+ /* 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
+ * \param[in] unknown set if does not know how to repair the inconsistency
+ *
+ * \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, bool *unknown)
+{
+ 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;
+
+ rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
+ info->lti_key,
+ sizeof(info->lti_key));
+ /* The unique linkEA entry with bad parent will be handled as orphan. */
+ if (rc != 0) {
+ 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, lfsck, 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);
+ 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);
+ } /* !dt_object_exists(parent) */
+
+ /* 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_dir(env, parent, cname->ln_name, &tfid);
+ 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, lfsck, 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);
+ 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);
+ } /* rc == -ENOENT */
+
+ 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);
+ }
+
+ /* Zero FID may because the remote directroy object has invalid linkEA,
+ * or lost linkEA. Under such case, the LFSCK on this MDT does not know
+ * how to repair the inconsistency, but the namespace LFSCK on the MDT
+ * where its name entry resides may has more information (name, FID) to
+ * repair such inconsistency. So here, keep the inconsistency to avoid
+ * some imporper repairing. */
+ if (fid_is_zero(pfid)) {
+ if (unknown)
+ *unknown = true;
+
+ 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/
+ * \param[in] unknown set if does not know how to repair the inconsistency
+ *
+ * \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, bool *unknown)
+{
+ 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 rc = 0;
+ bool once = true;
+ ENTRY;
+
+again:
+ while (ldata->ld_lee != NULL) {
+ rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
+ info->lti_key,
+ sizeof(info->lti_key));
+ /* Drop invalid linkEA entry. */
+ if (rc != 0) {
+ lfsck_linkea_del_buf(ldata, cname);
+ continue;
+ }
+
+ /* Drop repeated linkEA entries. */
+ lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
+
+ /* 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 && !fid_is_zero(pfid) &&
+ !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. */
+ lfsck_linkea_del_buf(ldata, cname);
+ continue;
+ }
+
+ break;
+ }
+
+ /* The linkEA entry with bad parent will be removed. */
+ if (unlikely(!dt_try_as_dir(env, parent))) {
+ lfsck_object_put(env, parent);
+ lfsck_linkea_del_buf(ldata, cname);
+ continue;
+ }
+
+ rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
+ *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 the parent (that is declared via linkEA entry)
+ * directory contains the specified child, but such
+ * parent does not match the dotdot name entry, then
+ * trust the linkEA. */
+ if (!fid_is_zero(pfid) && !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_links_new(&ldata_new, &info->lti_big_buf,
+ cname, pfid2);
+ if (rc != 0)
+ RETURN(rc);
+
+ rc = lfsck_namespace_rebuild_linkea(env, com, child,
+ &ldata_new);
+ if (rc < 0)
+ RETURN(rc);
+
+ lfsck_linkea_del_buf(ldata, cname);
+ 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) {
+ rc = lfsck_namespace_unpack_linkea_entry(ldata,
+ cname, &tfid, info->lti_key,
+ sizeof(info->lti_key));
+ if (rc != 0)
+ 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:
+ lfsck_linkea_del_buf(ldata, cname);
+ }
+
+ ns->ln_dirent_repaired += dirent_count;
+
+ RETURN(rc);
+ } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
+
+ 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;
+
+ lfsck_linkea_del_buf(ldata, cname);
+ } /* while (ldata->ld_lee != NULL) */
+
+ /* If there is still linkEA overflow, return. */
+ if (unlikely(ldata->ld_leh->leh_overflow_time))
+ RETURN(0);
+
+ linkea_first_entry(ldata);
+ if (ldata->ld_leh->leh_reccount == 1) {
+ rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
+ lh, type, NULL, unknown);
+
+ 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 many 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_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 rc = 0;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(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);
+
+ rc = dt_attr_get(env, obj, la);
+ if (rc != 0)
+ GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
+
+ rc = lfsck_links_read2_with_rec(env, obj, &ldata);
+ if (rc)
+ GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
+
+ /* XXX: Currently, we only update the nlink attribute if the known
+ * linkEA entries is larger than the nlink attribute. That is
+ * safe action. */
+ if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
+ unlikely(la->la_nlink == 0 ||
+ ldata.ld_leh->leh_overflow_time))
+ 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_dir(env, child, dotdot, pfid);
+ 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);
+ } /* rc != 0 */
+
+ 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, &unknown);
+ 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, &unknown);
+
+ 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;
+}
+
+static inline bool
+lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
+ struct lfsck_namespace *ns)
+{
+ /* Both the leh_overflow_time and ln_time_latest_reset are
+ * local time based, so need NOT to care about clock drift
+ * among the servers. */
+ return ldata->ld_leh->leh_overflow_time &&
+ ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
+}
+
+/**
+ * Clear the object's linkEA overflow timestamp.
