* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
- * or http://www.opensolaris.org/os/licensing.
+ * or https://opensource.org/licenses/CDDL-1.0.
* See the License for the specific language governing permissions
* and limitations under the License.
*
*/
/*
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2021 by Delphix. All rights reserved.
* Copyright 2016 Gary Mills
- * Copyright (c) 2017 Datto Inc.
- * Copyright 2017 Joyent, Inc.
+ * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
+ * Copyright (c) 2015, Nexenta Systems, Inc. All rights reserved.
+ * Copyright 2019 Joyent, Inc.
*/
#include <sys/dsl_scan.h>
#include <sys/dmu_tx.h>
#include <sys/dmu_objset.h>
#include <sys/arc.h>
+#include <sys/arc_impl.h>
#include <sys/zap.h>
#include <sys/zio.h>
#include <sys/zfs_context.h>
#include <sys/vdev_impl.h>
#include <sys/zil_impl.h>
#include <sys/zio_checksum.h>
+#include <sys/brt.h>
#include <sys/ddt.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
static void scan_ds_queue_insert(dsl_scan_t *scn, uint64_t dsobj, uint64_t txg);
static void scan_ds_queue_remove(dsl_scan_t *scn, uint64_t dsobj);
static void scan_ds_queue_sync(dsl_scan_t *scn, dmu_tx_t *tx);
-static uint64_t dsl_scan_count_leaves(vdev_t *vd);
+static uint64_t dsl_scan_count_data_disks(spa_t *spa);
-extern int zfs_vdev_async_write_active_min_dirty_percent;
+extern uint_t zfs_vdev_async_write_active_min_dirty_percent;
+static int zfs_scan_blkstats = 0;
+
+/*
+ * 'zpool status' uses bytes processed per pass to report throughput and
+ * estimate time remaining. We define a pass to start when the scanning
+ * phase completes for a sequential resilver. Optionally, this value
+ * may be used to reset the pass statistics every N txgs to provide an
+ * estimated completion time based on currently observed performance.
+ */
+static uint_t zfs_scan_report_txgs = 0;
/*
* By default zfs will check to ensure it is not over the hard memory
* this value can be set to 1 to enable checking before scanning each
* block.
*/
-int zfs_scan_strict_mem_lim = B_FALSE;
+static int zfs_scan_strict_mem_lim = B_FALSE;
/*
* Maximum number of parallelly executed bytes per leaf vdev. We attempt
* overload the drives with I/O, since that is protected by
* zfs_vdev_scrub_max_active.
*/
-unsigned long zfs_scan_vdev_limit = 4 << 20;
+static uint64_t zfs_scan_vdev_limit = 16 << 20;
+
+static uint_t zfs_scan_issue_strategy = 0;
-int zfs_scan_issue_strategy = 0;
-int zfs_scan_legacy = B_FALSE; /* don't queue & sort zios, go direct */
-unsigned long zfs_scan_max_ext_gap = 2 << 20; /* in bytes */
+/* don't queue & sort zios, go direct */
+static int zfs_scan_legacy = B_FALSE;
+static uint64_t zfs_scan_max_ext_gap = 2 << 20; /* in bytes */
/*
* fill_weight is non-tunable at runtime, so we copy it at module init from
* zfs_scan_fill_weight. Runtime adjustments to zfs_scan_fill_weight would
* break queue sorting.
*/
-int zfs_scan_fill_weight = 3;
+static uint_t zfs_scan_fill_weight = 3;
static uint64_t fill_weight;
/* See dsl_scan_should_clear() for details on the memory limit tunables */
-uint64_t zfs_scan_mem_lim_min = 16 << 20; /* bytes */
-uint64_t zfs_scan_mem_lim_soft_max = 128 << 20; /* bytes */
-int zfs_scan_mem_lim_fact = 20; /* fraction of physmem */
-int zfs_scan_mem_lim_soft_fact = 20; /* fraction of mem lim above */
-
-int zfs_scrub_min_time_ms = 1000; /* min millisecs to scrub per txg */
-int zfs_obsolete_min_time_ms = 500; /* min millisecs to obsolete per txg */
-int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */
-int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */
-int zfs_scan_checkpoint_intval = 7200; /* in seconds */
+static const uint64_t zfs_scan_mem_lim_min = 16 << 20; /* bytes */
+static const uint64_t zfs_scan_mem_lim_soft_max = 128 << 20; /* bytes */
+
+
+/* fraction of physmem */
+static uint_t zfs_scan_mem_lim_fact = 20;
+
+/* fraction of mem lim above */
+static uint_t zfs_scan_mem_lim_soft_fact = 20;
+
+/* minimum milliseconds to scrub per txg */
+static uint_t zfs_scrub_min_time_ms = 1000;
+
+/* minimum milliseconds to obsolete per txg */
+static uint_t zfs_obsolete_min_time_ms = 500;
+
+/* minimum milliseconds to free per txg */
+static uint_t zfs_free_min_time_ms = 1000;
+
+/* minimum milliseconds to resilver per txg */
+static uint_t zfs_resilver_min_time_ms = 3000;
+
+static uint_t zfs_scan_checkpoint_intval = 7200; /* in seconds */
int zfs_scan_suspend_progress = 0; /* set to prevent scans from progressing */
-int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
-int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
-enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
+static int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
+static int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */
+static const enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
/* max number of blocks to free in a single TXG */
-unsigned long zfs_async_block_max_blocks = 100000;
+static uint64_t zfs_async_block_max_blocks = UINT64_MAX;
+/* max number of dedup blocks to free in a single TXG */
+static uint64_t zfs_max_async_dedup_frees = 100000;
-int zfs_resilver_disable_defer = 0; /* set to disable resilver deferring */
+/* set to disable resilver deferring */
+static int zfs_resilver_disable_defer = B_FALSE;
/*
* We wait a few txgs after importing a pool to begin scanning so that
/*
* Enable/disable the processing of the free_bpobj object.
*/
-int zfs_free_bpobj_enabled = 1;
+static int zfs_free_bpobj_enabled = 1;
/* the order has to match pool_scan_type */
static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
/*
* This controls what conditions are placed on dsl_scan_sync_state():
- * SYNC_OPTIONAL) write out scn_phys iff scn_bytes_pending == 0
- * SYNC_MANDATORY) write out scn_phys always. scn_bytes_pending must be 0.
- * SYNC_CACHED) if scn_bytes_pending == 0, write out scn_phys. Otherwise
+ * SYNC_OPTIONAL) write out scn_phys iff scn_queues_pending == 0
+ * SYNC_MANDATORY) write out scn_phys always. scn_queues_pending must be 0.
+ * SYNC_CACHED) if scn_queues_pending == 0, write out scn_phys. Otherwise
* write out the scn_phys_cached version.
* See dsl_scan_sync_state for details.
*/
* event of an error. This array must go at the end of the
* struct to allow this for the variable number of elements.
*/
- dva_t sio_dva[0];
+ dva_t sio_dva[];
} scan_io_t;
#define SIO_SET_OFFSET(sio, x) DVA_SET_OFFSET(&(sio)->sio_dva[0], x)
struct dsl_scan_io_queue {
dsl_scan_t *q_scn; /* associated dsl_scan_t */
vdev_t *q_vd; /* top-level vdev that this queue represents */
+ zio_t *q_zio; /* scn_zio_root child for waiting on IO */
/* trees used for sorting I/Os and extents of I/Os */
range_tree_t *q_exts_by_addr;
- avl_tree_t q_exts_by_size;
+ zfs_btree_t q_exts_by_size;
avl_tree_t q_sios_by_addr;
uint64_t q_sio_memused;
+ uint64_t q_last_ext_addr;
/* members for zio rate limiting */
uint64_t q_maxinflight_bytes;
for (int i = 0; i < SPA_DVAS_PER_BP; i++) {
char name[36];
- (void) sprintf(name, "sio_cache_%d", i);
+ (void) snprintf(name, sizeof (name), "sio_cache_%d", i);
sio_cache[i] = kmem_cache_create(name,
(sizeof (scan_io_t) + ((i + 1) * sizeof (dva_t))),
0, NULL, NULL, NULL, NULL, NULL, 0);
static inline void
sio2bp(const scan_io_t *sio, blkptr_t *bp)
{
- bzero(bp, sizeof (*bp));
+ memset(bp, 0, sizeof (*bp));
bp->blk_prop = sio->sio_blk_prop;
bp->blk_phys_birth = sio->sio_phys_birth;
bp->blk_birth = sio->sio_birth;
ASSERT3U(sio->sio_nr_dvas, >, 0);
ASSERT3U(sio->sio_nr_dvas, <=, SPA_DVAS_PER_BP);
- bcopy(sio->sio_dva, bp->blk_dva, sio->sio_nr_dvas * sizeof (dva_t));
+ memcpy(bp->blk_dva, sio->sio_dva, sio->sio_nr_dvas * sizeof (dva_t));
}
static inline void
/*
* Calculate the max number of in-flight bytes for pool-wide
- * scanning operations (minimum 1MB). Limits for the issuing
- * phase are done per top-level vdev and are handled separately.
+ * scanning operations (minimum 1MB, maximum 1/4 of arc_c_max).
+ * Limits for the issuing phase are done per top-level vdev and
+ * are handled separately.
*/
- scn->scn_maxinflight_bytes = MAX(zfs_scan_vdev_limit *
- dsl_scan_count_leaves(spa->spa_root_vdev), 1ULL << 20);
+ scn->scn_maxinflight_bytes = MIN(arc_c_max / 4, MAX(1ULL << 20,
+ zfs_scan_vdev_limit * dsl_scan_count_data_disks(spa)));
avl_create(&scn->scn_queue, scan_ds_queue_compare, sizeof (scan_ds_t),
offsetof(scan_ds_t, sds_node));
* new-style scrub from the beginning.
*/
scn->scn_restart_txg = txg;
- zfs_dbgmsg("old-style scrub was in progress; "
+ zfs_dbgmsg("old-style scrub was in progress for %s; "
"restarting new-style scrub in txg %llu",
+ spa->spa_name,
(longlong_t)scn->scn_restart_txg);
/*
return (EOVERFLOW);
}
- bcopy(zaptmp, &scn->scn_phys,
+ memcpy(&scn->scn_phys, zaptmp,
SCAN_PHYS_NUMINTS * sizeof (uint64_t));
scn->scn_phys.scn_flags = overflow;
* the meantime.
*/
scn->scn_restart_txg = txg;
- zfs_dbgmsg("new-style scrub was modified "
+ zfs_dbgmsg("new-style scrub for %s was modified "
"by old software; restarting in txg %llu",
+ spa->spa_name,
(longlong_t)scn->scn_restart_txg);
+ } else if (dsl_scan_resilvering(dp)) {
+ /*
+ * If a resilver is in progress and there are already
+ * errors, restart it instead of finishing this scan and
+ * then restarting it. If there haven't been any errors
+ * then remember that the incore DTL is valid.
+ */
+ if (scn->scn_phys.scn_errors > 0) {
+ scn->scn_restart_txg = txg;
+ zfs_dbgmsg("resilver can't excise DTL_MISSING "
+ "when finished; restarting on %s in txg "
+ "%llu",
+ spa->spa_name,
+ (u_longlong_t)scn->scn_restart_txg);
+ } else {
+ /* it's safe to excise DTL when finished */
+ spa->spa_scrub_started = B_TRUE;
+ }
}
}
- bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys));
+ memcpy(&scn->scn_phys_cached, &scn->scn_phys, sizeof (scn->scn_phys));
/* reload the queue into the in-core state */
if (scn->scn_phys.scn_queue_obj != 0) {
}
spa_scan_stat_init(spa);
+ vdev_scan_stat_init(spa->spa_root_vdev);
+
return (0);
}
scn->scn_restart_txg <= tx->tx_txg);
}
+boolean_t
+dsl_scan_resilver_scheduled(dsl_pool_t *dp)
+{
+ return ((dp->dp_scan && dp->dp_scan->scn_restart_txg != 0) ||
+ (spa_async_tasks(dp->dp_spa) & SPA_ASYNC_RESILVER));
+}
+
boolean_t
dsl_scan_scrubbing(const dsl_pool_t *dp)
{
* Because we can be running in the block sorting algorithm, we do not always
* want to write out the record, only when it is "safe" to do so. This safety
* condition is achieved by making sure that the sorting queues are empty
- * (scn_bytes_pending == 0). When this condition is not true, the sync'd state
+ * (scn_queues_pending == 0). When this condition is not true, the sync'd state
* is inconsistent with how much actual scanning progress has been made. The
* kind of sync to be performed is specified by the sync_type argument. If the
* sync is optional, we only sync if the queues are empty. If the sync is
int i;
spa_t *spa = scn->scn_dp->dp_spa;
- ASSERT(sync_type != SYNC_MANDATORY || scn->scn_bytes_pending == 0);
- if (scn->scn_bytes_pending == 0) {
+ ASSERT(sync_type != SYNC_MANDATORY || scn->scn_queues_pending == 0);
+ if (scn->scn_queues_pending == 0) {
for (i = 0; i < spa->spa_root_vdev->vdev_children; i++) {
vdev_t *vd = spa->spa_root_vdev->vdev_child[i];
dsl_scan_io_queue_t *q = vd->vdev_scan_io_queue;
mutex_enter(&vd->vdev_scan_io_queue_lock);
ASSERT3P(avl_first(&q->q_sios_by_addr), ==, NULL);
- ASSERT3P(avl_first(&q->q_exts_by_size), ==, NULL);
+ ASSERT3P(zfs_btree_first(&q->q_exts_by_size, NULL), ==,
+ NULL);
ASSERT3P(range_tree_first(q->q_exts_by_addr), ==, NULL);
mutex_exit(&vd->vdev_scan_io_queue_lock);
}
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys, tx));
- bcopy(&scn->scn_phys, &scn->scn_phys_cached,
+ memcpy(&scn->scn_phys_cached, &scn->scn_phys,
sizeof (scn->scn_phys));
if (scn->scn_checkpointing)
- zfs_dbgmsg("finish scan checkpoint");
+ zfs_dbgmsg("finish scan checkpoint for %s",
+ spa->spa_name);
scn->scn_checkpointing = B_FALSE;
scn->scn_last_checkpoint = ddi_get_lbolt();
}
}
-/* ARGSUSED */
-static int
+int
dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
{
+ (void) arg;
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
+ vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev;
- if (dsl_scan_is_running(scn))
+ if (dsl_scan_is_running(scn) || vdev_rebuild_active(rvd))
return (SET_ERROR(EBUSY));
return (0);
}
-static void
+void
dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
ASSERT(!dsl_scan_is_running(scn));
ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
- bzero(&scn->scn_phys, sizeof (scn->scn_phys));
+ memset(&scn->scn_phys, 0, sizeof (scn->scn_phys));
scn->scn_phys.scn_func = *funcp;
scn->scn_phys.scn_state = DSS_SCANNING;
scn->scn_phys.scn_min_txg = 0;
scn->scn_last_checkpoint = 0;
scn->scn_checkpointing = B_FALSE;
spa_scan_stat_init(spa);
+ vdev_scan_stat_init(spa->spa_root_vdev);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
if (vdev_resilver_needed(spa->spa_root_vdev,
&scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
- spa_event_notify(spa, NULL, NULL,
+ nvlist_t *aux = fnvlist_alloc();
+ fnvlist_add_string(aux, ZFS_EV_RESILVER_TYPE,
+ "healing");
+ spa_event_notify(spa, NULL, aux,
ESC_ZFS_RESILVER_START);
+ nvlist_free(aux);
} else {
spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_START);
}
if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
+ /*
+ * When starting a resilver clear any existing rebuild state.
+ * This is required to prevent stale rebuild status from
+ * being reported when a rebuild is run, then a resilver and
+ * finally a scrub. In which case only the scrub status
+ * should be reported by 'zpool status'.
+ */
+ if (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) {
+ vdev_t *rvd = spa->spa_root_vdev;
+ for (uint64_t i = 0; i < rvd->vdev_children; i++) {
+ vdev_t *vd = rvd->vdev_child[i];
+ vdev_rebuild_clear_sync(
+ (void *)(uintptr_t)vd->vdev_id, tx);
+ }
+ }
}
/* back to the generic stuff */
- if (dp->dp_blkstats == NULL) {
- dp->dp_blkstats =
- vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
- mutex_init(&dp->dp_blkstats->zab_lock, NULL,
- MUTEX_DEFAULT, NULL);
+ if (zfs_scan_blkstats) {
+ if (dp->dp_blkstats == NULL) {
+ dp->dp_blkstats =
+ vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
+ }
+ memset(&dp->dp_blkstats->zab_type, 0,
+ sizeof (dp->dp_blkstats->zab_type));
+ } else {
+ if (dp->dp_blkstats) {
+ vmem_free(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
+ dp->dp_blkstats = NULL;
+ }
}
- bzero(&dp->dp_blkstats->zab_type, sizeof (dp->dp_blkstats->zab_type));
if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
ot = DMU_OT_ZAP_OTHER;
scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
- bcopy(&scn->scn_phys, &scn->scn_phys_cached, sizeof (scn->scn_phys));
+ memcpy(&scn->scn_phys_cached, &scn->scn_phys, sizeof (scn->scn_phys));
dsl_scan_sync_state(scn, tx, SYNC_MANDATORY);
spa_history_log_internal(spa, "scan setup", tx,
"func=%u mintxg=%llu maxtxg=%llu",
- *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
+ *funcp, (u_longlong_t)scn->scn_phys.scn_min_txg,
+ (u_longlong_t)scn->scn_phys.scn_max_txg);
}
/*
(void) spa_vdev_state_exit(spa, NULL, 0);
if (func == POOL_SCAN_RESILVER) {
- dsl_resilver_restart(spa->spa_dsl_pool, 0);
+ dsl_scan_restart_resilver(spa->spa_dsl_pool, 0);
return (0);
}
POOL_SCRUB_NORMAL);
if (err == 0) {
spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_RESUME);
- return (ECANCELED);
+ return (SET_ERROR(ECANCELED));
}
return (SET_ERROR(err));
dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_EXTRA_RESERVED));
}
-/*
- * Sets the resilver defer flag to B_FALSE on all leaf devs under vd. Returns
- * B_TRUE if we have devices that need to be resilvered and are available to
- * accept resilver I/Os.
- */
-static boolean_t
-dsl_scan_clear_deferred(vdev_t *vd, dmu_tx_t *tx)
-{
- boolean_t resilver_needed = B_FALSE;
- spa_t *spa = vd->vdev_spa;
-
- for (int c = 0; c < vd->vdev_children; c++) {
- resilver_needed |=
- dsl_scan_clear_deferred(vd->vdev_child[c], tx);
- }
-
- if (vd == spa->spa_root_vdev &&
- spa_feature_is_active(spa, SPA_FEATURE_RESILVER_DEFER)) {
- spa_feature_decr(spa, SPA_FEATURE_RESILVER_DEFER, tx);
- vdev_config_dirty(vd);
- spa->spa_resilver_deferred = B_FALSE;
- return (resilver_needed);
- }
-
- if (!vdev_is_concrete(vd) || vd->vdev_aux ||
- !vd->vdev_ops->vdev_op_leaf)
- return (resilver_needed);
-
- if (vd->vdev_resilver_deferred)
- vd->vdev_resilver_deferred = B_FALSE;
-
- return (!vdev_is_dead(vd) && !vd->vdev_offline &&
- vdev_resilver_needed(vd, NULL, NULL));
-}
-
-/* ARGSUSED */
static void
dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
{
scn->scn_phys.scn_state = complete ? DSS_FINISHED : DSS_CANCELED;
+ spa_notify_waiters(spa);
+
if (dsl_scan_restarting(scn, tx))
spa_history_log_internal(spa, "scan aborted, restarting", tx,
- "errors=%llu", spa_get_errlog_size(spa));
+ "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa));
else if (!complete)
spa_history_log_internal(spa, "scan cancelled", tx,
- "errors=%llu", spa_get_errlog_size(spa));
+ "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa));
else
spa_history_log_internal(spa, "scan done", tx,
- "errors=%llu", spa_get_errlog_size(spa));
+ "errors=%llu", (u_longlong_t)spa_approx_errlog_size(spa));
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
- spa->spa_scrub_started = B_FALSE;
spa->spa_scrub_active = B_FALSE;
/*
if (complete &&
!spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
- scn->scn_phys.scn_max_txg, B_TRUE);
-
- spa_event_notify(spa, NULL, NULL,
- scn->scn_phys.scn_min_txg ?
- ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
+ scn->scn_phys.scn_max_txg, B_TRUE, B_FALSE);
+
+ if (scn->scn_phys.scn_min_txg) {
+ nvlist_t *aux = fnvlist_alloc();
+ fnvlist_add_string(aux, ZFS_EV_RESILVER_TYPE,
+ "healing");
+ spa_event_notify(spa, NULL, aux,
+ ESC_ZFS_RESILVER_FINISH);
+ nvlist_free(aux);
+ } else {
+ spa_event_notify(spa, NULL, NULL,
+ ESC_ZFS_SCRUB_FINISH);
+ }
} else {
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
- 0, B_TRUE);
+ 0, B_TRUE, B_FALSE);
}
spa_errlog_rotate(spa);
+ /*
+ * Don't clear flag until after vdev_dtl_reassess to ensure that
+ * DTL_MISSING will get updated when possible.
+ */
+ spa->spa_scrub_started = B_FALSE;
+
/*
* We may have finished replacing a device.
* Let the async thread assess this and handle the detach.
spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
/*
- * Clear any deferred_resilver flags in the config.
+ * Clear any resilver_deferred flags in the config.
* If there are drives that need resilvering, kick
* off an asynchronous request to start resilver.
