/**
* Helper function to retrieve DMU object id from fid for accounting object
*/
uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
{
LASSERT(fid_is_acct(fid));
if (fid_oid(fid) == ACCT_GROUP_OID)
return DMU_GROUPUSED_OBJECT;
return DMU_USERUSED_OBJECT;
}
int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, struct osd_inode_id *id,
enum oi_check_flags flags)
{
if (unlikely(fid_is_last_id(fid)))
return osd_obj_spec_lookup(info, osd, fid, id);
if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
return osd_obj_map_lookup(info, osd, fid, id);
if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
int rc;
if (fid_is_fs_root(fid)) {
osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
osd_sb(osd)->s_root->d_inode->i_generation);
return 0;
}
if (unlikely(fid_is_acct(fid)))
return osd_acct_obj_lookup(info, osd, fid, id);
/* For other special FIDs, try OI first, then do spec lookup */
rc = __osd_oi_lookup(info, osd, fid, id);
if (rc == -ENOENT)
return osd_obj_spec_lookup(info, osd, fid, id);
return rc;
}
if (!osd->od_igif_inoi && fid_is_igif(fid)) {
osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
return 0;
}
return __osd_oi_lookup(info, osd, fid, id);
}
int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, struct osd_inode_id *id,
enum oi_check_flags flags)
{
if (unlikely(fid_is_last_id(fid)))
return osd_obj_spec_lookup(info, osd, fid, id);
if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
return osd_obj_map_lookup(info, osd, fid, id);
if (fid_is_fs_root(fid)) {
osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
osd_sb(osd)->s_root->d_inode->i_generation);
return 0;
}
if (unlikely(fid_is_acct(fid)))
return osd_acct_obj_lookup(info, osd, fid, id);
if (!osd->od_igif_inoi && fid_is_igif(fid)) {
osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
return 0;
}
return __osd_oi_lookup(info, osd, fid, id);
}
/*
* Concurrency: shouldn't matter.
*/
int osd_object_init0(const struct lu_env *env, struct osd_object *obj)
{
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
int rc = 0;
ENTRY;
if (obj->oo_db == NULL)
RETURN(0);
/* object exist */
rc = osd_object_sa_init(obj, osd);
if (rc)
RETURN(rc);
/* cache attrs in object */
rc = __osd_object_attr_get(env, osd, obj, &obj->oo_attr);
if (rc)
RETURN(rc);
if (likely(!fid_is_acct(fid)))
/* no body operations for accounting objects */
obj->oo_dt.do_body_ops = &osd_body_ops;
/*
* initialize object before marking it existing
*/
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;
RETURN(0);
}
int osd_index_try(const struct lu_env *env, struct dt_object *dt,
const struct dt_index_features *feat)
{
struct osd_object *obj = osd_dt_obj(dt);
ENTRY;
LASSERT(dt_object_exists(dt));
/*
* XXX: implement support for fixed-size keys sorted with natural
* numerical way (not using internal hash value)
*/
if (feat->dif_flags & DT_IND_RANGE)
RETURN(-ERANGE);
if (unlikely(feat == &dt_otable_features))
/* do not support oi scrub yet. */
RETURN(-ENOTSUPP);
LASSERT(obj->oo_db != NULL);
if (likely(feat == &dt_directory_features)) {
if (udmu_object_is_zap(obj->oo_db))
dt->do_index_ops = &osd_dir_ops;
else
RETURN(-ENOTDIR);
} else if (unlikely(feat == &dt_acct_features)) {
LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
dt->do_index_ops = &osd_acct_index_ops;
} else if (udmu_object_is_zap(obj->oo_db) &&
dt->do_index_ops == NULL) {
/* For index file, we don't support variable key & record sizes
* and the key has to be unique */
if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
RETURN(-EINVAL);
if (feat->dif_keysize_max > ZAP_MAXNAMELEN)
RETURN(-E2BIG);
if (feat->dif_keysize_max != feat->dif_keysize_min)
RETURN(-EINVAL);
/* As for the record size, it should be a multiple of 8 bytes
* and smaller than the maximum value length supported by ZAP.
*/
if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
RETURN(-E2BIG);
if (feat->dif_recsize_max != feat->dif_recsize_min)
RETURN(-EINVAL);
obj->oo_keysize = feat->dif_keysize_max;
obj->oo_recsize = feat->dif_recsize_max;
obj->oo_recusize = 1;
/* ZFS prefers to work with array of 64bits */
if ((obj->oo_recsize & 7) == 0) {
obj->oo_recsize >>= 3;
obj->oo_recusize = 8;
}
dt->do_index_ops = &osd_index_ops;
}
int osd_index_try(const struct lu_env *env, struct dt_object *dt,
const struct dt_index_features *feat)
{
struct osd_object *obj = osd_dt_obj(dt);
ENTRY;
LASSERT(dt_object_exists(dt));
/*
* XXX: implement support for fixed-size keys sorted with natural
* numerical way (not using internal hash value)
*/
if (feat->dif_flags & DT_IND_RANGE)
RETURN(-ERANGE);
if (unlikely(feat == &dt_otable_features))
/* do not support oi scrub yet. */
RETURN(-ENOTSUPP);
LASSERT(obj->oo_db != NULL);
if (likely(feat == &dt_directory_features)) {
if (udmu_object_is_zap(obj->oo_db))
dt->do_index_ops = &osd_dir_ops;
else
RETURN(-ENOTDIR);
} else if (unlikely(feat == &dt_acct_features)) {
LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
dt->do_index_ops = &osd_acct_index_ops;
} else if (udmu_object_is_zap(obj->oo_db) &&
dt->do_index_ops == NULL) {
/* For index file, we don't support variable key & record sizes
* and the key has to be unique */
if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
RETURN(-EINVAL);
/* Although the zap_*_uint64() primitives support large keys, we
* limit ourselves to 64-bit keys for now */
if (feat->dif_keysize_max != sizeof(__u64) ||
feat->dif_keysize_min != sizeof(__u64))
RETURN(-EINVAL);
/* As for the record size, it should be a multiple of 8 bytes
* and smaller than the maximum value length supported by ZAP.
*/
if (feat->dif_recsize_max > ZAP_MAXVALUELEN)
RETURN(-E2BIG);
if (feat->dif_recsize_max != feat->dif_recsize_min ||
(feat->dif_recsize_max & (sizeof(__u64) - 1)))
RETURN(-EINVAL);
obj->oo_recsize = feat->dif_recsize_max / sizeof(__u64);
dt->do_index_ops = &osd_index_ops;
}
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);
}