+ *
+ * If the MDT-object has too many hard links as to the linkEA cannot hold
+ * all of them, then overflow timestamp will be set in the linkEA header.
+ * If some hard links are removed after that, then it is possible to hold
+ * other missed linkEA entries. If the namespace LFSCK have added all the
+ * related linkEA entries, then it will remove the overflow timestamp.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] ldata pointer to the linkEA data for the given @obj
+ * \param[in] obj pointer to the dt_object to 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_linkea_clear_overflow(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct linkea_data *ldata,
+ struct dt_object *obj)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct thandle *th = NULL;
+ struct lustre_handle lh = { 0 };
+ struct lu_buf linkea_buf;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(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));
+
+ rc = dt_declare_xattr_set(env, obj,
+ lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
+ 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);
+ rc = lfsck_links_read(env, obj, ldata);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
+ GOTO(unlock, rc = 0);
+
+ ldata->ld_leh->leh_overflow_time = 0;
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock, rc = 1);
+
+ /* If all known entries are in the linkEA, then the 'leh_reccount'
+ * should NOT be zero. */
+ LASSERT(ldata->ld_leh->leh_reccount > 0);
+
+ lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
+ rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
+ if (unlikely(rc == -ENOSPC))
+ rc = 0;
+ else if (!rc)
+ rc = 1;
+
+ GOTO(unlock, 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: clear linkea overflow timestamp for the object "
+ DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
+
+ return rc;
+}
+
+/**
+ * Verify the object's agent entry.
+ *
+ * If the object claims to have agent entry but the linkEA does not contain
+ * remote parent, then remove the agent entry. Otherwise, if the object has
+ * no agent entry but its linkEA contains remote parent, then will generate
+ * agent entry for it.
+ *
+ * \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
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj)
+{
+ struct linkea_data ldata = { NULL };
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ 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);
+ __u32 idx = lfsck_dev_idx(lfsck);
+ int rc;
+ bool remote = false;
+ ENTRY;
+
+ if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
+ RETURN(0);
+
+ rc = lfsck_links_read_with_rec(env, obj, &ldata);
+ if (rc == -ENOENT || rc == -ENODATA)
+ RETURN(0);
+
+ if (rc && rc != -EINVAL)
+ GOTO(out, rc);
+
+ /* We check the agent entry again after verifying the linkEA
+ * successfully. So invalid linkEA should be dryrun mode. */
+ if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
+ RETURN(0);
+
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL && !remote) {
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+ cname, pfid);
+ if (!linkea_entry_is_valid(&ldata, cname, pfid))
+ GOTO(out, rc = 0);
+
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if (rc)
+ GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
+
+ if (range->lsr_index != idx)
+ remote = true;
+ else
+ linkea_next_entry(&ldata);
+ }
+
+ if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+ (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct linkea_data ldata2 = { NULL };
+ struct lustre_handle lh = { 0 };
+ struct lu_buf linkea_buf;
+ struct thandle *handle;
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(out, rc = 1);
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc)
+ GOTO(out, rc);
+
+ handle = dt_trans_create(env, dev);
+ if (IS_ERR(handle))
+ GOTO(unlock, rc = PTR_ERR(handle));
+
+ 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, LU_XATTR_REPLACE, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
+ if (rc) {
+ if (rc == -ENOENT || rc == -ENODATA)
+ rc = 0;
+ GOTO(unlock2, rc);
+ }
+
+ /* If someone changed linkEA by race, then the agent
+ * entry will be updated by lower layer automatically. */
+ if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
+ memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len) != 0)
+ GOTO(unlock2, rc = 0);
+
+ rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
+ LU_XATTR_REPLACE, handle);
+ if (!rc)
+ rc = 1;
+
+ GOTO(unlock2, rc);
+
+unlock2:
+ dt_write_unlock(env, obj);
+stop:
+ dt_trans_stop(env, dev, handle);
+unlock:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ }
+
+ GOTO(out, rc);
+
+out:
+ if (rc > 0)
+ ns->ln_agent_entries_repaired++;
+ if (rc)
+ CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
+ 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);
+ if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+ rc = lfsck_namespace_check_agent_entry(env, com, child);
+
+ RETURN(rc);
+ }
+
+ rc = lfsck_links_read(env, child, &ldata);
+ dt_read_unlock(env, child);
+
+ if (rc == -EINVAL) {
+ struct lustre_handle lh = { 0 };
+
+ rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc == 0) {
+ rc = lfsck_namespace_links_remove(env, com, child);
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ }
+
+ GOTO(out, rc);
+ }
+
+ if (rc != 0)
+ GOTO(out, rc);
+
+ if (!(ns->ln_flags & LF_INCOMPLETE) &&
+ unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
+ rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
+ child);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ ns->ln_linkea_overflow_cleared++;
+ }
+
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL) {
+ rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
+ info->lti_key,
+ sizeof(info->lti_key));
+ /* Invalid PFID in the linkEA entry. */
+ if (rc != 0) {
+ 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 = 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 */
+ }
+
+ 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);
+ 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;
+ } /* !dt_object_exists(parent) */
+
+ /* 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_dir(env, parent, cname->ln_name, cfid);
+ 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;
+ }
+
+ /* The following handles -ENOENT case */
+
+ 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 fewer 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, lfsck, 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);
+ 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 && 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) &&
+ (ldata.ld_leh == NULL ||
+ !ldata.ld_leh->leh_overflow_time)) {
+ /* 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) {
+ if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
+ !S_ISLNK(lfsck_object_type(child)))) {
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
+ "the object "DFID"'s nlink count %d "
+ "does not match linkEA count %d, "
+ "type %o, skip it.\n",
+ lfsck_lfsck2name(lfsck),
+ PFID(lfsck_dto2fid(child)),
+ la->la_nlink, count,
+ lfsck_object_type(child));
+ } else if (la->la_nlink < count &&
+ likely(!ldata.ld_leh->leh_overflow_time)) {
+ 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;
+ }
+
+ if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+ rc = lfsck_namespace_check_agent_entry(env, com, child);
+
+ return rc;
+}
+
+static void lfsck_namespace_dump_statistics(struct seq_file *m,
+ struct lfsck_namespace *ns,
+ __u64 checked_phase1,
+ __u64 checked_phase2,
+ time64_t time_phase1,
+ time64_t time_phase2, bool dryrun)
+{
+ const char *postfix = dryrun ? "inconsistent" : "repaired";
+
+ seq_printf(m, "checked_phase1: %llu\n"
+ "checked_phase2: %llu\n"
+ "%s_phase1: %llu\n"
+ "%s_phase2: %llu\n"
+ "failed_phase1: %llu\n"
+ "failed_phase2: %llu\n"
+ "directories: %llu\n"
+ "dirent_%s: %llu\n"
+ "linkea_%s: %llu\n"
+ "nlinks_%s: %llu\n"
+ "multiple_linked_checked: %llu\n"
+ "multiple_linked_%s: %llu\n"
+ "unknown_inconsistency: %llu\n"
+ "unmatched_pairs_%s: %llu\n"
+ "dangling_%s: %llu\n"
+ "multiple_referenced_%s: %llu\n"
+ "bad_file_type_%s: %llu\n"
+ "lost_dirent_%s: %llu\n"
+ "local_lost_found_scanned: %llu\n"
+ "local_lost_found_moved: %llu\n"
+ "local_lost_found_skipped: %llu\n"
+ "local_lost_found_failed: %llu\n"
+ "striped_dirs_scanned: %llu\n"
+ "striped_dirs_%s: %llu\n"
+ "striped_dirs_failed: %llu\n"
+ "striped_dirs_disabled: %llu\n"
+ "striped_dirs_skipped: %llu\n"
+ "striped_shards_scanned: %llu\n"
+ "striped_shards_%s: %llu\n"
+ "striped_shards_failed: %llu\n"
+ "striped_shards_skipped: %llu\n"
+ "name_hash_%s: %llu\n"
+ "linkea_overflow_%s: %llu\n"
+ "agent_entries_%s: %llu\n"
+ "success_count: %u\n"
+ "run_time_phase1: %lld seconds\n"
+ "run_time_phase2: %lld seconds\n",
+ checked_phase1,
+ checked_phase2,
+ dryrun ? "inconsistent" : "updated",
+ ns->ln_items_repaired,
+ dryrun ? "inconsistent" : "updated",
+ ns->ln_objs_repaired_phase2,
+ ns->ln_items_failed,
+ ns->ln_objs_failed_phase2,
+ ns->ln_dirs_checked,
+ postfix, ns->ln_dirent_repaired,
+ postfix, ns->ln_linkea_repaired,
+ postfix, ns->ln_objs_nlink_repaired,
+ ns->ln_mul_linked_checked,
+ postfix, ns->ln_mul_linked_repaired,
+ ns->ln_unknown_inconsistency,
+ postfix, ns->ln_unmatched_pairs_repaired,
+ postfix, ns->ln_dangling_repaired,
+ postfix, ns->ln_mul_ref_repaired,
+ postfix, ns->ln_bad_type_repaired,
+ postfix, 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,
+ postfix, 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,
+ postfix, ns->ln_striped_shards_repaired,
+ ns->ln_striped_shards_failed,
+ ns->ln_striped_shards_skipped,
+ postfix, ns->ln_name_hash_repaired,
+ dryrun ? "inconsistent" : "cleared",
+ ns->ln_linkea_overflow_cleared,
+ postfix, ns->ln_agent_entries_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;
+ 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;
+ time64_t 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;
+ ns->ln_time_latest_reset = ktime_get_real_seconds();
+
+ rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
+ &dt_lfsck_namespace_features,
+ LFSCK_NAMESPACE, true);
+ if (rc)
+ GOTO(out, rc);
+
+ rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
+ LFSCK_NAMESPACE, true);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
+ CFS_RESET_BITMAP(lad->lad_bitmap);
+
+ rc = lfsck_namespace_store(env, com);
+
+ 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;
+ struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
+ __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
+ int rc;
+ bool wakeup = false;
+ ENTRY;
+
+ if (llmv == NULL)
+ RETURN_EXIT;
+
+ rc = dt_attr_get(env, lfsck->li_obj_dir, la);
+ if (rc)
+ 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),
+ la, 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;
+ lnr->lnr_type = lso->lso_attr.la_mode;
+
+ spin_lock(&lad->lad_lock);
+ if (lad->lad_assistant_status < 0 ||
+ unlikely(!thread_is_running(&lfsck->li_thread) ||
+ !thread_is_running(&lad->lad_thread))) {
+ 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 += ktime_get_seconds() -
+ lfsck->li_time_last_checkpoint;
+ ns->ln_time_last_checkpoint = ktime_get_real_seconds();
+ ns->ln_items_checked += com->lc_new_checked;
+ com->lc_new_checked = 0;
+ }
+
+ rc = lfsck_namespace_store(env, com);
+ up_write(&com->lc_sem);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
+ ", "DFID", %#llx], status = %d: 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, ns->ln_status, 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 = ktime_get_real_seconds();
+ 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 [%llu, "
+ DFID", %#llx]: 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);
+ struct lu_attr la = { .la_valid = 0 };
+ bool remote = false;
+ int rc;
+ ENTRY;
+
+ rc = dt_attr_get(env, obj, &la);
+ if (unlikely(rc || (la.la_valid & LA_FLAGS &&
+ la.la_flags & LUSTRE_ORPHAN_FL))) {
+ CDEBUG(D_INFO,
+ "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(fid),
+ la.la_valid, la.la_flags, rc);
+
+ return rc;
+ }
+
+ 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 && rc != -ENODATA)
+ GOTO(out, rc);
+
+ if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
+ rc = lfsck_namespace_check_for_double_scan(env, com, obj);
+
+ GOTO(out, rc);
+ }
+
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL) {
+ 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 if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
+ GOTO(out, rc);
+ } else {
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if ((rc == -ENOENT) ||
+ (!rc && range->lsr_index != idx)) {
+ remote = true;
+ break;
+ }
+ }
+ if (rc)
+ GOTO(out, rc);
+
+ linkea_next_entry(&ldata);
+ }
+
+ if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+ (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_AGENT_ENTRY, true);
+ if (rc)
+ 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);
+ }
+
+ if (remote)
+ 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;
+ bool wakeup = false;
+
+ wait_event_idle(mthread->t_ctl_waitq,
+ lad->lad_prefetched < bk->lb_async_windows ||
+ !thread_is_running(mthread) ||
+ !thread_is_running(athread));
+
+ if (unlikely(!thread_is_running(mthread) ||
+ !thread_is_running(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 ||
+ unlikely(!thread_is_running(mthread) ||
+ !thread_is_running(athread))) {
+ 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 += ktime_get_seconds() -
+ lfsck->li_time_last_checkpoint;
+ ns->ln_time_last_checkpoint = ktime_get_real_seconds();
+ ns->ln_items_checked += com->lc_new_checked;
+ com->lc_new_checked = 0;
+ }
+
+ rc = lfsck_namespace_store(env, com);
+ up_write(&com->lc_sem);
+
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+
+ RETURN(rc);
+}
+
+static void
+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;
+
+ 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_status2name(ns->ln_status));
+
+ lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
+
+ lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
+
+ lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
+
+ lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
+
+ lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
+
+ lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
+
+ lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
+ "last_checkpoint_position");
+
+ lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
+ "first_failure_position");
+
+ if (ns->ln_status == LS_SCANNING_PHASE1) {
+ struct lfsck_position pos;
+ time64_t duration = ktime_get_seconds() -
+ 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;
+ time64_t rtime = ns->ln_run_time_phase1 + duration;
+
+ if (duration != 0)
+ new_checked = div64_s64(new_checked, duration);
+
+ if (rtime != 0)
+ speed = div64_s64(speed, rtime);
+
+ lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
+ bk->lb_param & LPF_DRYRUN);
+ seq_printf(m, "average_speed_phase1: %llu items/sec\n"
+ "average_speed_phase2: N/A\n"
+ "average_speed_total: %llu items/sec\n"
+ "real_time_speed_phase1: %llu items/sec\n"
+ "real_time_speed_phase2: N/A\n",
+ speed,
+ speed,
+ new_checked);
+
+ if (likely(lfsck->li_di_oit)) {
+ const struct dt_it_ops *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) {
+ 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);
+ } else {
+ pos = ns->ln_pos_last_checkpoint;
+ }
+
+ lfsck_pos_dump(m, &pos, "current_position");
+ } else if (ns->ln_status == LS_SCANNING_PHASE2) {
+ time64_t duration = ktime_get_seconds() -
+ 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;
+ time64_t rtime = ns->ln_run_time_phase2 + duration;
+ time64_t time0 = ns->ln_run_time_phase1 + rtime;
+
+ if (duration != 0)
+ new_checked = div64_s64(new_checked, duration);
+
+ if (ns->ln_run_time_phase1 != 0)
+ speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
+ else if (ns->ln_items_checked != 0)
+ time0++;
+
+ if (rtime != 0)
+ speed2 = div64_s64(speed2, rtime);
+ else if (checked != 0)
+ time0++;
+
+ if (time0 != 0)
+ speed0 = div64_s64(speed0, time0);
+
+ lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
+ checked,
+ ns->ln_run_time_phase1, rtime,
+ bk->lb_param & LPF_DRYRUN);
+ seq_printf(m, "average_speed_phase1: %llu items/sec\n"
+ "average_speed_phase2: %llu objs/sec\n"
+ "average_speed_total: %llu items/sec\n"
+ "real_time_speed_phase1: N/A\n"
+ "real_time_speed_phase2: %llu 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;
+ time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
+
+ if (ns->ln_run_time_phase1 != 0)
+ speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
+ else if (ns->ln_items_checked != 0)
+ time0++;