Deleted Added
sdiff udiff text old ( 177698 ) new ( 185029 )
full compact
dsl_pool.c (177698) dsl_pool.c (185029)
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

--- 5 unchanged lines hidden (view full) ---

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/*
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

--- 5 unchanged lines hidden (view full) ---

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/*
22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
23 * Use is subject to license terms.
24 */
25
26#pragma ident "%Z%%M% %I% %E% SMI"
27
28#include <sys/dsl_pool.h>
29#include <sys/dsl_dataset.h>
30#include <sys/dsl_dir.h>
31#include <sys/dsl_synctask.h>
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>
26#include <sys/dsl_pool.h>
27#include <sys/dsl_dataset.h>
28#include <sys/dsl_dir.h>
29#include <sys/dsl_synctask.h>
30#include <sys/dmu_tx.h>
31#include <sys/dmu_objset.h>
32#include <sys/arc.h>
33#include <sys/zap.h>
34#include <sys/zio.h>
35#include <sys/zfs_context.h>
36#include <sys/fs/zfs.h>
37#include <sys/zfs_znode.h>
38#include <sys/spa_impl.h>
39
39
40int zfs_no_write_throttle = 0;
41int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
42int zfs_txg_synctime = 5; /* target secs to sync a txg */
43
44uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
45uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
46uint64_t zfs_write_limit_inflated = 0;
47uint64_t zfs_write_limit_override = 0;
48extern uint64_t zfs_write_limit_min;
49
50kmutex_t zfs_write_limit_lock;
51
52static pgcnt_t old_physmem = 0;
53
40static int
54static int
41dsl_pool_open_mos_dir(dsl_pool_t *dp, dsl_dir_t **ddp)
55dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
42{
43 uint64_t obj;
44 int err;
45
46 err = zap_lookup(dp->dp_meta_objset,
47 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
56{
57 uint64_t obj;
58 int err;
59
60 err = zap_lookup(dp->dp_meta_objset,
61 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
48 MOS_DIR_NAME, sizeof (obj), 1, &obj);
62 name, sizeof (obj), 1, &obj);
49 if (err)
50 return (err);
51
63 if (err)
64 return (err);
65
52 return (dsl_dir_open_obj(dp, obj, MOS_DIR_NAME, dp, ddp));
66 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
53}
54
55static dsl_pool_t *
56dsl_pool_open_impl(spa_t *spa, uint64_t txg)
57{
58 dsl_pool_t *dp;
59 blkptr_t *bp = spa_get_rootblkptr(spa);
60
61 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
62 dp->dp_spa = spa;
63 dp->dp_meta_rootbp = *bp;
64 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
67}
68
69static dsl_pool_t *
70dsl_pool_open_impl(spa_t *spa, uint64_t txg)
71{
72 dsl_pool_t *dp;
73 blkptr_t *bp = spa_get_rootblkptr(spa);
74
75 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
76 dp->dp_spa = spa;
77 dp->dp_meta_rootbp = *bp;
78 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
79 dp->dp_write_limit = zfs_write_limit_min;
65 txg_init(dp, txg);
66
67 txg_list_create(&dp->dp_dirty_datasets,
68 offsetof(dsl_dataset_t, ds_dirty_link));
69 txg_list_create(&dp->dp_dirty_dirs,
70 offsetof(dsl_dir_t, dd_dirty_link));
71 txg_list_create(&dp->dp_sync_tasks,
72 offsetof(dsl_sync_task_group_t, dstg_node));
80 txg_init(dp, txg);
81
82 txg_list_create(&dp->dp_dirty_datasets,
83 offsetof(dsl_dataset_t, ds_dirty_link));
84 txg_list_create(&dp->dp_dirty_dirs,
85 offsetof(dsl_dir_t, dd_dirty_link));
86 txg_list_create(&dp->dp_sync_tasks,
87 offsetof(dsl_sync_task_group_t, dstg_node));
73 list_create(&dp->dp_synced_objsets, sizeof (dsl_dataset_t),
88 list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
74 offsetof(dsl_dataset_t, ds_synced_link));
75
89 offsetof(dsl_dataset_t, ds_synced_link));
90
91 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
92 mutex_init(&dp->dp_scrub_cancel_lock, NULL, MUTEX_DEFAULT, NULL);
93
76 return (dp);
77}
78
79int
80dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
81{
82 int err;
83 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
