[LU-9744] HS_DIRTY/CLF_HSM_DIRTY can be wrongly reported upon HSM action error

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.10.1, Lustre 2.11.0
• Severity: 3
• Assignee: Bruno Faccini (Inactive)
• Reporter: Bruno Faccini (Inactive)

Description

In hsm_cdt_request_completed() and upon error being returned from mdt_hsm_get_md_hsm(), where "struct md_hsm mh" has not been populated, HS_DIRTY can be wrongly detected and thus CLF_HSM_DIRTY flag will be reported by error :

/**
 * get obj and HSM attributes on a fid
 * \param mti [IN] context
 * \param fid [IN] object fid
 * \param hsm [OUT] HSM meta data
 * \retval obj or error (-ENOENT if not found)
 */
struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
                                      const struct lu_fid *fid,
                                      struct md_hsm *hsm)
{
        struct md_attr          *ma;
        struct mdt_object       *obj;
        int                      rc;
        ENTRY;

        ma = &mti->mti_attr;
        ma->ma_need = MA_HSM;
        ma->ma_valid = 0;

        /* find object by FID */
        obj = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
        if (IS_ERR(obj))
                RETURN(obj);

        if (!mdt_object_exists(obj)) {
                /* no more object */
                mdt_object_put(mti->mti_env, obj);
                RETURN(ERR_PTR(-ENOENT));
        }

        rc = mdt_attr_get_complex(mti, obj, ma);
        if (rc) {
                mdt_object_put(mti->mti_env, obj);
                RETURN(ERR_PTR(rc));
        }

        if (ma->ma_valid & MA_HSM)
                *hsm = ma->ma_hsm;
        else
                memset(hsm, 0, sizeof(*hsm));
        ma->ma_valid = 0;
        RETURN(obj);
}

...................

/**
 * update status of a completed request
 * \param mti [IN] context
 * \param pgs [IN] progress of the copy tool
 * \param update_record [IN] update llog record
 * \retval 0 success
 * \retval -ve failure
 */
static int hsm_cdt_request_completed(struct mdt_thread_info *mti,
                                     struct hsm_progress_kernel *pgs,
                                     const struct cdt_agent_req *car,
                                     enum agent_req_status *status)
{
        const struct lu_env     *env = mti->mti_env;
        struct mdt_device       *mdt = mti->mti_mdt;
        struct coordinator      *cdt = &mdt->mdt_coordinator;
        struct mdt_object       *obj = NULL;
        int                      cl_flags = 0, rc = 0;
        struct md_hsm            mh;
        bool                     is_mh_changed;
        bool                     need_changelog = true;
        ENTRY;

        /* default is to retry */
        *status = ARS_WAITING;

        /* find object by FID, mdt_hsm_get_md_hsm() returns obj or err
         * if error/removed continue anyway to get correct reporting done */
        obj = mdt_hsm_get_md_hsm(mti, &car->car_hai->hai_fid, &mh);
......................
        /* rc != 0 means error when analysing action, it may come from
         * a crasy CT no need to manage DIRTY
         */
        if (rc == 0)
                hsm_set_cl_flags(&cl_flags,
                                 mh.mh_flags & HS_DIRTY ? CLF_HSM_DIRTY : 0);
.......................

!IS_ERR(obj) should also be tested before adding CLF_HSM_DIRTY in cl_flags.

This can be the cause of errors in associated sanity-hsm sub-tests test_220a, test_222c, test_222d, tests_224a, when analyzing ChangeLog flags.

Issue Links

• is related to: LU-10043 sanity-hsm test_222c: Changelog flag is not 0x82