/**
* FLR: verify the layout version of object.
*
* \param[in] env execution environment
* \param[in] fo OFD object
* \param[in] oa OBDO structure with layout version
*
* \retval 0 on successful verification
* \retval -EINPROGRESS layout version is in transfer
* \retval -ESTALE the layout version on client is stale
*/
int ofd_verify_layout_version(const struct lu_env *env,
struct ofd_object *fo, const struct obdo *oa)
{
__u32 layout_version;
int rc;
ENTRY;
if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_OST_SKIP_LV_CHECK)))
GOTO(out, rc = 0);
rc = ofd_object_ff_load(env, fo);
if (rc < 0) {
if (rc == -ENODATA)
rc = -EINPROGRESS;
GOTO(out, rc);
}
layout_version = fo->ofo_ff.ff_layout_version;
if (oa->o_layout_version >= layout_version &&
oa->o_layout_version <= layout_version + fo->ofo_ff.ff_range)
GOTO(out, rc = 0);
/* normal traffic, decide if to return ESTALE or EINPROGRESS */
layout_version &= ~LU_LAYOUT_RESYNC;
/* this update is not legitimate */
if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) <= layout_version)
GOTO(out, rc = -ESTALE);
/* layout version may not be transmitted yet */
if ((oa->o_layout_version & ~LU_LAYOUT_RESYNC) > layout_version)
GOTO(out, rc = -EINPROGRESS);
EXIT;
out:
CDEBUG(D_INODE, DFID " verify layout version: %u vs. %u/%u, rc: %d\n",
PFID(lu_object_fid(&fo->ofo_obj.do_lu)),
oa->o_layout_version, fo->ofo_ff.ff_layout_version,
fo->ofo_ff.ff_range, rc);
return rc;
}
/**
* Verify parent FID of an object.
*
* Check the parent FID is sane and start extended
* verification procedure otherwise.
*
* \param[in] env execution environment
* \param[in] fo OFD object
* \param[in] oa OBDO structure with PFID
*
* \retval 0 on successful verification
* \retval -EINPROGRESS if PFID is being repaired
* \retval -EPERM if PFID was verified but still insane
*/
int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
struct obdo *oa)
{
struct lu_fid *pfid = &fo->ofo_pfid;
int rc = 0;
ENTRY;
if (fid_is_sane(pfid)) {
if (likely(oa->o_parent_seq == pfid->f_seq &&
oa->o_parent_oid == pfid->f_oid &&
oa->o_stripe_idx == pfid->f_stripe_idx))
RETURN(0);
if (fo->ofo_pfid_verified)
RETURN(-EPERM);
}
/* The OST-object may be inconsistent, and we need further verification.
* To avoid block the RPC service thread, return -EINPROGRESS to client
* and make it retry later. */
if (fo->ofo_pfid_checking)
RETURN(-EINPROGRESS);
rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
RETURN(0);
if (rc < 0)
RETURN(rc);
if (likely(oa->o_parent_seq == pfid->f_seq &&
oa->o_parent_oid == pfid->f_oid &&
oa->o_stripe_idx == pfid->f_stripe_idx))
RETURN(0);
/* Push it to the dedicated thread for further verification. */
ofd_add_inconsistency_item(env, fo, oa);
RETURN(-EINPROGRESS);
}
/**
* Set attributes of object during write bulk IO processing.
*
* Change object attributes and write parent FID into extended
* attributes when needed.
*
* \param[in] env execution environment
* \param[in] ofd OFD device
* \param[in] ofd_obj OFD object
* \param[in] la object attributes
* \param[in] ff parent FID
*
* \retval 0 on successful attributes update
* \retval negative value on error
*/
static int
ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
struct ofd_object *ofd_obj, struct lu_attr *la,
struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
__u64 valid = la->la_valid;
int rc;
struct thandle *th;
struct dt_object *dt_obj;
int ff_needed = 0;
ENTRY;
LASSERT(la);
dt_obj = ofd_object_child(ofd_obj);
LASSERT(dt_obj != NULL);
la->la_valid &= LA_UID | LA_GID;
rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(out, rc);
if (ff != NULL) {
rc = ofd_object_ff_load(env, ofd_obj);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(out, rc);
}
if (!la->la_valid && !ff_needed)
/* no attributes to set */
GOTO(out, rc = 0);
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
if (la->la_valid) {
rc = dt_declare_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
if (ff_needed) {
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
ff->ff_parent.f_oid =
cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, dt_obj, &info->fti_buf,
XATTR_NAME_FID, 0, th);
if (rc)
GOTO(out_tx, rc);
}
/* We don't need a transno for this operation which will be re-executed
* anyway when the OST_WRITE (with a transno assigned) is replayed */
rc = dt_trans_start_local(env, ofd->ofd_osd , th);
if (rc)
GOTO(out_tx, rc);
/* set uid/gid */
if (la->la_valid) {
