int lmv_fld_lookup(struct lmv_obd *lmv,
const struct lu_fid *fid,
mdsno_t *mds)
{
int rc;
/* FIXME: Currently ZFS still use local seq for ROOT unfortunately, and
* this fid_is_local check should be removed once LU-2240 is fixed */
LASSERTF((fid_seq_in_fldb(fid_seq(fid)) ||
fid_seq_is_local_file(fid_seq(fid))) &&
fid_is_sane(fid), DFID" is insane!\n", PFID(fid));
rc = fld_client_lookup(&lmv->lmv_fld, fid_seq(fid), mds,
LU_SEQ_RANGE_MDT, NULL);
if (rc) {
CERROR("Error while looking for mds number. Seq %#llx, err = %d\n",
fid_seq(fid), rc);
return rc;
}
CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
*mds, PFID(fid));
if (*mds >= lmv->desc.ld_tgt_count) {
CERROR("FLD lookup got invalid mds #%x (max: %x) "
"for fid="DFID"\n", *mds, lmv->desc.ld_tgt_count,
PFID(fid));
rc = -EINVAL;
}
return rc;
}
int lmv_fld_lookup(struct lmv_obd *lmv, const struct lu_fid *fid, u32 *mds)
{
struct obd_device *obd = lmv2obd_dev(lmv);
int rc;
/*
* FIXME: Currently ZFS still use local seq for ROOT unfortunately, and
* this fid_is_local check should be removed once LU-2240 is fixed
*/
if (!fid_is_sane(fid) || !(fid_seq_in_fldb(fid_seq(fid)) ||
fid_seq_is_local_file(fid_seq(fid)))) {
CERROR("%s: invalid FID " DFID "\n", obd->obd_name, PFID(fid));
return -EINVAL;
}
rc = fld_client_lookup(&lmv->lmv_fld, fid_seq(fid), mds,
LU_SEQ_RANGE_MDT, NULL);
if (rc) {
CERROR("Error while looking for mds number. Seq %#llx, err = %d\n",
fid_seq(fid), rc);
return rc;
}
CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
*mds, PFID(fid));
if (*mds >= lmv->desc.ld_tgt_count) {
CERROR("FLD lookup got invalid mds #%x (max: %x) for fid=" DFID "\n", *mds, lmv->desc.ld_tgt_count,
PFID(fid));
rc = -EINVAL;
}
return rc;
}
static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
struct lu_fid *fid)
{
ENTRY;
/* FID seqs not in FLDB, must be local seq */
if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
RETURN(0);
if (osd_seq_exists(env, osd, fid_seq(fid)))
RETURN(0);
RETURN(1);
}
int fid_is_on_ost(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, enum oi_check_flags flags)
{
struct lu_seq_range *range = &info->oti_seq_range;
int rc;
ENTRY;
if (flags & OI_KNOWN_ON_OST)
RETURN(1);
if (unlikely(fid_is_local_file(fid) || fid_is_igif(fid) ||
fid_is_llog(fid)) || fid_is_name_llog(fid) ||
fid_is_quota(fid))
RETURN(0);
if (fid_is_idif(fid) || fid_is_last_id(fid))
RETURN(1);
if (!(flags & OI_CHECK_FLD))
RETURN(0);
rc = osd_fld_lookup(info->oti_env, osd, fid_seq(fid), range);
if (rc != 0) {
if (rc != -ENOENT)
CERROR("%s: lookup FLD "DFID": rc = %d\n",
osd_name(osd), PFID(fid), rc);
RETURN(0);
}
if (fld_range_is_ost(range))
RETURN(1);
RETURN(0);
}
int osd_oi_update(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, const struct osd_inode_id *id,
handle_t *th, enum oi_check_flags flags)
{
struct lu_fid *oi_fid = &info->oti_fid2;
struct osd_inode_id *oi_id = &info->oti_id2;
int rc = 0;
if (unlikely(fid_is_last_id(fid)))
return osd_obj_spec_update(info, osd, fid, id, th);
if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
return osd_obj_map_update(info, osd, fid, id, th);
fid_cpu_to_be(oi_fid, fid);
osd_id_pack(oi_id, id);
rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
(const struct dt_rec *)oi_id,
(const struct dt_key *)oi_fid, th, false);
if (rc != 0)
return rc;
if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
rc = osd_obj_spec_update(info, osd, fid, id, th);
return rc;
}
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
* to a transaction group.
*/
int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la)
{
uint64_t oid;
int rc;
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
/* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
* would get an additional ditto copy */
if (unlikely(S_ISREG(la->la_mode) &&
fid_seq_is_local_file(fid_seq(fid))))
type = DMU_OTN_UINT8_METADATA;
/* Create a new DMU object using the default dnode size. */
oid = osd_dmu_object_alloc(osd->od_os, type, 0, 0, tx);
rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
LASSERTF(rc == 0, "sa_buf_hold %llu failed: %d\n", oid, rc);
LASSERT(la->la_valid & LA_MODE);
la->la_size = 0;
la->la_nlink = 1;
return 0;
}
int osd_oi_lookup(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, struct osd_inode_id *id,
enum oi_check_flags flags)
{
if (unlikely(fid_is_last_id(fid)))
return osd_obj_spec_lookup(info, osd, fid, id);
if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
return osd_obj_map_lookup(info, osd, fid, id);
if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
int rc;
if (fid_is_fs_root(fid)) {
osd_id_gen(id, osd_sb(osd)->s_root->d_inode->i_ino,
osd_sb(osd)->s_root->d_inode->i_generation);
return 0;
}
if (unlikely(fid_is_acct(fid)))
return osd_acct_obj_lookup(info, osd, fid, id);
/* For other special FIDs, try OI first, then do spec lookup */
rc = __osd_oi_lookup(info, osd, fid, id);
if (rc == -ENOENT)
return osd_obj_spec_lookup(info, osd, fid, id);
return rc;
}
if (!osd->od_igif_inoi && fid_is_igif(fid)) {
osd_id_gen(id, lu_igif_ino(fid), lu_igif_gen(fid));
return 0;
}
return __osd_oi_lookup(info, osd, fid, id);
}
void obdo_set_parent_fid(struct obdo *dst, const struct lu_fid *parent)
{
dst->o_parent_oid = fid_oid(parent);
dst->o_parent_seq = fid_seq(parent);
dst->o_parent_ver = fid_ver(parent);
dst->o_valid |= OBD_MD_FLGENER | OBD_MD_FLFID;
}
static int osp_rd_prealloc_last_seq(char *page, char **start, off_t off,
int count, int *eof, void *data)
{
struct obd_device *obd = data;
struct osp_device *osp = lu2osp_dev(obd->obd_lu_dev);
if (osp == NULL)
return 0;
return snprintf(page, count, LPX64"\n",
fid_seq(&osp->opd_pre_last_created_fid));
}
/* compatibility with orphan files created in versions before 2.11 */
static struct dt_key *mdd_orphan_key_fill_20(const struct lu_env *env,
const struct lu_fid *lf)
{
char *key = mdd_env_info(env)->mti_key;
LASSERT(key);
snprintf(key, sizeof(mdd_env_info(env)->mti_key),
ORPHAN_FILE_NAME_FORMAT_20,
fid_seq(lf), fid_oid(lf), fid_ver(lf), ORPH_OP_UNLINK);
return (struct dt_key *)key;
}
int osp_precreate_rollover_new_seq(struct lu_env *env, struct osp_device *osp)
{
struct lu_fid *fid = &osp_env_info(env)->osi_fid;
struct lu_fid *last_fid = &osp->opd_last_used_fid;
int rc;
ENTRY;
rc = seq_client_get_seq(env, osp->opd_obd->u.cli.cl_seq, &fid->f_seq);
if (rc != 0) {
