kmem_cache_t *zio_link_cache;
kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
+int zio_bulk_flags = 0;
#ifdef _KERNEL
extern vmem_t *zio_alloc_arena;
int zio_buf_debug_limit = 0;
#endif
+static inline void __zio_execute(zio_t *zio);
+
void
zio_init(void)
{
(void) sprintf(name, "zio_buf_%lu", (ulong_t)size);
zio_buf_cache[c] = kmem_cache_create(name, size,
align, NULL, NULL, NULL, NULL, NULL,
- size > zio_buf_debug_limit ? KMC_NODEBUG : 0);
+ (size > zio_buf_debug_limit ? KMC_NODEBUG : 0) |
+ zio_bulk_flags);
(void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size);
zio_data_buf_cache[c] = kmem_cache_create(name, size,
align, NULL, NULL, NULL, NULL, data_alloc_arena,
- size > zio_buf_debug_limit ? KMC_NODEBUG : 0);
+ (size > zio_buf_debug_limit ? KMC_NODEBUG : 0) |
+ zio_bulk_flags);
}
}
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);
return (waiting);
}
-static void
+__attribute__((always_inline))
+static inline void
zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait)
{
uint64_t *countp = &pio->io_children[zio->io_child_type][wait];
if (--*countp == 0 && pio->io_stall == countp) {
pio->io_stall = NULL;
mutex_exit(&pio->io_lock);
- zio_execute(pio);
+ __zio_execute(pio);
} else {
mutex_exit(&pio->io_lock);
}
*/
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 {
{
spa_t *spa = zio->io_spa;
zio_type_t t = zio->io_type;
- int flags = TQ_SLEEP | (cutinline ? TQ_FRONT : 0);
+ int flags = TQ_NOSLEEP | (cutinline ? TQ_FRONT : 0);
/*
* If we're a config writer or a probe, the normal issue and
q++;
ASSERT3U(q, <, ZIO_TASKQ_TYPES);
- (void) taskq_dispatch(spa->spa_zio_taskq[t][q],
- (task_func_t *)zio_execute, zio, flags);
+
+ while (taskq_dispatch(spa->spa_zio_taskq[t][q],
+ (task_func_t *)zio_execute, zio, flags) == 0); /* do nothing */
}
static boolean_t
{
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);
*/
static zio_pipe_stage_t *zio_pipeline[];
+/*
+ * zio_execute() is a wrapper around the static function
+ * __zio_execute() so that we can force __zio_execute() to be
+ * inlined. This reduces stack overhead which is important
+ * because __zio_execute() is called recursively in several zio
+ * code paths. zio_execute() itself cannot be inlined because
+ * it is externally visible.
+ */
void
zio_execute(zio_t *zio)
+{
+ __zio_execute(zio);
+}
+
+__attribute__((always_inline))
+static inline void
+__zio_execute(zio_t *zio)
{
zio->io_executor = curthread;
}
}
+
/*
* ==========================================================================
* Initiate I/O, either sync or async
zio->io_waiter = curthread;
- zio_execute(zio);
+ __zio_execute(zio);
mutex_enter(&zio->io_lock);
while (zio->io_executor != NULL)
zio_add_child(spa->spa_async_zio_root, zio);
}
- zio_execute(zio);
+ __zio_execute(zio);
}
/*
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);
* responsibility of the caller to wait on him.
*/
if (!(pio->io_flags & ZIO_FLAG_GODFATHER))
- zio_execute(pio);
+ __zio_execute(pio);
}
void
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 --
static int
zio_done(zio_t *zio)
{
- spa_t *spa = zio->io_spa;
- zio_t *lio = zio->io_logical;
- blkptr_t *bp = zio->io_bp;
- 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) {
- ASSERT(bp->blk_pad[0] == 0);
- ASSERT(bp->blk_pad[1] == 0);
- ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 ||
- (bp == zio_unique_parent(zio)->io_bp));
- if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) &&
+ if (zio->io_bp != NULL) {
+ ASSERT(zio->io_bp->blk_pad[0] == 0);
+ ASSERT(zio->io_bp->blk_pad[1] == 0);
+ ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 ||
+ (zio->io_bp == zio_unique_parent(zio)->io_bp));
+ if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) &&
zio->io_bp_override == NULL &&
!(zio->io_flags & ZIO_FLAG_IO_REPAIR)) {
- ASSERT(!BP_SHOULD_BYTESWAP(bp));
- ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp));
- ASSERT(BP_COUNT_GANG(bp) == 0 ||
- (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp)));
+ ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp));
+ ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp));
+ ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 ||
+ (BP_COUNT_GANG(zio->io_bp) == BP_GET_NDVAS(zio->io_bp)));
}
}
while (zio->io_cksum_report != NULL) {
zio_cksum_report_t *zcr = zio->io_cksum_report;
uint64_t align = zcr->zcr_align;
- uint64_t asize = P2ROUNDUP(psize, align);
+ uint64_t asize = P2ROUNDUP(zio->io_size, align);
char *abuf = zio->io_data;
- if (asize != psize) {
+ if (asize != zio->io_size) {
abuf = zio_buf_alloc(asize);
- bcopy(zio->io_data, abuf, psize);
- bzero(abuf + psize, asize - psize);
+ bcopy(zio->io_data, abuf, zio->io_size);
+ bzero(abuf + zio->io_size, asize - zio->io_size);
}
zio->io_cksum_report = zcr->zcr_next;
zcr->zcr_finish(zcr, abuf);
zfs_ereport_free_checksum(zcr);
- if (asize != psize)
+ if (asize != zio->io_size)
zio_buf_free(abuf, asize);
}
}
zio_pop_transforms(zio); /* note: may set zio->io_error */
- vdev_stat_update(zio, psize);
+ vdev_stat_update(zio, zio->io_size);
if (zio->io_error) {
/*
* at the block level. We ignore these errors if the
* device is currently unavailable.
*/
- if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd))
- zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0);
+ if (zio->io_error != ECKSUM && zio->io_vd != NULL &&
+ !vdev_is_dead(zio->io_vd))
+ zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa,
+ zio->io_vd, zio, 0, 0);
if ((zio->io_error == EIO || !(zio->io_flags &
(ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) &&
- zio == lio) {
+ zio == zio->io_logical) {
/*
* For logical I/O requests, tell the SPA to log the
* error and generate a logical data ereport.
*/
- spa_log_error(spa, zio);
- zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio,
+ spa_log_error(zio->io_spa, zio);
+ zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, NULL, zio,
0, 0);
}
}
- if (zio->io_error && zio == lio) {
+ if (zio->io_error && zio == zio->io_logical) {
/*
* Determine whether zio should be reexecuted. This will
* propagate all the way to the root via zio_notify_parent().
*/
- ASSERT(vd == NULL && bp != NULL);
+ ASSERT(zio->io_vd == NULL && zio->io_bp != NULL);
ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
if (IO_IS_ALLOCATING(zio) &&
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)
+ spa_load_state(zio->io_spa) == SPA_LOAD_NONE &&
+ spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE)
zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute)
if ((zio->io_error || zio->io_reexecute) &&
IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio &&
!(zio->io_flags & ZIO_FLAG_IO_REWRITE))
- zio_dva_unallocate(zio, zio->io_gang_tree, bp);
+ zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp);
zio_gang_tree_free(&zio->io_gang_tree);
* We'd fail again if we reexecuted now, so suspend
* until conditions improve (e.g. device comes online).
*/
- zio_suspend(spa, zio);
+ zio_suspend(zio->io_spa, zio);
} else {
/*
* Reexecution is potentially a huge amount of work.
* Hand it off to the otherwise-unused claim taskq.
*/
(void) taskq_dispatch(
- spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE],
+ zio->io_spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE],
(task_func_t *)zio_reexecute, zio, TQ_SLEEP);
}
return (ZIO_PIPELINE_STOP);
zio_checksum_verify,
zio_done
};
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+/* Fault injection */
+EXPORT_SYMBOL(zio_injection_enabled);
+EXPORT_SYMBOL(zio_inject_fault);
+EXPORT_SYMBOL(zio_inject_list_next);
+EXPORT_SYMBOL(zio_clear_fault);
+EXPORT_SYMBOL(zio_handle_fault_injection);
+EXPORT_SYMBOL(zio_handle_device_injection);
+EXPORT_SYMBOL(zio_handle_label_injection);
+EXPORT_SYMBOL(zio_priority_table);
+EXPORT_SYMBOL(zio_type_name);
+
+module_param(zio_bulk_flags, int, 0644);
+MODULE_PARM_DESC(zio_bulk_flags, "Additional flags to pass to bulk buffers");
+#endif
projects / mirror_zfs.git / blobdiff