]>
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. |
34dc7c2f BB |
23 | */ |
24 | ||
34dc7c2f BB |
25 | #include <sys/dsl_pool.h> |
26 | #include <sys/dsl_dataset.h> | |
428870ff | 27 | #include <sys/dsl_prop.h> |
34dc7c2f BB |
28 | #include <sys/dsl_dir.h> |
29 | #include <sys/dsl_synctask.h> | |
428870ff BB |
30 | #include <sys/dsl_scan.h> |
31 | #include <sys/dnode.h> | |
34dc7c2f BB |
32 | #include <sys/dmu_tx.h> |
33 | #include <sys/dmu_objset.h> | |
34 | #include <sys/arc.h> | |
35 | #include <sys/zap.h> | |
36 | #include <sys/zio.h> | |
37 | #include <sys/zfs_context.h> | |
38 | #include <sys/fs/zfs.h> | |
b128c09f BB |
39 | #include <sys/zfs_znode.h> |
40 | #include <sys/spa_impl.h> | |
428870ff | 41 | #include <sys/dsl_deadlist.h> |
34dc7c2f BB |
42 | |
43 | int zfs_no_write_throttle = 0; | |
b128c09f | 44 | int zfs_write_limit_shift = 3; /* 1/8th of physical memory */ |
572e2857 | 45 | int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */ |
b128c09f BB |
46 | |
47 | uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */ | |
48 | uint64_t zfs_write_limit_max = 0; /* max data payload per txg */ | |
49 | uint64_t zfs_write_limit_inflated = 0; | |
34dc7c2f BB |
50 | uint64_t zfs_write_limit_override = 0; |
51 | ||
b128c09f BB |
52 | kmutex_t zfs_write_limit_lock; |
53 | ||
54 | static pgcnt_t old_physmem = 0; | |
55 | ||
428870ff | 56 | int |
b128c09f | 57 | dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp) |
34dc7c2f BB |
58 | { |
59 | uint64_t obj; | |
60 | int err; | |
61 | ||
62 | err = zap_lookup(dp->dp_meta_objset, | |
63 | dp->dp_root_dir->dd_phys->dd_child_dir_zapobj, | |
b128c09f | 64 | name, sizeof (obj), 1, &obj); |
34dc7c2f BB |
65 | if (err) |
66 | return (err); | |
67 | ||
b128c09f | 68 | return (dsl_dir_open_obj(dp, obj, name, dp, ddp)); |
34dc7c2f BB |
69 | } |
70 | ||
71 | static dsl_pool_t * | |
72 | dsl_pool_open_impl(spa_t *spa, uint64_t txg) | |
73 | { | |
74 | dsl_pool_t *dp; | |
75 | blkptr_t *bp = spa_get_rootblkptr(spa); | |
34dc7c2f BB |
76 | |
77 | dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP); | |
78 | dp->dp_spa = spa; | |
79 | dp->dp_meta_rootbp = *bp; | |
80 | rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL); | |
81 | dp->dp_write_limit = zfs_write_limit_min; | |
82 | txg_init(dp, txg); | |
83 | ||
84 | txg_list_create(&dp->dp_dirty_datasets, | |
85 | offsetof(dsl_dataset_t, ds_dirty_link)); | |
86 | txg_list_create(&dp->dp_dirty_dirs, | |
87 | offsetof(dsl_dir_t, dd_dirty_link)); | |
88 | txg_list_create(&dp->dp_sync_tasks, | |
89 | offsetof(dsl_sync_task_group_t, dstg_node)); | |
90 | list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t), | |
91 | offsetof(dsl_dataset_t, ds_synced_link)); | |
92 | ||
93 | mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL); | |
94 | ||
3558fd73 | 95 | dp->dp_iput_taskq = taskq_create("zfs_iput_taskq", 1, minclsyspri, |
9babb374 BB |
96 | 1, 4, 0); |
97 | ||
34dc7c2f BB |
98 | return (dp); |
99 | } | |
100 | ||
101 | int | |
102 | dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp) | |
103 | { | |
104 | int err; | |
105 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
b128c09f BB |
106 | dsl_dir_t *dd; |
107 | dsl_dataset_t *ds; | |
428870ff | 108 | uint64_t obj; |
34dc7c2f | 109 | |
b128c09f | 110 | rw_enter(&dp->dp_config_rwlock, RW_WRITER); |
428870ff BB |
111 | err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, |
112 | &dp->dp_meta_objset); | |
34dc7c2f BB |
113 | if (err) |
114 | goto out; | |
34dc7c2f BB |
115 | |
116 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
117 | DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, | |
118 | &dp->dp_root_dir_obj); | |
119 | if (err) | |
120 | goto out; | |
121 | ||
122 | err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, | |
123 | NULL, dp, &dp->dp_root_dir); | |
124 | if (err) | |
125 | goto out; | |
126 | ||
b128c09f | 127 | err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir); |
34dc7c2f BB |
128 | if (err) |
129 | goto out; | |
130 | ||
b128c09f BB |
131 | if (spa_version(spa) >= SPA_VERSION_ORIGIN) { |
132 | err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd); | |
133 | if (err) | |
134 | goto out; | |
135 | err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj, | |
136 | FTAG, &ds); | |
9babb374 BB |
137 | if (err == 0) { |
138 | err = dsl_dataset_hold_obj(dp, | |
139 | ds->ds_phys->ds_prev_snap_obj, dp, | |
140 | &dp->dp_origin_snap); | |
141 | dsl_dataset_rele(ds, FTAG); | |
142 | } | |
143 | dsl_dir_close(dd, dp); | |
b128c09f BB |
144 | if (err) |
145 | goto out; | |
b128c09f BB |
146 | } |
147 | ||
428870ff BB |
148 | if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { |
149 | err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME, | |
150 | &dp->dp_free_dir); | |
b128c09f BB |
151 | if (err) |
152 | goto out; | |
428870ff | 153 | |
b128c09f | 154 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
428870ff | 155 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj); |
b128c09f BB |
156 | if (err) |
157 | goto out; | |
428870ff BB |
158 | VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, |
159 | dp->dp_meta_objset, obj)); | |
b128c09f BB |
160 | } |
161 | ||
428870ff BB |
162 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
163 | DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1, | |
164 | &dp->dp_tmp_userrefs_obj); | |
165 | if (err == ENOENT) | |
166 | err = 0; | |
167 | if (err) | |
168 | goto out; | |
169 | ||
170 | err = dsl_scan_init(dp, txg); | |
171 | ||
34dc7c2f BB |
172 | out: |
173 | rw_exit(&dp->dp_config_rwlock); | |
174 | if (err) | |
175 | dsl_pool_close(dp); | |
176 | else | |
177 | *dpp = dp; | |
178 | ||
179 | return (err); | |
180 | } | |
181 | ||
182 | void | |
183 | dsl_pool_close(dsl_pool_t *dp) | |
184 | { | |
b128c09f BB |
185 | /* drop our references from dsl_pool_open() */ |
186 | ||
187 | /* | |
188 | * Since we held the origin_snap from "syncing" context (which | |
189 | * includes pool-opening context), it actually only got a "ref" | |
190 | * and not a hold, so just drop that here. | |
191 | */ | |
192 | if (dp->dp_origin_snap) | |
193 | dsl_dataset_drop_ref(dp->dp_origin_snap, dp); | |
34dc7c2f BB |
194 | if (dp->dp_mos_dir) |
195 | dsl_dir_close(dp->dp_mos_dir, dp); | |
428870ff BB |
196 | if (dp->dp_free_dir) |
197 | dsl_dir_close(dp->dp_free_dir, dp); | |
34dc7c2f BB |
198 | if (dp->dp_root_dir) |
199 | dsl_dir_close(dp->dp_root_dir, dp); | |
200 | ||
428870ff BB |
201 | bpobj_close(&dp->dp_free_bpobj); |
202 | ||
34dc7c2f BB |
203 | /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */ |
204 | if (dp->dp_meta_objset) | |
428870ff | 205 | dmu_objset_evict(dp->dp_meta_objset); |
34dc7c2f BB |
206 | |
207 | txg_list_destroy(&dp->dp_dirty_datasets); | |
428870ff | 208 | txg_list_destroy(&dp->dp_sync_tasks); |
34dc7c2f BB |
209 | txg_list_destroy(&dp->dp_dirty_dirs); |
210 | list_destroy(&dp->dp_synced_datasets); | |
211 | ||
212 | arc_flush(dp->dp_spa); | |
213 | txg_fini(dp); | |
428870ff | 214 | dsl_scan_fini(dp); |
34dc7c2f BB |
215 | rw_destroy(&dp->dp_config_rwlock); |
216 | mutex_destroy(&dp->dp_lock); | |
3558fd73 | 217 | taskq_destroy(dp->dp_iput_taskq); |
b128c09f BB |
218 | if (dp->dp_blkstats) |
219 | kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); | |
34dc7c2f BB |
220 | kmem_free(dp, sizeof (dsl_pool_t)); |
221 | } | |
222 | ||
223 | dsl_pool_t * | |
b128c09f | 224 | dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg) |
34dc7c2f BB |
225 | { |
226 | int err; | |
227 | dsl_pool_t *dp = dsl_pool_open_impl(spa, txg); | |
228 | dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg); | |
428870ff | 229 | objset_t *os; |
b128c09f | 230 | dsl_dataset_t *ds; |
428870ff | 231 | uint64_t obj; |
b128c09f BB |
232 | |
