Re: Socket related stuff - patch available for testing
Looks good Matt,
Just my quick review,
I didn't notice anything obvious, although patch is fairly large.
Cheers,
Edward.
2009/11/12 Matthew Dillon <dillon@apollo.backplane.com>:
> =A0 =A0Here's a patch, it needs some serious testing:
>
> =A0 =A0 =A0 =A0fetch http://apollo.backplane.com/DFlyMisc/tcp01.patch
>
> =A0 =A0The patch:
>
> =A0 =A0 =A0 =A0* Moves the socket pointer into the netmsg base structure,=
even
> =A0 =A0 =A0 =A0 =A0though some things don't use it, I know
>
> =A0 =A0 =A0 =A0* Adds a lwkt_port pointer to the socket structure and ini=
tializes
> =A0 =A0 =A0 =A0 =A0it to cpu0_soport()
>
> =A0 =A0 =A0 =A0* Expects network protocols to set so->so_port in their at=
tach
> =A0 =A0 =A0 =A0 =A0functions, plus I do that for tcp and udp.
>
> =A0 =A0 =A0 =A0* Performs message chasing when the protocol port changes =
due to
> =A0 =A0 =A0 =A0 =A0e.g. =A0a connect() or an implied connect or (I think)=
also an
> =A0 =A0 =A0 =A0 =A0implied binding to INADDR_ANY.
>
> =A0 =A0 =A0 =A0 =A0If a number of messages for a socket have built up on =
a protocol
> =A0 =A0 =A0 =A0 =A0thread and some operation in the protocol changes the =
socket's
> =A0 =A0 =A0 =A0 =A0protocol thread (aka implied connect), then any other =
messages
> =A0 =A0 =A0 =A0 =A0queued to that protocol thread, or new messages which =
race the
> =A0 =A0 =A0 =A0 =A0change, will automatically be forwarded to the correct=
protocol
> =A0 =A0 =A0 =A0 =A0thread when they are encountered.
>
> =A0 =A0 =A0 =A0* Fixes implied connects for TCP. =A0This is when you use =
sendmsg()
> =A0 =A0 =A0 =A0 =A0with an address to imply a connect along with data, so=
data can
> =A0 =A0 =A0 =A0 =A0be sent along with the SYN.
>
> =A0 =A0 =A0 =A0 =A0Implied connects were completely broken and would cras=
h the kernel.
> =A0 =A0 =A0 =A0 =A0Example: =A0finger user@target =A0(instant crash). =A0=
finger uses the
> =A0 =A0 =A0 =A0 =A0implied connect feature.
>
> =A0 =A0 =A0 =A0* Cleans up a bunch of stuff. =A0I was almost able to remo=
ve
> =A0 =A0 =A0 =A0 =A0the proto->pr_mport field entirely but sendmsg() still=
needs
> =A0 =A0 =A0 =A0 =A0to use it in the case where the passed address is non-=
NULL (aka
> =A0 =A0 =A0 =A0 =A0sendto() style). =A0All other code that used to call p=
roto->pr_mport()
> =A0 =A0 =A0 =A0 =A0now simply snarf the port out of so->so_port.
>
> =A0 =A0 =A0 =A0 =A0The mport functions themselves now basically just retu=
rn so_port.
> =A0 =A0 =A0 =A0 =A0I left the sendmsg code intact just in case we wanted =
to optimize
> =A0 =A0 =A0 =A0 =A0it later on for UDP.
>
> =A0 =A0I found a few other bugs in the code but haven't fixed them yet. =
=A0UDP
> =A0 =A0is not MPSAFE due to the global inpcbinfo (udbinfo) structure it u=
ses.
>
> =A0 =A0This patch does need testing. =A0I don't know what I might have bl=
own up.
> =A0 =A0It's fairly straight forward so I would also appreciate a code rev=
iew.
>
> =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 =
=A0 =A0 =A0 =A0 =A0 =A0-Matt
>
>
--=20
--=20
Edward O'Callaghan
http://www.auroraux.org/
eocallaghan at auroraux dot org
討論串 (同標題文章)
完整討論串 (本文為第 2 之 11 篇)：