+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = cfid;
+ rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_fid = lfsck_dto2fid(parent);
+ rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th);
+ if (rc == 0)
+ rc = dt_declare_ref_add(env, orphan, th);
+
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_declare_ref_add(env, orphan, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (lmv != NULL) {
+ lmv->lmv_magic = LMV_MAGIC;
+ lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
+ lfsck_lmv_header_cpu_to_le(lmv2, lmv);
+ lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
+ rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
+ XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
+ rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_fid = cfid;
+ rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th);
+ if (rc == 0)
+ rc = dt_declare_ref_add(env, parent, th);
+
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, orphan, 0);
+ rc = dt_create(env, orphan, la, hint, dof, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ rec->rec_fid = cfid;
+ rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th, 1);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ rec->rec_fid = lfsck_dto2fid(parent);
+ rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th, 1);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ rc = dt_ref_add(env, orphan, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ if (lmv != NULL) {
+ rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+ }
+
+ rc = dt_xattr_set(env, orphan, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ dt_write_unlock(env, orphan);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_fid = cfid;
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th, 1);
+ if (rc == 0) {
+ dt_write_lock(env, parent, 0);
+ rc = dt_ref_add(env, parent, th);
+ dt_write_unlock(env, parent);
+ }
+
+ GOTO(stop, rc = (rc == 0 ? 1 : rc));
+
+unlock2:
+ dt_write_unlock(env, orphan);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+unlock1:
+ lfsck_unlock(llh);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
+ "the object "DFID", name = %s: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(cfid),
+ cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
+
+ if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
+ lfsck_object_put(env, parent);
+
+ if (rc != 0)
+ ns->ln_flags |= LF_INCONSISTENT;
+
+ return rc;
+}
+
+/**
+ * Remove the specified entry from the linkEA.
+ *
+ * Locate the linkEA entry with the given @cname and @pfid, then
+ * remove this entry or the other entries those are repeated with
+ * this entry.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ * \param[in,out]ldata pointer to the buffer that holds the linkEA
+ * \param[in] cname the name for the child in the parent directory
+ * \param[in] pfid the parent directory's FID for the linkEA
+ * \param[in] next if true, then remove the first found linkEA
+ * entry, and move the ldata->ld_lee to next entry
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj,
+ struct linkea_data *ldata,
+ struct lu_name *cname,
+ struct lu_fid *pfid,
+ bool next)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct thandle *th = NULL;
+ struct lustre_handle lh = { 0 };
+ struct linkea_data ldata_new = { NULL };
+ struct lu_buf linkea_buf;
+ int rc = 0;
+ ENTRY;
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
+ LCK_EX);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ if (next)
+ linkea_del_buf(ldata, cname);
+ else
+ lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
+ true);
+ lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
+
+again:
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(unlock1, rc = PTR_ERR(th));
+
+ rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ if (unlikely(lfsck_is_dead_obj(obj)))
+ GOTO(unlock2, rc = -ENOENT);
+
+ rc = lfsck_links_read2(env, obj, &ldata_new);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ /* The specified linkEA entry has been removed by race. */
+ rc = linkea_links_find(&ldata_new, cname, pfid);
+ if (rc != 0)
+ GOTO(unlock2, rc = 0);
+
+ if (bk->lb_param & LPF_DRYRUN)
+ GOTO(unlock2, rc = 1);
+
+ if (next)
+ linkea_del_buf(&ldata_new, cname);
+ else
+ lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
+ true);
+
+ if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
+ dt_write_unlock(env, obj);
+ dt_trans_stop(env, dev, th);
+ lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
+ ldata_new.ld_leh->leh_len);
+ goto again;
+ }
+
+ lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
+ ldata_new.ld_leh->leh_len);
+ rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
+
+ GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
+
+unlock2:
+ dt_write_unlock(env, obj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+unlock1:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
+ "for the object: "DFID", parent "DFID", name %.*s\n",
+ lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
+ PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
+ cname->ln_name);
+
+ if (rc != 0) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ ns->ln_flags |= LF_INCONSISTENT;
+ }
+
+ return rc;
+}
+
+/**
+ * Conditionally remove the specified entry from the linkEA.
+ *
+ * Take the parent lock firstly, then check whether the specified
+ * name entry exists or not: if yes, do nothing; otherwise, call
+ * lfsck_namespace_shrink_linkea() to remove the linkea entry.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] parent pointer to the parent directory
+ * \param[in] child pointer to the child object that holds the linkEA
+ * \param[in,out]ldata pointer to the buffer that holds the linkEA
+ * \param[in] cname the name for the child in the parent directory
+ * \param[in] pfid the parent directory's FID for the linkEA
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *parent,
+ struct dt_object *child,
+ struct linkea_data *ldata,
+ struct lu_name *cname,
+ struct lu_fid *pfid)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_fid *cfid = &info->lti_fid3;
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ int rc;
+ ENTRY;
+
+ rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
+ MDS_INODELOCK_UPDATE, LCK_PR);
+ if (rc != 0)
+ RETURN(rc);
+
+ dt_read_lock(env, parent, 0);
+ if (unlikely(lfsck_is_dead_obj(parent))) {
+ dt_read_unlock(env, parent);
+ lfsck_unlock(llh);
+ rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
+ cname, pfid, true);
+
+ RETURN(rc);
+ }
+
+ rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
+ (const struct dt_key *)cname->ln_name);
+ dt_read_unlock(env, parent);
+
+ /* It is safe to release the ldlm lock, because when the logic come
+ * here, we have got all the needed information above whether the
+ * linkEA entry is valid or not. It is not important that others
+ * may add new linkEA entry after the ldlm lock released. If other
+ * has removed the specified linkEA entry by race, then it is OK,
+ * because the subsequent lfsck_namespace_shrink_linkea() can handle
+ * such case. */
+ lfsck_unlock(llh);
+ if (rc == -ENOENT) {
+ rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
+ cname, pfid, true);
+
+ RETURN(rc);
+ }
+
+ if (rc != 0)
+ RETURN(rc);
+
+ /* The LFSCK just found some internal status of cross-MDTs
+ * create operation. That is normal. */
+ if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
+ linkea_next_entry(ldata);
+
+ RETURN(0);
+ }
+
+ rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
+ pfid, true);
+
+ RETURN(rc);
+}
+
+/**
+ * Conditionally replace name entry in the parent.
+ *
+ * As required, the LFSCK may re-create the lost MDT-object for dangling
+ * name entry, but such repairing may be wrong because of bad FID in the
+ * name entry. As the LFSCK processing, the real MDT-object may be found,
+ * then the LFSCK should check whether the former re-created MDT-object
+ * has been modified or not, if not, then destroy it and update the name
+ * entry in the parent to reference the real MDT-object.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] parent pointer to the parent directory
+ * \param[in] child pointer to the MDT-object that may be the real
+ * MDT-object corresponding to the name entry in parent
+ * \param[in] cfid the current FID in the name entry
+ * \param[in] cname contains the name of the child in the parent directory
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_replace_cond(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *parent,
+ struct dt_object *child,
+ const struct lu_fid *cfid,
+ const struct lu_name *cname)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lu_fid tfid;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ /* The child and its name may be on different MDTs. */
+ struct dt_device *dev = lfsck->li_next;
+ const char *name = cname->ln_name;
+ struct dt_object *pobj = NULL;
+ struct dt_object *cobj = NULL;
+ struct lfsck_lock_handle *pllh = &info->lti_llh;
+ struct lustre_handle clh = { 0 };
+ struct linkea_data ldata = { NULL };
+ struct thandle *th = NULL;
+ bool exist = true;
+ int rc = 0;
+ ENTRY;
+
+ /* @parent/@child may be based on lfsck->li_bottom,
+ * but here we need the object based on the lfsck->li_next. */
+
+ pobj = lfsck_object_locate(dev, parent);
+ if (IS_ERR(pobj))
+ GOTO(log, rc = PTR_ERR(pobj));
+
+ if (unlikely(!dt_try_as_dir(env, pobj)))
+ GOTO(log, rc = -ENOTDIR);
+
+ rc = lfsck_lock(env, lfsck, parent, name, pllh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ if (!fid_is_sane(cfid)) {
+ exist = false;
+ goto replace;
+ }
+
+ cobj = lfsck_object_find_by_dev(env, dev, cfid);
+ if (IS_ERR(cobj)) {
+ rc = PTR_ERR(cobj);
+ if (rc == -ENOENT) {
+ exist = false;
+ goto replace;
+ }
+
+ GOTO(log, rc);
+ }
+
+ if (!dt_object_exists(cobj)) {
+ exist = false;
+ goto replace;
+ }
+
+ rc = dt_lookup(env, pobj, (struct dt_rec *)&tfid,
+ (const struct dt_key *)name);
+ if (rc == -ENOENT) {
+ exist = false;
+ goto replace;
+ }
+
+ if (rc != 0)
+ GOTO(log, rc);
+
+ /* Someone changed the name entry, cannot replace it. */
+ if (!lu_fid_eq(cfid, &tfid))
+ GOTO(log, rc = 0);
+
+ /* lock the object to be destroyed. */
+ rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
+ LCK_EX);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ if (unlikely(lfsck_is_dead_obj(cobj))) {
+ exist = false;
+ goto replace;
+ }
+
+ rc = dt_attr_get(env, cobj, la);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ /* The object has been modified by other(s), or it is not created by
+ * LFSCK, the two cases are indistinguishable. So cannot replace it. */
+ if (la->la_ctime != 0)
+ GOTO(log, rc);
+
+ if (S_ISREG(la->la_mode)) {
+ rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
+ /* If someone has created related OST-object(s),
+ * then keep it. */
+ if ((rc > 0) || (rc < 0 && rc != -ENODATA))
+ GOTO(log, rc = (rc > 0 ? 0 : rc));
+ }
+
+replace:
+ dt_read_lock(env, child, 0);
+ rc = lfsck_links_read2(env, child, &ldata);
+ dt_read_unlock(env, child);
+
+ /* Someone changed the child, no need to replace. */
+ if (rc == -ENODATA)
+ GOTO(log, rc = 0);
+
+ if (rc != 0)
+ GOTO(log, rc);
+
+ rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(pobj));
+ /* Someone moved the child, no need to replace. */
+ if (rc != 0)
+ GOTO(log, rc = 0);
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(log, rc = 1);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ if (exist) {
+ rc = dt_declare_destroy(env, cobj, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ rc = dt_declare_delete(env, pobj, (const struct dt_key *)name, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = lfsck_dto2fid(child);
+ rc = dt_declare_insert(env, pobj, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (exist) {
+ rc = dt_destroy(env, cobj, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ /* The old name entry maybe not exist. */
+ rc = dt_delete(env, pobj, (const struct dt_key *)name, th);
+ if (rc != 0 && rc != -ENOENT)
+ GOTO(stop, rc);
+
+ rc = dt_insert(env, pobj, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th, 1);
+
+ GOTO(stop, rc = (rc == 0 ? 1 : rc));
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ lfsck_ibits_unlock(&clh, LCK_EX);
+ lfsck_unlock(pllh);
+
+ if (cobj != NULL && !IS_ERR(cobj))
+ lfsck_object_put(env, cobj);
+
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
+ "object "DFID" because of conflict with the object "DFID
+ " under the parent "DFID" with name %s: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(cfid),
+ PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
+ name, rc);
+
+ return rc;
+}
+
+/**
+ * Overwrite the linkEA for the object with the given ldata.
+ *
+ * The caller should take the ldlm lock before the calling.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ * \param[in] ldata pointer to the new linkEA data
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj,
+ struct linkea_data *ldata)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct thandle *th = NULL;
+ struct lu_buf linkea_buf;
+ int rc = 0;
+ ENTRY;
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
+ rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ if (unlikely(lfsck_is_dead_obj(obj)))
+ GOTO(unlock, rc = 0);
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock, rc = 1);
+
+ rc = dt_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+
+ GOTO(unlock, rc = (rc == 0 ? 1 : rc));
+
+unlock:
+ dt_write_unlock(env, obj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
+ "object "DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
+
+ if (rc != 0) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ ns->ln_flags |= LF_INCONSISTENT;
+ }
+
+ return rc;
+}
+
+/**
+ * Repair invalid name entry.
+ *
+ * If the name entry contains invalid information, such as bad file type
+ * or (and) corrupted object FID, then either remove the name entry or
+ * udpate the name entry with the given (right) information.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] parent pointer to the parent directory
+ * \param[in] child pointer to the object referenced by the name entry
+ * \param[in] name the old name of the child under the parent directory
+ * \param[in] name2 the new name of the child under the parent directory
+ * \param[in] type the type claimed by the name entry
+ * \param[in] update update the name entry if true; otherwise, remove it
+ * \param[in] dec decrease the parent nlink count if true
+ *
+ * \retval positive number for repaired successfully
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+int lfsck_namespace_repair_dirent(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *parent,
+ struct dt_object *child,
+ const char *name, const char *name2,
+ __u16 type, bool update, bool dec)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct lu_fid tfid;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_object *dto;
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ struct lustre_handle lh = { 0 };
+ int rc = 0;
+ ENTRY;
+
+ if (unlikely(!dt_try_as_dir(env, parent)))
+ GOTO(log, rc = -ENOTDIR);
+
+ if (!update || strcmp(name, name2) == 0)
+ rc = lfsck_lock(env, lfsck, parent, name, llh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
+ else
+ rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(unlock1, rc = PTR_ERR(th));
+
+ dto = dt_object_locate(parent, th->th_dev);
+ rc = dt_declare_delete(env, dto, (const struct dt_key *)name, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (update) {
+ rec->rec_type = lfsck_object_type(child) & S_IFMT;
+ rec->rec_fid = cfid;
+ rc = dt_declare_insert(env, dto,
+ (const struct dt_rec *)rec,
+ (const struct dt_key *)name2, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ if (dec) {
+ rc = dt_declare_ref_del(env, dto, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+
+ dt_write_lock(env, dto, 0);
+ rc = dt_lookup(env, dto, (struct dt_rec *)&tfid,
+ (const struct dt_key *)name);
+ /* Someone has removed the bad name entry by race. */
+ if (rc == -ENOENT)
+ GOTO(unlock2, rc = 0);
+
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ /* Someone has removed the bad name entry and reused it for other
+ * object by race. */
+ if (!lu_fid_eq(&tfid, cfid))
+ GOTO(unlock2, rc = 0);
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock2, rc = 1);
+
+ rc = dt_delete(env, dto, (const struct dt_key *)name, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ if (update) {
+ rc = dt_insert(env, dto,
+ (const struct dt_rec *)rec,
+ (const struct dt_key *)name2, th, 1);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+ }
+
+ if (dec) {
+ rc = dt_ref_del(env, dto, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+ }
+
+ GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
+
+unlock2:
+ dt_write_unlock(env, parent);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+ /* We are not sure whether the child will become orphan or not.
+ * Record it in the LFSCK trace file for further checking in
+ * the second-stage scanning. */
+ if (!update && !dec && rc == 0)
+ lfsck_namespace_trace_update(env, com, cfid,
+ LNTF_CHECK_LINKEA, true);
+
+unlock1:
+ /* It is harmless even if unlock the unused lock_handle */
+ lfsck_ibits_unlock(&lh, LCK_PW);
+ lfsck_unlock(llh);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
+ "entry for: parent "DFID", child "DFID", name %s, type "
+ "in name entry %o, type claimed by child %o. repair it "
+ "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
+ PFID(lfsck_dto2fid(parent)), PFID(cfid),
+ name, type, update ? lfsck_object_type(child) : 0,
+ update ? "updating" : "removing", name2, rc);
+
+ if (rc != 0) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ ns->ln_flags |= LF_INCONSISTENT;
+ }
+
+ return rc;
+}
+
+/**
+ * Update the ".." name entry for the given object.
+ *
+ * The object's ".." is corrupted, this function will update the ".." name
+ * entry with the given pfid, and the linkEA with the given ldata.
+ *
+ * The caller should take the ldlm lock before the calling.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ * \param[in] pfid the new fid for the object's ".." name entry
+ * \param[in] cname the name for the @obj in the parent directory
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj,
+ const struct lu_fid *pfid,
+ struct lu_name *cname)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct thandle *th = NULL;
+ struct linkea_data ldata = { NULL };
+ struct lu_buf linkea_buf;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(obj));
+ LASSERT(S_ISDIR(lfsck_object_type(obj)));
+
+ rc = linkea_data_new(&ldata, &info->lti_big_buf);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ rc = linkea_add_buf(&ldata, cname, pfid);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = pfid;
+ rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ if (unlikely(lfsck_is_dead_obj(obj)))
+ GOTO(unlock, rc = 0);
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock, rc = 1);
+
+ /* The old ".." name entry maybe not exist. */
+ dt_delete(env, obj, (const struct dt_key *)dotdot, th);
+
+ rc = dt_insert(env, obj, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th, 1);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ rc = dt_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+
+ GOTO(unlock, rc = (rc == 0 ? 1 : rc));
+
+unlock:
+ dt_write_unlock(env, obj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
+ "the object "DFID", new parent "DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
+ PFID(pfid), rc);
+
+ if (rc != 0) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ ns->ln_flags |= LF_INCONSISTENT;
+ }
+
+ return rc;
+}
+
+/**
+ * Handle orphan @obj during Double Scan Directory.
+ *
+ * Remove the @obj's current (invalid) linkEA entries, and insert
+ * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
+ * ${FID}-${PFID}-D-${conflict_version}
+ *
+ * The caller should take the ldlm lock before the calling.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the orphan object to be handled
+ * \param[in] pfid the new fid for the object's ".." name entry
+ * \param[in,out] lh ldlm lock handler for the given @obj
+ * \param[out] type to tell the caller what the inconsistency is
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int
+lfsck_namespace_dsd_orphan(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj,
+ const struct lu_fid *pfid,
+ struct lustre_handle *lh,
+ enum lfsck_namespace_inconsistency_type *type)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ int rc;
+ ENTRY;
+
+ /* Remove the unrecognized linkEA. */
+ rc = lfsck_namespace_links_remove(env, com, obj);
+ lfsck_ibits_unlock(lh, LCK_EX);
+ if (rc < 0 && rc != -ENODATA)
+ RETURN(rc);
+
+ *type = LNIT_MUL_REF;
+
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
+ * ever tried to verify some remote MDT-object that resides on this
+ * MDT, but this MDT failed to respond such request. So means there
+ * may be some remote name entry on other MDT that references this
+ * object with another name, so we cannot know whether this linkEA
+ * is valid or not. So keep it there and maybe resolved when next
+ * LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ RETURN(0);
+
+ /* The unique linkEA is invalid, even if the ".." name entry may be
+ * valid, we still cannot know via which name entry this directory
+ * will be referenced. Then handle it as pure orphan. */
+ snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com, obj,
+ info->lti_tmpbuf, "D", NULL);
+
+ RETURN(rc);
+}
+
+/**
+ * Double Scan Directory object for single linkEA entry case.
+ *
+ * The given @child has unique linkEA entry. If the linkEA entry is valid,
+ * then check whether the name is in the namespace or not, if not, add the
+ * missing name entry back to namespace. If the linkEA entry is invalid,
+ * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
+ * as an orphan.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] child pointer to the directory to be double scanned
+ * \param[in] pfid the FID corresponding to the ".." entry
+ * \param[in] ldata pointer to the linkEA data for the given @child
+ * \param[in,out] lh ldlm lock handler for the given @child
+ * \param[out] type to tell the caller what the inconsistency is
+ * \param[in] retry if found inconsistency, but the caller does not hold
+ * ldlm lock on the @child, then set @retry as true
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int
+lfsck_namespace_dsd_single(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *child,
+ const struct lu_fid *pfid,
+ struct linkea_data *ldata,
+ struct lustre_handle *lh,
+ enum lfsck_namespace_inconsistency_type *type,
+ bool *retry)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_name *cname = &info->lti_name;
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct lu_fid tfid;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_object *parent = NULL;
+ struct lmv_mds_md_v1 *lmv;
+ int rc = 0;
+ ENTRY;
+
+ lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
+ /* The unique linkEA entry with bad parent will be handled as orphan. */
+ if (!fid_is_sane(&tfid)) {
+ if (!lustre_handle_is_used(lh) && retry != NULL)
+ *retry = true;
+ else
+ rc = lfsck_namespace_dsd_orphan(env, com, child,
+ pfid, lh, type);
+
+ GOTO(out, rc);
+ }
+
+ parent = lfsck_object_find_bottom(env, lfsck, &tfid);
+ if (IS_ERR(parent))
+ GOTO(out, rc = PTR_ERR(parent));
+
+ /* We trust the unique linkEA entry in spite of whether it matches the
+ * ".." name entry or not. Because even if the linkEA entry is wrong
+ * and the ".." name entry is right, we still cannot know via which
+ * name entry the child will be referenced, since all known entries
+ * have been verified during the first-stage scanning. */
+ if (!dt_object_exists(parent)) {
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
+ * has ever tried to verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond such request. So
+ * means there may be some remote name entry on other MDT that
+ * references this object with another name, so we cannot know
+ * whether this linkEA is valid or not. So keep it there and
+ * maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(out, rc = 0);
+
+ if (!lustre_handle_is_used(lh) && retry != NULL) {
+ *retry = true;
+
+ GOTO(out, rc = 0);
+ }
+
+ lfsck_ibits_unlock(lh, LCK_EX);
+
+lost_parent:
+ lmv = &info->lti_lmv;
+ rc = lfsck_read_stripe_lmv(env, child, lmv);
+ if (rc != 0 && rc != -ENODATA)
+ GOTO(out, rc);
+
+ if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
+ lmv = NULL;
+ } else if (lfsck_shard_name_to_index(env,
+ cname->ln_name, cname->ln_namelen,
+ S_IFDIR, cfid) < 0) {
+ /* It is an invalid name entry, we
+ * cannot trust the parent also. */
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ ldata, cname, &tfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com, child,
+ info->lti_tmpbuf, "S", NULL);
+
+ GOTO(out, rc);
+ }
+
+ /* Create the lost parent as an orphan. */
+ rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
+ if (rc >= 0) {
+ /* Add the missing name entry to the parent. */
+ rc = lfsck_namespace_insert_normal(env, com, parent,
+ child, cname->ln_name);
+ if (unlikely(rc == -EEXIST)) {
+ /* Unfortunately, someone reused the name
+ * under the parent by race. So we have
+ * to remove the linkEA entry from
+ * current child object. It means that the
+ * LFSCK cannot recover the system
+ * totally back to its original status,
+ * but it is necessary to make the
+ * current system to be consistent. */
+ rc = lfsck_namespace_shrink_linkea(env,
+ com, child, ldata,
+ cname, &tfid, true);
+ if (rc >= 0) {
+ snprintf(info->lti_tmpbuf,
+ sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env,
+ com, child, info->lti_tmpbuf,
+ "D", NULL);
+ }
+ }
+ }
+
+ GOTO(out, rc);
+ }
+
+ /* The unique linkEA entry with bad parent will be handled as orphan. */
+ if (unlikely(!dt_try_as_dir(env, parent))) {
+ if (!lustre_handle_is_used(lh) && retry != NULL)
+ *retry = true;
+ else
+ rc = lfsck_namespace_dsd_orphan(env, com, child,
+ pfid, lh, type);
+
+ GOTO(out, rc);
+ }
+
+ rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
+ (const struct dt_key *)cname->ln_name);
+ if (rc == -ENOENT) {
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
+ * has ever tried to verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond such request. So
+ * means there may be some remote name entry on other MDT that
+ * references this object with another name, so we cannot know
+ * whether this linkEA is valid or not. So keep it there and
+ * maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(out, rc = 0);
+
+ if (!lustre_handle_is_used(lh) && retry != NULL) {
+ *retry = true;
+
+ GOTO(out, rc = 0);
+ }
+
+ lfsck_ibits_unlock(lh, LCK_EX);
+ rc = lfsck_namespace_check_name(env, parent, child, cname);
+ if (rc == -ENOENT)
+ goto lost_parent;
+
+ if (rc < 0)
+ GOTO(out, rc);
+
+ /* It is an invalid name entry, drop it. */
+ if (unlikely(rc > 0)) {
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ ldata, cname, &tfid, true);
+ if (rc >= 0) {
+ snprintf(info->lti_tmpbuf,
+ sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com,
+ child, info->lti_tmpbuf, "D", NULL);
+ }
+
+ GOTO(out, rc);
+ }
+
+ /* Add the missing name entry back to the namespace. */
+ rc = lfsck_namespace_insert_normal(env, com, parent, child,
+ cname->ln_name);
+ if (unlikely(rc == -ESTALE))
+ /* It may happen when the remote object has been
+ * removed, but the local MDT is not aware of that. */
+ goto lost_parent;
+
+ if (unlikely(rc == -EEXIST)) {
+ /* Unfortunately, someone reused the name under the
+ * parent by race. So we have to remove the linkEA
+ * entry from current child object. It means that the
+ * LFSCK cannot recover the system totally back to
+ * its original status, but it is necessary to make
+ * the current system to be consistent.