233 | /* create and open the MOS (meta-objset) */ | |
428870ff BB |
234 | dp->dp_meta_objset = dmu_objset_create_impl(spa, |
235 | NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx); | |
34dc7c2f BB |
236 | |
237 | /* create the pool directory */ | |
238 | err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
239 | DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx); | |
240 | ASSERT3U(err, ==, 0); | |
241 | ||
428870ff BB |
242 | /* Initialize scan structures */ |
243 | VERIFY3U(0, ==, dsl_scan_init(dp, txg)); | |
244 | ||
34dc7c2f | 245 | /* create and open the root dir */ |
b128c09f | 246 | dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx); |
34dc7c2f BB |
247 | VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj, |
248 | NULL, dp, &dp->dp_root_dir)); | |
249 | ||
250 | /* create and open the meta-objset dir */ | |
b128c09f BB |
251 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx); |
252 | VERIFY(0 == dsl_pool_open_special_dir(dp, | |
253 | MOS_DIR_NAME, &dp->dp_mos_dir)); | |
254 | ||
428870ff BB |
255 | if (spa_version(spa) >= SPA_VERSION_DEADLISTS) { |
256 | /* create and open the free dir */ | |
257 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, | |
258 | FREE_DIR_NAME, tx); | |
259 | VERIFY(0 == dsl_pool_open_special_dir(dp, | |
260 | FREE_DIR_NAME, &dp->dp_free_dir)); | |
261 | ||
262 | /* create and open the free_bplist */ | |
263 | obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx); | |
264 | VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
265 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0); | |
266 | VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, | |
267 | dp->dp_meta_objset, obj)); | |
268 | } | |
269 | ||
b128c09f BB |
270 | if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) |
271 | dsl_pool_create_origin(dp, tx); | |
272 | ||
273 | /* create the root dataset */ | |
428870ff | 274 | obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx); |
b128c09f BB |
275 | |
276 | /* create the root objset */ | |
428870ff | 277 | VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds)); |
0fe3d820 BB |
278 | VERIFY(NULL != (os = dmu_objset_create_impl(dp->dp_spa, ds, |
279 | dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx))); | |
b128c09f | 280 | #ifdef _KERNEL |
428870ff | 281 | zfs_create_fs(os, kcred, zplprops, tx); |
b128c09f BB |
282 | #endif |
283 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f BB |
284 | |
285 | dmu_tx_commit(tx); | |
286 | ||
287 | return (dp); | |
288 | } | |
289 | ||
428870ff BB |
290 | static int |
291 | deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
292 | { | |
293 | dsl_deadlist_t *dl = arg; | |
294 | dsl_deadlist_insert(dl, bp, tx); | |
295 | return (0); | |
296 | } | |
297 | ||
34dc7c2f BB |
298 | void |
299 | dsl_pool_sync(dsl_pool_t *dp, uint64_t txg) | |
300 | { | |
301 | zio_t *zio; | |
302 | dmu_tx_t *tx; | |
303 | dsl_dir_t *dd; | |
304 | dsl_dataset_t *ds; | |
305 | dsl_sync_task_group_t *dstg; | |
428870ff | 306 | objset_t *mos = dp->dp_meta_objset; |
b128c09f BB |
307 | hrtime_t start, write_time; |
308 | uint64_t data_written; | |
34dc7c2f BB |
309 | int err; |
310 | ||
428870ff BB |
311 | /* |
312 | * We need to copy dp_space_towrite() before doing | |
313 | * dsl_sync_task_group_sync(), because | |
314 | * dsl_dataset_snapshot_reserve_space() will increase | |
315 | * dp_space_towrite but not actually write anything. | |
316 | */ | |
317 | data_written = dp->dp_space_towrite[txg & TXG_MASK]; | |
318 | ||
34dc7c2f BB |
319 | tx = dmu_tx_create_assigned(dp, txg); |
320 | ||
b128c09f | 321 | dp->dp_read_overhead = 0; |
9babb374 BB |
322 | start = gethrtime(); |
323 | ||
34dc7c2f | 324 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
c65aa5b2 | 325 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) { |
9babb374 BB |
326 | /* |
327 | * We must not sync any non-MOS datasets twice, because | |
328 | * we may have taken a snapshot of them. However, we | |
329 | * may sync newly-created datasets on pass 2. | |
330 | */ | |
331 | ASSERT(!list_link_active(&ds->ds_synced_link)); | |
332 | list_insert_tail(&dp->dp_synced_datasets, ds); | |
34dc7c2f BB |
333 | dsl_dataset_sync(ds, zio, tx); |
334 | } | |
b128c09f | 335 | DTRACE_PROBE(pool_sync__1setup); |
34dc7c2f | 336 | err = zio_wait(zio); |
9babb374 | 337 | |
b128c09f | 338 | write_time = gethrtime() - start; |
34dc7c2f | 339 | ASSERT(err == 0); |
b128c09f | 340 | DTRACE_PROBE(pool_sync__2rootzio); |
34dc7c2f | 341 | |
9babb374 BB |
342 | for (ds = list_head(&dp->dp_synced_datasets); ds; |
343 | ds = list_next(&dp->dp_synced_datasets, ds)) | |
428870ff | 344 | dmu_objset_do_userquota_updates(ds->ds_objset, tx); |
9babb374 BB |
345 | |
346 | /* | |
347 | * Sync the datasets again to push out the changes due to | |
428870ff | 348 | * userspace updates. This must be done before we process the |
9babb374 BB |
349 | * sync tasks, because that could cause a snapshot of a dataset |
350 | * whose ds_bp will be rewritten when we do this 2nd sync. | |
351 | */ | |
352 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
c65aa5b2 | 353 | while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg))) { |
9babb374 BB |
354 | ASSERT(list_link_active(&ds->ds_synced_link)); |
355 | dmu_buf_rele(ds->ds_dbuf, ds); | |
356 | dsl_dataset_sync(ds, zio, tx); | |
357 | } | |
358 | err = zio_wait(zio); | |
359 | ||
428870ff BB |
360 | /* |
361 | * Move dead blocks from the pending deadlist to the on-disk | |
362 | * deadlist. | |
363 | */ | |
364 | for (ds = list_head(&dp->dp_synced_datasets); ds; | |
365 | ds = list_next(&dp->dp_synced_datasets, ds)) { | |
366 | bplist_iterate(&ds->ds_pending_deadlist, | |
367 | deadlist_enqueue_cb, &ds->ds_deadlist, tx); | |
368 | } | |
369 | ||
c65aa5b2 | 370 | while ((dstg = txg_list_remove(&dp->dp_sync_tasks, txg))) { |
9babb374 BB |
371 | /* |
372 | * No more sync tasks should have been added while we | |
373 | * were syncing. | |
374 | */ | |
375 | ASSERT(spa_sync_pass(dp->dp_spa) == 1); | |
34dc7c2f | 376 | dsl_sync_task_group_sync(dstg, tx); |
9babb374 | 377 | } |
b128c09f BB |
378 | DTRACE_PROBE(pool_sync__3task); |
379 | ||
380 | start = gethrtime(); | |
c65aa5b2 | 381 | while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg))) |
34dc7c2f | 382 | dsl_dir_sync(dd, tx); |
b128c09f BB |
383 | write_time += gethrtime() - start; |
384 | ||
b128c09f | 385 | start = gethrtime(); |
428870ff BB |
386 | if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL || |
387 | list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) { | |
34dc7c2f | 388 | zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); |
428870ff | 389 | dmu_objset_sync(mos, zio, tx); |
34dc7c2f BB |
390 | err = zio_wait(zio); |
391 | ASSERT(err == 0); | |
392 | dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", ""); | |
393 | spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); | |
394 | } | |
b128c09f BB |
395 | write_time += gethrtime() - start; |
396 | DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time, | |
397 | hrtime_t, dp->dp_read_overhead); | |
398 | write_time -= dp->dp_read_overhead; | |
34dc7c2f BB |
399 | |
400 | dmu_tx_commit(tx); | |
b128c09f | 401 | |
b128c09f BB |
402 | dp->dp_space_towrite[txg & TXG_MASK] = 0; |
403 | ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0); | |
404 | ||
405 | /* | |
406 | * If the write limit max has not been explicitly set, set it | |
407 | * to a fraction of available physical memory (default 1/8th). | |
408 | * Note that we must inflate the limit because the spa | |
409 | * inflates write sizes to account for data replication. | |
410 | * Check this each sync phase to catch changing memory size. | |
411 | */ | |
412 | if (physmem != old_physmem && zfs_write_limit_shift) { | |
413 | mutex_enter(&zfs_write_limit_lock); | |
