+/**
+ * Lookup an existing OI by the given name.
+ */
+static int
+osd_oi_lookup(const struct lu_env *env, struct osd_device *o,
+ uint64_t parent, const char *name, struct osd_oi *oi)
+{
+ struct zpl_direntry *zde = &osd_oti_get(env)->oti_zde.lzd_reg;
+ int rc;
+
+ rc = -zap_lookup(o->od_os, parent, name, 8, 1, (void *)zde);
+ if (rc)
+ return rc;
+
+ rc = strlcpy(oi->oi_name, name, sizeof(oi->oi_name));
+ if (rc >= sizeof(oi->oi_name))
+ return -E2BIG;
+
+ oi->oi_zapid = zde->zde_dnode;
+
+ return 0;
+}
+
+static int osd_obj_create(const struct lu_env *env, struct osd_device *o,
+ uint64_t parent, const char *name, uint64_t *child,
+ const struct lu_fid *fid, bool isdir)
+{
+ struct osd_thread_info *info = osd_oti_get(env);
+ struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
+ struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+ struct lu_attr *la = &info->oti_la;
+ sa_handle_t *sa_hdl = NULL;
+ nvlist_t *nvbuf = NULL;
+ dmu_tx_t *tx;
+ uint64_t oid;
+ __u32 compat = LMAC_NOT_IN_OI;
+ int rc;
+ ENTRY;
+
+ if (o->od_dt_dev.dd_rdonly)
+ RETURN(-EROFS);
+
+ memset(la, 0, sizeof(*la));
+ la->la_valid = LA_MODE | LA_UID | LA_GID;
+ la->la_mode = S_IRUGO | S_IWUSR | (isdir ? S_IXUGO | S_IFDIR : S_IFREG);
+
+ if (fid) {
+ rc = -nvlist_alloc(&nvbuf, NV_UNIQUE_NAME, KM_SLEEP);
+ if (rc)
+ RETURN(rc);
+
+ if (o->od_is_ost)
+ compat |= LMAC_FID_ON_OST;
+ lustre_lma_init(lma, fid, compat, 0);
+ lustre_lma_swab(lma);
+ rc = -nvlist_add_byte_array(nvbuf, XATTR_NAME_LMA,
+ (uchar_t *)lma, sizeof(*lma));
+ if (rc)
+ GOTO(out, rc);
+ }
+
+ /* create fid-to-dnode index */
+ tx = dmu_tx_create(o->od_os);
+ if (!tx)
+ GOTO(out, rc = -ENOMEM);
+
+ dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
+ dmu_tx_hold_bonus(tx, parent);
+ dmu_tx_hold_zap(tx, parent, TRUE, name);
+ dmu_tx_hold_sa_create(tx, ZFS_SA_BASE_ATTR_SIZE);
+ rc = -dmu_tx_assign(tx, TXG_WAIT);
+ if (rc) {
+ dmu_tx_abort(tx);
+ GOTO(out, rc);
+ }
+
+ if (isdir)
+ oid = osd_zap_create_flags(o->od_os, 0, ZAP_FLAG_HASH64,
+ DMU_OT_DIRECTORY_CONTENTS,
+ 14, DN_MAX_INDBLKSHIFT, 0, tx);
+ else
+ oid = osd_dmu_object_alloc(o->od_os, DMU_OTN_UINT8_METADATA,
+ 0, 0, tx);
+ rc = -sa_handle_get(o->od_os, oid, NULL, SA_HDL_PRIVATE, &sa_hdl);
+ if (rc)
+ GOTO(commit, rc);
+
+ rc = __osd_attr_init(env, o, NULL, sa_hdl, tx, la, parent, nvbuf);
+ sa_handle_destroy(sa_hdl);
+ if (rc)
+ GOTO(commit, rc);
+
+ zde->zde_dnode = oid;
+ zde->zde_pad = 0;
+ zde->zde_type = S_DT(isdir ? S_IFDIR : S_IFREG);
+ rc = -zap_add(o->od_os, parent, name, 8, 1, (void *)zde, tx);
+
+ GOTO(commit, rc);
+
+commit:
+ if (rc)
+ dmu_object_free(o->od_os, oid, tx);
+ else
+ *child = oid;
+ dmu_tx_commit(tx);
+out:
+ if (nvbuf)
+ nvlist_free(nvbuf);
+ return rc;
+}
+
+static int
+osd_oi_find_or_create(const struct lu_env *env, struct osd_device *o,
+ uint64_t parent, const char *name, uint64_t *child)
+{
+ struct osd_oi oi;
+ int rc;
+
+ rc = osd_oi_lookup(env, o, parent, name, &oi);
+ if (rc == 0)
+ *child = oi.oi_zapid;
+ else if (rc == -ENOENT)
+ rc = osd_obj_create(env, o, parent, name, child, NULL, true);
+
+ return rc;
+}
+
+int osd_obj_find_or_create(const struct lu_env *env, struct osd_device *o,
+ uint64_t parent, const char *name, uint64_t *child,
+ const struct lu_fid *fid, bool isdir)
+{
+ struct osd_oi oi;
+ int rc;
+
+ rc = osd_oi_lookup(env, o, parent, name, &oi);
+ if (!rc)
+ *child = oi.oi_zapid;
+ else if (rc == -ENOENT)
+ rc = osd_obj_create(env, o, parent, name, child, fid, isdir);
+
+ return rc;
+}
+
+/**
+ * Lookup the target index/flags of the fid, so it will know where
+ * the object is located (tgt index) and it is MDT or OST object.
+ */
+int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
+ u64 seq, struct lu_seq_range *range)
+{
+ struct seq_server_site *ss = osd_seq_site(osd);
+
+ if (fid_seq_is_idif(seq)) {
+ fld_range_set_ost(range);
+ range->lsr_index = idif_ost_idx(seq);
+ return 0;
+ }
+
+ if (!fid_seq_in_fldb(seq)) {
+ fld_range_set_mdt(range);
+ if (ss != NULL)
+ /* FIXME: If ss is NULL, it suppose not get lsr_index
+ * at all */
+ range->lsr_index = ss->ss_node_id;
+ return 0;
+ }
+
+ /* The seq_server_site may be NOT ready during initial OI scrub */
+ if (unlikely(!ss || !ss->ss_server_fld ||
+ !ss->ss_server_fld->lsf_cache))
+ return -ENOENT;
+
+ fld_range_set_any(range);
+ /* OSD will only do local fld lookup */
+ return fld_local_lookup(env, ss->ss_server_fld, seq, range);
+}
+
+int fid_is_on_ost(const struct lu_env *env, struct osd_device *osd,
+ const struct lu_fid *fid)
+{
+ struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
+ int rc;
+ ENTRY;
+
+ if (fid_is_idif(fid))
+ RETURN(1);
+
+ if (unlikely(fid_is_local_file(fid) || fid_is_llog(fid)) ||
+ fid_is_name_llog(fid) || fid_is_quota(fid) ||
+ fid_is_igif(fid))
+ RETURN(0);
+
+ rc = osd_fld_lookup(env, osd, fid_seq(fid), range);
+ if (rc != 0) {
+ /* During upgrade, OST FLDB might not be loaded because
+ * OST FLDB is not created until 2.6, so if some DNE
+ * filesystem upgrade from 2.5 to 2.7/2.8, they will
+ * not be able to find the sequence from local FLDB
+ * cache see fld_index_init(). */
+ if (rc == -ENOENT && osd->od_is_ost)
+ RETURN(1);
+
+ if (rc != -ENOENT)
+ CERROR("%s: "DFID" lookup failed: rc = %d\n",
+ osd_name(osd), PFID(fid), rc);
+ RETURN(0);
+ }
+
+ if (fld_range_is_ost(range))
+ RETURN(1);
+
+ RETURN(0);
+}
+
+static struct osd_seq *osd_seq_find_locked(struct osd_seq_list *seq_list,
+ u64 seq)
+{
+ struct osd_seq *osd_seq;
+
+ list_for_each_entry(osd_seq, &seq_list->osl_seq_list, os_seq_list) {
+ if (osd_seq->os_seq == seq)
+ return osd_seq;
+ }
+ return NULL;
+}
+
+static struct osd_seq *osd_seq_find(struct osd_seq_list *seq_list, u64 seq)
+{
+ struct osd_seq *osd_seq;
+
+ read_lock(&seq_list->osl_seq_list_lock);
+ osd_seq = osd_seq_find_locked(seq_list, seq);
+ read_unlock(&seq_list->osl_seq_list_lock);
+
+ return osd_seq;
+}
+
+static struct osd_seq *osd_find_or_add_seq(const struct lu_env *env,
+ struct osd_device *osd, u64 seq)
+{
+ struct osd_seq_list *seq_list = &osd->od_seq_list;
+ struct osd_seq *osd_seq;
+ char *key = osd_oti_get(env)->oti_buf;
+ char *seq_name = osd_oti_get(env)->oti_str;
+ struct osd_oi oi;
+ uint64_t sdb, odb;
+ int i;
+ int rc = 0;
+ ENTRY;
+
+ osd_seq = osd_seq_find(seq_list, seq);
+ if (osd_seq != NULL)
+ RETURN(osd_seq);
+
+ down(&seq_list->osl_seq_init_sem);
+ /* Check again, in case some one else already add it
+ * to the list */
+ osd_seq = osd_seq_find(seq_list, seq);
+ if (osd_seq != NULL)
+ GOTO(out, rc = 0);
+
+ OBD_ALLOC_PTR(osd_seq);
+ if (osd_seq == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ INIT_LIST_HEAD(&osd_seq->os_seq_list);
+ osd_seq->os_seq = seq;
+
+ /* Init subdir count to be 32, but each seq can have
+ * different subdir count */
+ osd_seq->os_subdir_count = OSD_OST_MAP_SIZE;
+ OBD_ALLOC_PTR_ARRAY(osd_seq->os_compat_dirs, osd_seq->os_subdir_count);
+ if (osd_seq->os_compat_dirs == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ oi.oi_zapid = osd->od_O_id;
+ sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
+ fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
+ fid_seq_is_idif(seq) ? 0 : seq);
+
+ rc = osd_oi_find_or_create(env, osd, oi.oi_zapid, seq_name, &odb);
+ if (rc != 0) {
+ CERROR("%s: Can not create %s : rc = %d\n",
+ osd_name(osd), seq_name, rc);
+ GOTO(out, rc);
+ }
+
+ osd_seq->os_oid = odb;
+ for (i = 0; i < OSD_OST_MAP_SIZE; i++) {
+ sprintf(key, "d%d", i);
+ rc = osd_oi_find_or_create(env, osd, odb, key, &sdb);
+ if (rc)
+ GOTO(out, rc);
+ osd_seq->os_compat_dirs[i] = sdb;
+ }
+
+ write_lock(&seq_list->osl_seq_list_lock);
+ list_add(&osd_seq->os_seq_list, &seq_list->osl_seq_list);
+ write_unlock(&seq_list->osl_seq_list_lock);
+out:
+ up(&seq_list->osl_seq_init_sem);
+ if (rc != 0) {
+ if (osd_seq != NULL && osd_seq->os_compat_dirs != NULL)
+ OBD_FREE_PTR_ARRAY(osd_seq->os_compat_dirs,
+ osd_seq->os_subdir_count);
+ if (osd_seq != NULL)
+ OBD_FREE_PTR(osd_seq);
+ osd_seq = ERR_PTR(rc);
+ }
+ RETURN(osd_seq);
+}
+
+static uint64_t
+osd_get_idx_for_ost_obj_compat(const struct lu_env *env, struct osd_device *osd,
+ const struct lu_fid *fid, char *buf, int bufsize)
+{
+ struct osd_seq *osd_seq;
+ unsigned long b;
+ u64 id;
+ int rc;
+
+ osd_seq = osd_find_or_add_seq(env, osd, fid_seq(fid));
+ if (IS_ERR(osd_seq)) {
+ CERROR("%s: Can not find seq group "DFID"\n", osd_name(osd),
+ PFID(fid));
+ return PTR_ERR(osd_seq);
+ }
+
+ if (fid_is_last_id(fid)) {
+ id = 0;
+ } else {
+ rc = fid_to_ostid(fid, &osd_oti_get(env)->oti_ostid);
+ LASSERT(rc == 0); /* we should not get here with IGIF */
+ id = ostid_id(&osd_oti_get(env)->oti_ostid);
+ }
+
+ b = id % OSD_OST_MAP_SIZE;
+ LASSERT(osd_seq->os_compat_dirs[b]);
+
+ if (buf)
+ snprintf(buf, bufsize, "%llu", id);
+
+ return osd_seq->os_compat_dirs[b];
+}
+