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.
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.
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]
22 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2016 Gary Mills
24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
25 * Copyright 2017 Joyent, Inc.
26 * Copyright (c) 2017 Datto Inc.
29 #include <sys/dsl_scan.h>
30 #include <sys/dsl_pool.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dnode.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/dmu_objset.h>
41 #include <sys/zfs_context.h>
42 #include <sys/fs/zfs.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/spa_impl.h>
45 #include <sys/vdev_impl.h>
46 #include <sys/zil_impl.h>
47 #include <sys/zio_checksum.h>
50 #include <sys/sa_impl.h>
51 #include <sys/zfeature.h>
54 #include <sys/zfs_vfsops.h>
57 typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *,
58 const zbookmark_phys_t *);
60 static scan_cb_t dsl_scan_scrub_cb;
61 static void dsl_scan_cancel_sync(void *, dmu_tx_t *);
62 static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *);
63 static boolean_t dsl_scan_restarting(dsl_scan_t *, dmu_tx_t *);
65 unsigned int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */
66 unsigned int zfs_resilver_delay = 2; /* number of ticks to delay resilver */
67 unsigned int zfs_scrub_delay = 4; /* number of ticks to delay scrub */
68 unsigned int zfs_scan_idle = 50; /* idle window in clock ticks */
70 unsigned int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */
71 unsigned int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */
72 unsigned int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver
74 boolean_t zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
75 boolean_t zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
77 SYSCTL_DECL(_vfs_zfs);
78 SYSCTL_UINT(_vfs_zfs, OID_AUTO, top_maxinflight, CTLFLAG_RWTUN,
79 &zfs_top_maxinflight, 0, "Maximum I/Os per top-level vdev");
80 SYSCTL_UINT(_vfs_zfs, OID_AUTO, resilver_delay, CTLFLAG_RWTUN,
81 &zfs_resilver_delay, 0, "Number of ticks to delay resilver");
82 SYSCTL_UINT(_vfs_zfs, OID_AUTO, scrub_delay, CTLFLAG_RWTUN,
83 &zfs_scrub_delay, 0, "Number of ticks to delay scrub");
84 SYSCTL_UINT(_vfs_zfs, OID_AUTO, scan_idle, CTLFLAG_RWTUN,
85 &zfs_scan_idle, 0, "Idle scan window in clock ticks");
86 SYSCTL_UINT(_vfs_zfs, OID_AUTO, scan_min_time_ms, CTLFLAG_RWTUN,
87 &zfs_scan_min_time_ms, 0, "Min millisecs to scrub per txg");
88 SYSCTL_UINT(_vfs_zfs, OID_AUTO, free_min_time_ms, CTLFLAG_RWTUN,
89 &zfs_free_min_time_ms, 0, "Min millisecs to free per txg");
90 SYSCTL_UINT(_vfs_zfs, OID_AUTO, resilver_min_time_ms, CTLFLAG_RWTUN,
91 &zfs_resilver_min_time_ms, 0, "Min millisecs to resilver per txg");
92 SYSCTL_INT(_vfs_zfs, OID_AUTO, no_scrub_io, CTLFLAG_RWTUN,
93 &zfs_no_scrub_io, 0, "Disable scrub I/O");
94 SYSCTL_INT(_vfs_zfs, OID_AUTO, no_scrub_prefetch, CTLFLAG_RWTUN,
95 &zfs_no_scrub_prefetch, 0, "Disable scrub prefetching");
97 enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
98 /* max number of blocks to free in a single TXG */
99 uint64_t zfs_free_max_blocks = UINT64_MAX;
100 SYSCTL_UQUAD(_vfs_zfs, OID_AUTO, free_max_blocks, CTLFLAG_RWTUN,
101 &zfs_free_max_blocks, 0, "Maximum number of blocks to free in one TXG");
104 #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
105 ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
106 (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
108 extern int zfs_txg_timeout;
111 * Enable/disable the processing of the free_bpobj object.
113 boolean_t zfs_free_bpobj_enabled = B_TRUE;
115 SYSCTL_INT(_vfs_zfs, OID_AUTO, free_bpobj_enabled, CTLFLAG_RWTUN,
116 &zfs_free_bpobj_enabled, 0, "Enable free_bpobj processing");
118 /* the order has to match pool_scan_type */
119 static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
121 dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */
122 dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */
126 dsl_scan_init(dsl_pool_t *dp, uint64_t txg)
130 spa_t *spa = dp->dp_spa;
133 scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP);
137 * It's possible that we're resuming a scan after a reboot so
138 * make sure that the scan_async_destroying flag is initialized
141 ASSERT(!scn->scn_async_destroying);
142 scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa,
143 SPA_FEATURE_ASYNC_DESTROY);
145 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
146 "scrub_func", sizeof (uint64_t), 1, &f);
149 * There was an old-style scrub in progress. Restart a
150 * new-style scrub from the beginning.
152 scn->scn_restart_txg = txg;
153 zfs_dbgmsg("old-style scrub was in progress; "
154 "restarting new-style scrub in txg %llu",
155 scn->scn_restart_txg);
158 * Load the queue obj from the old location so that it
159 * can be freed by dsl_scan_done().
161 (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
162 "scrub_queue", sizeof (uint64_t), 1,
163 &scn->scn_phys.scn_queue_obj);
165 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
166 DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
173 if (scn->scn_phys.scn_state == DSS_SCANNING &&
174 spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) {
176 * A new-type scrub was in progress on an old
177 * pool, and the pool was accessed by old
178 * software. Restart from the beginning, since
179 * the old software may have changed the pool in
182 scn->scn_restart_txg = txg;
183 zfs_dbgmsg("new-style scrub was modified "
184 "by old software; restarting in txg %llu",
185 scn->scn_restart_txg);
189 spa_scan_stat_init(spa);
194 dsl_scan_fini(dsl_pool_t *dp)
197 kmem_free(dp->dp_scan, sizeof (dsl_scan_t));
204 dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
206 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
208 if (scn->scn_phys.scn_state == DSS_SCANNING)
209 return (SET_ERROR(EBUSY));
215 dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
217 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
218 pool_scan_func_t *funcp = arg;
219 dmu_object_type_t ot = 0;
220 dsl_pool_t *dp = scn->scn_dp;
221 spa_t *spa = dp->dp_spa;
223 ASSERT(scn->scn_phys.scn_state != DSS_SCANNING);
224 ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
225 bzero(&scn->scn_phys, sizeof (scn->scn_phys));
226 scn->scn_phys.scn_func = *funcp;
227 scn->scn_phys.scn_state = DSS_SCANNING;
228 scn->scn_phys.scn_min_txg = 0;
229 scn->scn_phys.scn_max_txg = tx->tx_txg;
230 scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
231 scn->scn_phys.scn_start_time = gethrestime_sec();
232 scn->scn_phys.scn_errors = 0;
233 scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
234 scn->scn_restart_txg = 0;
235 scn->scn_done_txg = 0;
236 spa_scan_stat_init(spa);
238 if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
239 scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
241 /* rewrite all disk labels */
242 vdev_config_dirty(spa->spa_root_vdev);
244 if (vdev_resilver_needed(spa->spa_root_vdev,
245 &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
246 spa_event_notify(spa, NULL, NULL,
247 ESC_ZFS_RESILVER_START);
249 spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_START);
252 spa->spa_scrub_started = B_TRUE;
254 * If this is an incremental scrub, limit the DDT scrub phase
255 * to just the auto-ditto class (for correctness); the rest
256 * of the scrub should go faster using top-down pruning.
258 if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
259 scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
263 /* back to the generic stuff */
265 if (dp->dp_blkstats == NULL) {
267 kmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
269 bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
271 if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
272 ot = DMU_OT_ZAP_OTHER;
274 scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
275 ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
277 dsl_scan_sync_state(scn, tx);
279 spa_history_log_internal(spa, "scan setup", tx,
280 "func=%u mintxg=%llu maxtxg=%llu",
281 *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
286 dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
288 static const char *old_names[] = {
290 "scrub_ddt_bookmark",
291 "scrub_ddt_class_max",
300 dsl_pool_t *dp = scn->scn_dp;
301 spa_t *spa = dp->dp_spa;
304 /* Remove any remnants of an old-style scrub. */
305 for (i = 0; old_names[i]; i++) {
306 (void) zap_remove(dp->dp_meta_objset,
307 DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx);
310 if (scn->scn_phys.scn_queue_obj != 0) {
311 VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
312 scn->scn_phys.scn_queue_obj, tx));
313 scn->scn_phys.scn_queue_obj = 0;
316 scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED;
319 * If we were "restarted" from a stopped state, don't bother
320 * with anything else.
322 if (scn->scn_phys.scn_state != DSS_SCANNING)
326 scn->scn_phys.scn_state = DSS_FINISHED;
328 scn->scn_phys.scn_state = DSS_CANCELED;
330 if (dsl_scan_restarting(scn, tx))
331 spa_history_log_internal(spa, "scan aborted, restarting", tx,
332 "errors=%llu", spa_get_errlog_size(spa));
334 spa_history_log_internal(spa, "scan cancelled", tx,
335 "errors=%llu", spa_get_errlog_size(spa));
337 spa_history_log_internal(spa, "scan done", tx,
338 "errors=%llu", spa_get_errlog_size(spa));
340 if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
341 mutex_enter(&spa->spa_scrub_lock);
342 while (spa->spa_scrub_inflight > 0) {
343 cv_wait(&spa->spa_scrub_io_cv,
344 &spa->spa_scrub_lock);
346 mutex_exit(&spa->spa_scrub_lock);
347 spa->spa_scrub_started = B_FALSE;
348 spa->spa_scrub_active = B_FALSE;
351 * If the scrub/resilver completed, update all DTLs to
352 * reflect this. Whether it succeeded or not, vacate
353 * all temporary scrub DTLs.
355 vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
356 complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE);
358 spa_event_notify(spa, NULL, NULL,
359 scn->scn_phys.scn_min_txg ?
360 ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
362 spa_errlog_rotate(spa);
365 * We may have finished replacing a device.
366 * Let the async thread assess this and handle the detach.
368 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
371 scn->scn_phys.scn_end_time = gethrestime_sec();
376 dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
378 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
380 if (scn->scn_phys.scn_state != DSS_SCANNING)
381 return (SET_ERROR(ENOENT));
387 dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
389 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
391 dsl_scan_done(scn, B_FALSE, tx);
392 dsl_scan_sync_state(scn, tx);
393 spa_event_notify(scn->scn_dp->dp_spa, NULL, NULL, ESC_ZFS_SCRUB_ABORT);
397 dsl_scan_cancel(dsl_pool_t *dp)
399 return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
400 dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED));
404 dsl_scan_is_paused_scrub(const dsl_scan_t *scn)
406 if (dsl_scan_scrubbing(scn->scn_dp) &&
407 scn->scn_phys.scn_flags & DSF_SCRUB_PAUSED)
414 dsl_scrub_pause_resume_check(void *arg, dmu_tx_t *tx)
416 pool_scrub_cmd_t *cmd = arg;
417 dsl_pool_t *dp = dmu_tx_pool(tx);
418 dsl_scan_t *scn = dp->dp_scan;
420 if (*cmd == POOL_SCRUB_PAUSE) {
421 /* can't pause a scrub when there is no in-progress scrub */
422 if (!dsl_scan_scrubbing(dp))
423 return (SET_ERROR(ENOENT));
425 /* can't pause a paused scrub */
426 if (dsl_scan_is_paused_scrub(scn))
427 return (SET_ERROR(EBUSY));
428 } else if (*cmd != POOL_SCRUB_NORMAL) {
429 return (SET_ERROR(ENOTSUP));
436 dsl_scrub_pause_resume_sync(void *arg, dmu_tx_t *tx)
438 pool_scrub_cmd_t *cmd = arg;
439 dsl_pool_t *dp = dmu_tx_pool(tx);
440 spa_t *spa = dp->dp_spa;
441 dsl_scan_t *scn = dp->dp_scan;
443 if (*cmd == POOL_SCRUB_PAUSE) {
444 /* can't pause a scrub when there is no in-progress scrub */
445 spa->spa_scan_pass_scrub_pause = gethrestime_sec();
446 scn->scn_phys.scn_flags |= DSF_SCRUB_PAUSED;
447 dsl_scan_sync_state(scn, tx);
448 spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_PAUSED);
450 ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL);
451 if (dsl_scan_is_paused_scrub(scn)) {
453 * We need to keep track of how much time we spend
454 * paused per pass so that we can adjust the scrub rate
455 * shown in the output of 'zpool status'
457 spa->spa_scan_pass_scrub_spent_paused +=
458 gethrestime_sec() - spa->spa_scan_pass_scrub_pause;
459 spa->spa_scan_pass_scrub_pause = 0;
460 scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED;
461 dsl_scan_sync_state(scn, tx);
467 * Set scrub pause/resume state if it makes sense to do so
470 dsl_scrub_set_pause_resume(const dsl_pool_t *dp, pool_scrub_cmd_t cmd)
472 return (dsl_sync_task(spa_name(dp->dp_spa),
473 dsl_scrub_pause_resume_check, dsl_scrub_pause_resume_sync, &cmd, 3,
474 ZFS_SPACE_CHECK_RESERVED));
478 dsl_scan_scrubbing(const dsl_pool_t *dp)
480 dsl_scan_t *scn = dp->dp_scan;
482 if (scn->scn_phys.scn_state == DSS_SCANNING &&
483 scn->scn_phys.scn_func == POOL_SCAN_SCRUB)
489 static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb,
490 dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn,
491 dmu_objset_type_t ostype, dmu_tx_t *tx);
492 static void dsl_scan_visitdnode(dsl_scan_t *, dsl_dataset_t *ds,
493 dmu_objset_type_t ostype,
494 dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx);
497 dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp)
499 zio_free(dp->dp_spa, txg, bp);
503 dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp)
505 ASSERT(dsl_pool_sync_context(dp));
506 zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, BP_GET_PSIZE(bpp),
511 dsl_scan_ds_maxtxg(dsl_dataset_t *ds)
513 uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg;
514 if (ds->ds_is_snapshot)
515 return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg));
520 dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx)
522 VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
523 DMU_POOL_DIRECTORY_OBJECT,
524 DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
525 &scn->scn_phys, tx));
528 extern int zfs_vdev_async_write_active_min_dirty_percent;
531 dsl_scan_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb)
533 /* we never skip user/group accounting objects */
534 if (zb && (int64_t)zb->zb_object < 0)
537 if (scn->scn_suspending)
538 return (B_TRUE); /* we're already suspending */
540 if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark))
541 return (B_FALSE); /* we're resuming */
543 /* We only know how to resume from level-0 blocks. */
544 if (zb && zb->zb_level != 0)
549 * - we have scanned for the maximum time: an entire txg
550 * timeout (default 5 sec)
552 * - we have scanned for at least the minimum time (default 1 sec
553 * for scrub, 3 sec for resilver), and either we have sufficient
554 * dirty data that we are starting to write more quickly
555 * (default 30%), or someone is explicitly waiting for this txg
558 * - the spa is shutting down because this pool is being exported
559 * or the machine is rebooting.
561 int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
562 zfs_resilver_min_time_ms : zfs_scan_min_time_ms;
563 uint64_t elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
564 int dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max;
565 if (elapsed_nanosecs / NANOSEC >= zfs_txg_timeout ||
566 (NSEC2MSEC(elapsed_nanosecs) > mintime &&
567 (txg_sync_waiting(scn->scn_dp) ||
568 dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent)) ||
569 spa_shutting_down(scn->scn_dp->dp_spa)) {
571 dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n",
572 (longlong_t)zb->zb_objset,
573 (longlong_t)zb->zb_object,
574 (longlong_t)zb->zb_level,
575 (longlong_t)zb->zb_blkid);
576 scn->scn_phys.scn_bookmark = *zb;
578 dprintf("suspending at DDT bookmark %llx/%llx/%llx/%llx\n",
579 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
580 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
581 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
582 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
583 scn->scn_suspending = B_TRUE;
589 typedef struct zil_scan_arg {
591 zil_header_t *zsa_zh;
596 dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
598 zil_scan_arg_t *zsa = arg;
599 dsl_pool_t *dp = zsa->zsa_dp;
600 dsl_scan_t *scn = dp->dp_scan;
601 zil_header_t *zh = zsa->zsa_zh;
604 if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
608 * One block ("stubby") can be allocated a long time ago; we
609 * want to visit that one because it has been allocated
610 * (on-disk) even if it hasn't been claimed (even though for
611 * scrub there's nothing to do to it).
613 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa))
616 SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
617 ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
619 VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
625 dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
627 if (lrc->lrc_txtype == TX_WRITE) {
628 zil_scan_arg_t *zsa = arg;
629 dsl_pool_t *dp = zsa->zsa_dp;
630 dsl_scan_t *scn = dp->dp_scan;
631 zil_header_t *zh = zsa->zsa_zh;
632 lr_write_t *lr = (lr_write_t *)lrc;
633 blkptr_t *bp = &lr->lr_blkptr;
636 if (BP_IS_HOLE(bp) ||
637 bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
641 * birth can be < claim_txg if this record's txg is
642 * already txg sync'ed (but this log block contains
643 * other records that are not synced)
645 if (claim_txg == 0 || bp->blk_birth < claim_txg)
648 SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
649 lr->lr_foid, ZB_ZIL_LEVEL,
650 lr->lr_offset / BP_GET_LSIZE(bp));
652 VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
658 dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh)
660 uint64_t claim_txg = zh->zh_claim_txg;
661 zil_scan_arg_t zsa = { dp, zh };
665 * We only want to visit blocks that have been claimed but not yet
666 * replayed (or, in read-only mode, blocks that *would* be claimed).
668 if (claim_txg == 0 && spa_writeable(dp->dp_spa))
671 zilog = zil_alloc(dp->dp_meta_objset, zh);
673 (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa,
681 dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp,
682 uint64_t objset, uint64_t object, uint64_t blkid)
684 zbookmark_phys_t czb;
685 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
687 if (zfs_no_scrub_prefetch)
690 if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg ||
691 (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE))
694 SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid);
696 (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp,
697 NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
698 ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
702 dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp,
703 const zbookmark_phys_t *zb)
706 * We never skip over user/group accounting objects (obj<0)
708 if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) &&
709 (int64_t)zb->zb_object >= 0) {
711 * If we already visited this bp & everything below (in
712 * a prior txg sync), don't bother doing it again.
714 if (zbookmark_subtree_completed(dnp, zb,
715 &scn->scn_phys.scn_bookmark))
719 * If we found the block we're trying to resume from, or
720 * we went past it to a different object, zero it out to
721 * indicate that it's OK to start checking for suspending
724 if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 ||
725 zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) {
726 dprintf("resuming at %llx/%llx/%llx/%llx\n",
727 (longlong_t)zb->zb_objset,
728 (longlong_t)zb->zb_object,
729 (longlong_t)zb->zb_level,
730 (longlong_t)zb->zb_blkid);
731 bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb));
738 * Return nonzero on i/o error.
739 * Return new buf to write out in *bufp.
742 dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
743 dnode_phys_t *dnp, const blkptr_t *bp,
744 const zbookmark_phys_t *zb, dmu_tx_t *tx)
746 dsl_pool_t *dp = scn->scn_dp;
747 int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
750 if (BP_GET_LEVEL(bp) > 0) {
751 arc_flags_t flags = ARC_FLAG_WAIT;
754 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
757 err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
758 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
760 scn->scn_phys.scn_errors++;
763 for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) {
764 dsl_scan_prefetch(scn, buf, cbp, zb->zb_objset,
765 zb->zb_object, zb->zb_blkid * epb + i);
767 for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) {
768 zbookmark_phys_t czb;
770 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
772 zb->zb_blkid * epb + i);
773 dsl_scan_visitbp(cbp, &czb, dnp,
774 ds, scn, ostype, tx);
776 arc_buf_destroy(buf, &buf);
777 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
778 arc_flags_t flags = ARC_FLAG_WAIT;
781 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
784 err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
785 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
787 scn->scn_phys.scn_errors++;
790 for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) {
791 for (j = 0; j < cdnp->dn_nblkptr; j++) {
792 blkptr_t *cbp = &cdnp->dn_blkptr[j];
793 dsl_scan_prefetch(scn, buf, cbp,
794 zb->zb_objset, zb->zb_blkid * epb + i, j);
797 for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) {
798 dsl_scan_visitdnode(scn, ds, ostype,
799 cdnp, zb->zb_blkid * epb + i, tx);
802 arc_buf_destroy(buf, &buf);
803 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
804 arc_flags_t flags = ARC_FLAG_WAIT;
808 err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
809 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
811 scn->scn_phys.scn_errors++;
817 dsl_scan_visitdnode(scn, ds, osp->os_type,
818 &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx);
820 if (OBJSET_BUF_HAS_USERUSED(buf)) {
822 * We also always visit user/group accounting
823 * objects, and never skip them, even if we are
824 * suspending. This is necessary so that the space
825 * deltas from this txg get integrated.
827 dsl_scan_visitdnode(scn, ds, osp->os_type,
828 &osp->os_groupused_dnode,
829 DMU_GROUPUSED_OBJECT, tx);
830 dsl_scan_visitdnode(scn, ds, osp->os_type,
831 &osp->os_userused_dnode,
832 DMU_USERUSED_OBJECT, tx);
834 arc_buf_destroy(buf, &buf);
841 dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds,
842 dmu_objset_type_t ostype, dnode_phys_t *dnp,
843 uint64_t object, dmu_tx_t *tx)
847 for (j = 0; j < dnp->dn_nblkptr; j++) {
848 zbookmark_phys_t czb;
850 SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
851 dnp->dn_nlevels - 1, j);
852 dsl_scan_visitbp(&dnp->dn_blkptr[j],
853 &czb, dnp, ds, scn, ostype, tx);
856 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
857 zbookmark_phys_t czb;
858 SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
860 dsl_scan_visitbp(&dnp->dn_spill,
861 &czb, dnp, ds, scn, ostype, tx);
866 * The arguments are in this order because mdb can only print the
867 * first 5; we want them to be useful.
870 dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb,
871 dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn,
872 dmu_objset_type_t ostype, dmu_tx_t *tx)
874 dsl_pool_t *dp = scn->scn_dp;
875 arc_buf_t *buf = NULL;
876 blkptr_t bp_toread = *bp;
878 /* ASSERT(pbuf == NULL || arc_released(pbuf)); */
880 if (dsl_scan_check_suspend(scn, zb))
883 if (dsl_scan_check_resume(scn, dnp, zb))
889 scn->scn_visited_this_txg++;
892 "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p",
893 ds, ds ? ds->ds_object : 0,
894 zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
897 if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
900 if (dsl_scan_recurse(scn, ds, ostype, dnp, &bp_toread, zb, tx) != 0)
904 * If dsl_scan_ddt() has already visited this block, it will have
905 * already done any translations or scrubbing, so don't call the
908 if (ddt_class_contains(dp->dp_spa,
909 scn->scn_phys.scn_ddt_class_max, bp)) {
915 * If this block is from the future (after cur_max_txg), then we
916 * are doing this on behalf of a deleted snapshot, and we will
917 * revisit the future block on the next pass of this dataset.
918 * Don't scan it now unless we need to because something
919 * under it was modified.
921 if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) {
922 scan_funcs[scn->scn_phys.scn_func](dp, bp, zb);
927 dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp,
932 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
933 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
934 dsl_scan_visitbp(bp, &zb, NULL,
935 ds, scn, DMU_OST_NONE, tx);
937 dprintf_ds(ds, "finished scan%s", "");
941 dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
943 dsl_pool_t *dp = ds->ds_dir->dd_pool;
944 dsl_scan_t *scn = dp->dp_scan;
947 if (scn->scn_phys.scn_state != DSS_SCANNING)
950 if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
951 if (ds->ds_is_snapshot) {
954 * - scn_cur_{min,max}_txg stays the same.
955 * - Setting the flag is not really necessary if
956 * scn_cur_max_txg == scn_max_txg, because there
957 * is nothing after this snapshot that we care
958 * about. However, we set it anyway and then
959 * ignore it when we retraverse it in
960 * dsl_scan_visitds().
962 scn->scn_phys.scn_bookmark.zb_objset =
963 dsl_dataset_phys(ds)->ds_next_snap_obj;
964 zfs_dbgmsg("destroying ds %llu; currently traversing; "
965 "reset zb_objset to %llu",
966 (u_longlong_t)ds->ds_object,
967 (u_longlong_t)dsl_dataset_phys(ds)->
969 scn->scn_phys.scn_flags |= DSF_VISIT_DS_AGAIN;
971 SET_BOOKMARK(&scn->scn_phys.scn_bookmark,
972 ZB_DESTROYED_OBJSET, 0, 0, 0);
973 zfs_dbgmsg("destroying ds %llu; currently traversing; "
974 "reset bookmark to -1,0,0,0",
975 (u_longlong_t)ds->ds_object);
977 } else if (zap_lookup_int_key(dp->dp_meta_objset,
978 scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
979 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
980 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
981 scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
982 if (ds->ds_is_snapshot) {
984 * We keep the same mintxg; it could be >
985 * ds_creation_txg if the previous snapshot was
988 VERIFY(zap_add_int_key(dp->dp_meta_objset,
989 scn->scn_phys.scn_queue_obj,
990 dsl_dataset_phys(ds)->ds_next_snap_obj,
992 zfs_dbgmsg("destroying ds %llu; in queue; "
993 "replacing with %llu",
994 (u_longlong_t)ds->ds_object,
995 (u_longlong_t)dsl_dataset_phys(ds)->
998 zfs_dbgmsg("destroying ds %llu; in queue; removing",
999 (u_longlong_t)ds->ds_object);
1004 * dsl_scan_sync() should be called after this, and should sync
1005 * out our changed state, but just to be safe, do it here.
1007 dsl_scan_sync_state(scn, tx);
1011 dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx)
1013 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1014 dsl_scan_t *scn = dp->dp_scan;
1017 if (scn->scn_phys.scn_state != DSS_SCANNING)
1020 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1022 if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
1023 scn->scn_phys.scn_bookmark.zb_objset =
1024 dsl_dataset_phys(ds)->ds_prev_snap_obj;
1025 zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
1026 "reset zb_objset to %llu",
1027 (u_longlong_t)ds->ds_object,
1028 (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
1029 } else if (zap_lookup_int_key(dp->dp_meta_objset,
1030 scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
1031 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1032 scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
1033 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1034 scn->scn_phys.scn_queue_obj,
1035 dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0);
1036 zfs_dbgmsg("snapshotting ds %llu; in queue; "
1037 "replacing with %llu",
1038 (u_longlong_t)ds->ds_object,
1039 (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
1041 dsl_scan_sync_state(scn, tx);
1045 dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx)
1047 dsl_pool_t *dp = ds1->ds_dir->dd_pool;
1048 dsl_scan_t *scn = dp->dp_scan;
1051 if (scn->scn_phys.scn_state != DSS_SCANNING)
1054 if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) {
1055 scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object;
1056 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
1057 "reset zb_objset to %llu",
1058 (u_longlong_t)ds1->ds_object,
1059 (u_longlong_t)ds2->ds_object);
1060 } else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) {
1061 scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object;
1062 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
1063 "reset zb_objset to %llu",
1064 (u_longlong_t)ds2->ds_object,
1065 (u_longlong_t)ds1->ds_object);
1068 if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
1069 ds1->ds_object, &mintxg) == 0) {
1072 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
1073 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
1074 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1075 scn->scn_phys.scn_queue_obj, ds1->ds_object, tx));
1076 err = zap_add_int_key(dp->dp_meta_objset,
1077 scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx);
1078 VERIFY(err == 0 || err == EEXIST);
1079 if (err == EEXIST) {
1080 /* Both were there to begin with */
1081 VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
1082 scn->scn_phys.scn_queue_obj,
1083 ds1->ds_object, mintxg, tx));
1085 zfs_dbgmsg("clone_swap ds %llu; in queue; "
1086 "replacing with %llu",
1087 (u_longlong_t)ds1->ds_object,
1088 (u_longlong_t)ds2->ds_object);
1089 } else if (zap_lookup_int_key(dp->dp_meta_objset,
1090 scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) {
1091 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
1092 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
1093 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1094 scn->scn_phys.scn_queue_obj, ds2->ds_object, tx));
1095 VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
1096 scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx));
1097 zfs_dbgmsg("clone_swap ds %llu; in queue; "
1098 "replacing with %llu",
1099 (u_longlong_t)ds2->ds_object,
1100 (u_longlong_t)ds1->ds_object);
1103 dsl_scan_sync_state(scn, tx);
1106 struct enqueue_clones_arg {
1113 enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
1115 struct enqueue_clones_arg *eca = arg;
1118 dsl_scan_t *scn = dp->dp_scan;
1120 if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != eca->originobj)
1123 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
1127 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != eca->originobj) {
1128 dsl_dataset_t *prev;
1129 err = dsl_dataset_hold_obj(dp,
1130 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1132 dsl_dataset_rele(ds, FTAG);
1137 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1138 scn->scn_phys.scn_queue_obj, ds->ds_object,
1139 dsl_dataset_phys(ds)->ds_prev_snap_txg, eca->tx) == 0);
1140 dsl_dataset_rele(ds, FTAG);
1145 dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx)
1147 dsl_pool_t *dp = scn->scn_dp;
1151 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1153 if (scn->scn_phys.scn_cur_min_txg >=
1154 scn->scn_phys.scn_max_txg) {
1156 * This can happen if this snapshot was created after the
1157 * scan started, and we already completed a previous snapshot
1158 * that was created after the scan started. This snapshot
1159 * only references blocks with:
1161 * birth < our ds_creation_txg
1162 * cur_min_txg is no less than ds_creation_txg.
1163 * We have already visited these blocks.
1165 * birth > scn_max_txg
1166 * The scan requested not to visit these blocks.
1168 * Subsequent snapshots (and clones) can reference our
1169 * blocks, or blocks with even higher birth times.
1170 * Therefore we do not need to visit them either,
1171 * so we do not add them to the work queue.
1173 * Note that checking for cur_min_txg >= cur_max_txg
1174 * is not sufficient, because in that case we may need to
1175 * visit subsequent snapshots. This happens when min_txg > 0,
1176 * which raises cur_min_txg. In this case we will visit
1177 * this dataset but skip all of its blocks, because the
1178 * rootbp's birth time is < cur_min_txg. Then we will
1179 * add the next snapshots/clones to the work queue.
1181 char *dsname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
1182 dsl_dataset_name(ds, dsname);
1183 zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because "
1184 "cur_min_txg (%llu) >= max_txg (%llu)",
1186 scn->scn_phys.scn_cur_min_txg,
1187 scn->scn_phys.scn_max_txg);
1188 kmem_free(dsname, MAXNAMELEN);
1193 if (dmu_objset_from_ds(ds, &os))
1197 * Only the ZIL in the head (non-snapshot) is valid. Even though
1198 * snapshots can have ZIL block pointers (which may be the same
1199 * BP as in the head), they must be ignored. So we traverse the
1200 * ZIL here, rather than in scan_recurse(), because the regular
1201 * snapshot block-sharing rules don't apply to it.
1203 if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !ds->ds_is_snapshot)
1204 dsl_scan_zil(dp, &os->os_zil_header);
1207 * Iterate over the bps in this ds.
1209 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1210 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1211 dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx);
1212 rrw_exit(&ds->ds_bp_rwlock, FTAG);
1214 char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1215 dsl_dataset_name(ds, dsname);
1216 zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
1218 (longlong_t)dsobj, dsname,
1219 (longlong_t)scn->scn_phys.scn_cur_min_txg,
1220 (longlong_t)scn->scn_phys.scn_cur_max_txg,
1221 (int)scn->scn_suspending);
1222 kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN);
1224 if (scn->scn_suspending)
1228 * We've finished this pass over this dataset.
1232 * If we did not completely visit this dataset, do another pass.
1234 if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) {
1235 zfs_dbgmsg("incomplete pass; visiting again");
1236 scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN;
1237 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1238 scn->scn_phys.scn_queue_obj, ds->ds_object,
1239 scn->scn_phys.scn_cur_max_txg, tx) == 0);
1244 * Add descendent datasets to work queue.
1246 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
1247 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1248 scn->scn_phys.scn_queue_obj,
1249 dsl_dataset_phys(ds)->ds_next_snap_obj,
1250 dsl_dataset_phys(ds)->ds_creation_txg, tx) == 0);
1252 if (dsl_dataset_phys(ds)->ds_num_children > 1) {
1253 boolean_t usenext = B_FALSE;
1254 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1257 * A bug in a previous version of the code could
1258 * cause upgrade_clones_cb() to not set
1259 * ds_next_snap_obj when it should, leading to a
1260 * missing entry. Therefore we can only use the
1261 * next_clones_obj when its count is correct.
1263 int err = zap_count(dp->dp_meta_objset,
1264 dsl_dataset_phys(ds)->ds_next_clones_obj, &count);
1266 count == dsl_dataset_phys(ds)->ds_num_children - 1)
1271 VERIFY0(zap_join_key(dp->dp_meta_objset,
1272 dsl_dataset_phys(ds)->ds_next_clones_obj,
1273 scn->scn_phys.scn_queue_obj,
1274 dsl_dataset_phys(ds)->ds_creation_txg, tx));
1276 struct enqueue_clones_arg eca;
1278 eca.originobj = ds->ds_object;
1280 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1281 enqueue_clones_cb, &eca, DS_FIND_CHILDREN));
1286 dsl_dataset_rele(ds, FTAG);
1291 enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
1296 dsl_scan_t *scn = dp->dp_scan;
1298 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
1302 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1303 dsl_dataset_t *prev;
1304 err = dsl_dataset_hold_obj(dp,
1305 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1307 dsl_dataset_rele(ds, FTAG);
1312 * If this is a clone, we don't need to worry about it for now.
1314 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) {
1315 dsl_dataset_rele(ds, FTAG);
1316 dsl_dataset_rele(prev, FTAG);
1319 dsl_dataset_rele(ds, FTAG);
1323 VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
1324 ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx) == 0);
1325 dsl_dataset_rele(ds, FTAG);
1330 * Scrub/dedup interaction.
1332 * If there are N references to a deduped block, we don't want to scrub it
1333 * N times -- ideally, we should scrub it exactly once.
1335 * We leverage the fact that the dde's replication class (enum ddt_class)
1336 * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
1337 * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
1339 * To prevent excess scrubbing, the scrub begins by walking the DDT
1340 * to find all blocks with refcnt > 1, and scrubs each of these once.
1341 * Since there are two replication classes which contain blocks with
1342 * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
1343 * Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
1345 * There would be nothing more to say if a block's refcnt couldn't change
1346 * during a scrub, but of course it can so we must account for changes
1347 * in a block's replication class.
1349 * Here's an example of what can occur:
1351 * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
1352 * when visited during the top-down scrub phase, it will be scrubbed twice.
1353 * This negates our scrub optimization, but is otherwise harmless.
1355 * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
1356 * on each visit during the top-down scrub phase, it will never be scrubbed.
1357 * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
1358 * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
1359 * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
1360 * while a scrub is in progress, it scrubs the block right then.
1363 dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx)
1365 ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark;
1366 ddt_entry_t dde = { 0 };
1370 while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) {
1373 if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max)
1375 dprintf("visiting ddb=%llu/%llu/%llu/%llx\n",
1376 (longlong_t)ddb->ddb_class,
1377 (longlong_t)ddb->ddb_type,
1378 (longlong_t)ddb->ddb_checksum,
1379 (longlong_t)ddb->ddb_cursor);
1381 /* There should be no pending changes to the dedup table */
1382 ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum];
1383 ASSERT(avl_first(&ddt->ddt_tree) == NULL);
1385 dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx);
1388 if (dsl_scan_check_suspend(scn, NULL))
1392 zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; "
1393 "suspending=%u", (longlong_t)n,
1394 (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_suspending);
1396 ASSERT(error == 0 || error == ENOENT);
1397 ASSERT(error != ENOENT ||
1398 ddb->ddb_class > scn->scn_phys.scn_ddt_class_max);
1403 dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum,
1404 ddt_entry_t *dde, dmu_tx_t *tx)
1406 const ddt_key_t *ddk = &dde->dde_key;
1407 ddt_phys_t *ddp = dde->dde_phys;
1409 zbookmark_phys_t zb = { 0 };
1411 if (scn->scn_phys.scn_state != DSS_SCANNING)
1414 for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
1415 if (ddp->ddp_phys_birth == 0 ||
1416 ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg)
1418 ddt_bp_create(checksum, ddk, ddp, &bp);
1420 scn->scn_visited_this_txg++;
1421 scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb);
1426 dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx)
1428 dsl_pool_t *dp = scn->scn_dp;
1432 if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
1433 scn->scn_phys.scn_ddt_class_max) {
1434 scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
1435 scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
1436 dsl_scan_ddt(scn, tx);
1437 if (scn->scn_suspending)
1441 if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) {
1442 /* First do the MOS & ORIGIN */
1444 scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
1445 scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
1446 dsl_scan_visit_rootbp(scn, NULL,
1447 &dp->dp_meta_rootbp, tx);
1448 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
1449 if (scn->scn_suspending)
1452 if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
1453 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1454 enqueue_cb, tx, DS_FIND_CHILDREN));
1456 dsl_scan_visitds(scn,
1457 dp->dp_origin_snap->ds_object, tx);
1459 ASSERT(!scn->scn_suspending);
1460 } else if (scn->scn_phys.scn_bookmark.zb_objset !=
1461 ZB_DESTROYED_OBJSET) {
1463 * If we were suspended, continue from here. Note if the
1464 * ds we were suspended on was deleted, the zb_objset may
1465 * be -1, so we will skip this and find a new objset
1468 dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx);
1469 if (scn->scn_suspending)
1474 * In case we were suspended right at the end of the ds, zero the
1475 * bookmark so we don't think that we're still trying to resume.
1477 bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t));
1479 /* keep pulling things out of the zap-object-as-queue */
1480 while (zap_cursor_init(&zc, dp->dp_meta_objset,
1481 scn->scn_phys.scn_queue_obj),
1482 zap_cursor_retrieve(&zc, &za) == 0) {
1486 dsobj = zfs_strtonum(za.za_name, NULL);
1487 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1488 scn->scn_phys.scn_queue_obj, dsobj, tx));
1490 /* Set up min/max txg */
1491 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1492 if (za.za_first_integer != 0) {
1493 scn->scn_phys.scn_cur_min_txg =
1494 MAX(scn->scn_phys.scn_min_txg,
1495 za.za_first_integer);
1497 scn->scn_phys.scn_cur_min_txg =
1498 MAX(scn->scn_phys.scn_min_txg,
1499 dsl_dataset_phys(ds)->ds_prev_snap_txg);
1501 scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds);
1502 dsl_dataset_rele(ds, FTAG);
1504 dsl_scan_visitds(scn, dsobj, tx);
1505 zap_cursor_fini(&zc);
1506 if (scn->scn_suspending)
1509 zap_cursor_fini(&zc);
1513 dsl_scan_free_should_suspend(dsl_scan_t *scn)
1515 uint64_t elapsed_nanosecs;
1520 if (scn->scn_visited_this_txg >= zfs_free_max_blocks)
1523 elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
1524 return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
1525 (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms &&
1526 txg_sync_waiting(scn->scn_dp)) ||
1527 spa_shutting_down(scn->scn_dp->dp_spa));
1531 dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1533 dsl_scan_t *scn = arg;
1535 if (!scn->scn_is_bptree ||
1536 (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) {
1537 if (dsl_scan_free_should_suspend(scn))
1538 return (SET_ERROR(ERESTART));
1541 zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa,
1542 dmu_tx_get_txg(tx), bp, BP_GET_PSIZE(bp), 0));
1543 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
1544 -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp),
1545 -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
1546 scn->scn_visited_this_txg++;
1551 dsl_scan_active(dsl_scan_t *scn)
1553 spa_t *spa = scn->scn_dp->dp_spa;
1554 uint64_t used = 0, comp, uncomp;
1556 if (spa->spa_load_state != SPA_LOAD_NONE)
1558 if (spa_shutting_down(spa))
1560 if ((scn->scn_phys.scn_state == DSS_SCANNING &&
1561 !dsl_scan_is_paused_scrub(scn)) ||
1562 (scn->scn_async_destroying && !scn->scn_async_stalled))
1565 if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
1566 (void) bpobj_space(&scn->scn_dp->dp_free_bpobj,
1567 &used, &comp, &uncomp);
1572 /* Called whenever a txg syncs. */
1574 dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
1576 dsl_scan_t *scn = dp->dp_scan;
1577 spa_t *spa = dp->dp_spa;
1581 * Check for scn_restart_txg before checking spa_load_state, so
1582 * that we can restart an old-style scan while the pool is being
1583 * imported (see dsl_scan_init).
1585 if (dsl_scan_restarting(scn, tx)) {
1586 pool_scan_func_t func = POOL_SCAN_SCRUB;
1587 dsl_scan_done(scn, B_FALSE, tx);
1588 if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
1589 func = POOL_SCAN_RESILVER;
1590 zfs_dbgmsg("restarting scan func=%u txg=%llu",
1592 dsl_scan_setup_sync(&func, tx);
1596 * Only process scans in sync pass 1.
1598 if (spa_sync_pass(dp->dp_spa) > 1)
1602 * If the spa is shutting down, then stop scanning. This will
1603 * ensure that the scan does not dirty any new data during the
1606 if (spa_shutting_down(spa))
1610 * If the scan is inactive due to a stalled async destroy, try again.
1612 if (!scn->scn_async_stalled && !dsl_scan_active(scn))
1615 scn->scn_visited_this_txg = 0;
1616 scn->scn_suspending = B_FALSE;
1617 scn->scn_sync_start_time = gethrtime();
1618 spa->spa_scrub_active = B_TRUE;
1621 * First process the async destroys. If we suspend, don't do
1622 * any scrubbing or resilvering. This ensures that there are no
1623 * async destroys while we are scanning, so the scan code doesn't
1624 * have to worry about traversing it. It is also faster to free the
1625 * blocks than to scrub them.
1627 if (zfs_free_bpobj_enabled &&
1628 spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
1629 scn->scn_is_bptree = B_FALSE;
1630 scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
1631 NULL, ZIO_FLAG_MUSTSUCCEED);
1632 err = bpobj_iterate(&dp->dp_free_bpobj,
1633 dsl_scan_free_block_cb, scn, tx);
1634 VERIFY3U(0, ==, zio_wait(scn->scn_zio_root));
1636 if (err != 0 && err != ERESTART)
1637 zfs_panic_recover("error %u from bpobj_iterate()", err);
1640 if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
1641 ASSERT(scn->scn_async_destroying);
1642 scn->scn_is_bptree = B_TRUE;
1643 scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
1644 NULL, ZIO_FLAG_MUSTSUCCEED);
1645 err = bptree_iterate(dp->dp_meta_objset,
1646 dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx);
1647 VERIFY0(zio_wait(scn->scn_zio_root));
1649 if (err == EIO || err == ECKSUM) {
1651 } else if (err != 0 && err != ERESTART) {
1652 zfs_panic_recover("error %u from "
1653 "traverse_dataset_destroyed()", err);
1656 if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) {
1657 /* finished; deactivate async destroy feature */
1658 spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx);
1659 ASSERT(!spa_feature_is_active(spa,
1660 SPA_FEATURE_ASYNC_DESTROY));
1661 VERIFY0(zap_remove(dp->dp_meta_objset,
1662 DMU_POOL_DIRECTORY_OBJECT,
1663 DMU_POOL_BPTREE_OBJ, tx));
1664 VERIFY0(bptree_free(dp->dp_meta_objset,
1665 dp->dp_bptree_obj, tx));
1666 dp->dp_bptree_obj = 0;
1667 scn->scn_async_destroying = B_FALSE;
1668 scn->scn_async_stalled = B_FALSE;
1671 * If we didn't make progress, mark the async
1672 * destroy as stalled, so that we will not initiate
1673 * a spa_sync() on its behalf. Note that we only
1674 * check this if we are not finished, because if the
1675 * bptree had no blocks for us to visit, we can
1676 * finish without "making progress".
1678 scn->scn_async_stalled =
1679 (scn->scn_visited_this_txg == 0);
1682 if (scn->scn_visited_this_txg) {
1683 zfs_dbgmsg("freed %llu blocks in %llums from "
1684 "free_bpobj/bptree txg %llu; err=%d",
1685 (longlong_t)scn->scn_visited_this_txg,
1687 NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
1688 (longlong_t)tx->tx_txg, err);
1689 scn->scn_visited_this_txg = 0;
1692 * Write out changes to the DDT that may be required as a
1693 * result of the blocks freed. This ensures that the DDT
1694 * is clean when a scrub/resilver runs.
1696 ddt_sync(spa, tx->tx_txg);
1700 if (dp->dp_free_dir != NULL && !scn->scn_async_destroying &&
1701 zfs_free_leak_on_eio &&
1702 (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 ||
1703 dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 ||
1704 dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) {
1706 * We have finished background destroying, but there is still
1707 * some space left in the dp_free_dir. Transfer this leaked
1708 * space to the dp_leak_dir.
1710 if (dp->dp_leak_dir == NULL) {
1711 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
1712 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
1714 VERIFY0(dsl_pool_open_special_dir(dp,
1715 LEAK_DIR_NAME, &dp->dp_leak_dir));
1716 rrw_exit(&dp->dp_config_rwlock, FTAG);
1718 dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD,
1719 dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes,
1720 dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes,
1721 dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx);
1722 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1723 -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes,
1724 -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes,
1725 -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx);
1727 if (dp->dp_free_dir != NULL && !scn->scn_async_destroying) {
1728 /* finished; verify that space accounting went to zero */
1729 ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes);
1730 ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes);
1731 ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes);
1734 if (scn->scn_phys.scn_state != DSS_SCANNING)
1737 if (scn->scn_done_txg == tx->tx_txg) {
1738 ASSERT(!scn->scn_suspending);
1739 /* finished with scan. */
1740 zfs_dbgmsg("txg %llu scan complete", tx->tx_txg);
1741 dsl_scan_done(scn, B_TRUE, tx);
1742 ASSERT3U(spa->spa_scrub_inflight, ==, 0);
1743 dsl_scan_sync_state(scn, tx);
1747 if (dsl_scan_is_paused_scrub(scn))
1750 if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
1751 scn->scn_phys.scn_ddt_class_max) {
1752 zfs_dbgmsg("doing scan sync txg %llu; "
1753 "ddt bm=%llu/%llu/%llu/%llx",
1754 (longlong_t)tx->tx_txg,
1755 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
1756 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
1757 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
1758 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
1759 ASSERT(scn->scn_phys.scn_bookmark.zb_objset == 0);
1760 ASSERT(scn->scn_phys.scn_bookmark.zb_object == 0);
1761 ASSERT(scn->scn_phys.scn_bookmark.zb_level == 0);
1762 ASSERT(scn->scn_phys.scn_bookmark.zb_blkid == 0);
1764 zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
1765 (longlong_t)tx->tx_txg,
1766 (longlong_t)scn->scn_phys.scn_bookmark.zb_objset,
1767 (longlong_t)scn->scn_phys.scn_bookmark.zb_object,
1768 (longlong_t)scn->scn_phys.scn_bookmark.zb_level,
1769 (longlong_t)scn->scn_phys.scn_bookmark.zb_blkid);
1772 scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
1773 NULL, ZIO_FLAG_CANFAIL);
1774 dsl_pool_config_enter(dp, FTAG);
1775 dsl_scan_visit(scn, tx);
1776 dsl_pool_config_exit(dp, FTAG);
1777 (void) zio_wait(scn->scn_zio_root);
1778 scn->scn_zio_root = NULL;
1780 zfs_dbgmsg("visited %llu blocks in %llums",
1781 (longlong_t)scn->scn_visited_this_txg,
1782 (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time));
1784 if (!scn->scn_suspending) {
1785 scn->scn_done_txg = tx->tx_txg + 1;
1786 zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
1787 tx->tx_txg, scn->scn_done_txg);
1790 if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
1791 mutex_enter(&spa->spa_scrub_lock);
1792 while (spa->spa_scrub_inflight > 0) {
1793 cv_wait(&spa->spa_scrub_io_cv,
1794 &spa->spa_scrub_lock);
1796 mutex_exit(&spa->spa_scrub_lock);
1799 dsl_scan_sync_state(scn, tx);
1803 * This will start a new scan, or restart an existing one.
1806 dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg)
1810 tx = dmu_tx_create_dd(dp->dp_mos_dir);
1811 VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT));
1813 txg = dmu_tx_get_txg(tx);
1814 dp->dp_scan->scn_restart_txg = txg;
1817 dp->dp_scan->scn_restart_txg = txg;
1819 zfs_dbgmsg("restarting resilver txg=%llu", txg);
1823 dsl_scan_resilvering(dsl_pool_t *dp)
1825 return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING &&
1826 dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER);
1834 count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp)
1839 * If we resume after a reboot, zab will be NULL; don't record
1840 * incomplete stats in that case.
1845 for (i = 0; i < 4; i++) {
1846 int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS;
1847 int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL;
1848 if (t & DMU_OT_NEWTYPE)
1850 zfs_blkstat_t *zb = &zab->zab_type[l][t];
1854 zb->zb_asize += BP_GET_ASIZE(bp);
1855 zb->zb_lsize += BP_GET_LSIZE(bp);
1856 zb->zb_psize += BP_GET_PSIZE(bp);
1857 zb->zb_gangs += BP_COUNT_GANG(bp);
1859 switch (BP_GET_NDVAS(bp)) {
1861 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
1862 DVA_GET_VDEV(&bp->blk_dva[1]))
1863 zb->zb_ditto_2_of_2_samevdev++;
1866 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
1867 DVA_GET_VDEV(&bp->blk_dva[1])) +
1868 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
1869 DVA_GET_VDEV(&bp->blk_dva[2])) +
1870 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
1871 DVA_GET_VDEV(&bp->blk_dva[2]));
1873 zb->zb_ditto_2_of_3_samevdev++;
1874 else if (equal == 3)
1875 zb->zb_ditto_3_of_3_samevdev++;
1882 dsl_scan_scrub_done(zio_t *zio)
1884 spa_t *spa = zio->io_spa;
1886 abd_free(zio->io_abd);
1888 mutex_enter(&spa->spa_scrub_lock);
1889 spa->spa_scrub_inflight--;
1890 cv_broadcast(&spa->spa_scrub_io_cv);
1892 if (zio->io_error && (zio->io_error != ECKSUM ||
1893 !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) {
1894 spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors++;
1896 mutex_exit(&spa->spa_scrub_lock);
1900 dsl_scan_scrub_cb(dsl_pool_t *dp,
1901 const blkptr_t *bp, const zbookmark_phys_t *zb)
1903 dsl_scan_t *scn = dp->dp_scan;
1904 size_t size = BP_GET_PSIZE(bp);
1905 spa_t *spa = dp->dp_spa;
1906 uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp);
1908 int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
1909 unsigned int scan_delay = 0;
1911 if (phys_birth <= scn->scn_phys.scn_min_txg ||
1912 phys_birth >= scn->scn_phys.scn_max_txg)
1915 count_block(dp->dp_blkstats, bp);
1917 if (BP_IS_EMBEDDED(bp))
1920 ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn));
1921 if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) {
1922 zio_flags |= ZIO_FLAG_SCRUB;
1924 scan_delay = zfs_scrub_delay;
1926 ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER);
1927 zio_flags |= ZIO_FLAG_RESILVER;
1929 scan_delay = zfs_resilver_delay;
1932 /* If it's an intent log block, failure is expected. */
1933 if (zb->zb_level == ZB_ZIL_LEVEL)
1934 zio_flags |= ZIO_FLAG_SPECULATIVE;
1936 for (int d = 0; d < BP_GET_NDVAS(bp); d++) {
1937 vdev_t *vd = vdev_lookup_top(spa,
1938 DVA_GET_VDEV(&bp->blk_dva[d]));
1941 * Keep track of how much data we've examined so that
1942 * zpool(1M) status can make useful progress reports.
1944 scn->scn_phys.scn_examined += DVA_GET_ASIZE(&bp->blk_dva[d]);
1945 spa->spa_scan_pass_exam += DVA_GET_ASIZE(&bp->blk_dva[d]);
1947 /* if it's a resilver, this may not be in the target range */
1949 if (DVA_GET_GANG(&bp->blk_dva[d])) {
1951 * Gang members may be spread across multiple
1952 * vdevs, so the best estimate we have is the
1953 * scrub range, which has already been checked.
1954 * XXX -- it would be better to change our
1955 * allocation policy to ensure that all
1956 * gang members reside on the same vdev.
1960 needs_io = vdev_dtl_contains(vd, DTL_PARTIAL,
1966 if (needs_io && !zfs_no_scrub_io) {
1967 vdev_t *rvd = spa->spa_root_vdev;
1968 uint64_t maxinflight = rvd->vdev_children *
1969 MAX(zfs_top_maxinflight, 1);
1971 mutex_enter(&spa->spa_scrub_lock);
1972 while (spa->spa_scrub_inflight >= maxinflight)
1973 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1974 spa->spa_scrub_inflight++;
1975 mutex_exit(&spa->spa_scrub_lock);
1978 * If we're seeing recent (zfs_scan_idle) "important" I/Os
1979 * then throttle our workload to limit the impact of a scan.
1981 if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle)
1982 delay(MAX((int)scan_delay, 0));
1984 zio_nowait(zio_read(NULL, spa, bp,
1985 abd_alloc_for_io(size, B_FALSE), size, dsl_scan_scrub_done,
1986 NULL, ZIO_PRIORITY_SCRUB, zio_flags, zb));
1989 /* do not relocate this block */
1994 * Called by the ZFS_IOC_POOL_SCAN ioctl to start a scrub or resilver.
1995 * Can also be called to resume a paused scrub.
1998 dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
2000 spa_t *spa = dp->dp_spa;
2001 dsl_scan_t *scn = dp->dp_scan;
2004 * Purge all vdev caches and probe all devices. We do this here
2005 * rather than in sync context because this requires a writer lock
2006 * on the spa_config lock, which we can't do from sync context. The
2007 * spa_scrub_reopen flag indicates that vdev_open() should not
2008 * attempt to start another scrub.
2010 spa_vdev_state_enter(spa, SCL_NONE);
2011 spa->spa_scrub_reopen = B_TRUE;
2012 vdev_reopen(spa->spa_root_vdev);
2013 spa->spa_scrub_reopen = B_FALSE;
2014 (void) spa_vdev_state_exit(spa, NULL, 0);
2016 if (func == POOL_SCAN_SCRUB && dsl_scan_is_paused_scrub(scn)) {
2017 /* got scrub start cmd, resume paused scrub */
2018 int err = dsl_scrub_set_pause_resume(scn->scn_dp,
2021 spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_RESUME);
2025 return (SET_ERROR(err));
2028 return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
2029 dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_NONE));
2033 dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx)
2035 return (scn->scn_restart_txg != 0 &&
2036 scn->scn_restart_txg <= tx->tx_txg);