414 | old_physmem = physmem; | |
415 | zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift; | |
416 | zfs_write_limit_inflated = MAX(zfs_write_limit_min, | |
417 | spa_get_asize(dp->dp_spa, zfs_write_limit_max)); | |
418 | mutex_exit(&zfs_write_limit_lock); | |
419 | } | |
420 | ||
421 | /* | |
422 | * Attempt to keep the sync time consistent by adjusting the | |
423 | * amount of write traffic allowed into each transaction group. | |
424 | * Weight the throughput calculation towards the current value: | |
425 | * thru = 3/4 old_thru + 1/4 new_thru | |
428870ff BB |
426 | * |
427 | * Note: write_time is in nanosecs, so write_time/MICROSEC | |
428 | * yields millisecs | |
b128c09f BB |
429 | */ |
430 | ASSERT(zfs_write_limit_min > 0); | |
428870ff BB |
431 | if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) { |
432 | uint64_t throughput = data_written / (write_time / MICROSEC); | |
433 | ||
b128c09f BB |
434 | if (dp->dp_throughput) |
435 | dp->dp_throughput = throughput / 4 + | |
436 | 3 * dp->dp_throughput / 4; | |
437 | else | |
438 | dp->dp_throughput = throughput; | |
439 | dp->dp_write_limit = MIN(zfs_write_limit_inflated, | |
440 | MAX(zfs_write_limit_min, | |
428870ff | 441 | dp->dp_throughput * zfs_txg_synctime_ms)); |
b128c09f | 442 | } |
34dc7c2f BB |
443 | } |
444 | ||
445 | void | |
428870ff | 446 | dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg) |
34dc7c2f BB |
447 | { |
448 | dsl_dataset_t *ds; | |
428870ff | 449 | objset_t *os; |
34dc7c2f | 450 | |
c65aa5b2 | 451 | while ((ds = list_head(&dp->dp_synced_datasets))) { |
34dc7c2f | 452 | list_remove(&dp->dp_synced_datasets, ds); |
428870ff | 453 | os = ds->ds_objset; |
572e2857 | 454 | zil_clean(os->os_zil, txg); |
428870ff | 455 | ASSERT(!dmu_objset_is_dirty(os, txg)); |
34dc7c2f BB |
456 | dmu_buf_rele(ds->ds_dbuf, ds); |
457 | } | |
428870ff | 458 | ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg)); |
34dc7c2f BB |
459 | } |
460 | ||
461 | /* | |
462 | * TRUE if the current thread is the tx_sync_thread or if we | |
463 | * are being called from SPA context during pool initialization. | |
464 | */ | |
465 | int | |
466 | dsl_pool_sync_context(dsl_pool_t *dp) | |
467 | { | |
468 | return (curthread == dp->dp_tx.tx_sync_thread || | |
469 | spa_get_dsl(dp->dp_spa) == NULL); | |
470 | } | |
471 | ||
472 | uint64_t | |
473 | dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree) | |
474 | { | |
475 | uint64_t space, resv; | |
476 | ||
477 | /* | |
478 | * Reserve about 1.6% (1/64), or at least 32MB, for allocation | |
479 | * efficiency. | |
480 | * XXX The intent log is not accounted for, so it must fit | |
481 | * within this slop. | |
482 | * | |
483 | * If we're trying to assess whether it's OK to do a free, | |
484 | * cut the reservation in half to allow forward progress | |
485 | * (e.g. make it possible to rm(1) files from a full pool). | |
486 | */ | |
487 | space = spa_get_dspace(dp->dp_spa); | |
488 | resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1); | |
489 | if (netfree) | |
490 | resv >>= 1; | |
491 | ||
492 | return (space - resv); | |
493 | } | |
494 | ||
495 | int | |
496 | dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx) | |
497 | { | |
498 | uint64_t reserved = 0; | |
499 | uint64_t write_limit = (zfs_write_limit_override ? | |
500 | zfs_write_limit_override : dp->dp_write_limit); | |
501 | ||
502 | if (zfs_no_write_throttle) { | |
b128c09f BB |
503 | atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], |
504 | space); | |
34dc7c2f BB |
505 | return (0); |
506 | } | |
507 | ||
508 | /* | |
509 | * Check to see if we have exceeded the maximum allowed IO for | |
510 | * this transaction group. We can do this without locks since | |
511 | * a little slop here is ok. Note that we do the reserved check | |
512 | * with only half the requested reserve: this is because the | |
