Re: [RTnet-developers] Patches for SOCKET_RAW and ETH_P_ALL

[Jan K. <jan...@we...>]

Jorge Almeida wrote:
> OPS Sorry :)
>=20
> There it goes :)
>=20
>=20
> JAN,
>=20
> In attach goes the patches for the SOCKET_RAW and ETH_P_ALL functionali=
ties for RTnet.
> The only thing that was not done was the configurable option for the ET=
H_P_ALL.=20
>=20

Thanks, will have a look at them later.

>=20
> After some tests in the same computer, with two different cards, i find=
 something interesting that i need to talk with you (or anybody that that=
 wants to participate in the discution).

Both NICs attached to the same network?

> I have two small test programs, one for sending, and another one to rec=
eive the message. In both there's one socket opened with the ETH_P_ALL fe=
ature. Making two different filedescriptors in the same machine.
> After i send the message it is sended for both sockets (REmenber that w=
e are in the same machine), but i only read from one.=20
> This means, after some time the queue for the opened socket used to wri=
te only (not read) will fill.=20
> Wath is the behaviour for this problem?

Buffers for TX and RX are taken from the same pool. Thus, once incoming
unhandled packets consumed all of them, TX will fail.

>=20
> This queue should fill and start to give errors when no more memory is =
available?
> Or we should not send the message to one sockt thas is only writing? Ho=
w can we know if the socket is only writting?
> Or we can set a maximum value to the queued message in one determined q=
ueue? (I think this is the best option)

Simple solution: if you have a send-only socket, don't open it with
ETH_P_ALL. Use some arbitrary unused protocol instead. This also helps
to reduce the system load due to unnecessary packet delivery via costly
cloning.

Jan