3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * CDDL HEADER START
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * The contents of this file are subject to the terms of the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Common Development and Distribution License (the "License").
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * You may not use this file except in compliance with the License.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * See the License for the specific language governing permissions
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * and limitations under the License.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * When distributing Covered Code, include this CDDL HEADER in each
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If applicable, add the following below this CDDL HEADER, with the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * fields enclosed by brackets "[]" replaced with your own identifying
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * information: Portions Copyright [yyyy] [name of copyright owner]
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * CDDL HEADER END
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
139510fb6efa97dbe5f5479594b308d940cab8d1George Wilson * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
8c04a1fa3f7d569d48fe9b5342d0bd4c533179b9Gary Mills * Copyright 2016 Gary Mills
7802d7bf98dec568dadf72286893b1fe5abd8602Matthew Ahrenstypedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *,
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensstatic void dsl_scan_cancel_sync(void *, dmu_tx_t *);
1825bc56e5a1f7ef6f0dc3137f3b35f5850c1100Nav Ravindranathstatic void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *);
1825bc56e5a1f7ef6f0dc3137f3b35f5850c1100Nav Ravindranathstatic boolean_t dsl_scan_restarting(dsl_scan_t *, dmu_tx_t *);
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilsonint zfs_top_maxinflight = 32; /* maximum I/Os per top-level */
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilsonint zfs_resilver_delay = 2; /* number of ticks to delay resilver */
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilsonint zfs_scrub_delay = 4; /* number of ticks to delay scrub */
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilsonint zfs_scan_idle = 50; /* idle window in clock ticks */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingint zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrensint zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingint zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingboolean_t zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrensboolean_t zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingenum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingint dsl_scan_delay_completion = B_FALSE; /* set to delay scan completion */
af3465da8fa420c4ec701e3e57704d537a6f755bMax Grossman/* max number of blocks to free in a single TXG */
139510fb6efa97dbe5f5479594b308d940cab8d1George Wilson * Enable/disable the processing of the free_bpobj object.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* the order has to match pool_scan_type */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP);
4a92375985c37d61406d66cd2b10ee642eb1f5e7George Wilson * It's possible that we're resuming a scan after a reboot so
4a92375985c37d61406d66cd2b10ee642eb1f5e7George Wilson * make sure that the scan_async_destroying flag is initialized
4a92375985c37d61406d66cd2b10ee642eb1f5e7George Wilson * appropriately.
4a92375985c37d61406d66cd2b10ee642eb1f5e7George Wilson scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (err == 0) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * There was an old-style scrub in progress. Restart a
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * new-style scrub from the beginning.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "restarting new-style scrub in txg %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Load the queue obj from the old location so that it
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * can be freed by dsl_scan_done().
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * A new-type scrub was in progress on an old
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * pool, and the pool was accessed by old
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * software. Restart from the beginning, since
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * the old software may have changed the pool in
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * the meantime.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "by old software; restarting in txg %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* rewrite all disk labels */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling spa_event_notify(spa, NULL, ESC_ZFS_RESILVER_START);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If this is an incremental scrub, limit the DDT scrub phase
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * to just the auto-ditto class (for correctness); the rest
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * of the scrub should go faster using top-down pruning.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* back to the generic stuff */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
4445fffbbb1ea25fd0e9ea68b9380dd7a6709025Matthew Ahrens spa_history_log_internal(spa, "scan setup", tx,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "func=%u mintxg=%llu maxtxg=%llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_bookmark",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_ddt_bookmark",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_ddt_class_max",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_queue",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_min_txg",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_max_txg",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_func",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "scrub_errors",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* Remove any remnants of an old-style scrub. */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling for (i = 0; old_names[i]; i++) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If we were "restarted" from a stopped state, don't bother
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * with anything else.
1825bc56e5a1f7ef6f0dc3137f3b35f5850c1100Nav Ravindranath spa_history_log_internal(spa, "scan aborted, restarting", tx,
1825bc56e5a1f7ef6f0dc3137f3b35f5850c1100Nav Ravindranath spa_history_log_internal(spa, "scan cancelled", tx,
1825bc56e5a1f7ef6f0dc3137f3b35f5850c1100Nav Ravindranath spa_history_log_internal(spa, "scan done", tx,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If the scrub/resilver completed, update all DTLs to
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * reflect this. Whether it succeeded or not, vacate
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * all temporary scrub DTLs.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ?
