struct lfsck_layout_slave_target {
/* link into lfsck_layout_slave_data::llsd_master_list. */
struct list_head llst_list;
+ /* The position for next record in the rbtree for iteration. */
+ struct lu_fid llst_fid;
+ /* Dummy hash for iteration against the rbtree. */
+ __u64 llst_hash;
__u64 llst_gen;
atomic_t llst_ref;
__u32 llst_index;
static inline struct lfsck_layout_slave_target *
lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
- __u32 index)
+ __u32 index, bool unlink)
{
struct lfsck_layout_slave_target *llst;
spin_lock(&llsd->llsd_lock);
list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
if (llst->llst_index == index) {
- list_del_init(&llst->llst_list);
+ if (unlink)
+ list_del_init(&llst->llst_list);
+ else
+ atomic_inc(&llst->llst_ref);
spin_unlock(&llsd->llsd_lock);
return llst;
return 0;
}
-/* The caller should hold lock. */
+/* The caller should hold llsd->llsd_rb_lock. */
static struct lfsck_rbtree_node *
lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
- const struct lu_fid *fid)
+ const struct lu_fid *fid, bool *exact)
{
- struct rb_node *node = llsd->llsd_rb_root.rb_node;
- struct lfsck_rbtree_node *lrn;
- int rc;
+ struct rb_node *node = llsd->llsd_rb_root.rb_node;
+ struct rb_node *prev = NULL;
+ struct lfsck_rbtree_node *lrn = NULL;
+ int rc = 0;
+
+ if (exact != NULL)
+ *exact = true;
while (node != NULL) {
+ prev = node;
lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
if (rc < 0)
return lrn;
}
- return NULL;
+ if (exact == NULL)
+ return NULL;
+
+ /* If there is no exactly matched one, then to the next valid one. */
+ *exact = false;
+
+ /* The rbtree is empty. */
+ if (rc == 0)
+ return NULL;
+
+ if (rc < 0)
+ return lrn;
+
+ node = rb_next(prev);
+
+ /* The end of the rbtree. */
+ if (node == NULL)
+ return NULL;
+
+ lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
+
+ return lrn;
}
static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
return lrn;
}
+extern const struct dt_index_operations lfsck_orphan_index_ops;
+
static int lfsck_rbtree_setup(const struct lu_env *env,
struct lfsck_component *com)
{
if (IS_ERR(obj))
RETURN(PTR_ERR(obj));
- /* XXX: Generate an in-RAM object to stand for the layout rbtree.
- * Scanning the layout rbtree will be via the iteration over
- * the object. In the future, the rbtree may be written onto
- * disk with the object.
+ /* Generate an in-RAM object to stand for the layout rbtree.
+ * Scanning the layout rbtree will be via the iteration over
+ * the object. In the future, the rbtree may be written onto
+ * disk with the object.
*
- * Mark the object to be as exist. */
+ * Mark the object to be as exist. */
obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
+ obj->do_index_ops = &lfsck_orphan_index_ops;
llsd->llsd_rb_obj = obj;
llsd->llsd_rbtree_valid = 1;
dev->dd_record_fid_accessed = 1;
if (!llsd->llsd_rbtree_valid)
GOTO(unlock, rc = 0);
- lrn = lfsck_rbtree_search(llsd, fid);
+ lrn = lfsck_rbtree_search(llsd, fid, NULL);
if (lrn == NULL) {
struct lfsck_rbtree_node *tmp;
/* Any accessed object must be a known object. */
if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
atomic_inc(&lrn->lrn_known_count);
- if (accessed) {
- if (!test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
- atomic_inc(&lrn->lrn_accessed_count);
- }
+ if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
+ atomic_inc(&lrn->lrn_accessed_count);
GOTO(unlock, rc = 0);
return rc;
}
+static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_tgt_desc *ltd,
+ struct lu_orphan_rec *rec,
+ struct lu_fid *cfid)
+{
+ struct lfsck_layout *lo = com->lc_file_ram;
+ int rc = 0;
+
+ /* XXX: To be extended in other patch. */
+
+ down_write(&com->lc_sem);
+ com->lc_new_scanned++;
+ com->lc_new_checked++;
+ if (rc > 0) {
+ lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
+ rc = 0;
+ } else if (rc < 0) {
+ lo->ll_objs_failed_phase2++;
+ }
+ up_write(&com->lc_sem);
+
+ return rc;
+}
+
static int lfsck_layout_scan_orphan(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd)
{
- /* XXX: To be extended in other patch. */
+ struct lfsck_layout *lo = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct ost_id *oi = &info->lti_oi;
+ struct lu_fid *fid = &info->lti_fid;
+ struct dt_object *obj;
+ const struct dt_it_ops *iops;
+ struct dt_it *di;
+ int rc = 0;
+ ENTRY;
- return 0;
+ CDEBUG(D_LFSCK, "%s: start the orphan scanning for OST%04x\n",
+ lfsck_lfsck2name(lfsck), ltd->ltd_index);
+
+ ostid_set_seq(oi, FID_SEQ_IDIF);
+ ostid_set_id(oi, 0);
+ ostid_to_fid(fid, oi, ltd->ltd_index);
+ obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
+ if (unlikely(IS_ERR(obj)))
+ RETURN(PTR_ERR(obj));
+
+ rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
+ if (rc != 0)
+ GOTO(put, rc);
+
+ iops = &obj->do_index_ops->dio_it;
+ di = iops->init(env, obj, 0, BYPASS_CAPA);
+ if (IS_ERR(di))
+ GOTO(put, rc = PTR_ERR(di));
+
+ rc = iops->load(env, di, 0);
+ if (rc == -ESRCH) {
+ /* -ESRCH means that the orphan OST-objects rbtree has been
+ * cleanup because of the OSS server restart or other errors. */
+ lo->ll_flags |= LF_INCOMPLETE;
+ GOTO(fini, rc);
+ }
+
+ if (rc == 0)
+ rc = iops->next(env, di);
+ else if (rc > 0)
+ rc = 0;
+
+ if (rc < 0)
+ GOTO(fini, rc);
+
+ if (rc > 0)
+ GOTO(fini, rc = 0);
+
+ do {
+ struct dt_key *key;
+ struct lu_orphan_rec *rec = &info->lti_rec;
+
+ key = iops->key(env, di);
+ com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
+ rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
+ if (rc == 0)
+ rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
+ &com->lc_fid_latest_scanned_phase2);
+ if (rc != 0 && bk->lb_param & LPF_FAILOUT)
+ GOTO(fini, rc);
+
+ lfsck_control_speed_by_self(com);
+ do {
+ rc = iops->next(env, di);
+ } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
+ } while (rc == 0);
+
+ GOTO(fini, rc);
+
+fini:
+ iops->put(env, di);