+ *
+ * It also may be because of the LFSCK found some
+ * internal status of create operation. Under such
+ * case, nothing to be done. */
+ rc = lfsck_namespace_shrink_linkea_cond(env, com,
+ parent, child, ldata, cname, &tfid);
+ if (rc >= 0) {
+ snprintf(info->lti_tmpbuf,
+ sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com,
+ child, info->lti_tmpbuf, "D", NULL);
+ }
+ }
+
+ GOTO(out, rc);
+ }
+
+ if (rc != 0)
+ GOTO(out, rc);
+
+ if (!lu_fid_eq(&tfid, cfid)) {
+ if (!lustre_handle_is_used(lh) && retry != NULL) {
+ *retry = true;
+
+ GOTO(out, rc = 0);
+ }
+
+ lfsck_ibits_unlock(lh, LCK_EX);
+ /* The name entry references another MDT-object that
+ * may be created by the LFSCK for repairing dangling
+ * name entry. Try to replace it. */
+ rc = lfsck_namespace_replace_cond(env, com, parent, child,
+ &tfid, cname);
+ if (rc == 0)
+ rc = lfsck_namespace_dsd_orphan(env, com, child,
+ pfid, lh, type);
+
+ GOTO(out, rc);
+ }
+
+ if (fid_is_zero(pfid))
+ GOTO(out, rc = 0);
+
+ /* The ".." name entry is wrong, update it. */
+ if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
+ if (!lustre_handle_is_used(lh) && retry != NULL) {
+ *retry = true;
+
+ GOTO(out, rc = 0);
+ }
+
+ *type = LNIT_UNMATCHED_PAIRS;
+ rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
+ lfsck_dto2fid(parent), cname);
+ }
+
+ GOTO(out, rc);
+
+out:
+ if (parent != NULL && !IS_ERR(parent))
+ lfsck_object_put(env, parent);
+
+ return rc;
+}
+
+/**
+ * Double Scan Directory object for multiple linkEA entries case.
+ *
+ * The given @child has multiple linkEA entries. There is at most one linkEA
+ * entry will be valid, all the others will be removed. Firstly, the function
+ * will try to find out the linkEA entry for which the name entry exists under
+ * the given parent (@pfid). If there is no linkEA entry that matches the given
+ * ".." name entry, then tries to find out the first linkEA entry that both the
+ * parent and the name entry exist to rebuild a new ".." name entry.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] child pointer to the directory to be double scanned
+ * \param[in] pfid the FID corresponding to the ".." entry
+ * \param[in] ldata pointer to the linkEA data for the given @child
+ * \param[in,out] lh ldlm lock handler for the given @child
+ * \param[out] type to tell the caller what the inconsistency is
+ * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int
+lfsck_namespace_dsd_multiple(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *child,
+ const struct lu_fid *pfid,
+ struct linkea_data *ldata,
+ struct lustre_handle *lh,
+ enum lfsck_namespace_inconsistency_type *type,
+ bool lpf)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_name *cname = &info->lti_name;
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct lu_fid *pfid2 = &info->lti_fid3;
+ struct lu_fid tfid;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct dt_object *parent = NULL;
+ struct linkea_data ldata_new = { NULL };
+ int dirent_count = 0;
+ int linkea_count = 0;
+ int rc = 0;
+ bool once = true;
+ ENTRY;
+
+again:
+ while (ldata->ld_lee != NULL) {
+ lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
+ info->lti_key);
+ /* Drop repeated linkEA entries. */
+ lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
+ /* Drop invalid linkEA entry. */
+ if (!fid_is_sane(&tfid)) {
+ linkea_del_buf(ldata, cname);
+ linkea_count++;
+ continue;
+ }
+
+ /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
+ * then it is possible that: the directry object has ever
+ * been lost, but its name entry was there. In the former
+ * LFSCK run, during the first-stage scanning, the LFSCK
+ * found the dangling name entry, but it did not recreate
+ * the lost object, and when moved to the second-stage
+ * scanning, some children objects of the lost directory
+ * object were found, then the LFSCK recreated such lost
+ * directory object as an orphan.
+ *
+ * When the LFSCK runs again, if the dangling name is still
+ * there, the LFSCK should move the orphan directory object
+ * back to the normal namespace. */
+ if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
+ linkea_next_entry(ldata);
+ continue;
+ }
+
+ parent = lfsck_object_find_bottom(env, lfsck, &tfid);
+ if (IS_ERR(parent))
+ RETURN(PTR_ERR(parent));
+
+ if (!dt_object_exists(parent)) {
+ lfsck_object_put(env, parent);
+ if (ldata->ld_leh->leh_reccount > 1) {
+ /* If it is NOT the last linkEA entry, then
+ * there is still other chance to make the
+ * child to be visible via other parent, then
+ * remove this linkEA entry. */
+ linkea_del_buf(ldata, cname);
+ linkea_count++;
+ continue;
+ }
+
+ break;
+ }
+
+ /* The linkEA entry with bad parent will be removed. */
+ if (unlikely(!dt_try_as_dir(env, parent))) {
+ lfsck_object_put(env, parent);
+ linkea_del_buf(ldata, cname);
+ linkea_count++;
+ continue;
+ }
+
+ rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
+ (const struct dt_key *)cname->ln_name);
+ *pfid2 = *lfsck_dto2fid(parent);
+ if (rc == -ENOENT) {
+ lfsck_object_put(env, parent);
+ linkea_next_entry(ldata);
+ continue;
+ }
+
+ if (rc != 0) {
+ lfsck_object_put(env, parent);
+
+ RETURN(rc);
+ }
+
+ if (lu_fid_eq(&tfid, cfid)) {
+ lfsck_object_put(env, parent);
+ if (!lu_fid_eq(pfid, pfid2)) {
+ *type = LNIT_UNMATCHED_PAIRS;
+ rc = lfsck_namespace_repair_unmatched_pairs(env,
+ com, child, pfid2, cname);
+
+ RETURN(rc);
+ }
+
+rebuild:
+ /* It is the most common case that we find the
+ * name entry corresponding to the linkEA entry
+ * that matches the ".." name entry. */
+ rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
+ if (rc != 0)
+ RETURN(rc);
+
+ rc = linkea_add_buf(&ldata_new, cname, pfid2);
+ if (rc != 0)
+ RETURN(rc);
+
+ rc = lfsck_namespace_rebuild_linkea(env, com, child,
+ &ldata_new);
+ if (rc < 0)
+ RETURN(rc);
+
+ linkea_del_buf(ldata, cname);
+ linkea_count++;
+ linkea_first_entry(ldata);
+ /* There may be some invalid dangling name entries under
+ * other parent directories, remove all of them. */
+ while (ldata->ld_lee != NULL) {
+ lfsck_namespace_unpack_linkea_entry(ldata,
+ cname, &tfid, info->lti_key);
+ if (!fid_is_sane(&tfid))
+ goto next;
+
+ parent = lfsck_object_find_bottom(env, lfsck,
+ &tfid);
+ if (IS_ERR(parent)) {
+ rc = PTR_ERR(parent);
+ if (rc != -ENOENT &&
+ bk->lb_param & LPF_FAILOUT)
+ RETURN(rc);
+
+ goto next;
+ }
+
+ if (!dt_object_exists(parent)) {
+ lfsck_object_put(env, parent);
+ goto next;
+ }
+
+ rc = lfsck_namespace_repair_dirent(env, com,
+ parent, child, cname->ln_name,
+ cname->ln_name, S_IFDIR, false, true);
+ lfsck_object_put(env, parent);
+ if (rc < 0) {
+ if (bk->lb_param & LPF_FAILOUT)
+ RETURN(rc);
+
+ goto next;
+ }
+
+ dirent_count += rc;
+
+next:
+ linkea_del_buf(ldata, cname);
+ }
+
+ ns->ln_dirent_repaired += dirent_count;
+
+ RETURN(rc);
+ }
+
+ lfsck_ibits_unlock(lh, LCK_EX);
+ /* The name entry references another MDT-object that may be
+ * created by the LFSCK for repairing dangling name entry.
+ * Try to replace it. */
+ rc = lfsck_namespace_replace_cond(env, com, parent, child,
+ &tfid, cname);
+ lfsck_object_put(env, parent);
+ if (rc < 0)
+ RETURN(rc);
+
+ if (rc > 0)
+ goto rebuild;
+
+ linkea_del_buf(ldata, cname);
+ }
+
+ linkea_first_entry(ldata);
+ if (ldata->ld_leh->leh_reccount == 1) {
+ rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
+ lh, type, NULL);
+
+ if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
+ rc = lfsck_namespace_rebuild_linkea(env, com, child,
+ ldata);
+
+ RETURN(rc);
+ }
+
+ /* All linkEA entries are invalid and removed, then handle the @child
+ * as an orphan.*/
+ if (ldata->ld_leh->leh_reccount == 0) {
+ rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
+ type);
+
+ RETURN(rc);
+ }
+
+ /* If the dangling name entry for the orphan directory object has
+ * been remvoed, then just check whether the directory object is
+ * still under the .lustre/lost+found/MDTxxxx/ or not. */
+ if (lpf) {
+ lpf = false;
+ goto again;
+ }
+
+ /* There is no linkEA entry that matches the ".." name entry. Find
+ * the first linkEA entry that both parent and name entry exist to
+ * rebuild a new ".." name entry. */
+ if (once) {
+ once = false;
+ goto again;
+ }
+
+ RETURN(rc);
+}
+
+/**
+ * Repair the object's nlink attribute.
+ *
+ * If all the known name entries have been verified, then the object's hard
+ * link attribute should match the object's linkEA entries count unless the
+ * object's has too much hard link to be recorded in the linkEA. Such cases
+ * should have been marked in the LFSCK trace file. Otherwise, trust the
+ * linkEA to update the object's nlink attribute.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ * \param[in,out] la pointer to buffer to object's attribute before
+ * and after the repairing
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_repair_nlink(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj,
+ struct lu_attr *la)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_fid *tfid = &info->lti_fid3;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ const struct lu_fid *cfid = lfsck_dto2fid(obj);
+ struct thandle *th = NULL;
+ struct linkea_data ldata = { NULL };
+ struct lustre_handle lh = { 0 };
+ __u32 old = la->la_nlink;
+ int idx;
+ int rc = 0;
+ __u8 flags;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(obj));
+ LASSERT(S_ISREG(lfsck_object_type(obj)));
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE, LCK_PW);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ la->la_valid = LA_NLINK;
+ rc = dt_declare_attr_set(env, obj, la, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
+ * ever tried to verify some remote MDT-object that resides on this
+ * MDT, but this MDT failed to respond such request. So means there
+ * may be some remote name entry on other MDT that references this
+ * object with another name, so we cannot know whether this linkEA
+ * is valid or not. So keep it there and maybe resolved when next
+ * LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(unlock, rc = 0);
+
+ fid_cpu_to_be(tfid, cfid);
+ idx = lfsck_sub_trace_file_fid2idx(cfid);
+ rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
+ (struct dt_rec *)&flags, (const struct dt_key *)tfid);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ if (flags & LNTF_SKIP_NLINK)
+ GOTO(unlock, rc = 0);
+
+ rc = dt_attr_get(env, obj, la);
+ if (rc != 0)
+ GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
+
+ rc = lfsck_links_read2(env, obj, &ldata);
+ if (rc != 0)
+ GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
+
+ if (la->la_nlink == ldata.ld_leh->leh_reccount ||
+ unlikely(la->la_nlink == 0))
+ GOTO(unlock, rc = 0);
+
+ la->la_nlink = ldata.ld_leh->leh_reccount;
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock, rc = 1);
+
+ rc = dt_attr_set(env, obj, la, th);
+
+ GOTO(unlock, rc = (rc == 0 ? 1 : rc));
+
+unlock:
+ dt_write_unlock(env, obj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ lfsck_ibits_unlock(&lh, LCK_PW);
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
+ "nlink count from %u to %u: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
+
+ if (rc != 0)
+ ns->ln_flags |= LF_INCONSISTENT;
+
+ return rc;
+}
+
+/**
+ * Double scan the directory object for namespace LFSCK.
+ *
+ * This function will verify the <parent, child> pairs in the namespace tree:
+ * the parent references the child via some name entry that should be in the
+ * child's linkEA entry, the child should back references the parent via its
+ * ".." name entry.
+ *
+ * The LFSCK will scan every linkEA entry in turn until find out the first
+ * matched pairs. If found, then all other linkEA entries will be dropped.
+ * If all the linkEA entries cannot match the ".." name entry, then there
+ * are serveral possible cases:
+ *
+ * 1) If there is only one linkEA entry, then trust it as long as the PFID
+ * in the linkEA entry is valid.
+ *
+ * 2) If there are multiple linkEA entries, then try to find the linkEA
+ * that matches the ".." name entry. If found, then all other entries
+ * are invalid; otherwise, it is quite possible that the ".." name entry
+ * is corrupted. Under such case, the LFSCK will rebuild the ".." name
+ * entry according to the first valid linkEA entry (both the parent and
+ * the name entry should exist).
+ *
+ * 3) If the directory object has no (valid) linkEA entry, then the
+ * directory object will be handled as pure orphan and inserted
+ * in the .lustre/lost+found/MDTxxxx/ with the name:
+ * ${self_FID}-${PFID}-D-${conflict_version}
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] child pointer to the directory object to be handled
+ * \param[in] flags to indicate the specical checking on the @child
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *child, __u8 flags)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct lu_fid *pfid = &info->lti_fid2;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lustre_handle lh = { 0 };
+ struct linkea_data ldata = { NULL };
+ bool unknown = false;
+ bool lpf = false;
+ bool retry = false;
+ enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(child));
+
+ if (flags & LNTF_UNCERTAIN_LMV) {
+ if (flags & LNTF_RECHECK_NAME_HASH) {
+ rc = lfsck_namespace_scan_shard(env, com, child);
+ if (rc < 0)
+ RETURN(rc);
+
+ ns->ln_striped_shards_scanned++;
+ } else {
+ ns->ln_striped_shards_skipped++;
+ }
+ }
+
+ flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
+ if (flags == 0)
+ RETURN(0);
+
+ if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
+ !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
+ CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
+ "the namespace LFSCK, then the LFSCK cannot guarantee"
+ "all the name entries have been verified in first-stage"
+ "scanning. So have to skip orphan related handling for"
+ "the directory object "DFID" with remote name entry\n",
+ lfsck_lfsck2name(lfsck), PFID(cfid));
+
+ RETURN(0);
+ }
+
+ if (unlikely(!dt_try_as_dir(env, child)))
+ GOTO(out, rc = -ENOTDIR);
+
+ /* We only take ldlm lock on the @child when required. When the
+ * logic comes here for the first time, it is always false. */
+ if (0) {
+
+lock:
+ rc = lfsck_ibits_lock(env, lfsck, child, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
+ dt_read_lock(env, child, 0);
+ if (unlikely(lfsck_is_dead_obj(child))) {
+ dt_read_unlock(env, child);
+
+ GOTO(out, rc = 0);
+ }
+
+ rc = dt_lookup(env, child, (struct dt_rec *)pfid,
+ (const struct dt_key *)dotdot);
+ if (rc != 0) {
+ if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
+ dt_read_unlock(env, child);
+
+ GOTO(out, rc);
+ }
+
+ if (!lustre_handle_is_used(&lh)) {
+ dt_read_unlock(env, child);
+ goto lock;
+ }
+
+ fid_zero(pfid);
+ } else if (lfsck->li_lpf_obj != NULL &&
+ lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
+ lpf = true;
+ } else if (unlikely(!fid_is_sane(pfid))) {
+ fid_zero(pfid);
+ }
+
+ rc = lfsck_links_read(env, child, &ldata);
+ dt_read_unlock(env, child);
+ if (rc != 0) {
+ if (rc != -ENODATA && rc != -EINVAL)
+ GOTO(out, rc);
+
+ if (!lustre_handle_is_used(&lh))
+ goto lock;
+
+ if (rc == -EINVAL && !fid_is_zero(pfid)) {
+ /* Remove the corrupted linkEA. */
+ rc = lfsck_namespace_links_remove(env, com, child);
+ if (rc == 0)
+ /* Here, because of the crashed linkEA, we
+ * cannot know whether there is some parent
+ * that references the child directory via
+ * some name entry or not. So keep it there,
+ * when the LFSCK run next time, if there is
+ * some parent that references this object,
+ * then the LFSCK can rebuild the linkEA;
+ * otherwise, this object will be handled
+ * as orphan as above. */
+ unknown = true;
+ } else {
+ /* 1. If we have neither ".." nor linkEA,
+ * then it is an orphan.
+ *
+ * 2. If we only have the ".." name entry,
+ * but no parent references this child
+ * directory, then handle it as orphan. */
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ type = LNIT_MUL_REF;
+
+ /* If the LFSCK is marked as LF_INCOMPLETE,
+ * then means some MDT has ever tried to
+ * verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond
+ * such request. So means there may be some
+ * remote name entry on other MDT that
+ * references this object with another name,
+ * so we cannot know whether this linkEA is
+ * valid or not. So keep it there and maybe
+ * resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(out, rc = 0);
+
+ snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com, child,
+ info->lti_tmpbuf, "D", NULL);
+ }
+
+ GOTO(out, rc);
+ }
+
+ linkea_first_entry(&ldata);
+ /* This is the most common case: the object has unique linkEA entry. */
+ if (ldata.ld_leh->leh_reccount == 1) {
+ rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
+ &lh, &type, &retry);
+ if (retry) {
+ LASSERT(!lustre_handle_is_used(&lh));
+
+ retry = false;
+ goto lock;
+ }
+
+ GOTO(out, rc);
+ }
+
+ if (!lustre_handle_is_used(&lh))
+ goto lock;
+
+ if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
+ rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
+ &type);
+
+ GOTO(out, rc);
+ }
+
+ /* When we come here, the cases usually like that:
+ * 1) The directory object has a corrupted linkEA entry. During the
+ * first-stage scanning, the LFSCK cannot know such corruption,
+ * then it appends the right linkEA entry according to the found
+ * name entry after the bad one.
+ *
+ * 2) The directory object has a right linkEA entry. During the
+ * first-stage scanning, the LFSCK finds some bad name entry,
+ * but the LFSCK cannot aware that at that time, then it adds
+ * the bad linkEA entry for further processing. */
+ rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
+ &lh, &type, lpf);
+
+ GOTO(out, rc);
+
+out:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ if (rc > 0) {
+ switch (type) {
+ case LNIT_BAD_LINKEA:
+ ns->ln_linkea_repaired++;
+ break;
+ case LNIT_UNMATCHED_PAIRS:
+ ns->ln_unmatched_pairs_repaired++;
+ break;
+ case LNIT_MUL_REF:
+ ns->ln_mul_ref_repaired++;
+ break;
+ default:
+ break;
+ }
+ }
+
+ if (unknown)
+ ns->ln_unknown_inconsistency++;
+
+ return rc;
+}
+
+/**
+ * Double scan the MDT-object for namespace LFSCK.
+ *
+ * If the MDT-object contains invalid or repeated linkEA entries, then drop
+ * those entries from the linkEA; if the linkEA becomes empty or the object
+ * has no linkEA, then it is an orphan and will be added into the directory
+ * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
+ * the remote parent; if the name entry corresponding to some linkEA entry
+ * is lost, then add the name entry back to the namespace.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] child pointer to the dt_object to be handled
+ * \param[in] flags some hints to indicate how the @child should be handled
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_double_scan_one(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *child, __u8 flags)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct lu_name *cname = &info->lti_name;
+ struct lu_fid *pfid = &info->lti_fid;
+ struct lu_fid *cfid = &info->lti_fid2;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct dt_object *parent = NULL;
+ struct linkea_data ldata = { NULL };
+ bool repaired = false;
+ int count = 0;
+ int rc;
+ ENTRY;
+
+ dt_read_lock(env, child, 0);
+ if (unlikely(lfsck_is_dead_obj(child))) {
+ dt_read_unlock(env, child);
+
+ RETURN(0);
+ }
+
+ if (S_ISDIR(lfsck_object_type(child))) {
+ dt_read_unlock(env, child);
+ rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
+
+ RETURN(rc);
+ }
+
+ rc = lfsck_links_read(env, child, &ldata);
+ dt_read_unlock(env, child);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL) {
+ lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
+ info->lti_key);
+ rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
+ false);
+ /* Found repeated linkEA entries */
+ if (rc > 0) {
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ &ldata, cname, pfid, false);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc == 0)
+ continue;
+
+ repaired = true;
+
+ /* fall through */
+ }
+
+ /* Invalid PFID in the linkEA entry. */
+ if (!fid_is_sane(pfid)) {
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ &ldata, cname, pfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+
+ continue;
+ }
+
+ parent = lfsck_object_find_bottom(env, lfsck, pfid);
+ if (IS_ERR(parent))
+ GOTO(out, rc = PTR_ERR(parent));
+
+ if (!dt_object_exists(parent)) {
+
+lost_parent:
+ if (ldata.ld_leh->leh_reccount > 1) {
+ /* If it is NOT the last linkEA entry, then
+ * there is still other chance to make the
+ * child to be visible via other parent, then
+ * remove this linkEA entry. */
+ rc = lfsck_namespace_shrink_linkea(env, com,
+ child, &ldata, cname, pfid, true);
+ } else {
+ /* If the LFSCK is marked as LF_INCOMPLETE,
+ * then means some MDT has ever tried to
+ * verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond
+ * such request. So means there may be some
+ * remote name entry on other MDT that
+ * references this object with another name,
+ * so we cannot know whether this linkEA is
+ * valid or not. So keep it there and maybe
+ * resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc = 0);
+ }
+
+ /* Create the lost parent as an orphan. */
+ rc = lfsck_namespace_create_orphan_dir(env, com,
+ parent, NULL);
+ if (rc < 0) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc);
+ }
+
+ if (rc > 0)
+ repaired = true;
+
+ /* Add the missing name entry to the parent. */
+ rc = lfsck_namespace_insert_normal(env, com,
+ parent, child, cname->ln_name);
+ if (unlikely(rc == -EEXIST))
+ /* Unfortunately, someone reused the
+ * name under the parent by race. So we
+ * have to remove the linkEA entry from
+ * current child object. It means that
+ * the LFSCK cannot recover the system
+ * totally back to its original status,
+ * but it is necessary to make the
+ * current system to be consistent. */
+ rc = lfsck_namespace_shrink_linkea(env,
+ com, child, &ldata,
+ cname, pfid, true);
+ else
+ linkea_next_entry(&ldata);
+ }
+
+ lfsck_object_put(env, parent);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+
+ continue;
+ }
+
+ /* The linkEA entry with bad parent will be removed. */
+ if (unlikely(!dt_try_as_dir(env, parent))) {
+ lfsck_object_put(env, parent);
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ &ldata, cname, pfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+
+ continue;
+ }
+
+ rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
+ (const struct dt_key *)cname->ln_name);
+ if (rc != 0 && rc != -ENOENT) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc);
+ }
+
+ if (rc == 0) {
+ if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
+ /* It is the most common case that we
+ * find the name entry corresponding
+ * to the linkEA entry. */
+ lfsck_object_put(env, parent);
+ linkea_next_entry(&ldata);
+ } else {
+ /* The name entry references another
+ * MDT-object that may be created by
+ * the LFSCK for repairing dangling
+ * name entry. Try to replace it. */
+ rc = lfsck_namespace_replace_cond(env, com,
+ parent, child, cfid, cname);
+ lfsck_object_put(env, parent);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0) {
+ repaired = true;
+ linkea_next_entry(&ldata);
+ } else {
+ rc = lfsck_namespace_shrink_linkea(env,
+ com, child, &ldata,
+ cname, pfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+ }
+ }
+
+ continue;
+ }
+
+ rc = dt_attr_get(env, child, la);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ /* If there is no name entry in the parent dir and the object
+ * link count is less than the linkea entries count, then the
+ * linkea entry should be removed. */
+ if (ldata.ld_leh->leh_reccount > la->la_nlink) {
+ rc = lfsck_namespace_shrink_linkea_cond(env, com,
+ parent, child, &ldata, cname, pfid);
+ lfsck_object_put(env, parent);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+
+ continue;
+ }
+
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some
+ * MDT has ever tried to verify some remote MDT-object that
+ * resides on this MDT, but this MDT failed to respond such
+ * request. So means there may be some remote name entry on
+ * other MDT that references this object with another name,
+ * so we cannot know whether this linkEA is valid or not.
+ * So keep it there and maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc = 0);
+ }
+
+ rc = lfsck_namespace_check_name(env, parent, child, cname);
+ if (rc == -ENOENT)
+ goto lost_parent;
+
+ if (rc < 0) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc);
+ }
+
+ /* It is an invalid name entry, drop it. */
+ if (unlikely(rc > 0)) {
+ lfsck_object_put(env, parent);
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ &ldata, cname, pfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+
+ continue;
+ }
+
+ /* Add the missing name entry back to the namespace. */
+ rc = lfsck_namespace_insert_normal(env, com, parent, child,
+ cname->ln_name);
+ if (unlikely(rc == -ESTALE))
+ /* It may happen when the remote object has been
+ * removed, but the local MDT is not aware of that. */
+ goto lost_parent;
+
+ if (unlikely(rc == -EEXIST))
+ /* Unfortunately, someone reused the name under the
+ * parent by race. So we have to remove the linkEA
+ * entry from current child object. It means that the
+ * LFSCK cannot recover the system totally back to
+ * its original status, but it is necessary to make
+ * the current system to be consistent.
+ *
+ * It also may be because of the LFSCK found some
+ * internal status of create operation. Under such
+ * case, nothing to be done. */
+ rc = lfsck_namespace_shrink_linkea_cond(env, com,
+ parent, child, &ldata, cname, pfid);
+ else
+ linkea_next_entry(&ldata);
+
+ lfsck_object_put(env, parent);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+ }
+
+ GOTO(out, rc = 0);
+
+out:
+ if (rc < 0 && rc != -ENODATA)
+ return rc;
+
+ if (rc == 0) {
+ LASSERT(ldata.ld_leh != NULL);
+
+ count = ldata.ld_leh->leh_reccount;
+ }
+
+ if (count == 0) {
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some
+ * MDT has ever tried to verify some remote MDT-object that
+ * resides on this MDT, but this MDT failed to respond such
+ * request. So means there may be some remote name entry on
+ * other MDT that references this object with another name,
+ * so we cannot know whether this linkEA is valid or not.
+ * So keep it there and maybe resolved when next LFSCK run. */
+ if (!(ns->ln_flags & LF_INCOMPLETE)) {
+ /* If the child becomes orphan, then insert it into
+ * the global .lustre/lost+found/MDTxxxx directory. */
+ rc = lfsck_namespace_insert_orphan(env, com, child,
+ "", "O", &count);
+ if (rc < 0)
+ return rc;
+
+ if (rc > 0) {
+ ns->ln_mul_ref_repaired++;
+ repaired = true;
+ }
+ }
+ } else {
+ rc = dt_attr_get(env, child, la);
+ if (rc != 0)
+ return rc;
+
+ if (la->la_nlink != 0 && la->la_nlink != count) {
+ rc = lfsck_namespace_repair_nlink(env, com, child, la);
+ if (rc > 0) {
+ ns->ln_objs_nlink_repaired++;
+ rc = 0;
+ }
+ }
+ }
+
+ if (repaired) {
+ if (la->la_nlink > 1)
+ ns->ln_mul_linked_repaired++;
+
+ if (rc == 0)
+ rc = 1;
+ }
+
+ return rc;
+}
+
+static void lfsck_namespace_dump_statistics(struct seq_file *m,
+ struct lfsck_namespace *ns,
+ __u64 checked_phase1,
+ __u64 checked_phase2,
+ __u32 time_phase1,
+ __u32 time_phase2)
+{
+ seq_printf(m, "checked_phase1: "LPU64"\n"
+ "checked_phase2: "LPU64"\n"
+ "updated_phase1: "LPU64"\n"
+ "updated_phase2: "LPU64"\n"
+ "failed_phase1: "LPU64"\n"
+ "failed_phase2: "LPU64"\n"
+ "directories: "LPU64"\n"
+ "dirent_repaired: "LPU64"\n"
+ "linkea_repaired: "LPU64"\n"
+ "nlinks_repaired: "LPU64"\n"
+ "multiple_linked_checked: "LPU64"\n"
+ "multiple_linked_repaired: "LPU64"\n"
+ "unknown_inconsistency: "LPU64"\n"
+ "unmatched_pairs_repaired: "LPU64"\n"
+ "dangling_repaired: "LPU64"\n"
+ "multiple_referenced_repaired: "LPU64"\n"
+ "bad_file_type_repaired: "LPU64"\n"
+ "lost_dirent_repaired: "LPU64"\n"
+ "local_lost_found_scanned: "LPU64"\n"
+ "local_lost_found_moved: "LPU64"\n"
+ "local_lost_found_skipped: "LPU64"\n"
+ "local_lost_found_failed: "LPU64"\n"
+ "striped_dirs_scanned: "LPU64"\n"
+ "striped_dirs_repaired: "LPU64"\n"
+ "striped_dirs_failed: "LPU64"\n"
+ "striped_dirs_disabled: "LPU64"\n"
+ "striped_dirs_skipped: "LPU64"\n"
+ "striped_shards_scanned: "LPU64"\n"
+ "striped_shards_repaired: "LPU64"\n"
+ "striped_shards_failed: "LPU64"\n"
+ "striped_shards_skipped: "LPU64"\n"
+ "name_hash_repaired: "LPU64"\n"
+ "success_count: %u\n"
+ "run_time_phase1: %u seconds\n"
+ "run_time_phase2: %u seconds\n",
+ checked_phase1,
+ checked_phase2,
+ ns->ln_items_repaired,
+ ns->ln_objs_repaired_phase2,
+ ns->ln_items_failed,
+ ns->ln_objs_failed_phase2,
+ ns->ln_dirs_checked,
+ ns->ln_dirent_repaired,
+ ns->ln_linkea_repaired,
+ ns->ln_objs_nlink_repaired,
+ ns->ln_mul_linked_checked,
+ ns->ln_mul_linked_repaired,
+ ns->ln_unknown_inconsistency,
+ ns->ln_unmatched_pairs_repaired,
+ ns->ln_dangling_repaired,
+ ns->ln_mul_ref_repaired,
+ ns->ln_bad_type_repaired,
+ ns->ln_lost_dirent_repaired,
+ ns->ln_local_lpf_scanned,
+ ns->ln_local_lpf_moved,
+ ns->ln_local_lpf_skipped,
+ ns->ln_local_lpf_failed,
+ ns->ln_striped_dirs_scanned,
+ ns->ln_striped_dirs_repaired,
+ ns->ln_striped_dirs_failed,
+ ns->ln_striped_dirs_disabled,
+ ns->ln_striped_dirs_skipped,
+ ns->ln_striped_shards_scanned,
+ ns->ln_striped_shards_repaired,
+ ns->ln_striped_shards_failed,
+ ns->ln_striped_shards_skipped,
+ ns->ln_name_hash_repaired,
+ ns->ln_success_count,
+ time_phase1,
+ time_phase2);
+}
+
+static void lfsck_namespace_release_lmv(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ while (!list_empty(&lfsck->li_list_lmv)) {
+ struct lfsck_lmv_unit *llu;
+ struct lfsck_lmv *llmv;
+
+ llu = list_entry(lfsck->li_list_lmv.next,
+ struct lfsck_lmv_unit, llu_link);
+ llmv = &llu->llu_lmv;
+
+ LASSERTF(atomic_read(&llmv->ll_ref) == 1,
+ "still in using: %u\n",
+ atomic_read(&llmv->ll_ref));
+
+ ns->ln_striped_dirs_skipped++;
+ lfsck_lmv_put(env, llmv);
+ }
+}
+
+static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj)
+{
+ struct lu_attr *la = &lfsck_env_info(env)->lti_la;
+ int rc;
+
+ rc = dt_attr_get(env, obj, la);
+ if (rc != 0)
+ return rc;
+
+ /* zero-linkEA object may be orphan, but it also maybe because
+ * of upgrading. Currently, we cannot record it for double scan.
+ * Because it may cause the LFSCK trace file to be too large. */
+
+ /* "la_ctime" == 1 means that it has ever been removed from
+ * backend /lost+found directory but not been added back to
+ * the normal namespace yet. */
+
+ if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
+ unlikely(la->la_ctime == 1))
+ rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
+ LNTF_CHECK_LINKEA, true);
+
+ return rc;
+}
+
+/* namespace APIs */
+
+static int lfsck_namespace_reset(const struct lu_env *env,
+ struct lfsck_component *com, bool init)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct dt_object *root;
+ struct dt_object *dto;
+ int rc;
+ ENTRY;
+
+ root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
+ if (IS_ERR(root))
+ GOTO(log, rc = PTR_ERR(root));
+
+ if (unlikely(!dt_try_as_dir(env, root)))
+ GOTO(put, rc = -ENOTDIR);
+
+ down_write(&com->lc_sem);
+ if (init) {
+ memset(ns, 0, sizeof(*ns));
+ } else {
+ __u32 count = ns->ln_success_count;
+ __u64 last_time = ns->ln_time_last_complete;
+
+ memset(ns, 0, sizeof(*ns));
+ ns->ln_success_count = count;
+ ns->ln_time_last_complete = last_time;
+ }
+ ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
+ ns->ln_status = LS_INIT;
+
+ lfsck_object_put(env, com->lc_obj);
+ com->lc_obj = NULL;
+ dto = lfsck_namespace_load_one_trace_file(env, com, root,
+ LFSCK_NAMESPACE, true);
+ if (IS_ERR(dto))
+ GOTO(out, rc = PTR_ERR(dto));
+
+ com->lc_obj = dto;
+ rc = lfsck_namespace_load_sub_trace_files(env, com, true);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(lad->lad_bitmap);
+
+ rc = lfsck_namespace_store(env, com, true);
+
+ GOTO(out, rc);
+
+out:
+ up_write(&com->lc_sem);
+
+put:
+ lfsck_object_put(env, root);
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+ return rc;
+}
+
+static void
+lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
+ bool new_checked)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ down_write(&com->lc_sem);
+ if (new_checked)
+ com->lc_new_checked++;
+ lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
+ up_write(&com->lc_sem);
+}
+
+static void lfsck_namespace_close_dir(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_assistant_object *lso = NULL;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_lmv *llmv = lfsck->li_lmv;
+ struct lfsck_namespace_req *lnr;
+ __u32 size =
+ sizeof(*lnr) + LFSCK_TMPBUF_LEN;
+ bool wakeup = false;
+ ENTRY;
+
+ if (llmv == NULL)
+ RETURN_EXIT;
+
+ OBD_ALLOC(lnr, size);
+ if (lnr == NULL) {
+ ns->ln_striped_dirs_skipped++;
+
+ RETURN_EXIT;
+ }
+
+ lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
+ NULL, lfsck->li_pos_current.lp_oit_cookie, true);
+ if (IS_ERR(lso)) {
+ OBD_FREE(lnr, size);
+ ns->ln_striped_dirs_skipped++;
+
+ RETURN_EXIT;
+ }
+
+ /* Generate a dummy request to indicate that all shards' name entry
+ * in this striped directory has been scanned for the first time. */
+ INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
+ lnr->lnr_lar.lar_parent = lso;
+ lnr->lnr_lmv = lfsck_lmv_get(llmv);
+ lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
+ lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
+ lnr->lnr_size = size;
+
+ spin_lock(&lad->lad_lock);
+ if (lad->lad_assistant_status < 0) {
+ spin_unlock(&lad->lad_lock);
+ lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
+ ns->ln_striped_dirs_skipped++;
+
+ RETURN_EXIT;
+ }
+
+ list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
+ if (lad->lad_prefetched == 0)
+ wakeup = true;
+
+ lad->lad_prefetched++;
+ spin_unlock(&lad->lad_lock);
+ if (wakeup)
+ wake_up_all(&lad->lad_thread.t_ctl_waitq);
+
+ EXIT;
+}
+
+static int lfsck_namespace_open_dir(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_lmv *llmv = lfsck->li_lmv;
+ int rc = 0;
+ ENTRY;
+
+ if (llmv == NULL)
+ RETURN(0);
+
+ if (llmv->ll_lmv_master) {
+ struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
+
+ if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
+ lmv->lmv_master_mdt_index =
+ lfsck_dev_idx(lfsck);
+ ns->ln_flags |= LF_INCONSISTENT;
+ llmv->ll_lmv_updated = 1;
+ }
+ } else {
+ rc = lfsck_namespace_verify_stripe_slave(env, com,
+ lfsck->li_obj_dir, llmv);
+ }
+
+ RETURN(rc > 0 ? 0 : rc);
+}
+
+static int lfsck_namespace_checkpoint(const struct lu_env *env,
+ struct lfsck_component *com, bool init)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ int rc;
+
+ if (!init) {
+ rc = lfsck_checkpoint_generic(env, com);
+ if (rc != 0)
+ goto log;
+ }
+
+ down_write(&com->lc_sem);
+ if (init) {
+ ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
+ } else {
+ ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
+ ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
+ HALF_SEC - lfsck->li_time_last_checkpoint);
+ ns->ln_time_last_checkpoint = cfs_time_current_sec();
+ ns->ln_items_checked += com->lc_new_checked;
+ com->lc_new_checked = 0;
+ }
+
+ rc = lfsck_namespace_store(env, com, false);
+ up_write(&com->lc_sem);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
+ ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
+ lfsck->li_pos_current.lp_oit_cookie,
+ PFID(&lfsck->li_pos_current.lp_dir_parent),
+ lfsck->li_pos_current.lp_dir_cookie, rc);
+
+ return rc > 0 ? 0 : rc;
+}
+
+static int lfsck_namespace_prep(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_start_param *lsp)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_position *pos = &com->lc_pos_start;
+ int rc;
+
+ rc = lfsck_namespace_load_bitmap(env, com);
+ if (rc != 0 || ns->ln_status == LS_COMPLETED) {
+ rc = lfsck_namespace_reset(env, com, false);
+ if (rc == 0)
+ rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
+
+ if (rc != 0) {
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
+ "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
+
+ return rc;
+ }
+ }
+
+ down_write(&com->lc_sem);
+ ns->ln_time_latest_start = cfs_time_current_sec();
+ spin_lock(&lfsck->li_lock);
+
+ if (ns->ln_flags & LF_SCANNED_ONCE) {
+ if (!lfsck->li_drop_dryrun ||
+ lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
+ ns->ln_status = LS_SCANNING_PHASE2;
+ list_move_tail(&com->lc_link,
+ &lfsck->li_list_double_scan);
+ if (!list_empty(&com->lc_link_dir))
+ list_del_init(&com->lc_link_dir);
+ lfsck_pos_set_zero(pos);
+ } else {
+ ns->ln_status = LS_SCANNING_PHASE1;
+ ns->ln_run_time_phase1 = 0;
+ ns->ln_run_time_phase2 = 0;
+ ns->ln_items_checked = 0;
+ ns->ln_items_repaired = 0;
+ ns->ln_items_failed = 0;
+ ns->ln_dirs_checked = 0;
+ ns->ln_objs_checked_phase2 = 0;
+ ns->ln_objs_repaired_phase2 = 0;
+ ns->ln_objs_failed_phase2 = 0;
+ ns->ln_objs_nlink_repaired = 0;
+ ns->ln_dirent_repaired = 0;
+ ns->ln_linkea_repaired = 0;
+ ns->ln_mul_linked_checked = 0;
+ ns->ln_mul_linked_repaired = 0;
+ ns->ln_unknown_inconsistency = 0;
+ ns->ln_unmatched_pairs_repaired = 0;
+ ns->ln_dangling_repaired = 0;
+ ns->ln_mul_ref_repaired = 0;
+ ns->ln_bad_type_repaired = 0;
+ ns->ln_lost_dirent_repaired = 0;
+ ns->ln_striped_dirs_scanned = 0;
+ ns->ln_striped_dirs_repaired = 0;
+ ns->ln_striped_dirs_failed = 0;
+ ns->ln_striped_dirs_disabled = 0;
+ ns->ln_striped_dirs_skipped = 0;
+ ns->ln_striped_shards_scanned = 0;
+ ns->ln_striped_shards_repaired = 0;
+ ns->ln_striped_shards_failed = 0;
+ ns->ln_striped_shards_skipped = 0;
+ ns->ln_name_hash_repaired = 0;
+ fid_zero(&ns->ln_fid_latest_scanned_phase2);
+ if (list_empty(&com->lc_link_dir))
+ list_add_tail(&com->lc_link_dir,
+ &lfsck->li_list_dir);
+ *pos = ns->ln_pos_first_inconsistent;
+ }
+ } else {
+ ns->ln_status = LS_SCANNING_PHASE1;
+ if (list_empty(&com->lc_link_dir))
+ list_add_tail(&com->lc_link_dir,
+ &lfsck->li_list_dir);
+ if (!lfsck->li_drop_dryrun ||
+ lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
+ *pos = ns->ln_pos_last_checkpoint;
+ pos->lp_oit_cookie++;
+ } else {
+ *pos = ns->ln_pos_first_inconsistent;
+ }
+ }
+
+ spin_unlock(&lfsck->li_lock);
+ up_write(&com->lc_sem);
+
+ rc = lfsck_start_assistant(env, com, lsp);
+
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
+ DFID", "LPX64"]: rc = %d\n",
+ lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
+ PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
+
+ return rc;
+}
+
+static int lfsck_namespace_exec_oit(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ const struct lu_fid *fid = lfsck_dto2fid(obj);
+ struct lu_fid *pfid = &info->lti_fid2;
+ struct lu_name *cname = &info->lti_name;
+ struct lu_seq_range *range = &info->lti_range;
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
+ struct linkea_data ldata = { NULL };
+ __u32 idx = lfsck_dev_idx(lfsck);
+ int rc;
+ ENTRY;
+
+ rc = lfsck_links_read(env, obj, &ldata);
+ if (rc == -ENOENT)
+ GOTO(out, rc = 0);
+
+ /* -EINVAL means crashed linkEA, should be verified. */
+ if (rc == -EINVAL) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_LINKEA, true);
+ if (rc == 0) {
+ struct lustre_handle lh = { 0 };
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc == 0) {
+ rc = lfsck_namespace_links_remove(env, com,
+ obj);
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ }
+ }
+
+ GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
+ }
+
+ if (rc == -ENODATA) {
+ rc = lfsck_namespace_check_for_double_scan(env, com, obj);
+
+ GOTO(out, rc);
+ }
+
+ if (rc != 0)
+ GOTO(out, rc);
+
+ /* Record multiple-linked object. */
+ if (ldata.ld_leh->leh_reccount > 1) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_LINKEA, true);
+
+ GOTO(out, rc);
+ }
+
+ linkea_first_entry(&ldata);
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
+ if (!fid_is_sane(pfid)) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_PARENT, true);
+ } else {
+ fld_range_set_mdt(range);
+ rc = fld_local_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if ((rc == -ENOENT) ||
+ (rc == 0 && range->lsr_index != idx))
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_LINKEA, true);
+ else
+ rc = lfsck_namespace_check_for_double_scan(env, com,
+ obj);
+ }
+
+ GOTO(out, rc);
+
+out:
+ down_write(&com->lc_sem);
+ if (S_ISDIR(lfsck_object_type(obj)))
+ ns->ln_dirs_checked++;
+ if (rc != 0)
+ lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
+ up_write(&com->lc_sem);
+
+ return rc;
+}
+
+static int lfsck_namespace_exec_dir(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_assistant_object *lso,
+ struct lu_dirent *ent, __u16 type)
+{
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace_req *lnr;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct ptlrpc_thread *mthread = &lfsck->li_thread;
+ struct ptlrpc_thread *athread = &lad->lad_thread;
+ struct l_wait_info lwi = { 0 };
+ bool wakeup = false;
+
+ l_wait_event(mthread->t_ctl_waitq,
+ lad->lad_prefetched < bk->lb_async_windows ||
+ !thread_is_running(mthread) ||
+ thread_is_stopped(athread),
+ &lwi);
+
+ if (unlikely(!thread_is_running(mthread)) ||
+ thread_is_stopped(athread))
+ return 0;
+
+ if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
+ return 0;
+
+ lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
+ if (IS_ERR(lnr)) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
+ return PTR_ERR(lnr);
+ }
+
+ spin_lock(&lad->lad_lock);
+ if (lad->lad_assistant_status < 0) {
+ spin_unlock(&lad->lad_lock);
+ lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
+ return lad->lad_assistant_status;
+ }
+
+ list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
+ if (lad->lad_prefetched == 0)
+ wakeup = true;
+
+ lad->lad_prefetched++;
+ spin_unlock(&lad->lad_lock);
+ if (wakeup)
+ wake_up_all(&lad->lad_thread.t_ctl_waitq);
+
+ down_write(&com->lc_sem);
+ com->lc_new_checked++;
+ up_write(&com->lc_sem);
+
+ return 0;
+}
+
+static int lfsck_namespace_post(const struct lu_env *env,
+ struct lfsck_component *com,
+ int result, bool init)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ int rc;
+ ENTRY;
+
+ lfsck_post_generic(env, com, &result);
+
+ down_write(&com->lc_sem);
+ lfsck_namespace_release_lmv(env, com);
+
+ spin_lock(&lfsck->li_lock);
+ if (!init)
+ ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
+ if (result > 0) {
+ ns->ln_status = LS_SCANNING_PHASE2;
+ ns->ln_flags |= LF_SCANNED_ONCE;
+ ns->ln_flags &= ~LF_UPGRADE;
+ list_del_init(&com->lc_link_dir);
+ list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
+ } else if (result == 0) {
+ if (lfsck->li_status != 0)
+ ns->ln_status = lfsck->li_status;
+ else
+ ns->ln_status = LS_STOPPED;
+ if (ns->ln_status != LS_PAUSED) {
+ list_del_init(&com->lc_link_dir);
+ list_move_tail(&com->lc_link, &lfsck->li_list_idle);
+ }
+ } else {
+ ns->ln_status = LS_FAILED;
+ list_del_init(&com->lc_link_dir);
+ list_move_tail(&com->lc_link, &lfsck->li_list_idle);
+ }
+ spin_unlock(&lfsck->li_lock);
+
+ if (!init) {
+ ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
+ HALF_SEC - lfsck->li_time_last_checkpoint);
+ ns->ln_time_last_checkpoint = cfs_time_current_sec();
+ ns->ln_items_checked += com->lc_new_checked;
+ com->lc_new_checked = 0;
+ }
+
+ rc = lfsck_namespace_store(env, com, false);
+ up_write(&com->lc_sem);
+
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+
+ RETURN(rc);
+}
+
+static int
+lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
+ struct seq_file *m)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ int rc;
+
+ down_read(&com->lc_sem);
+ seq_printf(m, "name: lfsck_namespace\n"
+ "magic: %#x\n"
+ "version: %d\n"
+ "status: %s\n",
+ ns->ln_magic,
+ bk->lb_version,
+ lfsck_status2names(ns->ln_status));
+
+ rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_time_dump(m, ns->ln_time_last_complete,
+ "last_completed");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_time_dump(m, ns->ln_time_latest_start,
+ "latest_start");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
+ "last_checkpoint");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
+ "latest_start_position");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
+ "last_checkpoint_position");
+ if (rc < 0)
+ goto out;
+
+ rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
+ "first_failure_position");
+ if (rc < 0)
+ goto out;
+
+ if (ns->ln_status == LS_SCANNING_PHASE1) {
+ struct lfsck_position pos;
+ const struct dt_it_ops *iops;
+ cfs_duration_t duration = cfs_time_current() -
+ lfsck->li_time_last_checkpoint;
+ __u64 checked = ns->ln_items_checked + com->lc_new_checked;
+ __u64 speed = checked;
+ __u64 new_checked = com->lc_new_checked *
+ msecs_to_jiffies(MSEC_PER_SEC);
+ __u32 rtime = ns->ln_run_time_phase1 +
+ cfs_duration_sec(duration + HALF_SEC);
+
+ if (duration != 0)
+ do_div(new_checked, duration);
+
+ if (rtime != 0)
+ do_div(speed, rtime);
+
+ lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
+ seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
+ "average_speed_phase2: N/A\n"
+ "average_speed_total: "LPU64" items/sec\n"
+ "real_time_speed_phase1: "LPU64" items/sec\n"
+ "real_time_speed_phase2: N/A\n",
+ speed,
+ speed,
+ new_checked);
+
+ LASSERT(lfsck->li_di_oit != NULL);
+
+ iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
+
+ /* The low layer otable-based iteration position may NOT
+ * exactly match the namespace-based directory traversal
+ * cookie. Generally, it is not a serious issue. But the
+ * caller should NOT make assumption on that. */
+ pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
+ if (!lfsck->li_current_oit_processed)
+ pos.lp_oit_cookie--;
+
+ spin_lock(&lfsck->li_lock);
+ if (lfsck->li_di_dir != NULL) {
+ pos.lp_dir_cookie = lfsck->li_cookie_dir;
+ if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
+ fid_zero(&pos.lp_dir_parent);
+ pos.lp_dir_cookie = 0;
+ } else {
+ pos.lp_dir_parent =
+ *lfsck_dto2fid(lfsck->li_obj_dir);
+ }
+ } else {
+ fid_zero(&pos.lp_dir_parent);
+ pos.lp_dir_cookie = 0;
+ }
+ spin_unlock(&lfsck->li_lock);
+ lfsck_pos_dump(m, &pos, "current_position");
+ } else if (ns->ln_status == LS_SCANNING_PHASE2) {
+ cfs_duration_t duration = cfs_time_current() -
+ com->lc_time_last_checkpoint;
+ __u64 checked = ns->ln_objs_checked_phase2 +
+ com->lc_new_checked;
+ __u64 speed1 = ns->ln_items_checked;
+ __u64 speed2 = checked;
+ __u64 speed0 = speed1 + speed2;
+ __u64 new_checked = com->lc_new_checked *
+ msecs_to_jiffies(MSEC_PER_SEC);
+ __u32 rtime = ns->ln_run_time_phase2 +
+ cfs_duration_sec(duration + HALF_SEC);
+ __u32 time0 = ns->ln_run_time_phase1 + rtime;
+
+ if (duration != 0)
+ do_div(new_checked, duration);
+
+ if (ns->ln_run_time_phase1 != 0)
+ do_div(speed1, ns->ln_run_time_phase1);
+ else if (ns->ln_items_checked != 0)
+ time0++;
+
+ if (rtime != 0)
+ do_div(speed2, rtime);
+ else if (checked != 0)
+ time0++;
+
+ if (time0 != 0)
+ do_div(speed0, time0);
+
+ lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
+ checked,
+ ns->ln_run_time_phase1, rtime);
+ seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
+ "average_speed_phase2: "LPU64" objs/sec\n"
+ "average_speed_total: "LPU64" items/sec\n"
+ "real_time_speed_phase1: N/A\n"
+ "real_time_speed_phase2: "LPU64" objs/sec\n"
+ "current_position: "DFID"\n",
+ speed1,
+ speed2,
+ speed0,
+ new_checked,
+ PFID(&ns->ln_fid_latest_scanned_phase2));
+ } else {
+ __u64 speed1 = ns->ln_items_checked;
+ __u64 speed2 = ns->ln_objs_checked_phase2;
+ __u64 speed0 = speed1 + speed2;
+ __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
+
+ if (ns->ln_run_time_phase1 != 0)
+ do_div(speed1, ns->ln_run_time_phase1);
+ else if (ns->ln_items_checked != 0)
+ time0++;
+
+ if (ns->ln_run_time_phase2 != 0)
+ do_div(speed2, ns->ln_run_time_phase2);
+ else if (ns->ln_objs_checked_phase2 != 0)
+ time0++;
+
+ if (time0 != 0)
+ do_div(speed0, time0);
+
+ lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
+ ns->ln_objs_checked_phase2,
+ ns->ln_run_time_phase1,
+ ns->ln_run_time_phase2);
+ seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
+ "average_speed_phase2: "LPU64" objs/sec\n"
+ "average_speed_total: "LPU64" items/sec\n"
+ "real_time_speed_phase1: N/A\n"
+ "real_time_speed_phase2: N/A\n"
+ "current_position: N/A\n",
+ speed1,
+ speed2,
+ speed0);
+ }
+out:
+ up_read(&com->lc_sem);
+ return 0;
+}
+
+static int lfsck_namespace_double_scan(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct lfsck_tgt_desc *next;
+ int rc;
+
+ rc = lfsck_double_scan_generic(env, com, ns->ln_status);
+ if (thread_is_stopped(&lad->lad_thread)) {
+ LASSERT(list_empty(&lad->lad_req_list));
+ LASSERT(list_empty(&lad->lad_mdt_phase1_list));
+
+ spin_lock(<ds->ltd_lock);
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ spin_unlock(<ds->ltd_lock);
+ }
+
+ return rc;
+}
+
+static void lfsck_namespace_data_release(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct lfsck_tgt_desc *next;
+
+ LASSERT(lad != NULL);
+ LASSERT(thread_is_init(&lad->lad_thread) ||
+ thread_is_stopped(&lad->lad_thread));
+ LASSERT(list_empty(&lad->lad_req_list));
+
+ com->lc_data = NULL;
+ lfsck_namespace_release_lmv(env, com);
+
+ spin_lock(<ds->ltd_lock);
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
+ ltd_namespace_list) {
+ list_del_init(<d->ltd_namespace_list);
+ }
+ spin_unlock(<ds->ltd_lock);
+
+ if (likely(lad->lad_bitmap != NULL))
+ CFS_FREE_BITMAP(lad->lad_bitmap);
+
+ OBD_FREE_PTR(lad);
+}
+
+static void lfsck_namespace_quit(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct lfsck_tgt_desc *next;
+
+ LASSERT(lad != NULL);
+
+ lfsck_quit_generic(env, com);
+
+ LASSERT(thread_is_init(&lad->lad_thread) ||
+ thread_is_stopped(&lad->lad_thread));
+ LASSERT(list_empty(&lad->lad_req_list));
+
+ lfsck_namespace_release_lmv(env, com);
+
+ spin_lock(<ds->ltd_lock);
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ spin_unlock(<ds->ltd_lock);
+}
+
+static int lfsck_namespace_in_notify(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_request *lr,
+ struct thandle *th)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ int rc = 0;
+ bool fail = false;
+ ENTRY;
+
+ switch (lr->lr_event) {
+ case LE_SKIP_NLINK_DECLARE: {
+ struct dt_object *obj;
+ struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
+ int idx;
+ __u8 flags = 0;
+
+ LASSERT(th != NULL);
+
+ idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
+ obj = com->lc_sub_trace_objs[idx].lsto_obj;
+ fid_cpu_to_be(key, &lr->lr_fid);
+ mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
+ rc = dt_declare_delete(env, obj,
+ (const struct dt_key *)key, th);
+ if (rc == 0)
+ rc = dt_declare_insert(env, obj,
+ (const struct dt_rec *)&flags,
+ (const struct dt_key *)key, th);
+ mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
+
+ RETURN(rc);
+ }
+ case LE_SKIP_NLINK: {
+ struct dt_object *obj;
+ struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
+ int idx;
+ __u8 flags = 0;
+ bool exist = false;
+ ENTRY;
+
+ LASSERT(th != NULL);
+
+ idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
+ obj = com->lc_sub_trace_objs[idx].lsto_obj;
+ fid_cpu_to_be(key, &lr->lr_fid);
+ mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
+ rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
+ (const struct dt_key *)key);
+ if (rc == 0) {
+ if (flags & LNTF_SKIP_NLINK) {
+ mutex_unlock(
+ &com->lc_sub_trace_objs[idx].lsto_mutex);
+
+ RETURN(0);
+ }
+
+ exist = true;
+ } else if (rc != -ENOENT) {
+ GOTO(log, rc);
+ }
+
+ flags |= LNTF_SKIP_NLINK;
+ if (exist) {
+ rc = dt_delete(env, obj, (const struct dt_key *)key,
+ th);
+ if (rc != 0)
+ GOTO(log, rc);
+ }
+
+ rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
+ (const struct dt_key *)key, th, 1);
+
+ GOTO(log, rc);
+
+log:
+ mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
+ CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
+ " to be skipped for namespace double scan: rc = %d\n",
+ lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
+
+ if (rc != 0)
+ /* If we cannot record this object in the LFSCK tracing,
+ * we have to mark the LFSC as LF_INCOMPLETE, then the
+ * LFSCK will skip nlink attribute verification for
+ * all objects. */
+ ns->ln_flags |= LF_INCOMPLETE;
+
+ return 0;
+ }
+ case LE_SET_LMV_MASTER: {
+ struct dt_object *obj;
+
+ obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
+ if (IS_ERR(obj))
+ RETURN(PTR_ERR(obj));
+
+ if (likely(dt_object_exists(obj)))
+ rc = lfsck_namespace_notify_lmv_master_local(env, com,
+ obj);
+
+ lfsck_object_put(env, obj);
+
+ RETURN(rc > 0 ? 0 : rc);
+ }
+ case LE_SET_LMV_SLAVE: {
+ if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
+ ns->ln_striped_shards_repaired++;
+
+ rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
+ LNTF_RECHECK_NAME_HASH, true);
+
+ RETURN(rc > 0 ? 0 : rc);
+ }
+ case LE_PHASE1_DONE:
+ case LE_PHASE2_DONE:
+ case LE_PEER_EXIT:
+ break;
+ default:
+ RETURN(-EINVAL);
+ }
+
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
+ "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
+ lr->lr_index, lr->lr_status, lr->lr_flags2);
+
+ spin_lock(<ds->ltd_lock);
+ ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
+ if (ltd == NULL) {
+ spin_unlock(<ds->ltd_lock);
+
+ RETURN(-ENXIO);
+ }
+
+ list_del_init(<d->ltd_namespace_phase_list);
+ switch (lr->lr_event) {
+ case LE_PHASE1_DONE:
+ if (lr->lr_status <= 0) {
+ ltd->ltd_namespace_done = 1;
+ list_del_init(<d->ltd_namespace_list);
+ CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
+ "phase1 for namespace LFSCK: rc = %d.\n",
+ lfsck_lfsck2name(lfsck),
+ ltd->ltd_index, lr->lr_status);
+ ns->ln_flags |= LF_INCOMPLETE;
+ fail = true;
+ break;
+ }
+
+ if (lr->lr_flags2 & LF_INCOMPLETE)
+ ns->ln_flags |= LF_INCOMPLETE;
+
+ if (list_empty(<d->ltd_namespace_list))
+ list_add_tail(<d->ltd_namespace_list,
+ &lad->lad_mdt_list);
+ list_add_tail(<d->ltd_namespace_phase_list,
+ &lad->lad_mdt_phase2_list);
+ break;
+ case LE_PHASE2_DONE:
+ ltd->ltd_namespace_done = 1;
+ list_del_init(<d->ltd_namespace_list);
+ break;
+ case LE_PEER_EXIT:
+ fail = true;
+ ltd->ltd_namespace_done = 1;
+ list_del_init(<d->ltd_namespace_list);
+ if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
+ CDEBUG(D_LFSCK,
+ "%s: the peer MDT %x exit namespace LFSCK\n",
+ lfsck_lfsck2name(lfsck), ltd->ltd_index);
+ ns->ln_flags |= LF_INCOMPLETE;
+ }
+ break;
+ default:
+ break;
+ }
+ spin_unlock(<ds->ltd_lock);
+
+ if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
+ struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
+
+ memset(stop, 0, sizeof(*stop));
+ stop->ls_status = lr->lr_status;
+ stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
+ lfsck_stop(env, lfsck->li_bottom, stop);
+ } else if (lfsck_phase2_next_ready(lad)) {
+ wake_up_all(&lad->lad_thread.t_ctl_waitq);
+ }
+
+ RETURN(0);
+}
+
+static int lfsck_namespace_query(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ return ns->ln_status;
+}
+
+static struct lfsck_operations lfsck_namespace_ops = {
+ .lfsck_reset = lfsck_namespace_reset,
+ .lfsck_fail = lfsck_namespace_fail,
+ .lfsck_close_dir = lfsck_namespace_close_dir,
+ .lfsck_open_dir = lfsck_namespace_open_dir,
+ .lfsck_checkpoint = lfsck_namespace_checkpoint,
+ .lfsck_prep = lfsck_namespace_prep,
+ .lfsck_exec_oit = lfsck_namespace_exec_oit,
+ .lfsck_exec_dir = lfsck_namespace_exec_dir,
+ .lfsck_post = lfsck_namespace_post,
+ .lfsck_dump = lfsck_namespace_dump,
+ .lfsck_double_scan = lfsck_namespace_double_scan,
+ .lfsck_data_release = lfsck_namespace_data_release,
+ .lfsck_quit = lfsck_namespace_quit,
+ .lfsck_in_notify = lfsck_namespace_in_notify,
+ .lfsck_query = lfsck_namespace_query,
+};
+
+/**
+ * Repair dangling name entry.
+ *
+ * For the name entry with dangling reference, we need to repare the
+ * inconsistency according to the LFSCK sponsor's requirement:
+ *
+ * 1) Keep the inconsistency there and report the inconsistency case,
+ * then give the chance to the application to find related issues,
+ * and the users can make the decision about how to handle it with
+ * more human knownledge. (by default)
+ *
+ * 2) Re-create the missing MDT-object with the FID information.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] parent pointer to the dir object that contains the dangling
+ * name entry
+ * \param[in] child pointer to the object corresponding to the dangling
+ * name entry
+ * \param[in] lnr pointer to the namespace request that contains the
+ * name's name, parent object, parent's LMV, and ect.
+ *
+ * \retval positive number if no need to repair
+ * \retval zero for repaired successfully
+ * \retval negative error number on failure
+ */
+int lfsck_namespace_repair_dangling(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *parent,
+ struct dt_object *child,
+ struct lfsck_namespace_req *lnr)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_allocation_hint *hint = &info->lti_hint;
+ struct dt_object_format *dof = &info->lti_dof;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
+ struct dt_object *pobj = NULL;
+ struct dt_object *cobj = NULL;
+ const struct lu_name *cname;
+ struct linkea_data ldata = { NULL };
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ struct lu_buf linkea_buf;
+ struct lu_buf lmv_buf;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ int rc = 0;
+ __u16 type = lnr->lnr_type;
+ bool create;
+ ENTRY;
+
+ cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
+ if (bk->lb_param & LPF_CREATE_MDTOBJ)
+ create = true;
+ else
+ create = false;
+
+ if (!create || bk->lb_param & LPF_DRYRUN)
+ GOTO(log, rc = 0);
+
+ /* We may need to create the sub-objects of the @child via LOD,
+ * so make the modification based on lfsck->li_next. */
+
+ pobj = lfsck_object_locate(dev, parent);
+ if (IS_ERR(pobj))
+ GOTO(log, rc = PTR_ERR(pobj));
+
+ if (unlikely(!dt_try_as_dir(env, pobj)))
+ GOTO(log, rc = -ENOTDIR);
+
+ cobj = lfsck_object_locate(dev, child);
+ if (IS_ERR(cobj))
+ GOTO(log, rc = PTR_ERR(cobj));
+
+ rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(pobj));
+ if (rc != 0)
+ GOTO(log, rc);
+
+ rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
+ MDS_INODELOCK_UPDATE, LCK_PR);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ rc = lfsck_namespace_check_exist(env, pobj, cobj, lnr->lnr_name);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ /* Set the ctime as zero, then others can know it is created for
+ * repairing dangling name entry by LFSCK. And if the LFSCK made
+ * wrong decision and the real MDT-object has been found later,
+ * then the LFSCK has chance to fix the incosistency properly. */
+ memset(la, 0, sizeof(*la));
+ la->la_mode = (type & S_IFMT) | 0600;
+ la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
+ LA_ATIME | LA_MTIME | LA_CTIME;
+
+ cobj->do_ops->do_ah_init(env, hint, pobj, cobj,
+ la->la_mode & S_IFMT);
+
+ memset(dof, 0, sizeof(*dof));
+ dof->dof_type = dt_mode_to_dft(type);
+ /* If the target is a regular file, then the LFSCK will only create
+ * the MDT-object without stripes (dof->dof_reg.striped = 0). related
+ * OST-objects will be created when write open. */
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(log, rc = PTR_ERR(th));
+
+ /* 1a. create child. */
+ rc = dt_declare_create(env, cobj, la, hint, dof, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (S_ISDIR(type)) {
+ if (unlikely(!dt_try_as_dir(env, cobj)))
+ GOTO(stop, rc = -ENOTDIR);
+
+ /* 2a. insert dot into child dir */
+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = lfsck_dto2fid(cobj);
+ rc = dt_declare_insert(env, cobj,
+ (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ /* 3a. insert dotdot into child dir */
+ rec->rec_fid = lfsck_dto2fid(pobj);
+ rc = dt_declare_insert(env, cobj,
+ (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ /* 4a. increase child nlink */
+ rc = dt_declare_ref_add(env, cobj, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ /* 5a. generate slave LMV EA. */
+ if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
+ int idx;
+
+ idx = lfsck_shard_name_to_index(env,
+ lnr->lnr_name, lnr->lnr_namelen,
+ type, lfsck_dto2fid(cobj));
+ if (unlikely(idx < 0))
+ GOTO(stop, rc = idx);
+
+ *lmv2 = lnr->lnr_lmv->ll_lmv;
+ lmv2->lmv_magic = LMV_MAGIC_STRIPE;
+ lmv2->lmv_master_mdt_index = idx;
+
+ lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
+ lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
+ rc = dt_declare_xattr_set(env, cobj, &lmv_buf,
+ XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+ }
+
+ /* 6a. insert linkEA for child */
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
+ rc = dt_declare_xattr_set(env, cobj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
+
+ dt_write_lock(env, cobj, 0);
+ /* 1b. create child */
+ rc = dt_create(env, cobj, la, hint, dof, th);
+ if (rc != 0)
+ GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
+
+ if (S_ISDIR(type)) {
+ /* 2b. insert dot into child dir */
+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = lfsck_dto2fid(cobj);
+ rc = dt_insert(env, cobj, (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th, 1);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ /* 3b. insert dotdot into child dir */
+ rec->rec_fid = lfsck_dto2fid(pobj);
+ rc = dt_insert(env, cobj, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th, 1);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ /* 4b. increase child nlink */
+ rc = dt_ref_add(env, cobj, th);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ /* 5b. generate slave LMV EA. */
+ if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
+ rc = dt_xattr_set(env, cobj, &lmv_buf, XATTR_NAME_LMV,
+ 0, th);
+ if (rc != 0)
+ GOTO(unlock, rc);
+ }
+ }
+
+ /* 6b. insert linkEA for child. */
+ rc = dt_xattr_set(env, cobj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+
+ GOTO(unlock, rc);
+
+unlock:
+ dt_write_unlock(env, cobj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+log:
+ lfsck_unlock(llh);
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
+ "reference for: parent "DFID", child "DFID", type %u, "
+ "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
+ PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
+ type, cname->ln_name,
+ create ? "Create the lost MDT-object as required" :
+ "Keep the MDT-object there by default", rc);
+
+ if (rc <= 0) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ ns->ln_flags |= LF_INCONSISTENT;
+ }
+
+ return rc;
+}
+
+static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_assistant_req *lar)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct linkea_data ldata = { NULL };
+ const struct lu_name *cname;
+ struct thandle *handle = NULL;
+ struct lfsck_namespace_req *lnr =
+ container_of0(lar, struct lfsck_namespace_req, lnr_lar);
+ struct dt_object *dir = NULL;
+ struct dt_object *obj = NULL;
+ struct lfsck_assistant_object *lso = lar->lar_parent;
+ const struct lu_fid *pfid = &lso->lso_fid;
+ struct dt_device *dev = NULL;
+ struct lustre_handle lh = { 0 };
+ bool repaired = false;
+ bool dtlocked = false;
+ bool remove;
+ bool newdata;
+ bool log = false;
+ bool bad_hash = false;
+ bool bad_linkea = false;
+ int idx = 0;
+ int count = 0;
+ int rc = 0;
+ enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
+ ENTRY;
+
+ if (lso->lso_dead)
+ RETURN(0);
+
+ la->la_nlink = 0;
+ if (lnr->lnr_attr & LUDA_UPGRADE) {
+ ns->ln_flags |= LF_UPGRADE;
+ ns->ln_dirent_repaired++;
+ repaired = true;
+ } else if (lnr->lnr_attr & LUDA_REPAIR) {