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 (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright 2016 Gary Mills
27 #include <sys/dsl_scan.h>
28 #include <sys/dsl_pool.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_synctask.h>
33 #include <sys/dnode.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/dmu_objset.h>
39 #include <sys/zfs_context.h>
40 #include <sys/fs/zfs.h>
41 #include <sys/zfs_znode.h>
42 #include <sys/spa_impl.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/zil_impl.h>
45 #include <sys/zio_checksum.h>
48 #include <sys/sa_impl.h>
49 #include <sys/zfeature.h>
51 #include <sys/zfs_vfsops.h>
54 typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *,
55 const zbookmark_phys_t *);
57 static scan_cb_t dsl_scan_scrub_cb;
58 static void dsl_scan_cancel_sync(void *, dmu_tx_t *);
59 static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *);
60 static boolean_t dsl_scan_restarting(dsl_scan_t *, dmu_tx_t *);
62 int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */
63 int zfs_resilver_delay = 2; /* number of ticks to delay resilver */
64 int zfs_scrub_delay = 4; /* number of ticks to delay scrub */
65 int zfs_scan_idle = 50; /* idle window in clock ticks */
67 int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */
68 int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */
69 int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */
70 int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
71 int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
72 enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
73 int dsl_scan_delay_completion = B_FALSE; /* set to delay scan completion */
74 /* max number of blocks to free in a single TXG */
75 ulong zfs_free_max_blocks = 100000;
77 #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
78 ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
79 (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
82 * Enable/disable the processing of the free_bpobj object.
84 int zfs_free_bpobj_enabled = 1;
86 /* the order has to match pool_scan_type */
87 static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
89 dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */
90 dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */
94 dsl_scan_init(dsl_pool_t *dp, uint64_t txg)
98 spa_t *spa = dp->dp_spa;
101 scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP);
105 * It's possible that we're resuming a scan after a reboot so
106 * make sure that the scan_async_destroying flag is initialized
109 ASSERT(!scn->scn_async_destroying);
110 scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa,
111 SPA_FEATURE_ASYNC_DESTROY);
113 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
114 "scrub_func", sizeof (uint64_t), 1, &f);
117 * There was an old-style scrub in progress. Restart a
118 * new-style scrub from the beginning.
120 scn->scn_restart_txg = txg;
121 zfs_dbgmsg("old-style scrub was in progress; "
122 "restarting new-style scrub in txg %llu",
123 scn->scn_restart_txg);
126 * Load the queue obj from the old location so that it
127 * can be freed by dsl_scan_done().
129 (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
130 "scrub_queue", sizeof (uint64_t), 1,
131 &scn->scn_phys.scn_queue_obj);
133 err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
134 DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
137 * Detect if the pool contains the signature of #2094. If it
138 * does properly update the scn->scn_phys structure and notify
139 * the administrator by setting an errata for the pool.
141 if (err == EOVERFLOW) {
142 uint64_t zaptmp[SCAN_PHYS_NUMINTS + 1];
143 VERIFY3S(SCAN_PHYS_NUMINTS, ==, 24);
144 VERIFY3S(offsetof(dsl_scan_phys_t, scn_flags), ==,
145 (23 * sizeof (uint64_t)));
147 err = zap_lookup(dp->dp_meta_objset,
148 DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN,
149 sizeof (uint64_t), SCAN_PHYS_NUMINTS + 1, &zaptmp);
151 uint64_t overflow = zaptmp[SCAN_PHYS_NUMINTS];
153 if (overflow & ~DSL_SCAN_FLAGS_MASK ||
154 scn->scn_async_destroying) {
156 ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY;
160 bcopy(zaptmp, &scn->scn_phys,
161 SCAN_PHYS_NUMINTS * sizeof (uint64_t));
162 scn->scn_phys.scn_flags = overflow;
164 /* Required scrub already in progress. */
165 if (scn->scn_phys.scn_state == DSS_FINISHED ||
166 scn->scn_phys.scn_state == DSS_CANCELED)
168 ZPOOL_ERRATA_ZOL_2094_SCRUB;
177 if (scn->scn_phys.scn_state == DSS_SCANNING &&
178 spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) {
180 * A new-type scrub was in progress on an old
181 * pool, and the pool was accessed by old
182 * software. Restart from the beginning, since
183 * the old software may have changed the pool in
186 scn->scn_restart_txg = txg;
187 zfs_dbgmsg("new-style scrub was modified "
188 "by old software; restarting in txg %llu",
189 scn->scn_restart_txg);
193 spa_scan_stat_init(spa);
198 dsl_scan_fini(dsl_pool_t *dp)
201 kmem_free(dp->dp_scan, sizeof (dsl_scan_t));
208 dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
210 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
212 if (scn->scn_phys.scn_state == DSS_SCANNING)
213 return (SET_ERROR(EBUSY));
219 dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
221 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
222 pool_scan_func_t *funcp = arg;
223 dmu_object_type_t ot = 0;
224 dsl_pool_t *dp = scn->scn_dp;
225 spa_t *spa = dp->dp_spa;
227 ASSERT(scn->scn_phys.scn_state != DSS_SCANNING);
228 ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
229 bzero(&scn->scn_phys, sizeof (scn->scn_phys));
230 scn->scn_phys.scn_func = *funcp;
231 scn->scn_phys.scn_state = DSS_SCANNING;
232 scn->scn_phys.scn_min_txg = 0;
233 scn->scn_phys.scn_max_txg = tx->tx_txg;
234 scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
235 scn->scn_phys.scn_start_time = gethrestime_sec();
236 scn->scn_phys.scn_errors = 0;
237 scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
238 scn->scn_restart_txg = 0;
239 scn->scn_done_txg = 0;
240 spa_scan_stat_init(spa);
242 if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
243 scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
245 /* rewrite all disk labels */
246 vdev_config_dirty(spa->spa_root_vdev);
248 if (vdev_resilver_needed(spa->spa_root_vdev,
249 &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
250 spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
252 spa_event_notify(spa, NULL, ESC_ZFS_SCRUB_START);
255 spa->spa_scrub_started = B_TRUE;
257 * If this is an incremental scrub, limit the DDT scrub phase
258 * to just the auto-ditto class (for correctness); the rest
259 * of the scrub should go faster using top-down pruning.
261 if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
262 scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
266 /* back to the generic stuff */
268 if (dp->dp_blkstats == NULL) {
270 vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
272 bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
274 if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
275 ot = DMU_OT_ZAP_OTHER;
277 scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
278 ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
280 dsl_scan_sync_state(scn, tx);
282 spa_history_log_internal(spa, "scan setup", tx,
283 "func=%u mintxg=%llu maxtxg=%llu",
284 *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
289 dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
291 static const char *old_names[] = {
293 "scrub_ddt_bookmark",
294 "scrub_ddt_class_max",
303 dsl_pool_t *dp = scn->scn_dp;
304 spa_t *spa = dp->dp_spa;
307 /* Remove any remnants of an old-style scrub. */
308 for (i = 0; old_names[i]; i++) {
309 (void) zap_remove(dp->dp_meta_objset,
310 DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx);
313 if (scn->scn_phys.scn_queue_obj != 0) {
314 VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
315 scn->scn_phys.scn_queue_obj, tx));
316 scn->scn_phys.scn_queue_obj = 0;
320 * If we were "restarted" from a stopped state, don't bother
321 * with anything else.
323 if (scn->scn_phys.scn_state != DSS_SCANNING)
327 scn->scn_phys.scn_state = DSS_FINISHED;
329 scn->scn_phys.scn_state = DSS_CANCELED;
331 if (dsl_scan_restarting(scn, tx))
332 spa_history_log_internal(spa, "scan aborted, restarting", tx,
333 "errors=%llu", spa_get_errlog_size(spa));
335 spa_history_log_internal(spa, "scan cancelled", tx,
336 "errors=%llu", spa_get_errlog_size(spa));
338 spa_history_log_internal(spa, "scan done", tx,
339 "errors=%llu", spa_get_errlog_size(spa));
341 if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
342 mutex_enter(&spa->spa_scrub_lock);
343 while (spa->spa_scrub_inflight > 0) {
344 cv_wait(&spa->spa_scrub_io_cv,
345 &spa->spa_scrub_lock);
347 mutex_exit(&spa->spa_scrub_lock);
348 spa->spa_scrub_started = B_FALSE;
349 spa->spa_scrub_active = B_FALSE;
352 * If the scrub/resilver completed, update all DTLs to
353 * reflect this. Whether it succeeded or not, vacate
354 * all temporary scrub DTLs.
356 vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
357 complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE);
359 spa_event_notify(spa, NULL, 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();
373 if (spa->spa_errata == ZPOOL_ERRATA_ZOL_2094_SCRUB)
379 dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
381 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
383 if (scn->scn_phys.scn_state != DSS_SCANNING)
384 return (SET_ERROR(ENOENT));
390 dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
392 dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
394 dsl_scan_done(scn, B_FALSE, tx);
395 dsl_scan_sync_state(scn, tx);
399 dsl_scan_cancel(dsl_pool_t *dp)
401 return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
402 dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED));
405 static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb,
406 dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn,
407 dmu_objset_type_t ostype, dmu_tx_t *tx);
408 inline __attribute__((always_inline)) static void dsl_scan_visitdnode(
409 dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype,
410 dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx);
413 dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp)
415 zio_free(dp->dp_spa, txg, bp);
419 dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp)
421 ASSERT(dsl_pool_sync_context(dp));
422 zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags));
426 dsl_scan_ds_maxtxg(dsl_dataset_t *ds)
428 uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg;
429 if (ds->ds_is_snapshot)
430 return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg));
435 dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx)
437 VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
438 DMU_POOL_DIRECTORY_OBJECT,
439 DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
440 &scn->scn_phys, tx));
443 extern int zfs_vdev_async_write_active_min_dirty_percent;
446 dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_phys_t *zb)
448 uint64_t elapsed_nanosecs;
452 /* we never skip user/group accounting objects */
453 if (zb && (int64_t)zb->zb_object < 0)
456 if (scn->scn_pausing)
457 return (B_TRUE); /* we're already pausing */
459 if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark))
460 return (B_FALSE); /* we're resuming */
462 /* We only know how to resume from level-0 blocks. */
463 if (zb && zb->zb_level != 0)
468 * - we have scanned for the maximum time: an entire txg
469 * timeout (default 5 sec)
471 * - we have scanned for at least the minimum time (default 1 sec
472 * for scrub, 3 sec for resilver), and either we have sufficient
473 * dirty data that we are starting to write more quickly
474 * (default 30%), or someone is explicitly waiting for this txg
477 * - the spa is shutting down because this pool is being exported
478 * or the machine is rebooting.
480 mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
481 zfs_resilver_min_time_ms : zfs_scan_min_time_ms;
482 elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
483 dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max;
484 if (elapsed_nanosecs / NANOSEC >= zfs_txg_timeout ||
485 (NSEC2MSEC(elapsed_nanosecs) > mintime &&
486 (txg_sync_waiting(scn->scn_dp) ||
487 dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent)) ||
488 spa_shutting_down(scn->scn_dp->dp_spa)) {
490 dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n",
491 (longlong_t)zb->zb_objset,
492 (longlong_t)zb->zb_object,
493 (longlong_t)zb->zb_level,
494 (longlong_t)zb->zb_blkid);
495 scn->scn_phys.scn_bookmark = *zb;
497 dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n",
498 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
499 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
500 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
501 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
502 scn->scn_pausing = B_TRUE;
508 typedef struct zil_scan_arg {
510 zil_header_t *zsa_zh;
515 dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
517 zil_scan_arg_t *zsa = arg;
518 dsl_pool_t *dp = zsa->zsa_dp;
519 dsl_scan_t *scn = dp->dp_scan;
520 zil_header_t *zh = zsa->zsa_zh;
523 if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
527 * One block ("stubby") can be allocated a long time ago; we
528 * want to visit that one because it has been allocated
529 * (on-disk) even if it hasn't been claimed (even though for
530 * scrub there's nothing to do to it).
532 if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa))
535 SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
536 ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
538 VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
544 dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
546 if (lrc->lrc_txtype == TX_WRITE) {
547 zil_scan_arg_t *zsa = arg;
548 dsl_pool_t *dp = zsa->zsa_dp;
549 dsl_scan_t *scn = dp->dp_scan;
550 zil_header_t *zh = zsa->zsa_zh;
551 lr_write_t *lr = (lr_write_t *)lrc;
552 blkptr_t *bp = &lr->lr_blkptr;
555 if (BP_IS_HOLE(bp) ||
556 bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
560 * birth can be < claim_txg if this record's txg is
561 * already txg sync'ed (but this log block contains
562 * other records that are not synced)
564 if (claim_txg == 0 || bp->blk_birth < claim_txg)
567 SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
568 lr->lr_foid, ZB_ZIL_LEVEL,
569 lr->lr_offset / BP_GET_LSIZE(bp));
571 VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
577 dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh)
579 uint64_t claim_txg = zh->zh_claim_txg;
580 zil_scan_arg_t zsa = { dp, zh };
584 * We only want to visit blocks that have been claimed but not yet
585 * replayed (or, in read-only mode, blocks that *would* be claimed).
587 if (claim_txg == 0 && spa_writeable(dp->dp_spa))
590 zilog = zil_alloc(dp->dp_meta_objset, zh);
592 (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa,
600 dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp,
601 uint64_t objset, uint64_t object, uint64_t blkid)
603 zbookmark_phys_t czb;
604 arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
606 if (zfs_no_scrub_prefetch)
609 if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg ||
610 (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE))
613 SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid);
615 (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp,
616 NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
617 ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
621 dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp,
622 const zbookmark_phys_t *zb)
625 * We never skip over user/group accounting objects (obj<0)
627 if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) &&
628 (int64_t)zb->zb_object >= 0) {
630 * If we already visited this bp & everything below (in
631 * a prior txg sync), don't bother doing it again.
633 if (zbookmark_subtree_completed(dnp, zb,
634 &scn->scn_phys.scn_bookmark))
638 * If we found the block we're trying to resume from, or
639 * we went past it to a different object, zero it out to
640 * indicate that it's OK to start checking for pausing
643 if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 ||
644 zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) {
645 dprintf("resuming at %llx/%llx/%llx/%llx\n",
646 (longlong_t)zb->zb_objset,
647 (longlong_t)zb->zb_object,
648 (longlong_t)zb->zb_level,
649 (longlong_t)zb->zb_blkid);
650 bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb));
657 * Return nonzero on i/o error.
658 * Return new buf to write out in *bufp.
660 inline __attribute__((always_inline)) static int
661 dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
662 dnode_phys_t *dnp, const blkptr_t *bp,
663 const zbookmark_phys_t *zb, dmu_tx_t *tx)
665 dsl_pool_t *dp = scn->scn_dp;
666 int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
669 if (BP_GET_LEVEL(bp) > 0) {
670 arc_flags_t flags = ARC_FLAG_WAIT;
673 int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
676 err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
677 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
679 scn->scn_phys.scn_errors++;
682 for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) {
683 dsl_scan_prefetch(scn, buf, cbp, zb->zb_objset,
684 zb->zb_object, zb->zb_blkid * epb + i);
686 for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) {
687 zbookmark_phys_t czb;
689 SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
691 zb->zb_blkid * epb + i);
692 dsl_scan_visitbp(cbp, &czb, dnp,
693 ds, scn, ostype, tx);
695 (void) arc_buf_remove_ref(buf, &buf);
696 } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
697 arc_flags_t flags = ARC_FLAG_WAIT;
700 int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
703 err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
704 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
706 scn->scn_phys.scn_errors++;
709 for (i = 0, cdnp = buf->b_data; i < epb;
710 i += cdnp->dn_extra_slots + 1,
711 cdnp += cdnp->dn_extra_slots + 1) {
712 for (j = 0; j < cdnp->dn_nblkptr; j++) {
713 blkptr_t *cbp = &cdnp->dn_blkptr[j];
714 dsl_scan_prefetch(scn, buf, cbp,
715 zb->zb_objset, zb->zb_blkid * epb + i, j);
718 for (i = 0, cdnp = buf->b_data; i < epb;
719 i += cdnp->dn_extra_slots + 1,
720 cdnp += cdnp->dn_extra_slots + 1) {
721 dsl_scan_visitdnode(scn, ds, ostype,
722 cdnp, zb->zb_blkid * epb + i, tx);
725 (void) arc_buf_remove_ref(buf, &buf);
726 } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
727 arc_flags_t flags = ARC_FLAG_WAIT;
731 err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
732 ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
734 scn->scn_phys.scn_errors++;
740 dsl_scan_visitdnode(scn, ds, osp->os_type,
741 &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx);
743 if (OBJSET_BUF_HAS_USERUSED(buf)) {
745 * We also always visit user/group accounting
746 * objects, and never skip them, even if we are
747 * pausing. This is necessary so that the space
748 * deltas from this txg get integrated.
750 dsl_scan_visitdnode(scn, ds, osp->os_type,
751 &osp->os_groupused_dnode,
752 DMU_GROUPUSED_OBJECT, tx);
753 dsl_scan_visitdnode(scn, ds, osp->os_type,
754 &osp->os_userused_dnode,
755 DMU_USERUSED_OBJECT, tx);
757 (void) arc_buf_remove_ref(buf, &buf);
763 inline __attribute__((always_inline)) static void
764 dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds,
765 dmu_objset_type_t ostype, dnode_phys_t *dnp,
766 uint64_t object, dmu_tx_t *tx)
770 for (j = 0; j < dnp->dn_nblkptr; j++) {
771 zbookmark_phys_t czb;
773 SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
774 dnp->dn_nlevels - 1, j);
775 dsl_scan_visitbp(&dnp->dn_blkptr[j],
776 &czb, dnp, ds, scn, ostype, tx);
779 if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
780 zbookmark_phys_t czb;
781 SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
783 dsl_scan_visitbp(DN_SPILL_BLKPTR(dnp),
784 &czb, dnp, ds, scn, ostype, tx);
789 * The arguments are in this order because mdb can only print the
790 * first 5; we want them to be useful.
793 dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb,
794 dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn,
795 dmu_objset_type_t ostype, dmu_tx_t *tx)
797 dsl_pool_t *dp = scn->scn_dp;
800 bp_toread = kmem_alloc(sizeof (blkptr_t), KM_SLEEP);
803 /* ASSERT(pbuf == NULL || arc_released(pbuf)); */
805 if (dsl_scan_check_pause(scn, zb))
808 if (dsl_scan_check_resume(scn, dnp, zb))
814 scn->scn_visited_this_txg++;
817 * This debugging is commented out to conserve stack space. This
818 * function is called recursively and the debugging addes several
819 * bytes to the stack for each call. It can be commented back in
820 * if required to debug an issue in dsl_scan_visitbp().
823 * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p",
824 * ds, ds ? ds->ds_object : 0,
825 * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
829 if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
832 if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx) != 0)
836 * If dsl_scan_ddt() has aready visited this block, it will have
837 * already done any translations or scrubbing, so don't call the
840 if (ddt_class_contains(dp->dp_spa,
841 scn->scn_phys.scn_ddt_class_max, bp)) {
846 * If this block is from the future (after cur_max_txg), then we
847 * are doing this on behalf of a deleted snapshot, and we will
848 * revisit the future block on the next pass of this dataset.
849 * Don't scan it now unless we need to because something
850 * under it was modified.
852 if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) {
853 scan_funcs[scn->scn_phys.scn_func](dp, bp, zb);
856 kmem_free(bp_toread, sizeof (blkptr_t));
860 dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp,
865 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
866 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
867 dsl_scan_visitbp(bp, &zb, NULL,
868 ds, scn, DMU_OST_NONE, tx);
870 dprintf_ds(ds, "finished scan%s", "");
874 dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
876 dsl_pool_t *dp = ds->ds_dir->dd_pool;
877 dsl_scan_t *scn = dp->dp_scan;
880 if (scn->scn_phys.scn_state != DSS_SCANNING)
883 if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
884 if (ds->ds_is_snapshot) {
887 * - scn_cur_{min,max}_txg stays the same.
888 * - Setting the flag is not really necessary if
889 * scn_cur_max_txg == scn_max_txg, because there
890 * is nothing after this snapshot that we care
891 * about. However, we set it anyway and then
892 * ignore it when we retraverse it in
893 * dsl_scan_visitds().
895 scn->scn_phys.scn_bookmark.zb_objset =
896 dsl_dataset_phys(ds)->ds_next_snap_obj;
897 zfs_dbgmsg("destroying ds %llu; currently traversing; "
898 "reset zb_objset to %llu",
899 (u_longlong_t)ds->ds_object,
900 (u_longlong_t)dsl_dataset_phys(ds)->
902 scn->scn_phys.scn_flags |= DSF_VISIT_DS_AGAIN;
904 SET_BOOKMARK(&scn->scn_phys.scn_bookmark,
905 ZB_DESTROYED_OBJSET, 0, 0, 0);
906 zfs_dbgmsg("destroying ds %llu; currently traversing; "
907 "reset bookmark to -1,0,0,0",
908 (u_longlong_t)ds->ds_object);
910 } else if (zap_lookup_int_key(dp->dp_meta_objset,
911 scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
912 ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
913 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
914 scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
915 if (ds->ds_is_snapshot) {
917 * We keep the same mintxg; it could be >
918 * ds_creation_txg if the previous snapshot was
921 VERIFY(zap_add_int_key(dp->dp_meta_objset,
922 scn->scn_phys.scn_queue_obj,
923 dsl_dataset_phys(ds)->ds_next_snap_obj,
925 zfs_dbgmsg("destroying ds %llu; in queue; "
926 "replacing with %llu",
927 (u_longlong_t)ds->ds_object,
928 (u_longlong_t)dsl_dataset_phys(ds)->
931 zfs_dbgmsg("destroying ds %llu; in queue; removing",
932 (u_longlong_t)ds->ds_object);
937 * dsl_scan_sync() should be called after this, and should sync
938 * out our changed state, but just to be safe, do it here.
940 dsl_scan_sync_state(scn, tx);
944 dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx)
946 dsl_pool_t *dp = ds->ds_dir->dd_pool;
947 dsl_scan_t *scn = dp->dp_scan;
950 if (scn->scn_phys.scn_state != DSS_SCANNING)
953 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
955 if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
956 scn->scn_phys.scn_bookmark.zb_objset =
957 dsl_dataset_phys(ds)->ds_prev_snap_obj;
958 zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
959 "reset zb_objset to %llu",
960 (u_longlong_t)ds->ds_object,
961 (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
962 } else if (zap_lookup_int_key(dp->dp_meta_objset,
963 scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
964 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
965 scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
966 VERIFY(zap_add_int_key(dp->dp_meta_objset,
967 scn->scn_phys.scn_queue_obj,
968 dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0);
969 zfs_dbgmsg("snapshotting ds %llu; in queue; "
970 "replacing with %llu",
971 (u_longlong_t)ds->ds_object,
972 (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
974 dsl_scan_sync_state(scn, tx);
978 dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx)
980 dsl_pool_t *dp = ds1->ds_dir->dd_pool;
981 dsl_scan_t *scn = dp->dp_scan;
984 if (scn->scn_phys.scn_state != DSS_SCANNING)
987 if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) {
988 scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object;
989 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
990 "reset zb_objset to %llu",
991 (u_longlong_t)ds1->ds_object,
992 (u_longlong_t)ds2->ds_object);
993 } else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) {
994 scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object;
995 zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
996 "reset zb_objset to %llu",
997 (u_longlong_t)ds2->ds_object,
998 (u_longlong_t)ds1->ds_object);
1001 if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
1002 ds1->ds_object, &mintxg) == 0) {
1005 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
1006 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
1007 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1008 scn->scn_phys.scn_queue_obj, ds1->ds_object, tx));
1009 err = zap_add_int_key(dp->dp_meta_objset,
1010 scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx);
1011 VERIFY(err == 0 || err == EEXIST);
1012 if (err == EEXIST) {
1013 /* Both were there to begin with */
1014 VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
1015 scn->scn_phys.scn_queue_obj,
1016 ds1->ds_object, mintxg, tx));
1018 zfs_dbgmsg("clone_swap ds %llu; in queue; "
1019 "replacing with %llu",
1020 (u_longlong_t)ds1->ds_object,
1021 (u_longlong_t)ds2->ds_object);
1022 } else if (zap_lookup_int_key(dp->dp_meta_objset,
1023 scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) {
1024 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
1025 ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
1026 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1027 scn->scn_phys.scn_queue_obj, ds2->ds_object, tx));
1028 VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
1029 scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx));
1030 zfs_dbgmsg("clone_swap ds %llu; in queue; "
1031 "replacing with %llu",
1032 (u_longlong_t)ds2->ds_object,
1033 (u_longlong_t)ds1->ds_object);
1036 dsl_scan_sync_state(scn, tx);
1039 struct enqueue_clones_arg {
1046 enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
1048 struct enqueue_clones_arg *eca = arg;
1051 dsl_scan_t *scn = dp->dp_scan;
1053 if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != eca->originobj)
1056 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
1060 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != eca->originobj) {
1061 dsl_dataset_t *prev;
1062 err = dsl_dataset_hold_obj(dp,
1063 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1065 dsl_dataset_rele(ds, FTAG);
1070 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1071 scn->scn_phys.scn_queue_obj, ds->ds_object,
1072 dsl_dataset_phys(ds)->ds_prev_snap_txg, eca->tx) == 0);
1073 dsl_dataset_rele(ds, FTAG);
1078 dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx)
1080 dsl_pool_t *dp = scn->scn_dp;
1085 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1087 if (scn->scn_phys.scn_cur_min_txg >=
1088 scn->scn_phys.scn_max_txg) {
1090 * This can happen if this snapshot was created after the
1091 * scan started, and we already completed a previous snapshot
1092 * that was created after the scan started. This snapshot
1093 * only references blocks with:
1095 * birth < our ds_creation_txg
1096 * cur_min_txg is no less than ds_creation_txg.
1097 * We have already visited these blocks.
1099 * birth > scn_max_txg
1100 * The scan requested not to visit these blocks.
1102 * Subsequent snapshots (and clones) can reference our
1103 * blocks, or blocks with even higher birth times.
1104 * Therefore we do not need to visit them either,
1105 * so we do not add them to the work queue.
1107 * Note that checking for cur_min_txg >= cur_max_txg
1108 * is not sufficient, because in that case we may need to
1109 * visit subsequent snapshots. This happens when min_txg > 0,
1110 * which raises cur_min_txg. In this case we will visit
1111 * this dataset but skip all of its blocks, because the
1112 * rootbp's birth time is < cur_min_txg. Then we will
1113 * add the next snapshots/clones to the work queue.
1115 char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1116 dsl_dataset_name(ds, dsname);
1117 zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because "
1118 "cur_min_txg (%llu) >= max_txg (%llu)",
1120 scn->scn_phys.scn_cur_min_txg,
1121 scn->scn_phys.scn_max_txg);
1122 kmem_free(dsname, MAXNAMELEN);
1127 if (dmu_objset_from_ds(ds, &os))
1131 * Only the ZIL in the head (non-snapshot) is valid. Even though
1132 * snapshots can have ZIL block pointers (which may be the same
1133 * BP as in the head), they must be ignored. So we traverse the
1134 * ZIL here, rather than in scan_recurse(), because the regular
1135 * snapshot block-sharing rules don't apply to it.
1137 if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !ds->ds_is_snapshot)
1138 dsl_scan_zil(dp, &os->os_zil_header);
1141 * Iterate over the bps in this ds.
1143 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1144 dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx);
1146 dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1147 dsl_dataset_name(ds, dsname);
1148 zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
1150 (longlong_t)dsobj, dsname,
1151 (longlong_t)scn->scn_phys.scn_cur_min_txg,
1152 (longlong_t)scn->scn_phys.scn_cur_max_txg,
1153 (int)scn->scn_pausing);
1154 kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN);
1156 if (scn->scn_pausing)
1160 * We've finished this pass over this dataset.
1164 * If we did not completely visit this dataset, do another pass.
1166 if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) {
1167 zfs_dbgmsg("incomplete pass; visiting again");
1168 scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN;
1169 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1170 scn->scn_phys.scn_queue_obj, ds->ds_object,
1171 scn->scn_phys.scn_cur_max_txg, tx) == 0);
1176 * Add descendent datasets to work queue.
1178 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
1179 VERIFY(zap_add_int_key(dp->dp_meta_objset,
1180 scn->scn_phys.scn_queue_obj,
1181 dsl_dataset_phys(ds)->ds_next_snap_obj,
1182 dsl_dataset_phys(ds)->ds_creation_txg, tx) == 0);
1184 if (dsl_dataset_phys(ds)->ds_num_children > 1) {
1185 boolean_t usenext = B_FALSE;
1186 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1189 * A bug in a previous version of the code could
1190 * cause upgrade_clones_cb() to not set
1191 * ds_next_snap_obj when it should, leading to a
1192 * missing entry. Therefore we can only use the
1193 * next_clones_obj when its count is correct.
1195 int err = zap_count(dp->dp_meta_objset,
1196 dsl_dataset_phys(ds)->ds_next_clones_obj, &count);
1198 count == dsl_dataset_phys(ds)->ds_num_children - 1)
1203 VERIFY0(zap_join_key(dp->dp_meta_objset,
1204 dsl_dataset_phys(ds)->ds_next_clones_obj,
1205 scn->scn_phys.scn_queue_obj,
1206 dsl_dataset_phys(ds)->ds_creation_txg, tx));
1208 struct enqueue_clones_arg eca;
1210 eca.originobj = ds->ds_object;
1212 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1213 enqueue_clones_cb, &eca, DS_FIND_CHILDREN));
1218 dsl_dataset_rele(ds, FTAG);
1223 enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
1228 dsl_scan_t *scn = dp->dp_scan;
1230 err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
1234 while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1235 dsl_dataset_t *prev;
1236 err = dsl_dataset_hold_obj(dp,
1237 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1239 dsl_dataset_rele(ds, FTAG);
1244 * If this is a clone, we don't need to worry about it for now.
1246 if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) {
1247 dsl_dataset_rele(ds, FTAG);
1248 dsl_dataset_rele(prev, FTAG);
1251 dsl_dataset_rele(ds, FTAG);
1255 VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
1256 ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx) == 0);
1257 dsl_dataset_rele(ds, FTAG);
1262 * Scrub/dedup interaction.
1264 * If there are N references to a deduped block, we don't want to scrub it
1265 * N times -- ideally, we should scrub it exactly once.
1267 * We leverage the fact that the dde's replication class (enum ddt_class)
1268 * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
1269 * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
1271 * To prevent excess scrubbing, the scrub begins by walking the DDT
1272 * to find all blocks with refcnt > 1, and scrubs each of these once.
1273 * Since there are two replication classes which contain blocks with
1274 * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
1275 * Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
1277 * There would be nothing more to say if a block's refcnt couldn't change
1278 * during a scrub, but of course it can so we must account for changes
1279 * in a block's replication class.
1281 * Here's an example of what can occur:
1283 * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
1284 * when visited during the top-down scrub phase, it will be scrubbed twice.
1285 * This negates our scrub optimization, but is otherwise harmless.
1287 * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
1288 * on each visit during the top-down scrub phase, it will never be scrubbed.
1289 * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
1290 * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
1291 * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
1292 * while a scrub is in progress, it scrubs the block right then.
1295 dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx)
1297 ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark;
1302 bzero(&dde, sizeof (ddt_entry_t));
1304 while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) {
1307 if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max)
1309 dprintf("visiting ddb=%llu/%llu/%llu/%llx\n",
1310 (longlong_t)ddb->ddb_class,
1311 (longlong_t)ddb->ddb_type,
1312 (longlong_t)ddb->ddb_checksum,
1313 (longlong_t)ddb->ddb_cursor);
1315 /* There should be no pending changes to the dedup table */
1316 ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum];
1317 ASSERT(avl_first(&ddt->ddt_tree) == NULL);
1319 dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx);
1322 if (dsl_scan_check_pause(scn, NULL))
1326 zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u",
1327 (longlong_t)n, (int)scn->scn_phys.scn_ddt_class_max,
1328 (int)scn->scn_pausing);
1330 ASSERT(error == 0 || error == ENOENT);
1331 ASSERT(error != ENOENT ||
1332 ddb->ddb_class > scn->scn_phys.scn_ddt_class_max);
1337 dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum,
1338 ddt_entry_t *dde, dmu_tx_t *tx)
1340 const ddt_key_t *ddk = &dde->dde_key;
1341 ddt_phys_t *ddp = dde->dde_phys;
1343 zbookmark_phys_t zb = { 0 };
1346 if (scn->scn_phys.scn_state != DSS_SCANNING)
1349 for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
1350 if (ddp->ddp_phys_birth == 0 ||
1351 ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg)
1353 ddt_bp_create(checksum, ddk, ddp, &bp);
1355 scn->scn_visited_this_txg++;
1356 scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb);
1361 dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx)
1363 dsl_pool_t *dp = scn->scn_dp;
1365 zap_attribute_t *za;
1367 if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
1368 scn->scn_phys.scn_ddt_class_max) {
1369 scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
1370 scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
1371 dsl_scan_ddt(scn, tx);
1372 if (scn->scn_pausing)
1376 if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) {
1377 /* First do the MOS & ORIGIN */
1379 scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
1380 scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
1381 dsl_scan_visit_rootbp(scn, NULL,
1382 &dp->dp_meta_rootbp, tx);
1383 spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
1384 if (scn->scn_pausing)
1387 if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
1388 VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
1389 enqueue_cb, tx, DS_FIND_CHILDREN));
1391 dsl_scan_visitds(scn,
1392 dp->dp_origin_snap->ds_object, tx);
1394 ASSERT(!scn->scn_pausing);
1395 } else if (scn->scn_phys.scn_bookmark.zb_objset !=
1396 ZB_DESTROYED_OBJSET) {
1398 * If we were paused, continue from here. Note if the
1399 * ds we were paused on was deleted, the zb_objset may
1400 * be -1, so we will skip this and find a new objset
1403 dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx);
1404 if (scn->scn_pausing)
1409 * In case we were paused right at the end of the ds, zero the
1410 * bookmark so we don't think that we're still trying to resume.
1412 bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t));
1413 zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
1414 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1416 /* keep pulling things out of the zap-object-as-queue */
1417 while (zap_cursor_init(zc, dp->dp_meta_objset,
1418 scn->scn_phys.scn_queue_obj),
1419 zap_cursor_retrieve(zc, za) == 0) {
1423 dsobj = strtonum(za->za_name, NULL);
1424 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
1425 scn->scn_phys.scn_queue_obj, dsobj, tx));
1427 /* Set up min/max txg */
1428 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1429 if (za->za_first_integer != 0) {
1430 scn->scn_phys.scn_cur_min_txg =
1431 MAX(scn->scn_phys.scn_min_txg,
1432 za->za_first_integer);
1434 scn->scn_phys.scn_cur_min_txg =
1435 MAX(scn->scn_phys.scn_min_txg,
1436 dsl_dataset_phys(ds)->ds_prev_snap_txg);
1438 scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds);
1439 dsl_dataset_rele(ds, FTAG);
1441 dsl_scan_visitds(scn, dsobj, tx);
1442 zap_cursor_fini(zc);
1443 if (scn->scn_pausing)
1446 zap_cursor_fini(zc);
1448 kmem_free(za, sizeof (zap_attribute_t));
1449 kmem_free(zc, sizeof (zap_cursor_t));
1453 dsl_scan_free_should_pause(dsl_scan_t *scn)
1455 uint64_t elapsed_nanosecs;
1460 if (scn->scn_visited_this_txg >= zfs_free_max_blocks)
1463 elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
1464 return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
1465 (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms &&
1466 txg_sync_waiting(scn->scn_dp)) ||
1467 spa_shutting_down(scn->scn_dp->dp_spa));
1471 dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1473 dsl_scan_t *scn = arg;
1475 if (!scn->scn_is_bptree ||
1476 (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) {
1477 if (dsl_scan_free_should_pause(scn))
1478 return (SET_ERROR(ERESTART));
1481 zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa,
1482 dmu_tx_get_txg(tx), bp, 0));
1483 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
1484 -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp),
1485 -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
1486 scn->scn_visited_this_txg++;
1491 dsl_scan_active(dsl_scan_t *scn)
1493 spa_t *spa = scn->scn_dp->dp_spa;
1494 uint64_t used = 0, comp, uncomp;
1496 if (spa->spa_load_state != SPA_LOAD_NONE)
1498 if (spa_shutting_down(spa))
1500 if (scn->scn_phys.scn_state == DSS_SCANNING ||
1501 (scn->scn_async_destroying && !scn->scn_async_stalled))
1504 if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
1505 (void) bpobj_space(&scn->scn_dp->dp_free_bpobj,
1506 &used, &comp, &uncomp);
1512 dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
1514 dsl_scan_t *scn = dp->dp_scan;
1515 spa_t *spa = dp->dp_spa;
1519 * Check for scn_restart_txg before checking spa_load_state, so
1520 * that we can restart an old-style scan while the pool is being
1521 * imported (see dsl_scan_init).
1523 if (dsl_scan_restarting(scn, tx)) {
1524 pool_scan_func_t func = POOL_SCAN_SCRUB;
1525 dsl_scan_done(scn, B_FALSE, tx);
1526 if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
1527 func = POOL_SCAN_RESILVER;
1528 zfs_dbgmsg("restarting scan func=%u txg=%llu",
1530 dsl_scan_setup_sync(&func, tx);
1534 * Only process scans in sync pass 1.
1536 if (spa_sync_pass(dp->dp_spa) > 1)
1540 * If the spa is shutting down, then stop scanning. This will
1541 * ensure that the scan does not dirty any new data during the
1544 if (spa_shutting_down(spa))
1548 * If the scan is inactive due to a stalled async destroy, try again.
1550 if (!scn->scn_async_stalled && !dsl_scan_active(scn))
1553 scn->scn_visited_this_txg = 0;
1554 scn->scn_pausing = B_FALSE;
1555 scn->scn_sync_start_time = gethrtime();
1556 spa->spa_scrub_active = B_TRUE;
1559 * First process the async destroys. If we pause, don't do
1560 * any scrubbing or resilvering. This ensures that there are no
1561 * async destroys while we are scanning, so the scan code doesn't
1562 * have to worry about traversing it. It is also faster to free the
1563 * blocks than to scrub them.
1565 if (zfs_free_bpobj_enabled &&
1566 spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
1567 scn->scn_is_bptree = B_FALSE;
1568 scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
1569 NULL, ZIO_FLAG_MUSTSUCCEED);
1570 err = bpobj_iterate(&dp->dp_free_bpobj,
1571 dsl_scan_free_block_cb, scn, tx);
1572 VERIFY3U(0, ==, zio_wait(scn->scn_zio_root));
1574 if (err != 0 && err != ERESTART)
1575 zfs_panic_recover("error %u from bpobj_iterate()", err);
1578 if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
1579 ASSERT(scn->scn_async_destroying);
1580 scn->scn_is_bptree = B_TRUE;
1581 scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
1582 NULL, ZIO_FLAG_MUSTSUCCEED);
1583 err = bptree_iterate(dp->dp_meta_objset,
1584 dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx);
1585 VERIFY0(zio_wait(scn->scn_zio_root));
1587 if (err == EIO || err == ECKSUM) {
1589 } else if (err != 0 && err != ERESTART) {
1590 zfs_panic_recover("error %u from "
1591 "traverse_dataset_destroyed()", err);
1594 if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) {
1595 /* finished; deactivate async destroy feature */
1596 spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx);
1597 ASSERT(!spa_feature_is_active(spa,
1598 SPA_FEATURE_ASYNC_DESTROY));
1599 VERIFY0(zap_remove(dp->dp_meta_objset,
1600 DMU_POOL_DIRECTORY_OBJECT,
1601 DMU_POOL_BPTREE_OBJ, tx));
1602 VERIFY0(bptree_free(dp->dp_meta_objset,
1603 dp->dp_bptree_obj, tx));
1604 dp->dp_bptree_obj = 0;
1605 scn->scn_async_destroying = B_FALSE;
1606 scn->scn_async_stalled = B_FALSE;
1609 * If we didn't make progress, mark the async
1610 * destroy as stalled, so that we will not initiate
1611 * a spa_sync() on its behalf. Note that we only
1612 * check this if we are not finished, because if the
1613 * bptree had no blocks for us to visit, we can
1614 * finish without "making progress".
1616 scn->scn_async_stalled =
1617 (scn->scn_visited_this_txg == 0);
1620 if (scn->scn_visited_this_txg) {
1621 zfs_dbgmsg("freed %llu blocks in %llums from "
1622 "free_bpobj/bptree txg %llu; err=%u",
1623 (longlong_t)scn->scn_visited_this_txg,
1625 NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
1626 (longlong_t)tx->tx_txg, err);
1627 scn->scn_visited_this_txg = 0;
1630 * Write out changes to the DDT that may be required as a
1631 * result of the blocks freed. This ensures that the DDT
1632 * is clean when a scrub/resilver runs.
1634 ddt_sync(spa, tx->tx_txg);
1638 if (dp->dp_free_dir != NULL && !scn->scn_async_destroying &&
1639 zfs_free_leak_on_eio &&
1640 (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 ||
1641 dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 ||
1642 dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) {
1644 * We have finished background destroying, but there is still
1645 * some space left in the dp_free_dir. Transfer this leaked
1646 * space to the dp_leak_dir.
1648 if (dp->dp_leak_dir == NULL) {
1649 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
1650 (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
1652 VERIFY0(dsl_pool_open_special_dir(dp,
1653 LEAK_DIR_NAME, &dp->dp_leak_dir));
1654 rrw_exit(&dp->dp_config_rwlock, FTAG);
1656 dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD,
1657 dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes,
1658 dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes,
1659 dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx);
1660 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1661 -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes,
1662 -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes,
1663 -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx);
1665 if (dp->dp_free_dir != NULL && !scn->scn_async_destroying) {
1666 /* finished; verify that space accounting went to zero */
1667 ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes);
1668 ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes);
1669 ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes);
1672 if (scn->scn_phys.scn_state != DSS_SCANNING)
1675 if (scn->scn_done_txg == tx->tx_txg) {
1676 ASSERT(!scn->scn_pausing);
1677 /* finished with scan. */
1678 zfs_dbgmsg("txg %llu scan complete", tx->tx_txg);
1679 dsl_scan_done(scn, B_TRUE, tx);
1680 ASSERT3U(spa->spa_scrub_inflight, ==, 0);
1681 dsl_scan_sync_state(scn, tx);
1685 if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
1686 scn->scn_phys.scn_ddt_class_max) {
1687 zfs_dbgmsg("doing scan sync txg %llu; "
1688 "ddt bm=%llu/%llu/%llu/%llx",
1689 (longlong_t)tx->tx_txg,
1690 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
1691 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
1692 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
1693 (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
1694 ASSERT(scn->scn_phys.scn_bookmark.zb_objset == 0);
1695 ASSERT(scn->scn_phys.scn_bookmark.zb_object == 0);
1696 ASSERT(scn->scn_phys.scn_bookmark.zb_level == 0);
1697 ASSERT(scn->scn_phys.scn_bookmark.zb_blkid == 0);
1699 zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
1700 (longlong_t)tx->tx_txg,
1701 (longlong_t)scn->scn_phys.scn_bookmark.zb_objset,
1702 (longlong_t)scn->scn_phys.scn_bookmark.zb_object,
1703 (longlong_t)scn->scn_phys.scn_bookmark.zb_level,
1704 (longlong_t)scn->scn_phys.scn_bookmark.zb_blkid);
1707 scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
1708 NULL, ZIO_FLAG_CANFAIL);
1709 dsl_pool_config_enter(dp, FTAG);
1710 dsl_scan_visit(scn, tx);
1711 dsl_pool_config_exit(dp, FTAG);
1712 (void) zio_wait(scn->scn_zio_root);
1713 scn->scn_zio_root = NULL;
1715 zfs_dbgmsg("visited %llu blocks in %llums",
1716 (longlong_t)scn->scn_visited_this_txg,
1717 (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time));
1719 if (!scn->scn_pausing) {
1720 scn->scn_done_txg = tx->tx_txg + 1;
1721 zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
1722 tx->tx_txg, scn->scn_done_txg);
1725 if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
1726 mutex_enter(&spa->spa_scrub_lock);
1727 while (spa->spa_scrub_inflight > 0) {
1728 cv_wait(&spa->spa_scrub_io_cv,
1729 &spa->spa_scrub_lock);
1731 mutex_exit(&spa->spa_scrub_lock);
1734 dsl_scan_sync_state(scn, tx);
1738 * This will start a new scan, or restart an existing one.
1741 dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg)
1745 tx = dmu_tx_create_dd(dp->dp_mos_dir);
1746 VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT));
1748 txg = dmu_tx_get_txg(tx);
1749 dp->dp_scan->scn_restart_txg = txg;
1752 dp->dp_scan->scn_restart_txg = txg;
1754 zfs_dbgmsg("restarting resilver txg=%llu", txg);
1758 dsl_scan_resilvering(dsl_pool_t *dp)
1760 return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING &&
1761 dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER);
1769 count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp)
1774 * If we resume after a reboot, zab will be NULL; don't record
1775 * incomplete stats in that case.
1780 for (i = 0; i < 4; i++) {
1781 int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS;
1782 int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL;
1786 if (t & DMU_OT_NEWTYPE)
1789 zb = &zab->zab_type[l][t];
1791 zb->zb_asize += BP_GET_ASIZE(bp);
1792 zb->zb_lsize += BP_GET_LSIZE(bp);
1793 zb->zb_psize += BP_GET_PSIZE(bp);
1794 zb->zb_gangs += BP_COUNT_GANG(bp);
1796 switch (BP_GET_NDVAS(bp)) {
1798 if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
1799 DVA_GET_VDEV(&bp->blk_dva[1]))
1800 zb->zb_ditto_2_of_2_samevdev++;
1803 equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
1804 DVA_GET_VDEV(&bp->blk_dva[1])) +
1805 (DVA_GET_VDEV(&bp->blk_dva[0]) ==
1806 DVA_GET_VDEV(&bp->blk_dva[2])) +
1807 (DVA_GET_VDEV(&bp->blk_dva[1]) ==
1808 DVA_GET_VDEV(&bp->blk_dva[2]));
1810 zb->zb_ditto_2_of_3_samevdev++;
1811 else if (equal == 3)
1812 zb->zb_ditto_3_of_3_samevdev++;
1819 dsl_scan_scrub_done(zio_t *zio)
1821 spa_t *spa = zio->io_spa;
1823 zio_data_buf_free(zio->io_data, zio->io_size);
1825 mutex_enter(&spa->spa_scrub_lock);
1826 spa->spa_scrub_inflight--;
1827 cv_broadcast(&spa->spa_scrub_io_cv);
1829 if (zio->io_error && (zio->io_error != ECKSUM ||
1830 !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) {
1831 spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors++;
1833 mutex_exit(&spa->spa_scrub_lock);
1837 dsl_scan_scrub_cb(dsl_pool_t *dp,
1838 const blkptr_t *bp, const zbookmark_phys_t *zb)
1840 dsl_scan_t *scn = dp->dp_scan;
1841 size_t size = BP_GET_PSIZE(bp);
1842 spa_t *spa = dp->dp_spa;
1843 uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp);
1844 boolean_t needs_io = B_FALSE;
1845 int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
1849 if (phys_birth <= scn->scn_phys.scn_min_txg ||
1850 phys_birth >= scn->scn_phys.scn_max_txg)
1853 count_block(dp->dp_blkstats, bp);
1855 if (BP_IS_EMBEDDED(bp))
1858 ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn));
1859 if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) {
1860 zio_flags |= ZIO_FLAG_SCRUB;
1862 scan_delay = zfs_scrub_delay;
1864 ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER);
1865 zio_flags |= ZIO_FLAG_RESILVER;
1867 scan_delay = zfs_resilver_delay;
1870 /* If it's an intent log block, failure is expected. */
1871 if (zb->zb_level == ZB_ZIL_LEVEL)
1872 zio_flags |= ZIO_FLAG_SPECULATIVE;
1874 for (d = 0; d < BP_GET_NDVAS(bp); d++) {
1875 vdev_t *vd = vdev_lookup_top(spa,
1876 DVA_GET_VDEV(&bp->blk_dva[d]));
1879 * Keep track of how much data we've examined so that
1880 * zpool(1M) status can make useful progress reports.
1882 scn->scn_phys.scn_examined += DVA_GET_ASIZE(&bp->blk_dva[d]);
1883 spa->spa_scan_pass_exam += DVA_GET_ASIZE(&bp->blk_dva[d]);
1885 /* if it's a resilver, this may not be in the target range */
1887 if (DVA_GET_GANG(&bp->blk_dva[d])) {
1889 * Gang members may be spread across multiple
1890 * vdevs, so the best estimate we have is the
1891 * scrub range, which has already been checked.
1892 * XXX -- it would be better to change our
1893 * allocation policy to ensure that all
1894 * gang members reside on the same vdev.
1898 needs_io = vdev_dtl_contains(vd, DTL_PARTIAL,
1904 if (needs_io && !zfs_no_scrub_io) {
1905 vdev_t *rvd = spa->spa_root_vdev;
1906 uint64_t maxinflight = rvd->vdev_children * zfs_top_maxinflight;
1907 void *data = zio_data_buf_alloc(size);
1909 mutex_enter(&spa->spa_scrub_lock);
1910 while (spa->spa_scrub_inflight >= maxinflight)
1911 cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
1912 spa->spa_scrub_inflight++;
1913 mutex_exit(&spa->spa_scrub_lock);
1916 * If we're seeing recent (zfs_scan_idle) "important" I/Os
1917 * then throttle our workload to limit the impact of a scan.
1919 if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle)
1922 zio_nowait(zio_read(NULL, spa, bp, data, size,
1923 dsl_scan_scrub_done, NULL, ZIO_PRIORITY_SCRUB,
1927 /* do not relocate this block */
1932 dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
1934 spa_t *spa = dp->dp_spa;
1937 * Purge all vdev caches and probe all devices. We do this here
1938 * rather than in sync context because this requires a writer lock
1939 * on the spa_config lock, which we can't do from sync context. The
1940 * spa_scrub_reopen flag indicates that vdev_open() should not
1941 * attempt to start another scrub.
1943 spa_vdev_state_enter(spa, SCL_NONE);
1944 spa->spa_scrub_reopen = B_TRUE;
1945 vdev_reopen(spa->spa_root_vdev);
1946 spa->spa_scrub_reopen = B_FALSE;
1947 (void) spa_vdev_state_exit(spa, NULL, 0);
1949 return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
1950 dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_NONE));
1954 dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx)
1956 return (scn->scn_restart_txg != 0 &&
1957 scn->scn_restart_txg <= tx->tx_txg);
1960 #if defined(_KERNEL) && defined(HAVE_SPL)
1961 module_param(zfs_top_maxinflight, int, 0644);
1962 MODULE_PARM_DESC(zfs_top_maxinflight, "Max I/Os per top-level");
1964 module_param(zfs_resilver_delay, int, 0644);
1965 MODULE_PARM_DESC(zfs_resilver_delay, "Number of ticks to delay resilver");
1967 module_param(zfs_scrub_delay, int, 0644);
1968 MODULE_PARM_DESC(zfs_scrub_delay, "Number of ticks to delay scrub");
1970 module_param(zfs_scan_idle, int, 0644);
1971 MODULE_PARM_DESC(zfs_scan_idle, "Idle window in clock ticks");
1973 module_param(zfs_scan_min_time_ms, int, 0644);
1974 MODULE_PARM_DESC(zfs_scan_min_time_ms, "Min millisecs to scrub per txg");
1976 module_param(zfs_free_min_time_ms, int, 0644);
1977 MODULE_PARM_DESC(zfs_free_min_time_ms, "Min millisecs to free per txg");
1979 module_param(zfs_resilver_min_time_ms, int, 0644);
1980 MODULE_PARM_DESC(zfs_resilver_min_time_ms, "Min millisecs to resilver per txg");
1982 module_param(zfs_no_scrub_io, int, 0644);
1983 MODULE_PARM_DESC(zfs_no_scrub_io, "Set to disable scrub I/O");
1985 module_param(zfs_no_scrub_prefetch, int, 0644);
1986 MODULE_PARM_DESC(zfs_no_scrub_prefetch, "Set to disable scrub prefetching");
1988 module_param(zfs_free_max_blocks, ulong, 0644);
1989 MODULE_PARM_DESC(zfs_free_max_blocks, "Max number of blocks freed in one txg");
1991 module_param(zfs_free_bpobj_enabled, int, 0644);
1992 MODULE_PARM_DESC(zfs_free_bpobj_enabled, "Enable processing of the free_bpobj");