zio_add_child(zio_t *pio, zio_t *cio)
{
zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP);
+ int w;
/*
* Logical I/Os can have logical, gang, or vdev children.
ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0);
- for (int w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (w = 0; w < ZIO_WAIT_TYPES; w++)
pio->io_children[cio->io_child_type][w] += !cio->io_state[w];
list_insert_head(&pio->io_child_list, zl);
*/
if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize &&
pass > SYNC_PASS_REWRITE) {
- ASSERT(psize != 0);
enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
+ ASSERT(psize != 0);
zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
zio->io_flags |= ZIO_FLAG_IO_REWRITE;
} else {
{
kthread_t *executor = zio->io_executor;
spa_t *spa = zio->io_spa;
+ zio_type_t t;
- for (zio_type_t t = 0; t < ZIO_TYPES; t++)
+ for (t = 0; t < ZIO_TYPES; t++)
if (taskq_member(spa->spa_zio_taskq[t][q], executor))
return (B_TRUE);
zio_reexecute(zio_t *pio)
{
zio_t *cio, *cio_next;
+ int c, w;
ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL);
ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN);
pio->io_pipeline = pio->io_orig_pipeline;
pio->io_reexecute = 0;
pio->io_error = 0;
- for (int w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (w = 0; w < ZIO_WAIT_TYPES; w++)
pio->io_state[w] = 0;
- for (int c = 0; c < ZIO_CHILD_TYPES; c++)
+ for (c = 0; c < ZIO_CHILD_TYPES; c++)
pio->io_child_error[c] = 0;
if (IO_IS_ALLOCATING(pio))
for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) {
cio_next = zio_walk_children(pio);
mutex_enter(&pio->io_lock);
- for (int w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (w = 0; w < ZIO_WAIT_TYPES; w++)
pio->io_children[cio->io_child_type][w]++;
mutex_exit(&pio->io_lock);
zio_reexecute(cio);
zio_gang_node_free(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = *gnpp;
+ int g;
- for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
+ for (g = 0; g < SPA_GBH_NBLKPTRS; g++)
ASSERT(gn->gn_child[g] == NULL);
zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE);
zio_gang_tree_free(zio_gang_node_t **gnpp)
{
zio_gang_node_t *gn = *gnpp;
+ int g;
if (gn == NULL)
return;
- for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
+ for (g = 0; g < SPA_GBH_NBLKPTRS; g++)
zio_gang_tree_free(&gn->gn_child[g]);
zio_gang_node_free(gnpp);
zio_t *gio = zio->io_gang_leader;
zio_gang_node_t *gn = zio->io_private;
blkptr_t *bp = zio->io_bp;
+ int g;
ASSERT(gio == zio_unique_parent(zio));
ASSERT(zio->io_child_count == 0);
ASSERT(zio->io_size == SPA_GANGBLOCKSIZE);
ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
- for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
+ for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
if (!BP_IS_GANG(gbp))
continue;
{
zio_t *gio = pio->io_gang_leader;
zio_t *zio;
+ int g;
ASSERT(BP_IS_GANG(bp) == !!gn);
ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp));
if (gn != NULL) {
ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
- for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
+ for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
if (BP_IS_HOLE(gbp))
continue;
zio_write_gang_member_ready(zio_t *zio)
{
zio_t *pio = zio_unique_parent(zio);
- zio_t *gio = zio->io_gang_leader;
+ ASSERTV(zio_t *gio = zio->io_gang_leader;)
dva_t *cdva = zio->io_bp->blk_dva;
dva_t *pdva = pio->io_bp->blk_dva;
uint64_t asize;
+ int d;
if (BP_IS_HOLE(zio->io_bp))
return;
ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp));
mutex_enter(&pio->io_lock);
- for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) {
+ for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) {
ASSERT(DVA_GET_GANG(&pdva[d]));
asize = DVA_GET_ASIZE(&pdva[d]);
asize += DVA_GET_ASIZE(&cdva[d]);
int copies = gio->io_prop.zp_copies;
int gbh_copies = MIN(copies + 1, spa_max_replication(spa));
zio_prop_t zp;
- int error;
+ int g, error;
error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE,
bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp,
/*
* Create and nowait the gang children.
