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 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
24 */
25
26 #include <sys/dsl_pool.h>
27 #include <sys/dsl_dataset.h>
28 #include <sys/dsl_prop.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_synctask.h>
31 #include <sys/dsl_scan.h>
32 #include <sys/dnode.h>
33 #include <sys/dmu_tx.h>
34 #include <sys/dmu_objset.h>
35 #include <sys/arc.h>
36 #include <sys/zap.h>
37 #include <sys/zio.h>
38 #include <sys/zfs_context.h>
39 #include <sys/fs/zfs.h>
40 #include <sys/zfs_znode.h>
41 #include <sys/spa_impl.h>
42 #include <sys/dsl_deadlist.h>
43
44 int zfs_no_write_throttle = 0;
45 int zfs_write_limit_shift = 3; /* 1/8th of physical memory */
46 int zfs_txg_synctime_ms = 1000; /* target millisecs to sync a txg */
47
48 uint64_t zfs_write_limit_min = 32 << 20; /* min write limit is 32MB */
49 uint64_t zfs_write_limit_max = 0; /* max data payload per txg */
50 uint64_t zfs_write_limit_inflated = 0;
51 uint64_t zfs_write_limit_override = 0;
52
53 kmutex_t zfs_write_limit_lock;
54
55 static pgcnt_t old_physmem = 0;
56
57 int
58 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
59 {
60 uint64_t obj;
61 int err;
62
63 err = zap_lookup(dp->dp_meta_objset,
64 dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
65 name, sizeof (obj), 1, &obj);
66 if (err)
67 return (err);
68
69 return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
70 }
71
72 static dsl_pool_t *
73 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
74 {
75 dsl_pool_t *dp;
76 blkptr_t *bp = spa_get_rootblkptr(spa);
77
78 dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
79 dp->dp_spa = spa;
80 dp->dp_meta_rootbp = *bp;
81 rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
82 dp->dp_write_limit = zfs_write_limit_min;
83 txg_init(dp, txg);
84
85 txg_list_create(&dp->dp_dirty_datasets,
86 offsetof(dsl_dataset_t, ds_dirty_link));
87 txg_list_create(&dp->dp_dirty_dirs,
88 offsetof(dsl_dir_t, dd_dirty_link));
89 txg_list_create(&dp->dp_sync_tasks,
90 offsetof(dsl_sync_task_group_t, dstg_node));
91 list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
92 offsetof(dsl_dataset_t, ds_synced_link));
93
94 mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
95
96 dp->dp_vnrele_taskq = taskq_create("zfs_vn_rele_taskq", 1, minclsyspri,
97 1, 4, 0);
98
99 return (dp);
100 }
101
102 int
103 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
104 {
105 int err;
106 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
107 dsl_dir_t *dd;
108 dsl_dataset_t *ds;
109 uint64_t obj;
110
111 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
112 err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp,
113 &dp->dp_meta_objset);
114 if (err)
115 goto out;
116
117 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
118 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
119 &dp->dp_root_dir_obj);
120 if (err)
121 goto out;
122
123 err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
124 NULL, dp, &dp->dp_root_dir);
125 if (err)
126 goto out;
127
128 err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
129 if (err)
130 goto out;
131
132 if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
133 err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
134 if (err)
135 goto out;
136 err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
137 FTAG, &ds);
138 if (err == 0) {
139 err = dsl_dataset_hold_obj(dp,
140 ds->ds_phys->ds_prev_snap_obj, dp,
141 &dp->dp_origin_snap);
142 dsl_dataset_rele(ds, FTAG);
143 }
144 dsl_dir_close(dd, dp);
145 if (err)
146 goto out;
147 }
148
149 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
150 err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
151 &dp->dp_free_dir);
152 if (err)
153 goto out;
154
155 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
156 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
157 if (err)
158 goto out;
159 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
160 dp->dp_meta_objset, obj));
161 }
162
163 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
164 DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
165 &dp->dp_tmp_userrefs_obj);
166 if (err == ENOENT)
167 err = 0;
168 if (err)
169 goto out;
170
171 err = dsl_scan_init(dp, txg);
172
173 out:
174 rw_exit(&dp->dp_config_rwlock);
175 if (err)
176 dsl_pool_close(dp);
177 else
178 *dpp = dp;
179
180 return (err);
181 }
182
183 void
184 dsl_pool_close(dsl_pool_t *dp)
185 {
186 /* drop our references from dsl_pool_open() */
187
188 /*
189 * Since we held the origin_snap from "syncing" context (which
190 * includes pool-opening context), it actually only got a "ref"
191 * and not a hold, so just drop that here.
192 */
193 if (dp->dp_origin_snap)
194 dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
195 if (dp->dp_mos_dir)
196 dsl_dir_close(dp->dp_mos_dir, dp);
197 if (dp->dp_free_dir)
198 dsl_dir_close(dp->dp_free_dir, dp);
199 if (dp->dp_root_dir)
200 dsl_dir_close(dp->dp_root_dir, dp);
201
202 bpobj_close(&dp->dp_free_bpobj);
203
204 /* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
205 if (dp->dp_meta_objset)
206 dmu_objset_evict(dp->dp_meta_objset);
207
208 txg_list_destroy(&dp->dp_dirty_datasets);
209 txg_list_destroy(&dp->dp_sync_tasks);
210 txg_list_destroy(&dp->dp_dirty_dirs);
211 list_destroy(&dp->dp_synced_datasets);
212
213 arc_flush(dp->dp_spa);
214 txg_fini(dp);
215 dsl_scan_fini(dp);
216 rw_destroy(&dp->dp_config_rwlock);
217 mutex_destroy(&dp->dp_lock);
218 taskq_destroy(dp->dp_vnrele_taskq);
219 if (dp->dp_blkstats)
220 kmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
221 kmem_free(dp, sizeof (dsl_pool_t));
222 }
223
224 dsl_pool_t *
225 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
226 {
227 int err;
228 dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
229 dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
230 objset_t *os;
231 dsl_dataset_t *ds;
232 uint64_t obj;
233
234 /* create and open the MOS (meta-objset) */
235 dp->dp_meta_objset = dmu_objset_create_impl(spa,
236 NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
237
238 /* create the pool directory */
239 err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
240 DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
241 ASSERT3U(err, ==, 0);
242
243 /* Initialize scan structures */
244 VERIFY3U(0, ==, dsl_scan_init(dp, txg));
245
246 /* create and open the root dir */
247 dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
248 VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
249 NULL, dp, &dp->dp_root_dir));
250
251 /* create and open the meta-objset dir */
252 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
253 VERIFY(0 == dsl_pool_open_special_dir(dp,
254 MOS_DIR_NAME, &dp->dp_mos_dir));
255
256 if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
257 /* create and open the free dir */
258 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
259 FREE_DIR_NAME, tx);
260 VERIFY(0 == dsl_pool_open_special_dir(dp,
261 FREE_DIR_NAME, &dp->dp_free_dir));
262
263 /* create and open the free_bplist */
264 obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
265 VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
266 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
267 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
268 dp->dp_meta_objset, obj));
269 }
270
271 if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
272 dsl_pool_create_origin(dp, tx);
273
274 /* create the root dataset */
275 obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
276
277 /* create the root objset */
278 VERIFY(0 == dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
279 os = dmu_objset_create_impl(dp->dp_spa, ds,
280 dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
281 #ifdef _KERNEL
282 zfs_create_fs(os, kcred, zplprops, tx);
283 #endif
284 dsl_dataset_rele(ds, FTAG);
285
286 dmu_tx_commit(tx);
287
288 return (dp);
289 }
290
291 static int
292 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
293 {
294 dsl_deadlist_t *dl = arg;
295 dsl_pool_t *dp = dmu_objset_pool(dl->dl_os);
296 rw_enter(&dp->dp_config_rwlock, RW_READER);
297 dsl_deadlist_insert(dl, bp, tx);
298 rw_exit(&dp->dp_config_rwlock);
299 return (0);
300 }
301
302 void
303 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
304 {
305 zio_t *zio;
306 dmu_tx_t *tx;
307 dsl_dir_t *dd;
308 dsl_dataset_t *ds;
309 dsl_sync_task_group_t *dstg;
310 objset_t *mos = dp->dp_meta_objset;
311 hrtime_t start, write_time;
312 uint64_t data_written;
313 int err;
314
315 /*
316 * We need to copy dp_space_towrite() before doing
317 * dsl_sync_task_group_sync(), because
318 * dsl_dataset_snapshot_reserve_space() will increase
319 * dp_space_towrite but not actually write anything.
320 */
321 data_written = dp->dp_space_towrite[txg & TXG_MASK];
322
323 tx = dmu_tx_create_assigned(dp, txg);
324
325 dp->dp_read_overhead = 0;
326 start = gethrtime();
327
328 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
329 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
330 /*
331 * We must not sync any non-MOS datasets twice, because
332 * we may have taken a snapshot of them. However, we
333 * may sync newly-created datasets on pass 2.
334 */
335 ASSERT(!list_link_active(&ds->ds_synced_link));
336 list_insert_tail(&dp->dp_synced_datasets, ds);
337 dsl_dataset_sync(ds, zio, tx);
338 }
339 DTRACE_PROBE(pool_sync__1setup);
340 err = zio_wait(zio);
341
342 write_time = gethrtime() - start;
343 ASSERT(err == 0);
344 DTRACE_PROBE(pool_sync__2rootzio);
345
346 for (ds = list_head(&dp->dp_synced_datasets); ds;
347 ds = list_next(&dp->dp_synced_datasets, ds))
348 dmu_objset_do_userquota_updates(ds->ds_objset, tx);
349
350 /*
351 * Sync the datasets again to push out the changes due to
352 * userspace updates. This must be done before we process the
353 * sync tasks, because that could cause a snapshot of a dataset
354 * whose ds_bp will be rewritten when we do this 2nd sync.
355 */
356 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
357 while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
358 ASSERT(list_link_active(&ds->ds_synced_link));
359 dmu_buf_rele(ds->ds_dbuf, ds);
360 dsl_dataset_sync(ds, zio, tx);
361 }
362 err = zio_wait(zio);
363
364 /*
365 * Move dead blocks from the pending deadlist to the on-disk
366 * deadlist.
367 */
368 for (ds = list_head(&dp->dp_synced_datasets); ds;
369 ds = list_next(&dp->dp_synced_datasets, ds)) {
370 bplist_iterate(&ds->ds_pending_deadlist,
371 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
372 }
373
374 while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg)) {
375 /*
376 * No more sync tasks should have been added while we
377 * were syncing.
378 */
379 ASSERT(spa_sync_pass(dp->dp_spa) == 1);
380 dsl_sync_task_group_sync(dstg, tx);
381 }
382 DTRACE_PROBE(pool_sync__3task);
383
384 start = gethrtime();
385 while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
386 dsl_dir_sync(dd, tx);
387 write_time += gethrtime() - start;
388
389 start = gethrtime();
390 if (list_head(&mos->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
391 list_head(&mos->os_free_dnodes[txg & TXG_MASK]) != NULL) {
392 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
393 dmu_objset_sync(mos, zio, tx);
394 err = zio_wait(zio);
395 ASSERT(err == 0);
396 dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
397 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
398 }
399 write_time += gethrtime() - start;
400 DTRACE_PROBE2(pool_sync__4io, hrtime_t, write_time,
401 hrtime_t, dp->dp_read_overhead);
402 write_time -= dp->dp_read_overhead;
403
404 dmu_tx_commit(tx);
405
406 dp->dp_space_towrite[txg & TXG_MASK] = 0;
407 ASSERT(dp->dp_tempreserved[txg & TXG_MASK] == 0);
408
409 /*
410 * If the write limit max has not been explicitly set, set it
411 * to a fraction of available physical memory (default 1/8th).
412 * Note that we must inflate the limit because the spa
413 * inflates write sizes to account for data replication.
414 * Check this each sync phase to catch changing memory size.
415 */
416 if (physmem != old_physmem && zfs_write_limit_shift) {
417 mutex_enter(&zfs_write_limit_lock);
418 old_physmem = physmem;
419 zfs_write_limit_max = ptob(physmem) >> zfs_write_limit_shift;
420 zfs_write_limit_inflated = MAX(zfs_write_limit_min,
421 spa_get_asize(dp->dp_spa, zfs_write_limit_max));
422 mutex_exit(&zfs_write_limit_lock);
423 }
424
425 /*
426 * Attempt to keep the sync time consistent by adjusting the
427 * amount of write traffic allowed into each transaction group.
428 * Weight the throughput calculation towards the current value:
429 * thru = 3/4 old_thru + 1/4 new_thru
430 *
431 * Note: write_time is in nanosecs, so write_time/MICROSEC
432 * yields millisecs
433 */
434 ASSERT(zfs_write_limit_min > 0);
435 if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
436 uint64_t throughput = data_written / (write_time / MICROSEC);
437
438 if (dp->dp_throughput)
439 dp->dp_throughput = throughput / 4 +
440 3 * dp->dp_throughput / 4;
441 else
442 dp->dp_throughput = throughput;
443 dp->dp_write_limit = MIN(zfs_write_limit_inflated,
444 MAX(zfs_write_limit_min,
445 dp->dp_throughput * zfs_txg_synctime_ms));
446 }
447 }
448
449 void
450 dsl_pool_sync_done(dsl_pool_t *dp, uint64_t txg)
451 {
452 dsl_dataset_t *ds;
453 objset_t *os;
454
455 while (ds = list_head(&dp->dp_synced_datasets)) {
456 list_remove(&dp->dp_synced_datasets, ds);
457 os = ds->ds_objset;
458 zil_clean(os->os_zil, txg);
459 ASSERT(!dmu_objset_is_dirty(os, txg));
460 dmu_buf_rele(ds->ds_dbuf, ds);
461 }
462 ASSERT(!dmu_objset_is_dirty(dp->dp_meta_objset, txg));
463 }
464
465 /*
466 * TRUE if the current thread is the tx_sync_thread or if we
467 * are being called from SPA context during pool initialization.
468 */
469 int
470 dsl_pool_sync_context(dsl_pool_t *dp)
471 {
472 return (curthread == dp->dp_tx.tx_sync_thread ||
473 spa_get_dsl(dp->dp_spa) == NULL);
474 }
475
476 uint64_t
477 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
478 {
479 uint64_t space, resv;
480
481 /*
482 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
483 * efficiency.
484 * XXX The intent log is not accounted for, so it must fit
485 * within this slop.
486 *
487 * If we're trying to assess whether it's OK to do a free,
488 * cut the reservation in half to allow forward progress
489 * (e.g. make it possible to rm(1) files from a full pool).
490 */
491 space = spa_get_dspace(dp->dp_spa);
492 resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
493 if (netfree)
494 resv >>= 1;
495
496 return (space - resv);
497 }
498
499 int
500 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
501 {
502 uint64_t reserved = 0;
503 uint64_t write_limit = (zfs_write_limit_override ?
504 zfs_write_limit_override : dp->dp_write_limit);
505
506 if (zfs_no_write_throttle) {
507 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
508 space);
509 return (0);
510 }
511
512 /*
513 * Check to see if we have exceeded the maximum allowed IO for
514 * this transaction group. We can do this without locks since
515 * a little slop here is ok. Note that we do the reserved check
516 * with only half the requested reserve: this is because the
517 * reserve requests are worst-case, and we really don't want to
518 * throttle based off of worst-case estimates.
519 */
520 if (write_limit > 0) {
521 reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
522 + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
523
524 if (reserved && reserved > write_limit)
525 return (ERESTART);
526 }
527
528 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
529
530 /*
531 * If this transaction group is over 7/8ths capacity, delay
532 * the caller 1 clock tick. This will slow down the "fill"
533 * rate until the sync process can catch up with us.
534 */
535 if (reserved && reserved > (write_limit - (write_limit >> 3)))
536 txg_delay(dp, tx->tx_txg, 1);
537
538 return (0);
539 }
540
541 void
542 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
543 {
544 ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
545 atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
546 }
547
548 void
549 dsl_pool_memory_pressure(dsl_pool_t *dp)
550 {
551 uint64_t space_inuse = 0;
552 int i;
553
554 if (dp->dp_write_limit == zfs_write_limit_min)
555 return;
556
557 for (i = 0; i < TXG_SIZE; i++) {
558 space_inuse += dp->dp_space_towrite[i];
559 space_inuse += dp->dp_tempreserved[i];
560 }
561 dp->dp_write_limit = MAX(zfs_write_limit_min,
562 MIN(dp->dp_write_limit, space_inuse / 4));
563 }
564
565 void
566 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
567 {
568 if (space > 0) {
569 mutex_enter(&dp->dp_lock);
570 dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
571 mutex_exit(&dp->dp_lock);
572 }
573 }
574
575 /* ARGSUSED */
576 static int
577 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
578 {
579 dmu_tx_t *tx = arg;
580 dsl_dataset_t *ds, *prev = NULL;
581 int err;
582 dsl_pool_t *dp = spa_get_dsl(spa);
583
584 err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
585 if (err)
586 return (err);
587
588 while (ds->ds_phys->ds_prev_snap_obj != 0) {
589 err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
590 FTAG, &prev);
591 if (err) {
592 dsl_dataset_rele(ds, FTAG);
593 return (err);
594 }
595
596 if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
597 break;
598 dsl_dataset_rele(ds, FTAG);
599 ds = prev;
600 prev = NULL;
601 }
602
603 if (prev == NULL) {
604 prev = dp->dp_origin_snap;
605
606 /*
607 * The $ORIGIN can't have any data, or the accounting
608 * will be wrong.
