return ERR_PTR(-ENOMEM);
INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
- lu_object_get(&lfsck->li_obj_dir->do_lu);
- lnr->lnr_obj = lfsck->li_obj_dir;
+ lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
+ lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
lnr->lnr_fid = ent->lde_fid;
lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
lnr->lnr_dir_cookie = ent->lde_hash;
struct lfsck_namespace_req *lnr =
container_of0(lar, struct lfsck_namespace_req, lnr_lar);
+ if (lnr->lnr_lmv != NULL)
+ lfsck_lmv_put(env, lnr->lnr_lmv);
+
lu_object_put(env, &lnr->lnr_obj->do_lu);
OBD_FREE(lnr, lnr->lnr_size);
}
le64_to_cpu(src->ln_objs_repaired_phase2);
dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
- dst->ln_objs_lost_found = le64_to_cpu(src->ln_objs_lost_found);
fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
&src->ln_fid_latest_scanned_phase2);
dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
+ dst->ln_lost_dirent_repaired =
+ le64_to_cpu(src->ln_lost_dirent_repaired);
+ dst->ln_striped_dirs_scanned =
+ le64_to_cpu(src->ln_striped_dirs_scanned);
+ dst->ln_striped_dirs_repaired =
+ le64_to_cpu(src->ln_striped_dirs_repaired);
+ dst->ln_striped_dirs_failed =
+ le64_to_cpu(src->ln_striped_dirs_failed);
+ dst->ln_striped_dirs_disabled =
+ le64_to_cpu(src->ln_striped_dirs_disabled);
+ dst->ln_striped_dirs_skipped =
+ le64_to_cpu(src->ln_striped_dirs_skipped);
+ dst->ln_striped_shards_scanned =
+ le64_to_cpu(src->ln_striped_shards_scanned);
+ dst->ln_striped_shards_repaired =
+ le64_to_cpu(src->ln_striped_shards_repaired);
+ dst->ln_striped_shards_failed =
+ le64_to_cpu(src->ln_striped_shards_failed);
+ dst->ln_striped_shards_skipped =
+ le64_to_cpu(src->ln_striped_shards_skipped);
+ dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
+ dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
+ dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
+ dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
+ dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
+ dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
}
static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
cpu_to_le64(src->ln_objs_repaired_phase2);
dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
- dst->ln_objs_lost_found = cpu_to_le64(src->ln_objs_lost_found);
fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
&src->ln_fid_latest_scanned_phase2);
dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
+ dst->ln_lost_dirent_repaired =
+ cpu_to_le64(src->ln_lost_dirent_repaired);
+ dst->ln_striped_dirs_scanned =
+ cpu_to_le64(src->ln_striped_dirs_scanned);
+ dst->ln_striped_dirs_repaired =
+ cpu_to_le64(src->ln_striped_dirs_repaired);
+ dst->ln_striped_dirs_failed =
+ cpu_to_le64(src->ln_striped_dirs_failed);
+ dst->ln_striped_dirs_disabled =
+ cpu_to_le64(src->ln_striped_dirs_disabled);
+ dst->ln_striped_dirs_skipped =
+ cpu_to_le64(src->ln_striped_dirs_skipped);
+ dst->ln_striped_shards_scanned =
+ cpu_to_le64(src->ln_striped_shards_scanned);
+ dst->ln_striped_shards_repaired =
+ cpu_to_le64(src->ln_striped_shards_repaired);
+ dst->ln_striped_shards_failed =
+ cpu_to_le64(src->ln_striped_shards_failed);
+ dst->ln_striped_shards_skipped =
+ cpu_to_le64(src->ln_striped_shards_skipped);
+ dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
+ dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
+ dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
+ dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
+ dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
+ dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
}
static void lfsck_namespace_record_failure(const struct lu_env *env,
}
/**
+ * Load the MDT bitmap from the lfsck_namespace trace file.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure or data corruption
+ */
+static int lfsck_namespace_load_bitmap(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct dt_object *obj = com->lc_obj;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ cfs_bitmap_t *bitmap = lad->lad_bitmap;
+ ssize_t size;
+ __u32 nbits;
+ int rc;
+ ENTRY;
+
+ if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
+ ns->ln_bitmap_size)
+ nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
+ else
+ nbits = ns->ln_bitmap_size;
+
+ if (unlikely(nbits < BITS_PER_LONG))
+ nbits = BITS_PER_LONG;
+
+ if (nbits > bitmap->size) {
+ __u32 new_bits = bitmap->size;
+ cfs_bitmap_t *new_bitmap;
+
+ while (new_bits < nbits)
+ new_bits <<= 1;
+
+ new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
+ if (new_bitmap == NULL)
+ RETURN(-ENOMEM);
+
+ lad->lad_bitmap = new_bitmap;
+ CFS_FREE_BITMAP(bitmap);
+ bitmap = new_bitmap;
+ }
+
+ if (ns->ln_bitmap_size == 0) {
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(bitmap);
+
+ RETURN(0);
+ }
+
+ size = (ns->ln_bitmap_size + 7) >> 3;
+ rc = dt_xattr_get(env, obj,
+ lfsck_buf_get(env, bitmap->data, size),
+ XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
+ if (rc != size)
+ RETURN(rc >= 0 ? -EINVAL : rc);
+
+ if (cfs_bitmap_check_empty(bitmap))
+ lad->lad_incomplete = 0;
+ else
+ lad->lad_incomplete = 1;
+
+ RETURN(0);
+}
+
+/**
* \retval +ve: the lfsck_namespace is broken, the caller should reset it.
* \retval 0: succeed.
* \retval -ve: failed cases.
}
static int lfsck_namespace_store(const struct lu_env *env,
- struct lfsck_component *com, bool init)
+ struct lfsck_component *com)
{
- struct dt_object *obj = com->lc_obj;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct thandle *handle;
- int len = com->lc_file_size;
- int rc;
+ struct dt_object *obj = com->lc_obj;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ cfs_bitmap_t *bitmap = NULL;
+ struct thandle *handle;
+ __u32 nbits = 0;
+ int len = com->lc_file_size;
+ int rc;
ENTRY;
+ if (lad != NULL) {
+ bitmap = lad->lad_bitmap;
+ nbits = bitmap->size;
+
+ LASSERT(nbits > 0);
+ LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
+ }
+
+ ns->ln_bitmap_size = nbits;
lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
- (struct lfsck_namespace *)com->lc_file_ram);
+ ns);
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle))
GOTO(log, rc = PTR_ERR(handle));
if (rc != 0)
GOTO(out, rc);
+ if (bitmap != NULL) {
+ rc = dt_declare_xattr_set(env, obj,
+ lfsck_buf_get(env, bitmap->data, nbits >> 3),
+ XATTR_NAME_LFSCK_BITMAP, 0, handle);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_xattr_set(env, obj,
lfsck_buf_get(env, com->lc_file_disk, len),
- XATTR_NAME_LFSCK_NAMESPACE,
- init ? LU_XATTR_CREATE : LU_XATTR_REPLACE,
- handle, BYPASS_CAPA);
+ XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
+ if (rc == 0 && bitmap != NULL)
+ rc = dt_xattr_set(env, obj,
+ lfsck_buf_get(env, bitmap->data, nbits >> 3),
+ XATTR_NAME_LFSCK_BITMAP, 0, handle,
+ BYPASS_CAPA);
GOTO(out, rc);
ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
ns->ln_status = LS_INIT;
down_write(&com->lc_sem);
- rc = lfsck_namespace_store(env, com, true);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
return rc;
}
/**
- * Update the namespace LFSCK tracing file for the given @fid
+ * Update the namespace LFSCK trace file for the given @fid
*
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
* \param[in] fid the fid which flags to be updated in the lfsck
- * tracing file
+ * trace file
* \param[in] add true if add new flags, otherwise remove flags
*
* \retval 0 for succeed or nothing to be done
dt_trans_stop(env, dev, th);
CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
- "tracing file, flags %x, old %x, new %x: rc = %d\n",
+ "trace file, flags %x, old %x, new %x: rc = %d\n",
lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
(__u32)flags, (__u32)old, (__u32)new, rc);
* type "O": The MDT-object has no linkEA, and there is no name
* entry that references the MDT-object.
*
+ * type "S": The orphan MDT-object is a shard of a striped directory
+ *
* \see lfsck_layout_recreate_parent() for more types.
*
* The orphan name will be like:
struct lu_name *cname = &info->lti_name;
struct dt_insert_rec *rec = &info->lti_dt_rec;
struct lu_fid *tfid = &info->lti_fid5;
+ struct lu_attr *la = &info->lti_la3;
const struct lu_fid *cfid = lfsck_dto2fid(orphan);
const struct lu_fid *pfid;
struct lfsck_instance *lfsck = com->lc_lfsck;
}
}
+ memset(la, 0, sizeof(*la));
+ la->la_ctime = cfs_time_current_sec();
+ la->la_valid = LA_CTIME;
+ rc = dt_declare_attr_set(env, orphan, la, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
rc = dt_trans_start_local(env, dev, th);
if (rc != 0)
GOTO(stop, rc);
}
}
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
+
GOTO(stop, rc = (rc == 0 ? 1 : rc));
unlock:
return rc;
}
-static int lfsck_namespace_insert_normal(const struct lu_env *env,
- struct lfsck_component *com,
- struct dt_object *parent,
- struct dt_object *child,
- const char *name)
-{
- /* XXX: TBD */
- return 0;
-}
-
-static int lfsck_namespace_create_orphan(const struct lu_env *env,
- struct lfsck_component *com,
- struct dt_object *orphan)
-{
- /* XXX: TBD */
- return 0;
-}
-
/**
- * Remove the specified entry from the linkEA.
+ * Add the specified name entry back to namespace.
*
- * Locate the linkEA entry with the given @cname and @pfid, then
- * remove this entry or the other entries those are repeated with
- * this entry.
+ * If there is a linkEA entry that back references a name entry under
+ * some parent directory, but such parent directory does not have the
+ * claimed name entry. On the other hand, the linkEA entries count is
+ * not larger than the MDT-object's hard link count. Under such case,
+ * it is quite possible that the name entry is lost. Then the LFSCK
+ * should 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] obj pointer to the dt_object to be handled
- * \param[in,out]ldata pointer to the buffer that holds the linkEA
- * \param[in] cname the name for the child in the parent directory
- * \param[in] pfid the parent directory's FID for the linkEA
- * \param[in] next if true, then remove the first found linkEA
- * entry, and move the ldata->ld_lee to next entry
+ * \param[in] parent pointer to the directory under which the name entry
+ * will be inserted into
+ * \param[in] child pointer to the object referenced by the name entry
+ * that to be inserted into the parent
+ * \param[in] name the name for the child in the parent directory
*
* \retval positive number for repaired cases
* \retval 0 if nothing to be repaired
* \retval negative error number on failure
*/
-static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
+static int lfsck_namespace_insert_normal(const struct lu_env *env,
struct lfsck_component *com,
- struct dt_object *obj,
- struct linkea_data *ldata,
- struct lu_name *cname,
- struct lu_fid *pfid,
- bool next)
+ struct dt_object *parent,
+ struct dt_object *child,
+ const char *name)
{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct dt_device *dev = lfsck->li_bottom;
- struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct thandle *th = NULL;
- struct lustre_handle lh = { 0 };
- struct linkea_data ldata_new = { 0 };
- struct lu_buf linkea_buf;
- int rc = 0;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ struct lustre_handle lh = { 0 };
+ int rc = 0;
ENTRY;
- rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
- MDS_INODELOCK_UPDATE |
- MDS_INODELOCK_XATTR, LCK_EX);
+ if (unlikely(!dt_try_as_dir(env, parent)))
+ GOTO(log, rc = -ENOTDIR);
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(log, rc = 1);
+
+ /* Hold update lock on the parent to prevent others to access. */
+ rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
+ MDS_INODELOCK_UPDATE, LCK_EX);
if (rc != 0)
GOTO(log, rc);
- if (next)
- linkea_del_buf(ldata, cname);
- else
- lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
- true);
- lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
- ldata->ld_leh->leh_len);
-
-again:
th = dt_trans_create(env, dev);
if (IS_ERR(th))
- GOTO(unlock1, rc = PTR_ERR(th));
+ GOTO(unlock, rc = PTR_ERR(th));
- rc = dt_declare_xattr_set(env, obj, &linkea_buf,
- XATTR_NAME_LINK, 0, th);
+ rec->rec_type = lfsck_object_type(child) & S_IFMT;
+ rec->rec_fid = lfsck_dto2fid(child);
+ rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th);
if (rc != 0)
GOTO(stop, rc);
- rc = dt_trans_start_local(env, dev, th);
+ if (S_ISDIR(rec->rec_type)) {
+ rc = dt_declare_ref_add(env, parent, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ memset(la, 0, sizeof(*la));
+ la->la_ctime = cfs_time_current_sec();
+ la->la_valid = LA_CTIME;
+ rc = dt_declare_attr_set(env, parent, la, th);
if (rc != 0)
GOTO(stop, rc);
- dt_write_lock(env, obj, 0);
- if (unlikely(lfsck_is_dead_obj(obj)))
- GOTO(unlock2, rc = -ENOENT);
-
- rc = lfsck_links_read2(env, obj, &ldata_new);
+ rc = dt_declare_attr_set(env, child, la, th);
if (rc != 0)
- GOTO(unlock2, rc);
+ GOTO(stop, rc);
- /* The specified linkEA entry has been removed by race. */
- rc = linkea_links_find(&ldata_new, cname, pfid);
+ rc = dt_trans_start_local(env, dev, th);
if (rc != 0)
- GOTO(unlock2, rc = 0);
-
- if (bk->lb_param & LPF_DRYRUN)
- GOTO(unlock2, rc = 1);
+ GOTO(stop, rc);
- if (next)
- linkea_del_buf(&ldata_new, cname);
- else
- lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
- true);
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+ if (rc != 0)
+ GOTO(stop, rc);
- if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
- dt_write_unlock(env, obj);
- dt_trans_stop(env, dev, th);
- lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
- ldata_new.ld_leh->leh_len);
- goto again;
+ if (S_ISDIR(rec->rec_type)) {
+ dt_write_lock(env, parent, 0);
+ rc = dt_ref_add(env, parent, th);
+ dt_write_unlock(env, parent);
+ if (rc != 0)
+ GOTO(stop, rc);
}
- lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
- ldata_new.ld_leh->leh_len);
- rc = dt_xattr_set(env, obj, &linkea_buf,
- XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
+ la->la_ctime = cfs_time_current_sec();
+ rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(stop, rc);
- GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
+ rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
-unlock2:
- dt_write_unlock(env, obj);
+ GOTO(stop, rc = (rc == 0 ? 1 : rc));
stop:
dt_trans_stop(env, dev, th);
-unlock1:
+unlock:
lfsck_ibits_unlock(&lh, LCK_EX);
log:
- CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
- "for the object: "DFID", parent "DFID", name %.*s\n",
- lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
- PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
- cname->ln_name);
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
+ "the name %s and type %o to the parent "DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
+ lfsck_object_type(child) & S_IFMT,
+ PFID(lfsck_dto2fid(parent)), rc);
if (rc != 0) {
struct lfsck_namespace *ns = com->lc_file_ram;
ns->ln_flags |= LF_INCONSISTENT;
+ if (rc > 0)
+ ns->ln_lost_dirent_repaired++;
}
return rc;
}
/**
- * Conditionally remove the specified entry from the linkEA.
+ * Create the specified orphan MDT-object on remote MDT.
*
- * Take the parent lock firstly, then check whether the specified
- * name entry exists or not: if yes, do nothing; otherwise, call
- * lfsck_namespace_shrink_linkea() to remove the linkea entry.
+ * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
+ * ask the remote LFSCK instance to create the specified orphan object
+ * under .lustre/lost+found/MDTxxxx/ directory with the name:
+ * ${FID}-P-${conflict_version}.
*
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
- * \param[in] parent pointer to the parent directory
- * \param[in] child pointer to the child object that holds the linkEA
- * \param[in,out]ldata pointer to the buffer that holds the linkEA
- * \param[in] cname the name for the child in the parent directory
- * \param[in] pfid the parent directory's FID for the linkEA
+ * \param[in] orphan pointer to the orphan MDT-object
+ * \param[in] type the orphan's type to be created
+ *
+ * type "P": The orphan object to be created was a parent directory
+ * of some MDT-object which linkEA shows that the @orphan
+ * object is missing.
+ *
+ * \see lfsck_layout_recreate_parent() for more types.
+ *
+ * \param[in] lmv pointer to master LMV EA that will be set to the orphan
*
* \retval positive number for repaired cases
- * \retval 0 if nothing to be repaired
+ * \retval 0 if needs to repair nothing
* \retval negative error number on failure
*/
-static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
- struct lfsck_component *com,
- struct dt_object *parent,
- struct dt_object *child,
- struct linkea_data *ldata,
- struct lu_name *cname,
- struct lu_fid *pfid)
+static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *orphan,
+ __u32 type,
+ struct lmv_mds_md_v1 *lmv)
{
- struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
- struct lustre_handle lh = { 0 };
- int rc;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_request *lr = &info->lti_lr;
+ struct lu_seq_range *range = &info->lti_range;
+ const struct lu_fid *fid = lfsck_dto2fid(orphan);
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct seq_server_site *ss =
+ lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
+ struct lfsck_tgt_desc *ltd = NULL;
+ struct ptlrpc_request *req = NULL;
+ int rc;
ENTRY;
- rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
- MDS_INODELOCK_UPDATE, LCK_EX);
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(out, rc = 1);
+
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
if (rc != 0)
- RETURN(rc);
+ GOTO(out, rc);
- dt_read_lock(env, parent, 0);
- if (unlikely(lfsck_is_dead_obj(parent))) {
- dt_read_unlock(env, parent);
- lfsck_ibits_unlock(&lh, LCK_EX);
- rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
- cname, pfid, true);
+ ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
+ if (ltd == NULL) {
+ ns->ln_flags |= LF_INCOMPLETE;
- RETURN(rc);
+ GOTO(out, rc = -ENODEV);
}
- rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
- (const struct dt_key *)cname->ln_name,
- BYPASS_CAPA);
- dt_read_unlock(env, parent);
+ req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
+ &RQF_LFSCK_NOTIFY);
+ if (req == NULL)
+ GOTO(out, rc = -ENOMEM);
- /* It is safe to release the ldlm lock, because when the logic come
- * here, we have got all the needed information above whether the
- * linkEA entry is valid or not. It is not important that others
- * may add new linkEA entry after the ldlm lock released. If other
- * has removed the specified linkEA entry by race, then it is OK,
- * because the subsequent lfsck_namespace_shrink_linkea() can handle
- * such case. */
- lfsck_ibits_unlock(&lh, LCK_EX);
- if (rc == -ENOENT) {
- rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
- cname, pfid, true);
+ rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
+ if (rc != 0) {
+ ptlrpc_request_free(req);
- RETURN(rc);
+ GOTO(out, rc);
+ }
+
+ lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
+ memset(lr, 0, sizeof(*lr));
+ lr->lr_event = LE_CREATE_ORPHAN;
+ lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
+ lr->lr_active = LFSCK_TYPE_NAMESPACE;
+ lr->lr_fid = *fid;
+ lr->lr_type = type;
+ if (lmv != NULL) {
+ lr->lr_hash_type = lmv->lmv_hash_type;
+ lr->lr_stripe_count = lmv->lmv_stripe_count;
+ lr->lr_layout_version = lmv->lmv_layout_version;
+ memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
+ sizeof(lr->lr_pool_name));
+ }
+
+ ptlrpc_request_set_replen(req);
+ rc = ptlrpc_queue_wait(req);
+ ptlrpc_req_finished(req);
+
+ if (rc == 0)
+ rc = 1;
+ else if (rc == -EEXIST)
+ rc = 0;
+
+ GOTO(out, rc);
+
+out:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
+ DFID" on the MDT %x remotely: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(fid),
+ ltd != NULL ? ltd->ltd_index : -1, rc);
+
+ if (ltd != NULL)
+ lfsck_tgt_put(ltd);
+
+ return rc;
+}
+
+/**
+ * Create the specified orphan MDT-object locally.
+ *
+ * For the case that the parent MDT-object stored in some MDT-object's
+ * linkEA entry is lost, the LFSCK will re-create the parent object as
+ * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
+ * with the name ${FID}-P-${conflict_version}.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] orphan pointer to the orphan MDT-object to be created
+ * \param[in] type the orphan's type to be created
+ *
+ * type "P": The orphan object to be created was a parent directory
+ * of some MDT-object which linkEA shows that the @orphan
+ * object is missing.
+ *
+ * \see lfsck_layout_recreate_parent() for more types.
+ *
+ * \param[in] lmv pointer to master LMV EA that will be set to the orphan
+ *
+ * \retval positive number for repaired cases
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *orphan,
+ __u32 type,
+ struct lmv_mds_md_v1 *lmv)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_allocation_hint *hint = &info->lti_hint;
+ struct dt_object_format *dof = &info->lti_dof;
+ struct lu_name *cname = &info->lti_name2;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lu_fid *tfid = &info->lti_fid;
+ struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
+ const struct lu_fid *cfid = lfsck_dto2fid(orphan);
+ const struct lu_fid *pfid;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck->li_bottom;
+ struct dt_object *parent = NULL;
+ struct dt_object *child = NULL;
+ struct thandle *th = NULL;
+ struct lustre_handle lh = { 0 };
+ struct linkea_data ldata = { 0 };
+ struct lu_buf linkea_buf;
+ struct lu_buf lmv_buf;
+ char name[32];
+ int namelen;
+ int idx = 0;
+ int rc = 0;
+ ENTRY;
+
+ LASSERT(!dt_object_exists(orphan));
+ LASSERT(!dt_object_remote(orphan));
+
+ /* @orphan maybe not attached to lfsck->li_bottom */
+ child = lfsck_object_find_by_dev(env, dev, cfid);
+ if (IS_ERR(child))
+ GOTO(log, rc = PTR_ERR(child));
+
+ cname->ln_name = NULL;
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(log, rc = 1);
+
+ /* Create .lustre/lost+found/MDTxxxx when needed. */
+ if (unlikely(lfsck->li_lpf_obj == NULL)) {
+ rc = lfsck_create_lpf(env, lfsck);
+ if (rc != 0)
+ GOTO(log, rc);
+ }
+
+ parent = lfsck->li_lpf_obj;
+ pfid = lfsck_dto2fid(parent);
+
+ /* Hold update lock on the parent to prevent others to access. */
+ rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
+ MDS_INODELOCK_UPDATE, LCK_EX);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ do {
+ namelen = snprintf(name, 31, DFID"-P-%d",
+ PFID(cfid), idx++);
+ rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+ (const struct dt_key *)name, BYPASS_CAPA);
+ if (rc != 0 && rc != -ENOENT)
+ GOTO(unlock1, rc);
+ } while (rc == 0);
+
+ cname->ln_name = name;
+ cname->ln_namelen = namelen;
+
+ memset(la, 0, sizeof(*la));
+ la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
+ la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
+ LA_ATIME | LA_MTIME | LA_CTIME;
+
+ child->do_ops->do_ah_init(env, hint, parent, child,
+ la->la_mode & S_IFMT);
+
+ memset(dof, 0, sizeof(*dof));
+ dof->dof_type = dt_mode_to_dft(type);
+
+ rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
+ if (rc != 0)
+ GOTO(unlock1, rc);
+
+ rc = linkea_add_buf(&ldata, cname, pfid);
+ if (rc != 0)
+ GOTO(unlock1, rc);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(unlock1, rc = PTR_ERR(th));
+
+ rc = dt_declare_create(env, child, la, hint, dof, th);
+ if (rc == 0 && S_ISDIR(type))
+ rc = dt_declare_ref_add(env, child, th);
+
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (lmv != NULL) {
+ lmv->lmv_magic = LMV_MAGIC;
+ lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
+ lfsck_lmv_header_cpu_to_le(lmv2, lmv);
+ lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
+ rc = dt_declare_xattr_set(env, child, &lmv_buf,
+ XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
+ rc = dt_declare_xattr_set(env, child, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_type = type;
+ rec->rec_fid = cfid;
+ rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th);
+ if (rc == 0 && S_ISDIR(type))
+ rc = dt_declare_ref_add(env, parent, th);
+
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, child, 0);
+ rc = dt_create(env, child, la, hint, dof, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ if (S_ISDIR(type)) {
+ if (unlikely(!dt_try_as_dir(env, child)))
+ GOTO(unlock2, rc = -ENOTDIR);
+
+ rec->rec_type = S_IFDIR;
+ rec->rec_fid = cfid;
+ rc = dt_insert(env, child, (const struct dt_rec *)rec,
+ (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ rec->rec_fid = pfid;
+ rc = dt_insert(env, child, (const struct dt_rec *)rec,
+ (const struct dt_key *)dotdot, th,
+ BYPASS_CAPA, 1);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ rc = dt_ref_add(env, child, th);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+ }
+
+ if (lmv != NULL) {
+ rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
+ th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+ }
+
+ rc = dt_xattr_set(env, child, &linkea_buf,
+ XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
+ dt_write_unlock(env, child);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rec->rec_type = type;
+ rec->rec_fid = cfid;
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)name, th, BYPASS_CAPA, 1);
+ if (rc == 0 && S_ISDIR(type)) {
+ dt_write_lock(env, parent, 0);
+ rc = dt_ref_add(env, parent, th);
+ dt_write_unlock(env, parent);
+ }
+
+ GOTO(stop, rc = (rc == 0 ? 1 : rc));
+
+unlock2:
+ dt_write_unlock(env, child);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+unlock1:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
+ "the object "DFID", name = %s, type %o: rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(cfid),
+ cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
+
+ if (child != NULL && !IS_ERR(child))
+ lfsck_object_put(env, child);
+
+ return rc;
+}
+
+/**
+ * Create the specified orphan MDT-object.
+ *
+ * For the case that the parent MDT-object stored in some MDT-object's
+ * linkEA entry is lost, the LFSCK will re-create the parent object as
+ * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
+ * with the name: ${FID}-P-${conflict_version}.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] orphan pointer to the orphan MDT-object
+ *
+ * type "P": The orphan object to be created was a parent directory
+ * of some MDT-object which linkEA shows that the @orphan
+ * object is missing.
+ *
+ * \see lfsck_layout_recreate_parent() for more types.
+ *
+ * \param[in] lmv pointer to master LMV EA that will be set to the orphan
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if needs to repair nothing
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_create_orphan(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *orphan,
+ struct lmv_mds_md_v1 *lmv)
+{
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ int rc;
+
+ if (dt_object_remote(orphan))
+ rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
+ S_IFDIR, lmv);
+ else
+ rc = lfsck_namespace_create_orphan_local(env, com, orphan,
+ S_IFDIR, lmv);
+
+ if (rc != 0)
+ ns->ln_flags |= LF_INCONSISTENT;
+
+ return rc;
+}
+
+/**
+ * Remove the specified entry from the linkEA.
+ *
+ * Locate the linkEA entry with the given @cname and @pfid, then
+ * remove this entry or the other entries those are repeated with
+ * this entry.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ * \param[in,out]ldata pointer to the buffer that holds the linkEA
+ * \param[in] cname the name for the child in the parent directory
+ * \param[in] pfid the parent directory's FID for the linkEA
+ * \param[in] next if true, then remove the first found linkEA
+ * entry, and move the ldata->ld_lee to next entry
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj,
+ struct linkea_data *ldata,
+ struct lu_name *cname,
+ struct lu_fid *pfid,
+ bool next)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck->li_bottom;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct thandle *th = NULL;
+ struct lustre_handle lh = { 0 };
+ struct linkea_data ldata_new = { 0 };
+ struct lu_buf linkea_buf;
+ int rc = 0;
+ ENTRY;
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc != 0)
+ GOTO(log, rc);
+
+ if (next)
+ linkea_del_buf(ldata, cname);
+ else
+ lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
+ true);
+ lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
+ ldata->ld_leh->leh_len);
+
+again:
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(unlock1, rc = PTR_ERR(th));
+
+ rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ if (unlikely(lfsck_is_dead_obj(obj)))
+ GOTO(unlock2, rc = -ENOENT);
+
+ rc = lfsck_links_read2(env, obj, &ldata_new);
+ if (rc != 0)
+ GOTO(unlock2, rc);
+
+ /* The specified linkEA entry has been removed by race. */
+ rc = linkea_links_find(&ldata_new, cname, pfid);
+ if (rc != 0)
+ GOTO(unlock2, rc = 0);
+
+ if (bk->lb_param & LPF_DRYRUN)
+ GOTO(unlock2, rc = 1);
+
+ if (next)
+ linkea_del_buf(&ldata_new, cname);
+ else
+ lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
+ true);
+
+ if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
+ dt_write_unlock(env, obj);
+ dt_trans_stop(env, dev, th);
+ lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
+ ldata_new.ld_leh->leh_len);
+ goto again;
+ }
+
+ lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
+ ldata_new.ld_leh->leh_len);
+ rc = dt_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
+
+ GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
+
+unlock2:
+ dt_write_unlock(env, obj);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+unlock1:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+
+log:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
+ "for the object: "DFID", parent "DFID", name %.*s\n",
+ lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
+ PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
+ cname->ln_name);
+
+ if (rc != 0) {
+ struct lfsck_namespace *ns = com->lc_file_ram;
+
+ ns->ln_flags |= LF_INCONSISTENT;
+ }
+
+ return rc;
+}
+
+/**
+ * Conditionally remove the specified entry from the linkEA.
+ *
+ * Take the parent lock firstly, then check whether the specified
+ * name entry exists or not: if yes, do nothing; otherwise, call
+ * lfsck_namespace_shrink_linkea() to remove the linkea entry.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] parent pointer to the parent directory
+ * \param[in] child pointer to the child object that holds the linkEA
+ * \param[in,out]ldata pointer to the buffer that holds the linkEA
+ * \param[in] cname the name for the child in the parent directory
+ * \param[in] pfid the parent directory's FID for the linkEA
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *parent,
+ struct dt_object *child,
+ struct linkea_data *ldata,
+ struct lu_name *cname,
+ struct lu_fid *pfid)
+{
+ struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
+ struct lustre_handle lh = { 0 };
+ int rc;
+ ENTRY;
+
+ rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
+ MDS_INODELOCK_UPDATE, LCK_EX);
+ if (rc != 0)
+ RETURN(rc);
+
+ dt_read_lock(env, parent, 0);
+ if (unlikely(lfsck_is_dead_obj(parent))) {
+ dt_read_unlock(env, parent);
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
+ cname, pfid, true);
+
+ RETURN(rc);
+ }
+
+ rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
+ (const struct dt_key *)cname->ln_name,
+ BYPASS_CAPA);
+ dt_read_unlock(env, parent);
+
+ /* It is safe to release the ldlm lock, because when the logic come
+ * here, we have got all the needed information above whether the
+ * linkEA entry is valid or not. It is not important that others
+ * may add new linkEA entry after the ldlm lock released. If other
+ * has removed the specified linkEA entry by race, then it is OK,
+ * because the subsequent lfsck_namespace_shrink_linkea() can handle
+ * such case. */
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ if (rc == -ENOENT) {
+ rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
+ cname, pfid, true);
+
+ RETURN(rc);
}
if (rc != 0)
if (rc != 0)
GOTO(unlock2, rc);
- /* Someone has removed the bad name entry and reuse it for other
+ /* Someone has removed the bad name entry and reused it for other
* object by race. */
if (!lu_fid_eq(tfid, cfid))
GOTO(unlock2, rc = 0);
dt_trans_stop(env, dev, th);
/* We are not sure whether the child will become orphan or not.
- * Record it in the LFSCK tracing file for further checking in
+ * Record it in the LFSCK trace file for further checking in
* the second-stage scanning. */
if (!update && !dec && rc == 0)
lfsck_namespace_trace_update(env, com, cfid,
enum lfsck_namespace_inconsistency_type *type)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
int rc;
ENTRY;
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. */
struct lu_name *cname = &info->lti_name;
const struct lu_fid *cfid = lfsck_dto2fid(child);
struct lu_fid *tfid = &info->lti_fid3;
+ 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;
* 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;
}
lfsck_ibits_unlock(lh, LCK_EX);
+
+lost_parent:
+ lmv = &info->lti_lmv;
+ rc = lfsck_read_stripe_lmv(env, child, lmv);
+ if (rc != 0 && rc != -ENODATA)
+ GOTO(out, rc);
+
+ if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
+ lmv = NULL;
+ } else if (lfsck_shard_name_to_index(env,
+ cname->ln_name, cname->ln_namelen,
+ S_IFDIR, cfid) < 0) {
+ /* It is an invalid name entry, we
+ * cannot trust the parent also. */
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ ldata, cname, tfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com, child,
+ info->lti_tmpbuf, "S", NULL);
+
+ GOTO(out, rc);
+ }
+
/* Create the lost parent as an orphan. */
- rc = lfsck_namespace_create_orphan(env, com, parent);
- if (rc >= 0)
+ rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
+ if (rc >= 0) {
/* Add the missing name entry to the parent. */
rc = lfsck_namespace_insert_normal(env, com, parent,
child, cname->ln_name);
+ if (unlikely(rc == -EEXIST)) {
+ /* Unfortunately, someone reused the name
+ * under the parent by race. So we have
+ * to remove the linkEA entry from
+ * current child object. It means that the
+ * LFSCK cannot recover the system
+ * totally back to its original status,
+ * but it is necessary to make the
+ * current system to be consistent. */
+ rc = lfsck_namespace_shrink_linkea(env,
+ com, child, ldata,
+ cname, tfid, true);
+ if (rc >= 0) {
+ snprintf(info->lti_tmpbuf,
+ sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env,
+ com, child, info->lti_tmpbuf,
+ "D", NULL);
+ }
+ }
+ }
GOTO(out, rc);
}
rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
(const struct dt_key *)cname->ln_name, BYPASS_CAPA);
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;
}
lfsck_ibits_unlock(lh, LCK_EX);
+ rc = lfsck_namespace_check_name(env, parent, child, cname);
+ if (rc == -ENOENT)
+ goto lost_parent;
+
+ if (rc < 0)
+ GOTO(out, rc);
+
+ /* It is an invalid name entry, drop it. */
+ if (unlikely(rc > 0)) {
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ ldata, cname, tfid, true);
+ if (rc >= 0) {
+ snprintf(info->lti_tmpbuf,
+ sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com,
+ child, info->lti_tmpbuf, "D", NULL);
+ }
+
+ GOTO(out, rc);
+ }
+
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
cname->ln_name);
+ if (unlikely(rc == -ESTALE))
+ /* It may happen when the remote object has been
+ * removed, but the local MDT is not aware of that. */
+ goto lost_parent;
+
+ if (unlikely(rc == -EEXIST)) {
+ /* Unfortunately, someone reused the name under the
+ * parent by race. So we have to remove the linkEA
+ * entry from current child object. It means that the
+ * LFSCK cannot recover the system totally back to
+ * its original status, but it is necessary to make
+ * the current system to be consistent.
+ *
+ * It also may be because of the LFSCK found some
+ * internal status of create operation. Under such
+ * case, nothing to be done. */
+ rc = lfsck_namespace_shrink_linkea_cond(env, com,
+ parent, child, ldata, cname, tfid);
+ if (rc >= 0) {
+ snprintf(info->lti_tmpbuf,
+ sizeof(info->lti_tmpbuf),
+ "-"DFID, PFID(pfid));
+ rc = lfsck_namespace_insert_orphan(env, com,
+ child, info->lti_tmpbuf, "D", NULL);
+ }
+ }
GOTO(out, rc);
}
linkea_del_buf(ldata, cname);
}
- ns->ln_dirent_repaired += count;
+ ns->ln_dirent_repaired += count;
+
+ RETURN(rc);
+ }
+
+ lfsck_ibits_unlock(lh, LCK_EX);
+ /* The name entry references another MDT-object that may be
+ * created by the LFSCK for repairing dangling name entry.
+ * Try to replace it. */
+ rc = lfsck_namespace_replace_cond(env, com, parent, child,
+ tfid, cname);
+ lfsck_object_put(env, parent);
+ if (rc < 0)
+ RETURN(rc);
+
+ if (rc > 0)
+ goto rebuild;
+
+ linkea_del_buf(ldata, cname);
+ }
+
+ if (ldata->ld_leh->leh_reccount == 1) {
+ rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
+ lh, type, NULL);
+
+ 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);
+ }
+
+ linkea_first_entry(ldata);
+ /* If the dangling name entry for the orphan directory object has
+ * been remvoed, then just check whether the directory object is
+ * still under the .lustre/lost+found/MDTxxxx/ or not. */
+ if (lpf) {
+ lpf = false;
+ goto again;
+ }
+
+ /* There is no linkEA entry that matches the ".." name entry. Find
+ * the first linkEA entry that both parent and name entry exist to
+ * rebuild a new ".." name entry. */
+ if (once) {
+ once = false;
+ goto again;
+ }
+
+ RETURN(rc);
+}
+
+/**
+ * Repair the object's nlink attribute.
+ *
+ * If all the known name entries have been verified, then the object's hard
+ * link attribute should match the object's linkEA entries count unless the
+ * object's has too much hard link to be recorded in the linkEA. Such cases
+ * should have been marked in the LFSCK trace file. Otherwise, trust the
+ * linkEA to update the object's nlink attribute.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ * \param[in,out] nlink pointer to buffer to object's hard lock count 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, __u32 *nlink)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la3;
+ struct lu_fid *tfid = &info->lti_fid3;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_device *dev = lfsck->li_bottom;
+ const struct lu_fid *cfid = lfsck_dto2fid(obj);
+ struct dt_object *child = NULL;
+ struct thandle *th = NULL;
+ struct linkea_data ldata = { 0 };
+ struct lustre_handle lh = { 0 };
+ __u32 old = *nlink;
+ int rc = 0;
+ __u8 flags;
+ ENTRY;
+
+ LASSERT(!dt_object_remote(obj));
+ LASSERT(S_ISREG(lfsck_object_type(obj)));
+
+ child = lfsck_object_find_by_dev(env, dev, cfid);
+ if (IS_ERR(child))
+ GOTO(log, rc = PTR_ERR(child));
+
+ rc = lfsck_ibits_lock(env, lfsck, child, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ 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, child, 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, child, 0);
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
+ * ever tried to verify some remote MDT-object that resides on this
+ * MDT, but this MDT failed to respond such request. So means there
+ * may be some remote name entry on other MDT that references this
+ * object with another name, so we cannot know whether this linkEA
+ * is valid or not. So keep it there and maybe resolved when next
+ * LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(unlock, rc = 0);
+
+ fid_cpu_to_be(tfid, cfid);
+ rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
+ (const struct dt_key *)tfid, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ if (flags & LNTF_SKIP_NLINK)
+ GOTO(unlock, rc = 0);
+
+ rc = lfsck_links_read2(env, child, &ldata);
+ if (rc == -ENODATA)
+ GOTO(unlock, rc = 0);
- RETURN(rc);
- }
+ if (rc != 0)
+ GOTO(unlock, rc);
- lfsck_ibits_unlock(lh, LCK_EX);
- /* The name entry references another MDT-object that may be
- * created by the LFSCK for repairing dangling name entry.
- * Try to replace it. */
- rc = lfsck_namespace_replace_cond(env, com, parent, child,
- tfid, cname);
- lfsck_object_put(env, parent);
- if (rc < 0)
- RETURN(rc);
+ if (*nlink == ldata.ld_leh->leh_reccount)
+ GOTO(unlock, rc = 0);
- if (rc > 0)
- goto rebuild;
+ la->la_nlink = *nlink = ldata.ld_leh->leh_reccount;
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(unlock, rc = 1);
- linkea_del_buf(ldata, cname);
- }
+ rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
- if (ldata->ld_leh->leh_reccount == 1) {
- rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
- lh, type, NULL);
+ GOTO(unlock, rc = (rc == 0 ? 1 : rc));
- RETURN(rc);
- }
+unlock:
+ dt_write_unlock(env, child);
- /* 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);
+stop:
+ dt_trans_stop(env, dev, th);
- RETURN(rc);
- }
+log:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ if (child != NULL && !IS_ERR(child))
+ lfsck_object_put(env, child);
- linkea_first_entry(ldata);
- /* 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;
- }
+ 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, *nlink, rc);
- /* 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;
- }
+ if (rc != 0)
+ ns->ln_flags |= LF_INCONSISTENT;
- RETURN(rc);
+ return rc;
}
/**
LASSERT(!dt_object_remote(child));
- if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
+ 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"
* 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,
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
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(env, com,
- parent);
+ parent, NULL);
if (rc < 0) {
lfsck_object_put(env, parent);
/* Add the missing name entry to the parent. */
rc = lfsck_namespace_insert_normal(env, com,
parent, child, cname->ln_name);
- linkea_next_entry(&ldata);
+ 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);
continue;
}
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some
+ * MDT has ever tried to verify some remote MDT-object that
+ * resides on this MDT, but this MDT failed to respond such
+ * request. So means there may be some remote name entry on
+ * other MDT that references this object with another name,
+ * so we cannot know whether this linkEA is valid or not.
+ * So keep it there and maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc = 0);
+ }
+
+ rc = lfsck_namespace_check_name(env, parent, child, cname);
+ if (rc == -ENOENT)
+ goto lost_parent;
+
+ if (rc < 0) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc);
+ }
+
+ /* It is an invalid name entry, drop it. */
+ if (unlikely(rc > 0)) {
+ lfsck_object_put(env, parent);
+ rc = lfsck_namespace_shrink_linkea(env, com, child,
+ &ldata, cname, pfid, true);
+ if (rc < 0)
+ GOTO(out, rc);
+
+ if (rc > 0)
+ repaired = true;
+
+ continue;
+ }
+
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
cname->ln_name);
+ if (unlikely(rc == -ESTALE))
+ /* It may happen when the remote object has been
+ * removed, but the local MDT is not aware of that. */
+ goto lost_parent;
+
+ if (unlikely(rc == -EEXIST))
+ /* Unfortunately, someone reused the name under the
+ * parent by race. So we have to remove the linkEA
+ * entry from current child object. It means that the
+ * LFSCK cannot recover the system totally back to
+ * its original status, but it is necessary to make
+ * the current system to be consistent.
+ *
+ * It also may be because of the LFSCK found some
+ * internal status of create operation. Under such
+ * case, nothing to be done. */
+ rc = lfsck_namespace_shrink_linkea_cond(env, com,
+ parent, child, &ldata, cname, pfid);
+ else
+ linkea_next_entry(&ldata);
+
lfsck_object_put(env, parent);
if (rc < 0)
GOTO(out, rc);
if (rc > 0)
repaired = true;
-
- linkea_next_entry(&ldata);
}
GOTO(out, rc = 0);
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 (count == 0 && !(ns->ln_flags & LF_INCOMPLETE)) {
/* If the child becomes orphan, then insert it into
* the global .lustre/lost+found/MDTxxxx directory. */
rc = lfsck_namespace_insert_orphan(env, com, child, "", "O",
return rc;
if (la->la_nlink != count) {
- /* XXX: there will be other patch(es) for MDT-object
- * hard links verification. */
+ rc = lfsck_namespace_repair_nlink(env, com, child,
+ &la->la_nlink);
+ if (rc > 0) {
+ ns->ln_objs_nlink_repaired++;
+ rc = 0;
+ }
}
if (repaired) {
"dirent_repaired: "LPU64"\n"
"linkea_repaired: "LPU64"\n"
"nlinks_repaired: "LPU64"\n"
- "lost_found: "LPU64"\n"
"multiple_linked_checked: "LPU64"\n"
"multiple_linked_repaired: "LPU64"\n"
"unknown_inconsistency: "LPU64"\n"
"dangling_repaired: "LPU64"\n"
"multiple_referenced_repaired: "LPU64"\n"
"bad_file_type_repaired: "LPU64"\n"
+ "lost_dirent_repaired: "LPU64"\n"
+ "local_lost_found_scanned: "LPU64"\n"
+ "local_lost_found_moved: "LPU64"\n"
+ "local_lost_found_skipped: "LPU64"\n"
+ "local_lost_found_failed: "LPU64"\n"
+ "striped_dirs_scanned: "LPU64"\n"
+ "striped_dirs_repaired: "LPU64"\n"
+ "striped_dirs_failed: "LPU64"\n"
+ "striped_dirs_disabled: "LPU64"\n"
+ "striped_dirs_skipped: "LPU64"\n"
+ "striped_shards_scanned: "LPU64"\n"
+ "striped_shards_repaired: "LPU64"\n"
+ "striped_shards_failed: "LPU64"\n"
+ "striped_shards_skipped: "LPU64"\n"
+ "name_hash_repaired: "LPU64"\n"
"success_count: %u\n"
"run_time_phase1: %u seconds\n"
"run_time_phase2: %u seconds\n",
ns->ln_dirent_repaired,
ns->ln_linkea_repaired,
ns->ln_objs_nlink_repaired,
- ns->ln_objs_lost_found,
ns->ln_mul_linked_checked,
ns->ln_mul_linked_repaired,
ns->ln_unknown_inconsistency,
ns->ln_dangling_repaired,
ns->ln_mul_ref_repaired,
ns->ln_bad_type_repaired,
+ ns->ln_lost_dirent_repaired,
+ ns->ln_local_lpf_scanned,
+ ns->ln_local_lpf_moved,
+ ns->ln_local_lpf_skipped,
+ ns->ln_local_lpf_failed,
+ ns->ln_striped_dirs_scanned,
+ ns->ln_striped_dirs_repaired,
+ ns->ln_striped_dirs_failed,
+ ns->ln_striped_dirs_disabled,
+ ns->ln_striped_dirs_skipped,
+ ns->ln_striped_shards_scanned,
+ ns->ln_striped_shards_repaired,
+ ns->ln_striped_shards_failed,
+ ns->ln_striped_shards_skipped,
+ ns->ln_name_hash_repaired,
ns->ln_success_count,
time_phase1,
time_phase2);
static 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 dt_object *root;
- struct dt_object *dto;
- int rc;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct dt_object *root;
+ struct dt_object *dto;
+ int rc;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (rc != 0)
GOTO(out, rc);
- rc = lfsck_namespace_store(env, com, true);
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(lad->lad_bitmap);
+
+ rc = lfsck_namespace_store(env, com);
GOTO(out, rc);
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_instance *lfsck = com->lc_lfsck;
+ struct lfsck_lmv *llmv = lfsck->li_lmv;
+ struct lfsck_namespace_req *lnr;
+ __u32 size =
+ sizeof(*lnr) + LFSCK_TMPBUF_LEN;
+ bool wakeup = false;
+ ENTRY;
+
+ if (llmv == NULL)
+ RETURN_EXIT;
+
+ OBD_ALLOC(lnr, size);
+ if (lnr == NULL) {
+ ns->ln_striped_dirs_skipped++;
+
+ RETURN_EXIT;
+ }
+
+ /* 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_obj = lfsck_object_get(lfsck->li_obj_dir);
+ lnr->lnr_lmv = lfsck_lmv_get(llmv);
+ lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
+ lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
+ lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
+ lnr->lnr_size = size;
+
+ spin_lock(&lad->lad_lock);
+ if (lad->lad_assistant_status < 0) {
+ spin_unlock(&lad->lad_lock);
+ lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
+ ns->ln_striped_dirs_skipped++;
+
+ RETURN_EXIT;
+ }
+
+ list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
+ if (lad->lad_prefetched == 0)
+ wakeup = true;
+
+ lad->lad_prefetched++;
+ spin_unlock(&lad->lad_lock);
+ if (wakeup)
+ wake_up_all(&lad->lad_thread.t_ctl_waitq);
+
+ EXIT;
+}
+
+static int lfsck_namespace_open_dir(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_lmv *llmv = lfsck->li_lmv;
+ int rc = 0;
+ ENTRY;
+
+ if (llmv == NULL)
+ RETURN(0);
+
+ if (llmv->ll_lmv_master) {
+ struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
+
+ if (lmv->lmv_master_mdt_index !=
+ lfsck_dev_idx(lfsck->li_bottom)) {
+ lmv->lmv_master_mdt_index =
+ lfsck_dev_idx(lfsck->li_bottom);
+ 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)
{
com->lc_new_checked = 0;
}
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
log:
struct lfsck_position *pos = &com->lc_pos_start;
int rc;
- if (ns->ln_status == LS_COMPLETED) {
+ 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);
ns->ln_objs_repaired_phase2 = 0;
ns->ln_objs_failed_phase2 = 0;
ns->ln_objs_nlink_repaired = 0;
- ns->ln_objs_lost_found = 0;
ns->ln_dirent_repaired = 0;
ns->ln_linkea_repaired = 0;
ns->ln_mul_linked_checked = 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,
/* 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 tracing file to be too large. */
+ * Because it may cause the LFSCK trace file to be too large. */
if (rc == -ENODATA) {
if (S_ISDIR(lfsck_object_type(obj)))
GOTO(out, rc = 0);
if (rc != 0)
GOTO(out, rc);
- if (la->la_nlink > 1)
+ /* "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 (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
rc = lfsck_namespace_trace_update(env, com, fid,
LNTF_CHECK_LINKEA, true);
if (rc != 0)
GOTO(out, rc);
- if (la->la_nlink > 1)
+ /* "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 (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
rc = lfsck_namespace_trace_update(env, com,
fid, LNTF_CHECK_LINKEA, true);
}
list_del_init(&com->lc_link_dir);
list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
} else if (result == 0) {
- ns->ln_status = lfsck->li_status;
- if (ns->ln_status == 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);
com->lc_new_checked = 0;
}
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
static int lfsck_namespace_double_scan(const struct lu_env *env,
struct lfsck_component *com)
{
- struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct lfsck_tgt_desc *next;
+ int rc;
+
+ rc = lfsck_double_scan_generic(env, com, ns->ln_status);
+ if (thread_is_stopped(&lad->lad_thread)) {
+ LASSERT(list_empty(&lad->lad_req_list));
+ LASSERT(list_empty(&lad->lad_mdt_phase1_list));
+
+ spin_lock(<ds->ltd_lock);
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ spin_unlock(<ds->ltd_lock);
+ }
- return lfsck_double_scan_generic(env, com, ns->ln_status);
+ return rc;
}
static void lfsck_namespace_data_release(const struct lu_env *env,
}
spin_unlock(<ds->ltd_lock);
- CFS_FREE_BITMAP(lad->lad_bitmap);
+ if (likely(lad->lad_bitmap != NULL))
+ CFS_FREE_BITMAP(lad->lad_bitmap);
OBD_FREE_PTR(lad);
}
+static void lfsck_namespace_quit(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct lfsck_tgt_desc *next;
+
+ LASSERT(lad != NULL);
+
+ lfsck_quit_generic(env, com);
+
+ LASSERT(thread_is_init(&lad->lad_thread) ||
+ thread_is_stopped(&lad->lad_thread));
+ LASSERT(list_empty(&lad->lad_req_list));
+
+ spin_lock(<ds->ltd_lock);
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
+ ltd_namespace_phase_list) {
+ list_del_init(<d->ltd_namespace_phase_list);
+ }
+ spin_unlock(<ds->ltd_lock);
+}
+
static int lfsck_namespace_in_notify(const struct lu_env *env,
struct lfsck_component *com,
- struct lfsck_request *lr)
+ struct lfsck_request *lr,
+ struct thandle *th)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_assistant_data *lad = com->lc_data;
struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
struct lfsck_tgt_desc *ltd;
+ int rc;
bool fail = false;
ENTRY;
- if (lr->lr_event != LE_PHASE1_DONE &&
- lr->lr_event != LE_PHASE2_DONE &&
- lr->lr_event != LE_PEER_EXIT)
+ switch (lr->lr_event) {
+ case LE_CREATE_ORPHAN: {
+ struct dt_object *orphan = NULL;
+ struct lmv_mds_md_v1 *lmv;
+
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK handling notify from "
+ "MDT %x to create orphan"DFID" with type %o\n",
+ lfsck_lfsck2name(lfsck), lr->lr_index,
+ PFID(&lr->lr_fid), lr->lr_type);
+
+ orphan = lfsck_object_find(env, lfsck, &lr->lr_fid);
+ if (IS_ERR(orphan))
+ GOTO(out_create, rc = PTR_ERR(orphan));
+
+ if (dt_object_exists(orphan))
+ GOTO(out_create, rc = -EEXIST);
+
+ if (lr->lr_stripe_count > 0) {
+ lmv = &lfsck_env_info(env)->lti_lmv;
+ memset(lmv, 0, sizeof(*lmv));
+ lmv->lmv_hash_type = lr->lr_hash_type;
+ lmv->lmv_stripe_count = lr->lr_stripe_count;
+ lmv->lmv_layout_version = lr->lr_layout_version;
+ memcpy(lmv->lmv_pool_name, lr->lr_pool_name,
+ sizeof(lmv->lmv_pool_name));
+ } else {
+ lmv = NULL;
+ }
+
+ rc = lfsck_namespace_create_orphan_local(env, com, orphan,
+ lr->lr_type, lmv);
+
+ GOTO(out_create, rc = (rc == 1) ? 0 : rc);
+
+out_create:
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK handled notify from "
+ "MDT %x to create orphan"DFID" with type %o: rc = %d\n",
+ lfsck_lfsck2name(lfsck), lr->lr_index,
+ PFID(&lr->lr_fid), lr->lr_type, rc);
+
+ if (orphan != NULL && !IS_ERR(orphan))
+ lfsck_object_put(env, orphan);
+
+ return rc;
+ }
+ case LE_SKIP_NLINK_DECLARE: {
+ struct dt_object *obj = com->lc_obj;
+ struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
+ __u8 flags = 0;
+
+ LASSERT(th != NULL);
+
+ rc = dt_declare_delete(env, obj,
+ (const struct dt_key *)key, th);
+ if (rc == 0)
+ rc = dt_declare_insert(env, obj,
+ (const struct dt_rec *)&flags,
+ (const struct dt_key *)key, th);
+
+ RETURN(rc);
+ }
+ case LE_SKIP_NLINK: {
+ struct dt_object *obj = com->lc_obj;
+ struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
+ __u8 flags = 0;
+ bool exist = false;
+ ENTRY;
+
+ LASSERT(th != NULL);
+
+ fid_cpu_to_be(key, &lr->lr_fid);
+ rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
+ (const struct dt_key *)key, BYPASS_CAPA);
+ if (rc == 0) {
+ if (flags & LNTF_SKIP_NLINK)
+ RETURN(0);
+
+ exist = true;
+ } else if (rc != -ENOENT) {
+ GOTO(log, rc);
+ }
+
+ flags |= LNTF_SKIP_NLINK;
+ if (exist) {
+ rc = dt_delete(env, obj, (const struct dt_key *)key,
+ th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(log, rc);
+ }
+
+ rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
+ (const struct dt_key *)key, th, BYPASS_CAPA, 1);
+
+ GOTO(log, rc);
+
+log:
+ CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
+ " to be skipped for namespace double scan: rc = %d\n",
+ lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
+
+ if (rc != 0)
+ /* If we cannot record this object in the LFSCK tracing,
+ * we have to mark the LFSC as LF_INCOMPLETE, then the
+ * LFSCK will skip nlink attribute verification for
+ * all objects. */
+ ns->ln_flags |= LF_INCOMPLETE;
+
+ return 0;
+ }
+ case LE_PHASE1_DONE:
+ case LE_PHASE2_DONE:
+ case LE_PEER_EXIT:
+ break;
+ default:
RETURN(-EINVAL);
+ }
CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
- "status %d\n", lfsck_lfsck2name(lfsck), lr->lr_event,
- lr->lr_index, lr->lr_status);
+ "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
+ lr->lr_index, lr->lr_status, lr->lr_flags2);
spin_lock(<ds->ltd_lock);
ltd = LTD_TGT(ltds, lr->lr_index);
break;
}
+ if (lr->lr_flags2 & LF_INCOMPLETE)
+ ns->ln_flags |= LF_INCOMPLETE;
+
if (list_empty(<d->ltd_namespace_list))
list_add_tail(<d->ltd_namespace_list,
&lad->lad_mdt_list);
static struct lfsck_operations lfsck_namespace_ops = {
.lfsck_reset = lfsck_namespace_reset,
.lfsck_fail = lfsck_namespace_fail,
+ .lfsck_close_dir = lfsck_namespace_close_dir,
+ .lfsck_open_dir = lfsck_namespace_open_dir,
.lfsck_checkpoint = lfsck_namespace_checkpoint,
.lfsck_prep = lfsck_namespace_prep,
.lfsck_exec_oit = lfsck_namespace_exec_oit,
.lfsck_dump = lfsck_namespace_dump,
.lfsck_double_scan = lfsck_namespace_double_scan,
.lfsck_data_release = lfsck_namespace_data_release,
- .lfsck_quit = lfsck_quit_generic,
+ .lfsck_quit = lfsck_namespace_quit,
.lfsck_in_notify = lfsck_namespace_in_notify,
.lfsck_query = lfsck_namespace_query,
};
* and the users can make the decision about how to handle it with
* more human knownledge. (by default)
*
- * 2) Re-create the missed MDT-object with the FID information.
+ * 2) Re-create the missing MDT-object with the FID information.
*
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
struct dt_allocation_hint *hint = &info->lti_hint;
struct dt_object_format *dof = &info->lti_dof;
struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
struct dt_object *parent = lnr->lnr_obj;
const struct lu_name *cname;
struct linkea_data ldata = { 0 };
struct lustre_handle lh = { 0 };
struct lu_buf linkea_buf;
+ struct lu_buf lmv_buf;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct dt_device *dev = lfsck_obj2dt_dev(child);
rc = dt_declare_ref_add(env, child, th);
if (rc != 0)
GOTO(stop, rc);
+
+ /* 5a. generate slave LMV EA. */
+ if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
+ int idx;
+
+ idx = lfsck_shard_name_to_index(env,
+ lnr->lnr_name, lnr->lnr_namelen,
+ type, lfsck_dto2fid(child));
+ if (unlikely(idx < 0))
+ GOTO(stop, rc = idx);
+
+ *lmv2 = lnr->lnr_lmv->ll_lmv;
+ lmv2->lmv_magic = LMV_MAGIC_STRIPE;
+ lmv2->lmv_master_mdt_index = idx;
+
+ lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
+ lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
+ rc = dt_declare_xattr_set(env, child, &lmv_buf,
+ XATTR_NAME_LMV, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
}
- /* 5a. insert linkEA for child */
+ /* 6a. insert linkEA for child */
lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
ldata.ld_leh->leh_len);
rc = dt_declare_xattr_set(env, child, &linkea_buf,
rc = dt_ref_add(env, child, th);
if (rc != 0)
GOTO(unlock, rc);
+
+ /* 5b. generate slave LMV EA. */
+ if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
+ rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
+ 0, th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(unlock, rc);
+ }
}
- /* 5b. insert linkEA for child. */
+ /* 6b. insert linkEA for child. */
rc = dt_xattr_set(env, child, &linkea_buf,
XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
struct dt_object *dir = lnr->lnr_obj;
struct dt_object *obj = NULL;
const struct lu_fid *pfid = lfsck_dto2fid(dir);
- struct dt_device *dev;
+ struct dt_device *dev = NULL;
struct lustre_handle lh = { 0 };
bool repaired = false;
bool dtlocked = false;
bool remove;
bool newdata;
bool log = false;
- int idx;
+ bool bad_hash = false;
+ int idx = 0;
int count = 0;
int rc;
enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
GOTO(out, rc);
}
+ if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
+ rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
+
+ RETURN(rc);
+ }
+
if (lnr->lnr_name[0] == '.' &&
(lnr->lnr_namelen == 1 || fid_seq_is_dot(fid_seq(&lnr->lnr_fid))))
GOTO(out, rc = 0);
CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
"did not join the namespace LFSCK\n",
lfsck_lfsck2name(lfsck), idx);
- ns->ln_flags |= LF_INCOMPLETE;
+ lfsck_lad_set_bitmap(env, com, idx);
GOTO(out, rc = -ENODEV);
}
dangling:
rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
if (rc == 0) {
+ if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
+ lnr->lnr_name, lnr->lnr_namelen)) {
+ type = LNIT_BAD_DIRENT;
+
+ GOTO(out, rc);
+ }
+
type = LNIT_DANGLING;
rc = lfsck_namespace_repair_dangling(env, com,
obj, lnr);
ns->ln_flags |= LF_INCONSISTENT;
+ /* If the name entry hash does not match the slave striped
+ * directory, and the name entry does not match also, then
+ * it is quite possible that name entry is corrupted. */
+ if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
+ lnr->lnr_name, lnr->lnr_namelen)) {
+ type = LNIT_BAD_DIRENT;
+
+ GOTO(stop, rc = 0);
+ }
+
/* If the file type stored in the name entry does not match
* the file type claimed by the object, and the object does
* not recognize the name entry, then it is quite possible
GOTO(stop, rc);
rc = lfsck_links_write(env, obj, &ldata, handle);
+ if (unlikely(rc == -ENOSPC) &&
+ S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
+ if (handle != NULL) {
+ LASSERT(dt_write_locked(env, obj));
+
+ dt_write_unlock(env, obj);
+ dtlocked = false;
+
+ dt_trans_stop(env, dev, handle);
+ handle = NULL;
+
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ }
+
+ rc = lfsck_namespace_trace_update(env, com,
+ &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
+ if (rc != 0)
+ /* If we cannot record this object in the
+ * LFSCK tracing, we have to mark the LFSCK
+ * as LF_INCOMPLETE, then the LFSCK will
+ * skip nlink attribute verification for
+ * all objects. */
+ ns->ln_flags |= LF_INCOMPLETE;
+
+ GOTO(out, rc = 0);
+ }
+
if (rc != 0)
GOTO(stop, rc);
out:
lfsck_ibits_unlock(&lh, LCK_EX);
+ if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
+ !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
+ lnr->lnr_name, lnr->lnr_namelen) &&
+ type != LNIT_BAD_DIRENT) {
+ ns->ln_flags |= LF_INCONSISTENT;
+
+ log = false;
+ rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
+ lnr->lnr_lmv, lnr->lnr_name);
+ if (rc >= 0)
+ bad_hash = true;
+ }
+
if (rc >= 0) {
switch (type) {
case LNIT_BAD_TYPE:
lnr->lnr_namelen, lnr->lnr_name, rc);
lfsck_namespace_record_failure(env, lfsck, ns);
+ if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
+ rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
+ rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
+ dev != NULL && dev != lfsck->li_next)
+ lfsck_lad_set_bitmap(env, com, idx);
+
if (!(bk->lb_param & LPF_FAILOUT))
rc = 0;
} else {
false);
}
+ if (bad_hash) {
+ ns->ln_name_hash_repaired++;
+
+ /* Not count repeatedly. */
+ if (!repaired)
+ ns->ln_items_repaired++;
+
+ if (bk->lb_param & LPF_DRYRUN &&
+ lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
+ lfsck_pos_fill(env, lfsck,
+ &ns->ln_pos_first_inconsistent,
+ false);
+ }
+
+
rc = 0;
}
up_write(&com->lc_sem);
if (obj != NULL && !IS_ERR(obj))
lfsck_object_put(env, obj);
+
+ return rc;
+}
+
+/**
+ * Handle one orphan under the backend /lost+found directory
+ *
+ * Insert the orphan FID into the namespace LFSCK trace file for further
+ * processing (via the subsequent namespace LFSCK second-stage scanning).
+ * At the same time, remove the orphan name entry from backend /lost+found
+ * directory. There is an interval between the orphan name entry removed
+ * from the backend /lost+found directory and the orphan FID in the LFSCK
+ * trace file handled. In such interval, the LFSCK can be reset, then
+ * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
+ * To guarantee that the orphans can be found when LFSCK run next time
+ * without e2fsck again, when remove the orphan name entry, the LFSCK
+ * will set the orphan's ctime attribute as 1. Since normal applications
+ * cannot change the object's ctime attribute as 1. Then when LFSCK run
+ * next time, it can record the object (that ctime is 1) in the namespace
+ * LFSCK trace file during the first-stage scanning.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] parent pointer to the object for the backend /lost+found
+ * \param[in] ent pointer to the name entry for the target under the
+ * backend /lost+found
+ *
+ * \retval positive for repaired
+ * \retval 0 if needs to repair nothing
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *parent,
+ struct lu_dirent *ent)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_fid *key = &info->lti_fid;
+ struct lu_attr *la = &info->lti_la;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct dt_object *obj = com->lc_obj;
+ struct dt_device *dev = lfsck->li_bottom;
+ struct dt_object *child = NULL;
+ struct thandle *th = NULL;
+ int rc = 0;
+ __u8 flags = 0;
+ bool exist = false;
+ ENTRY;
+
+ child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
+ if (IS_ERR(child))
+ RETURN(PTR_ERR(child));
+
+ LASSERT(dt_object_exists(child));
+ LASSERT(!dt_object_remote(child));
+
+ fid_cpu_to_be(key, &ent->lde_fid);
+ rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
+ (const struct dt_key *)key, BYPASS_CAPA);
+ if (rc == 0) {
+ exist = true;
+ flags |= LNTF_CHECK_ORPHAN;
+ } else if (rc == -ENOENT) {
+ flags = LNTF_CHECK_ORPHAN;
+ } else {
+ GOTO(out, rc);
+ }
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(out, rc = PTR_ERR(th));
+
+ /* a1. remove name entry from backend /lost+found */
+ rc = dt_declare_delete(env, parent,
+ (const struct dt_key *)ent->lde_name, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (S_ISDIR(lfsck_object_type(child))) {
+ /* a2. decrease parent's nlink */
+ rc = dt_declare_ref_del(env, parent, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ if (exist) {
+ /* a3. remove child's FID from the LFSCK trace file. */
+ rc = dt_declare_delete(env, obj,
+ (const struct dt_key *)key, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ } else {
+ /* a4. set child's ctime as 1 */
+ memset(la, 0, sizeof(*la));
+ la->la_ctime = 1;
+ la->la_valid = LA_CTIME;
+ rc = dt_declare_attr_set(env, child, la, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ /* a5. insert child's FID into the LFSCK trace file. */
+ rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
+ (const struct dt_key *)key, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ /* b1. remove name entry from backend /lost+found */
+ rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th,
+ BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ if (S_ISDIR(lfsck_object_type(child))) {
+ /* b2. decrease parent's nlink */
+ dt_write_lock(env, parent, 0);
+ rc = dt_ref_del(env, parent, th);
+ dt_write_unlock(env, parent);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ if (exist) {
+ /* a3. remove child's FID from the LFSCK trace file. */
+ rc = dt_delete(env, obj, (const struct dt_key *)key, th,
+ BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(stop, rc);
+ } else {
+ /* b4. set child's ctime as 1 */
+ rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(stop, rc);
+ }
+
+ /* b5. insert child's FID into the LFSCK trace file. */
+ rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
+ (const struct dt_key *)key, th, BYPASS_CAPA, 1);
+
+ GOTO(stop, rc = (rc == 0 ? 1 : rc));
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+out:
+ lu_object_put(env, &child->do_lu);
+
return rc;
}
+/**
+ * Handle orphans under the backend /lost+found directory
+ *
+ * Some backend checker, such as e2fsck for ldiskfs may find some orphans
+ * and put them under the backend /lost+found directory that is invisible
+ * to client. The LFSCK will scan such directory, for the original client
+ * visible orphans, add their fids into the namespace LFSCK trace file,
+ * then the subsenquent namespace LFSCK second-stage scanning can handle
+ * them as other objects to be double scanned: either move back to normal
+ * namespace, or to the global visible orphan directory:
+ * /ROOT/.lustre/lost+found/MDTxxxx/
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ */
+static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_dirent *ent =
+ (struct lu_dirent *)info->lti_key;
+ struct lu_seq_range *range = &info->lti_range;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct ptlrpc_thread *thread = &lfsck->li_thread;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct dt_device *dev = lfsck->li_bottom;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct dt_object *parent;
+ const struct dt_it_ops *iops;
+ struct dt_it *di;
+ struct seq_server_site *ss =
+ lu_site2seq(dev->dd_lu_dev.ld_site);
+ __u64 cookie;
+ int rc = 0;
+ __u16 type;
+ ENTRY;
+
+ parent = lfsck_object_find_by_dev(env, dev, &LU_BACKEND_LPF_FID);
+ if (IS_ERR(parent)) {
+ CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
+ lfsck_lfsck2name(lfsck), PTR_ERR(parent));
+ RETURN_EXIT;
+ }
+
+ /* It is normal that the /lost+found does not exist for ZFS backend. */
+ if (!dt_object_exists(parent))
+ GOTO(out, rc = 0);
+
+ if (unlikely(!dt_try_as_dir(env, parent)))
+ GOTO(out, rc = -ENOTDIR);
+
+ CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
+ lfsck_lfsck2name(lfsck));
+
+ com->lc_new_scanned = 0;
+ iops = &parent->do_index_ops->dio_it;
+ di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE, BYPASS_CAPA);
+ if (IS_ERR(di))
+ GOTO(out, rc = PTR_ERR(di));
+
+ rc = iops->load(env, di, 0);
+ if (rc == 0)
+ rc = iops->next(env, di);
+ else if (rc > 0)
+ rc = 0;
+
+ while (rc == 0) {
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
+ cfs_fail_val > 0) {
+ struct l_wait_info lwi;
+
+ lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
+ NULL, NULL);
+ l_wait_event(thread->t_ctl_waitq,
+ !thread_is_running(thread),
+ &lwi);
+
+ if (unlikely(!thread_is_running(thread)))
+ break;
+ }
+
+ rc = iops->rec(env, di, (struct dt_rec *)ent,
+ LUDA_64BITHASH | LUDA_TYPE);
+ if (rc == 0)
+ rc = lfsck_unpack_ent(ent, &cookie, &type);
+
+ if (unlikely(rc != 0)) {
+ CDEBUG(D_LFSCK, "%s: fail to iterate backend "
+ "/lost+found: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+
+ goto skip;
+ }
+
+ /* skip dot and dotdot entries */
+ if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
+ goto next;
+
+ if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
+ goto skip;
+
+ if (fid_is_norm(&ent->lde_fid)) {
+ fld_range_set_mdt(range);
+ rc = fld_local_lookup(env, ss->ss_server_fld,
+ fid_seq(&ent->lde_fid), range);
+ if (rc != 0)
+ goto skip;
+ } else if (lfsck_dev_idx(dev) != 0) {
+ /* If the returned FID is IGIF, then there are three
+ * possible cases:
+ *
+ * 1) The object is upgraded from old Lustre-1.8 with
+ * IGIF assigned to such object.
+ * 2) The object is a backend local object and is
+ * invisible to client.
+ * 3) The object lost its LMV EA, and since there is
+ * no FID-in-dirent for the orphan in the backend
+ * /lost+found directory, then the low layer will
+ * return IGIF for such object.
+ *
+ * For MDTx (x != 0), it is either case 2) or case 3),
+ * but from the LFSCK view, they are indistinguishable.
+ * To be safe, the LFSCK will keep it there and report
+ * some message, then the adminstrator can handle that
+ * furtherly.
+ *
+ * For MDT0, it is more possible the case 1). The LFSCK
+ * will handle the orphan as an upgraded object. */
+ CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
+ "in the backend /lost+found on the MDT %04x, "
+ "to be safe, skip it.\n",
+ lfsck_lfsck2name(lfsck), ent->lde_namelen,
+ ent->lde_name, PFID(&ent->lde_fid),
+ lfsck_dev_idx(dev));
+ goto skip;
+ }
+
+ rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
+
+skip:
+ down_write(&com->lc_sem);
+ com->lc_new_scanned++;
+ ns->ln_local_lpf_scanned++;
+ if (rc > 0)
+ ns->ln_local_lpf_moved++;
+ else if (rc == 0)
+ ns->ln_local_lpf_skipped++;
+ else
+ ns->ln_local_lpf_failed++;
+ up_write(&com->lc_sem);
+
+ if (rc < 0 && bk->lb_param & LPF_FAILOUT)
+ break;
+
+next:
+ lfsck_control_speed_by_self(com);
+ if (unlikely(!thread_is_running(thread))) {
+ rc = 0;
+ break;
+ }
+
+ rc = iops->next(env, di);
+ }
+
+ iops->put(env, di);
+ iops->fini(env, di);
+
+ EXIT;
+
+out:
+ CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+
+ lu_object_put(env, &parent->do_lu);
+}
+
static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
struct lfsck_component *com)
{
CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
lfsck_lfsck2name(lfsck));
+ lfsck_namespace_scan_local_lpf(env, com);
+
com->lc_new_checked = 0;
com->lc_new_scanned = 0;
com->lc_time_last_checkpoint = cfs_time_current();
goto checkpoint;
}
- target = lfsck_object_find(env, lfsck, &fid);
+ target = lfsck_object_find_by_dev(env, lfsck->li_bottom, &fid);
if (IS_ERR(target)) {
rc = PTR_ERR(target);
goto checkpoint;
ns->ln_time_last_checkpoint = cfs_time_current_sec();
ns->ln_objs_checked_phase2 += com->lc_new_checked;
com->lc_new_checked = 0;
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
if (rc != 0)
GOTO(put, rc);
ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
ns->ln_success_count++;
} else if (rc == 0) {
- ns->ln_status = lfsck->li_status;
- if (ns->ln_status == 0)
+ if (lfsck->li_status != 0)
+ ns->ln_status = lfsck->li_status;
+ else
ns->ln_status = LS_STOPPED;
} else {
ns->ln_status = LS_FAILED;
}
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
return rc;
}
+static int
+lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
+ struct ptlrpc_request *req,
+ void *args, int rc)
+{
+ return 0;
+}
+
+/**
+ * Notify remote LFSCK instances about former failures.
+ *
+ * The local LFSCK instance has recorded which MDTs have ever failed to respond
+ * some LFSCK verification requests (maybe because of network issues or the MDT
+ * itself trouble). During the respond gap the MDT may missed some name entries
+ * verification, then the MDT cannot know whether related MDT-objects have been
+ * referenced by related name entries or not, then in the second-stage scanning,
+ * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
+ * linkEA for back reference, then it will misguide the LFSCK to generate wrong
+ * name entry for repairing the orphan.
+ *
+ * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
+ * it will scan the bitmap for the ever failed MDTs, and notify them that they
+ * have ever missed some name entries verification and should skip the handling
+ * for orphan MDT-objects.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] lr pointer to the lfsck request
+ */
static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_request *lr)
{
- /* XXX: TBD */
+ struct lfsck_async_interpret_args *laia =
+ &lfsck_env_info(env)->lti_laia2;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct ptlrpc_request_set *set;
+ int rc = 0;
+ ENTRY;
+
+ set = ptlrpc_prep_set();
+ if (set == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
+ memset(laia, 0, sizeof(*laia));
+ lad->lad_touch_gen++;
+
+ spin_lock(<ds->ltd_lock);
+ while (!list_empty(&lad->lad_mdt_list)) {
+ ltd = list_entry(lad->lad_mdt_list.next,
+ struct lfsck_tgt_desc,
+ ltd_namespace_list);
+ if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
+ break;
+
+ ltd->ltd_namespace_gen = lad->lad_touch_gen;
+ list_move_tail(<d->ltd_namespace_list,
+ &lad->lad_mdt_list);
+ if (!lad->lad_incomplete ||
+ !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
+ ltd->ltd_namespace_failed = 0;
+ continue;
+ }
+
+ ltd->ltd_namespace_failed = 1;
+ spin_unlock(<ds->ltd_lock);
+ rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
+ lfsck_namespace_assistant_sync_failures_interpret,
+ laia, LFSCK_NOTIFY);
+ if (rc != 0)
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
+ "to sync failure with MDT %x: rc = %d\n",
+ lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
+
+ spin_lock(<ds->ltd_lock);
+ }
+ spin_unlock(<ds->ltd_lock);
+
+ rc = ptlrpc_set_wait(set);
+ ptlrpc_set_destroy(set);
+
+ GOTO(out, rc);
+
+out:
+ if (rc != 0)
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
+ "to sync failure with MDTs, and related MDTs "
+ "may handle orphan improperly: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+
+ EXIT;
}
struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
git://git.whamcloud.com / fs / lustre-release.git / blobdiff