513 | * reserve requests are worst-case, and we really don't want to | |
514 | * throttle based off of worst-case estimates. | |
515 | */ | |
516 | if (write_limit > 0) { | |
517 | reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK] | |
518 | + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2; | |
519 | ||
520 | if (reserved && reserved > write_limit) | |
521 | return (ERESTART); | |
522 | } | |
523 | ||
524 | atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space); | |
525 | ||
526 | /* | |
527 | * If this transaction group is over 7/8ths capacity, delay | |
528 | * the caller 1 clock tick. This will slow down the "fill" | |
529 | * rate until the sync process can catch up with us. | |
530 | */ | |
b128c09f | 531 | if (reserved && reserved > (write_limit - (write_limit >> 3))) |
34dc7c2f BB |
532 | txg_delay(dp, tx->tx_txg, 1); |
533 | ||
534 | return (0); | |
535 | } | |
536 | ||
537 | void | |
538 | dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) | |
539 | { | |
540 | ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space); | |
541 | atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space); | |
542 | } | |
543 | ||
544 | void | |
545 | dsl_pool_memory_pressure(dsl_pool_t *dp) | |
546 | { | |
34dc7c2f BB |
547 | uint64_t space_inuse = 0; |
548 | int i; | |
549 | ||
550 | if (dp->dp_write_limit == zfs_write_limit_min) | |
551 | return; | |
552 | ||
553 | for (i = 0; i < TXG_SIZE; i++) { | |
554 | space_inuse += dp->dp_space_towrite[i]; | |
555 | space_inuse += dp->dp_tempreserved[i]; | |
556 | } | |
557 | dp->dp_write_limit = MAX(zfs_write_limit_min, | |
558 | MIN(dp->dp_write_limit, space_inuse / 4)); | |
559 | } | |
560 | ||
561 | void | |
562 | dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx) | |
563 | { | |
564 | if (space > 0) { | |
565 | mutex_enter(&dp->dp_lock); | |
566 | dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space; | |
567 | mutex_exit(&dp->dp_lock); | |
568 | } | |
569 | } | |
b128c09f BB |
570 | |
571 | /* ARGSUSED */ | |
572 | static int | |
573 | upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) | |
574 | { | |
575 | dmu_tx_t *tx = arg; | |
576 | dsl_dataset_t *ds, *prev = NULL; | |
577 | int err; | |
578 | dsl_pool_t *dp = spa_get_dsl(spa); | |
579 | ||
580 | err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); | |
581 | if (err) | |
582 | return (err); | |
583 | ||
584 | while (ds->ds_phys->ds_prev_snap_obj != 0) { | |
585 | err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, | |
586 | FTAG, &prev); | |
587 | if (err) { | |
588 | dsl_dataset_rele(ds, FTAG); | |
589 | return (err); | |
590 | } | |
591 | ||
592 | if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) | |
593 | break; | |
594 | dsl_dataset_rele(ds, FTAG); | |
595 | ds = prev; | |
596 | prev = NULL; | |
597 | } | |
598 | ||
599 | if (prev == NULL) { | |
600 | prev = dp->dp_origin_snap; | |
601 | ||
602 | /* | |
603 | * The $ORIGIN can't have any data, or the accounting | |
604 | * will be wrong. | |
605 | */ | |
606 | ASSERT(prev->ds_phys->ds_bp.blk_birth == 0); | |
607 | ||
608 | /* The origin doesn't get attached to itself */ | |
609 | if (ds->ds_object == prev->ds_object) { | |
610 | dsl_dataset_rele(ds, FTAG); | |
611 | return (0); | |
612 | } | |
613 | ||
614 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
615 | ds->ds_phys->ds_prev_snap_obj = prev->ds_object; | |
616 | ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg; | |
617 | ||
618 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); | |
619 | ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object; | |
620 | ||
621 | dmu_buf_will_dirty(prev->ds_dbuf, tx); | |
622 | prev->ds_phys->ds_num_children++; | |
623 | ||
624 | if (ds->ds_phys->ds_next_snap_obj == 0) { | |
625 | ASSERT(ds->ds_prev == NULL); | |
626 | VERIFY(0 == dsl_dataset_hold_obj(dp, | |
627 | ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); | |