609 */
610 ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
611
612 /* The origin doesn't get attached to itself */
613 if (ds->ds_object == prev->ds_object) {
614 dsl_dataset_rele(ds, FTAG);
615 return (0);
616 }
617
618 dmu_buf_will_dirty(ds->ds_dbuf, tx);
619 ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
620 ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
621
622 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
623 ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
624
625 dmu_buf_will_dirty(prev->ds_dbuf, tx);
626 prev->ds_phys->ds_num_children++;
627
628 if (ds->ds_phys->ds_next_snap_obj == 0) {
629 ASSERT(ds->ds_prev == NULL);
630 VERIFY(0 == dsl_dataset_hold_obj(dp,
631 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
632 }
633 }
634
635 ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
636 ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
637
638 if (prev->ds_phys->ds_next_clones_obj == 0) {
639 dmu_buf_will_dirty(prev->ds_dbuf, tx);
640 prev->ds_phys->ds_next_clones_obj =
641 zap_create(dp->dp_meta_objset,
642 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
643 }
644 VERIFY(0 == zap_add_int(dp->dp_meta_objset,
645 prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
646
647 dsl_dataset_rele(ds, FTAG);
648 if (prev != dp->dp_origin_snap)
649 dsl_dataset_rele(prev, FTAG);
650 return (0);
651 }
652
653 void
654 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
655 {
656 ASSERT(dmu_tx_is_syncing(tx));
657 ASSERT(dp->dp_origin_snap != NULL);
658
659 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
660 tx, DS_FIND_CHILDREN));
661 }
662
663 /* ARGSUSED */
664 static int
665 upgrade_dir_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
666 {
667 dmu_tx_t *tx = arg;
668 dsl_dataset_t *ds;
669 dsl_pool_t *dp = spa_get_dsl(spa);
670 objset_t *mos = dp->dp_meta_objset;
671
672 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
673
674 if (ds->ds_dir->dd_phys->dd_origin_obj) {
675 dsl_dataset_t *origin;
676
677 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
678 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin));
679
680 if (origin->ds_dir->dd_phys->dd_clones == 0) {
681 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
682 origin->ds_dir->dd_phys->dd_clones = zap_create(mos,
683 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
684 }
685
686 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
687 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
688
689 dsl_dataset_rele(origin, FTAG);
690 }
691
692 dsl_dataset_rele(ds, FTAG);
693 return (0);
694 }
695
696 void
697 dsl_pool_upgrade_dir_clones(dsl_pool_t *dp, dmu_tx_t *tx)
698 {
699 ASSERT(dmu_tx_is_syncing(tx));
700 uint64_t obj;
701
702 (void) dsl_dir_create_sync(dp, dp->dp_root_dir, FREE_DIR_NAME, tx);
703 VERIFY(0 == dsl_pool_open_special_dir(dp,
704 FREE_DIR_NAME, &dp->dp_free_dir));
705
706 /*
707 * We can't use bpobj_alloc(), because spa_version() still
708 * returns the old version, and we need a new-version bpobj with
709 * subobj support. So call dmu_object_alloc() directly.
