/* ARGSUSED */
int
zil_clear_log_chain(char *osname, void *txarg)
{
zilog_t *zilog;
zil_header_t *zh;
objset_t *os;
dmu_tx_t *tx;
int error;
error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_USER, &os);
if (error) {
cmn_err(CE_WARN, "can't open objset for %s", osname);
return (0);
}
zilog = dmu_objset_zil(os);
tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
zh = zil_header_in_syncing_context(zilog);
BP_ZERO(&zh->zh_log);
dsl_dataset_dirty(dmu_objset_ds(os), tx);
dmu_tx_commit(tx);
dmu_objset_close(os);
return (0);
}
/* called from dsl for meta-objset */
objset_t *
dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
dmu_objset_type_t type, dmu_tx_t *tx)
{
objset_t *os;
dnode_t *mdn;
ASSERT(dmu_tx_is_syncing(tx));
if (ds != NULL)
VERIFY0(dmu_objset_from_ds(ds, &os));
else
VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
mdn = DMU_META_DNODE(os);
dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
/*
* We don't want to have to increase the meta-dnode's nlevels
* later, because then we could do it in quescing context while
* we are also accessing it in open context.
*
* This precaution is not necessary for the MOS (ds == NULL),
* because the MOS is only updated in syncing context.
* This is most fortunate: the MOS is the only objset that
* needs to be synced multiple times as spa_sync() iterates
* to convergence, so minimizing its dn_nlevels matters.
*/
if (ds != NULL) {
int levels = 1;
/*
* Determine the number of levels necessary for the meta-dnode
* to contain DN_MAX_OBJECT dnodes.
*/
while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
(levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
DN_MAX_OBJECT * sizeof (dnode_phys_t))
levels++;
mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
mdn->dn_nlevels = levels;
}
ASSERT(type != DMU_OST_NONE);
ASSERT(type != DMU_OST_ANY);
ASSERT(type < DMU_OST_NUMTYPES);
os->os_phys->os_type = type;
if (dmu_objset_userused_enabled(os)) {
os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
os->os_flags = os->os_phys->os_flags;
}
dsl_dataset_dirty(ds, tx);
return (os);
}
/*
* In one tx, free all log blocks and clear the log header.
* If keep_first is set, then we're replaying a log with no content.
* We want to keep the first block, however, so that the first
* synchronous transaction doesn't require a txg_wait_synced()
* in zil_create(). We don't need to txg_wait_synced() here either
* when keep_first is set, because both zil_create() and zil_destroy()
* will wait for any in-progress destroys to complete.
*/
void
zil_destroy(zilog_t *zilog, boolean_t keep_first)
{
const zil_header_t *zh = zilog->zl_header;
lwb_t *lwb;
dmu_tx_t *tx;
uint64_t txg;
/*
* Wait for any previous destroy to complete.
*/
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
if (BP_IS_HOLE(&zh->zh_log))
return;
tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
mutex_enter(&zilog->zl_lock);
/*
* It is possible for the ZIL to get the previously mounted zilog
* structure of the same dataset if quickly remounted and the dbuf
* eviction has not completed. In this case we can see a non
* empty lwb list and keep_first will be set. We fix this by
* clearing the keep_first. This will be slower but it's very rare.
*/
if (!list_is_empty(&zilog->zl_lwb_list) && keep_first)
keep_first = B_FALSE;
ASSERT3U(zilog->zl_destroy_txg, <, txg);
zilog->zl_destroy_txg = txg;
zilog->zl_keep_first = keep_first;
if (!list_is_empty(&zilog->zl_lwb_list)) {
ASSERT(zh->zh_claim_txg == 0);
ASSERT(!keep_first);
while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
list_remove(&zilog->zl_lwb_list, lwb);
if (lwb->lwb_buf != NULL)
zio_buf_free(lwb->lwb_buf, lwb->lwb_sz);
zio_free_blk(zilog->zl_spa, &lwb->lwb_blk, txg);
kmem_cache_free(zil_lwb_cache, lwb);
}
} else {
if (!keep_first) {
(void) zil_parse(zilog, zil_free_log_block,
zil_free_log_record, tx, zh->zh_claim_txg);
}
}
mutex_exit(&zilog->zl_lock);
dmu_tx_commit(tx);
}
int
zil_claim(char *osname, void *txarg)
{
dmu_tx_t *tx = txarg;
uint64_t first_txg = dmu_tx_get_txg(tx);
zilog_t *zilog;
zil_header_t *zh;
objset_t *os;
int error;
error = dmu_objset_open(osname, DMU_OST_ANY, DS_MODE_USER, &os);
if (error) {
cmn_err(CE_WARN, "can't open objset for %s", osname);
return (0);
}
zilog = dmu_objset_zil(os);
zh = zil_header_in_syncing_context(zilog);
/*
* Record here whether the zil has any records to replay.
* If the header block pointer is null or the block points
* to the stubby then we know there are no valid log records.
* We use the header to store this state as the the zilog gets
* freed later in dmu_objset_close().
* The flags (and the rest of the header fields) are cleared in
* zil_sync() as a result of a zil_destroy(), after replaying the log.
*
* Note, the intent log can be empty but still need the
* stubby to be claimed.
*/
if (!zil_empty(zilog))
zh->zh_flags |= ZIL_REPLAY_NEEDED;
/*
* Claim all log blocks if we haven't already done so, and remember
* the highest claimed sequence number. This ensures that if we can
* read only part of the log now (e.g. due to a missing device),
* but we can read the entire log later, we will not try to replay
* or destroy beyond the last block we successfully claimed.
*/
ASSERT3U(zh->zh_claim_txg, <=, first_txg);
if (zh->zh_claim_txg == 0 && !BP_IS_HOLE(&zh->zh_log)) {
zh->zh_claim_txg = first_txg;
zh->zh_claim_seq = zil_parse(zilog, zil_claim_log_block,
zil_claim_log_record, tx, first_txg);
dsl_dataset_dirty(dmu_objset_ds(os), tx);
}
ASSERT3U(first_txg, ==, (spa_last_synced_txg(zilog->zl_spa) + 1));
dmu_objset_close(os);
return (0);
}
/*
* Close an intent log.
*/
void
zil_close(zilog_t *zilog)
{
/*
* If the log isn't already committed, mark the objset dirty
* (so zil_sync() will be called) and wait for that txg to sync.
*/
if (!zil_is_committed(zilog)) {
uint64_t txg;
dmu_tx_t *tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
dmu_tx_commit(tx);
txg_wait_synced(zilog->zl_dmu_pool, txg);
}
taskq_destroy(zilog->zl_clean_taskq);
zilog->zl_clean_taskq = NULL;
zilog->zl_get_data = NULL;
zil_itx_clean(zilog);
ASSERT(list_head(&zilog->zl_itx_list) == NULL);
}
/* called from dsl for meta-objset */
objset_t *
dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
dmu_objset_type_t type, dsl_crypto_ctx_t *dcc, dmu_tx_t *tx)
{
objset_t *os;
dnode_t *mdn;
ASSERT(dmu_tx_is_syncing(tx));
if (ds != NULL)
VERIFY(0 == dmu_objset_from_ds(ds, &os));
else
VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
mdn = DMU_META_DNODE(os);
dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
/*
* We don't want to have to increase the meta-dnode's nlevels
* later, because then we could do it in quescing context while
* we are also accessing it in open context.
*
* This precaution is not necessary for the MOS (ds == NULL),
* because the MOS is only updated in syncing context.
* This is most fortunate: the MOS is the only objset that
* needs to be synced multiple times as spa_sync() iterates
* to convergence, so minimizing its dn_nlevels matters.
*/
if (ds != NULL) {
int levels = 1;
/*
* Determine the number of levels necessary for the meta-dnode
* to contain DN_MAX_OBJECT dnodes.
*/
while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
(levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
DN_MAX_OBJECT * sizeof (dnode_phys_t))
levels++;
mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
mdn->dn_nlevels = levels;
}
ASSERT(type != DMU_OST_NONE);
ASSERT(type != DMU_OST_ANY);
ASSERT(type < DMU_OST_NUMTYPES);
/*
* Note: although we should not be dirtying the objset outside of
* sync context, the os_type is used directly from the os_phys
* in dsl_scan so it may not be save to put os_type in the in-core
* objset_t and set it in the phys in sync context (as with os_flags)
*/
dmu_objset_dirty(os);
os->os_phys->os_type = type;
if (dmu_objset_userused_enabled(os)) {
os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
// FIXME - removed in newer.
// os->os_flags = os->os_phys->os_flags;
}
if (dcc != NULL && dcc->dcc_crypt != ZIO_CRYPT_INHERIT) {
os->os_crypt = zio_crypt_select(dcc->dcc_crypt,
ZIO_CRYPT_ON_VALUE);
}
dsl_dataset_dirty(ds, tx);
return (os);
}
/*
* Start a log block write and advance to the next log block.
* Calls are serialized.
*/
static lwb_t *
zil_lwb_write_start(zilog_t *zilog, lwb_t *lwb)
{
lwb_t *nlwb;
zil_trailer_t *ztp = (zil_trailer_t *)(lwb->lwb_buf + lwb->lwb_sz) - 1;
spa_t *spa = zilog->zl_spa;
blkptr_t *bp = &ztp->zit_next_blk;
uint64_t txg;
uint64_t zil_blksz;
int error;
ASSERT(lwb->lwb_nused <= ZIL_BLK_DATA_SZ(lwb));
/*
* Allocate the next block and save its address in this block
* before writing it in order to establish the log chain.
* Note that if the allocation of nlwb synced before we wrote
* the block that points at it (lwb), we'd leak it if we crashed.
* Therefore, we don't do txg_rele_to_sync() until zil_lwb_write_done().
*/
txg = txg_hold_open(zilog->zl_dmu_pool, &lwb->lwb_txgh);
txg_rele_to_quiesce(&lwb->lwb_txgh);
/*
* Pick a ZIL blocksize. We request a size that is the
* maximum of the previous used size, the current used size and
* the amount waiting in the queue.
*/
zil_blksz = MAX(zilog->zl_prev_used,
zilog->zl_cur_used + sizeof (*ztp));
zil_blksz = MAX(zil_blksz, zilog->zl_itx_list_sz + sizeof (*ztp));
zil_blksz = P2ROUNDUP_TYPED(zil_blksz, ZIL_MIN_BLKSZ, uint64_t);
if (zil_blksz > ZIL_MAX_BLKSZ)
zil_blksz = ZIL_MAX_BLKSZ;
BP_ZERO(bp);
/* pass the old blkptr in order to spread log blocks across devs */
error = zio_alloc_blk(spa, zil_blksz, bp, &lwb->lwb_blk, txg);
if (error) {
dmu_tx_t *tx = dmu_tx_create_assigned(zilog->zl_dmu_pool, txg);
/*
* We dirty the dataset to ensure that zil_sync() will
* be called to remove this lwb from our zl_lwb_list.
* Failing to do so, may leave an lwb with a NULL lwb_buf
* hanging around on the zl_lwb_list.
*/
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
dmu_tx_commit(tx);
/*
* Since we've just experienced an allocation failure so we
* terminate the current lwb and send it on its way.
*/
ztp->zit_pad = 0;
ztp->zit_nused = lwb->lwb_nused;
ztp->zit_bt.zbt_cksum = lwb->lwb_blk.blk_cksum;
zio_nowait(lwb->lwb_zio);
/*
* By returning NULL the caller will call tx_wait_synced()
*/
return (NULL);
}
ASSERT3U(bp->blk_birth, ==, txg);
ztp->zit_pad = 0;
ztp->zit_nused = lwb->lwb_nused;
ztp->zit_bt.zbt_cksum = lwb->lwb_blk.blk_cksum;
bp->blk_cksum = lwb->lwb_blk.blk_cksum;
bp->blk_cksum.zc_word[ZIL_ZC_SEQ]++;
/*
* Allocate a new log write buffer (lwb).
*/
nlwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
nlwb->lwb_zilog = zilog;
nlwb->lwb_blk = *bp;
nlwb->lwb_nused = 0;
nlwb->lwb_sz = BP_GET_LSIZE(&nlwb->lwb_blk);
nlwb->lwb_buf = zio_buf_alloc(nlwb->lwb_sz);
nlwb->lwb_max_txg = txg;
nlwb->lwb_zio = NULL;
/*
* Put new lwb at the end of the log chain
*/
mutex_enter(&zilog->zl_lock);
list_insert_tail(&zilog->zl_lwb_list, nlwb);
mutex_exit(&zilog->zl_lock);
/* Record the block for later vdev flushing */
zil_add_block(zilog, &lwb->lwb_blk);
/*
* kick off the write for the old log block
*/
dprintf_bp(&lwb->lwb_blk, "lwb %p txg %llu: ", lwb, txg);
ASSERT(lwb->lwb_zio);
zio_nowait(lwb->lwb_zio);
return (nlwb);
}
/*
* Create an on-disk intent log.
*/
static void
zil_create(zilog_t *zilog)
{
const zil_header_t *zh = zilog->zl_header;
lwb_t *lwb;
uint64_t txg = 0;
dmu_tx_t *tx = NULL;
blkptr_t blk;
int error = 0;
/*
* Wait for any previous destroy to complete.
*/
txg_wait_synced(zilog->zl_dmu_pool, zilog->zl_destroy_txg);
ASSERT(zh->zh_claim_txg == 0);
ASSERT(zh->zh_replay_seq == 0);
blk = zh->zh_log;
/*
* If we don't already have an initial log block or we have one
* but it's the wrong endianness then allocate one.
*/
if (BP_IS_HOLE(&blk) || BP_SHOULD_BYTESWAP(&blk)) {
tx = dmu_tx_create(zilog->zl_os);
(void) dmu_tx_assign(tx, TXG_WAIT);
dsl_dataset_dirty(dmu_objset_ds(zilog->zl_os), tx);
txg = dmu_tx_get_txg(tx);
if (!BP_IS_HOLE(&blk)) {
zio_free_blk(zilog->zl_spa, &blk, txg);
BP_ZERO(&blk);
}
error = zio_alloc_blk(zilog->zl_spa, ZIL_MIN_BLKSZ, &blk,
NULL, txg);
if (error == 0)
zil_init_log_chain(zilog, &blk);
}
/*
* Allocate a log write buffer (lwb) for the first log block.
*/
if (error == 0) {
lwb = kmem_cache_alloc(zil_lwb_cache, KM_SLEEP);
lwb->lwb_zilog = zilog;
lwb->lwb_blk = blk;
lwb->lwb_nused = 0;
lwb->lwb_sz = BP_GET_LSIZE(&lwb->lwb_blk);
lwb->lwb_buf = zio_buf_alloc(lwb->lwb_sz);
lwb->lwb_max_txg = txg;
lwb->lwb_zio = NULL;
mutex_enter(&zilog->zl_lock);
list_insert_tail(&zilog->zl_lwb_list, lwb);
mutex_exit(&zilog->zl_lock);
}
/*
* If we just allocated the first log block, commit our transaction
* and wait for zil_sync() to stuff the block poiner into zh_log.
* (zh is part of the MOS, so we cannot modify it in open context.)
*/
if (tx != NULL) {
dmu_tx_commit(tx);
txg_wait_synced(zilog->zl_dmu_pool, txg);
}
ASSERT(bcmp(&blk, &zh->zh_log, sizeof (blk)) == 0);
}