94 return (dp);
95}
96
97int
98dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
99{
100 int err;
101 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
102 dsl_dir_t *dd;
103 dsl_dataset_t *ds;
84 objset_impl_t *osi;
85
104 objset_impl_t *osi;
105
86 rw_enter(&dp->dp_config_rwlock, RW_READER);
106 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
87 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
88 if (err)
89 goto out;
90 dp->dp_meta_objset = &osi->os;
91
92 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
93 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
94 &dp->dp_root_dir_obj);
95 if (err)
96 goto out;
97
98 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
99 NULL, dp, &dp->dp_root_dir);
100 if (err)
101 goto out;
102
107 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
108 if (err)
109 goto out;
110 dp->dp_meta_objset = &osi->os;
111
112 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
113 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
114 &dp->dp_root_dir_obj);
115 if (err)
116 goto out;
117
118 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
119 NULL, dp, &dp->dp_root_dir);
120 if (err)
121 goto out;
122
103 err = dsl_pool_open_mos_dir(dp, &dp->dp_mos_dir);
123 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
104 if (err)
105 goto out;
106
124 if (err)
125 goto out;
126
127 if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
128 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
129 if (err)
130 goto out;
131 err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
132 FTAG, &ds);
133 if (err)
134 goto out;
135 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
136 dp, &dp->dp_origin_snap);
137 if (err)
138 goto out;
139 dsl_dataset_rele(ds, FTAG);
140 dsl_dir_close(dd, dp);
141 }
142
143 /* get scrub status */
144 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
145 DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
146 &dp->dp_scrub_func);
147 if (err == 0) {
148 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
149 DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
150 &dp->dp_scrub_queue_obj);
151 if (err)
152 goto out;
153 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
154 DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
155 &dp->dp_scrub_min_txg);
156 if (err)
157 goto out;
158 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
159 DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
160 &dp->dp_scrub_max_txg);
161 if (err)
162 goto out;
163 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
164 DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
165 &dp->dp_scrub_bookmark);
166 if (err)
167 goto out;
168 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
169 DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
170 &spa->spa_scrub_errors);
171 if (err)
172 goto out;
173 if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
174 /*
175 * A new-type scrub was in progress on an old
176 * pool. Restart from the beginning, since the
177 * old software may have changed the pool in the
178 * meantime.
179 */
180 dsl_pool_scrub_restart(dp);
181 }
182 } else {
183 /*
184 * It's OK if there is no scrub in progress (and if
185 * there was an I/O error, ignore it).
186 */
187 err = 0;
188 }
189
107out:
108 rw_exit(&dp->dp_config_rwlock);
109 if (err)
110 dsl_pool_close(dp);
111 else
112 *dpp = dp;
113
114 return (err);
115}
116
117void
118dsl_pool_close(dsl_pool_t *dp)
119{
190out:
191 rw_exit(&dp->dp_config_rwlock);
192 if (err)
193 dsl_pool_close(dp);
194 else
195 *dpp = dp;
196
197 return (err);
198}
199
200void
201dsl_pool_close(dsl_pool_t *dp)
202{
120 /* drop our reference from dsl_pool_open() */
203 /* drop our references from dsl_pool_open() */
204
205 /*
206 * Since we held the origin_snap from "syncing" context (which
207 * includes pool-opening context), it actually only got a "ref"
208 * and not a hold, so just drop that here.
209 */
210 if (dp->dp_origin_snap)
211 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
121 if (dp->dp_mos_dir)
122 dsl_dir_close(dp->dp_mos_dir, dp);
123 if (dp->dp_root_dir)
124 dsl_dir_close(dp->dp_root_dir, dp);
125
126 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
127 if (dp->dp_meta_objset)
128 dmu_objset_evict(NULL, dp->dp_meta_objset->os);
129
130 txg_list_destroy(&dp->dp_dirty_datasets);
131 txg_list_destroy(&dp->dp_dirty_dirs);
132 txg_list_destroy(&dp->dp_sync_tasks);
212 if (dp->dp_mos_dir)
213 dsl_dir_close(dp->dp_mos_dir, dp);
214 if (dp->dp_root_dir)
215 dsl_dir_close(dp->dp_root_dir, dp);
216
217 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
218 if (dp->dp_meta_objset)
219 dmu_objset_evict(NULL, dp->dp_meta_objset->os);
220
221 txg_list_destroy(&dp->dp_dirty_datasets);
222 txg_list_destroy(&dp->dp_dirty_dirs);
223 txg_list_destroy(&dp->dp_sync_tasks);
133 list_destroy(&dp->dp_synced_objsets);
224 list_destroy(&dp->dp_synced_datasets);
134
225
135 arc_flush();
226 arc_flush(dp->dp_spa);
136 txg_fini(dp);
137 rw_destroy(&dp->dp_config_rwlock);
227 txg_fini(dp);
228 rw_destroy(&dp->dp_config_rwlock);
229 mutex_destroy(&dp->dp_lock);
230 mutex_destroy(&dp->dp_scrub_cancel_lock);
138 kmem_free(dp, sizeof (dsl_pool_t));
139}
140
141dsl_pool_t *
231 kmem_free(dp, sizeof (dsl_pool_t));
232}
233
234dsl_pool_t *
142dsl_pool_create(spa_t *spa, uint64_t txg)
235dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
143{
144 int err;
145 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
146 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
236{
237 int err;
238 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
239 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
240 objset_impl_t *osip;
241 dsl_dataset_t *ds;
242 uint64_t dsobj;
243
244 /* create and open the MOS (meta-objset) */
147 dp->dp_meta_objset = &dmu_objset_create_impl(spa,
148 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
149
150 /* create the pool directory */
151 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
152 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
153 ASSERT3U(err, ==, 0);
154
155 /* create and open the root dir */
245 dp->dp_meta_objset = &dmu_objset_create_impl(spa,
246 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
247
248 /* create the pool directory */
249 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
250 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
251 ASSERT3U(err, ==, 0);
252
253 /* create and open the root dir */
156 dsl_dataset_create_root(dp, &dp->dp_root_dir_obj, tx);
254 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
157 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
158 NULL, dp, &dp->dp_root_dir));
159
160 /* create and open the meta-objset dir */
255 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
256 NULL, dp, &dp->dp_root_dir));
257
258 /* create and open the meta-objset dir */
161 (void) dsl_dir_create_sync(dp->dp_root_dir, MOS_DIR_NAME, tx);
162 VERIFY(0 == dsl_pool_open_mos_dir(dp, &dp->dp_mos_dir));
259 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
260 VERIFY(0 == dsl_pool_open_special_dir(dp,
261 MOS_DIR_NAME, &dp->dp_mos_dir));
163
262
263 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
264 dsl_pool_create_origin(dp, tx);
265
266 /* create the root dataset */
267 dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
268
269 /* create the root objset */
270 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
271 osip = dmu_objset_create_impl(dp->dp_spa, ds,
272 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
273#ifdef _KERNEL
274 zfs_create_fs(&osip->os, kcred, zplprops, tx);
275#endif
276 dsl_dataset_rele(ds, FTAG);
277
164 dmu_tx_commit(tx);
165
166 return (dp);
167}
168
169void
170dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
171{
172 zio_t *zio;
173 dmu_tx_t *tx;
174 dsl_dir_t *dd;
175 dsl_dataset_t *ds;
176 dsl_sync_task_group_t *dstg;
177 objset_impl_t *mosi = dp->dp_meta_objset->os;
278 dmu_tx_commit(tx);
279
280 return (dp);
281}
282
283void
284dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
285{
286 zio_t *zio;
287 dmu_tx_t *tx;
288 dsl_dir_t *dd;
289 dsl_dataset_t *ds;
290 dsl_sync_task_group_t *dstg;
291 objset_impl_t *mosi = dp->dp_meta_objset->os;
292 hrtime_t start, write_time;
293 uint64_t data_written;
178 int err;
179
180 tx = dmu_tx_create_assigned(dp, txg);
181
294 int err;
295
296 tx = dmu_tx_create_assigned(dp, txg);
297
298 dp->dp_read_overhead = 0;
182 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
183 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
184 if (!list_link_active(&ds->ds_synced_link))
299 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
300 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
301 if (!list_link_active(&ds->ds_synced_link))
185 list_insert_tail(&dp->dp_synced_objsets, ds);
302 list_insert_tail(&dp->dp_synced_datasets, ds);
186 else
187 dmu_buf_rele(ds->ds_dbuf, ds);
188 dsl_dataset_sync(ds, zio, tx);
189 }
303 else
304 dmu_buf_rele(ds->ds_dbuf, ds);
305 dsl_dataset_sync(ds, zio, tx);
306 }
307 DTRACE_PROBE(pool_sync__1setup);
308
309 start = gethrtime();
190 err = zio_wait(zio);
310 err = zio_wait(zio);
311 write_time = gethrtime() - start;
191 ASSERT(err == 0);
312 ASSERT(err == 0);
313 DTRACE_PROBE(pool_sync__2rootzio);
192
193 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
194 dsl_sync_task_group_sync(dstg, tx);
314
315 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
316 dsl_sync_task_group_sync(dstg, tx);
317 DTRACE_PROBE(pool_sync__3task);
318
319 start = gethrtime();
195 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
196 dsl_dir_sync(dd, tx);
320 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
321 dsl_dir_sync(dd, tx);
322 write_time += gethrtime() - start;
197
323
324 if (spa_sync_pass(dp->dp_spa) == 1)
325 dsl_pool_scrub_sync(dp, tx);
326
327 start = gethrtime();
198 if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
199 list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
200 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
201 dmu_objset_sync(mosi, zio, tx);
202 err = zio_wait(zio);
203 ASSERT(err == 0);
204 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
205 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
206 }
328 if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
329 list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
330 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
331 dmu_objset_sync(mosi, zio, tx);
332 err = zio_wait(zio);
333 ASSERT(err == 0);
334 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
335 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
336 }
337 write_time += gethrtime() - start;
338 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
339 hrtime_t, dp->dp_read_overhead);
340 write_time -= dp->dp_read_overhead;
207
208 dmu_tx_commit(tx);
341
342 dmu_tx_commit(tx);
343
344 data_written = dp->dp_space_towrite[txg & TXG_MASK];
345 dp->dp_space_towrite[txg & TXG_MASK] = 0;
346 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
347
348 /*
349 * If the write limit max has not been explicitly set, set it
350 * to a fraction of available physical memory (default 1/8th).
351 * Note that we must inflate the limit because the spa
352 * inflates write sizes to account for data replication.
353 * Check this each sync phase to catch changing memory size.
354 */
355 if (physmem != old_physmem && zfs_write_limit_shift) {
356 mutex_enter(&zfs_write_limit_lock);
357 old_physmem = physmem;
358 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
359 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
360 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
361 mutex_exit(&zfs_write_limit_lock);
362 }
363
364 /*
365 * Attempt to keep the sync time consistent by adjusting the
366 * amount of write traffic allowed into each transaction group.
367 * Weight the throughput calculation towards the current value:
368 * thru = 3/4 old_thru + 1/4 new_thru
369 */
370 ASSERT(zfs_write_limit_min > 0);
371 if (data_written > zfs_write_limit_min / 8 && write_time > 0) {
372 uint64_t throughput = (data_written * NANOSEC) / write_time;
373 if (dp->dp_throughput)
374 dp->dp_throughput = throughput / 4 +
375 3 * dp->dp_throughput / 4;
376 else
377 dp->dp_throughput = throughput;
378 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
379 MAX(zfs_write_limit_min,
380 dp->dp_throughput * zfs_txg_synctime));
381 }
209}
210
211void
212dsl_pool_zil_clean(dsl_pool_t *dp)
213{
214 dsl_dataset_t *ds;
215
382}
383
384void
385dsl_pool_zil_clean(dsl_pool_t *dp)
386{
387 dsl_dataset_t *ds;
388
216 while (ds = list_head(&dp->dp_synced_objsets)) {
217 list_remove(&dp->dp_synced_objsets, ds);
389 while (ds = list_head(&dp->dp_synced_datasets)) {
390 list_remove(&dp->dp_synced_datasets, ds);
218 ASSERT(ds->ds_user_ptr != NULL);
219 zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
220 dmu_buf_rele(ds->ds_dbuf, ds);
221 }
222}
223
224/*
225 * TRUE if the current thread is the tx_sync_thread or if we

--- 23 unchanged lines hidden (view full) ---

249 */
250 space = spa_get_dspace(dp->dp_spa);
251 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
252 if (netfree)
253 resv >>= 1;
254
255 return (space - resv);
256}
391 ASSERT(ds->ds_user_ptr != NULL);
392 zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
393 dmu_buf_rele(ds->ds_dbuf, ds);
394 }
395}
396
397/*
398 * TRUE if the current thread is the tx_sync_thread or if we

--- 23 unchanged lines hidden (view full) ---

422 */
423 space = spa_get_dspace(dp->dp_spa);
424 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
425 if (netfree)
426 resv >>= 1;
427
428 return (space - resv);
429}
430
431int
432dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
433{
434 uint64_t reserved = 0;
435 uint64_t write_limit = (zfs_write_limit_override ?