710 */
711 obj = dmu_object_alloc(dp->dp_meta_objset, DMU_OT_BPOBJ,
712 SPA_MAXBLOCKSIZE, DMU_OT_BPOBJ_HDR, sizeof (bpobj_phys_t), tx);
713 VERIFY3U(0, ==, zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
714 DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
715 VERIFY3U(0, ==, bpobj_open(&dp->dp_free_bpobj,
716 dp->dp_meta_objset, obj));
717
718 VERIFY3U(0, ==, dmu_objset_find_spa(dp->dp_spa, NULL,
719 upgrade_dir_clones_cb, tx, DS_FIND_CHILDREN));
720 }
721
722 void
723 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
724 {
725 uint64_t dsobj;
726 dsl_dataset_t *ds;
727
728 ASSERT(dmu_tx_is_syncing(tx));
729 ASSERT(dp->dp_origin_snap == NULL);
730
731 /* create the origin dir, ds, & snap-ds */
732 rw_enter(&dp->dp_config_rwlock, RW_WRITER);
733 dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
734 NULL, 0, kcred, tx);
735 VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
736 dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, tx);
737 VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
738 dp, &dp->dp_origin_snap));
739 dsl_dataset_rele(ds, FTAG);
740 rw_exit(&dp->dp_config_rwlock);
741 }
742
743 taskq_t *
744 dsl_pool_vnrele_taskq(dsl_pool_t *dp)
745 {
746 return (dp->dp_vnrele_taskq);
747 }
748
749 /*
750 * Walk through the pool-wide zap object of temporary snapshot user holds
751 * and release them.
752 */
753 void
754 dsl_pool_clean_tmp_userrefs(dsl_pool_t *dp)
755 {
756 zap_attribute_t za;
757 zap_cursor_t zc;
758 objset_t *mos = dp->dp_meta_objset;
759 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
760
761 if (zapobj == 0)
762 return;
763 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
764
765 for (zap_cursor_init(&zc, mos, zapobj);
766 zap_cursor_retrieve(&zc, &za) == 0;
767 zap_cursor_advance(&zc)) {
768 char *htag;
769 uint64_t dsobj;
770
771 htag = strchr(za.za_name, '-');
772 *htag = '\0';
773 ++htag;
774 dsobj = strtonum(za.za_name, NULL);
775 (void) dsl_dataset_user_release_tmp(dp, dsobj, htag, B_FALSE);
776 }
777 zap_cursor_fini(&zc);
778 }
779
780 /*
781 * Create the pool-wide zap object for storing temporary snapshot holds.
782 */
783 void
784 dsl_pool_user_hold_create_obj(dsl_pool_t *dp, dmu_tx_t *tx)
785 {
786 objset_t *mos = dp->dp_meta_objset;
787
788 ASSERT(dp->dp_tmp_userrefs_obj == 0);
789 ASSERT(dmu_tx_is_syncing(tx));
790
791 dp->dp_tmp_userrefs_obj = zap_create(mos, DMU_OT_USERREFS,
792 DMU_OT_NONE, 0, tx);
793
794 VERIFY(zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_TMP_USERREFS,
795 sizeof (uint64_t), 1, &dp->dp_tmp_userrefs_obj, tx) == 0);
796 }
797
798 static int
799 dsl_pool_user_hold_rele_impl(dsl_pool_t *dp, uint64_t dsobj,
800 const char *tag, uint64_t *now, dmu_tx_t *tx, boolean_t holding)
801 {
802 objset_t *mos = dp->dp_meta_objset;
803 uint64_t zapobj = dp->dp_tmp_userrefs_obj;
804 char *name;
805 int error;
806
807 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
808 ASSERT(dmu_tx_is_syncing(tx));
809
810 /*
811 * If the pool was created prior to SPA_VERSION_USERREFS, the
812 * zap object for temporary holds might not exist yet.
813 */
814 if (zapobj == 0) {
815 if (holding) {
816 dsl_pool_user_hold_create_obj(dp, tx);
817 zapobj = dp->dp_tmp_userrefs_obj;
818 } else {
819 return (ENOENT);
820 }
821 }
822
823 name = kmem_asprintf("%llx-%s", (u_longlong_t)dsobj, tag);
824 if (holding)
825 error = zap_add(mos, zapobj, name, 8, 1, now, tx);
826 else
827 error = zap_remove(mos, zapobj, name, tx);
828 strfree(name);
829
830 return (error);
831 }
832
833 /*
834 * Add a temporary hold for the given dataset object and tag.
835 */
836 int
837 dsl_pool_user_hold(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
838 uint64_t *now, dmu_tx_t *tx)
839 {
840 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, now, tx, B_TRUE));
841 }
842
843 /*
844 * Release a temporary hold for the given dataset object and tag.
845 */
846 int
847 dsl_pool_user_release(dsl_pool_t *dp, uint64_t dsobj, const char *tag,
848 dmu_tx_t *tx)
849 {
850 return (dsl_pool_user_hold_rele_impl(dp, dsobj, tag, NULL,
851 tx, B_FALSE));
852 }