/*
* The transaction passed to this routine must have
* dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
* to a transaction group.
*
* Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
* This is fine for directories today, because storing the FID in the dirent
* will also require a FAT ZAP. If there is a new type of micro ZAP created
* then we might need to re-evaluate the use of this flag and instead do
* a conversion from the different internal ZAP hash formats being used. */
int __osd_zap_create(const struct lu_env *env, udmu_objset_t *uos,
dmu_buf_t **zap_dbp, dmu_tx_t *tx,
struct lu_attr *la, void *tag, zap_flags_t flags)
{
uint64_t oid;
int rc;
LASSERT(tag);
spin_lock(&uos->lock);
uos->objects++;
spin_unlock(&uos->lock);
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
oid = zap_create_flags(uos->os, 0, flags | ZAP_FLAG_HASH64,
DMU_OT_DIRECTORY_CONTENTS, 12, 12,
DMU_OT_SA, DN_MAX_BONUSLEN, tx);
rc = -sa_buf_hold(uos->os, oid, tag, zap_dbp);
if (rc)
return rc;
LASSERT(la->la_valid & LA_MODE);
la->la_size = 2;
la->la_nlink = 1;
return __osd_attr_init(env, uos, oid, tx, la);
}
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
* to a transaction group.
*/
int __osd_object_create(const struct lu_env *env, udmu_objset_t *uos,
dmu_buf_t **dbp, dmu_tx_t *tx,
struct lu_attr *la, void *tag)
{
uint64_t oid;
int rc;
LASSERT(tag);
spin_lock(&uos->lock);
uos->objects++;
spin_unlock(&uos->lock);
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
/* Create a new DMU object. */
oid = dmu_object_alloc(uos->os, DMU_OT_PLAIN_FILE_CONTENTS, 0,
DMU_OT_SA, DN_MAX_BONUSLEN, tx);
rc = -sa_buf_hold(uos->os, oid, tag, dbp);
if (rc)
return rc;
LASSERT(la->la_valid & LA_MODE);
la->la_size = 0;
la->la_nlink = 1;
return __osd_attr_init(env, uos, oid, tx, la);
}
/*
* The transaction passed to this routine must have
* dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, ...) called and then assigned
* to a transaction group.
*
* Using ZAP_FLAG_HASH64 will force the ZAP to always be a FAT ZAP.
* This is fine for directories today, because storing the FID in the dirent
* will also require a FAT ZAP. If there is a new type of micro ZAP created
* then we might need to re-evaluate the use of this flag and instead do
* a conversion from the different internal ZAP hash formats being used. */
int __osd_zap_create(const struct lu_env *env, struct osd_device *osd,
dmu_buf_t **zap_dbp, dmu_tx_t *tx,
struct lu_attr *la, uint64_t parent, zap_flags_t flags)
{
uint64_t oid;
int rc;
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
oid = zap_create_flags(osd->od_os, 0, flags | ZAP_FLAG_HASH64,
DMU_OT_DIRECTORY_CONTENTS,
14, /* == ZFS fzap_default_block_shift */
DN_MAX_INDBLKSHIFT, /* indirect block shift */
DMU_OT_SA, DN_MAX_BONUSLEN, tx);
rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, zap_dbp);
if (rc)
return rc;
LASSERT(la->la_valid & LA_MODE);
la->la_size = 2;
la->la_nlink = 1;
return __osd_attr_init(env, osd, oid, tx, la, parent);
}
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
* to a transaction group.
*/
int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
uint64_t parent)
{
uint64_t oid;
int rc;
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
/* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
* would get an additional ditto copy */
if (unlikely(S_ISREG(la->la_mode) &&
fid_seq_is_local_file(fid_seq(fid))))
type = DMU_OTN_UINT8_METADATA;
/* Create a new DMU object. */
oid = dmu_object_alloc(osd->od_os, type, 0,
DMU_OT_SA, DN_MAX_BONUSLEN, tx);
rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
LASSERT(la->la_valid & LA_MODE);
la->la_size = 0;
la->la_nlink = 1;
rc = __osd_attr_init(env, osd, oid, tx, la, parent);
if (rc != 0) {
sa_buf_rele(*dbp, osd_obj_tag);
*dbp = NULL;
dmu_object_free(osd->od_os, oid, tx);
return rc;
}
return 0;
}
/*
* Concurrency: @dt is write locked.
*/
static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
struct lu_attr *attr,
struct dt_allocation_hint *hint,
struct dt_object_format *dof,
struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
char *buf = info->oti_str;
struct osd_thandle *oh;
dmu_buf_t *db = NULL;
uint64_t zapid, parent = 0;
int rc;
ENTRY;
/* concurrent create declarations should not see
* the object inconsistent (db, attr, etc).
* in regular cases acquisition should be cheap */
down_write(&obj->oo_guard);
if (unlikely(dt_object_exists(dt)))
GOTO(out, rc = -EEXIST);
LASSERT(osd_invariant(obj));
LASSERT(dof != NULL);
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
/*
* XXX missing: Quote handling.
*/
LASSERT(obj->oo_db == NULL);
/* to follow ZFS on-disk format we need
* to initialize parent dnode properly */
if (hint != NULL && hint->dah_parent != NULL &&
!dt_object_remote(hint->dah_parent))
parent = osd_dt_obj(hint->dah_parent)->oo_db->db_object;
/* we may fix some attributes, better do not change the source */
obj->oo_attr = *attr;
obj->oo_attr.la_valid |= LA_SIZE | LA_NLINK | LA_TYPE;
db = osd_create_type_f(dof->dof_type)(env, obj, &obj->oo_attr, oh);
if (IS_ERR(db)) {
rc = PTR_ERR(db);
db = NULL;
GOTO(out, rc);
}
zde->zde_pad = 0;
zde->zde_dnode = db->db_object;
zde->zde_type = IFTODT(attr->la_mode & S_IFMT);
zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
rc = -zap_add(osd->od_os, zapid, buf, 8, 1, zde, oh->ot_tx);
if (rc)
GOTO(out, rc);
/* Now add in all of the "SA" attributes */
rc = -sa_handle_get(osd->od_os, db->db_object, NULL,
SA_HDL_PRIVATE, &obj->oo_sa_hdl);
if (rc)
GOTO(out, rc);
/* configure new osd object */
obj->oo_db = db;
parent = parent != 0 ? parent : zapid;
rc = __osd_attr_init(env, osd, obj->oo_sa_hdl, oh->ot_tx,
&obj->oo_attr, parent);
if (rc)
GOTO(out, rc);
/* XXX: oo_lma_flags */
obj->oo_dt.do_lu.lo_header->loh_attr |= obj->oo_attr.la_mode & S_IFMT;
smp_mb();
obj->oo_dt.do_lu.lo_header->loh_attr |= LOHA_EXISTS;
if (likely(!fid_is_acct(lu_object_fid(&obj->oo_dt.do_lu))))
/* no body operations for accounting objects */
obj->oo_dt.do_body_ops = &osd_body_ops;
rc = -nvlist_alloc(&obj->oo_sa_xattr, NV_UNIQUE_NAME, KM_SLEEP);
if (rc)
GOTO(out, rc);
/* initialize LMA */
lustre_lma_init(lma, lu_object_fid(&obj->oo_dt.do_lu), 0, 0);
lustre_lma_swab(lma);
rc = -nvlist_add_byte_array(obj->oo_sa_xattr, XATTR_NAME_LMA,
(uchar_t *)lma, sizeof(*lma));
if (rc)
GOTO(out, rc);
rc = __osd_sa_xattr_update(env, obj, oh);
if (rc)
GOTO(out, rc);
/* Add new object to inode accounting.
* Errors are not considered as fatal */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
(attr->la_valid & LA_UID) ? attr->la_uid : 0, 1,
oh->ot_tx);
if (rc)
CERROR("%s: failed to add "DFID" to accounting ZAP for usr %d "
"(%d)\n", osd->od_svname, PFID(fid), attr->la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
(attr->la_valid & LA_GID) ? attr->la_gid : 0, 1,
oh->ot_tx);
if (rc)
CERROR("%s: failed to add "DFID" to accounting ZAP for grp %d "
"(%d)\n", osd->od_svname, PFID(fid), attr->la_gid, rc);
out:
if (unlikely(rc && db)) {
dmu_object_free(osd->od_os, db->db_object, oh->ot_tx);
sa_buf_rele(db, osd_obj_tag);
obj->oo_db = NULL;
}
up_write(&obj->oo_guard);
RETURN(rc);
}