628 | } | |
629 | } | |
630 | ||
631 | ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object); | |
632 | ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object); | |
633 | ||
634 | if (prev->ds_phys->ds_next_clones_obj == 0) { | |
428870ff | 635 | dmu_buf_will_dirty(prev->ds_dbuf, tx); |
b128c09f BB |
636 | prev->ds_phys->ds_next_clones_obj = |
637 | zap_create(dp->dp_meta_objset, | |
638 | DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); | |
639 | } | |
640 | VERIFY(0 == zap_add_int(dp->dp_meta_objset, | |
641 | prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx)); | |
642 | ||
643 | dsl_dataset_rele(ds, FTAG); | |
644 | if (prev != dp->dp_origin_snap) | |
645 | dsl_dataset_rele(prev, FTAG); | |
646 | return (0); | |
647 | } | |
648 | ||
649 | void | |
650 | dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
651 | { | |
652 | ASSERT(dmu_tx_is_syncing(tx)); | |
653 | ASSERT(dp->dp_origin_snap != NULL); | |
654 | ||
428870ff BB |
655 | VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb, |
656 | tx, DS_FIND_CHILDREN)); | |
657 | } | |
658 | ||
659 | /* ARGSUSED */ | |
660 | static int | |
661 | upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg) | |
662 | { | |
663 | dmu_tx_t *tx = arg; | |
664 | dsl_dataset_t *ds; | |
665 | dsl_pool_t *dp = spa_get_dsl(spa); | |
666 | objset_t *mos = dp->dp_meta_objset; | |
667 | ||
668 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); | |
669 | ||
670 | if (ds->ds_dir->dd_phys->dd_origin_obj) { | |
671 | dsl_dataset_t *origin; | |
672 | ||
673 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, | |
674 | ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin)); | |
675 | ||
676 | if (origin->ds_dir->dd_phys->dd_clones == 0) { | |
677 | dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); | |
678 | origin->ds_dir->dd_phys->dd_clones = zap_create(mos, | |
679 | DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); | |
680 | } | |
681 | ||
682 | VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, | |
683 | origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); | |
684 | ||
685 | dsl_dataset_rele(origin, FTAG); | |
686 | } | |
687 | ||
688 | dsl_dataset_rele(ds, FTAG); | |
689 | return (0); | |
690 | } | |
691 | ||
692 | void | |
693 | dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx) | |
694 | { | |
428870ff BB |
695 | uint64_t obj; |
696 | ||
d6320ddb BB |
697 | ASSERT(dmu_tx_is_syncing(tx)); |
698 | ||
428870ff BB |
699 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx); |
700 | VERIFY(0 == dsl_pool_open_special_dir(dp, | |
701 | FREE_DIR_NAME, &dp->dp_free_dir)); | |
702 | ||
703 | /* | |
704 | * We can't use bpobj_alloc(), because spa_version() still | |
705 | * returns the old version, and we need a new-version bpobj with | |
706 | * subobj support. So call dmu_object_alloc() directly. | |
707 | */ | |
708 | obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ, | |
709 | SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx); | |
710 | VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
711 | DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx)); | |
712 | VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj, | |
713 | dp->dp_meta_objset, obj)); | |
714 | ||
715 | VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, | |
716 | upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN)); | |
b128c09f BB |
717 | } |
718 | ||
719 | void | |
720 | dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx) | |
721 | { | |
722 | uint64_t dsobj; | |
723 | dsl_dataset_t *ds; | |
724 | ||
725 | ASSERT(dmu_tx_is_syncing(tx)); | |
726 | ASSERT(dp->dp_origin_snap == NULL); | |
727 | ||
728 | /* create the origin dir, ds, & snap-ds */ | |
729 | rw_enter(&dp->dp_config_rwlock, RW_WRITER); | |
730 | dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME, | |
731 | NULL, 0, kcred, tx); | |
732 | VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); | |
428870ff | 733 | dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx); |
b128c09f BB |
734 | VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj, |
735 | dp, &dp->dp_origin_snap)); | |
736 | dsl_dataset_rele(ds, FTAG); | |
737 | rw_exit(&dp->dp_config_rwlock); | |
738 | } | |
9babb374 BB |
739 | |
740 | taskq_t * | |
3558fd73 | 741 | dsl_pool_iput_taskq(dsl_pool_t *dp) |
9babb374 | 742 | { |
3558fd73 | 743 | return (dp->dp_iput_taskq); |
9babb374 | 744 | } |
428870ff BB |
745 | |
746 | /* | |
747 | * Walk through the pool-wide zap object of temporary snapshot user holds | |
748 | * and release them. | |
749 | */ | |
750 | void | |
751 | dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp) | |
752 | { | |
753 | zap_attribute_t za; | |
754 | zap_cursor_t zc; | |
755 | objset_t *mos = dp->dp_meta_objset; | |
756 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
757 | ||
758 | if (zapobj == 0) | |
759 | return; | |
760 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
761 | ||
762 | for (zap_cursor_init(&zc, mos, zapobj); | |
763 | zap_cursor_retrieve(&zc, &za) == 0; | |
764 | zap_cursor_advance(&zc)) { | |
765 | char *htag; | |
766 | uint64_t dsobj; | |
767 | ||
768 | htag = strchr(za.za_name, '-'); | |
769 | *htag = '\0'; | |
770 | ++htag; | |
771 | dsobj = strtonum(za.za_name, NULL); | |
572e2857 | 772 | (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE); |
428870ff BB |
773 | } |
774 | zap_cursor_fini(&zc); | |
775 | } | |
776 | ||
777 | /* | |
778 | * Create the pool-wide zap object for storing temporary snapshot holds. | |
779 | */ | |
780 | void | |
781 | dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx) | |
782 | { | |
783 | objset_t *mos = dp->dp_meta_objset; | |
784 | ||
785 | ASSERT(dp->dp_tmp_userrefs_obj == 0); | |
786 | ASSERT(dmu_tx_is_syncing(tx)); | |
787 | ||
788 | dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS, | |
789 | DMU_OT_NONE, 0, tx); | |
790 | ||
791 | VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, | |
792 | sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0); | |
793 | } | |
794 | ||
795 | static int | |
796 | dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj, | |
797 | const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding) | |
798 | { | |
799 | objset_t *mos = dp->dp_meta_objset; | |
800 | uint64_t zapobj = dp->dp_tmp_userrefs_obj; | |
801 | char *name; | |
802 | int error; | |
803 | ||
804 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); | |
805 | ASSERT(dmu_tx_is_syncing(tx)); | |
806 | ||
807 | /* | |
808 | * If the pool was created prior to SPA_VERSION_USERREFS, the | |
809 | * zap object for temporary holds might not exist yet. | |
810 | */ | |
811 | if (zapobj == 0) { | |
812 | if (holding) { | |
813 | dsl_pool_user_hold_create_obj(dp, tx); | |
814 | zapobj = dp->dp_tmp_userrefs_obj; | |
815 | } else { | |
816 | return (ENOENT); | |
817 | } | |
818 | } | |
819 | ||
820 | name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag); | |
821 | if (holding) | |
822 | error = zap_add(mos, zapobj, name, 8, 1, now, tx); | |
823 | else | |
824 | error = zap_remove(mos, zapobj, name, tx); | |
825 | strfree(name); | |
826 | ||
827 | return (error); | |
828 | } | |
829 | ||
830 | /* | |
831 | * Add a temporary hold for the given dataset object and tag. | |
832 | */ | |
833 | int | |
834 | dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
835 | uint64_t *now, dmu_tx_t *tx) | |
836 | { | |
837 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE)); | |
838 | } | |
839 | ||
840 | /* | |
841 | * Release a temporary hold for the given dataset object and tag. | |
842 | */ | |
843 | int | |
844 | dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag, | |
845 | dmu_tx_t *tx) | |
846 | { | |
847 | return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL, | |
848 | tx, B_FALSE)); | |
849 | } |