Print this page
5056 ZFS deadlock on db_mtx and dn_holds
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/dsl_pool.c
+++ new/usr/src/uts/common/fs/zfs/dsl_pool.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013 Steven Hartland. All rights reserved.
25 + * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 26 */
26 27
27 28 #include <sys/dsl_pool.h>
28 29 #include <sys/dsl_dataset.h>
29 30 #include <sys/dsl_prop.h>
30 31 #include <sys/dsl_dir.h>
31 32 #include <sys/dsl_synctask.h>
32 33 #include <sys/dsl_scan.h>
33 34 #include <sys/dnode.h>
34 35 #include <sys/dmu_tx.h>
35 36 #include <sys/dmu_objset.h>
36 37 #include <sys/arc.h>
37 38 #include <sys/zap.h>
38 39 #include <sys/zio.h>
39 40 #include <sys/zfs_context.h>
40 41 #include <sys/fs/zfs.h>
41 42 #include <sys/zfs_znode.h>
42 43 #include <sys/spa_impl.h>
43 44 #include <sys/dsl_deadlist.h>
44 45 #include <sys/bptree.h>
45 46 #include <sys/zfeature.h>
46 47 #include <sys/zil_impl.h>
47 48 #include <sys/dsl_userhold.h>
48 49
49 50 /*
50 51 * ZFS Write Throttle
51 52 * ------------------
52 53 *
53 54 * ZFS must limit the rate of incoming writes to the rate at which it is able
54 55 * to sync data modifications to the backend storage. Throttling by too much
55 56 * creates an artificial limit; throttling by too little can only be sustained
56 57 * for short periods and would lead to highly lumpy performance. On a per-pool
57 58 * basis, ZFS tracks the amount of modified (dirty) data. As operations change
58 59 * data, the amount of dirty data increases; as ZFS syncs out data, the amount
59 60 * of dirty data decreases. When the amount of dirty data exceeds a
60 61 * predetermined threshold further modifications are blocked until the amount
61 62 * of dirty data decreases (as data is synced out).
62 63 *
63 64 * The limit on dirty data is tunable, and should be adjusted according to
64 65 * both the IO capacity and available memory of the system. The larger the
65 66 * window, the more ZFS is able to aggregate and amortize metadata (and data)
66 67 * changes. However, memory is a limited resource, and allowing for more dirty
67 68 * data comes at the cost of keeping other useful data in memory (for example
68 69 * ZFS data cached by the ARC).
69 70 *
70 71 * Implementation
71 72 *
72 73 * As buffers are modified dsl_pool_willuse_space() increments both the per-
73 74 * txg (dp_dirty_pertxg[]) and poolwide (dp_dirty_total) accounting of
74 75 * dirty space used; dsl_pool_dirty_space() decrements those values as data
75 76 * is synced out from dsl_pool_sync(). While only the poolwide value is
76 77 * relevant, the per-txg value is useful for debugging. The tunable
77 78 * zfs_dirty_data_max determines the dirty space limit. Once that value is
78 79 * exceeded, new writes are halted until space frees up.
79 80 *
80 81 * The zfs_dirty_data_sync tunable dictates the threshold at which we
81 82 * ensure that there is a txg syncing (see the comment in txg.c for a full
82 83 * description of transaction group stages).
83 84 *
84 85 * The IO scheduler uses both the dirty space limit and current amount of
85 86 * dirty data as inputs. Those values affect the number of concurrent IOs ZFS
86 87 * issues. See the comment in vdev_queue.c for details of the IO scheduler.
87 88 *
88 89 * The delay is also calculated based on the amount of dirty data. See the
89 90 * comment above dmu_tx_delay() for details.
90 91 */
91 92
92 93 /*
93 94 * zfs_dirty_data_max will be set to zfs_dirty_data_max_percent% of all memory,
94 95 * capped at zfs_dirty_data_max_max. It can also be overridden in /etc/system.
95 96 */
96 97 uint64_t zfs_dirty_data_max;
97 98 uint64_t zfs_dirty_data_max_max = 4ULL * 1024 * 1024 * 1024;
98 99 int zfs_dirty_data_max_percent = 10;
99 100
100 101 /*
101 102 * If there is at least this much dirty data, push out a txg.
102 103 */
103 104 uint64_t zfs_dirty_data_sync = 64 * 1024 * 1024;
104 105
105 106 /*
106 107 * Once there is this amount of dirty data, the dmu_tx_delay() will kick in
107 108 * and delay each transaction.
108 109 * This value should be >= zfs_vdev_async_write_active_max_dirty_percent.
109 110 */
110 111 int zfs_delay_min_dirty_percent = 60;
111 112
112 113 /*
113 114 * This controls how quickly the delay approaches infinity.
114 115 * Larger values cause it to delay more for a given amount of dirty data.
115 116 * Therefore larger values will cause there to be less dirty data for a
116 117 * given throughput.
117 118 *
118 119 * For the smoothest delay, this value should be about 1 billion divided
119 120 * by the maximum number of operations per second. This will smoothly
120 121 * handle between 10x and 1/10th this number.
