688 | /* ARGSUSED */ | |
689 | static int | |
13fe0198 | 690 | upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) |
b128c09f BB |
691 | { |
692 | dmu_tx_t *tx = arg; | |
693 | dsl_dataset_t *ds, *prev = NULL; | |
694 | int err; | |
b128c09f | 695 | |
13fe0198 | 696 | err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); |
b128c09f BB |
697 | if (err) |
698 | return (err); | |
699 | ||
d683ddbb JG |
700 | while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { |
701 | err = dsl_dataset_hold_obj(dp, | |
702 | dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); | |
b128c09f BB |
703 | if (err) { |
704 | dsl_dataset_rele(ds, FTAG); | |
705 | return (err); | |
706 | } | |
707 | ||
d683ddbb | 708 | if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) |
b128c09f BB |
709 | break; |
710 | dsl_dataset_rele(ds, FTAG); | |
711 | ds = prev; | |
712 | prev = NULL; | |
713 | } | |
714 | ||
715 | if (prev == NULL) { | |
716 | prev = dp->dp_origin_snap; | |
717 | ||
718 | /* | |
719 | * The $ORIGIN can't have any data, or the accounting | |
720 | * will be wrong. | |
721 | */ | |
cc9bb3e5 | 722 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
d683ddbb | 723 | ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth); |
cc9bb3e5 | 724 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
b128c09f BB |
725 | |
726 | /* The origin doesn't get attached to itself */ | |
727 | if (ds->ds_object == prev->ds_object) { | |
728 | dsl_dataset_rele(ds, FTAG); | |
729 | return (0); | |
730 | } | |
731 | ||
732 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
d683ddbb JG |
733 | dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object; |
734 | dsl_dataset_phys(ds)->ds_prev_snap_txg = | |
735 | dsl_dataset_phys(prev)->ds_creation_txg; | |
b128c09f BB |
736 | |
737 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); | |
d683ddbb | 738 | dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object; |
b128c09f BB |
739 | |
740 | dmu_buf_will_dirty(prev->ds_dbuf, tx); | |
d683ddbb | 741 | dsl_dataset_phys(prev)->ds_num_children++; |
b128c09f | 742 | |
d683ddbb | 743 | if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) { |
b128c09f | 744 | ASSERT(ds->ds_prev == NULL); |
13fe0198 | 745 | VERIFY0(dsl_dataset_hold_obj(dp, |
d683ddbb JG |
746 | dsl_dataset_phys(ds)->ds_prev_snap_obj, |
747 | ds, &ds->ds_prev)); | |
b128c09f BB |
748 | } |
749 | } | |
750 | ||
d683ddbb JG |
751 | ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object); |
752 | ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object); | |
b128c09f | 753 | |
d683ddbb | 754 | if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) { |
428870ff | 755 | dmu_buf_will_dirty(prev->ds_dbuf, tx); |
d683ddbb | 756 | dsl_dataset_phys(prev)->ds_next_clones_obj = |
b128c09f BB |
757 | zap_create(dp->dp_meta_objset, |
758 | DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); | |
759 | } | |
13fe0198 | 760 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
d683ddbb | 761 | dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx)); |
b128c09f BB |
762 | |
763 | dsl_dataset_rele(ds, FTAG); | |
764 | if (prev != dp->dp_origin_snap) | |
765 | dsl_dataset_rele(prev, FTAG); | |
766 | return (0); | |
767 | } | |
768 | ||
769 | void | |
770 | dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
771 | { | |
772 | ASSERT(dmu_tx_is_syncing(tx)); | |
773 | ASSERT(dp->dp_origin_snap != NULL); | |
774 | ||
13fe0198 | 775 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, |
