[PATCH] nvmet-tcp: bound sgl->length check in nvmet_tcp_map_data()

[Shivam Kumar]

Hi Christoph, Maurizio

Just following up on this thread.
Would you like me to send a v2 patch that sets a sane default MDTS for
the TCP transport? If so, is there a preferred value I should use, or
should I follow what the other transports do?

Thanks,
Shivam Kumar

On Fri, Mar 20, 2026 at 3:48 AM Christoph Hellwig <hch at lst.de> wrote:
>
> On Thu, Mar 19, 2026 at 02:00:27PM -0400, Shivam Kumar wrote:
> > Hi Maurizio
> >
> > Thank you for the review. You're right, I was conflating the per-PDU
> > H2CData limit with the total transfer size for the command.
> > And the error handling should trigger a fatal transport error rather
> > than completing the request with an NVMe status code.
> >
> > Would setting a sane default for MDTS be the preferred approach here?
> > If so, I'm happy to send a v2 implementing that instead.
>
> Yes, if we want to limit the I/O size we have to limit MDTS.  Which all
> other transports do anyway, so I'm kinda surprised TCP doesn't.
>