int lfsck_bookmark_setup(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct dt_object *root;
struct dt_object *obj;
int rc;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
if (unlikely(!dt_try_as_dir(env, root))) {
lfsck_object_put(env, root);
RETURN(-ENOTDIR);
}
obj = local_file_find_or_create(env, lfsck->li_los, root,
LFSCK_BOOKMARK,
S_IFREG | S_IRUGO | S_IWUSR);
lfsck_object_put(env, root);
if (IS_ERR(obj))
RETURN(PTR_ERR(obj));
lfsck->li_bookmark_obj = obj;
rc = lfsck_bookmark_load(env, lfsck);
if (rc == 0) {
struct lfsck_bookmark *mb = &lfsck->li_bookmark_ram;
/* It is upgraded from old release, set it as
* LFSCK_ASYNC_WIN_DEFAULT to avoid memory pressure. */
if (unlikely(mb->lb_async_windows == 0)) {
mb->lb_async_windows = LFSCK_ASYNC_WIN_DEFAULT;
mutex_lock(&lfsck->li_mutex);
rc = lfsck_bookmark_store(env, lfsck);
mutex_unlock(&lfsck->li_mutex);
}
} else if (rc == -ENODATA) {
rc = lfsck_bookmark_init(env, lfsck);
}
RETURN(rc);
}
static void lfsck_close_dir(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct dt_object *dir_obj = lfsck->li_obj_dir;
const struct dt_it_ops *dir_iops = &dir_obj->do_index_ops->dio_it;
struct dt_it *dir_di = lfsck->li_di_dir;
lfsck_di_dir_put(env, lfsck);
dir_iops->fini(env, dir_di);
lfsck->li_obj_dir = NULL;
lfsck_object_put(env, dir_obj);
}
static int lfsck_master_oit_engine(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
const struct dt_it_ops *iops =
&lfsck->li_obj_oit->do_index_ops->dio_it;
struct dt_it *di = lfsck->li_di_oit;
struct lu_fid *fid = &info->lti_fid;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct ptlrpc_thread *thread = &lfsck->li_thread;
__u32 idx =
lfsck_dev_idx(lfsck->li_bottom);
int rc;
ENTRY;
do {
struct dt_object *target;
bool update_lma = false;
if (lfsck->li_di_dir != NULL) {
rc = lfsck_master_dir_engine(env, lfsck);
if (rc <= 0)
RETURN(rc);
}
if (unlikely(lfsck->li_oit_over))
RETURN(1);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
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 (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
RETURN(0);
lfsck->li_current_oit_processed = 1;
lfsck->li_new_scanned++;
rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
if (rc != 0) {
lfsck_fail(env, lfsck, true);
if (rc < 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
else
goto checkpoint;
}
if (fid_is_idif(fid)) {
__u32 idx1 = fid_idif_ost_idx(fid);
LASSERT(!lfsck->li_master);
/* It is an old format device, update the LMA. */
if (idx != idx1) {
struct ost_id *oi = &info->lti_oi;
fid_to_ostid(fid, oi);
ostid_to_fid(fid, oi, idx);
update_lma = true;
}
} else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
!fid_is_last_id(fid) && !fid_is_root(fid) &&
!fid_seq_is_dot(fid_seq(fid))) {
/* If the FID/object is only used locally and invisible
* to external nodes, then LFSCK will not handle it. */
goto checkpoint;
}
target = lfsck_object_find(env, lfsck, fid);
if (target == NULL) {
goto checkpoint;
} else if (IS_ERR(target)) {
lfsck_fail(env, lfsck, true);
if (bk->lb_param & LPF_FAILOUT)
RETURN(PTR_ERR(target));
else
goto checkpoint;
}
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_exists(target) && !dt_object_remote(target)) {
if (update_lma)
rc = lfsck_update_lma(env, lfsck, target);
if (rc == 0)
rc = lfsck_exec_oit(env, lfsck, target);
}
lfsck_object_put(env, target);
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
checkpoint:
rc = lfsck_checkpoint(env, lfsck);
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
/* Rate control. */
lfsck_control_speed(lfsck);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
spin_lock(&lfsck->li_lock);
thread_set_flags(thread, SVC_STOPPING);
spin_unlock(&lfsck->li_lock);
RETURN(-EINVAL);
}
rc = iops->next(env, di);
if (unlikely(rc > 0))
lfsck->li_oit_over = 1;
else if (likely(rc == 0))
lfsck->li_current_oit_processed = 0;
if (unlikely(!thread_is_running(thread)))
RETURN(0);
} while (rc == 0 || lfsck->li_di_dir != NULL);
RETURN(rc);
}
static int lfsck_master_dir_engine(const struct lu_env *env,
struct lfsck_instance *lfsck)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
const struct dt_it_ops *iops =
&lfsck->li_obj_dir->do_index_ops->dio_it;
struct dt_it *di = lfsck->li_di_dir;
struct lu_dirent *ent = &info->lti_ent;
struct lu_fid *fid = &info->lti_fid;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct ptlrpc_thread *thread = &lfsck->li_thread;
int rc;
ENTRY;
do {
struct dt_object *child;
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY2) &&
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);
}
lfsck->li_new_scanned++;
rc = iops->rec(env, di, (struct dt_rec *)ent,
lfsck->li_args_dir);
lfsck_unpack_ent(ent, &lfsck->li_cookie_dir);
if (rc != 0) {
lfsck_fail(env, lfsck, true);
if (bk->lb_param & LPF_FAILOUT)
RETURN(rc);
else
goto checkpoint;
}
if (ent->lde_attrs & LUDA_IGNORE)
goto checkpoint;
*fid = ent->lde_fid;
child = lfsck_object_find(env, lfsck, fid);
if (child == NULL) {
goto checkpoint;
} else if (IS_ERR(child)) {
lfsck_fail(env, lfsck, true);
if (bk->lb_param & LPF_FAILOUT)
RETURN(PTR_ERR(child));
else
goto checkpoint;
}
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_exists(child) && !dt_object_remote(child))
rc = lfsck_exec_dir(env, lfsck, child, ent);
lfsck_object_put(env, child);
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
checkpoint:
rc = lfsck_checkpoint(env, lfsck);
if (rc != 0 && bk->lb_param & LPF_FAILOUT)
RETURN(rc);
/* Rate control. */
lfsck_control_speed(lfsck);
if (unlikely(!thread_is_running(thread)))
RETURN(0);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
spin_lock(&lfsck->li_lock);
thread_set_flags(thread, SVC_STOPPING);
spin_unlock(&lfsck->li_lock);
RETURN(-EINVAL);
}
rc = iops->next(env, di);
} while (rc == 0);
if (rc > 0 && !lfsck->li_oit_over)
lfsck_close_dir(env, lfsck);
RETURN(rc);
}
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;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
if (unlikely(!dt_try_as_dir(env, root))) {
lu_object_put(env, &root->do_lu);
RETURN(-ENOTDIR);
}
down_write(&com->lc_sem);
if (init) {
memset(ns, 0, sizeof(*ns));
} else {
__u32 count = ns->ln_success_count;
__u64 last_time = ns->ln_time_last_complete;
memset(ns, 0, sizeof(*ns));
ns->ln_success_count = count;
ns->ln_time_last_complete = last_time;
}
ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
ns->ln_status = LS_INIT;
rc = local_object_unlink(env, lfsck->li_bottom, root,
lfsck_namespace_name);
if (rc != 0)
GOTO(out, rc);
lfsck_object_put(env, com->lc_obj);
com->lc_obj = NULL;
dto = local_index_find_or_create(env, lfsck->li_los, root,
lfsck_namespace_name,
S_IFREG | S_IRUGO | S_IWUSR,
&dt_lfsck_features);
if (IS_ERR(dto))
GOTO(out, rc = PTR_ERR(dto));
com->lc_obj = dto;
rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
if (rc != 0)
GOTO(out, rc);
rc = lfsck_namespace_store(env, com, true);
GOTO(out, rc);
out:
up_write(&com->lc_sem);
lu_object_put(env, &root->do_lu);
return rc;
}
/**
* \retval +ve repaired
* \retval 0 no need to repair
* \retval -ve error cases
*/
static int lfsck_namespace_double_scan_one(const struct lu_env *env,
struct lfsck_component *com,
struct dt_object *child, __u8 flags)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_attr *la = &info->lti_la;
struct lu_name *cname = &info->lti_name;
struct lu_fid *pfid = &info->lti_fid;
struct lu_fid *cfid = &info->lti_fid2;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_namespace *ns = com->lc_file_ram;
struct linkea_data ldata = { 0 };
struct thandle *handle = NULL;
bool locked = false;
bool update = false;
int rc;
ENTRY;
if (com->lc_journal) {
again:
LASSERT(!locked);
update = false;
com->lc_journal = 1;
handle = dt_trans_create(env, lfsck->li_next);
if (IS_ERR(handle))
RETURN(rc = PTR_ERR(handle));
rc = dt_declare_xattr_set(env, child,
lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
XATTR_NAME_LINK, 0, handle);
if (rc != 0)
GOTO(stop, rc);
rc = dt_trans_start(env, lfsck->li_next, handle);
if (rc != 0)
GOTO(stop, rc);
dt_write_lock(env, child, MOR_TGT_CHILD);
locked = true;
}
if (unlikely(lfsck_is_dead_obj(child)))
GOTO(stop, rc = 0);
rc = dt_attr_get(env, child, la, BYPASS_CAPA);
if (rc == 0)
rc = lfsck_links_read(env, child, &ldata);
if (rc != 0) {
if ((bk->lb_param & LPF_DRYRUN) &&
(rc == -EINVAL || rc == -ENODATA))
rc = 1;
GOTO(stop, rc);
}
linkea_first_entry(&ldata);
while (ldata.ld_lee != NULL) {
struct dt_object *parent = NULL;
rc = lfsck_linkea_entry_unpack(lfsck, &ldata, cname, pfid);
if (rc > 0)
update = true;
if (!fid_is_sane(pfid))
goto shrink;
parent = lfsck_object_find(env, lfsck, pfid);
if (parent == NULL)
goto shrink;
else if (IS_ERR(parent))
GOTO(stop, rc = PTR_ERR(parent));
if (!dt_object_exists(parent))
goto shrink;
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_remote(parent)) {
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
}
if (unlikely(!dt_try_as_dir(env, parent)))
goto shrink;
/* To guarantee the 'name' is terminated with '0'. */
memcpy(info->lti_key, cname->ln_name, cname->ln_namelen);
info->lti_key[cname->ln_namelen] = 0;
cname->ln_name = info->lti_key;
rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
(const struct dt_key *)cname->ln_name,
BYPASS_CAPA);
if (rc != 0 && rc != -ENOENT) {
lfsck_object_put(env, parent);
GOTO(stop, rc);
}
if (rc == 0) {
if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
}
goto shrink;
}
/* If there is no name entry in the parent dir and the object
* link count is less than the linkea entries count, then the
* linkea entry should be removed. */
if (ldata.ld_leh->leh_reccount > la->la_nlink)
goto shrink;
/* XXX: For the case of there is a linkea entry, but without
* name entry pointing to the object and its hard links
* count is not less than the object name entries count,
* then seems we should add the 'missed' name entry back
* to namespace, but before LFSCK phase III finished, we
* do not know whether the object has some inconsistency
* on other MDTs. So now, do NOT add the name entry back
* to the namespace, but keep the linkEA entry. LU-2914 */
lfsck_object_put(env, parent);
linkea_next_entry(&ldata);
continue;
shrink:
if (parent != NULL)
lfsck_object_put(env, parent);
if (bk->lb_param & LPF_DRYRUN)
RETURN(1);
CDEBUG(D_LFSCK, "Remove linkEA: "DFID"[%.*s], "DFID"\n",
PFID(lfsck_dto2fid(child)), cname->ln_namelen, cname->ln_name,
PFID(pfid));
linkea_del_buf(&ldata, cname);
update = true;
}
if (update) {
if (!com->lc_journal) {
com->lc_journal = 1;
goto again;
}
rc = lfsck_links_write(env, child, &ldata, handle);
}
GOTO(stop, rc);
stop:
if (locked) {
/* XXX: For the case linkea entries count does not match the object hard
* links count, we cannot update the later one simply. Before LFSCK
* phase III finished, we cannot know whether there are some remote
* name entries to be repaired or not. LU-2914 */
if (rc == 0 && !lfsck_is_dead_obj(child) &&
ldata.ld_leh != NULL &&
ldata.ld_leh->leh_reccount != la->la_nlink)
CWARN("%s: the object "DFID" linkEA entry count %u "
"may not match its hardlink count %u\n",
lfsck_lfsck2name(lfsck), PFID(cfid),
ldata.ld_leh->leh_reccount, la->la_nlink);
dt_write_unlock(env, child);
}
if (handle != NULL)
dt_trans_stop(env, lfsck->li_next, handle);
if (rc == 0 && update) {
ns->ln_objs_nlink_repaired++;
rc = 1;
}
return rc;
}
static int lfsck_namespace_double_scan_main(void *args)
{
struct lfsck_thread_args *lta = args;
const struct lu_env *env = <a->lta_env;
struct lfsck_component *com = lta->lta_com;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct ptlrpc_thread *thread = &lfsck->li_thread;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_namespace *ns = com->lc_file_ram;
struct dt_object *obj = com->lc_obj;
const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
struct dt_object *target;
struct dt_it *di;
struct dt_key *key;
struct lu_fid fid;
int rc;
__u8 flags = 0;
ENTRY;
com->lc_new_checked = 0;
com->lc_new_scanned = 0;
com->lc_time_last_checkpoint = cfs_time_current();
com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
di = iops->init(env, obj, 0, BYPASS_CAPA);
if (IS_ERR(di))
GOTO(out, rc = PTR_ERR(di));
fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
rc = iops->get(env, di, (const struct dt_key *)&fid);
if (rc < 0)
GOTO(fini, rc);
/* Skip the start one, which either has been processed or non-exist. */
rc = iops->next(env, di);
if (rc != 0)
GOTO(put, rc);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
GOTO(put, rc = 0);
do {
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);
}
key = iops->key(env, di);
fid_be_to_cpu(&fid, (const struct lu_fid *)key);
target = lfsck_object_find(env, lfsck, &fid);
down_write(&com->lc_sem);
if (target == NULL) {
rc = 0;
goto checkpoint;
} else if (IS_ERR(target)) {
rc = PTR_ERR(target);
goto checkpoint;
}
/* XXX: Currently, skip remote object, the consistency for
* remote object will be processed in LFSCK phase III. */
if (dt_object_exists(target) && !dt_object_remote(target)) {
rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
if (rc == 0)
rc = lfsck_namespace_double_scan_one(env, com,
target, flags);
}
lfsck_object_put(env, target);
checkpoint:
com->lc_new_checked++;
com->lc_new_scanned++;
ns->ln_fid_latest_scanned_phase2 = fid;
if (rc > 0)
ns->ln_objs_repaired_phase2++;
else if (rc < 0)
ns->ln_objs_failed_phase2++;
up_write(&com->lc_sem);
if ((rc == 0) || ((rc > 0) && !(bk->lb_param & LPF_DRYRUN))) {
lfsck_namespace_delete(env, com, &fid);
} else if (rc < 0) {
flags |= LLF_REPAIR_FAILED;
lfsck_namespace_update(env, com, &fid, flags, true);
}
if (rc < 0 && bk->lb_param & LPF_FAILOUT)
GOTO(put, rc);
if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
cfs_time_current())) &&
com->lc_new_checked != 0) {
down_write(&com->lc_sem);
ns->ln_run_time_phase2 +=
cfs_duration_sec(cfs_time_current() +
HALF_SEC - com->lc_time_last_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);
up_write(&com->lc_sem);
if (rc != 0)
GOTO(put, rc);
com->lc_time_last_checkpoint = cfs_time_current();
com->lc_time_next_checkpoint =
com->lc_time_last_checkpoint +
cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
}
lfsck_control_speed_by_self(com);
if (unlikely(!thread_is_running(thread)))
GOTO(put, rc = 0);
rc = iops->next(env, di);
} while (rc == 0);
GOTO(put, rc);
put:
iops->put(env, di);
fini:
iops->fini(env, di);
out:
down_write(&com->lc_sem);
ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
HALF_SEC - lfsck->li_time_last_checkpoint);
ns->ln_time_last_checkpoint = cfs_time_current_sec();
ns->ln_objs_checked_phase2 += com->lc_new_checked;
com->lc_new_checked = 0;
if (rc > 0) {
com->lc_journal = 0;
ns->ln_status = LS_COMPLETED;
if (!(bk->lb_param & LPF_DRYRUN))
ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
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)
ns->ln_status = LS_STOPPED;
} else {
ns->ln_status = LS_FAILED;
}
if (ns->ln_status != LS_PAUSED) {
spin_lock(&lfsck->li_lock);
cfs_list_del_init(&com->lc_link);
cfs_list_add_tail(&com->lc_link, &lfsck->li_list_idle);
spin_unlock(&lfsck->li_lock);
}
rc = lfsck_namespace_store(env, com, false);
up_write(&com->lc_sem);
if (atomic_dec_and_test(&lfsck->li_double_scan_count))
wake_up_all(&thread->t_ctl_waitq);
lfsck_thread_args_fini(lta);
return rc;
}
