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);
}
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);
}
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);
}
}
/**
- * Load the MDT bitmap from the lfsck_namespace tracing file.
+ * 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
}
/**
- * 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:
if (rc != 0)
GOTO(stop, rc);
+ 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);
la->la_ctime = cfs_time_current_sec();
rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
GOTO(stop, rc = (rc == 0 ? 1 : rc));
*
* \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_remote(const struct lu_env *env,
struct lfsck_component *com,
struct dt_object *orphan,
- __u32 type)
+ __u32 type,
+ struct lmv_mds_md_v1 *lmv)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lfsck_request *lr = &info->lti_lr;
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);
*
* \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)
+ __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 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 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;
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,
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);
*
* \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 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);
+ S_IFDIR, lmv);
else
rc = lfsck_namespace_create_orphan_local(env, com, orphan,
- S_IFDIR);
+ S_IFDIR, lmv);
if (rc != 0)
ns->ln_flags |= LF_INCONSISTENT;
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,
struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct dt_object *parent = NULL;
+ struct lmv_mds_md_v1 *lmv;
int rc = 0;
ENTRY;
lfsck_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);
+ 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,
}
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 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 tracing file. Otherwise, trust the
+ * 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
/* 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);
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);
"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_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);
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)
{
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);
}
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",
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);
+ lr->lr_type, lmv);
GOTO(out_create, rc = (rc == 1) ? 0 : rc);
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,
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);
bool remove;
bool newdata;
bool log = false;
+ bool bad_hash = false;
int idx = 0;
int count = 0;
int rc;
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);
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
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:
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);
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;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff