On Thu, Apr 23, 2020 at 08:25:24PM +0300, Max Gurtovoy wrote: > > On 4/21/2020 6:33 PM, Christoph Hellwig wrote: >> On Fri, Mar 27, 2020 at 08:15:43PM +0300, Max Gurtovoy wrote: >>> - if (!nvmet_check_transfer_len(req, nvmet_rw_data_len(req))) >>> + if (!nvmet_check_transfer_len(req, >>> + nvmet_rw_data_len(req) + req->md_len)) >> Shouldn't we also calculate the actual metadata length on the fly here? > > we calculate it during nvmet_init_req. Indeed. > >> >>> blk_start_plug(&plug); >>> + if (req->use_md) >> Can't we use a non-NULL req->md_sg or non-null req->md_sg_cnt as a >> metadata supported indicator and remove the use_md flag? Maybe wrap >> them in a helper function that also checks for blk integrity support >> using IS_ENABLED and we can skip the stubs as well. > > I'll replace it with: > > static inline bool nvmet_req_has_pi(struct nvmet_req *req) > { > return req->sq->ctrl->pi_support && nvmet_ns_has_pi(req->ns); > } Actually I think you can simply check for a non-0 md_len, as we always set them at the same time.