- * dsl_scan_clear_deferred() may update the config
+ * vdev_clear_resilver_deferred() may update the config
* before the resilver can restart. In the event of
* a crash during this period, the spa loading code
* will find the drives that need to be resilvered
- * when the machine reboots and start the resilver then.
+ * and start the resilver then.
*/
- boolean_t resilver_needed =
- dsl_scan_clear_deferred(spa->spa_root_vdev, tx);
- if (resilver_needed) {
+ if (spa_feature_is_enabled(spa, SPA_FEATURE_RESILVER_DEFER) &&
+ vdev_clear_resilver_deferred(spa->spa_root_vdev, tx)) {
spa_history_log_internal(spa,
- "starting deferred resilver", tx,
- "errors=%llu", spa_get_errlog_size(spa));
+ "starting deferred resilver", tx, "errors=%llu",
+ (u_longlong_t)spa_approx_errlog_size(spa));
spa_async_request(spa, SPA_ASYNC_RESILVER);
}
+
+ /* Clear recent error events (i.e. duplicate events tracking) */
+ if (complete)
+ zfs_ereport_clear(spa, NULL);
}
scn->scn_phys.scn_end_time = gethrestime_sec();
ASSERT(!dsl_scan_is_running(scn));
}
-/* ARGSUSED */
static int
dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
{
+ (void) arg;
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (!dsl_scan_is_running(scn))
return (0);
}
-/* ARGSUSED */
static void
dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
{
+ (void) arg;
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
dsl_scan_done(scn, B_FALSE, tx);
scn->scn_phys_cached.scn_flags |= DSF_SCRUB_PAUSED;
dsl_scan_sync_state(scn, tx, SYNC_CACHED);
spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_PAUSED);
+ spa_notify_waiters(spa);
} else {
ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL);
if (dsl_scan_is_paused_scrub(scn)) {
/* start a new scan, or restart an existing one. */
void
-dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg)
+dsl_scan_restart_resilver(dsl_pool_t *dp, uint64_t txg)
{
if (txg == 0) {
dmu_tx_t *tx;
} else {
dp->dp_scan->scn_restart_txg = txg;
}
- zfs_dbgmsg("restarting resilver txg=%llu", (longlong_t)txg);
+ zfs_dbgmsg("restarting resilver for %s at txg=%llu",
+ dp->dp_spa->spa_name, (longlong_t)txg);
}
void
dmu_object_type_t ot = (spa_version(spa) >= SPA_VERSION_DSL_SCRUB) ?
DMU_OT_SCAN_QUEUE : DMU_OT_ZAP_OTHER;
- ASSERT0(scn->scn_bytes_pending);
+ ASSERT0(scn->scn_queues_pending);
ASSERT(scn->scn_phys.scn_queue_obj != 0);
VERIFY0(dmu_object_free(dp->dp_meta_objset,
static boolean_t
dsl_scan_should_clear(dsl_scan_t *scn)
{
+ spa_t *spa = scn->scn_dp->dp_spa;
vdev_t *rvd = scn->scn_dp->dp_spa->spa_root_vdev;
- uint64_t mlim_hard, mlim_soft, mused;
- uint64_t alloc = metaslab_class_get_alloc(spa_normal_class(
- scn->scn_dp->dp_spa));
+ uint64_t alloc, mlim_hard, mlim_soft, mused;
+
+ alloc = metaslab_class_get_alloc(spa_normal_class(spa));
+ alloc += metaslab_class_get_alloc(spa_special_class(spa));
+ alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
mlim_hard = MAX((physmem / zfs_scan_mem_lim_fact) * PAGESIZE,
zfs_scan_mem_lim_min);
mutex_enter(&tvd->vdev_scan_io_queue_lock);
queue = tvd->vdev_scan_io_queue;
if (queue != NULL) {
- /* # extents in exts_by_size = # in exts_by_addr */
- mused += avl_numnodes(&queue->q_exts_by_size) *
- sizeof (range_seg_t) + queue->q_sio_memused;
+ /*
+ * # of extents in exts_by_addr = # in exts_by_size.
+ * B-tree efficiency is ~75%, but can be as low as 50%.
+ */
+ mused += zfs_btree_numnodes(&queue->q_exts_by_size) *
+ ((sizeof (range_seg_gap_t) + sizeof (uint64_t)) *
+ 3 / 2) + queue->q_sio_memused;
}
mutex_exit(&tvd->vdev_scan_io_queue_lock);
}
dprintf("current scan memory usage: %llu bytes\n", (longlong_t)mused);
if (mused == 0)
- ASSERT0(scn->scn_bytes_pending);
+ ASSERT0(scn->scn_queues_pending);
/*
* If we are above our hard limit, we need to clear out memory.
if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark))
return (B_FALSE); /* we're resuming */
- /* We only know how to resume from level-0 blocks. */
- if (zb && zb->zb_level != 0)
+ /* We only know how to resume from level-0 and objset blocks. */
+ if (zb && (zb->zb_level != 0 && zb->zb_level != ZB_ROOT_LEVEL))
return (B_FALSE);
/*
uint64_t scan_time_ns = curr_time_ns - scn->scn_sync_start_time;
uint64_t sync_time_ns = curr_time_ns -
scn->scn_dp->dp_spa->spa_sync_starttime;
- int dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max;
- int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
+ uint64_t dirty_min_bytes = zfs_dirty_data_max *
+ zfs_vdev_async_write_active_min_dirty_percent / 100;
+ uint_t mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
zfs_resilver_min_time_ms : zfs_scrub_min_time_ms;
if ((NSEC2MSEC(scan_time_ns) > mintime &&
- (dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent ||
+ (scn->scn_dp->dp_dirty_total >= dirty_min_bytes ||
txg_sync_waiting(scn->scn_dp) ||
NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) ||
spa_shutting_down(scn->scn_dp->dp_spa) ||
(zfs_scan_strict_mem_lim && dsl_scan_should_clear(scn))) {
- if (zb) {
+ if (zb && zb->zb_level == ZB_ROOT_LEVEL) {
+ dprintf("suspending at first available bookmark "
+ "%llx/%llx/%llx/%llx\n",
+ (longlong_t)zb->zb_objset,
+ (longlong_t)zb->zb_object,
+ (longlong_t)zb->zb_level,
+ (longlong_t)zb->zb_blkid);
+ SET_BOOKMARK(&scn->scn_phys.scn_bookmark,
+ zb->zb_objset, 0, 0, 0);
+ } else if (zb != NULL) {
dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n",
(longlong_t)zb->zb_objset,
(longlong_t)zb->zb_object,
zil_header_t *zsa_zh;
} zil_scan_arg_t;
-/* ARGSUSED */
static int
-dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
+dsl_scan_zil_block(zilog_t *zilog, const blkptr_t *bp, void *arg,
+ uint64_t claim_txg)
{
+ (void) zilog;
zil_scan_arg_t *zsa = arg;
dsl_pool_t *dp = zsa->zsa_dp;
dsl_scan_t *scn = dp->dp_scan;
return (0);
}
-/* ARGSUSED */
static int
-dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
+dsl_scan_zil_record(zilog_t *zilog, const lr_t *lrc, void *arg,
+ uint64_t claim_txg)
{
+ (void) zilog;
if (lrc->lrc_txtype == TX_WRITE) {
zil_scan_arg_t *zsa = arg;
dsl_pool_t *dp = zsa->zsa_dp;
dsl_scan_t *scn = dp->dp_scan;
zil_header_t *zh = zsa->zsa_zh;
- lr_write_t *lr = (lr_write_t *)lrc;
- blkptr_t *bp = &lr->lr_blkptr;
+ const lr_write_t *lr = (const lr_write_t *)lrc;
+ const blkptr_t *bp = &lr->lr_blkptr;
zbookmark_phys_t zb;
ASSERT(!BP_IS_REDACTED(bp));
if (claim_txg == 0 || bp->blk_birth < claim_txg)
return (0);
+ ASSERT3U(BP_GET_LSIZE(bp), !=, 0);
SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
lr->lr_foid, ZB_ZIL_LEVEL,
lr->lr_offset / BP_GET_LSIZE(bp));
}
static void
-scan_prefetch_ctx_rele(scan_prefetch_ctx_t *spc, void *tag)
+scan_prefetch_ctx_rele(scan_prefetch_ctx_t *spc, const void *tag)
{
if (zfs_refcount_remove(&spc->spc_refcnt, tag) == 0) {
zfs_refcount_destroy(&spc->spc_refcnt);
}
static scan_prefetch_ctx_t *
-scan_prefetch_ctx_create(dsl_scan_t *scn, dnode_phys_t *dnp, void *tag)
+scan_prefetch_ctx_create(dsl_scan_t *scn, dnode_phys_t *dnp, const void *tag)
{
scan_prefetch_ctx_t *spc;
}
static void
-scan_prefetch_ctx_add_ref(scan_prefetch_ctx_t *spc, void *tag)
+scan_prefetch_ctx_add_ref(scan_prefetch_ctx_t *spc, const void *tag)
{
zfs_refcount_add(&spc->spc_refcnt, tag);
}
scan_prefetch_ctx_rele(spc, FTAG);
}
-void
+static void
dsl_scan_prefetch_cb(zio_t *zio, const zbookmark_phys_t *zb, const blkptr_t *bp,
arc_buf_t *buf, void *private)
{
+ (void) zio;
scan_prefetch_ctx_t *spc = private;
dsl_scan_t *scn = spc->spc_scn;
spa_t *spa = scn->scn_dp->dp_spa;
scan_prefetch_ctx_rele(spc, scn);
}
-/* ARGSUSED */
static void
dsl_scan_prefetch_thread(void *arg)
{
/*
* If we found the block we're trying to resume from, or
- * we went past it to a different object, zero it out to
- * indicate that it's OK to start checking for suspending
- * again.
+ * we went past it, zero it out to indicate that it's OK
+ * to start checking for suspending again.
*/
- if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 ||
- zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) {
+ if (zbookmark_subtree_tbd(dnp, zb,
+ &scn->scn_phys.scn_bookmark)) {
dprintf("resuming at %llx/%llx/%llx/%llx\n",
(longlong_t)zb->zb_objset,
(longlong_t)zb->zb_object,
(longlong_t)zb->zb_level,
(longlong_t)zb->zb_blkid);
- bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb));
+ memset(&scn->scn_phys.scn_bookmark, 0, sizeof (*zb));
}
}
return (B_FALSE);
const zbookmark_phys_t *zb, dmu_tx_t *tx)
{
dsl_pool_t *dp = scn->scn_dp;
+ spa_t *spa = dp->dp_spa;
int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
int err;
ASSERT(!BP_IS_REDACTED(bp));
+ /*
+ * There is an unlikely case of encountering dnodes with contradicting
+ * dn_bonuslen and DNODE_FLAG_SPILL_BLKPTR flag before in files created
+ * or modified before commit 4254acb was merged. As it is not possible
+ * to know which of the two is correct, report an error.