+ iops->fini(env, di);
+put:
+ lu_object_put(env, &obj->do_lu);
+
+ CDEBUG(D_LFSCK, "%s: finish the orphan scanning for OST%04x, rc = %d\n",
+ lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
+
+ return rc > 0 ? 0 : rc;
}
/* For the MDT-object with dangling reference, we need to re-create
* mark the LFSCK as INCOMPLETE. */
if (rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -ETIMEDOUT ||
rc == -EHOSTDOWN || rc == -EHOSTUNREACH) {
- CERROR("%s: Fail to take with OST %x: rc = %d.\n",
+ CERROR("%s: Fail to talk with OST %x: rc = %d.\n",
lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
lo->ll_flags |= LF_INCOMPLETE;
lo->ll_objs_skipped++;
llmd->llmd_in_double_scan = 1;
wake_up_all(&mthread->t_ctl_waitq);
+ com->lc_new_checked = 0;
+ com->lc_new_scanned = 0;
+ com->lc_time_last_checkpoint = cfs_time_current();
+ com->lc_time_next_checkpoint =
+ com->lc_time_last_checkpoint +
+ cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+
while (llmd->llmd_in_double_scan) {
struct lfsck_tgt_descs *ltds =
&lfsck->li_ost_descs;
<d->ltd_layout_phase_list);
spin_unlock(<ds->ltd_lock);
- rc = lfsck_layout_scan_orphan(env, com,
- ltd);
- if (rc != 0 &&
- bk->lb_param & LPF_FAILOUT)
- GOTO(cleanup2, rc);
+ if (bk->lb_param & LPF_ALL_TGT) {
+ rc = lfsck_layout_scan_orphan(
+ env, com, ltd);
+ if (rc != 0 &&
+ bk->lb_param & LPF_FAILOUT)
+ GOTO(cleanup2, rc);
+ }
if (unlikely(llmd->llmd_exit ||
!thread_is_running(mthread)))
lo->ll_objs_repaired[i] = 0;
pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
+ fid_zero(&com->lc_fid_latest_scanned_phase2);
}
} else {
lo->ll_status = LS_SCANNING_PHASE1;
ostid_to_fid(fid, oi, index);
tgt = lfsck_tgt_get(ltds, index);
if (unlikely(tgt == NULL)) {
- CERROR("%s: Cannot talk with OST %x which is not join "
+ CERROR("%s: Cannot talk with OST %x which did not join "
"the layout LFSCK.\n",
lfsck_lfsck2name(lfsck), index);
lo->ll_flags |= LF_INCOMPLETE;
buf += rc;
len -= rc;
+ } else if (lo->ll_status == LS_SCANNING_PHASE2) {
+ cfs_duration_t duration = cfs_time_current() -
+ lfsck->li_time_last_checkpoint;
+ __u64 checked = lo->ll_objs_checked_phase1 + com->lc_new_checked;
+ __u64 speed = checked;
+ __u64 new_checked = com->lc_new_checked * HZ;
+ __u32 rtime = lo->ll_run_time_phase1 +
+ cfs_duration_sec(duration + HALF_SEC);
+
+ if (duration != 0)
+ do_div(new_checked, duration);
+ if (rtime != 0)
+ do_div(speed, rtime);
+ rc = snprintf(buf, len,
+ "checked_phase1: "LPU64"\n"
+ "checked_phase2: "LPU64"\n"
+ "run_time_phase1: %u seconds\n"
+ "run_time_phase2: %u seconds\n"
+ "average_speed_phase1: "LPU64" items/sec\n"
+ "average_speed_phase2: N/A\n"
+ "real-time_speed_phase1: "LPU64" items/sec\n"
+ "real-time_speed_phase2: N/A\n"
+ "current_position: "DFID"\n",
+ checked,
+ lo->ll_objs_checked_phase2,
+ rtime,
+ lo->ll_run_time_phase2,
+ speed,
+ new_checked,
+ PFID(&com->lc_fid_latest_scanned_phase2));
+ if (rc <= 0)
+ goto out;
+
+ buf += rc;
+ len -= rc;
} else {
- /* XXX: LS_SCANNING_PHASE2 will be handled in the future. */
__u64 speed1 = lo->ll_objs_checked_phase1;
__u64 speed2 = lo->ll_objs_checked_phase2;
if (lr->lr_event != LE_PHASE2_DONE && lr->lr_event != LE_PEER_EXIT)
RETURN(-EINVAL);
- llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index);
+ llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
if (llst == NULL)
RETURN(-ENODEV);
spin_lock(&lfsck->li_lock);
if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
spin_unlock(&lfsck->li_lock);
- llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index);
+ llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
+ true);
if (llst != NULL)
lfsck_layout_llst_put(llst);
spin_lock(&lfsck->li_lock);
return rc;
}
+
+struct lfsck_orphan_it {
+ struct lfsck_component *loi_com;
+ struct lfsck_rbtree_node *loi_lrn;
+ struct lfsck_layout_slave_target *loi_llst;
+ struct lu_fid loi_key;
+ struct lu_orphan_rec loi_rec;
+ __u64 loi_hash;
+ unsigned int loi_over:1;
+};
+
+static int lfsck_fid_match_idx(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ const struct lu_fid *fid, int idx)
+{
+ struct seq_server_site *ss;
+ struct lu_server_fld *sf;
+ struct lu_seq_range range = { 0 };
+ int rc;
+
+ /* All abnormal cases will be returned to MDT0. */
+ if (!fid_is_norm(fid)) {
+ if (idx == 0)
+ return 1;
+
+ return 0;
+ }
+
+ ss = lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
+ if (unlikely(ss == NULL))
+ return -ENOTCONN;
+
+ sf = ss->ss_server_fld;
+ LASSERT(sf != NULL);
+
+ fld_range_set_any(&range);
+ rc = fld_server_lookup(env, sf, fid_seq(fid), &range);
+ if (rc != 0)
+ return rc;
+
+ if (!fld_range_is_mdt(&range))
+ return -EINVAL;
+
+ if (range.lsr_index == idx)
+ return 1;
+
+ return 0;
+}
+
+static void lfsck_layout_destroy_orphan(const struct lu_env *env,
+ struct dt_device *dev,
+ struct dt_object *obj)
+{
+ struct thandle *handle;
+ int rc;
+ ENTRY;
+
+ handle = dt_trans_create(env, dev);
+ if (IS_ERR(handle))
+ RETURN_EXIT;
+
+ rc = dt_declare_ref_del(env, obj, handle);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_declare_destroy(env, obj, handle);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, handle);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ rc = dt_ref_del(env, obj, handle);
+ if (rc == 0)
+ rc = dt_destroy(env, obj, handle);
+ dt_write_unlock(env, obj);
+
+ GOTO(stop, rc);
+
+stop:
+ dt_trans_stop(env, dev, handle);
+
+ RETURN_EXIT;
+}
+
+static int lfsck_orphan_index_lookup(const struct lu_env *env,
+ struct dt_object *dt,
+ struct dt_rec *rec,
+ const struct dt_key *key,
+ struct lustre_capa *capa)
+{
+ return -EOPNOTSUPP;
+}
+
+static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct dt_rec *rec,
+ const struct dt_key *key,
+ struct thandle *handle)
+{
+ return -EOPNOTSUPP;
+}
+
+static int lfsck_orphan_index_insert(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct dt_rec *rec,
+ const struct dt_key *key,
+ struct thandle *handle,
+ struct lustre_capa *capa,
+ int ignore_quota)
+{
+ return -EOPNOTSUPP;
+}
+
+static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct dt_key *key,
+ struct thandle *handle)
+{
+ return -EOPNOTSUPP;
+}
+
+static int lfsck_orphan_index_delete(const struct lu_env *env,
+ struct dt_object *dt,
+ const struct dt_key *key,
+ struct thandle *handle,
+ struct lustre_capa *capa)
+{
+ return -EOPNOTSUPP;
+}
+
+static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
+ struct dt_object *dt,
+ __u32 attr,
+ struct lustre_capa *capa)
+{
+ struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
+ struct lfsck_instance *lfsck;
+ struct lfsck_component *com = NULL;
