]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
2e528b49 | 23 | * Copyright (c) 2013 by Delphix. All rights reserved. |
34dc7c2f BB |
24 | */ |
25 | ||
34dc7c2f BB |
26 | #include <sys/dsl_pool.h> |
27 | #include <sys/dsl_dataset.h> | |
428870ff | 28 | #include <sys/dsl_prop.h> |
34dc7c2f BB |
29 | #include <sys/dsl_dir.h> |
30 | #include <sys/dsl_synctask.h> | |
428870ff BB |
31 | #include <sys/dsl_scan.h> |
32 | #include <sys/dnode.h> | |
34dc7c2f BB |
33 | #include <sys/dmu_tx.h> |
34 | #include <sys/dmu_objset.h> | |
35 | #include <sys/arc.h> | |
36 | #include <sys/zap.h> | |
37 | #include <sys/zio.h> | |
38 | #include <sys/zfs_context.h> | |
39 | #include <sys/fs/zfs.h> | |
b128c09f BB |
40 | #include <sys/zfs_znode.h> |
41 | #include <sys/spa_impl.h> | |
428870ff | 42 | #include <sys/dsl_deadlist.h> |
9ae529ec CS |
43 | #include <sys/bptree.h> |
44 | #include <sys/zfeature.h> | |
29809a6c | 45 | #include <sys/zil_impl.h> |
13fe0198 | 46 | #include <sys/dsl_userhold.h> |
34dc7c2f BB |
47 | |
48 | int zfs_no_write_throttle = 0; | |
b128c09f | 49 | int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ |
572e2857 | 50 | int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */ |
b128c09f | 51 | |
c409e464 BB |
52 | unsigned long zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */ |
53 | unsigned long zfs_write_limit_max = 0; /* max data payload per txg */ | |
54 | unsigned long zfs_write_limit_inflated = 0; | |
55 | unsigned long zfs_write_limit_override = 0; | |
34dc7c2f | 56 | |
b128c09f BB |
57 | kmutex_t zfs_write_limit_lock; |
58 | ||
59 | static pgcnt_t old_physmem = 0; | |
60 | ||
63fd3c6c AL |
61 | hrtime_t zfs_throttle_delay = MSEC2NSEC(10); |
62 | hrtime_t zfs_throttle_resolution = MSEC2NSEC(10); | |
63 | ||
428870ff | 64 | int |
b128c09f | 65 | dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) |
34dc7c2f BB |
66 | { |
67 | uint64_t obj; | |
68 | int err; | |
69 | ||
70 | err = zap_lookup(dp->dp_meta_objset, | |
71 | dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, | |
b128c09f | 72 | name, sizeof (obj), 1, &obj); |
34dc7c2f BB |
73 | if (err) |
74 | return (err); | |
75 | ||
13fe0198 | 76 | return (dsl_dir_hold_obj(dp, obj, name, dp, ddp)); |
34dc7c2f BB |
77 | } |
78 | ||
79 | static dsl_pool_t * | |
80 | dsl_pool_open_impl(spa_t *spa, uint64_t txg) | |
81 | { | |
82 | dsl_pool_t *dp; | |
83 | blkptr_t *bp = spa_get_rootblkptr(spa); | |
34dc7c2f BB |
84 | |
85 | dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); | |
86 | dp->dp_spa = spa; | |
87 | dp->dp_meta_rootbp = *bp; | |
13fe0198 | 88 | rrw_init(&dp->dp_config_rwlock, B_TRUE); |
34dc7c2f BB |
89 | dp->dp_write_limit = zfs_write_limit_min; |
90 | txg_init(dp, txg); | |
91 | ||
92 | txg_list_create(&dp->dp_dirty_datasets, | |
93 | offsetof(dsl_dataset_t, ds_dirty_link)); | |
29809a6c MA |
94 | txg_list_create(&dp->dp_dirty_zilogs, |
95 | offsetof(zilog_t, zl_dirty_link)); | |
34dc7c2f BB |
96 | txg_list_create(&dp->dp_dirty_dirs, |
97 | offsetof(dsl_dir_t, dd_dirty_link)); | |
98 | txg_list_create(&dp->dp_sync_tasks, | |
13fe0198 | 99 | offsetof(dsl_sync_task_t, dst_node)); |
34dc7c2f BB |
100 | |
101 | mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); | |
102 | ||
3558fd73 | 103 | dp->dp_iput_taskq = taskq_create("zfs_iput_taskq", 1, minclsyspri, |
9babb374 BB |
104 | 1, 4, 0); |
105 | ||
34dc7c2f BB |
106 | return (dp); |
107 | } | |
108 | ||
109 | int | |
9ae529ec | 110 | dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) |
34dc7c2f BB |
111 | { |
112 | int err; | |
113 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
9ae529ec CS |
114 | |
115 | err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, | |
116 | &dp->dp_meta_objset); | |
117 | if (err != 0) | |
118 | dsl_pool_close(dp); | |
119 | else | |
120 | *dpp = dp; | |
121 | ||
122 | return (err); | |
123 | } | |
124 | ||
125 | int | |
126 | dsl_pool_open(dsl_pool_t *dp) | |
127 | { | |
128 | int err; | |
b128c09f BB |
129 | dsl_dir_t *dd; |
130 | dsl_dataset_t *ds; | |
428870ff | 131 | uint64_t obj; |
34dc7c2f | 132 | |
13fe0198 | 133 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
34dc7c2f BB |
134 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
135 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, | |
136 | &dp->dp_root_dir_obj); | |
137 | if (err) | |
138 | goto out; | |
139 | ||
13fe0198 | 140 | err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
141 | NULL, dp, &dp->dp_root_dir); |
142 | if (err) | |
143 | goto out; | |
144 | ||
b128c09f | 145 | err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); |
34dc7c2f BB |
146 | if (err) |
147 | goto out; | |
148 | ||
9ae529ec | 149 | if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) { |
b128c09f BB |
150 | err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); |
151 | if (err) | |
152 | goto out; | |
153 | err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, | |
154 | FTAG, &ds); | |
9babb374 BB |
155 | if (err == 0) { |
156 | err = dsl_dataset_hold_obj(dp, | |
157 | ds->ds_phys->ds_prev_snap_obj, dp, | |
158 | &dp->dp_origin_snap); | |
159 | dsl_dataset_rele(ds, FTAG); | |
160 | } | |
13fe0198 | 161 | dsl_dir_rele(dd, dp); |
b128c09f BB |
162 | if (err) |
163 | goto out; | |
b128c09f BB |
164 | } |
165 | ||
9ae529ec | 166 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { |
428870ff BB |
167 | err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, |
168 | &dp->dp_free_dir); | |
b128c09f BB |
169 | if (err) |
170 | goto out; | |
428870ff | 171 | |
b128c09f | 172 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 173 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); |
b128c09f BB |
174 | if (err) |
175 | goto out; | |
13fe0198 | 176 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff | 177 | dp->dp_meta_objset, obj)); |
b128c09f BB |
178 | } |
179 | ||
9ae529ec CS |
180 | if (spa_feature_is_active(dp->dp_spa, |
181 | &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) { | |
182 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
183 | DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1, | |
184 | &dp->dp_bptree_obj); | |
185 | if (err != 0) | |
186 | goto out; | |
187 | } | |
188 | ||
753c3839 MA |
189 | if (spa_feature_is_active(dp->dp_spa, |
190 | &spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ])) { | |
191 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
192 | DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1, | |
193 | &dp->dp_empty_bpobj); | |
194 | if (err != 0) | |
195 | goto out; | |
196 | } | |
197 | ||
428870ff BB |
198 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
199 | DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, | |
200 | &dp->dp_tmp_userrefs_obj); | |
201 | if (err == ENOENT) | |
202 | err = 0; | |
203 | if (err) | |
204 | goto out; | |
205 | ||
9ae529ec | 206 | err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg); |
428870ff | 207 | |
34dc7c2f | 208 | out: |
13fe0198 | 209 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
34dc7c2f BB |
210 | return (err); |
211 | } | |
212 | ||
213 | void | |
214 | dsl_pool_close(dsl_pool_t *dp) | |
215 | { | |
b128c09f BB |
216 | /* drop our references from dsl_pool_open() */ |
217 | ||
218 | /* | |
219 | * Since we held the origin_snap from "syncing" context (which | |
220 | * includes pool-opening context), it actually only got a "ref" | |
221 | * and not a hold, so just drop that here. | |
222 | */ | |
223 | if (dp->dp_origin_snap) | |
13fe0198 | 224 | dsl_dataset_rele(dp->dp_origin_snap, dp); |
34dc7c2f | 225 | if (dp->dp_mos_dir) |
13fe0198 | 226 | dsl_dir_rele(dp->dp_mos_dir, dp); |
428870ff | 227 | if (dp->dp_free_dir) |
13fe0198 | 228 | dsl_dir_rele(dp->dp_free_dir, dp); |
34dc7c2f | 229 | if (dp->dp_root_dir) |
13fe0198 | 230 | dsl_dir_rele(dp->dp_root_dir, dp); |
34dc7c2f | 231 | |
428870ff BB |
232 | bpobj_close(&dp->dp_free_bpobj); |
233 | ||
34dc7c2f BB |
234 | /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ |
235 | if (dp->dp_meta_objset) | |
428870ff | 236 | dmu_objset_evict(dp->dp_meta_objset); |
34dc7c2f BB |
237 | |
238 | txg_list_destroy(&dp->dp_dirty_datasets); | |
29809a6c | 239 | txg_list_destroy(&dp->dp_dirty_zilogs); |
428870ff | 240 | txg_list_destroy(&dp->dp_sync_tasks); |
34dc7c2f | 241 | txg_list_destroy(&dp->dp_dirty_dirs); |
34dc7c2f BB |
242 | |
243 | arc_flush(dp->dp_spa); | |
244 | txg_fini(dp); | |
428870ff | 245 | dsl_scan_fini(dp); |
13fe0198 | 246 | rrw_destroy(&dp->dp_config_rwlock); |
34dc7c2f | 247 | mutex_destroy(&dp->dp_lock); |
3558fd73 | 248 | taskq_destroy(dp->dp_iput_taskq); |
b128c09f BB |
249 | if (dp->dp_blkstats) |
250 | kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); | |
34dc7c2f BB |
251 | kmem_free(dp, sizeof (dsl_pool_t)); |
252 | } | |
253 | ||
254 | dsl_pool_t * | |
b128c09f | 255 | dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) |
34dc7c2f BB |
256 | { |
257 | int err; | |
258 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
259 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
428870ff | 260 | objset_t *os; |
b128c09f | 261 | dsl_dataset_t *ds; |
428870ff | 262 | uint64_t obj; |
b128c09f | 263 | |
13fe0198 MA |
264 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); |
265 | ||
b128c09f | 266 | /* create and open the MOS (meta-objset) */ |
428870ff BB |
267 | dp->dp_meta_objset = dmu_objset_create_impl(spa, |
268 | NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); | |
34dc7c2f BB |
269 | |
270 | /* create the pool directory */ | |
271 | err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
272 | DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); | |
c99c9001 | 273 | ASSERT0(err); |
34dc7c2f | 274 | |
428870ff | 275 | /* Initialize scan structures */ |
13fe0198 | 276 | VERIFY0(dsl_scan_init(dp, txg)); |
428870ff | 277 | |
34dc7c2f | 278 | /* create and open the root dir */ |
b128c09f | 279 | dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); |
13fe0198 | 280 | VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, |
34dc7c2f BB |
281 | NULL, dp, &dp->dp_root_dir)); |
282 | ||
283 | /* create and open the meta-objset dir */ | |
b128c09f | 284 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); |
13fe0198 | 285 | VERIFY0(dsl_pool_open_special_dir(dp, |
b128c09f BB |
286 | MOS_DIR_NAME, &dp->dp_mos_dir)); |
287 | ||
428870ff BB |
