static int osd_declare_object_destroy(const struct lu_env *env,
struct dt_object *dt,
struct thandle *th)
{
char *buf = osd_oti_get(env)->oti_str;
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);
struct osd_thandle *oh;
int rc;
uint64_t zapid;
ENTRY;
LASSERT(th != NULL);
LASSERT(dt_object_exists(dt));
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
/* declare that we'll remove object from fid-dnode mapping */
zapid = osd_get_name_n_idx(env, osd, fid, buf);
dmu_tx_hold_bonus(oh->ot_tx, zapid);
dmu_tx_hold_zap(oh->ot_tx, zapid, FALSE, buf);
osd_declare_xattrs_destroy(env, obj, oh);
/* declare that we'll remove object from inode accounting ZAPs */
dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, buf);
dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, buf);
/* one less inode */
rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
obj->oo_attr.la_gid, -1, oh, false, NULL, false);
if (rc)
RETURN(rc);
/* data to be truncated */
rc = osd_declare_quota(env, osd, obj->oo_attr.la_uid,
obj->oo_attr.la_gid, 0, oh, true, NULL, false);
if (rc)
RETURN(rc);
osd_object_set_destroy_type(obj);
if (obj->oo_destroy == OSD_DESTROY_SYNC)
dmu_tx_hold_free(oh->ot_tx, obj->oo_db->db_object,
0, DMU_OBJECT_END);
else
dmu_tx_hold_zap(oh->ot_tx, osd->od_unlinkedid, TRUE, NULL);
RETURN(0);
}
static int osd_object_destroy(const struct lu_env *env,
struct dt_object *dt,
struct thandle *th)
{
char *buf = osd_oti_get(env)->oti_str;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_thandle *oh;
int rc;
uint64_t zapid;
ENTRY;
LASSERT(obj->oo_db != NULL);
LASSERT(dt_object_exists(dt));
LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh != NULL);
LASSERT(oh->ot_tx != NULL);
zapid = osd_get_name_n_idx(env, osd, fid, buf);
/* remove obj ref from index dir (it depends) */
rc = -zap_remove(osd->od_objset.os, zapid, buf, oh->ot_tx);
if (rc) {
CERROR("%s: zap_remove() failed: rc = %d\n",
osd->od_svname, rc);
GOTO(out, rc);
}
/* Remove object from inode accounting. It is not fatal for the destroy
* operation if something goes wrong while updating accounting, but we
* still log an error message to notify the administrator */
rc = -zap_increment_int(osd->od_objset.os, osd->od_iusr_oid,
obj->oo_attr.la_uid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_uid, rc);
rc = -zap_increment_int(osd->od_objset.os, osd->od_igrp_oid,
obj->oo_attr.la_gid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_gid, rc);
/* kill object */
rc = __osd_object_destroy(env, obj, oh->ot_tx, osd_obj_tag);
if (rc) {
CERROR("%s: __osd_object_destroy() failed: rc = %d\n",
osd->od_svname, rc);
GOTO(out, rc);
}
out:
/* not needed in the cache anymore */
set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
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 zpl_direntry *zde = &osd_oti_get(env)->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 = osd_oti_get(env)->oti_str;
struct osd_thandle *oh;
dmu_buf_t *db;
uint64_t zapid;
int rc;
ENTRY;
/* concurrent create declarations should not see
* the object inconsistent (db, attr, etc).
* in regular cases acquisition should be cheap */
down(&obj->oo_guard);
LASSERT(osd_invariant(obj));
LASSERT(!dt_object_exists(dt));
LASSERT(dof != NULL);
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
/*
* XXX missing: Quote handling.
*/
LASSERT(obj->oo_db == NULL);
db = osd_create_type_f(dof->dof_type)(env, osd, attr, oh);
if (IS_ERR(db))
GOTO(out, rc = PTR_ERR(th));
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);
rc = -zap_add(osd->od_objset.os, zapid, buf, 8, 1, zde, oh->ot_tx);
if (rc)
GOTO(out, rc);
/* Add new object to inode accounting.