121 122 *
122 123 * Note: zfs_delay_scale * zfs_dirty_data_max must be < 2^64, due to the
123 124 * multiply in dmu_tx_delay().
124 125 */
125 126 uint64_t zfs_delay_scale = 1000 * 1000 * 1000 / 2000;
126 127
127 128
128 129 hrtime_t zfs_throttle_delay = MSEC2NSEC(10);
129 130 hrtime_t zfs_throttle_resolution = MSEC2NSEC(10);
130 131
131 132 int
132 133 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
133 134 {
134 135 uint64_t obj;
135 136 int err;
136 137
137 138 err = zap_lookup(dp->dp_meta_objset,
138 139 dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj,
139 140 name, sizeof (obj), 1, &obj);
140 141 if (err)
141 142 return (err);
142 143
143 144 return (dsl_dir_hold_obj(dp, obj, name, dp, ddp));
144 145 }
145 146
146 147 static dsl_pool_t *
147 148 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
148 149 {
149 150 dsl_pool_t *dp;
150 151 blkptr_t *bp = spa_get_rootblkptr(spa);
151 152
152 153 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
153 154 dp->dp_spa = spa;
154 155 dp->dp_meta_rootbp = *bp;
155 156 rrw_init(&dp->dp_config_rwlock, B_TRUE);
156 157 txg_init(dp, txg);
157 158
158 159 txg_list_create(&dp->dp_dirty_datasets,
159 160 offsetof(dsl_dataset_t, ds_dirty_link));
160 161 txg_list_create(&dp->dp_dirty_zilogs,
161 162 offsetof(zilog_t, zl_dirty_link));
162 163 txg_list_create(&dp->dp_dirty_dirs,
163 164 offsetof(dsl_dir_t, dd_dirty_link));
164 165 txg_list_create(&dp->dp_sync_tasks,
165 166 offsetof(dsl_sync_task_t, dst_node));
166 167
167 168 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
168 169 cv_init(&dp->dp_spaceavail_cv, NULL, CV_DEFAULT, NULL);
169 170
170 171 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
171 172 1, 4, 0);
172 173
173 174 return (dp);
174 175 }
175 176
176 177 int
177 178 dsl_pool_init(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
178 179 {
179 180 int err;
180 181 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
181 182
182 183 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
183 184 &dp->dp_meta_objset);
184 185 if (err != 0)
185 186 dsl_pool_close(dp);
186 187 else
187 188 *dpp = dp;
188 189
189 190 return (err);
190 191 }
191 192
192 193 int
193 194 dsl_pool_open(dsl_pool_t *dp)
194 195 {
195 196 int err;
196 197 dsl_dir_t *dd;
197 198 dsl_dataset_t *ds;
198 199 uint64_t obj;
199 200
200 201 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
201 202 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
202 203 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
203 204 &dp->dp_root_dir_obj);
204 205 if (err)
205 206 goto out;
206 207
207 208 err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
208 209 NULL, dp, &dp->dp_root_dir);
209 210 if (err)
210 211 goto out;
211 212
212 213 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
213 214 if (err)
214 215 goto out;
215 216
216 217 if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) {
217 218 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
218 219 if (err)
219 220 goto out;
220 221 err = dsl_dataset_hold_obj(dp,
221 222 dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds);
222 223 if (err == 0) {
223 224 err = dsl_dataset_hold_obj(dp,
224 225 dsl_dataset_phys(ds)->ds_prev_snap_obj, dp,
225 226 &dp->dp_origin_snap);
226 227 dsl_dataset_rele(ds, FTAG);
227 228 }
228 229 dsl_dir_rele(dd, dp);
229 230 if (err)
230 231 goto out;
231 232 }
232 233
233 234 if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
234 235 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
235 236 &dp->dp_free_dir);
236 237 if (err)
237 238 goto out;
238 239
239 240 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
240 241 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
241 242 if (err)
242 243 goto out;
243 244 VERIFY0(bpobj_open(&dp->dp_free_bpobj,
244 245 dp->dp_meta_objset, obj));
245 246 }
246 247
247 248 /*
248 249 * Note: errors ignored, because the leak dir will not exist if we
249 250 * have not encountered a leak yet.
250 251 */
251 252 (void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME,
252 253 &dp->dp_leak_dir);
253 254
254 255 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
255 256 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
256 257 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
257 258 &dp->dp_bptree_obj);
258 259 if (err != 0)
259 260 goto out;
260 261 }
261 262
262 263 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) {
263 264 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
264 265 DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
265 266 &dp->dp_empty_bpobj);
266 267 if (err != 0)
267 268 goto out;
268 269 }
269 270
270 271 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
271 272 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
272 273 &dp->dp_tmp_userrefs_obj);
273 274 if (err == ENOENT)
274 275 err = 0;
275 276 if (err)
276 277 goto out;
277 278
278 279 err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
279 280
280 281 out:
281 282 rrw_exit(&dp->dp_config_rwlock, FTAG);
282 283 return (err);
283 284 }
284 285
285 286 void
286 287 dsl_pool_close(dsl_pool_t *dp)
287 288 {
288 289 /*
289 290 * Drop our references from dsl_pool_open().