288 | if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { |
289 | /* create and open the free dir */ | |
290 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, | |
291 | FREE_DIR_NAME, tx); | |
13fe0198 | 292 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
293 | FREE_DIR_NAME, &dp->dp_free_dir)); |
294 | ||
295 | /* create and open the free_bplist */ | |
296 | obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); | |
297 | VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
298 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); | |
13fe0198 | 299 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, |
428870ff BB |
300 | dp->dp_meta_objset, obj)); |
301 | } | |
302 | ||
b128c09f BB |
303 | if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) |
304 | dsl_pool_create_origin(dp, tx); | |
305 | ||
306 | /* create the root dataset */ | |
428870ff | 307 | obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); |
b128c09f BB |
308 | |
309 | /* create the root objset */ | |
13fe0198 | 310 | VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); |
0fe3d820 BB |
311 | VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds, |
312 | dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx))); | |
b128c09f | 313 | #ifdef _KERNEL |
428870ff | 314 | zfs_create_fs(os, kcred, zplprops, tx); |
b128c09f BB |
315 | #endif |
316 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
317 | |
318 | dmu_tx_commit(tx); | |
319 | ||
13fe0198 MA |
320 | rrw_exit(&dp->dp_config_rwlock, FTAG); |
321 | ||
34dc7c2f BB |
322 | return (dp); |
323 | } | |
324 | ||
29809a6c MA |
325 | /* |
326 | * Account for the meta-objset space in its placeholder dsl_dir. | |
327 | */ | |
328 | void | |
329 | dsl_pool_mos_diduse_space(dsl_pool_t *dp, | |
330 | int64_t used, int64_t comp, int64_t uncomp) | |
331 | { | |
332 | ASSERT3U(comp, ==, uncomp); /* it's all metadata */ | |
333 | mutex_enter(&dp->dp_lock); | |
334 | dp->dp_mos_used_delta += used; | |
335 | dp->dp_mos_compressed_delta += comp; | |
336 | dp->dp_mos_uncompressed_delta += uncomp; | |
337 | mutex_exit(&dp->dp_lock); | |
338 | } | |
339 | ||
428870ff BB |
340 | static int |
341 | deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
342 | { | |
343 | dsl_deadlist_t *dl = arg; | |
344 | dsl_deadlist_insert(dl, bp, tx); | |
345 | return (0); | |
346 | } | |
347 | ||
34dc7c2f BB |
348 | void |
349 | dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) | |
350 | { | |
351 | zio_t *zio; | |
352 | dmu_tx_t *tx; | |
353 | dsl_dir_t *dd; | |
354 | dsl_dataset_t *ds; | |
428870ff | 355 | objset_t *mos = dp->dp_meta_objset; |
b128c09f BB |
356 | hrtime_t start, write_time; |
357 | uint64_t data_written; | |
34dc7c2f | 358 | int err; |
29809a6c MA |
359 | list_t synced_datasets; |
360 | ||
361 | list_create(&synced_datasets, sizeof (dsl_dataset_t), | |
362 | offsetof(dsl_dataset_t, ds_synced_link)); | |
34dc7c2f | 363 | |
428870ff BB |
364 | /* |
365 | * We need to copy dp_space_towrite() before doing | |
13fe0198 | 366 | * dsl_sync_task_sync(), because |
428870ff BB |
367 | * dsl_dataset_snapshot_reserve_space() will increase |
368 | * dp_space_towrite but not actually write anything. | |
369 | */ | |
370 | data_written = dp->dp_space_towrite[txg & TXG_MASK]; | |
371 | ||
34dc7c2f BB |
372 | tx = dmu_tx_create_assigned(dp, txg); |
373 | ||
b128c09f | 374 | dp->dp_read_overhead = 0; |
9babb374 BB |
375 | start = gethrtime(); |
376 | ||
34dc7c2f | 377 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
c65aa5b2 | 378 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) { |
9babb374 BB |
379 | /* |
380 | * We must not sync any non-MOS datasets twice, because | |
381 | * we may have taken a snapshot of them. However, we | |
382 | * may sync newly-created datasets on pass 2. | |
383 | */ | |
384 | ASSERT(!list_link_active(&ds->ds_synced_link)); | |
29809a6c | 385 | list_insert_tail(&synced_datasets, ds); |
34dc7c2f BB |
386 | dsl_dataset_sync(ds, zio, tx); |
387 | } | |
b128c09f | 388 | DTRACE_PROBE(pool_sync__1setup); |
34dc7c2f | 389 | err = zio_wait(zio); |
9babb374 | 390 | |
b128c09f | 391 | write_time = gethrtime() - start; |
34dc7c2f | 392 | ASSERT(err == 0); |
b128c09f | 393 | DTRACE_PROBE(pool_sync__2rootzio); |
34dc7c2f | 394 | |
29809a6c MA |
395 | /* |
396 | * After the data blocks have been written (ensured by the zio_wait() | |
397 | * above), update the user/group space accounting. | |
398 | */ | |
399 | for (ds = list_head(&synced_datasets); ds; | |
400 | ds = list_next(&synced_datasets, ds)) | |
428870ff | 401 | dmu_objset_do_userquota_updates(ds->ds_objset, tx); |
9babb374 BB |
402 | |
403 | /* | |
404 | * Sync the datasets again to push out the changes due to | |
428870ff | 405 | * userspace updates. This must be done before we process the |
29809a6c MA |
406 | * sync tasks, so that any snapshots will have the correct |
407 | * user accounting information (and we won't get confused | |
408 | * about which blocks are part of the snapshot). | |
9babb374 BB |
409 | */ |
410 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
c65aa5b2 | 411 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) { |
9babb374 BB |
412 | ASSERT(list_link_active(&ds->ds_synced_link)); |