+ struct lfsck_layout_slave_data *llsd;
+ struct lfsck_orphan_it *it = NULL;
+ int rc = 0;
+ ENTRY;
+
+ lfsck = lfsck_instance_find(dev, true, false);
+ if (unlikely(lfsck == NULL))
+ RETURN(ERR_PTR(-ENODEV));
+
+ com = lfsck_component_find(lfsck, LT_LAYOUT);
+ if (unlikely(com == NULL))
+ GOTO(out, rc = -ENOENT);
+
+ llsd = com->lc_data;
+ if (!llsd->llsd_rbtree_valid)
+ GOTO(out, rc = -ESRCH);
+
+ OBD_ALLOC_PTR(it);
+ if (it == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
+ if (it->loi_llst == NULL)
+ GOTO(out, rc = -ENODEV);
+
+ if (dev->dd_record_fid_accessed) {
+ /* The first iteratino against the rbtree, scan the whole rbtree
+ * to remove the nodes which do NOT need to be handled. */
+ write_lock(&llsd->llsd_rb_lock);
+ if (dev->dd_record_fid_accessed) {
+ struct rb_node *node;
+ struct rb_node *next;
+ struct lfsck_rbtree_node *lrn;
+
+ /* No need to record the fid accessing anymore. */
+ dev->dd_record_fid_accessed = 0;
+
+ node = rb_first(&llsd->llsd_rb_root);
+ while (node != NULL) {
+ next = rb_next(node);
+ lrn = rb_entry(node, struct lfsck_rbtree_node,
+ lrn_node);
+ if (atomic_read(&lrn->lrn_known_count) <=
+ atomic_read(&lrn->lrn_accessed_count)) {
+ rb_erase(node, &llsd->llsd_rb_root);
+ lfsck_rbtree_free(lrn);
+ }
+ node = next;
+ }
+ }
+ write_unlock(&llsd->llsd_rb_lock);
+ }
+
+ /* read lock the rbtree when init, and unlock when fini */
+ read_lock(&llsd->llsd_rb_lock);
+ it->loi_com = com;
+ com = NULL;
+
+ GOTO(out, rc = 0);
+
+out:
+ if (com != NULL)
+ lfsck_component_put(env, com);
+ lfsck_instance_put(env, lfsck);
+ if (rc != 0) {
+ if (it != NULL)
+ OBD_FREE_PTR(it);
+
+ it = (struct lfsck_orphan_it *)ERR_PTR(rc);
+ }
+
+ return (struct dt_it *)it;
+}
+
+static void lfsck_orphan_it_fini(const struct lu_env *env,
+ struct dt_it *di)
+{
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+ struct lfsck_component *com = it->loi_com;
+ struct lfsck_layout_slave_data *llsd;
+ struct lfsck_layout_slave_target *llst;
+
+ if (com != NULL) {
+ llsd = com->lc_data;
+ read_unlock(&llsd->llsd_rb_lock);
+ llst = it->loi_llst;
+ LASSERT(llst != NULL);
+
+ /* Save the key and hash for iterate next. */
+ llst->llst_fid = it->loi_key;
+ llst->llst_hash = it->loi_hash;
+ lfsck_layout_llst_put(llst);
+ lfsck_component_put(env, com);
+ }
+ OBD_FREE_PTR(it);
+}
+
+/**
+ * \retval +1: the iteration finished
+ * \retval 0: on success, not finished
+ * \retval -ve: on error
+ */
+static int lfsck_orphan_it_next(const struct lu_env *env,
+ struct dt_it *di)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct filter_fid_old *pfid = &info->lti_old_pfid;
+ struct lu_attr *la = &info->lti_la;
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+ struct lu_fid *key = &it->loi_key;
+ struct lu_orphan_rec *rec = &it->loi_rec;
+ struct lfsck_component *com = it->loi_com;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_layout_slave_data *llsd = com->lc_data;
+ struct dt_object *obj;
+ struct lfsck_rbtree_node *lrn;
+ int pos;
+ int rc;
+ __u32 save;
+ __u32 idx = it->loi_llst->llst_index;
+ bool exact = false;
+ ENTRY;
+
+ if (it->loi_over)
+ RETURN(1);
+
+again0:
+ lrn = it->loi_lrn;
+ if (lrn == NULL) {
+ lrn = lfsck_rbtree_search(llsd, key, &exact);
+ if (lrn == NULL) {
+ it->loi_over = 1;
+ RETURN(1);
+ }
+
+ it->loi_lrn = lrn;
+ if (!exact) {
+ key->f_seq = lrn->lrn_seq;
+ key->f_oid = lrn->lrn_first_oid;
+ key->f_ver = 0;
+ }
+ } else {
+ key->f_oid++;
+ if (unlikely(key->f_oid == 0)) {
+ key->f_seq++;
+ it->loi_lrn = NULL;
+ goto again0;
+ }
+
+ if (key->f_oid >=
+ lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
+ it->loi_lrn = NULL;
+ goto again0;
+ }
+ }
+
+ if (unlikely(atomic_read(&lrn->lrn_known_count) <=
+ atomic_read(&lrn->lrn_accessed_count))) {
+ struct rb_node *next = rb_next(&lrn->lrn_node);
+
+ while (next != NULL) {
+ lrn = rb_entry(next, struct lfsck_rbtree_node,
+ lrn_node);
+ if (atomic_read(&lrn->lrn_known_count) >
+ atomic_read(&lrn->lrn_accessed_count))
+ break;
+ next = rb_next(next);
+ }
+
+ if (next == NULL) {
+ it->loi_over = 1;
+ RETURN(1);
+ }
+
+ it->loi_lrn = lrn;
+ key->f_seq = lrn->lrn_seq;
+ key->f_oid = lrn->lrn_first_oid;
+ key->f_ver = 0;
+ }
+
+ pos = key->f_oid - lrn->lrn_first_oid;
+
+again1:
+ pos = find_next_bit(lrn->lrn_known_bitmap,
+ LFSCK_RBTREE_BITMAP_WIDTH, pos);
+ if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
+ key->f_oid = lrn->lrn_first_oid + pos;
+ if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
+ key->f_seq++;
+ key->f_oid = 0;
+ }
+ it->loi_lrn = NULL;
+ goto again0;
+ }
+
+ if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
+ pos++;
+ goto again1;
+ }
+
+ key->f_oid = lrn->lrn_first_oid + pos;
+ obj = lfsck_object_find(env, lfsck, key);
+ if (IS_ERR(obj)) {
+ rc = PTR_ERR(obj);
+ if (rc == -ENOENT) {
+ pos++;
+ goto again1;
+ }
+ RETURN(rc);
+ }
+
+ dt_read_lock(env, obj, 0);
+ if (!dt_object_exists(obj)) {
+ dt_read_unlock(env, obj);
+ lfsck_object_put(env, obj);
+ pos++;
+ goto again1;
+ }
+
+ rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
+ XATTR_NAME_FID, BYPASS_CAPA);
+ if (rc == -ENODATA) {
+ /* For the pre-created OST-object, update the bitmap to avoid
+ * others LFSCK (second phase) iteration to touch it again. */
+ if (la->la_ctime == 0) {
+ if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
+ atomic_inc(&lrn->lrn_accessed_count);
+
+ /* For the race between repairing dangling referenced
+ * MDT-object and unlink the file, it may left orphan
+ * OST-object there. Destroy it now! */
+ if (unlikely(!(la->la_mode & S_ISUID))) {
+ dt_read_unlock(env, obj);
+ lfsck_layout_destroy_orphan(env,
+ lfsck->li_bottom,
+ obj);
+ lfsck_object_put(env, obj);
+ pos++;
+ goto again1;
+ }
+ } else if (idx == 0) {
+ /* If the orphan OST-object has no parent information,
+ * regard it as referenced by the MDT-object on MDT0. */
+ fid_zero(&rec->lor_fid);
+ rec->lor_uid = la->la_uid;
+ rec->lor_gid = la->la_gid;
+ GOTO(out, rc = 0);
+ }
+
+ dt_read_unlock(env, obj);
+ lfsck_object_put(env, obj);
+ pos++;
+ goto again1;
+ }
+
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc != sizeof(struct filter_fid) &&
+ rc != sizeof(struct filter_fid_old))
+ GOTO(out, rc = -EINVAL);
+
+ fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
+ /* In fact, the ff_parent::f_ver is not the real parent FID::f_ver,
+ * instead, it is the OST-object index in its parent MDT-object
+ * layout EA. */
+ save = rec->lor_fid.f_ver;
+ rec->lor_fid.f_ver = 0;
+ rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
+ /* If the orphan OST-object does not claim the MDT, then next.
+ *
+ * If we do not know whether it matches or not, then return it
+ * to the MDT for further check. */
+ if (rc == 0) {
+ dt_read_unlock(env, obj);
+ lfsck_object_put(env, obj);
+ pos++;
+ goto again1;
+ }
+
+ rec->lor_fid.f_ver = save;
+ rec->lor_uid = la->la_uid;
+ rec->lor_gid = la->la_gid;
+
+ CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
+ lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
+ rec->lor_uid, rec->lor_gid);
+
+ GOTO(out, rc = 0);
+
+out:
+ dt_read_unlock(env, obj);
+ lfsck_object_put(env, obj);
+ if (rc == 0)
+ it->loi_hash++;
+
+ return rc;
+}
+
+/**
+ * \retval +1: locate to the exactly position
+ * \retval 0: cannot locate to the exactly position,
+ * call next() to move to a valid position.
+ * \retval -ve: on error
+ */
+static int lfsck_orphan_it_get(const struct lu_env *env,
+ struct dt_it *di,
+ const struct dt_key *key)
+{
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+ int rc;
+
+ it->loi_key = *(struct lu_fid *)key;
+ rc = lfsck_orphan_it_next(env, di);
+ if (rc == 1)
+ return 0;
+
+ if (rc == 0)
+ return 1;
+
+ return rc;
+}
+
+static void lfsck_orphan_it_put(const struct lu_env *env,
+ struct dt_it *di)
+{
+}
+
+static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+
+ return (struct dt_key *)&it->loi_key;
+}
+
+static int lfsck_orphan_it_key_size(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ return sizeof(struct lu_fid);
+}
+
+static int lfsck_orphan_it_rec(const struct lu_env *env,
+ const struct dt_it *di,
+ struct dt_rec *rec,
+ __u32 attr)
+{
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+
+ *(struct lu_orphan_rec *)rec = it->loi_rec;
+
+ return 0;
+}
+
+static __u64 lfsck_orphan_it_store(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+
+ return it->loi_hash;
+}
+
+/**
+ * \retval +1: locate to the exactly position
+ * \retval 0: cannot locate to the exactly position,
+ * call next() to move to a valid position.
+ * \retval -ve: on error
+ */
+static int lfsck_orphan_it_load(const struct lu_env *env,
+ const struct dt_it *di,
+ __u64 hash)
+{
+ struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
+ struct lfsck_layout_slave_target *llst = it->loi_llst;
+ int rc;
+
+ LASSERT(llst != NULL);
+
+ if (hash != llst->llst_hash) {
+ CWARN("%s: the given hash "LPU64" for orphan iteration does "
+ "not match the one when fini "LPU64", to be reset.\n",
+ lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
+ llst->llst_hash);
+ fid_zero(&llst->llst_fid);
+ llst->llst_hash = 0;
+ }
+
+ it->loi_key = llst->llst_fid;
+ it->loi_hash = llst->llst_hash;
+ rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
+ if (rc == 1)
+ return 0;
+
+ if (rc == 0)
+ return 1;
+
+ return rc;
+}
+
+static int lfsck_orphan_it_key_rec(const struct lu_env *env,
+ const struct dt_it *di,
+ void *key_rec)
+{
+ return 0;
+}
+
+const struct dt_index_operations lfsck_orphan_index_ops = {
+ .dio_lookup = lfsck_orphan_index_lookup,
+ .dio_declare_insert = lfsck_orphan_index_declare_insert,
+ .dio_insert = lfsck_orphan_index_insert,
+ .dio_declare_delete = lfsck_orphan_index_declare_delete,
+ .dio_delete = lfsck_orphan_index_delete,
+ .dio_it = {
+ .init = lfsck_orphan_it_init,
+ .fini = lfsck_orphan_it_fini,
+ .get = lfsck_orphan_it_get,
+ .put = lfsck_orphan_it_put,
+ .next = lfsck_orphan_it_next,
+ .key = lfsck_orphan_it_key,
+ .key_size = lfsck_orphan_it_key_size,
+ .rec = lfsck_orphan_it_rec,
+ .store = lfsck_orphan_it_store,
+ .load = lfsck_orphan_it_load,
+ .key_rec = lfsck_orphan_it_key_rec,
+ }
+};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff