*/
/*
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2013 by Delphix. All rights reserved.
*/
#include <sys/dsl_scan.h>
#include <sys/ddt.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
+#include <sys/zfeature.h>
#ifdef _KERNEL
#include <sys/zfs_vfsops.h>
#endif
typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_t *);
-static scan_cb_t dsl_scan_defrag_cb;
static scan_cb_t dsl_scan_scrub_cb;
-static scan_cb_t dsl_scan_remove_cb;
-static dsl_syncfunc_t dsl_scan_cancel_sync;
+static void dsl_scan_cancel_sync(void *, dmu_tx_t *);
static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *tx);
int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */
int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub 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 */
-boolean_t zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
-boolean_t zfs_no_scrub_prefetch = B_FALSE; /* set to disable srub prefetching */
+int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
+int zfs_no_scrub_prefetch = B_FALSE; /* set to disable srub prefetching */
enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
int dsl_scan_delay_completion = B_FALSE; /* set to delay scan completion */
((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
(scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
-extern int zfs_txg_timeout;
-
/* the order has to match pool_scan_type */
static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
NULL,
scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP);
scn->scn_dp = dp;
+ /*
+ * It's possible that we're resuming a scan after a reboot so
+ * make sure that the scan_async_destroying flag is initialized
+ * appropriately.
+ */
+ ASSERT(!scn->scn_async_destroying);
+ scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa,
+ &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]);
+
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
"scrub_func", sizeof (uint64_t), 1, &f);
if (err == 0) {
/* ARGSUSED */
static int
-dsl_scan_setup_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
{
- dsl_scan_t *scn = arg1;
+ dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (scn->scn_phys.scn_state == DSS_SCANNING)
- return (EBUSY);
+ return (SET_ERROR(EBUSY));
return (0);
}
-/* ARGSUSED */
static void
-dsl_scan_setup_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
{
- dsl_scan_t *scn = arg1;
- pool_scan_func_t *funcp = arg2;
+ dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
+ pool_scan_func_t *funcp = arg;
dmu_object_type_t ot = 0;
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
scn->scn_phys.scn_errors = 0;
scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
scn->scn_restart_txg = 0;
+ scn->scn_done_txg = 0;
spa_scan_stat_init(spa);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
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, ESC_ZFS_RESILVER_START);
+ spa_event_notify(spa, NULL,
+ FM_EREPORT_ZFS_RESILVER_START);
} else {
- spa_event_notify(spa, NULL, ESC_ZFS_SCRUB_START);
+ spa_event_notify(spa, NULL,
+ FM_EREPORT_ZFS_SCRUB_START);
}
spa->spa_scrub_started = B_TRUE;
/* back to the generic stuff */
if (dp->dp_blkstats == NULL) {
- dp->dp_blkstats =
- kmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP);
+ dp->dp_blkstats = kmem_alloc(sizeof (zfs_all_blkstats_t),
+ KM_PUSHPAGE | KM_NODEBUG);
}
bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
dsl_scan_sync_state(scn, tx);
- spa_history_log_internal(LOG_POOL_SCAN, spa, tx,
+ 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);
}
else
scn->scn_phys.scn_state = DSS_CANCELED;
- spa_history_log_internal(LOG_POOL_SCAN_DONE, spa, tx,
+ spa_history_log_internal(spa, "scan done", tx,
"complete=%u", complete);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE);
if (complete) {
spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ?
- ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH);
+ FM_EREPORT_ZFS_RESILVER_FINISH :
+ FM_EREPORT_ZFS_SCRUB_FINISH);
}
spa_errlog_rotate(spa);
/* ARGSUSED */
static int
-dsl_scan_cancel_check(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
{
- dsl_scan_t *scn = arg1;
+ dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (scn->scn_phys.scn_state != DSS_SCANNING)
- return (ENOENT);
+ return (SET_ERROR(ENOENT));
return (0);
}
/* ARGSUSED */
static void
-dsl_scan_cancel_sync(void *arg1, void *arg2, dmu_tx_t *tx)
+dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
{
- dsl_scan_t *scn = arg1;
+ dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
dsl_scan_done(scn, B_FALSE, tx);
dsl_scan_sync_state(scn, tx);
int
dsl_scan_cancel(dsl_pool_t *dp)
{
- boolean_t complete = B_FALSE;
- int err;
-
- err = dsl_sync_task_do(dp, dsl_scan_cancel_check,
- dsl_scan_cancel_sync, dp->dp_scan, &complete, 3);
- return (err);
+ return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
+ dsl_scan_cancel_sync, NULL, 3));
}
static void dsl_scan_visitbp(blkptr_t *bp,
const zbookmark_t *zb, dnode_phys_t *dnp, arc_buf_t *pbuf,
dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype,
dmu_tx_t *tx);
-static void dsl_scan_visitdnode(dsl_scan_t *, dsl_dataset_t *ds,
- dmu_objset_type_t ostype,
+inline __attribute__((always_inline)) static void dsl_scan_visitdnode(
+ dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype,
dnode_phys_t *dnp, arc_buf_t *buf, uint64_t object, dmu_tx_t *tx);
void
zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags));
}
-int
-dsl_read(zio_t *pio, spa_t *spa, const blkptr_t *bpp, arc_buf_t *pbuf,
- arc_done_func_t *done, void *private, int priority, int zio_flags,
- uint32_t *arc_flags, const zbookmark_t *zb)
-{
- return (arc_read(pio, spa, bpp, pbuf, done, private,
- priority, zio_flags, arc_flags, zb));
-}
-
-int
-dsl_read_nolock(zio_t *pio, spa_t *spa, const blkptr_t *bpp,
- arc_done_func_t *done, void *private, int priority, int zio_flags,
- uint32_t *arc_flags, const zbookmark_t *zb)
-{
- return (arc_read_nolock(pio, spa, bpp, done, private,
- priority, zio_flags, arc_flags, zb));
-}
-
-static boolean_t
-bookmark_is_zero(const zbookmark_t *zb)
-{
- return (zb->zb_objset == 0 && zb->zb_object == 0 &&
- zb->zb_level == 0 && zb->zb_blkid == 0);
-}
-
-/* dnp is the dnode for zb1->zb_object */
-static boolean_t
-bookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1,
- const zbookmark_t *zb2)
-{
- uint64_t zb1nextL0, zb2thisobj;
-
- ASSERT(zb1->zb_objset == zb2->zb_objset);
- ASSERT(zb2->zb_level == 0);
-
- /*
- * A bookmark in the deadlist is considered to be after
- * everything else.
- */
- if (zb2->zb_object == DMU_DEADLIST_OBJECT)
- return (B_TRUE);
-
- /* The objset_phys_t isn't before anything. */
- if (dnp == NULL)
- return (B_FALSE);
-
- zb1nextL0 = (zb1->zb_blkid + 1) <<
- ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT));
-
- zb2thisobj = zb2->zb_object ? zb2->zb_object :
- zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT);
-
- if (zb1->zb_object == DMU_META_DNODE_OBJECT) {
- uint64_t nextobj = zb1nextL0 *
- (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT;
- return (nextobj <= zb2thisobj);
- }
-
- if (zb1->zb_object < zb2thisobj)
- return (B_TRUE);
- if (zb1->zb_object > zb2thisobj)
- return (B_FALSE);
- if (zb2->zb_object == DMU_META_DNODE_OBJECT)
- return (B_FALSE);
- return (zb1nextL0 <= zb2->zb_blkid);
-}
-
static uint64_t
dsl_scan_ds_maxtxg(dsl_dataset_t *ds)
{
static void
dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx)
{
- VERIFY(0 == zap_update(scn->scn_dp->dp_meta_objset,
+ VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys, tx));
if (scn->scn_pausing)
return (B_TRUE); /* we're already pausing */
- if (!bookmark_is_zero(&scn->scn_phys.scn_bookmark))
+ 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. */
zfs_resilver_min_time_ms : zfs_scan_min_time_ms;
elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
- (elapsed_nanosecs / MICROSEC > mintime &&
+ (NSEC2MSEC(elapsed_nanosecs) > mintime &&
txg_sync_waiting(scn->scn_dp)) ||
spa_shutting_down(scn->scn_dp->dp_spa)) {
if (zb) {
SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid);
- /*
- * XXX need to make sure all of these arc_read() prefetches are
- * done before setting xlateall (similar to dsl_read())
- */
(void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp,
- buf, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
+ NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
}
/*
* We never skip over user/group accounting objects (obj<0)
*/
- if (!bookmark_is_zero(&scn->scn_phys.scn_bookmark) &&
+ if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) &&
(int64_t)zb->zb_object >= 0) {
/*
* If we already visited this bp & everything below (in
* a prior txg sync), don't bother doing it again.
*/
- if (bookmark_is_before(dnp, zb, &scn->scn_phys.scn_bookmark))
+ if (zbookmark_is_before(dnp, zb, &scn->scn_phys.scn_bookmark))
return (B_TRUE);
/*
* Return nonzero on i/o error.
* Return new buf to write out in *bufp.
*/
-static int
+inline __attribute__((always_inline)) static int
dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
dnode_phys_t *dnp, const blkptr_t *bp,
const zbookmark_t *zb, dmu_tx_t *tx, arc_buf_t **bufp)
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
- err = arc_read_nolock(NULL, dp->dp_spa, bp,
- arc_getbuf_func, bufp,
+ err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
} else if (BP_GET_TYPE(bp) == DMU_OT_USERGROUP_USED) {
uint32_t flags = ARC_WAIT;
- err = arc_read_nolock(NULL, dp->dp_spa, bp,
- arc_getbuf_func, bufp,
+ err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
int i, j;
int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
- err = arc_read_nolock(NULL, dp->dp_spa, bp,
- arc_getbuf_func, bufp,
+ err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
uint32_t flags = ARC_WAIT;
objset_phys_t *osp;
- err = arc_read_nolock(NULL, dp->dp_spa, bp,
- arc_getbuf_func, bufp,
+ err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (0);
}
-static void
+inline __attribute__((always_inline)) static void
dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds,
dmu_objset_type_t ostype, dnode_phys_t *dnp, arc_buf_t *buf,
uint64_t object, dmu_tx_t *tx)
{
dsl_pool_t *dp = scn->scn_dp;
arc_buf_t *buf = NULL;
- blkptr_t bp_toread = *bp;
+ blkptr_t *bp_toread;
+
+ bp_toread = kmem_alloc(sizeof (blkptr_t), KM_PUSHPAGE);
+ *bp_toread = *bp;
/* ASSERT(pbuf == NULL || arc_released(pbuf)); */
if (dsl_scan_check_pause(scn, zb))
- return;
+ goto out;
if (dsl_scan_check_resume(scn, dnp, zb))
- return;
+ goto out;
if (bp->blk_birth == 0)
- return;
+ goto out;
scn->scn_visited_this_txg++;
- dprintf_bp(bp,
- "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx buf=%p bp=%p",
- ds, ds ? ds->ds_object : 0,
- zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
- pbuf, bp);
+ /*
+ * This debugging is commented out to conserve stack space. This
+ * function is called recursively and the debugging addes 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 buf=%p bp=%p",
+ * ds, ds ? ds->ds_object : 0,
+ * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
+ * pbuf, bp);
+ */
if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
- return;
-
- if (BP_GET_TYPE(bp) != DMU_OT_USERGROUP_USED) {
- /*
- * For non-user-accounting blocks, we need to read the
- * new bp (from a deleted snapshot, found in
- * check_existing_xlation). If we used the old bp,
- * pointers inside this block from before we resumed
- * would be untranslated.
- *
- * For user-accounting blocks, we need to read the old
- * bp, because we will apply the entire space delta to
- * it (original untranslated -> translations from
- * deleted snap -> now).
- */
- bp_toread = *bp;
- }
+ goto out;
- if (dsl_scan_recurse(scn, ds, ostype, dnp, &bp_toread, zb, tx,
+ if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx,
&buf) != 0)
- return;
+ goto out;
/*
* If dsl_scan_ddt() has aready visited this block, it will have
if (ddt_class_contains(dp->dp_spa,
scn->scn_phys.scn_ddt_class_max, bp)) {
ASSERT(buf == NULL);
- return;
+ goto out;
}
/*
* Don't scan it now unless we need to because something
* under it was modified.
*/
- if (bp->blk_birth <= scn->scn_phys.scn_cur_max_txg) {
+ if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) {
scan_funcs[scn->scn_phys.scn_func](dp, bp, zb);
}
if (buf)
(void) arc_buf_remove_ref(buf, &buf);
+out:
+ kmem_free(bp_toread, sizeof (blkptr_t));
}
static void
/* ARGSUSED */
static int
-enqueue_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
{
struct enqueue_clones_arg *eca = arg;
dsl_dataset_t *ds;
int err;
- dsl_pool_t *dp = spa->spa_dsl_pool;
dsl_scan_t *scn = dp->dp_scan;
- err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+ if (hds->ds_dir->dd_phys->dd_origin_obj != eca->originobj)
+ return (0);
+
+ err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
if (err)
return (err);
- if (ds->ds_dir->dd_phys->dd_origin_obj == eca->originobj) {
- while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) {
- dsl_dataset_t *prev;
- err = dsl_dataset_hold_obj(dp,
- ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
+ while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) {
+ dsl_dataset_t *prev;
+ err = dsl_dataset_hold_obj(dp,
+ ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
- dsl_dataset_rele(ds, FTAG);
- if (err)
- return (err);
- ds = prev;
- }
- VERIFY(zap_add_int_key(dp->dp_meta_objset,
- scn->scn_phys.scn_queue_obj, ds->ds_object,
- ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0);
+ dsl_dataset_rele(ds, FTAG);
+ if (err)
+ return (err);
+ ds = prev;
}
+ VERIFY(zap_add_int_key(dp->dp_meta_objset,
+ scn->scn_phys.scn_queue_obj, ds->ds_object,
+ ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0);
dsl_dataset_rele(ds, FTAG);
return (0);
}
dsl_pool_t *dp = scn->scn_dp;
dsl_dataset_t *ds;
objset_t *os;
+ char *dsname;
VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
dmu_buf_will_dirty(ds->ds_dbuf, tx);
dsl_scan_visit_rootbp(scn, ds, &ds->ds_phys->ds_bp, tx);
- char *dsname = kmem_alloc(ZFS_MAXNAMELEN, KM_SLEEP);
+ dsname = kmem_alloc(ZFS_MAXNAMELEN, KM_PUSHPAGE);
dsl_dataset_name(ds, dsname);
zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
"pausing=%u",
}
if (usenext) {
- VERIFY(zap_join_key(dp->dp_meta_objset,
+ VERIFY0(zap_join_key(dp->dp_meta_objset,
ds->ds_phys->ds_next_clones_obj,
scn->scn_phys.scn_queue_obj,
- ds->ds_phys->ds_creation_txg, tx) == 0);
+ ds->ds_phys->ds_creation_txg, tx));
} else {
struct enqueue_clones_arg eca;
eca.tx = tx;
eca.originobj = ds->ds_object;
- (void) dmu_objset_find_spa(ds->ds_dir->dd_pool->dp_spa,
- NULL, enqueue_clones_cb, &eca, DS_FIND_CHILDREN);
+ VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
+ enqueue_clones_cb, &eca, DS_FIND_CHILDREN));
}
}
/* ARGSUSED */
static int
-enqueue_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
+enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
{
dmu_tx_t *tx = arg;
dsl_dataset_t *ds;
int err;
- dsl_pool_t *dp = spa->spa_dsl_pool;
dsl_scan_t *scn = dp->dp_scan;
- err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
+ err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
if (err)
return (err);
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 = { 0 };
+ ddt_entry_t dde;
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;
ddt_phys_t *ddp = dde->dde_phys;
blkptr_t bp;
zbookmark_t zb = { 0 };
+ int p;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
- for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
+ for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (ddp->ddp_phys_birth == 0 ||
- ddp->ddp_phys_birth > scn->scn_phys.scn_cur_max_txg)
+ ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg)
continue;
ddt_bp_create(checksum, ddk, ddp, &bp);
dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx)
{
dsl_pool_t *dp = scn->scn_dp;
- zap_cursor_t zc;
- zap_attribute_t za;
+ zap_cursor_t *zc;
+ zap_attribute_t *za;
if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
scn->scn_phys.scn_ddt_class_max) {
return;
if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
- VERIFY(0 == dmu_objset_find_spa(dp->dp_spa,
- NULL, enqueue_cb, tx, DS_FIND_CHILDREN));
+ VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
+ enqueue_cb, tx, DS_FIND_CHILDREN));
} else {
dsl_scan_visitds(scn,
dp->dp_origin_snap->ds_object, tx);
* bookmark so we don't think that we're still trying to resume.
*/
bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_t));
+ zc = kmem_alloc(sizeof (zap_cursor_t), KM_PUSHPAGE);
+ za = kmem_alloc(sizeof (zap_attribute_t), KM_PUSHPAGE);
/* keep pulling things out of the zap-object-as-queue */
- while (zap_cursor_init(&zc, dp->dp_meta_objset,
+ while (zap_cursor_init(zc, dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj),
- zap_cursor_retrieve(&zc, &za) == 0) {
+ zap_cursor_retrieve(zc, za) == 0) {
dsl_dataset_t *ds;
uint64_t dsobj;
- dsobj = strtonum(za.za_name, NULL);
+ dsobj = strtonum(za->za_name, NULL);
VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, dsobj, tx));
/* Set up min/max txg */
VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
- if (za.za_first_integer != 0) {
+ if (za->za_first_integer != 0) {
scn->scn_phys.scn_cur_min_txg =
MAX(scn->scn_phys.scn_min_txg,
- za.za_first_integer);
+ za->za_first_integer);
} else {
scn->scn_phys.scn_cur_min_txg =
MAX(scn->scn_phys.scn_min_txg,
dsl_dataset_rele(ds, FTAG);
dsl_scan_visitds(scn, dsobj, tx);
- zap_cursor_fini(&zc);
+ zap_cursor_fini(zc);
if (scn->scn_pausing)
- return;
+ goto out;
}
- zap_cursor_fini(&zc);
+ zap_cursor_fini(zc);
+out:
+ kmem_free(za, sizeof (zap_attribute_t));
+ kmem_free(zc, sizeof (zap_cursor_t));
}
-static int
-dsl_scan_free_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
+static boolean_t
+dsl_scan_free_should_pause(dsl_scan_t *scn)
{
- dsl_scan_t *scn = arg;
uint64_t elapsed_nanosecs;
- elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
+ if (zfs_recover)
+ return (B_FALSE);
- if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
- (elapsed_nanosecs / MICROSEC > zfs_free_min_time_ms &&
+ elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
+ return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
+ (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms &&
txg_sync_waiting(scn->scn_dp)) ||
- spa_shutting_down(scn->scn_dp->dp_spa))
- return (ERESTART);
+ spa_shutting_down(scn->scn_dp->dp_spa));
+}
+
+static int
+dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
+{
+ dsl_scan_t *scn = arg;
+
+ if (!scn->scn_is_bptree ||
+ (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) {
+ if (dsl_scan_free_should_pause(scn))
+ return (SET_ERROR(ERESTART));
+ }
zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa,
dmu_tx_get_txg(tx), bp, 0));
if (spa_shutting_down(spa))
return (B_FALSE);
- if (scn->scn_phys.scn_state == DSS_SCANNING)
+ if (scn->scn_phys.scn_state == DSS_SCANNING ||
+ scn->scn_async_destroying)
return (B_TRUE);
if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
func = POOL_SCAN_RESILVER;
zfs_dbgmsg("restarting scan func=%u txg=%llu",
func, tx->tx_txg);
- dsl_scan_setup_sync(scn, &func, tx);
+ dsl_scan_setup_sync(&func, tx);
}
if (!dsl_scan_active(scn) ||
* traversing it.
*/
if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
+ scn->scn_is_bptree = B_FALSE;
scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
NULL, ZIO_FLAG_MUSTSUCCEED);
err = bpobj_iterate(&dp->dp_free_bpobj,
- dsl_scan_free_cb, scn, tx);
+ dsl_scan_free_block_cb, scn, tx);
VERIFY3U(0, ==, zio_wait(scn->scn_zio_root));
+
+ if (err == 0 && spa_feature_is_active(spa,
+ &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
+ ASSERT(scn->scn_async_destroying);
+ scn->scn_is_bptree = B_TRUE;
+ scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
+ NULL, ZIO_FLAG_MUSTSUCCEED);
+ err = bptree_iterate(dp->dp_meta_objset,
+ dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb,
+ scn, tx);
+ VERIFY0(zio_wait(scn->scn_zio_root));
+
+ if (err == 0) {
+ zfeature_info_t *feat = &spa_feature_table
+ [SPA_FEATURE_ASYNC_DESTROY];
+ /* finished; deactivate async destroy feature */
+ spa_feature_decr(spa, feat, tx);
+ ASSERT(!spa_feature_is_active(spa, feat));
+ VERIFY0(zap_remove(dp->dp_meta_objset,
+ DMU_POOL_DIRECTORY_OBJECT,
+ DMU_POOL_BPTREE_OBJ, tx));
+ VERIFY0(bptree_free(dp->dp_meta_objset,
+ dp->dp_bptree_obj, tx));
+ dp->dp_bptree_obj = 0;
+ scn->scn_async_destroying = B_FALSE;
+ }
+ }
if (scn->scn_visited_this_txg) {
zfs_dbgmsg("freed %llu blocks in %llums from "
- "free_bpobj txg %llu",
+ "free_bpobj/bptree txg %llu",
(longlong_t)scn->scn_visited_this_txg,
(longlong_t)
- (gethrtime() - scn->scn_sync_start_time) / MICROSEC,
+ NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
(longlong_t)tx->tx_txg);
scn->scn_visited_this_txg = 0;
/*
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
+ if (scn->scn_done_txg == tx->tx_txg) {
+ ASSERT(!scn->scn_pausing);
+ /* finished with scan. */
+ zfs_dbgmsg("txg %llu scan complete", tx->tx_txg);
+ dsl_scan_done(scn, B_TRUE, tx);
+ ASSERT3U(spa->spa_scrub_inflight, ==, 0);
+ dsl_scan_sync_state(scn, tx);
+ return;
+ }
+
if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
scn->scn_phys.scn_ddt_class_max) {
zfs_dbgmsg("doing scan sync txg %llu; "
scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
NULL, ZIO_FLAG_CANFAIL);
+ dsl_pool_config_enter(dp, FTAG);
dsl_scan_visit(scn, tx);
+ dsl_pool_config_exit(dp, FTAG);
(void) zio_wait(scn->scn_zio_root);
scn->scn_zio_root = NULL;
zfs_dbgmsg("visited %llu blocks in %llums",
(longlong_t)scn->scn_visited_this_txg,
- (longlong_t)(gethrtime() - scn->scn_sync_start_time) / MICROSEC);
+ (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time));
if (!scn->scn_pausing) {
- /* finished with scan. */
- zfs_dbgmsg("finished scan txg %llu", (longlong_t)tx->tx_txg);
- dsl_scan_done(scn, B_TRUE, tx);
+ scn->scn_done_txg = tx->tx_txg + 1;
+ zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
+ tx->tx_txg, scn->scn_done_txg);
}
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
for (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;
- zfs_blkstat_t *zb = &zab->zab_type[l][t];
int equal;
+ zfs_blkstat_t *zb;
+ if (t & DMU_OT_NEWTYPE)
+ t = DMU_OT_OTHER;
+
+ zb = &zab->zab_type[l][t];
zb->zb_count++;
zb->zb_asize += BP_GET_ASIZE(bp);
zb->zb_lsize += BP_GET_LSIZE(bp);
size_t size = BP_GET_PSIZE(bp);
spa_t *spa = dp->dp_spa;
uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp);
- boolean_t needs_io;
+ boolean_t needs_io = B_FALSE;
int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
- int zio_priority;
int scan_delay = 0;
+ int d;
if (phys_birth <= scn->scn_phys.scn_min_txg ||
phys_birth >= scn->scn_phys.scn_max_txg)
ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn));
if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) {
zio_flags |= ZIO_FLAG_SCRUB;
- zio_priority = ZIO_PRIORITY_SCRUB;
needs_io = B_TRUE;
scan_delay = zfs_scrub_delay;
- } else if (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) {
+ } else {
+ ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER);
zio_flags |= ZIO_FLAG_RESILVER;
- zio_priority = ZIO_PRIORITY_RESILVER;
needs_io = B_FALSE;
scan_delay = zfs_resilver_delay;
}
if (zb->zb_level == ZB_ZIL_LEVEL)
zio_flags |= ZIO_FLAG_SPECULATIVE;
- for (int d = 0; d < BP_GET_NDVAS(bp); d++) {
+ for (d = 0; d < BP_GET_NDVAS(bp); d++) {
vdev_t *vd = vdev_lookup_top(spa,
DVA_GET_VDEV(&bp->blk_dva[d]));
delay(scan_delay);
zio_nowait(zio_read(NULL, spa, bp, data, size,
- dsl_scan_scrub_done, NULL, zio_priority,
+ dsl_scan_scrub_done, NULL, ZIO_PRIORITY_SCRUB,
zio_flags, zb));
}
spa->spa_scrub_reopen = B_FALSE;
(void) spa_vdev_state_exit(spa, NULL, 0);
- return (dsl_sync_task_do(dp, dsl_scan_setup_check,
- dsl_scan_setup_sync, dp->dp_scan, &func, 0));
+ return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
+ dsl_scan_setup_sync, &func, 0));
}
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+module_param(zfs_top_maxinflight, int, 0644);
+MODULE_PARM_DESC(zfs_top_maxinflight, "Max I/Os per top-level");
+
+module_param(zfs_resilver_delay, int, 0644);
+MODULE_PARM_DESC(zfs_resilver_delay, "Number of ticks to delay resilver");
+
+module_param(zfs_scrub_delay, int, 0644);
+MODULE_PARM_DESC(zfs_scrub_delay, "Number of ticks to delay scrub");
+
+module_param(zfs_scan_idle, int, 0644);
+MODULE_PARM_DESC(zfs_scan_idle, "Idle window in clock ticks");
+
+module_param(zfs_scan_min_time_ms, int, 0644);
+MODULE_PARM_DESC(zfs_scan_min_time_ms, "Min millisecs to scrub 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");
+
+module_param(zfs_resilver_min_time_ms, int, 0644);
+MODULE_PARM_DESC(zfs_resilver_min_time_ms, "Min millisecs to resilver per txg");
+
+module_param(zfs_no_scrub_io, int, 0644);
+MODULE_PARM_DESC(zfs_no_scrub_io, "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");
+#endif
projects / mirror_zfs.git / blobdiff