*/
- for (int g = 0; resid != 0; resid -= lsize, g++) {
+ for (g = 0; resid != 0; resid -= lsize, g++) {
lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g),
SPA_MINBLOCKSIZE);
ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid);
zio_ddt_read_start(zio_t *zio)
{
blkptr_t *bp = zio->io_bp;
+ int p;
ASSERT(BP_GET_DEDUP(bp));
ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
if (ddp_self == NULL)
return (ZIO_PIPELINE_CONTINUE);
- 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_self)
continue;
ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp,
zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde)
{
spa_t *spa = zio->io_spa;
+ int p;
/*
* Note: we compare the original data, not the transformed data,
* pushed the I/O transforms. That's an important optimization
* because otherwise we'd compress/encrypt all dmu_sync() data twice.
*/
- for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
+ for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
zio_t *lio = dde->dde_lead_zio[p];
if (lio != NULL) {
}
}
- for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
+ for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
ddt_phys_t *ddp = &dde->dde_phys[p];
if (ddp->ddp_phys_birth != 0) {
zio_ddt_ditto_write_done(zio_t *zio)
{
int p = DDT_PHYS_DITTO;
- zio_prop_t *zp = &zio->io_prop;
blkptr_t *bp = zio->io_bp;
ddt_t *ddt = ddt_select(zio->io_spa, bp);
ddt_entry_t *dde = zio->io_private;
ddt_phys_t *ddp = &dde->dde_phys[p];
ddt_key_t *ddk = &dde->dde_key;
+ ASSERTV(zio_prop_t *zp = &zio->io_prop);
ddt_enter(ddt);
static void
zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp)
{
+ int g;
+
ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
ASSERT(zio->io_bp_override == NULL);
metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE);
if (gn != NULL) {
- for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
+ for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
zio_dva_unallocate(zio, gn->gn_child[g],
&gn->gn_gbh->zg_blkptr[g]);
}
return (vdev_mirror_ops.vdev_op_io_start(zio));
}
+ /*
+ * We keep track of time-sensitive I/Os so that the scan thread
+ * can quickly react to certain workloads. In particular, we care
+ * about non-scrubbing, top-level reads and writes with the following
+ * characteristics:
+ * - synchronous writes of user data to non-slog devices
+ * - any reads of user data
+ * When these conditions are met, adjust the timestamp of spa_last_io
+ * which allows the scan thread to adjust its workload accordingly.
+ */
+ if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL &&
+ vd == vd->vdev_top && !vd->vdev_islog &&
+ zio->io_bookmark.zb_objset != DMU_META_OBJSET &&
+ zio->io_txg != spa_syncing_txg(spa)) {
+ uint64_t old = spa->spa_last_io;
+ uint64_t new = ddi_get_lbolt64();
+ if (old != new)
+ (void) atomic_cas_64(&spa->spa_last_io, old, new);
+ }
+
align = 1ULL << vd->vdev_top->vdev_ashift;
if (P2PHASE(zio->io_size, align) != 0) {
ASSERT(P2PHASE(zio->io_offset, align) == 0);
ASSERT(P2PHASE(zio->io_size, align) == 0);
- ASSERT(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa));
+ VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa));
/*
* If this is a repair I/O, and there's no self-healing involved --
vdev_t *vd = zio->io_vd;
uint64_t psize = zio->io_size;
zio_t *pio, *pio_next;
+ int c, w;
/*
* If our children haven't all completed,
zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE))
return (ZIO_PIPELINE_STOP);
- for (int c = 0; c < ZIO_CHILD_TYPES; c++)
- for (int w = 0; w < ZIO_WAIT_TYPES; w++)
+ for (c = 0; c < ZIO_CHILD_TYPES; c++)
+ for (w = 0; w < ZIO_WAIT_TYPES; w++)
ASSERT(zio->io_children[c][w] == 0);
if (bp != NULL) {
if ((zio->io_type == ZIO_TYPE_READ ||
zio->io_type == ZIO_TYPE_FREE) &&
+ !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) &&
zio->io_error == ENXIO &&
spa_load_state(spa) == SPA_LOAD_NONE &&
spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE)
git://git.camperquake.de / zfs.git / blobdiff