/* 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);
}
/*
* Create and return an immediate write ZIL transaction.
*/
itx_t *
zvol_immediate_itx(offset_t off, ssize_t len, char *addr)
{
itx_t *itx;
lr_write_t *lr;
itx = zil_itx_create(TX_WRITE, sizeof (*lr) + len);
lr = (lr_write_t *)&itx->itx_lr;
lr->lr_foid = ZVOL_OBJ;
lr->lr_offset = off;
lr->lr_length = len;
lr->lr_blkoff = 0;
BP_ZERO(&lr->lr_blkptr);
bcopy(addr, (char *)itx + offsetof(itx_t, itx_lr) +
sizeof (*lr), len);
itx->itx_wr_state = WR_COPIED;
return (itx);
}
void
zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
znode_t *zp, offset_t off, ssize_t resid, int ioflag,
zil_callback_t callback, void *callback_data)
{
itx_wr_state_t write_state;
boolean_t slogging;
uintptr_t fsync_cnt;
ssize_t immediate_write_sz;
if (zil_replaying(zilog, tx) || zp->z_unlinked) {
if (callback != NULL)
callback(callback_data);
return;
}
immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
? 0 : (ssize_t)zfs_immediate_write_sz;
slogging = spa_has_slogs(zilog->zl_spa) &&
(zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
if (resid > immediate_write_sz && !slogging && resid <= zp->z_blksz)
write_state = WR_INDIRECT;
else if (ioflag & (FSYNC | FDSYNC))
write_state = WR_COPIED;
else
write_state = WR_NEED_COPY;
if ((fsync_cnt = (uintptr_t)tsd_get(zfs_fsyncer_key)) != 0) {
(void) tsd_set(zfs_fsyncer_key, (void *)(fsync_cnt - 1));
}
while (resid) {
itx_t *itx;
lr_write_t *lr;
ssize_t len;
/*
* If the write would overflow the largest block then split it.
*/
if (write_state != WR_INDIRECT && resid > ZIL_MAX_LOG_DATA)
len = SPA_MAXBLOCKSIZE >> 1;
else
len = resid;
itx = zil_itx_create(txtype, sizeof (*lr) +
(write_state == WR_COPIED ? len : 0));
lr = (lr_write_t *)&itx->itx_lr;
if (write_state == WR_COPIED && dmu_read(ZTOZSB(zp)->z_os,
zp->z_id, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
zil_itx_destroy(itx);
itx = zil_itx_create(txtype, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
write_state = WR_NEED_COPY;
}
itx->itx_wr_state = write_state;
if (write_state == WR_NEED_COPY)
itx->itx_sod += len;
lr->lr_foid = zp->z_id;
lr->lr_offset = off;
lr->lr_length = len;
lr->lr_blkoff = 0;
BP_ZERO(&lr->lr_blkptr);
itx->itx_private = ZTOZSB(zp);
if (!(ioflag & (FSYNC | FDSYNC)) && (zp->z_sync_cnt == 0) &&
(fsync_cnt == 0))
itx->itx_sync = B_FALSE;
itx->itx_callback = callback;
itx->itx_callback_data = callback_data;
zil_itx_assign(zilog, itx, tx);
off += len;
resid -= len;
}
static void
zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
uint64_t size, int sync)
{
uint32_t blocksize = zv->zv_volblocksize;
zilog_t *zilog = zv->zv_zilog;
boolean_t slogging;
ssize_t immediate_write_sz;
if (zil_replaying(zilog, tx))
return;
immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
? 0 : zvol_immediate_write_sz;
slogging = spa_has_slogs(zilog->zl_spa) &&
(zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
while (size) {
itx_t *itx;
lr_write_t *lr;
ssize_t len;
itx_wr_state_t write_state;
/*
* Unlike zfs_log_write() we can be called with
* up to DMU_MAX_ACCESS/2 (5MB) writes.
*/
if (blocksize > immediate_write_sz && !slogging &&
size >= blocksize && offset % blocksize == 0) {
write_state = WR_INDIRECT; /* uses dmu_sync */
len = blocksize;
} else if (sync) {
write_state = WR_COPIED;
len = MIN(ZIL_MAX_LOG_DATA, size);
} else {
write_state = WR_NEED_COPY;
len = MIN(ZIL_MAX_LOG_DATA, size);
}
itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
(write_state == WR_COPIED ? len : 0));
lr = (lr_write_t *)&itx->itx_lr;
if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
zil_itx_destroy(itx);
itx = zil_itx_create(TX_WRITE, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
write_state = WR_NEED_COPY;
}
itx->itx_wr_state = write_state;
if (write_state == WR_NEED_COPY)
itx->itx_sod += len;
lr->lr_foid = ZVOL_OBJ;
lr->lr_offset = offset;
lr->lr_length = len;
lr->lr_blkoff = 0;
BP_ZERO(&lr->lr_blkptr);
itx->itx_private = zv;
itx->itx_sync = sync;
(void) zil_itx_assign(zilog, itx, tx);
offset += len;
size -= len;
}
}
void
zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
znode_t *zp, offset_t off, ssize_t resid, int ioflag)
{
uint32_t blocksize = zp->z_blksz;
itx_wr_state_t write_state;
uintptr_t fsync_cnt;
if (zil_replaying(zilog, tx) || zp->z_unlinked)
return;
if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
write_state = WR_INDIRECT;
else if (!spa_has_slogs(zilog->zl_spa) &&
resid >= zfs_immediate_write_sz)
write_state = WR_INDIRECT;
else if (ioflag & (FSYNC | FDSYNC))
write_state = WR_COPIED;
else
write_state = WR_NEED_COPY;
if ((fsync_cnt = (uintptr_t)tsd_get(zfs_fsyncer_key)) != 0) {
(void) tsd_set(zfs_fsyncer_key, (void *)(fsync_cnt - 1));
}
while (resid) {
itx_t *itx;
lr_write_t *lr;
itx_wr_state_t wr_state = write_state;
ssize_t len = resid;
if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
wr_state = WR_NEED_COPY;
else if (wr_state == WR_INDIRECT)
len = MIN(blocksize - P2PHASE(off, blocksize), resid);
itx = zil_itx_create(txtype, sizeof (*lr) +
(wr_state == WR_COPIED ? len : 0));
lr = (lr_write_t *)&itx->itx_lr;
if (wr_state == WR_COPIED && dmu_read(zp->z_zfsvfs->z_os,
zp->z_id, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
zil_itx_destroy(itx);
itx = zil_itx_create(txtype, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
wr_state = WR_NEED_COPY;
}
itx->itx_wr_state = wr_state;
lr->lr_foid = zp->z_id;
lr->lr_offset = off;
lr->lr_length = len;
lr->lr_blkoff = 0;
BP_ZERO(&lr->lr_blkptr);
itx->itx_private = zp->z_zfsvfs;
if (!(ioflag & (FSYNC | FDSYNC)) && (zp->z_sync_cnt == 0) &&
(fsync_cnt == 0))
itx->itx_sync = B_FALSE;
zil_itx_assign(zilog, itx, tx);
off += len;
resid -= len;
}
}
int
zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t len,
char *addr)
{
dmu_object_info_t doi;
ssize_t nbytes;
itx_t *itx;
lr_write_t *lr;
objset_t *os;
dmu_buf_t *db;
uint64_t txg;
uint64_t boff;
int error;
uint32_t blocksize;
/* handle common case */
if (len <= zvol_immediate_write_sz) {
itx = zvol_immediate_itx(off, len, addr);
(void) zil_itx_assign(zv->zv_zilog, itx, tx);
return (0);
}
txg = dmu_tx_get_txg(tx);
os = zv->zv_objset;
/*
* We need to dmu_sync() each block in the range.
* For this we need the blocksize.
*/
error = dmu_object_info(os, ZVOL_OBJ, &doi);
if (error)
return (error);
blocksize = doi.doi_data_block_size;
/*
* We need to immediate write or dmu_sync() each block in the range.
*/
while (len) {
nbytes = MIN(len, blocksize - P2PHASE(off, blocksize));
if (nbytes <= zvol_immediate_write_sz) {
itx = zvol_immediate_itx(off, nbytes, addr);
} else {
boff = P2ALIGN_TYPED(off, blocksize, uint64_t);
itx = zil_itx_create(TX_WRITE, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
lr->lr_foid = ZVOL_OBJ;
lr->lr_offset = off;
lr->lr_length = nbytes;
lr->lr_blkoff = off - boff;
BP_ZERO(&lr->lr_blkptr);
/* XXX - we should do these IOs in parallel */
VERIFY(0 == dmu_buf_hold(os, ZVOL_OBJ, boff,
FTAG, &db));
ASSERT(boff == db->db_offset);
error = dmu_sync(NULL, db, &lr->lr_blkptr,
txg, NULL, NULL);
dmu_buf_rele(db, FTAG);
if (error) {
kmem_free(itx, offsetof(itx_t, itx_lr));
return (error);
}
itx->itx_wr_state = WR_COPIED;
}
(void) zil_itx_assign(zv->zv_zilog, itx, tx);
len -= nbytes;
off += nbytes;
}
return (0);
}
void
zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
znode_t *zp, offset_t off, ssize_t resid, int ioflag)
{
itx_wr_state_t write_state;
boolean_t slogging;
uintptr_t fsync_cnt;
if (zilog == NULL || zp->z_unlinked)
return;
/*
* Writes are handled in three different ways:
*
* WR_INDIRECT:
* If the write is greater than zfs_immediate_write_sz and there are
* no separate logs in this pool then later *if* we need to log the
* write then dmu_sync() is used to immediately write the block and
* its block pointer is put in the log record.
* WR_COPIED:
* If we know we'll immediately be committing the
* transaction (FDSYNC (O_DSYNC)), the we allocate a larger
* log record here for the data and copy the data in.
* WR_NEED_COPY:
* Otherwise we don't allocate a buffer, and *if* we need to
* flush the write later then a buffer is allocated and
* we retrieve the data using the dmu.
*/
slogging = spa_has_slogs(zilog->zl_spa);
if (resid > zfs_immediate_write_sz && !slogging)
write_state = WR_INDIRECT;
else if (ioflag & FDSYNC)
write_state = WR_COPIED;
else
write_state = WR_NEED_COPY;
#ifndef __APPLE__
if ((fsync_cnt = (uintptr_t)tsd_get(zfs_fsyncer_key)) != 0) {
(void) tsd_set(zfs_fsyncer_key, (void *)(fsync_cnt - 1));
}
#endif
while (resid) {
itx_t *itx;
lr_write_t *lr;
ssize_t len;
/*
* If there are slogs and the write would overflow the largest
* block, then because we don't want to use the main pool
* to dmu_sync, we have to split the write.
*/
if (slogging && resid > ZIL_MAX_LOG_DATA)
len = SPA_MAXBLOCKSIZE >> 1;
else
len = resid;
itx = zil_itx_create(txtype, sizeof (*lr) +
(write_state == WR_COPIED ? len : 0));
lr = (lr_write_t *)&itx->itx_lr;
if (write_state == WR_COPIED && dmu_read(zp->z_zfsvfs->z_os,
zp->z_id, off, len, lr + 1) != 0) {
kmem_free(itx, offsetof(itx_t, itx_lr) +
itx->itx_lr.lrc_reclen);
itx = zil_itx_create(txtype, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
write_state = WR_NEED_COPY;
}
itx->itx_wr_state = write_state;
lr->lr_foid = zp->z_id;
lr->lr_offset = off;
lr->lr_length = len;
lr->lr_blkoff = 0;
BP_ZERO(&lr->lr_blkptr);
itx->itx_private = zp->z_zfsvfs;
if ((zp->z_sync_cnt != 0) || (fsync_cnt != 0))
itx->itx_sync = B_TRUE;
else
itx->itx_sync = B_FALSE;
zp->z_last_itx = zil_itx_assign(zilog, itx, tx);
off += len;
resid -= len;
}
/*
* 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);
}
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);
if (zilog->zl_spa->spa_log_state == SPA_LOG_CLEAR) {
if (!BP_IS_HOLE(&zh->zh_log))
zio_free_blk(zilog->zl_spa, &zh->zh_log, first_txg);
BP_ZERO(&zh->zh_log);
dsl_dataset_dirty(dmu_objset_ds(os), tx);
}
/*
* 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;
dsl_dataset_dirty(dmu_objset_ds(os), tx);
}
/*
* 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);
}
/*
* 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);
}
/*
* Called in syncing context to free committed log blocks and update log header.
*/
void
zil_sync(zilog_t *zilog, dmu_tx_t *tx)
{
zil_header_t *zh = zil_header_in_syncing_context(zilog);
uint64_t txg = dmu_tx_get_txg(tx);
spa_t *spa = zilog->zl_spa;
lwb_t *lwb;
/*
* We don't zero out zl_destroy_txg, so make sure we don't try
* to destroy it twice.
*/
if (spa_sync_pass(spa) != 1)
return;
mutex_enter(&zilog->zl_lock);
ASSERT(zilog->zl_stop_sync == 0);
zh->zh_replay_seq = zilog->zl_replayed_seq[txg & TXG_MASK];
if (zilog->zl_destroy_txg == txg) {
blkptr_t blk = zh->zh_log;
ASSERT(list_head(&zilog->zl_lwb_list) == NULL);
bzero(zh, sizeof (zil_header_t));
bzero(zilog->zl_replayed_seq, sizeof (zilog->zl_replayed_seq));
if (zilog->zl_keep_first) {
/*
* If this block was part of log chain that couldn't
* be claimed because a device was missing during
* zil_claim(), but that device later returns,
* then this block could erroneously appear valid.
* To guard against this, assign a new GUID to the new
* log chain so it doesn't matter what blk points to.
*/
zil_init_log_chain(zilog, &blk);
zh->zh_log = blk;
}
}
while ((lwb = list_head(&zilog->zl_lwb_list)) != NULL) {
zh->zh_log = lwb->lwb_blk;
if (lwb->lwb_buf != NULL || lwb->lwb_max_txg > txg)
break;
list_remove(&zilog->zl_lwb_list, lwb);
zio_free_blk(spa, &lwb->lwb_blk, txg);
kmem_cache_free(zil_lwb_cache, lwb);
/*
* If we don't have anything left in the lwb list then
* we've had an allocation failure and we need to zero
* out the zil_header blkptr so that we don't end
* up freeing the same block twice.
*/
if (list_head(&zilog->zl_lwb_list) == NULL)
BP_ZERO(&zh->zh_log);
}
mutex_exit(&zilog->zl_lock);
}
void
zfs_log_write(zilog_t *zilog, dmu_tx_t *tx, int txtype,
znode_t *zp, offset_t off, ssize_t resid, int ioflag)
{
itx_wr_state_t write_state;
boolean_t slogging;
uintptr_t fsync_cnt;
if (zilog == NULL || zp->z_unlinked)
return;
ZFS_HANDLE_REPLAY(zilog, tx); /* exits if replay */
slogging = spa_has_slogs(zilog->zl_spa);
if (resid > zfs_immediate_write_sz && !slogging && resid <= zp->z_blksz)
write_state = WR_INDIRECT;
else if (ioflag & (FSYNC | FDSYNC))
write_state = WR_COPIED;
else
write_state = WR_NEED_COPY;
if ((fsync_cnt = (uintptr_t)tsd_get(zfs_fsyncer_key)) != 0) {
(void) tsd_set(zfs_fsyncer_key, (void *)(fsync_cnt - 1));
}
while (resid) {
itx_t *itx;
lr_write_t *lr;
ssize_t len;
/*
* If the write would overflow the largest block then split it.
*/
if (write_state != WR_INDIRECT && resid > ZIL_MAX_LOG_DATA)
len = SPA_MAXBLOCKSIZE >> 1;
else
len = resid;
itx = zil_itx_create(txtype, sizeof (*lr) +
(write_state == WR_COPIED ? len : 0));
lr = (lr_write_t *)&itx->itx_lr;
if (write_state == WR_COPIED && dmu_read(zp->z_zfsvfs->z_os,
zp->z_id, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
kmem_free(itx, offsetof(itx_t, itx_lr) +
itx->itx_lr.lrc_reclen);
itx = zil_itx_create(txtype, sizeof (*lr));
lr = (lr_write_t *)&itx->itx_lr;
write_state = WR_NEED_COPY;
}
itx->itx_wr_state = write_state;
if (write_state == WR_NEED_COPY)
itx->itx_sod += len;
lr->lr_foid = zp->z_id;
lr->lr_offset = off;
lr->lr_length = len;
lr->lr_blkoff = 0;
BP_ZERO(&lr->lr_blkptr);
itx->itx_private = zp->z_zfsvfs;
if ((zp->z_sync_cnt != 0) || (fsync_cnt != 0) ||
(ioflag & (FSYNC | FDSYNC)))
itx->itx_sync = B_TRUE;
else
itx->itx_sync = B_FALSE;
zp->z_last_itx = zil_itx_assign(zilog, itx, tx);
off += len;
resid -= len;
}