CERROR("%s: alloc fid error: rc = %d\n",
osp->opd_obd->obd_name, rc);
RETURN(rc);
}
fid->f_oid = 1;
fid->f_ver = 0;
LASSERTF(fid_seq(fid) != fid_seq(last_fid),
"fid "DFID", last_fid "DFID"\n", PFID(fid),
PFID(last_fid));
rc = osp_write_last_oid_seq_files(env, osp, fid, 1);
if (rc != 0) {
CERROR("%s: Can not update oid/seq file: rc = %d\n",
osp->opd_obd->obd_name, rc);
RETURN(rc);
}
LCONSOLE_INFO("%s: update sequence from "LPX64" to "LPX64"\n",
osp->opd_obd->obd_name, fid_seq(last_fid),
fid_seq(fid));
/* Update last_xxx to the new seq */
spin_lock(&osp->opd_pre_lock);
osp->opd_last_used_fid = *fid;
osp->opd_gap_start_fid = *fid;
osp->opd_pre_used_fid = *fid;
osp->opd_pre_last_created_fid = *fid;
spin_unlock(&osp->opd_pre_lock);
RETURN(rc);
}
static struct dt_key* orph_key_fill_18(const struct lu_env *env,
const struct lu_fid *lf)
{
char *key = mdd_env_info(env)->mti_key;
int rc;
LASSERT(key);
rc = snprintf(key, NAME_MAX + 1, ORPHAN_FILE_NAME_FORMAT_18,
(unsigned long long)fid_seq(lf), fid_oid(lf));
if (rc > 0)
return (struct dt_key*) key;
else
return ERR_PTR(rc);
}
static inline int osp_objs_precreated(const struct lu_env *env,
struct osp_device *osp)
{
struct lu_fid *fid1 = &osp->opd_pre_last_created_fid;
struct lu_fid *fid2 = &osp->opd_pre_used_fid;
LASSERTF(fid_seq(fid1) == fid_seq(fid2),
"Created fid"DFID" Next fid "DFID"\n", PFID(fid1), PFID(fid2));
if (fid_is_idif(fid1)) {
struct ost_id *oi1 = &osp_env_info(env)->osi_oi;
struct ost_id *oi2 = &osp_env_info(env)->osi_oi2;
LASSERT(fid_is_idif(fid1) && fid_is_idif(fid2));
fid_to_ostid(fid1, oi1);
fid_to_ostid(fid2, oi2);
LASSERT(ostid_id(oi1) >= ostid_id(oi2));
return ostid_id(oi1) - ostid_id(oi2);
}
return fid_oid(fid1) - fid_oid(fid2);
}
int lmv_fld_lookup(struct lmv_obd *lmv,
const struct lu_fid *fid,
mdsno_t *mds)
{
int rc;
ENTRY;
LASSERTF(fid_is_sane(fid), DFID" is insane!\n", PFID(fid));
/* FIXME: Because ZFS still use LOCAL fid sequence for root,
* and root will always be in MDT0, for local fid, it will
* return 0 directly. And it should be removed once the root
* FID has been assigned with special sequence */
if (fid_seq(fid) == FID_SEQ_LOCAL_FILE) {
*mds = 0;
RETURN(0);
}
rc = fld_client_lookup(&lmv->lmv_fld, fid_seq(fid), mds,
LU_SEQ_RANGE_MDT, NULL);
if (rc) {
CERROR("Error while looking for mds number. Seq "LPX64
", err = %d\n", fid_seq(fid), rc);
RETURN(rc);
}
CDEBUG(D_INODE, "FLD lookup got mds #%x for fid="DFID"\n",
*mds, PFID(fid));
if (*mds >= lmv->desc.ld_tgt_count) {
CERROR("FLD lookup got invalid mds #%x (max: %x) "
"for fid="DFID"\n", *mds, lmv->desc.ld_tgt_count,
PFID(fid));
rc = -EINVAL;
}
RETURN(rc);
}
/*
* The transaction passed to this routine must have
* dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT) called and then assigned
* to a transaction group.
*/
int __osd_object_create(const struct lu_env *env, struct osd_object *obj,
dmu_buf_t **dbp, dmu_tx_t *tx, struct lu_attr *la,
uint64_t parent)
{
uint64_t oid;
int rc;
struct osd_device *osd = osd_obj2dev(obj);
const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
dmu_object_type_t type = DMU_OT_PLAIN_FILE_CONTENTS;
/* Assert that the transaction has been assigned to a
transaction group. */
LASSERT(tx->tx_txg != 0);
/* Use DMU_OTN_UINT8_METADATA for local objects so their data blocks
* would get an additional ditto copy */
if (unlikely(S_ISREG(la->la_mode) &&
fid_seq_is_local_file(fid_seq(fid))))
type = DMU_OTN_UINT8_METADATA;
/* Create a new DMU object. */
oid = dmu_object_alloc(osd->od_os, type, 0,
DMU_OT_SA, DN_MAX_BONUSLEN, tx);
rc = -sa_buf_hold(osd->od_os, oid, osd_obj_tag, dbp);
LASSERTF(rc == 0, "sa_buf_hold "LPU64" failed: %d\n", oid, rc);
LASSERT(la->la_valid & LA_MODE);
la->la_size = 0;
la->la_nlink = 1;
rc = __osd_attr_init(env, osd, oid, tx, la, parent);
if (rc != 0) {
sa_buf_rele(*dbp, osd_obj_tag);
*dbp = NULL;
dmu_object_free(osd->od_os, oid, tx);
return rc;
}
return 0;
}
/*
* Returns true, if fid is local to this server node.
*
* WARNING: this function is *not* guaranteed to return false if fid is
* remote: it makes an educated conservative guess only.
*
* fid_is_local() is supposed to be used in assertion checks only.
*/
int fid_is_local(const struct lu_env *env,
struct lu_site *site, const struct lu_fid *fid)
{
int result;
struct seq_server_site *ss_site;
struct lu_seq_range *range;
struct fld_thread_info *info;
ENTRY;
info = lu_context_key_get(&env->le_ctx, &fld_thread_key);
range = &info->fti_lrange;
result = 1; /* conservatively assume fid is local */
ss_site = lu_site2seq(site);
if (ss_site->ss_client_fld != NULL) {
int rc;
rc = fld_cache_lookup(ss_site->ss_client_fld->lcf_cache,
fid_seq(fid), range);
if (rc == 0)
result = (range->lsr_index == ss_site->ss_node_id);
}
return result;
}
static int osp_precreate_thread(void *_arg)
{
struct osp_device *d = _arg;
struct ptlrpc_thread *thread = &d->opd_pre_thread;
struct l_wait_info lwi = { 0 };
struct lu_env env;
int rc;
ENTRY;
rc = lu_env_init(&env, d->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
if (rc) {
CERROR("%s: init env error: rc = %d\n", d->opd_obd->obd_name,
rc);
RETURN(rc);
}
spin_lock(&d->opd_pre_lock);
thread->t_flags = SVC_RUNNING;
spin_unlock(&d->opd_pre_lock);
wake_up(&thread->t_ctl_waitq);
while (osp_precreate_running(d)) {
/*
* need to be connected to OST
*/
while (osp_precreate_running(d)) {
l_wait_event(d->opd_pre_waitq,
!osp_precreate_running(d) ||
d->opd_new_connection,
&lwi);
if (!d->opd_new_connection)
continue;
d->opd_new_connection = 0;
d->opd_got_disconnected = 0;
break;
}
if (!osp_precreate_running(d))
break;
LASSERT(d->opd_obd->u.cli.cl_seq != NULL);
if (d->opd_obd->u.cli.cl_seq->lcs_exp == NULL) {
/* Get new sequence for client first */
LASSERT(d->opd_exp != NULL);
d->opd_obd->u.cli.cl_seq->lcs_exp =
class_export_get(d->opd_exp);
rc = osp_init_pre_fid(d);
if (rc != 0) {
class_export_put(d->opd_exp);
d->opd_obd->u.cli.cl_seq->lcs_exp = NULL;
CERROR("%s: init pre fid error: rc = %d\n",
d->opd_obd->obd_name, rc);
continue;
}
}
osp_statfs_update(d);
/*
* Clean up orphans or recreate missing objects.