290 291 *
291 292 * Since we held the origin_snap from "syncing" context (which
292 293 * includes pool-opening context), it actually only got a "ref"
293 294 * and not a hold, so just drop that here.
294 295 */
295 296 if (dp->dp_origin_snap)
296 297 dsl_dataset_rele(dp->dp_origin_snap, dp);
297 298 if (dp->dp_mos_dir)
298 299 dsl_dir_rele(dp->dp_mos_dir, dp);
299 300 if (dp->dp_free_dir)
300 301 dsl_dir_rele(dp->dp_free_dir, dp);
301 302 if (dp->dp_leak_dir)
302 303 dsl_dir_rele(dp->dp_leak_dir, dp);
303 304 if (dp->dp_root_dir)
304 305 dsl_dir_rele(dp->dp_root_dir, dp);
305 306
306 307 bpobj_close(&dp->dp_free_bpobj);
307 308
308 309 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
309 310 if (dp->dp_meta_objset)
|
↓ open down ↓ |
275 lines elided |
↑ open up ↑ |
310 311 dmu_objset_evict(dp->dp_meta_objset);
311 312
312 313 txg_list_destroy(&dp->dp_dirty_datasets);
313 314 txg_list_destroy(&dp->dp_dirty_zilogs);
314 315 txg_list_destroy(&dp->dp_sync_tasks);
315 316 txg_list_destroy(&dp->dp_dirty_dirs);
316 317
317 318 arc_flush(dp->dp_spa);
318 319 txg_fini(dp);
319 320 dsl_scan_fini(dp);
321 + dmu_buf_user_evict_wait();
322 +
320 323 rrw_destroy(&dp->dp_config_rwlock);
321 324 mutex_destroy(&dp->dp_lock);
322 325 taskq_destroy(dp->dp_vnrele_taskq);
323 326 if (dp->dp_blkstats)
324 327 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
325 328 kmem_free(dp, sizeof (dsl_pool_t));
326 329 }
327 330
328 331 dsl_pool_t *
329 332 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
330 333 {
331 334 int err;
332 335 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
333 336 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
334 337 objset_t *os;
335 338 dsl_dataset_t *ds;
336 339 uint64_t obj;
337 340
338 341 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
339 342
340 343 /* create and open the MOS (meta-objset) */
341 344 dp->dp_meta_objset = dmu_objset_create_impl(spa,
342 345 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
343 346
344 347 /* create the pool directory */
345 348 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
346 349 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
347 350 ASSERT0(err);
348 351
349 352 /* Initialize scan structures */
350 353 VERIFY0(dsl_scan_init(dp, txg));
351 354
352 355 /* create and open the root dir */
353 356 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
354 357 VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
355 358 NULL, dp, &dp->dp_root_dir));
356 359
357 360 /* create and open the meta-objset dir */
358 361 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
359 362 VERIFY0(dsl_pool_open_special_dir(dp,
360 363 MOS_DIR_NAME, &dp->dp_mos_dir));
361 364
362 365 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
363 366 /* create and open the free dir */
364 367 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
365 368 FREE_DIR_NAME, tx);
366 369 VERIFY0(dsl_pool_open_special_dir(dp,
367 370 FREE_DIR_NAME, &dp->dp_free_dir));
368 371
369 372 /* create and open the free_bplist */
370 373 obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx);
371 374 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
372 375 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
373 376 VERIFY0(bpobj_open(&dp->dp_free_bpobj,
374 377 dp->dp_meta_objset, obj));
375 378 }
376 379
377 380 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
378 381 dsl_pool_create_origin(dp, tx);
379 382
380 383 /* create the root dataset */
381 384 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
382 385
383 386 /* create the root objset */
384 387 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
385 388 os = dmu_objset_create_impl(dp->dp_spa, ds,
386 389 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
387 390 #ifdef _KERNEL
388 391 zfs_create_fs(os, kcred, zplprops, tx);
389 392 #endif
390 393 dsl_dataset_rele(ds, FTAG);
391 394
392 395 dmu_tx_commit(tx);
393 396
394 397 rrw_exit(&dp->dp_config_rwlock, FTAG);
395 398
396 399 return (dp);
397 400 }
398 401
399 402 /*
400 403 * Account for the meta-objset space in its placeholder dsl_dir.
401 404 */
402 405 void
403 406 dsl_pool_mos_diduse_space(dsl_pool_t *dp,
404 407 int64_t used, int64_t comp, int64_t uncomp)
405 408 {
406 409 ASSERT3U(comp, ==, uncomp); /* it's all metadata */
407 410 mutex_enter(&dp->dp_lock);
408 411 dp->dp_mos_used_delta += used;
409 412 dp->dp_mos_compressed_delta += comp;
410 413 dp->dp_mos_uncompressed_delta += uncomp;
411 414 mutex_exit(&dp->dp_lock);
412 415 }
413 416
414 417 static int
415 418 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
416 419 {
417 420 dsl_deadlist_t *dl = arg;
418 421 dsl_deadlist_insert(dl, bp, tx);
419 422 return (0);
420 423 }
421 424
422 425 static void
423 426 dsl_pool_sync_mos(dsl_pool_t *dp, dmu_tx_t *tx)
424 427 {
425 428 zio_t *zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
426 429 dmu_objset_sync(dp->dp_meta_objset, zio, tx);
427 430 VERIFY0(zio_wait(zio));
428 431 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
429 432 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
430 433 }
431 434
432 435 static void
433 436 dsl_pool_dirty_delta(dsl_pool_t *dp, int64_t delta)
434 437 {
435 438 ASSERT(MUTEX_HELD(&dp->dp_lock));
436 439
437 440 if (delta < 0)
438 441 ASSERT3U(-delta, <=, dp->dp_dirty_total);
439 442
440 443 dp->dp_dirty_total += delta;
441 444
442 445 /*
443 446 * Note: we signal even when increasing dp_dirty_total.
444 447 * This ensures forward progress -- each thread wakes the next waiter.
445 448 */
446 449 if (dp->dp_dirty_total <= zfs_dirty_data_max)
447 450 cv_signal(&dp->dp_spaceavail_cv);
448 451 }
449 452
450 453 void
451 454 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
452 455 {
453 456 zio_t *zio;
454 457 dmu_tx_t *tx;
455 458 dsl_dir_t *dd;
456 459 dsl_dataset_t *ds;
457 460 objset_t *mos = dp->dp_meta_objset;
458 461 list_t synced_datasets;
459 462
460 463 list_create(&synced_datasets, sizeof (dsl_dataset_t),
461 464 offsetof(dsl_dataset_t, ds_synced_link));
462 465
463 466 tx = dmu_tx_create_assigned(dp, txg);
464 467
465 468 /*
466 469 * Write out all dirty blocks of dirty datasets.
467 470 */
468 471 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
469 472 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) {
470 473 /*
471 474 * We must not sync any non-MOS datasets twice, because
472 475 * we may have taken a snapshot of them. However, we
473 476 * may sync newly-created datasets on pass 2.
474 477 */
475 478 ASSERT(!list_link_active(&ds->ds_synced_link));
476 479 list_insert_tail(&synced_datasets, ds);
477 480 dsl_dataset_sync(ds, zio, tx);
478 481 }
479 482 VERIFY0(zio_wait(zio));
480 483
481 484 /*
482 485 * We have written all of the accounted dirty data, so our
483 486 * dp_space_towrite should now be zero. However, some seldom-used
484 487 * code paths do not adhere to this (e.g. dbuf_undirty(), also
485 488 * rounding error in dbuf_write_physdone).
486 489 * Shore up the accounting of any dirtied space now.
487 490 */
488 491 dsl_pool_undirty_space(dp, dp->dp_dirty_pertxg[txg & TXG_MASK], txg);
489 492
490 493 /*
491 494 * After the data blocks have been written (ensured by the zio_wait()
492 495 * above), update the user/group space accounting.
493 496 */
494 497 for (ds = list_head(&synced_datasets); ds != NULL;
495 498 ds = list_next(&synced_datasets, ds)) {
496 499 dmu_objset_do_userquota_updates(ds->ds_objset, tx);
497 500 }
498 501
499 502 /*
500 503 * Sync the datasets again to push out the changes due to
501 504 * userspace updates. This must be done before we process the
502 505 * sync tasks, so that any snapshots will have the correct
503 506 * user accounting information (and we won't get confused
504 507 * about which blocks are part of the snapshot).
505 508 */
506 509 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
507 510 while ((ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) != NULL) {
508 511 ASSERT(list_link_active(&ds->ds_synced_link));
509 512 dmu_buf_rele(ds->ds_dbuf, ds);
510 513 dsl_dataset_sync(ds, zio, tx);
511 514 }
512 515 VERIFY0(zio_wait(zio));
513 516
514 517 /*
515 518 * Now that the datasets have been completely synced, we can
516 519 * clean up our in-memory structures accumulated while syncing:
517 520 *
518 521 * - move dead blocks from the pending deadlist to the on-disk deadlist
519 522 * - release hold from dsl_dataset_dirty()
520 523 */
521 524 while ((ds = list_remove_head(&synced_datasets)) != NULL) {
522 525 objset_t *os = ds->ds_objset;
523 526 bplist_iterate(&ds->ds_pending_deadlist,
524 527 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
525 528 ASSERT(!dmu_objset_is_dirty(os, txg));
526 529 dmu_buf_rele(ds->ds_dbuf, ds);
527 530 }
528 531 while ((dd = txg_list_remove(&dp->dp_dirty_dirs, txg)) != NULL) {
529 532 dsl_dir_sync(dd, tx);
530 533 }
531 534
532 535 /*
533 536 * The MOS's space is accounted for in the pool/$MOS
534 537 * (dp_mos_dir). We can't modify the mos while we're syncing
535 538 * it, so we remember the deltas and apply them here.
536 539 */
537 540 if (dp->dp_mos_used_delta != 0 || dp->dp_mos_compressed_delta != 0 ||
538 541 dp->dp_mos_uncompressed_delta != 0) {
539 542 dsl_dir_diduse_space(dp->dp_mos_dir, DD_USED_HEAD,
540 543 dp->dp_mos_used_delta,
541 544 dp->dp_mos_compressed_delta,
542 545 dp->dp_mos_uncompressed_delta, tx);
543 546 dp->dp_mos_used_delta = 0;
544 547 dp->dp_mos_compressed_delta = 0;
545 548 dp->dp_mos_uncompressed_delta = 0;
546 549 }
547 550
548 551 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
549 552 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
550 553 dsl_pool_sync_mos(dp, tx);
551 554 }
552 555
553 556 /*
554 557 * If we modify a dataset in the same txg that we want to destroy it,
555 558 * its dsl_dir's dd_dbuf will be dirty, and thus have a hold on it.
556 559 * dsl_dir_destroy_check() will fail if there are unexpected holds.
557 560 * Therefore, we want to sync the MOS (thus syncing the dd_dbuf
558 561 * and clearing the hold on it) before we process the sync_tasks.
559 562 * The MOS data dirtied by the sync_tasks will be synced on the next
560 563 * pass.
561 564 */
562 565 if (!txg_list_empty(&dp->dp_sync_tasks, txg)) {
563 566 dsl_sync_task_t *dst;
564 567 /*
565 568 * No more sync tasks should have been added while we
566 569 * were syncing.
567 570 */
568 571 ASSERT3U(spa_sync_pass(dp->dp_spa), ==, 1);
569 572 while ((dst = txg_list_remove(&dp->dp_sync_tasks, txg)) != NULL)
570 573 dsl_sync_task_sync(dst, tx);
571 574 }
572 575
573 576 dmu_tx_commit(tx);
574 577
575 578 DTRACE_PROBE2(dsl_pool_sync__done, dsl_pool_t *dp, dp, uint64_t, txg);
576 579 }
577 580
578 581 void
579 582 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
580 583 {
581 584 zilog_t *zilog;
582 585
583 586 while (zilog = txg_list_remove(&dp->dp_dirty_zilogs, txg)) {
584 587 dsl_dataset_t *ds = dmu_objset_ds(zilog->zl_os);
585 588 zil_clean(zilog, txg);
586 589 ASSERT(!dmu_objset_is_dirty(zilog->zl_os, txg));
587 590 dmu_buf_rele(ds->ds_dbuf, zilog);
588 591 }
589 592 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
590 593 }
591 594
592 595 /*
593 596 * TRUE if the current thread is the tx_sync_thread or if we
594 597 * are being called from SPA context during pool initialization.
595 598 */
596 599 int
597 600 dsl_pool_sync_context(dsl_pool_t *dp)
598 601 {
599 602 return (curthread == dp->dp_tx.tx_sync_thread ||
600 603 spa_is_initializing(dp->dp_spa));
601 604 }
602 605
603 606 uint64_t
604 607 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
605 608 {
606 609 uint64_t space, resv;
607 610
608 611 /*
609 612 * If we're trying to assess whether it's OK to do a free,
610 613 * cut the reservation in half to allow forward progress
611 614 * (e.g. make it possible to rm(1) files from a full pool).
612 615 */
613 616 space = spa_get_dspace(dp->dp_spa);
614 617 resv = spa_get_slop_space(dp->dp_spa);
615 618 if (netfree)
616 619 resv >>= 1;
617 620
618 621 return (space - resv);
619 622 }
620 623
621 624 boolean_t
622 625 dsl_pool_need_dirty_delay(dsl_pool_t *dp)
623 626 {
624 627 uint64_t delay_min_bytes =
625 628 zfs_dirty_data_max * zfs_delay_min_dirty_percent / 100;
626 629 boolean_t rv;
627 630
628 631 mutex_enter(&dp->dp_lock);
629 632 if (dp->dp_dirty_total > zfs_dirty_data_sync)
630 633 txg_kick(dp);
631 634 rv = (dp->dp_dirty_total > delay_min_bytes);
632 635 mutex_exit(&dp->dp_lock);
633 636 return (rv);
634 637 }
635 638
636 639 void
637 640 dsl_pool_dirty_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
638 641 {
639 642 if (space > 0) {
640 643 mutex_enter(&dp->dp_lock);
641 644 dp->dp_dirty_pertxg[tx->tx_txg & TXG_MASK] += space;
642 645 dsl_pool_dirty_delta(dp, space);
643 646 mutex_exit(&dp->dp_lock);
644 647 }
645 648 }
646 649
647 650 void
648 651 dsl_pool_undirty_space(dsl_pool_t *dp, int64_t space, uint64_t txg)
649 652 {
650 653 ASSERT3S(space, >=, 0);
651 654 if (space == 0)
652 655 return;
653 656 mutex_enter(&dp->dp_lock);
654 657 if (dp->dp_dirty_pertxg[txg & TXG_MASK] < space) {
655 658 /* XXX writing something we didn't dirty? */
656 659 space = dp->dp_dirty_pertxg[txg & TXG_MASK];
657 660 }
658 661 ASSERT3U(dp->dp_dirty_pertxg[txg & TXG_MASK], >=, space);
659 662 dp->dp_dirty_pertxg[txg & TXG_MASK] -= space;
660 663 ASSERT3U(dp->dp_dirty_total, >=, space);
661 664 dsl_pool_dirty_delta(dp, -space);
662 665 mutex_exit(&dp->dp_lock);
663 666 }
664 667
665 668 /* ARGSUSED */
666 669 static int
667 670 upgrade_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
668 671 {
669 672 dmu_tx_t *tx = arg;
670 673 dsl_dataset_t *ds, *prev = NULL;
671 674 int err;
672 675
673 676 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
674 677 if (err)
675 678 return (err);
676 679
677 680 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
678 681 err = dsl_dataset_hold_obj(dp,
679 682 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
680 683 if (err) {
681 684 dsl_dataset_rele(ds, FTAG);
682 685 return (err);
683 686 }
684 687
685 688 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object)
686 689 break;
687 690 dsl_dataset_rele(ds, FTAG);
688 691 ds = prev;
689 692 prev = NULL;
690 693 }
691 694
692 695 if (prev == NULL) {
693 696 prev = dp->dp_origin_snap;
694 697
695 698 /*
696 699 * The $ORIGIN can't have any data, or the accounting
697 700 * will be wrong.
698 701 */
699 702 ASSERT0(dsl_dataset_phys(prev)->ds_bp.blk_birth);
700 703
701 704 /* The origin doesn't get attached to itself */
702 705 if (ds->ds_object == prev->ds_object) {
703 706 dsl_dataset_rele(ds, FTAG);
704 707 return (0);
705 708 }
706 709
707 710 dmu_buf_will_dirty(ds->ds_dbuf, tx);
708 711 dsl_dataset_phys(ds)->ds_prev_snap_obj = prev->ds_object;
709 712 dsl_dataset_phys(ds)->ds_prev_snap_txg =
710 713 dsl_dataset_phys(prev)->ds_creation_txg;
711 714
712 715 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
713 716 dsl_dir_phys(ds->ds_dir)->dd_origin_obj = prev->ds_object;
714 717
715 718 dmu_buf_will_dirty(prev->ds_dbuf, tx);
716 719 dsl_dataset_phys(prev)->ds_num_children++;
717 720
718 721 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0) {
719 722 ASSERT(ds->ds_prev == NULL);
720 723 VERIFY0(dsl_dataset_hold_obj(dp,
721 724 dsl_dataset_phys(ds)->ds_prev_snap_obj,
722 725 ds, &ds->ds_prev));
723 726 }
724 727 }
725 728
726 729 ASSERT3U(dsl_dir_phys(ds->ds_dir)->dd_origin_obj, ==, prev->ds_object);
727 730 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_obj, ==, prev->ds_object);
728 731
729 732 if (dsl_dataset_phys(prev)->ds_next_clones_obj == 0) {
730 733 dmu_buf_will_dirty(prev->ds_dbuf, tx);
731 734 dsl_dataset_phys(prev)->ds_next_clones_obj =
732 735 zap_create(dp->dp_meta_objset,
733 736 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
734 737 }
735 738 VERIFY0(zap_add_int(dp->dp_meta_objset,
736 739 dsl_dataset_phys(prev)->ds_next_clones_obj, ds->ds_object, tx));
737 740
738 741 dsl_dataset_rele(ds, FTAG);
739 742 if (prev != dp->dp_origin_snap)
740 743 dsl_dataset_rele(prev, FTAG);
741 744 return (0);
742 745 }
743 746
744 747 void
745 748 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
746 749 {
747 750 ASSERT(dmu_tx_is_syncing(tx));
748 751 ASSERT(dp->dp_origin_snap != NULL);
749 752
750 753 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, upgrade_clones_cb,
751 754 tx, DS_FIND_CHILDREN));
752 755 }
753 756
754 757 /* ARGSUSED */
755 758 static int
756 759 upgrade_dir_clones_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
757 760 {
758 761 dmu_tx_t *tx = arg;
759 762 objset_t *mos = dp->dp_meta_objset;
760 763
761 764 if (dsl_dir_phys(ds->ds_dir)->dd_origin_obj != 0) {
762 765 dsl_dataset_t *origin;
763 766
764 767 VERIFY0(dsl_dataset_hold_obj(dp,
765 768 dsl_dir_phys(ds->ds_dir)->dd_origin_obj, FTAG, &origin));
766 769
767 770 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
768 771 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
769 772 dsl_dir_phys(origin->ds_dir)->dd_clones =
770 773 zap_create(mos, DMU_OT_DSL_CLONES, DMU_OT_NONE,
771 774 0, tx);
772 775 }
773 776
774 777 VERIFY0(zap_add_int(dp->dp_meta_objset,
775 778 dsl_dir_phys(origin->ds_dir)->dd_clones,
776 779 ds->ds_object, tx));
777 780
778 781 dsl_dataset_rele(origin, FTAG);
779 782 }
780 783 return (0);
781 784 }
782 785
783 786 void
784 787 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
785 788 {
786 789 ASSERT(dmu_tx_is_syncing(tx));
787 790 uint64_t obj;
788 791
789 792 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
790 793 VERIFY0(dsl_pool_open_special_dir(dp,
791 794 FREE_DIR_NAME, &dp->dp_free_dir));
792 795
793 796 /*
794 797 * We can't use bpobj_alloc(), because spa_version() still
795 798 * returns the old version, and we need a new-version bpobj with
796 799 * subobj support. So call dmu_object_alloc() directly.
797 800 */
798 801 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
799 802 SPA_OLD_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
800 803 VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
801 804 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
802 805 VERIFY0(bpobj_open(&dp->dp_free_bpobj, dp->dp_meta_objset, obj));
803 806
804 807 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
805 808 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
806 809 }
807 810
808 811 void
809 812 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
810 813 {
811 814 uint64_t dsobj;
812 815 dsl_dataset_t *ds;
813 816
814 817 ASSERT(dmu_tx_is_syncing(tx));
815 818 ASSERT(dp->dp_origin_snap == NULL);
816 819 ASSERT(rrw_held(&dp->dp_config_rwlock, RW_WRITER));
817 820
818 821 /* create the origin dir, ds, & snap-ds */
819 822 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
820 823 NULL, 0, kcred, tx);
821 824 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
822 825 dsl_dataset_snapshot_sync_impl(ds, ORIGIN_DIR_NAME, tx);
823 826 VERIFY0(dsl_dataset_hold_obj(dp, dsl_dataset_phys(ds)->ds_prev_snap_obj,
824 827 dp, &dp->dp_origin_snap));
825 828 dsl_dataset_rele(ds, FTAG);
826 829 }
827 830
828 831 taskq_t *
829 832 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
830 833 {
831 834 return (dp->dp_vnrele_taskq);
832 835 }
833 836
834 837 /*
835 838 * Walk through the pool-wide zap object of temporary snapshot user holds
836 839 * and release them.
837 840 */
838 841 void
839 842 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
840 843 {
841 844 zap_attribute_t za;
842 845 zap_cursor_t zc;
843 846 objset_t *mos = dp->dp_meta_objset;
844 847 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
845 848 nvlist_t *holds;
846 849
847 850 if (zapobj == 0)
848 851 return;
849 852 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
850 853
851 854 holds = fnvlist_alloc();
852 855
853 856 for (zap_cursor_init(&zc, mos, zapobj);
854 857 zap_cursor_retrieve(&zc, &za) == 0;
855 858 zap_cursor_advance(&zc)) {
856 859 char *htag;
857 860 nvlist_t *tags;
858 861
859 862 htag = strchr(za.za_name, '-');
860 863 *htag = '\0';
861 864 ++htag;
862 865 if (nvlist_lookup_nvlist(holds, za.za_name, &tags) != 0) {
863 866 tags = fnvlist_alloc();
864 867 fnvlist_add_boolean(tags, htag);
865 868 fnvlist_add_nvlist(holds, za.za_name, tags);
866 869 fnvlist_free(tags);
867 870 } else {
868 871 fnvlist_add_boolean(tags, htag);
869 872 }
870 873 }
871 874 dsl_dataset_user_release_tmp(dp, holds);
872 875 fnvlist_free(holds);
873 876 zap_cursor_fini(&zc);
874 877 }
875 878
876 879 /*
877 880 * Create the pool-wide zap object for storing temporary snapshot holds.
878 881 */
879 882 void
880 883 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
881 884 {
882 885 objset_t *mos = dp->dp_meta_objset;
883 886
884 887 ASSERT(dp->dp_tmp_userrefs_obj == 0);
885 888 ASSERT(dmu_tx_is_syncing(tx));
886 889
887 890 dp->dp_tmp_userrefs_obj = zap_create_link(mos, DMU_OT_USERREFS,
888 891 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS, tx);
889 892 }
890 893
891 894 static int
892 895 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
893 896 const char *tag, uint64_t now, dmu_tx_t *tx, boolean_t holding)
894 897 {
895 898 objset_t *mos = dp->dp_meta_objset;
896 899 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
897 900 char *name;
898 901 int error;
899 902
900 903 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
901 904 ASSERT(dmu_tx_is_syncing(tx));
902 905
903 906 /*
904 907 * If the pool was created prior to SPA_VERSION_USERREFS, the
905 908 * zap object for temporary holds might not exist yet.
906 909 */
907 910 if (zapobj == 0) {
908 911 if (holding) {
909 912 dsl_pool_user_hold_create_obj(dp, tx);
910 913 zapobj = dp->dp_tmp_userrefs_obj;
911 914 } else {
912 915 return (SET_ERROR(ENOENT));
913 916 }
914 917 }
915 918
916 919 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
917 920 if (holding)
918 921 error = zap_add(mos, zapobj, name, 8, 1, &now, tx);
919 922 else
920 923 error = zap_remove(mos, zapobj, name, tx);
921 924 strfree(name);
922 925
923 926 return (error);
924 927 }
925 928
926 929 /*
927 930 * Add a temporary hold for the given dataset object and tag.
928 931 */
929 932 int
930 933 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
931 934 uint64_t now, dmu_tx_t *tx)
932 935 {
933 936 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
934 937 }
935 938
936 939 /*
937 940 * Release a temporary hold for the given dataset object and tag.