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We may have finished replacing a device.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Let the async thread assess this and handle the detach.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensdsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
7d46dc6ca63a6f3f0d51aa655bfcf10cf2405a9eMatthew Ahrens dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED));
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrensstatic void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb,
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingstatic void dsl_scan_visitdnode(dsl_scan_t *, dsl_dataset_t *ds,
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg;
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg));
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrensextern int zfs_vdev_async_write_active_min_dirty_percent;
7802d7bf98dec568dadf72286893b1fe5abd8602Matthew Ahrensdsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_phys_t *zb)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* we never skip user/group accounting objects */
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Siden if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark))
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* We only know how to resume from level-0 blocks. */
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * We pause if:
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * - we have scanned for the maximum time: an entire txg
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * timeout (default 5 sec)
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * - we have scanned for at least the minimum time (default 1 sec
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * for scrub, 3 sec for resilver), and either we have sufficient
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * dirty data that we are starting to write more quickly
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * (default 30%), or someone is explicitly waiting for this txg
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * to complete.
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * - the spa is shutting down because this pool is being exported
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens * or the machine is rebooting.
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens uint64_t elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens int dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max;
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens if (elapsed_nanosecs / NANOSEC >= zfs_txg_timeout ||
6f6a76adacda33b10633476dc6c5d66d7f17dd94Matthew Ahrens dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent)) ||
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
43466aae47bfcd2ad9bf501faec8e75c08095e4fMax Grossman if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * One block ("stubby") can be allocated a long time ago; we
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * want to visit that one because it has been allocated
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * (on-disk) even if it hasn't been claimed (even though for
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * scrub there's nothing to do to it).
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa))
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
43466aae47bfcd2ad9bf501faec8e75c08095e4fMax Grossman bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * birth can be < claim_txg if this record's txg is
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * already txg sync'ed (but this log block contains
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * other records that are not synced)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We only want to visit blocks that have been claimed but not yet
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * replayed (or, in read-only mode, blocks that *would* be claimed).
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp,
7adb730b589e553bf3b1ccfd9bae2df91c5c1061George Wilson arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg ||
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE))
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid);
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp,
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We never skip over user/group accounting objects (obj<0)
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Siden if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) &&
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If we already visited this bp & everything below (in
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * a prior txg sync), don't bother doing it again.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If we found the block we're trying to resume from, or
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * we went past it to a different object, zero it out to
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * indicate that it's OK to start checking for pausing
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 ||
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Return nonzero on i/o error.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Return new buf to write out in *bufp.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) {
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens dsl_scan_prefetch(scn, buf, cbp, zb->zb_objset,
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) {
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) {
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens for (i = 0, cdnp = buf->b_data; i < epb; i++, cdnp++) {
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We also always visit user/group accounting
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * objects, and never skip them, even if we are
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * pausing. This is necessary so that the space
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * deltas from this txg get integrated.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * The arguments are in this order because mdb can only print the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * first 5; we want them to be useful.
7802d7bf98dec568dadf72286893b1fe5abd8602Matthew Ahrensdsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb,
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* ASSERT(pbuf == NULL || arc_released(pbuf)); */
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
5f37736ac8f99922368294d745d3fefa22b49d11Matthew Ahrens if (dsl_scan_recurse(scn, ds, ostype, dnp, &bp_toread, zb, tx) != 0)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If dsl_scan_ddt() has aready visited this block, it will have
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * already done any translations or scrubbing, so don't call the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * callback again.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If this block is from the future (after cur_max_txg), then we
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * are doing this on behalf of a deleted snapshot, and we will
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * revisit the future block on the next pass of this dataset.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Don't scan it now unless we need to because something
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * under it was modified.
b4952e17e8858d3225793b28788278de9fe6038dGeorge Wilson if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * - scn_cur_{min,max}_txg stays the same.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * - Setting the flag is not really necessary if
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * scn_cur_max_txg == scn_max_txg, because there
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * is nothing after this snapshot that we care
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * about. However, we set it anyway and then
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * ignore it when we retraverse it in
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * dsl_scan_visitds().