*/
rc = osp_precreate_cleanup_orphans(&env, d);
if (rc != 0)
continue;
/*
* connected, can handle precreates now
*/
while (osp_precreate_running(d)) {
l_wait_event(d->opd_pre_waitq,
!osp_precreate_running(d) ||
osp_precreate_near_empty(&env, d) ||
osp_statfs_need_update(d) ||
d->opd_got_disconnected, &lwi);
if (!osp_precreate_running(d))
break;
/* something happened to the connection
* have to start from the beginning */
if (d->opd_got_disconnected)
break;
if (osp_statfs_need_update(d))
osp_statfs_update(d);
/* To avoid handling different seq in precreate/orphan
* cleanup, it will hold precreate until current seq is
* used up. */
if (unlikely(osp_precreate_end_seq(&env, d) &&
!osp_create_end_seq(&env, d)))
continue;
if (unlikely(osp_precreate_end_seq(&env, d) &&
osp_create_end_seq(&env, d))) {
LCONSOLE_INFO("%s:"LPX64" is used up."
" Update to new seq\n",
d->opd_obd->obd_name,
fid_seq(&d->opd_pre_last_created_fid));
rc = osp_precreate_rollover_new_seq(&env, d);
if (rc)
continue;
}
if (osp_precreate_near_empty(&env, d)) {
rc = osp_precreate_send(&env, d);
/* osp_precreate_send() sets opd_pre_status
* in case of error, that prevent the using of
* failed device. */
if (rc < 0 && rc != -ENOSPC &&
rc != -ETIMEDOUT && rc != -ENOTCONN)
CERROR("%s: cannot precreate objects:"
" rc = %d\n",
d->opd_obd->obd_name, rc);
}
}
}
thread->t_flags = SVC_STOPPED;
lu_env_fini(&env);
wake_up(&thread->t_ctl_waitq);
RETURN(0);
}
int mdd_compat_fixes(const struct lu_env *env, struct mdd_device *mdd)
{
struct mdd_thread_info *info = mdd_env_info(env);
struct mdd_object *root;
struct dt_object *o;
struct lustre_mdt_attrs *lma;
struct lu_buf buf;
int rc;
ENTRY;
/* IGIF FIDS are valid for old 1.8 and 2.[123] ROOT and are kept.
* Normal FIDs used by Xyratex 1.8->2.1 upgrade tool are also kept. */
if (fid_is_igif(&mdd->mdd_root_fid) || fid_is_norm(&mdd->mdd_root_fid))
RETURN(0);
/*
* FID is supposed to be FID_SEQ_ROOT for:
* - new ldiskfs fs
* - new ZFS fs
* - old ZFS fs, by now processed with osd_convert_root_to_new_seq()
*/
if (fid_seq(&mdd->mdd_root_fid) != FID_SEQ_ROOT) {
CERROR("%s: wrong FID "DFID" is used for /ROOT\n",
mdd2obd_dev(mdd)->obd_name,
PFID(&mdd->mdd_root_fid));
RETURN(-EINVAL);
}
root = mdd_object_find(env, mdd, &mdd->mdd_root_fid);
if (IS_ERR(root))
RETURN(PTR_ERR(root));
o = mdd_object_child(root);
CDEBUG(D_OTHER, "/ROOT = "DFID"\n", PFID(&mdd->mdd_root_fid));
if (dt_try_as_dir(env, o) == 0) {
CERROR("%s: not a directory\n", mdd2obd_dev(mdd)->obd_name);
GOTO(out, rc = -ENOTDIR);
}
lma = (struct lustre_mdt_attrs *)&info->mti_xattr_buf;
CLASSERT(sizeof(info->mti_xattr_buf) >= LMA_OLD_SIZE);
buf.lb_len = LMA_OLD_SIZE;
buf.lb_buf = lma;
rc = mdo_xattr_get(env, root, &buf, XATTR_NAME_LMA, BYPASS_CAPA);
if (rc < 0 && rc != -ENODATA) {
CERROR("%s: can't fetch LMA: rc = %d\n",
mdd2obd_dev(mdd)->obd_name, rc);
GOTO(out, rc);
}
lustre_lma_swab(lma);
if (lu_fid_eq(&lma->lma_self_fid, &mdd->mdd_root_fid)) {
/* /ROOT has been converted already
* or was correct from the beginning */
CDEBUG(D_OTHER, "%s: converted already\n",
mdd2obd_dev(mdd)->obd_name);
GOTO(out, rc = 0);
}
/* this is supposed to happen only on pre-production ZFS backend */
if (strcmp(mdd->mdd_bottom->dd_lu_dev.ld_type->ldt_name,
LUSTRE_OSD_ZFS_NAME) != 0) {
CERROR("%s: "DFID" is used on ldiskfs?!\n",
mdd2obd_dev(mdd)->obd_name, PFID(&mdd->mdd_root_fid));
GOTO(out, rc = -ENOTSUPP);
}
LCONSOLE_INFO("%s: FID of /ROOT has been changed. "
"Please remount the clients.\n",
mdd2obd_dev(mdd)->obd_name);
/* Fill FLDB first */
rc = mdd_fill_fldb(env, mdd);
if (rc)
GOTO(out, rc);
/* remove ./.. from /ROOT */
rc = mdd_convert_remove_dots(env, mdd, root);
if (rc)
GOTO(out, rc);
/* go over the directory, fix all the objects */
rc = mdd_fix_children(env, mdd, o);
if (rc)
GOTO(out, rc);
/* Update LMA on /ROOT. Done for simplicity in MDD, not in osd-zfs.
* Correct LMA will imply the whole directory has been coverted
* successfully, otherwise it will be retried on next mount. */
rc = mdd_convert_lma(env, mdd, root);
out:
mdd_object_put(env, root);
RETURN(rc);
}
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);
}
/**
* Prepare buffers for write request processing.
*
* This function converts remote buffers from client to local buffers
* and prepares the latter. If there is recovery in progress and required
* object is missing then it can be re-created before write.
*
* \param[in] env execution environment
* \param[in] exp OBD export of client
* \param[in] ofd OFD device
* \param[in] fid FID of object
* \param[in] la object attributes
* \param[in] oa OBDO structure from client
* \param[in] objcount always 1
* \param[in] obj object data
* \param[in] rnb remote buffers
* \param[in] nr_local number of local buffers
* \param[in] lnb local buffers
* \param[in] jobid job ID name
*
* \retval 0 on successful prepare
* \retval negative value on error
*/
static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
struct ofd_device *ofd, const struct lu_fid *fid,
struct lu_attr *la, struct obdo *oa,
int objcount, struct obd_ioobj *obj,
struct niobuf_remote *rnb, int *nr_local,
struct niobuf_local *lnb, char *jobid)
{
struct ofd_object *fo;
int i, j, k, rc = 0, tot_bytes = 0;
ENTRY;
LASSERT(env != NULL);
LASSERT(objcount == 1);
if (unlikely(exp->exp_obd->obd_recovering)) {
u64 seq = fid_seq(fid);
u64 oid = fid_oid(fid);
struct ofd_seq *oseq;
oseq = ofd_seq_load(env, ofd, seq);
if (IS_ERR(oseq)) {
CERROR("%s: Can't find FID Sequence "LPX64": rc = %d\n",
ofd_name(ofd), seq, (int)PTR_ERR(oseq));
GOTO(out, rc = -EINVAL);
}
if (oid > ofd_seq_last_oid(oseq)) {
int sync = 0;
int diff;
mutex_lock(&oseq->os_create_lock);
diff = oid - ofd_seq_last_oid(oseq);
/* Do sync create if the seq is about to used up */
if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq)) {
if (unlikely(oid >= IDIF_MAX_OID - 1))
sync = 1;
} else if (fid_seq_is_norm(seq)) {
if (unlikely(oid >=
LUSTRE_DATA_SEQ_MAX_WIDTH - 1))
sync = 1;
} else {
CERROR("%s : invalid o_seq "DOSTID"\n",
ofd_name(ofd), POSTID(&oa->o_oi));
mutex_unlock(&oseq->os_create_lock);
ofd_seq_put(env, oseq);
GOTO(out, rc = -EINVAL);
}
while (diff > 0) {
u64 next_id = ofd_seq_last_oid(oseq) + 1;
int count = ofd_precreate_batch(ofd, diff);
rc = ofd_precreate_objects(env, ofd, next_id,
oseq, count, sync);
if (rc < 0) {
mutex_unlock(&oseq->os_create_lock);
ofd_seq_put(env, oseq);
GOTO(out, rc);
}
diff -= rc;
}
mutex_unlock(&oseq->os_create_lock);
}
ofd_seq_put(env, oseq);
}
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo))
GOTO(out, rc = PTR_ERR(fo));
LASSERT(fo != NULL);
ofd_read_lock(env, fo);
if (!ofd_object_exists(fo)) {
CERROR("%s: BRW to missing obj "DOSTID"\n",
exp->exp_obd->obd_name, POSTID(&obj->ioo_oid));
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
GOTO(out, rc = -ENOENT);
}
if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
rc = ofd_verify_ff(env, fo, oa);
if (rc != 0) {
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
GOTO(out, rc);
}
}
/* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
* space back if possible */
ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
/* parse remote buffers to local buffers and prepare the latter */
*nr_local = 0;
for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
rc = dt_bufs_get(env, ofd_object_child(fo),
rnb + i, lnb + j, 1);
if (unlikely(rc < 0))
GOTO(err, rc);
LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
/* correct index for local buffers to continue with */
for (k = 0; k < rc; k++) {
lnb[j+k].lnb_flags = rnb[i].rnb_flags;
if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
lnb[j+k].lnb_rc = -ENOSPC;
}
j += rc;
*nr_local += rc;
LASSERT(j <= PTLRPC_MAX_BRW_PAGES);
tot_bytes += rnb[i].rnb_len;
}
LASSERT(*nr_local > 0 && *nr_local <= PTLRPC_MAX_BRW_PAGES);
rc = dt_write_prep(env, ofd_object_child(fo), lnb, *nr_local);
if (unlikely(rc != 0))
GOTO(err, rc);
ofd_counter_incr(exp, LPROC_OFD_STATS_WRITE, jobid, tot_bytes);
RETURN(0);
err:
dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
ofd_read_unlock(env, fo);
ofd_object_put(env, fo);
/* ofd_grant_prepare_write() was called, so we must commit */
ofd_grant_commit(exp, oa->o_grant_used, rc);
out:
/* let's still process incoming grant information packed in the oa,
* but without enforcing grant since we won't proceed with the write.
* Just like a read request actually. */
ofd_grant_prepare_read(env, exp, oa);
return rc;
}
/**
* Implementation of ldlm_valblock_ops::lvbo_init for OFD.
*
* This function allocates and initializes new LVB data for the given
* LDLM resource if it is not allocated yet. New LVB is filled with attributes
* of the object associated with that resource. Function does nothing if LVB
* for the given LDLM resource is allocated already.
*
* Called with res->lr_lvb_sem held.
*
* \param[in] lock LDLM lock on resource
*
* \retval 0 on successful setup
* \retval negative value on error
*/
static int ofd_lvbo_init(const struct lu_env *env, struct ldlm_resource *res)
{
struct ost_lvb *lvb;
struct ofd_device *ofd;
struct ofd_object *fo;
struct ofd_thread_info *info;
struct lu_env _env;
int rc = 0;
ENTRY;
LASSERT(res);
LASSERT(mutex_is_locked(&res->lr_lvb_mutex));
if (res->lr_lvb_data != NULL)
RETURN(0);
ofd = ldlm_res_to_ns(res)->ns_lvbp;
LASSERT(ofd != NULL);
if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_OST_LVB))
RETURN(-ENOMEM);
if (!env) {
rc = lu_env_init(&_env, LCT_DT_THREAD);
if (rc)
RETURN(rc);
env = &_env;
}
OBD_ALLOC_PTR(lvb);
if (lvb == NULL)
GOTO(out, rc = -ENOMEM);
info = ofd_info(env);
res->lr_lvb_data = lvb;
res->lr_lvb_len = sizeof(*lvb);
ost_fid_from_resid(&info->fti_fid, &res->lr_name,
ofd->ofd_lut.lut_lsd.lsd_osd_index);
fo = ofd_object_find(env, ofd, &info->fti_fid);
if (IS_ERR(fo))
GOTO(out_lvb, rc = PTR_ERR(fo));
rc = ofd_attr_get(env, fo, &info->fti_attr);
if (rc) {
struct ofd_seq *oseq;
__u64 seq;
/* Object could be recreated during the first
* CLEANUP_ORPHAN request. */
if (rc == -ENOENT) {
seq = fid_seq(&info->fti_fid);
oseq = ofd_seq_load(env, ofd, fid_seq_is_idif(seq) ?