436 zfs_write_limit_override : dp->dp_write_limit);
437
438 if (zfs_no_write_throttle) {
439 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
440 space);
441 return (0);
442 }
443
444 /*
445 * Check to see if we have exceeded the maximum allowed IO for
446 * this transaction group. We can do this without locks since
447 * a little slop here is ok. Note that we do the reserved check
448 * with only half the requested reserve: this is because the
449 * reserve requests are worst-case, and we really don't want to
450 * throttle based off of worst-case estimates.
451 */
452 if (write_limit > 0) {
453 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
454 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
455
456 if (reserved && reserved > write_limit)
457 return (ERESTART);
458 }
459
460 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
461
462 /*
463 * If this transaction group is over 7/8ths capacity, delay
464 * the caller 1 clock tick. This will slow down the "fill"
465 * rate until the sync process can catch up with us.
466 */
467 if (reserved && reserved > (write_limit - (write_limit >> 3)))
468 txg_delay(dp, tx->tx_txg, 1);
469
470 return (0);
471}
472
473void
474dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
475{
476 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
477 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
478}
479
480void
481dsl_pool_memory_pressure(dsl_pool_t *dp)
482{
483 uint64_t space_inuse = 0;
484 int i;
485
486 if (dp->dp_write_limit == zfs_write_limit_min)
487 return;
488
489 for (i = 0; i < TXG_SIZE; i++) {
490 space_inuse += dp->dp_space_towrite[i];
491 space_inuse += dp->dp_tempreserved[i];
492 }
493 dp->dp_write_limit = MAX(zfs_write_limit_min,
494 MIN(dp->dp_write_limit, space_inuse / 4));
495}
496
497void
498dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
499{
500 if (space > 0) {
501 mutex_enter(&dp->dp_lock);
502 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
503 mutex_exit(&dp->dp_lock);
504 }
505}
506
507/* ARGSUSED */
508static int
509upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
510{
511 dmu_tx_t *tx = arg;
512 dsl_dataset_t *ds, *prev = NULL;
513 int err;
514 dsl_pool_t *dp = spa_get_dsl(spa);
515
516 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
517 if (err)
518 return (err);
519
520 while (ds->ds_phys->ds_prev_snap_obj != 0) {
521 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
522 FTAG, &prev);
523 if (err) {
524 dsl_dataset_rele(ds, FTAG);
525 return (err);
526 }
527
528 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
529 break;
530 dsl_dataset_rele(ds, FTAG);
531 ds = prev;
532 prev = NULL;
533 }
534
535 if (prev == NULL) {
536 prev = dp->dp_origin_snap;
537
538 /*
539 * The $ORIGIN can't have any data, or the accounting
540 * will be wrong.
541 */
542 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
543
544 /* The origin doesn't get attached to itself */
545 if (ds->ds_object == prev->ds_object) {
546 dsl_dataset_rele(ds, FTAG);
547 return (0);
548 }
549
550 dmu_buf_will_dirty(ds->ds_dbuf, tx);
551 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
552 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
553
554 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
555 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
556
557 dmu_buf_will_dirty(prev->ds_dbuf, tx);
558 prev->ds_phys->ds_num_children++;
559
560 if (ds->ds_phys->ds_next_snap_obj == 0) {
561 ASSERT(ds->ds_prev == NULL);
562 VERIFY(0 == dsl_dataset_hold_obj(dp,
563 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
564 }
565 }
566
567 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
568 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
569
570 if (prev->ds_phys->ds_next_clones_obj == 0) {
571 prev->ds_phys->ds_next_clones_obj =
572 zap_create(dp->dp_meta_objset,
573 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
574 }
575 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
576 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
577
578 dsl_dataset_rele(ds, FTAG);
579 if (prev != dp->dp_origin_snap)
580 dsl_dataset_rele(prev, FTAG);
581 return (0);
582}
583
584void
585dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
586{
587 ASSERT(dmu_tx_is_syncing(tx));
588 ASSERT(dp->dp_origin_snap != NULL);
589
590 (void) dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
591 tx, DS_FIND_CHILDREN);
592}
593
594void
595dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
596{
597 uint64_t dsobj;
598 dsl_dataset_t *ds;
599
600 ASSERT(dmu_tx_is_syncing(tx));
601 ASSERT(dp->dp_origin_snap == NULL);
602
603 /* create the origin dir, ds, & snap-ds */
604 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
605 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
606 NULL, 0, kcred, tx);
607 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
608 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx);
609 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
610 dp, &dp->dp_origin_snap));
611 dsl_dataset_rele(ds, FTAG);
612 rw_exit(&dp->dp_config_rwlock);
613}