413 | dmu_buf_rele(ds->ds_dbuf, ds); | |
414 | dsl_dataset_sync(ds, zio, tx); | |
415 | } | |
416 | err = zio_wait(zio); | |
417 | ||
428870ff | 418 | /* |
29809a6c MA |
419 | * Now that the datasets have been completely synced, we can |
420 | * clean up our in-memory structures accumulated while syncing: | |
421 | * | |
422 | * - move dead blocks from the pending deadlist to the on-disk deadlist | |
423 | * - clean up zil records | |
424 | * - release hold from dsl_dataset_dirty() | |
428870ff | 425 | */ |
29809a6c MA |
426 | while ((ds = list_remove_head(&synced_datasets))) { |
427 | ASSERTV(objset_t *os = ds->ds_objset); | |
428870ff BB |
428 | bplist_iterate(&ds->ds_pending_deadlist, |
429 | deadlist_enqueue_cb, &ds->ds_deadlist, tx); | |
29809a6c MA |
430 | ASSERT(!dmu_objset_is_dirty(os, txg)); |
431 | dmu_buf_rele(ds->ds_dbuf, ds); | |
428870ff BB |
432 | } |
433 | ||
b128c09f | 434 | start = gethrtime(); |
c65aa5b2 | 435 | while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg))) |
34dc7c2f | 436 | dsl_dir_sync(dd, tx); |
b128c09f BB |
437 | write_time += gethrtime() - start; |
438 | ||
29809a6c MA |
439 | /* |
440 | * The MOS's space is accounted for in the pool/$MOS | |
441 | * (dp_mos_dir). We can't modify the mos while we're syncing | |
442 | * it, so we remember the deltas and apply them here. | |
443 | */ | |
444 | if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 || | |
445 | dp->dp_mos_uncompressed_delta != 0) { | |
446 | dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD, | |
447 | dp->dp_mos_used_delta, | |
448 | dp->dp_mos_compressed_delta, | |
449 | dp->dp_mos_uncompressed_delta, tx); | |
450 | dp->dp_mos_used_delta = 0; | |
451 | dp->dp_mos_compressed_delta = 0; | |
452 | dp->dp_mos_uncompressed_delta = 0; | |
453 | } | |
454 | ||
b128c09f | 455 | start = gethrtime(); |
428870ff BB |
456 | if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || |
457 | list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { | |
34dc7c2f | 458 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
428870ff | 459 | dmu_objset_sync(mos, zio, tx); |
34dc7c2f BB |
460 | err = zio_wait(zio); |
461 | ASSERT(err == 0); | |
462 | dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); | |
463 | spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); | |
464 | } | |
b128c09f BB |
465 | write_time += gethrtime() - start; |
466 | DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time, | |
467 | hrtime_t, dp->dp_read_overhead); | |
468 | write_time -= dp->dp_read_overhead; | |
34dc7c2f | 469 | |
29809a6c MA |
470 | /* |
471 | * If we modify a dataset in the same txg that we want to destroy it, | |
472 | * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it. | |
473 | * dsl_dir_destroy_check() will fail if there are unexpected holds. | |
474 | * Therefore, we want to sync the MOS (thus syncing the dd_dbuf | |
475 | * and clearing the hold on it) before we process the sync_tasks. | |
476 | * The MOS data dirtied by the sync_tasks will be synced on the next | |
477 | * pass. | |
478 | */ | |
479 | DTRACE_PROBE(pool_sync__3task); | |
480 | if (!txg_list_empty(&dp->dp_sync_tasks, txg)) { | |
13fe0198 | 481 | dsl_sync_task_t *dst; |
29809a6c MA |
482 | /* |
483 | * No more sync tasks should have been added while we | |
484 | * were syncing. | |
485 | */ | |
486 | ASSERT(spa_sync_pass(dp->dp_spa) == 1); | |
13fe0198 MA |
487 | while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg))) |
488 | dsl_sync_task_sync(dst, tx); | |
29809a6c MA |
489 | } |
490 | ||
34dc7c2f | 491 | dmu_tx_commit(tx); |
b128c09f | 492 | |
b128c09f BB |
493 | dp->dp_space_towrite[txg & TXG_MASK] = 0; |
494 | ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0); | |
495 | ||
496 | /* | |
497 | * If the write limit max has not been explicitly set, set it | |
498 | * to a fraction of available physical memory (default 1/8th). | |
499 | * Note that we must inflate the limit because the spa | |
500 | * inflates write sizes to account for data replication. | |
501 | * Check this each sync phase to catch changing memory size. | |
502 | */ | |
503 | if (physmem != old_physmem && zfs_write_limit_shift) { | |
504 | mutex_enter(&zfs_write_limit_lock); | |
505 | old_physmem = physmem; | |
506 | zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift; | |
507 | zfs_write_limit_inflated = MAX(zfs_write_limit_min, | |
508 | spa_get_asize(dp->dp_spa, zfs_write_limit_max)); | |
509 | mutex_exit(&zfs_write_limit_lock); | |
510 | } | |
511 | ||
512 | /* | |
513 | * Attempt to keep the sync time consistent by adjusting the | |
514 | * amount of write traffic allowed into each transaction group. | |
515 | * Weight the throughput calculation towards the current value: | |
516 | * thru = 3/4 old_thru + 1/4 new_thru | |
428870ff | 517 | * |
63fd3c6c AL |
518 | * Note: write_time is in nanosecs while dp_throughput is expressed in |
519 | * bytes per millisecond. | |
b128c09f BB |
520 | */ |
521 | ASSERT(zfs_write_limit_min > 0); | |
63fd3c6c AL |
522 | if (data_written > zfs_write_limit_min / 8 && |
523 | write_time > MSEC2NSEC(1)) { | |
524 | uint64_t throughput = data_written / NSEC2MSEC(write_time); | |
428870ff | 525 | |
b128c09f BB |
526 | if (dp->dp_throughput) |
527 | dp->dp_throughput = throughput / 4 + | |
528 | 3 * dp->dp_throughput / 4; | |
529 | else | |
530 | dp->dp_throughput = throughput; | |
531 | dp->dp_write_limit = MIN(zfs_write_limit_inflated, | |
532 | MAX(zfs_write_limit_min, | |
428870ff | 533 | dp->dp_throughput * zfs_txg_synctime_ms)); |
b128c09f | 534 | } |
34dc7c2f BB |
535 | } |
536 | ||
537 | void | |
428870ff | 538 | dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) |
34dc7c2f | 539 | { |
29809a6c | 540 | zilog_t *zilog; |
34dc7c2f BB |
541 | dsl_dataset_t *ds; |
542 | ||
29809a6c MA |
543 | while ((zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg))) { |
544 | ds = dmu_objset_ds(zilog->zl_os); | |
545 | zil_clean(zilog, txg); | |
546 | ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg)); | |
547 | dmu_buf_rele(ds->ds_dbuf, zilog); | |
34dc7c2f | 548 | } |
428870ff | 549 | ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); |
34dc7c2f BB |
550 | } |
551 | ||
552 | /* | |
553 | * TRUE if the current thread is the tx_sync_thread or if we | |
554 | * are being called from SPA context during pool initialization. | |
555 | */ | |
556 | int | |
557 | dsl_pool_sync_context(dsl_pool_t *dp) | |
558 | { | |
559 | return (curthread == dp->dp_tx.tx_sync_thread || | |
9ae529ec | 560 | spa_is_initializing(dp->dp_spa)); |
34dc7c2f BB |
561 | } |
562 | ||
563 | uint64_t | |
564 | dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) | |
565 | { | |
566 | uint64_t space, resv; | |
567 | ||
568 | /* | |
569 | * Reserve about 1.6% (1/64), or at least 32MB, for allocation | |
570 | * efficiency. | |
571 | * XXX The intent log is not accounted for, so it must fit | |
572 | * within this slop. | |
573 | * | |
574 | * If we're trying to assess whether it's OK to do a free, | |
575 | * cut the reservation in half to allow forward progress | |
576 | * (e.g. make it possible to rm(1) files from a full pool). | |
577 | */ | |
578 | space = spa_get_dspace(dp->dp_spa); | |
579 | resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1); | |
580 | if (netfree) | |
581 | resv >>= 1; | |
582 | ||
583 | return (space - resv); | |
584 | } | |
585 | ||
586 | int | |
587 | dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx) | |
588 | { | |
589 | uint64_t reserved = 0; | |
590 | uint64_t write_limit = (zfs_write_limit_override ? | |
591 | zfs_write_limit_override : dp->dp_write_limit); | |
592 | ||
593 | if (zfs_no_write_throttle) { | |
b128c09f BB |
594 | atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], |
595 | space); | |
34dc7c2f BB |
596 | return (0); |
597 | } | |
598 | ||
599 | /* | |
600 | * Check to see if we have exceeded the maximum allowed IO for | |
601 | * this transaction group. We can do this without locks since | |
602 | * a little slop here is ok. Note that we do the reserved check | |
603 | * with only half the requested reserve: this is because the | |
604 | * reserve requests are worst-case, and we really don't want to | |
605 | * throttle based off of worst-case estimates. | |
606 | */ | |
607 | if (write_limit > 0) { | |
608 | reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK] | |
609 | + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2; | |
610 | ||
570827e1 BB |
611 | if (reserved && reserved > write_limit) { |
612 | DMU_TX_STAT_BUMP(dmu_tx_write_limit); | |
2e528b49 | 613 | return (SET_ERROR(ERESTART)); |
570827e1 | 614 | } |
34dc7c2f BB |
615 | } |
616 | ||
617 | atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space); | |
618 | ||
619 | /* | |
620 | * If this transaction group is over 7/8ths capacity, delay | |
621 | * the caller 1 clock tick. This will slow down the "fill" | |
622 | * rate until the sync process can catch up with us. | |
623 | */ | |
63fd3c6c AL |
624 | if (reserved && reserved > (write_limit - (write_limit >> 3))) { |
625 | txg_delay(dp, tx->tx_txg, zfs_throttle_delay, | |
626 | zfs_throttle_resolution); | |
627 | } | |
34dc7c2f BB |
628 | |
629 | return (0); | |
630 | } | |
631 | ||
632 | void | |
633 | dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) | |
634 | { | |
635 | ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space); | |
636 | atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space); | |
637 | } | |
638 | ||
639 | void | |
640 | dsl_pool_memory_pressure(dsl_pool_t *dp) | |
641 | { | |
34dc7c2f BB |
642 | uint64_t space_inuse = 0; |
643 | int i; | |
644 | ||
645 | if (dp->dp_write_limit == zfs_write_limit_min) | |
646 | return; | |
647 | ||
648 | for (i = 0; i < TXG_SIZE; i++) { | |
649 | space_inuse += dp->dp_space_towrite[i]; | |
650 | space_inuse += dp->dp_tempreserved[i]; | |
651 | } | |
652 | dp->dp_write_limit = MAX(zfs_write_limit_min, | |
653 | MIN(dp->dp_write_limit, space_inuse / 4)); | |
654 | } | |
655 | ||
656 | void | |
657 | dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) | |
658 | { | |
659 | if (space > 0) { | |
660 | mutex_enter(&dp->dp_lock); | |
661 | dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space; | |
662 | mutex_exit(&dp->dp_lock); | |
663 | } | |
664 | } | |
b128c09f BB |
665 | |
666 | /* ARGSUSED */ | |
667 | static int | |
13fe0198 | 668 | upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) |
b128c09f BB |
669 | { |
670 | dmu_tx_t *tx = arg; | |
671 | dsl_dataset_t *ds, *prev = NULL; | |
672 | int err; | |
b128c09f | 673 | |
13fe0198 | 674 | err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); |
b128c09f BB |
675 | if (err) |
676 | return (err); | |
677 | ||
678 | while (ds->ds_phys->ds_prev_snap_obj != 0) { | |
679 | err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, | |
680 | FTAG, &prev); | |
681 | if (err) { | |
682 | dsl_dataset_rele(ds, FTAG); | |
683 | return (err); | |
684 | } | |
685 | ||
686 | if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) | |
687 | break; | |
688 | dsl_dataset_rele(ds, FTAG); | |
689 | ds = prev; | |
690 | prev = NULL; | |
691 | } | |
692 | ||
693 | if (prev == NULL) { | |
694 | prev = dp->dp_origin_snap; | |
695 | ||
696 | /* | |
697 | * The $ORIGIN can't have any data, or the accounting | |
698 | * will be wrong. | |
699 | */ | |
13fe0198 | 700 | ASSERT0(prev->ds_phys->ds_bp.blk_birth); |
b128c09f BB |
701 | |
702 | /* The origin doesn't get attached to itself */ | |
703 | if (ds->ds_object == prev->ds_object) { | |
704 | dsl_dataset_rele(ds, FTAG); | |
705 | return (0); | |
706 | } | |
707 | ||
708 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
709 | ds->ds_phys->ds_prev_snap_obj = prev->ds_object; | |
710 | ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; | |
711 | ||
712 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); | |
713 | ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; | |
714 | ||
715 | dmu_buf_will_dirty(prev->ds_dbuf, tx); | |
716 | prev->ds_phys->ds_num_children++; | |
717 | ||
718 | if (ds->ds_phys->ds_next_snap_obj == 0) { | |
719 | ASSERT(ds->ds_prev == NULL); | |
13fe0198 | 720 | VERIFY0(dsl_dataset_hold_obj(dp, |
b128c09f BB |
721 | ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); |
722 | } | |
723 | } | |
724 | ||
13fe0198 MA |
725 | ASSERT3U(ds->ds_dir->dd_phys->dd_origin_obj, ==, prev->ds_object); |
726 | ASSERT3U(ds->ds_phys->ds_prev_snap_obj, ==, prev->ds_object); | |
b128c09f BB |
727 | |
728 | if (prev->ds_phys->ds_next_clones_obj == 0) { | |
428870ff | 729 | dmu_buf_will_dirty(prev->ds_dbuf, tx); |
b128c09f BB |
730 | prev->ds_phys->ds_next_clones_obj = |
731 | zap_create(dp->dp_meta_objset, | |
732 | DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); | |
733 | } | |
13fe0198 | 734 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
b128c09f BB |
735 | prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); |
736 | ||
737 | dsl_dataset_rele(ds, FTAG); | |
738 | if (prev != dp->dp_origin_snap) | |
739 | dsl_dataset_rele(prev, FTAG); | |
740 | return (0); | |
741 | } | |
742 | ||
743 | void | |
744 | dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
745 | { | |
746 | ASSERT(dmu_tx_is_syncing(tx)); | |
747 | ASSERT(dp->dp_origin_snap != NULL); | |
748 | ||
13fe0198 | 749 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb, |
428870ff BB |
750 | tx, DS_FIND_CHILDREN)); |
751 | } | |
752 | ||
753 | /* ARGSUSED */ | |
754 | static int | |
13fe0198 | 755 | upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) |
428870ff BB |
756 | { |
757 | dmu_tx_t *tx = arg; | |
428870ff BB |
758 | objset_t *mos = dp->dp_meta_objset; |
759 | ||
13fe0198 | 760 | if (ds->ds_dir->dd_phys->dd_origin_obj != 0) { |
428870ff BB |
761 | dsl_dataset_t *origin; |
762 | ||
13fe0198 | 763 | VERIFY0(dsl_dataset_hold_obj(dp, |
428870ff BB |
764 | ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin)); |
765 | ||
766 | if (origin->ds_dir->dd_phys->dd_clones == 0) { | |
767 | dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); | |
768 | origin->ds_dir->dd_phys->dd_clones = zap_create(mos, | |
769 | DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); | |
770 | } | |
771 | ||
13fe0198 MA |
772 | VERIFY0(zap_add_int(dp->dp_meta_objset, |
773 | origin->ds_dir->dd_phys->dd_clones, ds->ds_object, tx)); | |
428870ff BB |
774 | |
775 | dsl_dataset_rele(origin, FTAG); | |
776 | } | |
428870ff BB |
777 | return (0); |
778 | } | |
779 | ||
780 | void | |
781 | dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
782 | { | |
428870ff BB |
783 | uint64_t obj; |
784 | ||
d6320ddb BB |
785 | ASSERT(dmu_tx_is_syncing(tx)); |
786 | ||
428870ff | 787 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); |
13fe0198 | 788 | VERIFY0(dsl_pool_open_special_dir(dp, |
428870ff BB |
789 | FREE_DIR_NAME, &dp->dp_free_dir)); |
790 | ||
791 | /* | |
792 | * We can't use bpobj_alloc(), because spa_version() still | |
793 | * returns the old version, and we need a new-version bpobj with | |
794 | * subobj support. So call dmu_object_alloc() directly. | |
795 | */ | |
796 | obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, | |
797 | SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); | |
13fe0198 | 798 | VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 799 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); |
13fe0198 | 800 | VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj)); |
428870ff | 801 | |
13fe0198 | 802 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
428870ff | 803 | upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); |
b128c09f BB |
804 | } |
805 | ||
806 | void | |
807 | dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) | |
808 | { | |
809 | uint64_t dsobj; | |
810 | dsl_dataset_t *ds; | |
811 | ||
812 | ASSERT(dmu_tx_is_syncing(tx)); | |
813 | ASSERT(dp->dp_origin_snap == NULL); | |
13fe0198 | 814 | ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER)); |
b128c09f BB |
815 | |
816 | /* create the origin dir, ds, & snap-ds */ | |
b128c09f BB |
817 | dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, |
818 | NULL, 0, kcred, tx); | |
13fe0198 MA |
819 | VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); |
820 | dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx); | |
821 | VERIFY0(dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, | |
b128c09f BB |
822 | dp, &dp->dp_origin_snap)); |
823 | dsl_dataset_rele(ds, FTAG); | |
b128c09f | 824 | } |
9babb374 BB |
825 | |
826 | taskq_t * | |
3558fd73 | 827 | dsl_pool_iput_taskq(dsl_pool_t *dp) |
9babb374 | 828 | { |
3558fd73 | 829 | return (dp->dp_iput_taskq); |
9babb374 | 830 | } |
428870ff BB |
831 | |
832 | /* | |
833 | * Walk through the pool-wide zap object of temporary snapshot user holds | |
834 | * and release them. | |
835 | */ | |
836 | void | |
837 | dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) | |
838 | { | |
839 | zap_attribute_t za; | |
840 | zap_cursor_t zc; | |
841 | objset_t *mos = dp->dp_meta_objset; | |
842 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
843 | ||
844 | if (zapobj == 0) | |
845 | return; | |
846 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
847 | ||
848 | for (zap_cursor_init(&zc, mos, zapobj); | |
849 | zap_cursor_retrieve(&zc, &za) == 0; | |
850 | zap_cursor_advance(&zc)) { | |
851 | char *htag; | |
852 | uint64_t dsobj; | |
853 | ||
854 | htag = strchr(za.za_name, '-'); | |
855 | *htag = '\0'; | |
856 | ++htag; | |
857 | dsobj = strtonum(za.za_name, NULL); | |
13fe0198 | 858 | dsl_dataset_user_release_tmp(dp, dsobj, htag); |
428870ff BB |
859 | } |
860 | zap_cursor_fini(&zc); | |
861 | } | |
862 | ||
863 | /* | |
864 | * Create the pool-wide zap object for storing temporary snapshot holds. | |
865 | */ | |
866 | void | |
867 | dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) | |
868 | { | |
869 | objset_t *mos = dp->dp_meta_objset; | |
870 | ||
871 | ASSERT(dp->dp_tmp_userrefs_obj == 0); | |
872 | ASSERT(dmu_tx_is_syncing(tx)); | |
873 | ||
9ae529ec CS |
874 | dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS, |
875 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx); | |
428870ff BB |
876 | } |
877 | ||
878 | static int | |
879 | dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, | |
13fe0198 | 880 | const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding) |
428870ff BB |
881 | { |
882 | objset_t *mos = dp->dp_meta_objset; | |
883 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
884 | char *name; | |
885 | int error; | |
886 | ||
887 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
888 | ASSERT(dmu_tx_is_syncing(tx)); | |
889 | ||
890 | /* | |
891 | * If the pool was created prior to SPA_VERSION_USERREFS, the | |
892 | * zap object for temporary holds might not exist yet. | |
893 | */ | |
894 | if (zapobj == 0) { | |
895 | if (holding) { | |
896 | dsl_pool_user_hold_create_obj(dp, tx); | |
897 | zapobj = dp->dp_tmp_userrefs_obj; | |
898 | } else { | |
2e528b49 | 899 | return (SET_ERROR(ENOENT)); |
428870ff BB |
900 | } |
901 | } | |
902 | ||
903 | name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); | |
904 | if (holding) | |
13fe0198 | 905 | error = zap_add(mos, zapobj, name, 8, 1, &now, tx); |
428870ff BB |
906 | else |
907 | error = zap_remove(mos, zapobj, name, tx); | |
908 | strfree(name); | |
909 | ||
910 | return (error); | |
911 | } | |
912 | ||
913 | /* | |
914 | * Add a temporary hold for the given dataset object and tag. | |
915 | */ | |
916 | int | |
917 | dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
13fe0198 | 918 | uint64_t now, dmu_tx_t *tx) |
428870ff BB |
919 | { |
920 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); | |
921 | } | |
922 | ||
923 | /* | |
924 | * Release a temporary hold for the given dataset object and tag. | |
925 | */ | |
926 | int | |
927 | dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
928 | dmu_tx_t *tx) | |
929 | { | |
13fe0198 | 930 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, 0, |
428870ff BB |
931 | tx, B_FALSE)); |
932 | } | |
c409e464 | 933 | |
13fe0198 MA |
934 | /* |
935 | * DSL Pool Configuration Lock | |
936 | * | |
937 | * The dp_config_rwlock protects against changes to DSL state (e.g. dataset | |
938 | * creation / destruction / rename / property setting). It must be held for | |
939 | * read to hold a dataset or dsl_dir. I.e. you must call | |
940 | * dsl_pool_config_enter() or dsl_pool_hold() before calling | |
941 | * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock | |
942 | * must be held continuously until all datasets and dsl_dirs are released. | |
943 | * | |
944 | * The only exception to this rule is that if a "long hold" is placed on | |
945 | * a dataset, then the dp_config_rwlock may be dropped while the dataset | |
946 | * is still held. The long hold will prevent the dataset from being | |
947 | * destroyed -- the destroy will fail with EBUSY. A long hold can be | |
948 | * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset | |
949 | * (by calling dsl_{dataset,objset}_{try}own{_obj}). | |
950 | * | |
951 | * Legitimate long-holders (including owners) should be long-running, cancelable | |
952 | * tasks that should cause "zfs destroy" to fail. This includes DMU | |
953 | * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open), | |
954 | * "zfs send", and "zfs diff". There are several other long-holders whose | |
955 | * uses are suboptimal (e.g. "zfs promote", and zil_suspend()). | |
956 | * | |
957 | * The usual formula for long-holding would be: | |
958 | * dsl_pool_hold() | |
959 | * dsl_dataset_hold() | |
960 | * ... perform checks ... | |
961 | * dsl_dataset_long_hold() | |
962 | * dsl_pool_rele() | |
963 | * ... perform long-running task ... | |
964 | * dsl_dataset_long_rele() | |
965 | * dsl_dataset_rele() | |
966 | * | |
967 | * Note that when the long hold is released, the dataset is still held but | |
968 | * the pool is not held. The dataset may change arbitrarily during this time | |
969 | * (e.g. it could be destroyed). Therefore you shouldn't do anything to the | |
970 | * dataset except release it. | |
971 | * | |
972 | * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only | |
973 | * or modifying operations. | |
974 | * | |
975 | * Modifying operations should generally use dsl_sync_task(). The synctask | |
976 | * infrastructure enforces proper locking strategy with respect to the | |
977 | * dp_config_rwlock. See the comment above dsl_sync_task() for details. | |
978 | * | |
979 | * Read-only operations will manually hold the pool, then the dataset, obtain | |
980 | * information from the dataset, then release the pool and dataset. | |
981 | * dmu_objset_{hold,rele}() are convenience routines that also do the pool | |
982 | * hold/rele. | |
983 | */ | |
984 | ||
985 | int | |
986 | dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp) | |
987 | { | |
988 | spa_t *spa; | |
989 | int error; | |
990 | ||
991 | error = spa_open(name, &spa, tag); | |
992 | if (error == 0) { | |
993 | *dp = spa_get_dsl(spa); | |
994 | dsl_pool_config_enter(*dp, tag); | |
995 | } | |
996 | return (error); | |
997 | } | |
998 | ||
999 | void | |
1000 | dsl_pool_rele(dsl_pool_t *dp, void *tag) | |
1001 | { | |
1002 | dsl_pool_config_exit(dp, tag); | |
1003 | spa_close(dp->dp_spa, tag); | |
1004 | } | |
1005 | ||
1006 | void | |
1007 | dsl_pool_config_enter(dsl_pool_t *dp, void *tag) | |
1008 | { | |
1009 | /* | |
1010 | * We use a "reentrant" reader-writer lock, but not reentrantly. | |
1011 | * | |
1012 | * The rrwlock can (with the track_all flag) track all reading threads, | |
1013 | * which is very useful for debugging which code path failed to release | |
1014 | * the lock, and for verifying that the *current* thread does hold | |
1015 | * the lock. | |
1016 | * | |
1017 | * (Unlike a rwlock, which knows that N threads hold it for | |
1018 | * read, but not *which* threads, so rw_held(RW_READER) returns TRUE | |
1019 | * if any thread holds it for read, even if this thread doesn't). | |
1020 | */ | |
1021 | ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER)); | |
1022 | rrw_enter(&dp->dp_config_rwlock, RW_READER, tag); | |
1023 | } | |
1024 | ||
1025 | void | |
1026 | dsl_pool_config_exit(dsl_pool_t *dp, void *tag) | |
1027 | { | |
1028 | rrw_exit(&dp->dp_config_rwlock, tag); | |
1029 | } | |
1030 | ||
1031 | boolean_t | |
1032 | dsl_pool_config_held(dsl_pool_t *dp) | |
1033 | { | |
1034 | return (RRW_LOCK_HELD(&dp->dp_config_rwlock)); | |
1035 | } | |
1036 | ||
c409e464 | 1037 | #if defined(_KERNEL) && defined(HAVE_SPL) |
40a806df NB |
1038 | EXPORT_SYMBOL(dsl_pool_config_enter); |
1039 | EXPORT_SYMBOL(dsl_pool_config_exit); | |
1040 | ||
c409e464 BB |
1041 | module_param(zfs_no_write_throttle, int, 0644); |
1042 | MODULE_PARM_DESC(zfs_no_write_throttle, "Disable write throttling"); | |
1043 | ||
1044 | module_param(zfs_write_limit_shift, int, 0444); | |
1045 | MODULE_PARM_DESC(zfs_write_limit_shift, "log2(fraction of memory) per txg"); | |
1046 | ||
1047 | module_param(zfs_txg_synctime_ms, int, 0644); | |
5b7e5b5a | 1048 | MODULE_PARM_DESC(zfs_txg_synctime_ms, "Target milliseconds between txg sync"); |
c409e464 BB |
1049 | |
1050 | module_param(zfs_write_limit_min, ulong, 0444); | |
5b7e5b5a | 1051 | MODULE_PARM_DESC(zfs_write_limit_min, "Min txg write limit"); |
c409e464 BB |
1052 | |
1053 | module_param(zfs_write_limit_max, ulong, 0444); | |
5b7e5b5a | 1054 | MODULE_PARM_DESC(zfs_write_limit_max, "Max txg write limit"); |
c409e464 BB |
1055 | |
1056 | module_param(zfs_write_limit_inflated, ulong, 0444); | |
5b7e5b5a | 1057 | MODULE_PARM_DESC(zfs_write_limit_inflated, "Inflated txg write limit"); |
c409e464 BB |
1058 | |
1059 | module_param(zfs_write_limit_override, ulong, 0444); | |
5b7e5b5a | 1060 | MODULE_PARM_DESC(zfs_write_limit_override, "Override txg write limit"); |
c409e464 | 1061 | #endif |