* Errors are not considered as fatal */
rc = -zap_increment_int(osd->od_objset.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_objset.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);
/* configure new osd object */
obj->oo_db = db;
rc = osd_object_init0(env, obj);
LASSERT(ergo(rc == 0, dt_object_exists(dt)));
LASSERT(osd_invariant(obj));
rc = osd_init_lma(env, obj, fid, oh);
if (rc) {
CERROR("%s: can not set LMA on "DFID": rc = %d\n",
osd->od_svname, PFID(fid), rc);
/* ignore errors during LMA initialization */
rc = 0;
}
out:
up(&obj->oo_guard);
RETURN(rc);
}
static int osd_declare_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 *handle)
{
char *buf = osd_oti_get(env)->oti_str;
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);
struct osd_thandle *oh;
uint64_t zapid;
int rc;
ENTRY;
LASSERT(dof);
switch (dof->dof_type) {
case DFT_REGULAR:
case DFT_SYM:
case DFT_NODE:
if (obj->oo_dt.do_body_ops == NULL)
obj->oo_dt.do_body_ops = &osd_body_ops;
break;
default:
break;
}
LASSERT(handle != NULL);
oh = container_of0(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
switch (dof->dof_type) {
case DFT_DIR:
case DFT_INDEX:
/* for zap create */
dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, 1, NULL);
break;
case DFT_REGULAR:
case DFT_SYM:
case DFT_NODE:
/* first, we'll create new object */
dmu_tx_hold_bonus(oh->ot_tx, DMU_NEW_OBJECT);
break;
default:
LBUG();
break;
}
/* and we'll add it to some mapping */
zapid = osd_get_name_n_idx(env, osd, fid, buf);
dmu_tx_hold_bonus(oh->ot_tx, zapid);
dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, buf);
/* we will also update inode accounting ZAPs */
dmu_tx_hold_bonus(oh->ot_tx, osd->od_iusr_oid);
dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, TRUE, buf);
dmu_tx_hold_bonus(oh->ot_tx, osd->od_igrp_oid);
dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, TRUE, buf);
dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
__osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
XATTR_NAME_LMA, oh);
rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
false, NULL, false);
RETURN(rc);
}
/**
* update/insert/delete the specified OI mapping (@fid @id) according to the ops
*
* \retval 1, changed nothing
* \retval 0, changed successfully
* \retval -ve, on error
*/
static int osd_scrub_refresh_mapping(const struct lu_env *env,
struct osd_device *dev,
const struct lu_fid *fid,
uint64_t oid, int ops,
bool force, const char *name)
{
struct osd_thread_info *info = osd_oti_get(env);
struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
char *buf = info->oti_str;
dmu_tx_t *tx = NULL;
dnode_t *dn = NULL;
uint64_t zapid;
int rc;
ENTRY;
if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
GOTO(log, rc = 0);
tx = dmu_tx_create(dev->od_os);
if (!tx)
GOTO(log, rc = -ENOMEM);
zapid = osd_get_name_n_idx(env, dev, fid, buf,
sizeof(info->oti_str), &dn);
osd_tx_hold_zap(tx, zapid, dn,
ops == DTO_INDEX_INSERT ? TRUE : FALSE, NULL);
rc = -dmu_tx_assign(tx, TXG_WAIT);
if (rc) {
dmu_tx_abort(tx);
GOTO(log, rc);
}
switch (ops) {
case DTO_INDEX_UPDATE:
zde->zde_pad = 0;
zde->zde_dnode = oid;
zde->zde_type = 0; /* The type in OI mapping is useless. */
rc = -zap_update(dev->od_os, zapid, buf, 8, sizeof(*zde) / 8,
zde, tx);
if (unlikely(rc == -ENOENT)) {
/* Some unlink thread may removed the OI mapping. */
rc = 1;
}
break;
case DTO_INDEX_INSERT:
zde->zde_pad = 0;
zde->zde_dnode = oid;
zde->zde_type = 0; /* The type in OI mapping is useless. */
rc = osd_zap_add(dev, zapid, dn, buf, 8, sizeof(*zde) / 8,
zde, tx);
if (unlikely(rc == -EEXIST))
rc = 1;
break;
case DTO_INDEX_DELETE:
rc = osd_zap_remove(dev, zapid, dn, buf, tx);
if (rc == -ENOENT) {
/* It is normal that the unlink thread has removed the
* OI mapping already. */
rc = 1;
}
break;
default:
LASSERTF(0, "Unexpected ops %d\n", ops);
rc = -EINVAL;
break;
}
dmu_tx_commit(tx);
GOTO(log, rc);
log:
CDEBUG(D_LFSCK, "%s: refresh OI map for scrub, op %d, force %s, "
DFID" => %llu (%s): rc = %d\n", osd_name(dev), ops,
force ? "yes" : "no", PFID(fid), oid, name ? name : "null", rc);
return rc;
}
static int osd_object_destroy(const struct lu_env *env,
struct dt_object *dt, struct thandle *th)
{
struct osd_thread_info *info = osd_oti_get(env);
char *buf = info->oti_str;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
struct osd_thandle *oh;
int rc;
uint64_t oid, zapid;
ENTRY;
down_write(&obj->oo_guard);
if (unlikely(!dt_object_exists(dt) || obj->oo_destroyed))
GOTO(out, rc = -ENOENT);
LASSERT(obj->oo_db != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(oh != NULL);
LASSERT(oh->ot_tx != NULL);
/* remove obj ref from index dir (it depends) */
zapid = osd_get_name_n_idx(env, osd, fid, buf, sizeof(info->oti_str));
rc = -zap_remove(osd->od_os, zapid, buf, oh->ot_tx);
if (rc) {
CERROR("%s: zap_remove(%s) failed: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
rc = osd_xattrs_destroy(env, obj, oh);
if (rc) {
CERROR("%s: cannot destroy xattrs for %s: rc = %d\n",
osd->od_svname, buf, rc);
GOTO(out, rc);
}
/* Remove object from inode accounting. It is not fatal for the destroy
* operation if something goes wrong while updating accounting, but we
* still log an error message to notify the administrator */
rc = -zap_increment_int(osd->od_os, osd->od_iusr_oid,
obj->oo_attr.la_uid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for usr"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_uid, rc);
rc = -zap_increment_int(osd->od_os, osd->od_igrp_oid,
obj->oo_attr.la_gid, -1, oh->ot_tx);
if (rc)
CERROR("%s: failed to remove "DFID" from accounting ZAP for grp"
" %d: rc = %d\n", osd->od_svname, PFID(fid),
obj->oo_attr.la_gid, rc);
oid = obj->oo_db->db_object;
if (unlikely(obj->oo_destroy == OSD_DESTROY_NONE)) {
/* this may happen if the destroy wasn't declared
* e.g. when the object is created and then destroyed
* in the same transaction - we don't need additional
* space for destroy specifically */
LASSERT(obj->oo_attr.la_size <= osd_sync_destroy_max_size);
rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
if (rc)
CERROR("%s: failed to free %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
} else if (obj->oo_destroy == OSD_DESTROY_SYNC) {
rc = -dmu_object_free(osd->od_os, oid, oh->ot_tx);
if (rc)
CERROR("%s: failed to free %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
} else { /* asynchronous destroy */
rc = osd_object_unlinked_add(obj, oh);
if (rc)
GOTO(out, rc);
rc = -zap_add_int(osd->od_os, osd->od_unlinkedid,
oid, oh->ot_tx);
if (rc)
CERROR("%s: zap_add_int() failed %s %llu: rc = %d\n",
osd->od_svname, buf, oid, rc);
}
out:
/* not needed in the cache anymore */
set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
if (rc == 0)
obj->oo_destroyed = 1;
up_write(&obj->oo_guard);
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);
}
static int osd_declare_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 *handle)
{
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);
struct osd_thandle *oh;
uint64_t zapid;
int rc, dnode_size;
ENTRY;
LASSERT(dof);
switch (dof->dof_type) {
case DFT_REGULAR:
case DFT_SYM:
case DFT_NODE:
if (obj->oo_dt.do_body_ops == NULL)
obj->oo_dt.do_body_ops = &osd_body_ops;
break;
default:
break;
}
LASSERT(handle != NULL);
oh = container_of0(handle, struct osd_thandle, ot_super);
LASSERT(oh->ot_tx != NULL);
/* this is the minimum set of EAs on every Lustre object */
obj->oo_ea_in_bonus = ZFS_SA_BASE_ATTR_SIZE +
sizeof(__u64) + /* VBR VERSION */
sizeof(struct lustre_mdt_attrs); /* LMA */
/* reserve 32 bytes for extra stuff like ACLs */
dnode_size = size_roundup_power2(obj->oo_ea_in_bonus + 32);
switch (dof->dof_type) {
case DFT_DIR:
dt->do_index_ops = &osd_dir_ops;
case DFT_INDEX:
/* for zap create */
dmu_tx_hold_zap(oh->ot_tx, DMU_NEW_OBJECT, FALSE, NULL);
dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
break;
case DFT_REGULAR:
case DFT_SYM:
case DFT_NODE:
/* first, we'll create new object */
dmu_tx_hold_sa_create(oh->ot_tx, dnode_size);
break;
default:
LBUG();
break;
}
/* and we'll add it to some mapping */
zapid = osd_get_name_n_idx(env, osd, fid, NULL, 0);
dmu_tx_hold_zap(oh->ot_tx, zapid, TRUE, NULL);
/* we will also update inode accounting ZAPs */
dmu_tx_hold_zap(oh->ot_tx, osd->od_iusr_oid, FALSE, NULL);
dmu_tx_hold_zap(oh->ot_tx, osd->od_igrp_oid, FALSE, NULL);
rc = osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
false, NULL, false);
RETURN(rc);
}