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("destroying ds %llu; currently traversing; "
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "reset zb_objset to %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("destroying ds %llu; currently traversing; "
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "reset bookmark to -1,0,0,0",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We keep the same mintxg; it could be >
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * ds_creation_txg if the previous snapshot was
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * deleted too.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "replacing with %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("destroying ds %llu; in queue; removing",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * dsl_scan_sync() should be called after this, and should sync
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * out our changed state, but just to be safe, do it here.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx)
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "reset zb_objset to %llu",
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "replacing with %llu",
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "reset zb_objset to %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling } else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "reset zb_objset to %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* Both were there to begin with */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "replacing with %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) {
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "replacing with %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensenqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != eca->originobj)
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs while (dsl_dataset_phys(ds)->ds_prev_snap_obj != eca->originobj) {
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_txg, eca->tx) == 0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx)
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * This can happen if this snapshot was created after the
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * scan started, and we already completed a previous snapshot
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * that was created after the scan started. This snapshot
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * only references blocks with:
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * birth < our ds_creation_txg
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * cur_min_txg is no less than ds_creation_txg.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * We have already visited these blocks.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * birth > scn_max_txg
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * The scan requested not to visit these blocks.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * Subsequent snapshots (and clones) can reference our
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * blocks, or blocks with even higher birth times.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * Therefore we do not need to visit them either,
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * so we do not add them to the work queue.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * Note that checking for cur_min_txg >= cur_max_txg
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * is not sufficient, because in that case we may need to
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * visit subsequent snapshots. This happens when min_txg > 0,
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * which raises cur_min_txg. In this case we will visit
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * this dataset but skip all of its blocks, because the
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * rootbp's birth time is < cur_min_txg. Then we will
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens * add the next snapshots/clones to the work queue.
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens char *dsname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because "
38d61036746e2273cc18f6698392e1e29f87d1bfMatthew Ahrens "cur_min_txg (%llu) >= max_txg (%llu)",
6e0cbcaa0c6f2bc34634a4cc17b099f9ecef03d1Matthew Ahrens * Only the ZIL in the head (non-snapshot) is valid. Even though
6e0cbcaa0c6f2bc34634a4cc17b099f9ecef03d1Matthew Ahrens * snapshots can have ZIL block pointers (which may be the same
6e0cbcaa0c6f2bc34634a4cc17b099f9ecef03d1Matthew Ahrens * BP as in the head), they must be ignored. So we traverse the
6e0cbcaa0c6f2bc34634a4cc17b099f9ecef03d1Matthew Ahrens * ZIL here, rather than in scan_recurse(), because the regular
6e0cbcaa0c6f2bc34634a4cc17b099f9ecef03d1Matthew Ahrens * snapshot block-sharing rules don't apply to it.
bc9014e6a81272073b9854d9f65dd59e18d18c35Justin Gibbs if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !ds->ds_is_snapshot)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Iterate over the bps in this ds.
c166b69d29138aed7a415fe7cef698e54c6ae945Paul Dagnelie rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx);
9adfa60d484ce2435f5af77cc99dcd4e692b6660Matthew Ahrens char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "pausing=%u",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We've finished this pass over this dataset.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If we did not completely visit this dataset, do another pass.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Add descendent datasets to work queue.
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_creation_txg, tx) == 0);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_num_children > 1) {
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * A bug in a previous version of the code could
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * cause upgrade_clones_cb() to not set
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * ds_next_snap_obj when it should, leading to a
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * missing entry. Therefore we can only use the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * next_clones_obj when its count is correct.
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_next_clones_obj, &count);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (err == 0 &&
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs count == dsl_dataset_phys(ds)->ds_num_children - 1)
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_creation_txg, tx));
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrensenqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If this is a clone, we don't need to worry about it for now.
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx) == 0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Scrub/dedup interaction.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If there are N references to a deduped block, we don't want to scrub it
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * N times -- ideally, we should scrub it exactly once.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * We leverage the fact that the dde's replication class (enum ddt_class)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * To prevent excess scrubbing, the scrub begins by walking the DDT
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * to find all blocks with refcnt > 1, and scrubs each of these once.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Since there are two replication classes which contain blocks with
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * There would be nothing more to say if a block's refcnt couldn't change
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * during a scrub, but of course it can so we must account for changes
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * in a block's replication class.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Here's an example of what can occur:
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * when visited during the top-down scrub phase, it will be scrubbed twice.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * This negates our scrub optimization, but is otherwise harmless.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * on each visit during the top-down scrub phase, it will never be scrubbed.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * while a scrub is in progress, it scrubs the block right then.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* There should be no pending changes to the dedup table */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum];
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)n, (int)scn->scn_phys.scn_ddt_class_max,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling/* ARGSUSED */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingdsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum,
b4952e17e8858d3225793b28788278de9fe6038dGeorge Wilson ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* First do the MOS & ORIGIN */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If we were paused, continue from here. Note if the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * ds we were paused on was deleted, the zb_objset may
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * be -1, so we will skip this and find a new objset
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * In case we were paused right at the end of the ds, zero the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * bookmark so we don't think that we're still trying to resume.
7802d7bf98dec568dadf72286893b1fe5abd8602Matthew Ahrens bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* keep pulling things out of the zap-object-as-queue */
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* Set up min/max txg */
b420f3adeb349714478d1a7813d2c0e069d41555Richard Lowe VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds);
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Sidendsl_scan_free_should_pause(dsl_scan_t *scn)
af3465da8fa420c4ec701e3e57704d537a6f755bMax Grossman if (scn->scn_visited_this_txg >= zfs_free_max_blocks)
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Siden return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
0689f76c08c5e553ff25ac43a852b56c430bb61eAdam Leventhal (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms &&
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Sidendsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
ad135b5d644628e791c3188a6ecbd9c257961ef8Christopher Siden (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) {
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa,
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
4a92375985c37d61406d66cd2b10ee642eb1f5e7George Wilson if (scn->scn_phys.scn_state == DSS_SCANNING ||
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens (scn->scn_async_destroying && !scn->scn_async_stalled))
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens (void) bpobj_space(&scn->scn_dp->dp_free_bpobj,
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens return (used != 0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Check for scn_restart_txg before checking spa_load_state, so
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * that we can restart an old-style scan while the pool is being
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * imported (see dsl_scan_init).
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
a443cc80c742af740aa82130db840f02b4389365George Wilson * Only process scans in sync pass 1.
a443cc80c742af740aa82130db840f02b4389365George Wilson * If the spa is shutting down, then stop scanning. This will
a443cc80c742af740aa82130db840f02b4389365George Wilson * ensure that the scan does not dirty any new data during the
a443cc80c742af740aa82130db840f02b4389365George Wilson * shutdown phase.
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * If the scan is inactive due to a stalled async destroy, try again.
a443cc80c742af740aa82130db840f02b4389365George Wilson if (!scn->scn_async_stalled && !dsl_scan_active(scn))
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * First process the async destroys. If we pause, don't do
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * any scrubbing or resilvering. This ensures that there are no
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * async destroys while we are scanning, so the scan code doesn't
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * have to worry about traversing it. It is also faster to free the
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * blocks than to scrub them.
139510fb6efa97dbe5f5479594b308d940cab8d1George Wilson spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens zfs_panic_recover("error %u from bpobj_iterate()", err);
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx);
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) {
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens /* finished; deactivate async destroy feature */
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx);
231aab857f14a3e7a0ed5f2879399c3cd6ae92eaMatthew Ahrens * If we didn't make progress, mark the async
231aab857f14a3e7a0ed5f2879399c3cd6ae92eaMatthew Ahrens * destroy as stalled, so that we will not initiate
231aab857f14a3e7a0ed5f2879399c3cd6ae92eaMatthew Ahrens * a spa_sync() on its behalf. Note that we only
231aab857f14a3e7a0ed5f2879399c3cd6ae92eaMatthew Ahrens * check this if we are not finished, because if the
231aab857f14a3e7a0ed5f2879399c3cd6ae92eaMatthew Ahrens * bptree had no blocks for us to visit, we can
231aab857f14a3e7a0ed5f2879399c3cd6ae92eaMatthew Ahrens * finish without "making progress".