9c43027b | 776 | tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); |
428870ff BB |
777 | } |
778 | ||
779 | /* ARGSUSED */ | |
780 | static int | |
13fe0198 | 781 | upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
428870ff BB |
782 | { |
783 | dmu_tx_t *tx = arg; | |
428870ff BB |
784 | objset_t *mos = dp->dp_meta_objset; |
785 | ||
d683ddbb | 786 | if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) { |
428870ff BB |
787 | dsl_dataset_t *origin; |
788 | ||
13fe0198 | 789 | VERIFY0(dsl_dataset_hold_obj(dp, |
d683ddbb | 790 | dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin)); |
428870ff | 791 | |
d683ddbb | 792 | if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) { |
428870ff | 793 | dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); |
d683ddbb JG |
794 | dsl_dir_phys(origin->ds_dir)->dd_clones = |
795 | zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE, | |
796 | 0, tx); | |
428870ff BB |
797 | } |
798 | ||
13fe0198 | 799 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
d683ddbb JG |
800 | dsl_dir_phys(origin->ds_dir)->dd_clones, |
801 | ds->ds_object, tx)); | |
428870ff BB |
802 | |
803 | dsl_dataset_rele(origin, FTAG); | |
804 | } | |
428870ff BB |
805 | return (0); |
806 | } | |
807 | ||
808 | void | |
809 | dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
810 | { | |
428870ff BB |
811 | uint64_t obj; |
812 | ||
d6320ddb BB |
813 | ASSERT(dmu_tx_is_syncing(tx)); |
814 | ||
428870ff | 815 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); |
13fe0198 | 816 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
817 | FREE_DIR_NAME, &dp->dp_free_dir)); |
818 | ||
819 | /* | |
820 | * We can't use bpobj_alloc(), because spa_version() still | |
821 | * returns the old version, and we need a new-version bpobj with | |
822 | * subobj support. So call dmu_object_alloc() directly. | |
823 | */ | |
824 | obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, | |
f1512ee6 | 825 | SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); |
13fe0198 | 826 | VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 827 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); |
13fe0198 | 828 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); |
428870ff | 829 | |
13fe0198 | 830 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
9c43027b | 831 | upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN | DS_FIND_SERIALIZE)); |
b128c09f BB |
832 | } |
833 | ||
834 | void | |
835 | dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) | |
836 | { | |
837 | uint64_t dsobj; | |
838 | dsl_dataset_t *ds; | |
839 | ||
840 | ASSERT(dmu_tx_is_syncing(tx)); | |
841 | ASSERT(dp->dp_origin_snap == NULL); | |
13fe0198 | 842 | ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); |
b128c09f BB |
843 | |
844 | /* create the origin dir, ds, & snap-ds */ | |
b128c09f BB |
845 | dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, |
846 | NULL, 0, kcred, tx); | |
13fe0198 MA |
847 | VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); |
848 | dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); | |
d683ddbb | 849 | VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj, |
b128c09f BB |
850 | dp, &dp->dp_origin_snap)); |
851 | dsl_dataset_rele(ds, FTAG); | |
b128c09f | 852 | } |
9babb374 BB |
853 | |
854 | taskq_t * | |
3558fd73 | 855 | dsl_pool_iput_taskq(dsl_pool_t *dp) |
9babb374 | 856 | { |
3558fd73 | 857 | return (dp->dp_iput_taskq); |
9babb374 | 858 | } |
428870ff BB |
859 | |
860 | /* | |
861 | * Walk through the pool-wide zap object of temporary snapshot user holds | |
862 | * and release them. | |
863 | */ | |
864 | void | |
865 | dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) | |
866 | { | |
867 | zap_attribute_t za; | |
868 | zap_cursor_t zc; | |
869 | objset_t *mos = dp->dp_meta_objset; | |
870 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
95fd54a1 | 871 | nvlist_t *holds; |
428870ff BB |
872 | |
873 | if (zapobj == 0) | |
874 | return; | |
875 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
876 | ||
95fd54a1 SH |
877 | holds = fnvlist_alloc(); |
878 | ||
428870ff BB |
879 | for (zap_cursor_init(&zc, mos, zapobj); |
880 | zap_cursor_retrieve(&zc, &za) == 0; | |
881 | zap_cursor_advance(&zc)) { | |
882 | char *htag; | |
95fd54a1 | 883 | nvlist_t *tags; |
428870ff BB |
884 | |
885 | htag = strchr(za.za_name, '-'); | |
886 | *htag = '\0'; | |
887 | ++htag; | |
95fd54a1 SH |
888 | if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) { |
889 | tags = fnvlist_alloc(); | |
890 | fnvlist_add_boolean(tags, htag); | |
891 | fnvlist_add_nvlist(holds, za.za_name, tags); | |
892 | fnvlist_free(tags); | |
893 | } else { | |
894 | fnvlist_add_boolean(tags, htag); | |
895 | } | |
428870ff | 896 | } |
95fd54a1 SH |
897 | dsl_dataset_user_release_tmp(dp, holds); |
898 | fnvlist_free(holds); | |
428870ff BB |
899 | zap_cursor_fini(&zc); |
900 | } | |
901 | ||
902 | /* | |
903 | * Create the pool-wide zap object for storing temporary snapshot holds. | |
904 | */ | |
905 | void | |
906 | dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) | |
907 | { | |
908 | objset_t *mos = dp->dp_meta_objset; | |
909 | ||
910 | ASSERT(dp->dp_tmp_userrefs_obj == 0); | |
911 | ASSERT(dmu_tx_is_syncing(tx)); | |
912 | ||
9ae529ec CS |
913 | dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, |
914 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); | |
428870ff BB |
915 | } |
916 | ||
917 | static int | |
918 | dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, | |
13fe0198 | 919 | const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) |
428870ff BB |
920 | { |
921 | objset_t *mos = dp->dp_meta_objset; | |
922 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
923 | char *name; | |
924 | int error; | |
925 | ||
926 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
927 | ASSERT(dmu_tx_is_syncing(tx)); | |
928 | ||
929 | /* | |
930 | * If the pool was created prior to SPA_VERSION_USERREFS, the | |
931 | * zap object for temporary holds might not exist yet. | |
932 | */ | |
933 | if (zapobj == 0) { | |
934 | if (holding) { | |
935 | dsl_pool_user_hold_create_obj(dp, tx); | |
936 | zapobj = dp->dp_tmp_userrefs_obj; | |
937 | } else { | |
2e528b49 | 938 | return (SET_ERROR(ENOENT)); |
428870ff BB |
939 | } |
940 | } | |
941 | ||
942 | name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); | |
943 | if (holding) | |
13fe0198 | 944 | error = zap_add(mos, zapobj, name, 8, 1, &now, tx); |
428870ff BB |
945 | else |
946 | error = zap_remove(mos, zapobj, name, tx); | |
947 | strfree(name); | |
948 | ||
949 | return (error); | |
950 | } | |
951 | ||
952 | /* | |
953 | * Add a temporary hold for the given dataset object and tag. | |
954 | */ | |
955 | int | |
956 | dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
13fe0198 | 957 | uint64_t now, dmu_tx_t *tx) |
428870ff BB |
958 | { |
959 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); | |
960 | } | |
961 | ||
962 | /* | |
963 | * Release a temporary hold for the given dataset object and tag. | |
964 | */ | |
965 | int | |
966 | dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
967 | dmu_tx_t *tx) | |
968 | { | |
13fe0198 | 969 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, |
428870ff BB |
970 | tx, B_FALSE)); |
971 | } | |
c409e464 | 972 | |
13fe0198 MA |
973 | /* |
974 | * DSL Pool Configuration Lock | |
975 | * | |
976 | * The dp_config_rwlock protects against changes to DSL state (e.g. dataset | |
977 | * creation / destruction / rename / property setting). It must be held for | |
978 | * read to hold a dataset or dsl_dir. I.e. you must call | |
979 | * dsl_pool_config_enter() or dsl_pool_hold() before calling | |
980 | * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock | |
981 | * must be held continuously until all datasets and dsl_dirs are released. | |
982 | * | |
983 | * The only exception to this rule is that if a "long hold" is placed on | |
984 | * a dataset, then the dp_config_rwlock may be dropped while the dataset | |
985 | * is still held. The long hold will prevent the dataset from being | |
986 | * destroyed -- the destroy will fail with EBUSY. A long hold can be | |
987 | * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset | |
988 | * (by calling dsl_{dataset,objset}_{try}own{_obj}). | |
989 | * | |
990 | * Legitimate long-holders (including owners) should be long-running, cancelable | |
991 | * tasks that should cause "zfs destroy" to fail. This includes DMU | |
992 | * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), | |
993 | * "zfs send", and "zfs diff". There are several other long-holders whose | |
994 | * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). | |
995 | * | |
996 | * The usual formula for long-holding would be: | |
997 | * dsl_pool_hold() | |
998 | * dsl_dataset_hold() | |
999 | * ... perform checks ... | |
1000 | * dsl_dataset_long_hold() | |
1001 | * dsl_pool_rele() | |
1002 | * ... perform long-running task ... | |
1003 | * dsl_dataset_long_rele() | |
1004 | * dsl_dataset_rele() | |
1005 | * | |
1006 | * Note that when the long hold is released, the dataset is still held but | |
1007 | * the pool is not held. The dataset may change arbitrarily during this time | |
1008 | * (e.g. it could be destroyed). Therefore you shouldn't do anything to the | |
1009 | * dataset except release it. | |
1010 | * | |
1011 | * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only | |
1012 | * or modifying operations. | |
1013 | * | |
1014 | * Modifying operations should generally use dsl_sync_task(). The synctask | |
1015 | * infrastructure enforces proper locking strategy with respect to the | |
1016 | * dp_config_rwlock. See the comment above dsl_sync_task() for details. | |
1017 | * | |
1018 | * Read-only operations will manually hold the pool, then the dataset, obtain | |
1019 | * information from the dataset, then release the pool and dataset. | |
1020 | * dmu_objset_{hold,rele}() are convenience routines that also do the pool | |
1021 | * hold/rele. | |
1022 | */ | |
1023 | ||
1024 | int | |
1025 | dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) | |
1026 | { | |
1027 | spa_t *spa; | |
1028 | int error; | |
1029 | ||
1030 | error = spa_open(name, &spa, tag); | |
1031 | if (error == 0) { | |
1032 | *dp = spa_get_dsl(spa); | |
1033 | dsl_pool_config_enter(*dp, tag); | |
1034 | } | |
1035 | return (error); | |
1036 | } | |
1037 | ||
1038 | void | |
1039 | dsl_pool_rele(dsl_pool_t *dp, void *tag) | |
1040 | { | |
1041 | dsl_pool_config_exit(dp, tag); | |
1042 | spa_close(dp->dp_spa, tag); | |
1043 | } | |
1044 | ||
1045 | void | |
1046 | dsl_pool_config_enter(dsl_pool_t *dp, void *tag) | |
1047 | { | |
1048 | /* | |
1049 | * We use a "reentrant" reader-writer lock, but not reentrantly. | |
1050 | * | |
1051 | * The rrwlock can (with the track_all flag) track all reading threads, | |
1052 | * which is very useful for debugging which code path failed to release | |
1053 | * the lock, and for verifying that the *current* thread does hold | |
1054 | * the lock. | |
1055 | * | |
1056 | * (Unlike a rwlock, which knows that N threads hold it for | |
1057 | * read, but not *which* threads, so rw_held(RW_READER) returns TRUE | |
1058 | * if any thread holds it for read, even if this thread doesn't). | |
1059 | */ | |
1060 | ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); | |
1061 | rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); | |
1062 | } | |
1063 | ||
5e8cd5d1 AJ |
1064 | void |
1065 | dsl_pool_config_enter_prio(dsl_pool_t *dp, void *tag) | |
1066 | { | |
1067 | ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); | |
1068 | rrw_enter_read_prio(&dp->dp_config_rwlock, tag); | |
1069 | } | |
1070 | ||
13fe0198 MA |
1071 | void |
1072 | dsl_pool_config_exit(dsl_pool_t *dp, void *tag) | |
1073 | { | |
1074 | rrw_exit(&dp->dp_config_rwlock, tag); | |
1075 | } | |
1076 | ||
1077 | boolean_t | |
1078 | dsl_pool_config_held(dsl_pool_t *dp) | |
1079 | { | |
1080 | return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); | |
1081 | } | |
1082 | ||
9c43027b AJ |
1083 | boolean_t |
1084 | dsl_pool_config_held_writer(dsl_pool_t *dp) | |
1085 | { | |
1086 | return (RRW_WRITE_HELD(&dp->dp_config_rwlock)); | |
1087 | } | |
1088 | ||
c409e464 | 1089 | #if defined(_KERNEL) && defined(HAVE_SPL) |
40a806df NB |
1090 | EXPORT_SYMBOL(dsl_pool_config_enter); |
1091 | EXPORT_SYMBOL(dsl_pool_config_exit); | |
1092 | ||
02730c33 | 1093 | /* BEGIN CSTYLED */ |
d1d7e268 | 1094 | /* zfs_dirty_data_max_percent only applied at module load in arc_init(). */ |
e8b96c60 MA |
1095 | module_param(zfs_dirty_data_max_percent, int, 0444); |
1096 | MODULE_PARM_DESC(zfs_dirty_data_max_percent, "percent of ram can be dirty"); | |
c409e464 | 1097 | |
d1d7e268 | 1098 | /* zfs_dirty_data_max_max_percent only applied at module load in arc_init(). */ |
e8b96c60 MA |
1099 | module_param(zfs_dirty_data_max_max_percent, int, 0444); |
1100 | MODULE_PARM_DESC(zfs_dirty_data_max_max_percent, | |
d1d7e268 | 1101 | "zfs_dirty_data_max upper bound as % of RAM"); |
c409e464 | 1102 | |
e8b96c60 MA |
1103 | module_param(zfs_delay_min_dirty_percent, int, 0644); |
1104 | MODULE_PARM_DESC(zfs_delay_min_dirty_percent, "transaction delay threshold"); | |
c409e464 | 1105 | |
e8b96c60 MA |
1106 | module_param(zfs_dirty_data_max, ulong, 0644); |
1107 | MODULE_PARM_DESC(zfs_dirty_data_max, "determines the dirty space limit"); | |
c409e464 | 1108 | |
d1d7e268 | 1109 | /* zfs_dirty_data_max_max only applied at module load in arc_init(). */ |
e8b96c60 MA |
1110 | module_param(zfs_dirty_data_max_max, ulong, 0444); |
1111 | MODULE_PARM_DESC(zfs_dirty_data_max_max, | |
d1d7e268 | 1112 | "zfs_dirty_data_max upper bound in bytes"); |
c409e464 | 1113 | |
e8b96c60 MA |
1114 | module_param(zfs_dirty_data_sync, ulong, 0644); |
1115 | MODULE_PARM_DESC(zfs_dirty_data_sync, "sync txg when this much dirty data"); | |
c409e464 | 1116 | |
e8b96c60 MA |
1117 | module_param(zfs_delay_scale, ulong, 0644); |
1118 | MODULE_PARM_DESC(zfs_delay_scale, "how quickly delay approaches infinity"); | |
02730c33 | 1119 | /* END CSTYLED */ |
c409e464 | 1120 | #endif |