938 941 */
939 942 int
940 943 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
941 944 dmu_tx_t *tx)
942 945 {
943 946 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,
944 947 tx, B_FALSE));
945 948 }
946 949
947 950 /*
948 951 * DSL Pool Configuration Lock
949 952 *
950 953 * The dp_config_rwlock protects against changes to DSL state (e.g. dataset
951 954 * creation / destruction / rename / property setting). It must be held for
952 955 * read to hold a dataset or dsl_dir. I.e. you must call
953 956 * dsl_pool_config_enter() or dsl_pool_hold() before calling
954 957 * dsl_{dataset,dir}_hold{_obj}. In most circumstances, the dp_config_rwlock
955 958 * must be held continuously until all datasets and dsl_dirs are released.
956 959 *
957 960 * The only exception to this rule is that if a "long hold" is placed on
958 961 * a dataset, then the dp_config_rwlock may be dropped while the dataset
959 962 * is still held. The long hold will prevent the dataset from being
960 963 * destroyed -- the destroy will fail with EBUSY. A long hold can be
961 964 * obtained by calling dsl_dataset_long_hold(), or by "owning" a dataset
962 965 * (by calling dsl_{dataset,objset}_{try}own{_obj}).
963 966 *
964 967 * Legitimate long-holders (including owners) should be long-running, cancelable
965 968 * tasks that should cause "zfs destroy" to fail. This includes DMU
966 969 * consumers (i.e. a ZPL filesystem being mounted or ZVOL being open),
967 970 * "zfs send", and "zfs diff". There are several other long-holders whose
968 971 * uses are suboptimal (e.g. "zfs promote", and zil_suspend()).
969 972 *
970 973 * The usual formula for long-holding would be:
971 974 * dsl_pool_hold()
972 975 * dsl_dataset_hold()
973 976 * ... perform checks ...
974 977 * dsl_dataset_long_hold()
975 978 * dsl_pool_rele()
976 979 * ... perform long-running task ...
977 980 * dsl_dataset_long_rele()
978 981 * dsl_dataset_rele()
979 982 *
980 983 * Note that when the long hold is released, the dataset is still held but
981 984 * the pool is not held. The dataset may change arbitrarily during this time
982 985 * (e.g. it could be destroyed). Therefore you shouldn't do anything to the
983 986 * dataset except release it.
984 987 *
985 988 * User-initiated operations (e.g. ioctls, zfs_ioc_*()) are either read-only
986 989 * or modifying operations.
987 990 *
988 991 * Modifying operations should generally use dsl_sync_task(). The synctask
989 992 * infrastructure enforces proper locking strategy with respect to the
990 993 * dp_config_rwlock. See the comment above dsl_sync_task() for details.
991 994 *
992 995 * Read-only operations will manually hold the pool, then the dataset, obtain
993 996 * information from the dataset, then release the pool and dataset.
994 997 * dmu_objset_{hold,rele}() are convenience routines that also do the pool
995 998 * hold/rele.
996 999 */
997 1000
998 1001 int
999 1002 dsl_pool_hold(const char *name, void *tag, dsl_pool_t **dp)
1000 1003 {
1001 1004 spa_t *spa;
1002 1005 int error;
1003 1006
1004 1007 error = spa_open(name, &spa, tag);
1005 1008 if (error == 0) {
1006 1009 *dp = spa_get_dsl(spa);
1007 1010 dsl_pool_config_enter(*dp, tag);
1008 1011 }
1009 1012 return (error);
1010 1013 }
1011 1014
1012 1015 void
1013 1016 dsl_pool_rele(dsl_pool_t *dp, void *tag)
1014 1017 {
1015 1018 dsl_pool_config_exit(dp, tag);
1016 1019 spa_close(dp->dp_spa, tag);
1017 1020 }
1018 1021
1019 1022 void
1020 1023 dsl_pool_config_enter(dsl_pool_t *dp, void *tag)
1021 1024 {
1022 1025 /*
1023 1026 * We use a "reentrant" reader-writer lock, but not reentrantly.
1024 1027 *
1025 1028 * The rrwlock can (with the track_all flag) track all reading threads,
1026 1029 * which is very useful for debugging which code path failed to release
1027 1030 * the lock, and for verifying that the *current* thread does hold
1028 1031 * the lock.
1029 1032 *
1030 1033 * (Unlike a rwlock, which knows that N threads hold it for
1031 1034 * read, but not *which* threads, so rw_held(RW_READER) returns TRUE
1032 1035 * if any thread holds it for read, even if this thread doesn't).
1033 1036 */
1034 1037 ASSERT(!rrw_held(&dp->dp_config_rwlock, RW_READER));
1035 1038 rrw_enter(&dp->dp_config_rwlock, RW_READER, tag);
1036 1039 }
1037 1040
1038 1041 void
1039 1042 dsl_pool_config_exit(dsl_pool_t *dp, void *tag)
1040 1043 {
1041 1044 rrw_exit(&dp->dp_config_rwlock, tag);
1042 1045 }
1043 1046
1044 1047 boolean_t
1045 1048 dsl_pool_config_held(dsl_pool_t *dp)
1046 1049 {
1047 1050 return (RRW_LOCK_HELD(&dp->dp_config_rwlock));
1048 1051 }
|
↓ open down ↓ |
719 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX