/* ARGSUSED */
int
dmu_objset_prefetch(char *name, void *arg)
{
dsl_dataset_t *ds;
if (dsl_dataset_hold(name, FTAG, &ds))
return (0);
if (!BP_IS_HOLE(&ds->ds_phys->ds_bp)) {
mutex_enter(&ds->ds_opening_lock);
if (!dsl_dataset_get_user_ptr(ds)) {
uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;
zbookmark_t zb;
zb.zb_objset = ds->ds_object;
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = 0;
(void) arc_read_nolock(NULL, dsl_dataset_get_spa(ds),
&ds->ds_phys->ds_bp, NULL, NULL,
ZIO_PRIORITY_ASYNC_READ,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
&aflags, &zb);
}
mutex_exit(&ds->ds_opening_lock);
}
dsl_dataset_rele(ds, FTAG);
return (0);
}
/*
* Read a log block, make sure it's valid, and byteswap it if necessary.
*/
static int
zil_read_log_block(zilog_t *zilog, const blkptr_t *bp, arc_buf_t **abufpp)
{
blkptr_t blk = *bp;
zbookmark_t zb;
uint32_t aflags = ARC_WAIT;
int error;
zb.zb_objset = bp->blk_cksum.zc_word[ZIL_ZC_OBJSET];
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = bp->blk_cksum.zc_word[ZIL_ZC_SEQ];
*abufpp = NULL;
/*
* We shouldn't be doing any scrubbing while we're doing log
* replay, it's OK to not lock.
*/
error = arc_read_nolock(NULL, zilog->zl_spa, &blk,
arc_getbuf_func, abufpp, ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL |
ZIO_FLAG_SPECULATIVE | ZIO_FLAG_SCRUB, &aflags, &zb);
if (error == 0) {
char *data = (*abufpp)->b_data;
uint64_t blksz = BP_GET_LSIZE(bp);
zil_trailer_t *ztp = (zil_trailer_t *)(data + blksz) - 1;
zio_cksum_t cksum = bp->blk_cksum;
/*
* Validate the checksummed log block.
*
* Sequence numbers should be... sequential. The checksum
* verifier for the next block should be bp's checksum plus 1.
*
* Also check the log chain linkage and size used.
*/
cksum.zc_word[ZIL_ZC_SEQ]++;
if (bcmp(&cksum, &ztp->zit_next_blk.blk_cksum,
sizeof (cksum)) || BP_IS_HOLE(&ztp->zit_next_blk) ||
(ztp->zit_nused > (blksz - sizeof (zil_trailer_t)))) {
error = ECKSUM;
}
if (error) {
VERIFY(arc_buf_remove_ref(*abufpp, abufpp) == 1);
*abufpp = NULL;
}
}
dprintf("error %d on %llu:%llu\n", error, zb.zb_objset, zb.zb_blkid);
return (error);
}
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);
}
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);
}