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens zfs_dbgmsg("freed %llu blocks in %llums from "
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens "free_bpobj/bptree txg %llu; err=%u",
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * Write out changes to the DDT that may be required as a
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * result of the blocks freed. This ensures that the DDT
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * is clean when a scrub/resilver runs.
8c04a1fa3f7d569d48fe9b5342d0bd4c533179b9Gary Mills if (dp->dp_free_dir != NULL && !scn->scn_async_destroying &&
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 ||
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 ||
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) {
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * We have finished background destroying, but there is still
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * some space left in the dp_free_dir. Transfer this leaked
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens * space to the dp_leak_dir.
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens (void) dsl_dir_create_sync(dp, dp->dp_root_dir,
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx);
7fd05ac4dec0c343d2f68f310d3718b715ecfbafMatthew Ahrens dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes,
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx);
8c04a1fa3f7d569d48fe9b5342d0bd4c533179b9Gary Mills if (dp->dp_free_dir != NULL && !scn->scn_async_destroying) {
5d7b4d438c4a51eccc95e77a83a437b4d48380ebMatthew Ahrens /* finished; verify that space accounting went to zero */
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes);
c1379625401dfbe1c39b79136dd384a571d47fdeJustin T. Gibbs ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes);
b4952e17e8858d3225793b28788278de9fe6038dGeorge Wilson /* finished with scan. */
b4952e17e8858d3225793b28788278de9fe6038dGeorge Wilson zfs_dbgmsg("txg %llu scan complete", tx->tx_txg);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling "ddt bm=%llu/%llu/%llu/%llx",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
cde58dbc6a23d4d38db7c8866312be83221c765fMatthew Ahrens scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
0689f76c08c5e553ff25ac43a852b56c430bb61eAdam Leventhal (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time));
b4952e17e8858d3225793b28788278de9fe6038dGeorge Wilson zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * This will start a new scan, or restart an existing one.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling if (txg == 0) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING &&
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * scrub consumers
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Lingcount_block(zfs_all_blkstats_t *zab, const blkptr_t *bp)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * If we resume after a reboot, zab will be NULL; don't record
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * incomplete stats in that case.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling for (i = 0; i < 4; i++) {
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS;
7802d7bf98dec568dadf72286893b1fe5abd8602Matthew Ahrens const blkptr_t *bp, const zbookmark_phys_t *zb)
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
d5285cae913f4e01ffa0e6693a6d8ef1fbea30baGeorge Wilson ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* If it's an intent log block, failure is expected. */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Keep track of how much data we've examined so that
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * zpool(1M) status can make useful progress reports.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling scn->scn_phys.scn_examined += DVA_GET_ASIZE(&bp->blk_dva[d]);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling spa->spa_scan_pass_exam += DVA_GET_ASIZE(&bp->blk_dva[d]);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* if it's a resilver, this may not be in the target range */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Gang members may be spread across multiple
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * vdevs, so the best estimate we have is the
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * scrub range, which has already been checked.
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * XXX -- it would be better to change our
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * allocation policy to ensure that all
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * gang members reside on the same vdev.
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson uint64_t maxinflight = rvd->vdev_children * zfs_top_maxinflight;
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson while (spa->spa_scrub_inflight >= maxinflight)
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson * If we're seeing recent (zfs_scan_idle) "important" I/Os
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson * then throttle our workload to limit the impact of a scan.
44ecc5327ab4ce0750dcca2a17e05566bf2812e2George Wilson if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle)
69962b5647e4a8b9b14998733b765925381b727eMatthew Ahrens dsl_scan_scrub_done, NULL, ZIO_PRIORITY_SCRUB,
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling /* do not relocate this block */
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling return (0);
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * Purge all vdev caches and probe all devices. We do this here
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * rather than in sync context because this requires a writer lock
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * on the spa_config lock, which we can't do from sync context. The
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * spa_scrub_reopen flag indicates that vdev_open() should not
3f9d6ad73e45c6823b409f93b0c8d4f62861d2d5Lin Ling * attempt to start another scrub.
3b2aab18808792cbd248a12f1edf139b89833c13Matthew Ahrens return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
7d46dc6ca63a6f3f0d51aa655bfcf10cf2405a9eMatthew Ahrens dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_NONE));