+ */
+ if (dnp != NULL &&
+ dnp->dn_bonuslen > DN_MAX_BONUS_LEN(dnp)) {
+ scn->scn_phys.scn_errors++;
+ spa_log_error(spa, zb, &bp->blk_birth);
+ return (SET_ERROR(EINVAL));
+ }
+
if (BP_GET_LEVEL(bp) > 0) {
arc_flags_t flags = ARC_FLAG_WAIT;
int i;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
arc_buf_t *buf;
- err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
+ err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
zio_flags |= ZIO_FLAG_RAW;
}
- err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
+ err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
objset_phys_t *osp;
arc_buf_t *buf;
- err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf,
+ err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_SCRUB, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
DMU_USERUSED_OBJECT, tx);
}
arc_buf_destroy(buf, &buf);
+ } else if (!zfs_blkptr_verify(spa, bp,
+ BLK_CONFIG_NEEDED, BLK_VERIFY_LOG)) {
+ /*
+ * Sanity check the block pointer contents, this is handled
+ * by arc_read() for the cases above.
+ */
+ scn->scn_phys.scn_errors++;
+ spa_log_error(spa, zb, &bp->blk_birth);
+ return (SET_ERROR(EINVAL));
}
return (0);
scn->scn_visited_this_txg++;
- /*
- * This debugging is commented out to conserve stack space. This
- * function is called recursively and the debugging adds several
- * bytes to the stack for each call. It can be commented back in
- * if required to debug an issue in dsl_scan_visitbp().
- *
- * dprintf_bp(bp,
- * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p",
- * ds, ds ? ds->ds_object : 0,
- * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
- * bp);
- */
-
if (BP_IS_HOLE(bp)) {
scn->scn_holes_this_txg++;
return;
return;
}
+ /*
+ * Check if this block contradicts any filesystem flags.
+ */
+ spa_feature_t f = SPA_FEATURE_LARGE_BLOCKS;
+ if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE)
+ ASSERT(dsl_dataset_feature_is_active(ds, f));
+
+ f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
+ if (f != SPA_FEATURE_NONE)
+ ASSERT(dsl_dataset_feature_is_active(ds, f));
+
+ f = zio_compress_to_feature(BP_GET_COMPRESS(bp));
+ if (f != SPA_FEATURE_NONE)
+ ASSERT(dsl_dataset_feature_is_active(ds, f));
+
if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) {
scn->scn_lt_min_this_txg++;
return;
*/
scn_phys->scn_bookmark.zb_objset =
dsl_dataset_phys(ds)->ds_next_snap_obj;
- zfs_dbgmsg("destroying ds %llu; currently traversing; "
- "reset zb_objset to %llu",
+ zfs_dbgmsg("destroying ds %llu on %s; currently "
+ "traversing; reset zb_objset to %llu",
(u_longlong_t)ds->ds_object,
+ ds->ds_dir->dd_pool->dp_spa->spa_name,
(u_longlong_t)dsl_dataset_phys(ds)->
ds_next_snap_obj);
scn_phys->scn_flags |= DSF_VISIT_DS_AGAIN;
} else {
SET_BOOKMARK(&scn_phys->scn_bookmark,
ZB_DESTROYED_OBJSET, 0, 0, 0);
- zfs_dbgmsg("destroying ds %llu; currently traversing; "
- "reset bookmark to -1,0,0,0",
- (u_longlong_t)ds->ds_object);
+ zfs_dbgmsg("destroying ds %llu on %s; currently "
+ "traversing; reset bookmark to -1,0,0,0",
+ (u_longlong_t)ds->ds_object,
+ ds->ds_dir->dd_pool->dp_spa->spa_name);
}
}
}
scn->scn_phys.scn_queue_obj,
dsl_dataset_phys(ds)->ds_next_snap_obj,
mintxg, tx) == 0);
- zfs_dbgmsg("destroying ds %llu; in queue; "
+ zfs_dbgmsg("destroying ds %llu on %s; in queue; "
"replacing with %llu",
(u_longlong_t)ds->ds_object,
+ dp->dp_spa->spa_name,
(u_longlong_t)dsl_dataset_phys(ds)->
ds_next_snap_obj);
} else {
- zfs_dbgmsg("destroying ds %llu; in queue; removing",
- (u_longlong_t)ds->ds_object);
+ zfs_dbgmsg("destroying ds %llu on %s; in queue; "
+ "removing",
+ (u_longlong_t)ds->ds_object,
+ dp->dp_spa->spa_name);
}
}
if (scn_bookmark->zb_objset == ds->ds_object) {
scn_bookmark->zb_objset =
dsl_dataset_phys(ds)->ds_prev_snap_obj;
- zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
+ zfs_dbgmsg("snapshotting ds %llu on %s; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds->ds_object,
+ ds->ds_dir->dd_pool->dp_spa->spa_name,
(u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
}
}
VERIFY(zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj,
dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0);
- zfs_dbgmsg("snapshotting ds %llu; in queue; "
+ zfs_dbgmsg("snapshotting ds %llu on %s; in queue; "
"replacing with %llu",
(u_longlong_t)ds->ds_object,
+ dp->dp_spa->spa_name,
(u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj);
}
{
if (scn_bookmark->zb_objset == ds1->ds_object) {
scn_bookmark->zb_objset = ds2->ds_object;
- zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
+ zfs_dbgmsg("clone_swap ds %llu on %s; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds1->ds_object,
+ ds1->ds_dir->dd_pool->dp_spa->spa_name,
(u_longlong_t)ds2->ds_object);
} else if (scn_bookmark->zb_objset == ds2->ds_object) {
scn_bookmark->zb_objset = ds1->ds_object;
- zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
+ zfs_dbgmsg("clone_swap ds %llu on %s; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds2->ds_object,
+ ds2->ds_dir->dd_pool->dp_spa->spa_name,
(u_longlong_t)ds1->ds_object);
}
}
/*
- * Called when a parent dataset and its clone are swapped. If we were
+ * Called when an origin dataset and its clone are swapped. If we were
* currently traversing the dataset, we need to switch to traversing the
- * newly promoted parent.
+ * newly promoted clone.
*/
void
dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx)
{
dsl_pool_t *dp = ds1->ds_dir->dd_pool;
dsl_scan_t *scn = dp->dp_scan;
- uint64_t mintxg;
+ uint64_t mintxg1, mintxg2;
+ boolean_t ds1_queued, ds2_queued;
if (!dsl_scan_is_running(scn))
return;
ds_clone_swapped_bookmark(ds1, ds2, &scn->scn_phys.scn_bookmark);
ds_clone_swapped_bookmark(ds1, ds2, &scn->scn_phys_cached.scn_bookmark);
- if (scan_ds_queue_contains(scn, ds1->ds_object, &mintxg)) {
- scan_ds_queue_remove(scn, ds1->ds_object);
- scan_ds_queue_insert(scn, ds2->ds_object, mintxg);
+ /*
+ * Handle the in-memory scan queue.
+ */
+ ds1_queued = scan_ds_queue_contains(scn, ds1->ds_object, &mintxg1);
+ ds2_queued = scan_ds_queue_contains(scn, ds2->ds_object, &mintxg2);
+
+ /* Sanity checking. */
+ if (ds1_queued) {
+ ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
+ ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
+ }
+ if (ds2_queued) {
+ ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
+ ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
}
- if (scan_ds_queue_contains(scn, ds2->ds_object, &mintxg)) {
+
+ if (ds1_queued && ds2_queued) {
+ /*
+ * If both are queued, we don't need to do anything.
+ * The swapping code below would not handle this case correctly,
+ * since we can't insert ds2 if it is already there. That's
+ * because scan_ds_queue_insert() prohibits a duplicate insert
+ * and panics.
+ */
+ } else if (ds1_queued) {
+ scan_ds_queue_remove(scn, ds1->ds_object);
+ scan_ds_queue_insert(scn, ds2->ds_object, mintxg1);
+ } else if (ds2_queued) {
scan_ds_queue_remove(scn, ds2->ds_object);
- scan_ds_queue_insert(scn, ds1->ds_object, mintxg);
+ scan_ds_queue_insert(scn, ds1->ds_object, mintxg2);
}
- if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
- ds1->ds_object, &mintxg) == 0) {
- int err;
- ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
- ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
- VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
+ /*
+ * Handle the on-disk scan queue.
+ * The on-disk state is an out-of-date version of the in-memory state,
+ * so the in-memory and on-disk values for ds1_queued and ds2_queued may
+ * be different. Therefore we need to apply the swap logic to the
+ * on-disk state independently of the in-memory state.
+ */
+ ds1_queued = zap_lookup_int_key(dp->dp_meta_objset,
+ scn->scn_phys.scn_queue_obj, ds1->ds_object, &mintxg1) == 0;
+ ds2_queued = zap_lookup_int_key(dp->dp_meta_objset,
+ scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg2) == 0;
+
+ /* Sanity checking. */
+ if (ds1_queued) {
+ ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
+ ASSERT3U(mintxg1, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
+ }
+ if (ds2_queued) {
+ ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
+ ASSERT3U(mintxg2, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
+ }
+
+ if (ds1_queued && ds2_queued) {
+ /*
+ * If both are queued, we don't need to do anything.
+ * Alternatively, we could check for EEXIST from
+ * zap_add_int_key() and back out to the original state, but
+ * that would be more work than checking for this case upfront.
+ */
+ } else if (ds1_queued) {
+ VERIFY3S(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds1->ds_object, tx));
- err = zap_add_int_key(dp->dp_meta_objset,
- scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx);
- VERIFY(err == 0 || err == EEXIST);
- if (err == EEXIST) {
- /* Both were there to begin with */
- VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
- scn->scn_phys.scn_queue_obj,
- ds1->ds_object, mintxg, tx));
- }
- zfs_dbgmsg("clone_swap ds %llu; in queue; "
+ VERIFY3S(0, ==, zap_add_int_key(dp->dp_meta_objset,
+ scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg1, tx));
+ zfs_dbgmsg("clone_swap ds %llu on %s; in queue; "
"replacing with %llu",
(u_longlong_t)ds1->ds_object,
+ dp->dp_spa->spa_name,
(u_longlong_t)ds2->ds_object);
- }
- if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
- ds2->ds_object, &mintxg) == 0) {
- ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg);
- ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg);
- VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
+ } else if (ds2_queued) {
+ VERIFY3S(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds2->ds_object, tx));
- VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
- scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx));
- zfs_dbgmsg("clone_swap ds %llu; in queue; "
+ VERIFY3S(0, ==, zap_add_int_key(dp->dp_meta_objset,
+ scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg2, tx));
+ zfs_dbgmsg("clone_swap ds %llu on %s; in queue; "
"replacing with %llu",
(u_longlong_t)ds2->ds_object,
+ dp->dp_spa->spa_name,
(u_longlong_t)ds1->ds_object);
}
dsl_scan_sync_state(scn, tx, SYNC_CACHED);
}
-/* ARGSUSED */
static int
enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
{
* If we did not completely visit this dataset, do another pass.
*/
if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) {
- zfs_dbgmsg("incomplete pass; visiting again");
+ zfs_dbgmsg("incomplete pass on %s; visiting again",
+ dp->dp_spa->spa_name);
scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN;
scan_ds_queue_insert(scn, ds->ds_object,
scn->scn_phys.scn_cur_max_txg);
dsl_dataset_rele(ds, FTAG);
}
-/* ARGSUSED */
static int
enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
{
+ (void) arg;
dsl_dataset_t *ds;
int err;
dsl_scan_t *scn = dp->dp_scan;
return (0);
}
-/* ARGSUSED */
void
dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum,
ddt_entry_t *dde, dmu_tx_t *tx)
{
+ (void) tx;
const ddt_key_t *ddk = &dde->dde_key;
ddt_phys_t *ddp = dde->dde_phys;
blkptr_t bp;
zbookmark_phys_t zb = { 0 };
- int p;
if (!dsl_scan_is_running(scn))
return;
if (scn->scn_done_txg != 0)
return;
- for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
+ for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (ddp->ddp_phys_birth == 0 ||
ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg)
continue;
dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx)
{
ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark;
- ddt_entry_t dde;
+ ddt_entry_t dde = {{{{0}}}};
int error;
uint64_t n = 0;
- bzero(&dde, sizeof (ddt_entry_t));
-
while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) {
ddt_t *ddt;
break;
}
- zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; "
- "suspending=%u", (longlong_t)n,
+ zfs_dbgmsg("scanned %llu ddt entries on %s with class_max = %u; "
+ "suspending=%u", (longlong_t)n, scn->scn_dp->dp_spa->spa_name,
(int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_suspending);
ASSERT(error == 0 || error == ENOENT);
* In case we suspended right at the end of the ds, zero the
* bookmark so we don't think that we're still trying to resume.
*/
- bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t));
+ memset(&scn->scn_phys.scn_bookmark, 0, sizeof (zbookmark_phys_t));
/*
* Keep pulling things out of the dataset avl queue. Updates to the
}
static uint64_t
-dsl_scan_count_leaves(vdev_t *vd)
+dsl_scan_count_data_disks(spa_t *spa)
{
+ vdev_t *rvd = spa->spa_root_vdev;
uint64_t i, leaves = 0;
- /* we only count leaves that belong to the main pool and are readable */
- if (vd->vdev_islog || vd->vdev_isspare ||
- vd->vdev_isl2cache || !vdev_readable(vd))
- return (0);
-
- if (vd->vdev_ops->vdev_op_leaf)
- return (1);
-
- for (i = 0; i < vd->vdev_children; i++) {
- leaves += dsl_scan_count_leaves(vd->vdev_child[i]);
+ for (i = 0; i < rvd->vdev_children; i++) {
+ vdev_t *vd = rvd->vdev_child[i];
+ if (vd->vdev_islog || vd->vdev_isspare || vd->vdev_isl2cache)
+ continue;
+ leaves += vdev_get_ndisks(vd) - vdev_get_nparity(vd);
}
-
return (leaves);
}
uint64_t scan_time_ns = curr_time_ns - scn->scn_sync_start_time;
uint64_t sync_time_ns = curr_time_ns -
scn->scn_dp->dp_spa->spa_sync_starttime;
- int dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max;
- int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
+ uint64_t dirty_min_bytes = zfs_dirty_data_max *
+ zfs_vdev_async_write_active_min_dirty_percent / 100;
+ uint_t mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
zfs_resilver_min_time_ms : zfs_scrub_min_time_ms;
return ((NSEC2MSEC(scan_time_ns) > mintime &&
- (dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent ||
+ (scn->scn_dp->dp_dirty_total >= dirty_min_bytes ||
txg_sync_waiting(scn->scn_dp) ||
NSEC2SEC(sync_time_ns) >= zfs_txg_timeout)) ||
spa_shutting_down(scn->scn_dp->dp_spa));
{
dsl_scan_t *scn = queue->q_scn;
scan_io_t *sio;
- int64_t bytes_issued = 0;
boolean_t suspended = B_FALSE;
while ((sio = list_head(io_list)) != NULL) {
}
sio2bp(sio, &bp);
- bytes_issued += SIO_GET_ASIZE(sio);
scan_exec_io(scn->scn_dp, &bp, sio->sio_flags,
&sio->sio_zb, queue);
(void) list_remove_head(io_list);
scan_io_queues_update_zio_stats(queue, &bp);
sio_free(sio);
}
-
- atomic_add_64(&scn->scn_bytes_pending, -bytes_issued);
-
return (suspended);
}
srch_sio = sio_alloc(1);
srch_sio->sio_nr_dvas = 1;
- SIO_SET_OFFSET(srch_sio, rs->rs_start);
+ SIO_SET_OFFSET(srch_sio, rs_get_start(rs, queue->q_exts_by_addr));
/*
* The exact start of the extent might not contain any matching zios,
if (sio == NULL)
sio = avl_nearest(&queue->q_sios_by_addr, idx, AVL_AFTER);
- while (sio != NULL &&
- SIO_GET_OFFSET(sio) < rs->rs_end && num_sios <= 32) {
- ASSERT3U(SIO_GET_OFFSET(sio), >=, rs->rs_start);
- ASSERT3U(SIO_GET_END_OFFSET(sio), <=, rs->rs_end);
+ while (sio != NULL && SIO_GET_OFFSET(sio) < rs_get_end(rs,
+ queue->q_exts_by_addr) && num_sios <= 32) {
+ ASSERT3U(SIO_GET_OFFSET(sio), >=, rs_get_start(rs,
+ queue->q_exts_by_addr));
+ ASSERT3U(SIO_GET_END_OFFSET(sio), <=, rs_get_end(rs,
+ queue->q_exts_by_addr));
next_sio = AVL_NEXT(&queue->q_sios_by_addr, sio);
avl_remove(&queue->q_sios_by_addr, sio);
+ if (avl_is_empty(&queue->q_sios_by_addr))
+ atomic_add_64(&queue->q_scn->scn_queues_pending, -1);
queue->q_sio_memused -= SIO_GET_MUSED(sio);
bytes_issued += SIO_GET_ASIZE(sio);
* in the segment we update it to reflect the work we were able to
* complete. Otherwise, we remove it from the range tree entirely.
*/
- if (sio != NULL && SIO_GET_OFFSET(sio) < rs->rs_end) {
+ if (sio != NULL && SIO_GET_OFFSET(sio) < rs_get_end(rs,
+ queue->q_exts_by_addr)) {
range_tree_adjust_fill(queue->q_exts_by_addr, rs,
-bytes_issued);
range_tree_resize_segment(queue->q_exts_by_addr, rs,
- SIO_GET_OFFSET(sio), rs->rs_end - SIO_GET_OFFSET(sio));
-
+ SIO_GET_OFFSET(sio), rs_get_end(rs,
+ queue->q_exts_by_addr) - SIO_GET_OFFSET(sio));
+ queue->q_last_ext_addr = SIO_GET_OFFSET(sio);
return (B_TRUE);
} else {
- range_tree_remove(queue->q_exts_by_addr, rs->rs_start,
- rs->rs_end - rs->rs_start);
+ uint64_t rstart = rs_get_start(rs, queue->q_exts_by_addr);
+ uint64_t rend = rs_get_end(rs, queue->q_exts_by_addr);
+ range_tree_remove(queue->q_exts_by_addr, rstart, rend - rstart);
+ queue->q_last_ext_addr = -1;
return (B_FALSE);
}
}
scan_io_queue_fetch_ext(dsl_scan_io_queue_t *queue)
{
dsl_scan_t *scn = queue->q_scn;
+ range_tree_t *rt = queue->q_exts_by_addr;
ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock));
ASSERT(scn->scn_is_sorted);
- /* handle tunable overrides */
- if (scn->scn_checkpointing || scn->scn_clearing) {
- if (zfs_scan_issue_strategy == 1) {
- return (range_tree_first(queue->q_exts_by_addr));
- } else if (zfs_scan_issue_strategy == 2) {
- return (avl_first(&queue->q_exts_by_size));
- }
- }
+ if (!scn->scn_checkpointing && !scn->scn_clearing)
+ return (NULL);
/*
* During normal clearing, we want to issue our largest segments
* so the way we are sorted now is as good as it will ever get.
* In this case, we instead switch to issuing extents in LBA order.
*/
- if (scn->scn_checkpointing) {
- return (range_tree_first(queue->q_exts_by_addr));
- } else if (scn->scn_clearing) {
- return (avl_first(&queue->q_exts_by_size));
- } else {
- return (NULL);
+ if ((zfs_scan_issue_strategy < 1 && scn->scn_checkpointing) ||
+ zfs_scan_issue_strategy == 1)
+ return (range_tree_first(rt));
+
+ /*
+ * Try to continue previous extent if it is not completed yet. After
+ * shrink in scan_io_queue_gather() it may no longer be the best, but
+ * otherwise we leave shorter remnant every txg.
+ */
+ uint64_t start;
+ uint64_t size = 1ULL << rt->rt_shift;
+ range_seg_t *addr_rs;
+ if (queue->q_last_ext_addr != -1) {
+ start = queue->q_last_ext_addr;
+ addr_rs = range_tree_find(rt, start, size);
+ if (addr_rs != NULL)
+ return (addr_rs);
}
+
+ /*
+ * Nothing to continue, so find new best extent.
+ */
+ uint64_t *v = zfs_btree_first(&queue->q_exts_by_size, NULL);
+ if (v == NULL)
+ return (NULL);
+ queue->q_last_ext_addr = start = *v << rt->rt_shift;
+
+ /*
+ * We need to get the original entry in the by_addr tree so we can
+ * modify it.
+ */
+ addr_rs = range_tree_find(rt, start, size);
+ ASSERT3P(addr_rs, !=, NULL);
+ ASSERT3U(rs_get_start(addr_rs, rt), ==, start);
+ ASSERT3U(rs_get_end(addr_rs, rt), >, start);
+ return (addr_rs);
}
static void
dsl_scan_io_queue_t *queue = arg;
kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock;
boolean_t suspended = B_FALSE;
- range_seg_t *rs = NULL;
- scan_io_t *sio = NULL;
+ range_seg_t *rs;
+ scan_io_t *sio;
+ zio_t *zio;
list_t sio_list;
- uint64_t bytes_per_leaf = zfs_scan_vdev_limit;
- uint64_t nr_leaves = dsl_scan_count_leaves(queue->q_vd);
ASSERT(queue->q_scn->scn_is_sorted);
list_create(&sio_list, sizeof (scan_io_t),
offsetof(scan_io_t, sio_nodes.sio_list_node));
+ zio = zio_null(queue->q_scn->scn_zio_root, queue->q_scn->scn_dp->dp_spa,
+ NULL, NULL, NULL, ZIO_FLAG_CANFAIL);
mutex_enter(q_lock);
+ queue->q_zio = zio;
- /* calculate maximum in-flight bytes for this txg (min 1MB) */
- queue->q_maxinflight_bytes =
- MAX(nr_leaves * bytes_per_leaf, 1ULL << 20);
+ /* Calculate maximum in-flight bytes for this vdev. */
+ queue->q_maxinflight_bytes = MAX(1, zfs_scan_vdev_limit *
+ (vdev_get_ndisks(queue->q_vd) - vdev_get_nparity(queue->q_vd)));
/* reset per-queue scan statistics for this txg */
queue->q_total_seg_size_this_txg = 0;
/* loop until we run out of time or sios */
while ((rs = scan_io_queue_fetch_ext(queue)) != NULL) {
uint64_t seg_start = 0, seg_end = 0;
- boolean_t more_left = B_TRUE;
+ boolean_t more_left;
ASSERT(list_is_empty(&sio_list));
/* loop while we still have sios left to process in this rs */
- while (more_left) {
+ do {
scan_io_t *first_sio, *last_sio;
/*
if (suspended)
break;
- }
+ } while (more_left);
/* update statistics for debugging purposes */
scan_io_queues_update_seg_stats(queue, seg_start, seg_end);
scan_io_queue_insert_impl(queue, sio);
}
+ queue->q_zio = NULL;
mutex_exit(q_lock);
+ zio_nowait(zio);
list_destroy(&sio_list);
}
ASSERT(scn->scn_is_sorted);
ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
- if (scn->scn_bytes_pending == 0)
+ if (scn->scn_queues_pending == 0)
return;
if (scn->scn_taskq == NULL) {
return (B_TRUE);
}
+ if (zfs_max_async_dedup_frees != 0 &&
+ scn->scn_dedup_frees_this_txg >= zfs_max_async_dedup_frees) {
+ return (B_TRUE);
+ }
+
elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
(NSEC2MSEC(elapsed_nanosecs) > scn->scn_async_block_min_time_ms &&
-bp_get_dsize_sync(scn->scn_dp->dp_spa, bp),
-BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
scn->scn_visited_this_txg++;
+ if (BP_GET_DEDUP(bp))
+ scn->scn_dedup_frees_this_txg++;
return (0);
}
return (B_TRUE);
}
- /*
- * Check if the txg falls within the range which must be
- * resilvered. DVAs outside this range can always be skipped.
- */
- if (!vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1))
- return (B_FALSE);
-
/*
* Check if the top-level vdev must resilver this offset.
* When the offset does not intersect with a dirty leaf DTL
* then it may be possible to skip the resilver IO. The psize
* is provided instead of asize to simplify the check for RAIDZ.
*/
- if (!vdev_dtl_need_resilver(vd, DVA_GET_OFFSET(dva), psize))
+ if (!vdev_dtl_need_resilver(vd, dva, psize, phys_birth))
return (B_FALSE);
/*
}
if (scn->scn_visited_this_txg) {
zfs_dbgmsg("freed %llu blocks in %llums from "
- "free_bpobj/bptree txg %llu; err=%u",
+ "free_bpobj/bptree on %s in txg %llu; err=%u",
(longlong_t)scn->scn_visited_this_txg,
(longlong_t)
NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
- (longlong_t)tx->tx_txg, err);
+ spa->spa_name, (longlong_t)tx->tx_txg, err);
scn->scn_visited_this_txg = 0;
+ scn->scn_dedup_frees_this_txg = 0;
/*
- * Write out changes to the DDT that may be required as a
- * result of the blocks freed. This ensures that the DDT
- * is clean when a scrub/resilver runs.
+ * Write out changes to the DDT and the BRT that may be required
+ * as a result of the blocks freed. This ensures that the DDT
+ * and the BRT are clean when a scrub/resilver runs.
*/
ddt_sync(spa, tx->tx_txg);
+ brt_sync(spa, tx->tx_txg);
}
if (err != 0)
return (err);
ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes);
}
+ spa_notify_waiters(spa);
+
EQUIV(bpobj_is_open(&dp->dp_obsolete_bpobj),
0 == zap_contains(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_OBSOLETE_BPOBJ));
dsl_scan_done(scn, B_FALSE, tx);
if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
func = POOL_SCAN_RESILVER;
- zfs_dbgmsg("restarting scan func=%u txg=%llu",
- func, (longlong_t)tx->tx_txg);
+ zfs_dbgmsg("restarting scan func=%u on %s txg=%llu",
+ func, dp->dp_spa->spa_name, (longlong_t)tx->tx_txg);
dsl_scan_setup_sync(&func, tx);
}
/* reset scan statistics */
scn->scn_visited_this_txg = 0;
+ scn->scn_dedup_frees_this_txg = 0;
scn->scn_holes_this_txg = 0;
scn->scn_lt_min_this_txg = 0;
scn->scn_gt_max_this_txg = 0;
*/
if (zfs_scan_suspend_progress) {
uint64_t scan_time_ns = gethrtime() - scn->scn_sync_start_time;
- int mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
- zfs_resilver_min_time_ms : zfs_scrub_min_time_ms;
+ uint_t mintime = (scn->scn_phys.scn_func ==
+ POOL_SCAN_RESILVER) ? zfs_resilver_min_time_ms :
+ zfs_scrub_min_time_ms;
while (zfs_scan_suspend_progress &&
!txg_sync_waiting(scn->scn_dp) &&
return;
}
+ /*
+ * Disabled by default, set zfs_scan_report_txgs to report
+ * average performance over the last zfs_scan_report_txgs TXGs.
+ */
+ if (!dsl_scan_is_paused_scrub(scn) && zfs_scan_report_txgs != 0 &&
+ tx->tx_txg % zfs_scan_report_txgs == 0) {
+ scn->scn_issued_before_pass += spa->spa_scan_pass_issued;
+ spa_scan_stat_init(spa);
+ }
+
/*
* It is possible to switch from unsorted to sorted at any time,
* but afterwards the scan will remain sorted unless reloaded from
ddi_get_lbolt() - scn->scn_last_checkpoint >
SEC_TO_TICK(zfs_scan_checkpoint_intval)) {
if (!scn->scn_checkpointing)
- zfs_dbgmsg("begin scan checkpoint");
+ zfs_dbgmsg("begin scan checkpoint for %s",
+ spa->spa_name);
scn->scn_checkpointing = B_TRUE;
scn->scn_clearing = B_TRUE;
} else {
boolean_t should_clear = dsl_scan_should_clear(scn);
if (should_clear && !scn->scn_clearing) {
- zfs_dbgmsg("begin scan clearing");
+ zfs_dbgmsg("begin scan clearing for %s",
+ spa->spa_name);
scn->scn_clearing = B_TRUE;
} else if (!should_clear && scn->scn_clearing) {
- zfs_dbgmsg("finish scan clearing");
+ zfs_dbgmsg("finish scan clearing for %s",
+ spa->spa_name);
scn->scn_clearing = B_FALSE;
}
}
/* Need to scan metadata for more blocks to scrub */
dsl_scan_phys_t *scnp = &scn->scn_phys;
taskqid_t prefetch_tqid;
- uint64_t bytes_per_leaf = zfs_scan_vdev_limit;
- uint64_t nr_leaves = dsl_scan_count_leaves(spa->spa_root_vdev);
/*
- * Recalculate the max number of in-flight bytes for pool-wide
- * scanning operations (minimum 1MB). Limits for the issuing
- * phase are done per top-level vdev and are handled separately.
+ * Calculate the max number of in-flight bytes for pool-wide
+ * scanning operations (minimum 1MB, maximum 1/4 of arc_c_max).
+ * Limits for the issuing phase are done per top-level vdev and
+ * are handled separately.
*/
- scn->scn_maxinflight_bytes =
- MAX(nr_leaves * bytes_per_leaf, 1ULL << 20);
+ scn->scn_maxinflight_bytes = MIN(arc_c_max / 4, MAX(1ULL << 20,
+ zfs_scan_vdev_limit * dsl_scan_count_data_disks(spa)));
if (scnp->scn_ddt_bookmark.ddb_class <=
scnp->scn_ddt_class_max) {
ASSERT(ZB_IS_ZERO(&scnp->scn_bookmark));
- zfs_dbgmsg("doing scan sync txg %llu; "
+ zfs_dbgmsg("doing scan sync for %s txg %llu; "
"ddt bm=%llu/%llu/%llu/%llx",
+ spa->spa_name,
(longlong_t)tx->tx_txg,
(longlong_t)scnp->scn_ddt_bookmark.ddb_class,
(longlong_t)scnp->scn_ddt_bookmark.ddb_type,
(longlong_t)scnp->scn_ddt_bookmark.ddb_checksum,
(longlong_t)scnp->scn_ddt_bookmark.ddb_cursor);
} else {
- zfs_dbgmsg("doing scan sync txg %llu; "
+ zfs_dbgmsg("doing scan sync for %s txg %llu; "
"bm=%llu/%llu/%llu/%llu",
+ spa->spa_name,
(longlong_t)tx->tx_txg,
(longlong_t)scnp->scn_bookmark.zb_objset,
(longlong_t)scnp->scn_bookmark.zb_object,
(void) zio_wait(scn->scn_zio_root);
scn->scn_zio_root = NULL;
- zfs_dbgmsg("scan visited %llu blocks in %llums "
+ zfs_dbgmsg("scan visited %llu blocks of %s in %llums "
"(%llu os's, %llu holes, %llu < mintxg, "
"%llu in ddt, %llu > maxtxg)",
(longlong_t)scn->scn_visited_this_txg,
+ spa->spa_name,
(longlong_t)NSEC2MSEC(gethrtime() -
scn->scn_sync_start_time),
(longlong_t)scn->scn_objsets_visited_this_txg,
if (scn->scn_is_sorted) {
scn->scn_checkpointing = B_TRUE;
scn->scn_clearing = B_TRUE;
+ scn->scn_issued_before_pass +=
+ spa->spa_scan_pass_issued;
+ spa_scan_stat_init(spa);
}
- zfs_dbgmsg("scan complete txg %llu",
+ zfs_dbgmsg("scan complete for %s txg %llu",
+ spa->spa_name,
(longlong_t)tx->tx_txg);
}
- } else if (scn->scn_is_sorted && scn->scn_bytes_pending != 0) {
+ } else if (scn->scn_is_sorted && scn->scn_queues_pending != 0) {
ASSERT(scn->scn_clearing);
/* need to issue scrubbing IOs from per-vdev queues */
(void) dsl_scan_should_clear(scn);
dsl_scan_update_stats(scn);
- zfs_dbgmsg("scan issued %llu blocks (%llu segs) in %llums "
- "(avg_block_size = %llu, avg_seg_size = %llu)",
+ zfs_dbgmsg("scan issued %llu blocks for %s (%llu segs) "
+ "in %llums (avg_block_size = %llu, avg_seg_size = %llu)",
(longlong_t)scn->scn_zios_this_txg,
+ spa->spa_name,
(longlong_t)scn->scn_segs_this_txg,
(longlong_t)NSEC2MSEC(gethrtime() -
scn->scn_sync_start_time),
(longlong_t)scn->scn_avg_seg_size_this_txg);
} else if (scn->scn_done_txg != 0 && scn->scn_done_txg <= tx->tx_txg) {
/* Finished with everything. Mark the scrub as complete */
- zfs_dbgmsg("scan issuing complete txg %llu",
- (longlong_t)tx->tx_txg);
+ zfs_dbgmsg("scan issuing complete txg %llu for %s",
+ (longlong_t)tx->tx_txg,
+ spa->spa_name);
ASSERT3U(scn->scn_done_txg, !=, 0);
ASSERT0(spa->spa_scrub_inflight);
- ASSERT0(scn->scn_bytes_pending);
+ ASSERT0(scn->scn_queues_pending);
dsl_scan_done(scn, B_TRUE, tx);
sync_type = SYNC_MANDATORY;
}
}
static void
-count_block(dsl_scan_t *scn, zfs_all_blkstats_t *zab, const blkptr_t *bp)
+count_block_issued(spa_t *spa, const blkptr_t *bp, boolean_t all)
{
- int i;
-
/*
* Don't count embedded bp's, since we already did the work of
* scanning these when we scanned the containing block.
* zio code will only try the first one unless there is an issue.
* Therefore, we should only count the first DVA for these IOs.
*/
- if (scn->scn_is_sorted) {
- atomic_add_64(&scn->scn_dp->dp_spa->spa_scan_pass_issued,
- DVA_GET_ASIZE(&bp->blk_dva[0]));
- } else {
- spa_t *spa = scn->scn_dp->dp_spa;
-
- for (i = 0; i < BP_GET_NDVAS(bp); i++) {
- atomic_add_64(&spa->spa_scan_pass_issued,
- DVA_GET_ASIZE(&bp->blk_dva[i]));
- }
- }
+ atomic_add_64(&spa->spa_scan_pass_issued,
+ all ? BP_GET_ASIZE(bp) : DVA_GET_ASIZE(&bp->blk_dva[0]));
+}
+static void
+count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp)
+{
/*
* If we resume after a reboot, zab will be NULL; don't record
* incomplete stats in that case.
if (zab == NULL)
return;
- mutex_enter(&zab->zab_lock);
-
- for (i = 0; i < 4; i++) {
+ for (int i = 0; i < 4; i++) {
int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS;
int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL;
break;
}
}
-
- mutex_exit(&zab->zab_lock);
}
static void
scan_io_queue_insert_impl(dsl_scan_io_queue_t *queue, scan_io_t *sio)
{
avl_index_t idx;
- int64_t asize = SIO_GET_ASIZE(sio);
dsl_scan_t *scn = queue->q_scn;
ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock));
+ if (unlikely(avl_is_empty(&queue->q_sios_by_addr)))
+ atomic_add_64(&scn->scn_queues_pending, 1);
if (avl_find(&queue->q_sios_by_addr, sio, &idx) != NULL) {
/* block is already scheduled for reading */
- atomic_add_64(&scn->scn_bytes_pending, -asize);
sio_free(sio);
return;
}
avl_insert(&queue->q_sios_by_addr, sio, idx);
queue->q_sio_memused += SIO_GET_MUSED(sio);
- range_tree_add(queue->q_exts_by_addr, SIO_GET_OFFSET(sio), asize);
+ range_tree_add(queue->q_exts_by_addr, SIO_GET_OFFSET(sio),
+ SIO_GET_ASIZE(sio));
}
/*
scan_io_queue_insert(dsl_scan_io_queue_t *queue, const blkptr_t *bp, int dva_i,
int zio_flags, const zbookmark_phys_t *zb)
{
- dsl_scan_t *scn = queue->q_scn;
scan_io_t *sio = sio_alloc(BP_GET_NDVAS(bp));
ASSERT0(BP_IS_GANG(bp));
sio->sio_flags = zio_flags;
sio->sio_zb = *zb;
- /*
- * Increment the bytes pending counter now so that we can't
- * get an integer underflow in case the worker processes the
- * zio before we get to incrementing this counter.
- */
- atomic_add_64(&scn->scn_bytes_pending, SIO_GET_ASIZE(sio));
-
+ queue->q_last_ext_addr = -1;
scan_io_queue_insert_impl(queue, sio);
}
boolean_t needs_io = B_FALSE;
int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
-
+ count_block(dp->dp_blkstats, bp);
if (phys_birth <= scn->scn_phys.scn_min_txg ||
phys_birth >= scn->scn_phys.scn_max_txg) {
- count_block(scn, dp->dp_blkstats, bp);
+ count_block_issued(spa, bp, B_TRUE);
return (0);
}
/*
* Keep track of how much data we've examined so that
- * zpool(1M) status can make useful progress reports.
+ * zpool(8) status can make useful progress reports.
*/
- scn->scn_phys.scn_examined += DVA_GET_ASIZE(dva);
- spa->spa_scan_pass_exam += DVA_GET_ASIZE(dva);
+ uint64_t asize = DVA_GET_ASIZE(dva);
+ scn->scn_phys.scn_examined += asize;
+ spa->spa_scan_pass_exam += asize;
/* if it's a resilver, this may not be in the target range */
if (!needs_io)
if (needs_io && !zfs_no_scrub_io) {
dsl_scan_enqueue(dp, bp, zio_flags, zb);
} else {
- count_block(scn, dp->dp_blkstats, bp);
+ count_block_issued(spa, bp, B_TRUE);
}
/* do not relocate this block */
dsl_scan_t *scn = dp->dp_scan;
size_t size = BP_GET_PSIZE(bp);
abd_t *data = abd_alloc_for_io(size, B_FALSE);
-
- ASSERT3U(scn->scn_maxinflight_bytes, >, 0);
+ zio_t *pio;
if (queue == NULL) {
+ ASSERT3U(scn->scn_maxinflight_bytes, >, 0);
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight >= scn->scn_maxinflight_bytes)
cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
spa->spa_scrub_inflight += BP_GET_PSIZE(bp);
mutex_exit(&spa->spa_scrub_lock);
+ pio = scn->scn_zio_root;
} else {
kmutex_t *q_lock = &queue->q_vd->vdev_scan_io_queue_lock;
+ ASSERT3U(queue->q_maxinflight_bytes, >, 0);
mutex_enter(q_lock);
while (queue->q_inflight_bytes >= queue->q_maxinflight_bytes)
cv_wait(&queue->q_zio_cv, q_lock);
queue->q_inflight_bytes += BP_GET_PSIZE(bp);
+ pio = queue->q_zio;
mutex_exit(q_lock);
}
- count_block(scn, dp->dp_blkstats, bp);
- zio_nowait(zio_read(scn->scn_zio_root, spa, bp, data, size,
- dsl_scan_scrub_done, queue, ZIO_PRIORITY_SCRUB, zio_flags, zb));
+ ASSERT(pio != NULL);
+ count_block_issued(spa, bp, queue == NULL);
+ zio_nowait(zio_read(pio, spa, bp, data, size, dsl_scan_scrub_done,
+ queue, ZIO_PRIORITY_SCRUB, zio_flags, zb));
}
/*
* extents that are more completely filled (in a 3:2 ratio) vs just larger.
* Note that as an optimization, we replace multiplication and division by
* 100 with bitshifting by 7 (which effectively multiplies and divides by 128).
+ *
+ * Since we do not care if one extent is only few percent better than another,
+ * compress the score into 6 bits via binary logarithm AKA highbit64() and
+ * put into otherwise unused due to ashift high bits of offset. This allows
+ * to reduce q_exts_by_size B-tree elements to only 64 bits and compare them
+ * with single operation. Plus it makes scrubs more sequential and reduces
+ * chances that minor extent change move it within the B-tree.
*/
static int
ext_size_compare(const void *x, const void *y)
{
- const range_seg_t *rsa = x, *rsb = y;
- uint64_t sa = rsa->rs_end - rsa->rs_start,
- sb = rsb->rs_end - rsb->rs_start;
- uint64_t score_a, score_b;
+ const uint64_t *a = x, *b = y;
- score_a = rsa->rs_fill + ((((rsa->rs_fill << 7) / sa) *
- fill_weight * rsa->rs_fill) >> 7);
- score_b = rsb->rs_fill + ((((rsb->rs_fill << 7) / sb) *
- fill_weight * rsb->rs_fill) >> 7);
+ return (TREE_CMP(*a, *b));
+}
- if (score_a > score_b)
- return (-1);
- if (score_a == score_b) {
- if (rsa->rs_start < rsb->rs_start)
- return (-1);
- if (rsa->rs_start == rsb->rs_start)
- return (0);
- return (1);
- }
- return (1);
+static void
+ext_size_create(range_tree_t *rt, void *arg)
+{
+ (void) rt;
+ zfs_btree_t *size_tree = arg;
+
+ zfs_btree_create(size_tree, ext_size_compare, sizeof (uint64_t));
+}
+
+static void
+ext_size_destroy(range_tree_t *rt, void *arg)
+{
+ (void) rt;
+ zfs_btree_t *size_tree = arg;
+ ASSERT0(zfs_btree_numnodes(size_tree));
+
+ zfs_btree_destroy(size_tree);
}
+static uint64_t
+ext_size_value(range_tree_t *rt, range_seg_gap_t *rsg)
+{
+ (void) rt;
+ uint64_t size = rsg->rs_end - rsg->rs_start;
+ uint64_t score = rsg->rs_fill + ((((rsg->rs_fill << 7) / size) *
+ fill_weight * rsg->rs_fill) >> 7);
+ ASSERT3U(rt->rt_shift, >=, 8);
+ return (((uint64_t)(64 - highbit64(score)) << 56) | rsg->rs_start);
+}
+
+static void
+ext_size_add(range_tree_t *rt, range_seg_t *rs, void *arg)
+{
+ zfs_btree_t *size_tree = arg;
+ ASSERT3U(rt->rt_type, ==, RANGE_SEG_GAP);
+ uint64_t v = ext_size_value(rt, (range_seg_gap_t *)rs);
+ zfs_btree_add(size_tree, &v);
+}
+
+static void
+ext_size_remove(range_tree_t *rt, range_seg_t *rs, void *arg)
+{
+ zfs_btree_t *size_tree = arg;
+ ASSERT3U(rt->rt_type, ==, RANGE_SEG_GAP);
+ uint64_t v = ext_size_value(rt, (range_seg_gap_t *)rs);
+ zfs_btree_remove(size_tree, &v);
+}
+
+static void
+ext_size_vacate(range_tree_t *rt, void *arg)
+{
+ zfs_btree_t *size_tree = arg;
+ zfs_btree_clear(size_tree);
+ zfs_btree_destroy(size_tree);
+
+ ext_size_create(rt, arg);
+}
+
+static const range_tree_ops_t ext_size_ops = {
+ .rtop_create = ext_size_create,
+ .rtop_destroy = ext_size_destroy,
+ .rtop_add = ext_size_add,
+ .rtop_remove = ext_size_remove,
+ .rtop_vacate = ext_size_vacate
+};
+
/*
* Comparator for the q_sios_by_addr tree. Sorting is simply performed
* based on LBA-order (from lowest to highest).
{
const scan_io_t *a = x, *b = y;
- return (AVL_CMP(SIO_GET_OFFSET(a), SIO_GET_OFFSET(b)));
+ return (TREE_CMP(SIO_GET_OFFSET(a), SIO_GET_OFFSET(b)));
}
/* IO queues are created on demand when they are needed. */
q->q_scn = scn;
q->q_vd = vd;
q->q_sio_memused = 0;
+ q->q_last_ext_addr = -1;
cv_init(&q->q_zio_cv, NULL, CV_DEFAULT, NULL);
- q->q_exts_by_addr = range_tree_create_impl(&rt_avl_ops,
- &q->q_exts_by_size, ext_size_compare, zfs_scan_max_ext_gap);
+ q->q_exts_by_addr = range_tree_create_gap(&ext_size_ops, RANGE_SEG_GAP,
+ &q->q_exts_by_size, 0, vd->vdev_ashift, zfs_scan_max_ext_gap);
avl_create(&q->q_sios_by_addr, sio_addr_compare,
sizeof (scan_io_t), offsetof(scan_io_t, sio_nodes.sio_addr_node));
dsl_scan_t *scn = queue->q_scn;
scan_io_t *sio;
void *cookie = NULL;
- int64_t bytes_dequeued = 0;
ASSERT(MUTEX_HELD(&queue->q_vd->vdev_scan_io_queue_lock));
+ if (!avl_is_empty(&queue->q_sios_by_addr))
+ atomic_add_64(&scn->scn_queues_pending, -1);
while ((sio = avl_destroy_nodes(&queue->q_sios_by_addr, &cookie)) !=
NULL) {
ASSERT(range_tree_contains(queue->q_exts_by_addr,
SIO_GET_OFFSET(sio), SIO_GET_ASIZE(sio)));
- bytes_dequeued += SIO_GET_ASIZE(sio);
queue->q_sio_memused -= SIO_GET_MUSED(sio);
sio_free(sio);
}
ASSERT0(queue->q_sio_memused);
- atomic_add_64(&scn->scn_bytes_pending, -bytes_dequeued);
range_tree_vacate(queue->q_exts_by_addr, NULL, queue);
range_tree_destroy(queue->q_exts_by_addr);
avl_destroy(&queue->q_sios_by_addr);
sio_free(srch_sio);
if (sio != NULL) {
- int64_t asize = SIO_GET_ASIZE(sio);
blkptr_t tmpbp;
/* Got it while it was cold in the queue */
ASSERT3U(start, ==, SIO_GET_OFFSET(sio));
- ASSERT3U(size, ==, asize);
+ ASSERT3U(size, ==, SIO_GET_ASIZE(sio));
avl_remove(&queue->q_sios_by_addr, sio);
+ if (avl_is_empty(&queue->q_sios_by_addr))
+ atomic_add_64(&scn->scn_queues_pending, -1);
queue->q_sio_memused -= SIO_GET_MUSED(sio);
ASSERT(range_tree_contains(queue->q_exts_by_addr, start, size));
range_tree_remove_fill(queue->q_exts_by_addr, start, size);
- /*
- * We only update scn_bytes_pending in the cold path,
- * otherwise it will already have been accounted for as
- * part of the zio's execution.
- */
- atomic_add_64(&scn->scn_bytes_pending, -asize);
-
/* count the block as though we issued it */
sio2bp(sio, &tmpbp);
- count_block(scn, dp->dp_blkstats, &tmpbp);
+ count_block_issued(spa, &tmpbp, B_FALSE);
sio_free(sio);
}
dsl_scan_freed_dva(spa, bp, i);
}
-#if defined(_KERNEL)
-/* CSTYLED */
-module_param(zfs_scan_vdev_limit, ulong, 0644);
-MODULE_PARM_DESC(zfs_scan_vdev_limit,
+/*
+ * Check if a vdev needs resilvering (non-empty DTL), if so, and resilver has
+ * not started, start it. Otherwise, only restart if max txg in DTL range is
+ * greater than the max txg in the current scan. If the DTL max is less than
+ * the scan max, then the vdev has not missed any new data since the resilver
+ * started, so a restart is not needed.
+ */
+void
+dsl_scan_assess_vdev(dsl_pool_t *dp, vdev_t *vd)
+{
+ uint64_t min, max;
+
+ if (!vdev_resilver_needed(vd, &min, &max))
+ return;
+
+ if (!dsl_scan_resilvering(dp)) {
+ spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER);
+ return;
+ }
+
+ if (max <= dp->dp_scan->scn_phys.scn_max_txg)
+ return;
+
+ /* restart is needed, check if it can be deferred */
+ if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_RESILVER_DEFER))
+ vdev_defer_resilver(vd);
+ else
+ spa_async_request(dp->dp_spa, SPA_ASYNC_RESILVER);
+}
+
+ZFS_MODULE_PARAM(zfs, zfs_, scan_vdev_limit, U64, ZMOD_RW,
"Max bytes in flight per leaf vdev for scrubs and resilvers");
-module_param(zfs_scrub_min_time_ms, int, 0644);
-MODULE_PARM_DESC(zfs_scrub_min_time_ms, "Min millisecs to scrub per txg");
+ZFS_MODULE_PARAM(zfs, zfs_, scrub_min_time_ms, UINT, ZMOD_RW,
+ "Min millisecs to scrub per txg");
-module_param(zfs_obsolete_min_time_ms, int, 0644);
-MODULE_PARM_DESC(zfs_obsolete_min_time_ms, "Min millisecs to obsolete per txg");
+ZFS_MODULE_PARAM(zfs, zfs_, obsolete_min_time_ms, UINT, ZMOD_RW,
+ "Min millisecs to obsolete per txg");
-module_param(zfs_free_min_time_ms, int, 0644);
-MODULE_PARM_DESC(zfs_free_min_time_ms, "Min millisecs to free per txg");
+ZFS_MODULE_PARAM(zfs, zfs_, free_min_time_ms, UINT, ZMOD_RW,
+ "Min millisecs to free per txg");
-module_param(zfs_resilver_min_time_ms, int, 0644);
-MODULE_PARM_DESC(zfs_resilver_min_time_ms, "Min millisecs to resilver per txg");
+ZFS_MODULE_PARAM(zfs, zfs_, resilver_min_time_ms, UINT, ZMOD_RW,
+ "Min millisecs to resilver per txg");
-module_param(zfs_scan_suspend_progress, int, 0644);
-MODULE_PARM_DESC(zfs_scan_suspend_progress,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_suspend_progress, INT, ZMOD_RW,
"Set to prevent scans from progressing");
-module_param(zfs_no_scrub_io, int, 0644);
-MODULE_PARM_DESC(zfs_no_scrub_io, "Set to disable scrub I/O");
+ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_io, INT, ZMOD_RW,
+ "Set to disable scrub I/O");
-module_param(zfs_no_scrub_prefetch, int, 0644);
-MODULE_PARM_DESC(zfs_no_scrub_prefetch, "Set to disable scrub prefetching");
+ZFS_MODULE_PARAM(zfs, zfs_, no_scrub_prefetch, INT, ZMOD_RW,
+ "Set to disable scrub prefetching");
-/* CSTYLED */
-module_param(zfs_async_block_max_blocks, ulong, 0644);
-MODULE_PARM_DESC(zfs_async_block_max_blocks,
+ZFS_MODULE_PARAM(zfs, zfs_, async_block_max_blocks, U64, ZMOD_RW,
"Max number of blocks freed in one txg");
-module_param(zfs_free_bpobj_enabled, int, 0644);
-MODULE_PARM_DESC(zfs_free_bpobj_enabled, "Enable processing of the free_bpobj");
+ZFS_MODULE_PARAM(zfs, zfs_, max_async_dedup_frees, U64, ZMOD_RW,
+ "Max number of dedup blocks freed in one txg");
+
+ZFS_MODULE_PARAM(zfs, zfs_, free_bpobj_enabled, INT, ZMOD_RW,
+ "Enable processing of the free_bpobj");
-module_param(zfs_scan_mem_lim_fact, int, 0644);
-MODULE_PARM_DESC(zfs_scan_mem_lim_fact, "Fraction of RAM for scan hard limit");
+ZFS_MODULE_PARAM(zfs, zfs_, scan_blkstats, INT, ZMOD_RW,
+ "Enable block statistics calculation during scrub");
-module_param(zfs_scan_issue_strategy, int, 0644);
-MODULE_PARM_DESC(zfs_scan_issue_strategy,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_mem_lim_fact, UINT, ZMOD_RW,
+ "Fraction of RAM for scan hard limit");
+
+ZFS_MODULE_PARAM(zfs, zfs_, scan_issue_strategy, UINT, ZMOD_RW,
"IO issuing strategy during scrubbing. 0 = default, 1 = LBA, 2 = size");
-module_param(zfs_scan_legacy, int, 0644);
-MODULE_PARM_DESC(zfs_scan_legacy, "Scrub using legacy non-sequential method");
+ZFS_MODULE_PARAM(zfs, zfs_, scan_legacy, INT, ZMOD_RW,
+ "Scrub using legacy non-sequential method");
-module_param(zfs_scan_checkpoint_intval, int, 0644);
-MODULE_PARM_DESC(zfs_scan_checkpoint_intval,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_checkpoint_intval, UINT, ZMOD_RW,
"Scan progress on-disk checkpointing interval");
-/* CSTYLED */
-module_param(zfs_scan_max_ext_gap, ulong, 0644);
-MODULE_PARM_DESC(zfs_scan_max_ext_gap,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_max_ext_gap, U64, ZMOD_RW,
"Max gap in bytes between sequential scrub / resilver I/Os");
-module_param(zfs_scan_mem_lim_soft_fact, int, 0644);
-MODULE_PARM_DESC(zfs_scan_mem_lim_soft_fact,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_mem_lim_soft_fact, UINT, ZMOD_RW,
"Fraction of hard limit used as soft limit");
-module_param(zfs_scan_strict_mem_lim, int, 0644);
-MODULE_PARM_DESC(zfs_scan_strict_mem_lim,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_strict_mem_lim, INT, ZMOD_RW,
"Tunable to attempt to reduce lock contention");
-module_param(zfs_scan_fill_weight, int, 0644);
-MODULE_PARM_DESC(zfs_scan_fill_weight,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_fill_weight, UINT, ZMOD_RW,
"Tunable to adjust bias towards more filled segments during scans");
-module_param(zfs_resilver_disable_defer, int, 0644);
-MODULE_PARM_DESC(zfs_resilver_disable_defer,
+ZFS_MODULE_PARAM(zfs, zfs_, scan_report_txgs, UINT, ZMOD_RW,
+ "Tunable to report resilver performance over the last N txgs");
+
+ZFS_MODULE_PARAM(zfs, zfs_, resilver_disable_defer, INT, ZMOD_RW,
"Process all resilvers immediately");
-#endif