On Tue, Sep 10, 2019 at 4:39 PM Eric Dumazet <eduma...@google.com> wrote:
>
> On Tue, Sep 10, 2019 at 10:11 PM Thomas Higdon <t...@fb.com> wrote:
> >
> >
> ...
> > Because an additional 32-bit member in struct tcp_info would cause
> > a hole on 64-bit systems, we reserve a struct member '_reserved'.
> ...
> > diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
> > index b3564f85a762..990a5bae3ac1 100644
> > --- a/include/uapi/linux/tcp.h
> > +++ b/include/uapi/linux/tcp.h
> > @@ -270,6 +270,9 @@ struct tcp_info {
> > __u64 tcpi_bytes_retrans; /* RFC4898 tcpEStatsPerfOctetsRetrans
> > */
> > __u32 tcpi_dsack_dups; /* RFC4898 tcpEStatsStackDSACKDups */
> > __u32 tcpi_reord_seen; /* reordering events seen */
> > +
> > + __u32 _reserved; /* Reserved for future 32-bit member.
> > */
> > + __u32 tcpi_rcv_ooopack; /* Out-of-order packets received */
> > };
> >
>
> Unfortunately we won't be able to use this hole, because the way the
> TCP_INFO works,
>
> The kernel will report the same size after the reserved field is
> renamed to something else.
>
> User space code is able to detect which fields are there or not based
> on what the kernel
> returns for the size of the structure.
If we are looking for something else useful to use for a __u32 to pair
up with this new field, I would suggest we export tp->snd_wnd
(send-side receive window) and tp->rcv_wnd (receive-side receive
window) in tcp_info. We could export one of them now, and the other
the next time we need to add a field and need some useful "padding".
These fields could help folks diagnose whether a flow is
receive-window-limited at a given instant, using "ss", etc. I think
there was some interest in this internally in our team a while ago.
neal