FID_SEQ_OST_MDT0 : seq);
if (!IS_ERR_OR_NULL(oseq)) {
if (!oseq->os_last_id_synced)
rc = -EAGAIN;
ofd_seq_put(env, oseq);
}
}
GOTO(out_obj, rc);
}
lvb->lvb_size = info->fti_attr.la_size;
lvb->lvb_blocks = info->fti_attr.la_blocks;
lvb->lvb_mtime = info->fti_attr.la_mtime;
lvb->lvb_atime = info->fti_attr.la_atime;
lvb->lvb_ctime = info->fti_attr.la_ctime;
CDEBUG(D_DLMTRACE, "res: "DFID" initial lvb size: %llu, "
"mtime: %#llx, blocks: %#llx\n",
PFID(&info->fti_fid), lvb->lvb_size,
lvb->lvb_mtime, lvb->lvb_blocks);
info->fti_attr.la_valid = 0;
EXIT;
out_obj:
ofd_object_put(env, fo);
out_lvb:
if (rc != 0)
OST_LVB_SET_ERR(lvb->lvb_blocks, rc);
out:
/* Don't free lvb data on lookup error */
if (env && env == &_env)
lu_env_fini(&_env);
return rc;
}
int osd_oi_insert(struct osd_thread_info *info, struct osd_device *osd,
const struct lu_fid *fid, const struct osd_inode_id *id,
handle_t *th, enum oi_check_flags flags)
{
struct lu_fid *oi_fid = &info->oti_fid2;
struct osd_inode_id *oi_id = &info->oti_id2;
int rc = 0;
if (unlikely(fid_is_last_id(fid)))
return osd_obj_spec_insert(info, osd, fid, id, th);
if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
return osd_obj_map_insert(info, osd, fid, id, th);
fid_cpu_to_be(oi_fid, fid);
osd_id_pack(oi_id, id);
rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
(const struct dt_rec *)oi_id,
(const struct dt_key *)oi_fid, th, true);
if (rc != 0) {
struct inode *inode;
struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
if (rc != -EEXIST)
return rc;
rc = osd_oi_lookup(info, osd, fid, oi_id, 0);
if (rc != 0)
return rc;
if (unlikely(osd_id_eq(id, oi_id)))
return 1;
/* Check whether the mapping for oi_id is valid or not. */
inode = osd_iget(info, osd, oi_id);
if (IS_ERR(inode)) {
rc = PTR_ERR(inode);
if (rc == -ENOENT || rc == -ESTALE)
goto update;
return rc;
}
/* The EA inode should NOT be in OI, old OI scrub may added
* such OI mapping by wrong, replace it. */
if (unlikely(osd_is_ea_inode(inode))) {
iput(inode);
goto update;
}
rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
iput(inode);
if (rc == -ENODATA)
goto update;
if (rc != 0)
return rc;
if (!(lma->lma_compat & LMAC_NOT_IN_OI) &&
lu_fid_eq(fid, &lma->lma_self_fid)) {
CERROR("%.16s: the FID "DFID" is used by two objects: "
"%u/%u %u/%u\n",
LDISKFS_SB(osd_sb(osd))->s_es->s_volume_name,
PFID(fid), oi_id->oii_ino, oi_id->oii_gen,
id->oii_ino, id->oii_gen);
return -EEXIST;
}
update:
osd_id_pack(oi_id, id);
rc = osd_oi_iam_refresh(info, osd_fid2oi(osd, fid),
(const struct dt_rec *)oi_id,
(const struct dt_key *)oi_fid, th, false);
if (rc != 0)
return rc;
}
if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE))
rc = osd_obj_spec_insert(info, osd, fid, id, th);
return rc;
}
static inline int fid_is_fs_root(const struct lu_fid *fid)
{
/* Map root inode to special local object FID */
return (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE &&
fid_oid(fid) == OSD_FS_ROOT_OID));
}