int ofd_trans_start(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *obj, struct thandle *th)
{
struct ofd_thread_info *info = ofd_info(env);
int rc;
if (info->fti_exp == NULL)
return 0;
/* declare last_rcvd update */
rc = dt_declare_record_write(env, ofd->ofd_lut.lut_last_rcvd,
sizeof(struct lsd_client_data),
info->fti_exp->exp_target_data.ted_lr_off,
th);
if (rc)
RETURN(rc);
/* declare last_rcvd header update */
rc = dt_declare_record_write(env, ofd->ofd_lut.lut_last_rcvd,
sizeof(ofd->ofd_lut.lut_lsd), 0, th);
if (rc)
RETURN(rc);
/* version change is required for this object */
if (obj) {
ofd_info(env)->fti_obj = obj;
rc = dt_declare_version_set(env, ofd_object_child(obj), th);
if (rc)
RETURN(rc);
}
return dt_trans_start(env, ofd->ofd_osd, th);
}
static int out_trans_start(const struct lu_env *env,
struct thandle_exec_args *ta)
{
/* Always do sync commit for Phase I */
LASSERT(ta->ta_handle->th_sync != 0);
return dt_trans_start(env, ta->ta_dev, ta->ta_handle);
}
static int out_trans_start(const struct lu_env *env, struct dt_device *dt,
struct thandle *th)
{
/* Always do sync commit for Phase I */
LASSERT(th->th_sync != 0);
return dt_trans_start(env, dt, th);
}
/**
* Start transaction in OFD.
*
* This function updates the given \a obj object version and calls
* dt_trans_start().
*
* \param[in] env execution environment
* \param[in] ofd OFD device
* \param[in] obj OFD object affected by this transaction
* \param[in] th transaction handle
*
* \retval 0 if successful
* \retval negative value in case of error
*/
int ofd_trans_start(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *obj, struct thandle *th)
{
/* version change is required for this object */
if (obj != NULL)
tgt_vbr_obj_set(env, ofd_object_child(obj));
return dt_trans_start(env, ofd->ofd_osd, th);
}
static int lfsck_update_lma(const struct lu_env *env,
struct lfsck_instance *lfsck, struct dt_object *obj)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct dt_device *dt = lfsck->li_bottom;
struct lustre_mdt_attrs *lma = &info->lti_lma;
struct lu_buf *buf;
struct thandle *th;
int fl;
int rc;
ENTRY;
if (bk->lb_param & LPF_DRYRUN)
RETURN(0);
buf = lfsck_buf_get(env, info->lti_lma_old, LMA_OLD_SIZE);
rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMA, BYPASS_CAPA);
if (rc < 0) {
if (rc != -ENODATA)
RETURN(rc);
fl = LU_XATTR_CREATE;
lustre_lma_init(lma, lfsck_dto2fid(obj), LMAC_FID_ON_OST, 0);
} else {
if (rc != LMA_OLD_SIZE && rc != sizeof(struct lustre_mdt_attrs))
RETURN(-EINVAL);
fl = LU_XATTR_REPLACE;
lustre_lma_swab(lma);
lustre_lma_init(lma, lfsck_dto2fid(obj),
lma->lma_compat | LMAC_FID_ON_OST,
lma->lma_incompat);
}
lustre_lma_swab(lma);
th = dt_trans_create(env, dt);
if (IS_ERR(th))
RETURN(PTR_ERR(th));
buf = lfsck_buf_get(env, lma, sizeof(*lma));
rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
if (rc != 0)
GOTO(stop, rc);
rc = dt_trans_start(env, dt, th);
if (rc != 0)
GOTO(stop, rc);
rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th, BYPASS_CAPA);
GOTO(stop, rc);
stop:
dt_trans_stop(env, dt, th);
return rc;
}
static int out_trans_start(const struct lu_env *env,
struct thandle_exec_args *ta)
{
return dt_trans_start(env, ta->ta_handle->th_dev, ta->ta_handle);
}
static int lfsck_namespace_exec_dir(const struct lu_env *env,
struct lfsck_component *com,
struct dt_object *obj,
struct lu_dirent *ent)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_attr *la = &info->lti_la;
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 };
const struct lu_fid *pfid = lfsck_dto2fid(lfsck->li_obj_dir);
const struct lu_fid *cfid = lfsck_dto2fid(obj);
const struct lu_name *cname;
struct thandle *handle = NULL;
bool repaired = false;
bool locked = false;
bool remove;
bool newdata;
int count = 0;
int rc;
ENTRY;
cname = lfsck_name_get_const(env, ent->lde_name, ent->lde_namelen);
down_write(&com->lc_sem);
com->lc_new_checked++;
if (ent->lde_attrs & LUDA_UPGRADE) {
ns->ln_flags |= LF_UPGRADE;
repaired = true;
} else if (ent->lde_attrs & LUDA_REPAIR) {
ns->ln_flags |= LF_INCONSISTENT;
repaired = true;
}
if (ent->lde_name[0] == '.' &&
(ent->lde_namelen == 1 ||
(ent->lde_namelen == 2 && ent->lde_name[1] == '.') ||
fid_is_dot_lustre(&ent->lde_fid)))
GOTO(out, rc = 0);
if (!(bk->lb_param & LPF_DRYRUN) &&
(com->lc_journal || repaired)) {
again:
LASSERT(!locked);
com->lc_journal = 1;
handle = dt_trans_create(env, lfsck->li_next);
if (IS_ERR(handle))
GOTO(out, rc = PTR_ERR(handle));
rc = lfsck_declare_namespace_exec_dir(env, obj, 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, obj, MOR_TGT_CHILD);
locked = true;
}
rc = lfsck_namespace_check_exist(env, lfsck, obj, ent->lde_name);
if (rc != 0)
GOTO(stop, rc);
rc = lfsck_links_read(env, obj, &ldata);
if (rc == 0) {
count = ldata.ld_leh->leh_reccount;
rc = linkea_links_find(&ldata, cname, pfid);
if ((rc == 0) &&
(count == 1 || !S_ISDIR(lfsck_object_type(obj))))
goto record;
ns->ln_flags |= LF_INCONSISTENT;
/* For dir, if there are more than one linkea entries, or the
* linkea entry does not match the name entry, then remove all
* and add the correct one. */
if (S_ISDIR(lfsck_object_type(obj))) {
remove = true;
newdata = true;
} else {
remove = false;
newdata = false;
}
goto nodata;
} else if (unlikely(rc == -EINVAL)) {
count = 1;
ns->ln_flags |= LF_INCONSISTENT;
/* The magic crashed, we are not sure whether there are more
* corrupt data in the linkea, so remove all linkea entries. */
remove = true;
newdata = true;
goto nodata;
} else if (rc == -ENODATA) {
count = 1;
ns->ln_flags |= LF_UPGRADE;
remove = false;
newdata = true;
nodata:
if (bk->lb_param & LPF_DRYRUN) {
repaired = true;
goto record;
}
if (!com->lc_journal)
goto again;
if (remove) {
LASSERT(newdata);
rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
BYPASS_CAPA);
if (rc != 0)
GOTO(stop, rc);
}
if (newdata) {
rc = linkea_data_new(&ldata,
&lfsck_env_info(env)->lti_linkea_buf);
if (rc != 0)
GOTO(stop, rc);
}
rc = linkea_add_buf(&ldata, cname, pfid);
if (rc != 0)
GOTO(stop, rc);
rc = lfsck_links_write(env, obj, &ldata, handle);
if (rc != 0)
GOTO(stop, rc);
count = ldata.ld_leh->leh_reccount;
repaired = true;
} else {
GOTO(stop, rc);
}
record:
LASSERT(count > 0);
rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
if (rc != 0)
GOTO(stop, rc);
if ((count == 1) &&
(la->la_nlink == 1 || S_ISDIR(lfsck_object_type(obj))))
/* Usually, it is for single linked object or dir, do nothing.*/
GOTO(stop, rc);
/* Following modification will be in another transaction. */
if (handle != NULL) {
LASSERT(dt_write_locked(env, obj));
dt_write_unlock(env, obj);
locked = false;
dt_trans_stop(env, lfsck->li_next, handle);
handle = NULL;
}
ns->ln_mlinked_checked++;
rc = lfsck_namespace_update(env, com, cfid,
count != la->la_nlink ? LLF_UNMATCH_NLINKS : 0, false);
GOTO(out, rc);
stop:
if (locked)
dt_write_unlock(env, obj);
if (handle != NULL)
dt_trans_stop(env, lfsck->li_next, handle);
out:
if (rc < 0) {
ns->ln_items_failed++;
if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
lfsck_pos_fill(env, lfsck,
&ns->ln_pos_first_inconsistent, false);
if (!(bk->lb_param & LPF_FAILOUT))
rc = 0;
} else {
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);
} else {
com->lc_journal = 0;
}
rc = 0;
}
up_write(&com->lc_sem);
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;
}
