/*
* The nvlist will be consumed by this call.
*/
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
char *msg;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
* Likewise if the pool is not writeable.
*/
if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
fnvlist_free(nvl);
return;
}
msg = kmem_vasprintf(fmt, adx);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
strfree(msg);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(nvl, tx);
} else {
dsl_sync_task_nowait(spa_get_dsl(spa),
spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
}
/* spa_history_log_sync() will free nvl */
}
int
spa_history_log_nvl(spa_t *spa, nvlist_t *nvl)
{
int err = 0;
dmu_tx_t *tx;
nvlist_t *nvarg;
if (spa_version(spa) < SPA_VERSION_ZPOOL_HISTORY || !spa_writeable(spa))
return (SET_ERROR(EINVAL));
tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
err = dmu_tx_assign(tx, TXG_WAIT);
if (err) {
dmu_tx_abort(tx);
return (err);
}
VERIFY0(nvlist_dup(nvl, &nvarg, KM_SLEEP));
if (spa_history_zone() != NULL) {
fnvlist_add_string(nvarg, ZPOOL_HIST_ZONE,
spa_history_zone());
}
fnvlist_add_uint64(nvarg, ZPOOL_HIST_WHO, crgetruid(CRED()));
/* Kick this off asynchronously; errors are ignored. */
dsl_sync_task_nowait(spa_get_dsl(spa), spa_history_log_sync,
nvarg, 0, ZFS_SPACE_CHECK_NONE, tx);
dmu_tx_commit(tx);
/* spa_history_log_sync will free nvl */
return (err);
}
/*
* Open-context function to add one entry to the new mapping. The new
* entry will be remembered and written from syncing context.
*/
static void
spa_condense_indirect_commit_entry(spa_t *spa,
vdev_indirect_mapping_entry_phys_t *vimep, uint32_t count)
{
spa_condensing_indirect_t *sci = spa->spa_condensing_indirect;
ASSERT3U(count, <, DVA_GET_ASIZE(&vimep->vimep_dst));
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
dmu_tx_hold_space(tx, sizeof (*vimep) + sizeof (count));
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
int txgoff = dmu_tx_get_txg(tx) & TXG_MASK;
/*
* If we are the first entry committed this txg, kick off the sync
* task to write to the MOS on our behalf.
*/
if (list_is_empty(&sci->sci_new_mapping_entries[txgoff])) {
dsl_sync_task_nowait(dmu_tx_pool(tx),
spa_condense_indirect_commit_sync, sci,
0, ZFS_SPACE_CHECK_NONE, tx);
}
vdev_indirect_mapping_entry_t *vime =
kmem_alloc(sizeof (*vime), KM_SLEEP);
vime->vime_mapping = *vimep;
vime->vime_obsolete_count = count;
list_insert_tail(&sci->sci_new_mapping_entries[txgoff], vime);
dmu_tx_commit(tx);
}
/* Takes care of physical writing and limiting # of concurrent ZIOs. */
static int
vdev_initialize_write(vdev_t *vd, uint64_t start, uint64_t size, abd_t *data)
{
spa_t *spa = vd->vdev_spa;
/* Limit inflight initializing I/Os */
mutex_enter(&vd->vdev_initialize_io_lock);
while (vd->vdev_initialize_inflight >= zfs_initialize_limit) {
cv_wait(&vd->vdev_initialize_io_cv,
&vd->vdev_initialize_io_lock);
}
vd->vdev_initialize_inflight++;
mutex_exit(&vd->vdev_initialize_io_lock);
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
uint64_t txg = dmu_tx_get_txg(tx);
spa_config_enter(spa, SCL_STATE_ALL, vd, RW_READER);
mutex_enter(&vd->vdev_initialize_lock);
if (vd->vdev_initialize_offset[txg & TXG_MASK] == 0) {
uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
*guid = vd->vdev_guid;
/* This is the first write of this txg. */
dsl_sync_task_nowait(spa_get_dsl(spa),
vdev_initialize_zap_update_sync, guid, 2,
ZFS_SPACE_CHECK_RESERVED, tx);
}
/*
* We know the vdev struct will still be around since all
* consumers of vdev_free must stop the initialization first.
*/
if (vdev_initialize_should_stop(vd)) {
mutex_enter(&vd->vdev_initialize_io_lock);
ASSERT3U(vd->vdev_initialize_inflight, >, 0);
vd->vdev_initialize_inflight--;
mutex_exit(&vd->vdev_initialize_io_lock);
spa_config_exit(vd->vdev_spa, SCL_STATE_ALL, vd);
mutex_exit(&vd->vdev_initialize_lock);
dmu_tx_commit(tx);
return (SET_ERROR(EINTR));
}
mutex_exit(&vd->vdev_initialize_lock);
vd->vdev_initialize_offset[txg & TXG_MASK] = start + size;
zio_nowait(zio_write_phys(spa->spa_txg_zio[txg & TXG_MASK], vd, start,
size, data, ZIO_CHECKSUM_OFF, vdev_initialize_cb, NULL,
ZIO_PRIORITY_INITIALIZING, ZIO_FLAG_CANFAIL, B_FALSE));
/* vdev_initialize_cb releases SCL_STATE_ALL */
dmu_tx_commit(tx);
return (0);
}
static void
vdev_initialize_change_state(vdev_t *vd, vdev_initializing_state_t new_state)
{
ASSERT(MUTEX_HELD(&vd->vdev_initialize_lock));
spa_t *spa = vd->vdev_spa;
if (new_state == vd->vdev_initialize_state)
return;
/*
* Copy the vd's guid, this will be freed by the sync task.
*/
uint64_t *guid = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
*guid = vd->vdev_guid;
/*
* If we're suspending, then preserving the original start time.
*/
if (vd->vdev_initialize_state != VDEV_INITIALIZE_SUSPENDED) {
vd->vdev_initialize_action_time = gethrestime_sec();
}
vd->vdev_initialize_state = new_state;
dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
dsl_sync_task_nowait(spa_get_dsl(spa), vdev_initialize_zap_update_sync,
guid, 2, ZFS_SPACE_CHECK_RESERVED, tx);
switch (new_state) {
case VDEV_INITIALIZE_ACTIVE:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s activated", vd->vdev_path);
break;
case VDEV_INITIALIZE_SUSPENDED:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s suspended", vd->vdev_path);
break;
case VDEV_INITIALIZE_CANCELED:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s canceled", vd->vdev_path);
break;
case VDEV_INITIALIZE_COMPLETE:
spa_history_log_internal(spa, "initialize", tx,
"vdev=%s complete", vd->vdev_path);
break;
default:
panic("invalid state %llu", (unsigned long long)new_state);
}
dmu_tx_commit(tx);
}
/*
* The nvlist will be consumed by this call.
*/
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
char *msg;
va_list adx1;
int size;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
* Likewise if the pool is not writeable.
*/
if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
fnvlist_free(nvl);
return;
}
va_copy(adx1, adx);
size = vsnprintf(NULL, 0, fmt, adx1) + 1;
msg = kmem_alloc(size, KM_PUSHPAGE);
va_end(adx1);
va_copy(adx1, adx);
(void) vsprintf(msg, fmt, adx1);
va_end(adx1);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
kmem_free(msg, size);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(nvl, tx);
} else {
dsl_sync_task_nowait(spa_get_dsl(spa),
spa_history_log_sync, nvl, 0, tx);
}
/* spa_history_log_sync() will free nvl */
}
