static void
copies_changed_cb(void *arg, uint64_t newval)
{
objset_t *os = arg;
/*
* Inheritance and range checking should have been done by now.
*/
ASSERT(newval > 0);
ASSERT(newval <= spa_max_replication(os->os_spa));
os->os_copies = newval;
}
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
objset_t **osp)
{
objset_t *os;
int i, err;
ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
os->os_dsl_dataset = ds;
os->os_spa = spa;
os->os_rootbp = bp;
if (!BP_IS_HOLE(os->os_rootbp)) {
arc_flags_t aflags = ARC_FLAG_WAIT;
zbookmark_phys_t zb;
SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
if (DMU_OS_IS_L2CACHEABLE(os))
aflags |= ARC_FLAG_L2CACHE;
if (DMU_OS_IS_L2COMPRESSIBLE(os))
aflags |= ARC_FLAG_L2COMPRESS;
dprintf_bp(os->os_rootbp, "reading %s", "");
err = arc_read(NULL, spa, os->os_rootbp,
arc_getbuf_func, &os->os_phys_buf,
ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
if (err != 0) {
kmem_free(os, sizeof (objset_t));
/* convert checksum errors into IO errors */
if (err == ECKSUM)
err = SET_ERROR(EIO);
return (err);
}
/* Increase the blocksize if we are permitted. */
if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
arc_buf_t *buf = arc_buf_alloc(spa,
sizeof (objset_phys_t), &os->os_phys_buf,
ARC_BUFC_METADATA);
bzero(buf->b_data, sizeof (objset_phys_t));
bcopy(os->os_phys_buf->b_data, buf->b_data,
arc_buf_size(os->os_phys_buf));
(void) arc_buf_remove_ref(os->os_phys_buf,
&os->os_phys_buf);
os->os_phys_buf = buf;
}
os->os_phys = os->os_phys_buf->b_data;
os->os_flags = os->os_phys->os_flags;
} else {
int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
os->os_phys_buf = arc_buf_alloc(spa, size,
&os->os_phys_buf, ARC_BUFC_METADATA);
os->os_phys = os->os_phys_buf->b_data;
bzero(os->os_phys, size);
}
/*
* Note: the changed_cb will be called once before the register
* func returns, thus changing the checksum/compression from the
* default (fletcher2/off). Snapshots don't need to know about
* checksum/compression/copies.
*/
if (ds != NULL) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
primary_cache_changed_cb, os);
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
secondary_cache_changed_cb, os);
}
if (!ds->ds_is_snapshot) {
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_CHECKSUM),
checksum_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_COMPRESSION),
compression_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_COPIES),
copies_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_DEDUP),
dedup_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_LOGBIAS),
logbias_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SYNC),
sync_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(
ZFS_PROP_REDUNDANT_METADATA),
redundant_metadata_changed_cb, os);
}
if (err == 0) {
err = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
recordsize_changed_cb, os);
}
}
if (err != 0) {
VERIFY(arc_buf_remove_ref(os->os_phys_buf,
&os->os_phys_buf));
kmem_free(os, sizeof (objset_t));
return (err);
}
} else {
/* It's the meta-objset. */
os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
os->os_compress = ZIO_COMPRESS_ON;
os->os_copies = spa_max_replication(spa);
os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
os->os_dedup_verify = B_FALSE;
os->os_logbias = ZFS_LOGBIAS_LATENCY;
os->os_sync = ZFS_SYNC_STANDARD;
os->os_primary_cache = ZFS_CACHE_ALL;
os->os_secondary_cache = ZFS_CACHE_ALL;
}
if (ds == NULL || !ds->ds_is_snapshot)
os->os_zil_header = os->os_phys->os_zil_header;
os->os_zil = zil_alloc(os, &os->os_zil_header);
for (i = 0; i < TXG_SIZE; i++) {
list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
}
list_create(&os->os_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_link));
list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_link));
mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
dnode_special_open(os, &os->os_phys->os_meta_dnode,
DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
dnode_special_open(os, &os->os_phys->os_userused_dnode,
DMU_USERUSED_OBJECT, &os->os_userused_dnode);
dnode_special_open(os, &os->os_phys->os_groupused_dnode,
DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
}
*osp = os;
return (0);
}
/* called from dsl */
void
dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_t zb;
zio_prop_t zp;
zio_t *zio;
list_t *list;
list_t *newlist = NULL;
dbuf_dirty_record_t *dr;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
SET_BOOKMARK(&zb, os->os_dsl_dataset ?
os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
arc_release(os->os_phys_buf, &os->os_phys_buf);
dmu_write_policy(os, NULL, 0, 0, &zp);
zio = arc_write(pio, os->os_spa, tx->tx_txg,
os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
dmu_objset_write_ready, dmu_objset_write_done, os,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
/*
* Sync special dnodes - the parent IO for the sync is the root block
*/
DMU_META_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_META_DNODE(os), tx);
os->os_phys->os_flags = os->os_flags;
if (DMU_USERUSED_DNODE(os) &&
DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
DMU_USERUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_USERUSED_DNODE(os), tx);
DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
}
txgoff = tx->tx_txg & TXG_MASK;
if (dmu_objset_userused_enabled(os)) {
newlist = &os->os_synced_dnodes;
/*
* We must create the list here because it uses the
* dn_dirty_link[] of this txg.
*/
list_create(newlist, sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[txgoff]));
}
dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
while (dr = list_head(list)) {
ASSERT0(dr->dr_dbuf->db_level);
list_remove(list, dr);
if (dr->dr_zio)
zio_nowait(dr->dr_zio);
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
objset_t **osp)
{
objset_t *os;
int i, err = 0;
ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
os = kmem_zalloc(sizeof (objset_t), KM_PUSHPAGE);
os->os_dsl_dataset = ds;
os->os_spa = spa;
os->os_rootbp = bp;
if (!BP_IS_HOLE(os->os_rootbp)) {
uint32_t aflags = ARC_WAIT;
zbookmark_t zb;
SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
if (DMU_OS_IS_L2CACHEABLE(os))
aflags |= ARC_L2CACHE;
dprintf_bp(os->os_rootbp, "reading %s", "");
/*
* XXX when bprewrite scrub can change the bp,
* and this is called from dmu_objset_open_ds_os, the bp
* could change, and we'll need a lock.
*/
err = dsl_read_nolock(NULL, spa, os->os_rootbp,
arc_getbuf_func, &os->os_phys_buf,
ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
if (err) {
kmem_free(os, sizeof (objset_t));
/* convert checksum errors into IO errors */
if (err == ECKSUM)
err = EIO;
return (err);
}
/* Increase the blocksize if we are permitted. */
if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
arc_buf_t *buf = arc_buf_alloc(spa,
sizeof (objset_phys_t), &os->os_phys_buf,
ARC_BUFC_METADATA);
bzero(buf->b_data, sizeof (objset_phys_t));
bcopy(os->os_phys_buf->b_data, buf->b_data,
arc_buf_size(os->os_phys_buf));
(void) arc_buf_remove_ref(os->os_phys_buf,
&os->os_phys_buf);
os->os_phys_buf = buf;
}
os->os_phys = os->os_phys_buf->b_data;
os->os_flags = os->os_phys->os_flags;
} else {
int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
os->os_phys_buf = arc_buf_alloc(spa, size,
&os->os_phys_buf, ARC_BUFC_METADATA);
os->os_phys = os->os_phys_buf->b_data;
bzero(os->os_phys, size);
}
/*
* Note: the changed_cb will be called once before the register
* func returns, thus changing the checksum/compression from the
* default (fletcher2/off). Snapshots don't need to know about
* checksum/compression/copies. But they do need to know about
* encryption so that clones from the snaphost inherit the
* same encryption property regardless of where in the namespace
* they get created.
*/
if (ds) {
err = dsl_prop_register(ds, "primarycache",
primary_cache_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "secondarycache",
secondary_cache_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "encryption",
crypt_changed_cb, os);
if (!dsl_dataset_is_snapshot(ds)) {
if (err == 0)
err = dsl_prop_register(ds, "checksum",
checksum_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "compression",
compression_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "copies",
copies_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "dedup",
dedup_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "logbias",
logbias_changed_cb, os);
if (err == 0)
err = dsl_prop_register(ds, "sync",
sync_changed_cb, os);
}
if (err) {
VERIFY(arc_buf_remove_ref(os->os_phys_buf,
&os->os_phys_buf) == 1);
kmem_free(os, sizeof (objset_t));
return (err);
}
} else if (ds == NULL) {
/*
* It's the meta-objset.
* Encryption is off for ZFS metadata but on for ZPL metadata
* and file/zvol contents.
*/
os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
os->os_compress = ZIO_COMPRESS_LZJB;
os->os_copies = spa_max_replication(spa);
os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
os->os_dedup_verify = 0;
os->os_logbias = 0;
os->os_sync = 0;
os->os_primary_cache = ZFS_CACHE_ALL;
os->os_secondary_cache = ZFS_CACHE_ALL;
os->os_crypt = ZIO_CRYPT_OFF;
}
if (ds == NULL || !dsl_dataset_is_snapshot(ds))
os->os_zil_header = os->os_phys->os_zil_header;
os->os_zil = zil_alloc(os, &os->os_zil_header);
for (i = 0; i < TXG_SIZE; i++) {
list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
}
list_create(&os->os_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_link));
list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_link));
mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
DMU_META_DNODE(os) = dnode_special_open(os,
&os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
&os->os_meta_dnode);
if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
DMU_USERUSED_DNODE(os) = dnode_special_open(os,
&os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
&os->os_userused_dnode);
DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
&os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
&os->os_groupused_dnode);
}
/*
* We should be the only thread trying to do this because we
* have ds_opening_lock
*/
if (ds) {
mutex_enter(&ds->ds_lock);
ASSERT(ds->ds_objset == NULL);
ds->ds_objset = os;
mutex_exit(&ds->ds_lock);
}
*osp = os;
return (0);
}
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
objset_impl_t **osip)
{
objset_impl_t *winner, *osi;
int i, err, checksum;
osi = kmem_zalloc(sizeof (objset_impl_t), KM_SLEEP);
osi->os.os = osi;
osi->os_dsl_dataset = ds;
osi->os_spa = spa;
osi->os_rootbp = bp;
if (!BP_IS_HOLE(osi->os_rootbp)) {
uint32_t aflags = ARC_WAIT;
zbookmark_t zb;
zb.zb_objset = ds ? ds->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = 0;
dprintf_bp(osi->os_rootbp, "reading %s", "");
err = arc_read(NULL, spa, osi->os_rootbp,
dmu_ot[DMU_OT_OBJSET].ot_byteswap,
arc_getbuf_func, &osi->os_phys_buf,
ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
if (err) {
kmem_free(osi, sizeof (objset_impl_t));
return (err);
}
osi->os_phys = osi->os_phys_buf->b_data;
arc_release(osi->os_phys_buf, &osi->os_phys_buf);
} else {
osi->os_phys_buf = arc_buf_alloc(spa, sizeof (objset_phys_t),
&osi->os_phys_buf, ARC_BUFC_METADATA);
osi->os_phys = osi->os_phys_buf->b_data;
bzero(osi->os_phys, sizeof (objset_phys_t));
}
/*
* Note: the changed_cb will be called once before the register
* func returns, thus changing the checksum/compression from the
* default (fletcher2/off). Snapshots don't need to know, and
* registering would complicate clone promotion.
*/
if (ds && ds->ds_phys->ds_num_children == 0) {
err = dsl_prop_register(ds, "checksum",
checksum_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "compression",
compression_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "copies",
copies_changed_cb, osi);
if (err) {
VERIFY(arc_buf_remove_ref(osi->os_phys_buf,
&osi->os_phys_buf) == 1);
kmem_free(osi, sizeof (objset_impl_t));
return (err);
}
} else if (ds == NULL) {
/* It's the meta-objset. */
osi->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
osi->os_compress = ZIO_COMPRESS_LZJB;
osi->os_copies = spa_max_replication(spa);
}
osi->os_zil = zil_alloc(&osi->os, &osi->os_phys->os_zil_header);
/*
* Metadata always gets compressed and checksummed.
* If the data checksum is multi-bit correctable, and it's not
* a ZBT-style checksum, then it's suitable for metadata as well.
* Otherwise, the metadata checksum defaults to fletcher4.
*/
checksum = osi->os_checksum;
if (zio_checksum_table[checksum].ci_correctable &&
!zio_checksum_table[checksum].ci_zbt)
osi->os_md_checksum = checksum;
else
osi->os_md_checksum = ZIO_CHECKSUM_FLETCHER_4;
osi->os_md_compress = ZIO_COMPRESS_LZJB;
for (i = 0; i < TXG_SIZE; i++) {
list_create(&osi->os_dirty_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
list_create(&osi->os_free_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
}
list_create(&osi->os_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_link));
list_create(&osi->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_link));
mutex_init(&osi->os_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&osi->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
osi->os_meta_dnode = dnode_special_open(osi,
&osi->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT);
if (ds != NULL) {
winner = dsl_dataset_set_user_ptr(ds, osi, dmu_objset_evict);
if (winner) {
dmu_objset_evict(ds, osi);
osi = winner;
}
}
*osip = osi;
return (0);
}
/* called from dsl */
void
dmu_objset_sync(objset_impl_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_t zb;
writeprops_t wp = { 0 };
zio_t *zio;
list_t *list;
dbuf_dirty_record_t *dr;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
zb.zb_objset = os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = 0;
if (BP_IS_OLDER(os->os_rootbp, tx->tx_txg)) {
(void) dsl_dataset_block_kill(os->os_dsl_dataset,
os->os_rootbp, pio, tx);
}
wp.wp_type = DMU_OT_OBJSET;
wp.wp_copies = os->os_copies;
wp.wp_level = (uint8_t)-1;
wp.wp_oschecksum = os->os_checksum;
wp.wp_oscompress = os->os_compress;
arc_release(os->os_phys_buf, &os->os_phys_buf);
zio = arc_write(pio, os->os_spa, &wp,
tx->tx_txg, os->os_rootbp, os->os_phys_buf, ready, NULL, os,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_METADATA,
&zb);
/*
* Sync meta-dnode - the parent IO for the sync is the root block
*/
os->os_meta_dnode->dn_zio = zio;
dnode_sync(os->os_meta_dnode, tx);
txgoff = tx->tx_txg & TXG_MASK;
dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], tx);
dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], tx);
list = &os->os_meta_dnode->dn_dirty_records[txgoff];
while (dr = list_head(list)) {
ASSERT(dr->dr_dbuf->db_level == 0);
list_remove(list, dr);
if (dr->dr_zio)
zio_nowait(dr->dr_zio);
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
objset_impl_t **osip)
{
objset_impl_t *osi;
int i, err;
ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
osi = kmem_zalloc(sizeof (objset_impl_t), KM_SLEEP);
osi->os.os = osi;
osi->os_dsl_dataset = ds;
osi->os_spa = spa;
osi->os_rootbp = bp;
if (!BP_IS_HOLE(osi->os_rootbp)) {
uint32_t aflags = ARC_WAIT;
zbookmark_t zb;
zb.zb_objset = ds ? ds->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = 0;
dprintf_bp(osi->os_rootbp, "reading %s", "");
/*
* NB: when bprewrite scrub can change the bp,
* and this is called from dmu_objset_open_ds_os, the bp
* could change, and we'll need a lock.
*/
err = arc_read_nolock(NULL, spa, osi->os_rootbp,
arc_getbuf_func, &osi->os_phys_buf,
ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
if (err) {
kmem_free(osi, sizeof (objset_impl_t));
return (err);
}
osi->os_phys = osi->os_phys_buf->b_data;
} else {
#ifdef __APPLE_KERNEL__
osi->os_phys_buf = arc_buf_alloc(spa, sizeof (objset_phys_t),
&osi->os_phys_buf, ARC_BUFC_METADATA, TRUE/*alloc_buf*/);
#else
osi->os_phys_buf = arc_buf_alloc(spa, sizeof (objset_phys_t),
&osi->os_phys_buf, ARC_BUFC_METADATA);
#endif
osi->os_phys = osi->os_phys_buf->b_data;
bzero(osi->os_phys, sizeof (objset_phys_t));
}
/*
* Note: the changed_cb will be called once before the register
* func returns, thus changing the checksum/compression from the
* default (fletcher2/off). Snapshots don't need to know, and
* registering would complicate clone promotion.
*/
if (ds && ds->ds_phys->ds_num_children == 0) {
err = dsl_prop_register(ds, "checksum",
checksum_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "compression",
compression_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "copies",
copies_changed_cb, osi);
if (err) {
VERIFY(arc_buf_remove_ref(osi->os_phys_buf,
&osi->os_phys_buf) == 1);
kmem_free(osi, sizeof (objset_impl_t));
return (err);
}
} else if (ds == NULL) {
/* It's the meta-objset. */
osi->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
osi->os_compress = ZIO_COMPRESS_LZJB;
osi->os_copies = spa_max_replication(spa);
}
osi->os_zil_header = osi->os_phys->os_zil_header;
osi->os_zil = zil_alloc(&osi->os, &osi->os_zil_header);
for (i = 0; i < TXG_SIZE; i++) {
list_create(&osi->os_dirty_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
list_create(&osi->os_free_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
}
list_create(&osi->os_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_link));
list_create(&osi->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_link));
mutex_init(&osi->os_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&osi->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&osi->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
osi->os_meta_dnode = dnode_special_open(osi,
&osi->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT);
/*
* We should be the only thread trying to do this because we
* have ds_opening_lock
*/
if (ds) {
VERIFY(NULL == dsl_dataset_set_user_ptr(ds, osi,
dmu_objset_evict));
}
*osip = osi;
return (0);
}
/* called from dsl */
void
dmu_objset_sync(objset_impl_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_t zb;
writeprops_t wp = { 0 };
zio_t *zio;
list_t *list;
list_t *newlist = NULL;
dbuf_dirty_record_t *dr;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
zb.zb_objset = os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1; /* for block ordering; it's level 0 on disk */
zb.zb_blkid = 0;
wp.wp_type = DMU_OT_OBJSET;
wp.wp_level = 0; /* on-disk BP level; see above */
wp.wp_copies = os->os_copies;
wp.wp_oschecksum = os->os_checksum;
wp.wp_oscompress = os->os_compress;
if (BP_IS_OLDER(os->os_rootbp, tx->tx_txg)) {
(void) dsl_dataset_block_kill(os->os_dsl_dataset,
os->os_rootbp, pio, tx);
}
arc_release(os->os_phys_buf, &os->os_phys_buf);
zio = arc_write(pio, os->os_spa, &wp, DMU_OS_IS_L2CACHEABLE(os),
tx->tx_txg, os->os_rootbp, os->os_phys_buf, ready, NULL, os,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
/*
* Sync special dnodes - the parent IO for the sync is the root block
*/
os->os_meta_dnode->dn_zio = zio;
dnode_sync(os->os_meta_dnode, tx);
os->os_phys->os_flags = os->os_flags;
if (os->os_userused_dnode &&
os->os_userused_dnode->dn_type != DMU_OT_NONE) {
os->os_userused_dnode->dn_zio = zio;
dnode_sync(os->os_userused_dnode, tx);
os->os_groupused_dnode->dn_zio = zio;
dnode_sync(os->os_groupused_dnode, tx);
}
txgoff = tx->tx_txg & TXG_MASK;
if (dmu_objset_userused_enabled(os)) {
newlist = &os->os_synced_dnodes;
/*
* We must create the list here because it uses the
* dn_dirty_link[] of this txg.
*/
list_create(newlist, sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[txgoff]));
}
dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
list = &os->os_meta_dnode->dn_dirty_records[txgoff];
while ((dr = list_head(list))) {
ASSERT(dr->dr_dbuf->db_level == 0);
list_remove(list, dr);
if (dr->dr_zio)
zio_nowait(dr->dr_zio);
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
int
dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
objset_impl_t **osip)
{
objset_impl_t *osi;
int i, err;
ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
osi = kmem_zalloc(sizeof (objset_impl_t), KM_SLEEP);
osi->os.os = osi;
osi->os_dsl_dataset = ds;
osi->os_spa = spa;
osi->os_rootbp = bp;
if (!BP_IS_HOLE(osi->os_rootbp)) {
uint32_t aflags = ARC_WAIT;
zbookmark_t zb;
zb.zb_objset = ds ? ds->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = 0;
if (DMU_OS_IS_L2CACHEABLE(osi))
aflags |= ARC_L2CACHE;
dprintf_bp(osi->os_rootbp, "reading %s", "");
/*
* NB: when bprewrite scrub can change the bp,
* and this is called from dmu_objset_open_ds_os, the bp
* could change, and we'll need a lock.
*/
err = arc_read_nolock(NULL, spa, osi->os_rootbp,
arc_getbuf_func, &osi->os_phys_buf,
ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
if (err) {
kmem_free(osi, sizeof (objset_impl_t));
/* convert checksum errors into IO errors */
if (err == ECKSUM)
err = EIO;
return (err);
}
/* Increase the blocksize if we are permitted. */
if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
arc_buf_size(osi->os_phys_buf) < sizeof (objset_phys_t)) {
arc_buf_t *buf = arc_buf_alloc(spa,
sizeof (objset_phys_t), &osi->os_phys_buf,
ARC_BUFC_METADATA);
bzero(buf->b_data, sizeof (objset_phys_t));
bcopy(osi->os_phys_buf->b_data, buf->b_data,
arc_buf_size(osi->os_phys_buf));
(void) arc_buf_remove_ref(osi->os_phys_buf,
&osi->os_phys_buf);
osi->os_phys_buf = buf;
}
osi->os_phys = osi->os_phys_buf->b_data;
osi->os_flags = osi->os_phys->os_flags;
} else {
int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
osi->os_phys_buf = arc_buf_alloc(spa, size,
&osi->os_phys_buf, ARC_BUFC_METADATA);
osi->os_phys = osi->os_phys_buf->b_data;
bzero(osi->os_phys, size);
}
/*
* Note: the changed_cb will be called once before the register
* func returns, thus changing the checksum/compression from the
* default (fletcher2/off). Snapshots don't need to know about
* checksum/compression/copies.
*/
if (ds) {
err = dsl_prop_register(ds, "primarycache",
primary_cache_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "secondarycache",
secondary_cache_changed_cb, osi);
if (!dsl_dataset_is_snapshot(ds)) {
if (err == 0)
err = dsl_prop_register(ds, "checksum",
checksum_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "compression",
compression_changed_cb, osi);
if (err == 0)
err = dsl_prop_register(ds, "copies",
copies_changed_cb, osi);
}
if (err) {
VERIFY(arc_buf_remove_ref(osi->os_phys_buf,
&osi->os_phys_buf) == 1);
kmem_free(osi, sizeof (objset_impl_t));
return (err);
}
} else if (ds == NULL) {
/* It's the meta-objset. */
osi->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
osi->os_compress = ZIO_COMPRESS_LZJB;
osi->os_copies = spa_max_replication(spa);
osi->os_primary_cache = ZFS_CACHE_ALL;
osi->os_secondary_cache = ZFS_CACHE_ALL;
}
osi->os_zil_header = osi->os_phys->os_zil_header;
osi->os_zil = zil_alloc(&osi->os, &osi->os_zil_header);
for (i = 0; i < TXG_SIZE; i++) {
list_create(&osi->os_dirty_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
list_create(&osi->os_free_dnodes[i], sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[i]));
}
list_create(&osi->os_dnodes, sizeof (dnode_t),
offsetof(dnode_t, dn_link));
list_create(&osi->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
offsetof(dmu_buf_impl_t, db_link));
mutex_init(&osi->os_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&osi->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&osi->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
osi->os_meta_dnode = dnode_special_open(osi,
&osi->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT);
if (arc_buf_size(osi->os_phys_buf) >= sizeof (objset_phys_t)) {
osi->os_userused_dnode = dnode_special_open(osi,
&osi->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT);
osi->os_groupused_dnode = dnode_special_open(osi,
&osi->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT);
}
/*
* We should be the only thread trying to do this because we
* have ds_opening_lock
*/
if (ds) {
VERIFY(NULL == dsl_dataset_set_user_ptr(ds, osi,
dmu_objset_evict));
}
*osip = osi;
return (0);
}