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);
}
static struct dt_it *osd_index_it_init(const struct lu_env *env,
struct dt_object *dt,
__u32 unused,
struct lustre_capa *capa)
{
struct osd_thread_info *info = osd_oti_get(env);
struct osd_zap_it *it;
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct lu_object *lo = &dt->do_lu;
ENTRY;
/* XXX: check capa ? */
LASSERT(lu_object_exists(lo));
LASSERT(obj->oo_db);
LASSERT(udmu_object_is_zap(obj->oo_db));
LASSERT(info);
it = &info->oti_it_zap;
if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
obj->oo_db->db_object, 0))
RETURN(ERR_PTR(-ENOMEM));
it->ozi_obj = obj;
it->ozi_capa = capa;
it->ozi_reset = 1;
lu_object_get(lo);
RETURN((struct dt_it *)it);
}
static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
struct dt_rec *rec, const struct dt_key *key,
struct lustre_capa *capa)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
char *name = (char *)key;
int rc;
ENTRY;
LASSERT(udmu_object_is_zap(obj->oo_db));
if (name[0] == '.') {
if (name[1] == 0) {
const struct lu_fid *f = lu_object_fid(&dt->do_lu);
memcpy(rec, f, sizeof(*f));
RETURN(1);
} else if (name[1] == '.' && name[2] == 0) {
rc = osd_find_parent_fid(env, dt, (struct lu_fid *)rec);
RETURN(rc == 0 ? 1 : rc);
}
}
rc = -zap_lookup(osd->od_objset.os, obj->oo_db->db_object,
(char *)key, 8, sizeof(oti->oti_zde) / 8,
(void *)&oti->oti_zde);
memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
RETURN(rc == 0 ? 1 : rc);
}
static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
const struct dt_key *key, struct thandle *th,
struct lustre_capa *capa)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
dmu_buf_t *zap_db = obj->oo_db;
int rc;
ENTRY;
LASSERT(obj->oo_db);
LASSERT(udmu_object_is_zap(obj->oo_db));
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
/* Remove key from the ZAP */
rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
(char *) key, oh->ot_tx);
if (rc && rc != -ENOENT)
CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
RETURN(rc);
}
/**
* Inserts (key, value) pair in \a directory object.
*
* \param dt osd index object
* \param key key for index
* \param rec record reference
* \param th transaction handler
* \param capa capability descriptor
* \param ignore_quota update should not affect quota
*
* \retval 0 success
* \retval -ve failure
*/
static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
const struct dt_rec *rec, const struct dt_key *key,
struct thandle *th, struct lustre_capa *capa,
int ignore_quota)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *parent = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(parent);
struct lu_fid *fid = (struct lu_fid *)rec;
struct osd_thandle *oh;
struct osd_object *child;
__u32 attr;
int rc;
ENTRY;
LASSERT(parent->oo_db);
LASSERT(udmu_object_is_zap(parent->oo_db));
LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(parent));
/*
* zfs_readdir() generates ./.. on fly, but
* we want own entries (.. at least) with a fid
*/
#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 3, 61, 0)
#warning "fix '.' and '..' handling"
#endif
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
child = osd_object_find(env, dt, fid);
if (IS_ERR(child))
RETURN(PTR_ERR(child));
LASSERT(child->oo_db);
CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
attr = child->oo_dt.do_lu.lo_header ->loh_attr;
oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
oti->oti_zde.lzd_fid = *fid;
/* Insert (key,oid) into ZAP */
rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
(char *)key, 8, sizeof(oti->oti_zde) / 8,
(void *)&oti->oti_zde, oh->ot_tx);
osd_object_put(env, child);
RETURN(rc);
}
static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
const struct dt_key *key, struct thandle *th,
struct lustre_capa *capa)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
struct osd_thandle *oh;
dmu_buf_t *zap_db = obj->oo_db;
char *name = (char *)key;
int rc;
ENTRY;
LASSERT(obj->oo_db);
LASSERT(udmu_object_is_zap(obj->oo_db));
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
/*
* In Orion . and .. were stored in the directory (not generated upon
* request as now). we preserve them for backward compatibility
*/
if (name[0] == '.') {
if (name[1] == 0) {
RETURN(0);
} else if (name[1] == '.' && name[2] == 0) {
RETURN(0);
}
}
/* Remove key from the ZAP */
rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
(char *) key, oh->ot_tx);
#if LUSTRE_VERSION_CODE <= OBD_OCD_VERSION(2, 4, 53, 0)
if (unlikely(rc == -ENOENT && name[0] == '.' &&
(name[1] == 0 || (name[1] == '.' && name[2] == 0))))
rc = 0;
#endif
if (unlikely(rc && rc != -ENOENT))
CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
RETURN(rc);
}
static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
struct dt_rec *rec, const struct dt_key *key,
struct lustre_capa *capa)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
int rc;
ENTRY;
LASSERT(udmu_object_is_zap(obj->oo_db));
rc = -zap_lookup(osd->od_objset.os, obj->oo_db->db_object,
(char *)key, 8, sizeof(oti->oti_zde) / 8,
(void *)&oti->oti_zde);
memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
RETURN(rc == 0 ? 1 : rc);
}
static int osd_declare_dir_insert(const struct lu_env *env,
struct dt_object *dt,
const struct dt_rec *rec,
const struct dt_key *key,
struct thandle *th)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thandle *oh;
ENTRY;
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
LASSERT(obj->oo_db);
LASSERT(udmu_object_is_zap(obj->oo_db));
dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
RETURN(0);
}
/**
* Inserts (key, value) pair in \a directory object.
*
* \param dt osd index object
* \param key key for index
* \param rec record reference
* \param th transaction handler
* \param capa capability descriptor
* \param ignore_quota update should not affect quota
*
* \retval 0 success
* \retval -ve failure
*/
static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
const struct dt_rec *rec, const struct dt_key *key,
struct thandle *th, struct lustre_capa *capa,
int ignore_quota)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *parent = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(parent);
struct lu_fid *fid = (struct lu_fid *)rec;
struct osd_thandle *oh;
struct osd_object *child = NULL;
__u32 attr;
char *name = (char *)key;
int rc;
ENTRY;
LASSERT(parent->oo_db);
LASSERT(udmu_object_is_zap(parent->oo_db));
LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(parent));
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
rc = osd_remote_fid(env, osd, fid);
if (rc < 0) {
CERROR("%s: Can not find object "DFID": rc = %d\n",
osd->od_svname, PFID(fid), rc);
RETURN(rc);
}
if (unlikely(rc == 1)) {
/* Insert remote entry */
memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
oti->oti_zde.lzd_reg.zde_type = IFTODT(S_IFDIR & S_IFMT);
} else {
/*
* To simulate old Orion setups with ./.. stored in the
* directories
*/
/* Insert local entry */
child = osd_object_find(env, dt, fid);
if (IS_ERR(child))
RETURN(PTR_ERR(child));
LASSERT(child->oo_db);
if (name[0] == '.') {
if (name[1] == 0) {
/* do not store ".", instead generate it
* during iteration */
GOTO(out, rc = 0);
} else if (name[1] == '.' && name[2] == 0) {
/* update parent dnode in the child.
* later it will be used to generate ".." */
udmu_objset_t *uos = &osd->od_objset;
rc = osd_object_sa_update(parent,
SA_ZPL_PARENT(uos),
&child->oo_db->db_object,
8, oh);
GOTO(out, rc);
}
}
CLASSERT(sizeof(oti->oti_zde.lzd_reg) == 8);
CLASSERT(sizeof(oti->oti_zde) % 8 == 0);
attr = child->oo_dt.do_lu.lo_header ->loh_attr;
oti->oti_zde.lzd_reg.zde_type = IFTODT(attr & S_IFMT);
oti->oti_zde.lzd_reg.zde_dnode = child->oo_db->db_object;
}
oti->oti_zde.lzd_fid = *fid;
/* Insert (key,oid) into ZAP */
rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
(char *)key, 8, sizeof(oti->oti_zde) / 8,
(void *)&oti->oti_zde, oh->ot_tx);
out:
if (child != NULL)
osd_object_put(env, child);
RETURN(rc);
}
