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_sync_state(dsl_scan_t *, dmu_tx_t *tx);
+int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */
+int zfs_resilver_delay = 2; /* number of ticks to delay resilver */
+int zfs_scrub_delay = 4; /* number of ticks to delay scrub */
+int zfs_scan_idle = 50; /* idle window in clock ticks */
+
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,
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));
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);
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
* 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, ZIO_FLAG_CANFAIL,
- &flags, &czb);
+ buf, NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
+ ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
}
static boolean_t
* 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)
{
dsl_pool_t *dp = scn->scn_dp;
+ int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
int err;
if (BP_GET_LEVEL(bp) > 0) {
err = arc_read_nolock(NULL, dp->dp_spa, bp,
arc_getbuf_func, bufp,
- ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
+ ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
err = arc_read_nolock(NULL, dp->dp_spa, bp,
arc_getbuf_func, bufp,
- ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
+ ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
err = arc_read_nolock(NULL, dp->dp_spa, bp,
arc_getbuf_func, bufp,
- ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
+ ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
err = arc_read_nolock(NULL, dp->dp_spa, bp,
arc_getbuf_func, bufp,
- ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
+ ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
osp = (*bufp)->b_data;
- if (DSL_SCAN_IS_SCRUB_RESILVER(scn))
- dsl_scan_zil(dp, &osp->os_zil_header);
-
dsl_scan_visitdnode(scn, ds, osp->os_type,
&osp->os_meta_dnode, *bufp, DMU_META_DNODE_OBJECT, tx);
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;
+ goto out;
if (BP_GET_TYPE(bp) != DMU_OT_USERGROUP_USED) {
/*
* it (original untranslated -> translations from
* deleted snap -> now).
*/
- bp_toread = *bp;
+ *bp_toread = *bp;
}
- 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;
}
/*
}
if (buf)
(void) arc_buf_remove_ref(buf, &buf);
+out:
+ kmem_free(bp_toread, sizeof(blkptr_t));
}
static void
{
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));
+ if (dmu_objset_from_ds(ds, &os))
+ goto out;
+
+ /*
+ * Only the ZIL in the head (non-snapshot) is valid. Even though
+ * snapshots can have ZIL block pointers (which may be the same
+ * BP as in the head), they must be ignored. So we traverse the
+ * ZIL here, rather than in scan_recurse(), because the regular
+ * snapshot block-sharing rules don't apply to it.
+ */
+ if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !dsl_dataset_is_snapshot(ds))
+ dsl_scan_zil(dp, &os->os_zil_header);
+
/*
* Iterate over the bps in this 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",
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)
continue;
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) {
* 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_setup_sync(scn, &func, tx);
}
-
if (!dsl_scan_active(scn) ||
spa_sync_pass(dp->dp_spa) > 1)
return;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
-
if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
scn->scn_phys.scn_ddt_class_max) {
zfs_dbgmsg("doing scan sync txg %llu; "
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;
- int zio_flags = ZIO_FLAG_SCRUB_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
- int zio_priority;
+ boolean_t needs_io = B_FALSE;
+ int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
+ int zio_priority = 0;
+ int scan_delay = 0;
+ int d;
if (phys_birth <= scn->scn_phys.scn_min_txg ||
phys_birth >= scn->scn_phys.scn_max_txg)
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) {
zio_flags |= ZIO_FLAG_RESILVER;
zio_priority = ZIO_PRIORITY_RESILVER;
needs_io = B_FALSE;
+ scan_delay = zfs_resilver_delay;
}
/* If it's an intent log block, failure is expected. */
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]));
}
if (needs_io && !zfs_no_scrub_io) {
+ vdev_t *rvd = spa->spa_root_vdev;
+ uint64_t maxinflight = rvd->vdev_children * zfs_top_maxinflight;
void *data = zio_data_buf_alloc(size);
mutex_enter(&spa->spa_scrub_lock);
- while (spa->spa_scrub_inflight >= spa->spa_scrub_maxinflight)
+ while (spa->spa_scrub_inflight >= maxinflight)
cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
spa->spa_scrub_inflight++;
mutex_exit(&spa->spa_scrub_lock);
+ /*
+ * If we're seeing recent (zfs_scan_idle) "important" I/Os
+ * then throttle our workload to limit the impact of a scan.
+ */
+ if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle)
+ delay(scan_delay);
+
zio_nowait(zio_read(NULL, spa, bp, data, size,
dsl_scan_scrub_done, NULL, zio_priority,
zio_flags, zb));
return (dsl_sync_task_do(dp, dsl_scan_setup_check,
dsl_scan_setup_sync, dp->dp_scan, &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
git://git.camperquake.de / zfs.git / blobdiff