rc = dt_attr_set(env, dt_obj, la, th);
if (rc)
GOTO(out_tx, rc);
}
/* set filter fid EA */
if (ff_needed) {
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NOPFID))
GOTO(out_tx, rc);
rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
0, th);
if (rc == 0) {
ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not
* the real parent MDT-object's FID::f_ver, instead it
* is the OST-object index in its parent MDT-object's
* layout EA. */
ofd_obj->ofo_pfid.f_stripe_idx =
le32_to_cpu(ff->ff_parent.f_stripe_idx);
}
}
GOTO(out_tx, rc);
out_tx:
dt_trans_stop(env, ofd->ofd_osd, th);
out:
la->la_valid = valid;
return rc;
}
int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
__u64 start, __u64 end, struct lu_attr *la,
struct filter_fid *ff, struct obdo *oa)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct ofd_mod_data *fmd;
struct dt_object *dob = ofd_object_child(fo);
struct thandle *th;
int ff_needed = 0;
int rc;
ENTRY;
/* we support truncate, not punch yet */
LASSERT(end == OBD_OBJECT_EOF);
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
if (ofd->ofd_lfsck_verify_pfid && oa->o_valid & OBD_MD_FLFID) {
rc = ofd_verify_ff(env, fo, oa);
if (rc != 0)
GOTO(unlock, rc);
}
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
rc = ofd_attr_handle_ugid(env, fo, la, 0 /* !is_setattr */);
if (rc != 0)
GOTO(unlock, rc);
if (ff != NULL) {
rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(unlock, rc);
}
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_attr_set(env, dob, la, th);
if (rc)
GOTO(stop, rc);
rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
if (rc)
GOTO(stop, rc);
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, ofd_object_child(fo),
&info->fti_buf, XATTR_NAME_FID, 0,
th);
if (rc)
GOTO(stop, rc);
}
rc = ofd_trans_start(env, ofd, fo, th);
if (rc)
GOTO(stop, rc);
rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th,
ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, dob, la, th, ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
if (ff_needed) {
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
XATTR_NAME_FID, 0, th, BYPASS_CAPA);
if (rc == 0) {
fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not
* the real parent MDT-object's FID::f_ver, instead it
* is the OST-object index in its parent MDT-object's
* layout EA. */
fo->ofo_pfid.f_stripe_idx =
le32_to_cpu(ff->ff_parent.f_stripe_idx);
}
}
GOTO(stop, rc);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
return rc;
}
int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
struct lu_attr *la, struct filter_fid *ff)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_device *ofd = ofd_obj2dev(fo);
struct thandle *th;
struct ofd_mod_data *fmd;
int ff_needed = 0;
int rc;
ENTRY;
ofd_write_lock(env, fo);
if (!ofd_object_exists(fo))
GOTO(unlock, rc = -ENOENT);
if (la->la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
fmd = ofd_fmd_get(info->fti_exp, &fo->ofo_header.loh_fid);
if (fmd && fmd->fmd_mactime_xid < info->fti_xid)
fmd->fmd_mactime_xid = info->fti_xid;
ofd_fmd_put(info->fti_exp, fmd);
}
/* VBR: version recovery check */
rc = ofd_version_get_check(info, fo);
if (rc)
GOTO(unlock, rc);
rc = ofd_attr_handle_ugid(env, fo, la, 1 /* is_setattr */);
if (rc != 0)
GOTO(unlock, rc);
if (ff != NULL) {
rc = ofd_object_ff_load(env, fo);
if (rc == -ENODATA)
ff_needed = 1;
else if (rc < 0)
GOTO(unlock, rc);
}
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_attr_set(env, ofd_object_child(fo), la, th);
if (rc)
GOTO(stop, rc);
if (ff_needed) {
info->fti_buf.lb_buf = ff;
info->fti_buf.lb_len = sizeof(*ff);
rc = dt_declare_xattr_set(env, ofd_object_child(fo),
&info->fti_buf, XATTR_NAME_FID, 0,
th);
if (rc)
GOTO(stop, rc);
}
rc = ofd_trans_start(env, ofd, la->la_valid & LA_SIZE ? fo : NULL, th);
if (rc)
GOTO(stop, rc);
rc = dt_attr_set(env, ofd_object_child(fo), la, th,
ofd_object_capa(env, fo));
if (rc)
GOTO(stop, rc);
if (ff_needed) {
rc = dt_xattr_set(env, ofd_object_child(fo), &info->fti_buf,
XATTR_NAME_FID, 0, th, BYPASS_CAPA);
if (rc == 0) {
fo->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
fo->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not
* the real parent MDT-object's FID::f_ver, instead it
* is the OST-object index in its parent MDT-object's
* layout EA. */
fo->ofo_pfid.f_stripe_idx =
le32_to_cpu(ff->ff_parent.f_stripe_idx);
}
}
GOTO(stop, rc);
stop:
ofd_trans_stop(env, ofd, th, rc);
unlock:
ofd_write_unlock(env, fo);
return rc;
}
