Re: [dpdk-dev] [PATCH v2 02/15] crypto: add total raw buffer length

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

Hi Akhil,

> > > On 9/21/2021 12:58 AM, Akhil Goyal wrote:
> > > >>> From: Gagandeep Singh <g.singh@nxp.com>
> > > >>>
> > > >>> The current crypto raw data vectors is extended to support
> > > >>> rte_security usecases, where we need total data length to know
> > > >>> how much additional memory space is available in buffer other
> > > >>> than data length so that driver/HW can write expanded size
> > > >>> data after encryption.
> > > >>>
> > > >>> Signed-off-by: Gagandeep Singh <g.singh@nxp.com>
> > > >>> Acked-by: Akhil Goyal <gakhil@marvell.com>
> > > >>> ---
> > > >>>   lib/cryptodev/rte_crypto_sym.h | 6 ++++++
> > > >>>   1 file changed, 6 insertions(+)
> > > >>>
> > > >>> diff --git a/lib/cryptodev/rte_crypto_sym.h
> > > >> b/lib/cryptodev/rte_crypto_sym.h
> > > >>> index dcc0bd5933..e5cef1fb72 100644
> > > >>> --- a/lib/cryptodev/rte_crypto_sym.h
> > > >>> +++ b/lib/cryptodev/rte_crypto_sym.h
> > > >>> @@ -37,6 +37,8 @@ struct rte_crypto_vec {
> > > >>>   	rte_iova_t iova;
> > > >>>   	/** length of the data buffer */
> > > >>>   	uint32_t len;
> > > >>> +	/** total buffer length*/
> > > >>> +	uint32_t tot_len;
> > > >>>   };
> > > >>>
> > > >>>   /**
> > > >>> @@ -980,12 +982,14 @@ rte_crypto_mbuf_to_vec(const struct
> > rte_mbuf
> > > >> *mb, uint32_t ofs, uint32_t len,
> > > >>>   	seglen = mb->data_len - ofs;
> > > >>>   	if (len <= seglen) {
> > > >>>   		vec[0].len = len;
> > > >>> +		vec[0].tot_len = mb->buf_len;
> > > >> That doesn't look right.
> > > >> We should take into a count mbuf headroom and input offset.
> > > >> Something like:
> > > >> vec[0].tot_len = mb->buf_len - rte_pktmbuf_headroom(m) - ofs;
> > > >> Same in other places below.
> > > >>
> > > > I believe the packet can expand into headroom based on the protocol
> > support.
> > > Yes, total length is representing the total buffer length available. The
> > > security protocol shall take care of the headroom and offsets.
> >
> > Hmm, and how it will now how many bytes are in head-room, and how
> > many are in tail-room?
> > We either need to provide values for both, or assume that only tail-room is
> > available for the driver.
> I believe it should be starting point where output can be written till the end of buffer.

Right, that's:
base = rte_pktmbuf_mtod_offset(mb, void *, ofs); 

> There should not be any headroom and tailroom for raw buffers.

I am not talking about raw buffers, what I am saying that some space in the mbuf
might be already occupied, that's why we have data_off inside rte_mbuf, etc. 

> It should be mbuf->buf_len - ofs.

No, it should be:
len = mb->buf_len - rte_pktmbuf_headroom(m) - ofs;
Otherwise PMD can overwrite memory beyond its buf_len.