linux-decnet-user — Discussion for users of DECnet on Linux

Re: [Linux-decnet-user] Question about "Rejected by object" (DNSTAT_REJECTED)

[Steven W. <st...@gw...>]

Hi,

>=20
> Le Lundi 29 Septembre 2003 18:24, Steven Whitehouse a =E9crit :
> > > If you talk about the sockaddr_dn structure, i think it's correctly
> > > configured (it's a named object) :
> > >
> > >     sockaddr.sdn_family =3D AF_DECnet;
> > >     sockaddr.sdn_flags  =3D 0x00;
> > >     sockaddr.sdn_objnum  =3D 0x00; /* for a named object ? */
> > >     sockaddr.sdn_objname  =3D "TASK";
> > >     sockaddr.sdn_objnamel  =3D 4;
> > >     memcpy(sockaddr.sdn_add.a_addr, np->n_addr,2);
> > >     ...
> >
> > That looks ok, and you are right that objnum =3D 0 for named objects.
>=20
> In fact, since the last email, we have found that the ADA program rejec=
ted our=20
> connexion because of some data wanted from the optdata_dn->opt_data str=
ucture=20
> and that the failure was from the software and not from the network lay=
er.=20
> So, you were right.
>=20
> DNSTAT_REJECTED seems really generated by the software layer.
>=20
> But ! I have a question about the origin of that kind of structure=20
> (optdata_dn->opt_data, max size of 16 bytes) ? It's seems to be very=20
> different from ip network where you don't have such optional structure =
of=20
> data (i think). How is it used usually ?
>=20
> We used setsockopt()  witch DSO_CONDATA : Get/set connect data (struct=20
> optdata_dn)
>
Thats the usual way to do it. There is no equivalent in TCP/IP as you
say. The reason for it is so that your application can pass a few bytes
of data in a connect init message. You can "half accept" a socket in
DECnet where the listening end sends no response until after the accept()
call has returned. This allows the application to send an accept/reject
based on the contents of the connect init message and thus its useful
for the connecting application to be able to send some extra info.

> > It depends on what the other end expects to see. Mostly it appears th=
at
> > it is not optional from the DEC applications I've seen interacting
> > with Linux DECnet.
>=20
> For the moment, we need to use accounting information. It should depend=
 of=20
> applications.
>
Yes, it will.
=20
> > Well ADA has bindings to the POSIX network API afaik, so if you are u=
sing
> > those, all the features of Linux DECnet should be available. As for A=
DA
> > on VAXen, I have no experience at all.
>=20
> It's a very good information and from now, it seems to be correct. In f=
act,=20
> Linux seems perfect to recycle old VAX program, even written in ADA :-)=
 You,=20
> and other Decnet developpers, have done a really good job !
>
Well its not perfect, but we try :-)
=20
> > Ok. If you have a way to set up a correctly working connection (witho=
ut
> > Linux DECnet) you should be able to use tcpdump or some similar tool =
to
> > work out whats going on and reverse engineer it. All the packet forma=
ts
> > (bar one!) are listed in the DECnet documents. The one missing bit re=
lates
> > to the formatting of the "end user names" where the formats are shown=
, but
> > the uses are not explained that well - the Linux code knows when to s=
end
> > which type.
>=20
> It's the next step. Try to understand a little more all the packet laye=
r. We=20
> have found a tool (unfortunatly under Windows, and not gpl) which is ab=
le to=20
> dump decnet packet into xml files : guess that it would be much easier =
to=20
> understand packets. Of course, tcpdump is a good candidate.
>=20
> Thanks for your comments and your support,
> Lionel
>=20
Ok. Beware that some (all?) versions of tcpdump don't dump DECnet
correctly. You might find it easier to dump it out as raw data and
deocde it yourself using the specs - thats what I tend to do,

Steve.

Re: [Linux-decnet-user] Question about "Rejected by object" (DNSTAT_REJECTED)

[Lionel T. <lio...@fr...>]

Le Lundi 29 Septembre 2003 18:24, Steven Whitehouse a =E9crit :
> > If you talk about the sockaddr_dn structure, i think it's correctly
> > configured (it's a named object) :
> >
> >     sockaddr.sdn_family =3D AF_DECnet;
> >     sockaddr.sdn_flags  =3D 0x00;
> >     sockaddr.sdn_objnum  =3D 0x00; /* for a named object ? */
> >     sockaddr.sdn_objname  =3D "TASK";
> >     sockaddr.sdn_objnamel  =3D 4;
> >     memcpy(sockaddr.sdn_add.a_addr, np->n_addr,2);
> >     ...
>
> That looks ok, and you are right that objnum =3D 0 for named objects.

In fact, since the last email, we have found that the ADA program rejected =
our=20
connexion because of some data wanted from the optdata_dn->opt_data structu=
re=20
and that the failure was from the software and not from the network layer.=
=20
So, you were right.

DNSTAT_REJECTED seems really generated by the software layer.

But ! I have a question about the origin of that kind of structure=20
(optdata_dn->opt_data, max size of 16 bytes) ? It's seems to be very=20
different from ip network where you don't have such optional structure of=20
data (i think). How is it used usually ?

We used setsockopt()  witch DSO_CONDATA : Get/set connect data (struct=20
optdata_dn)

> It depends on what the other end expects to see. Mostly it appears that
> it is not optional from the DEC applications I've seen interacting
> with Linux DECnet.

=46or the moment, we need to use accounting information. It should depend o=
f=20
applications.

> Well ADA has bindings to the POSIX network API afaik, so if you are using
> those, all the features of Linux DECnet should be available. As for ADA
> on VAXen, I have no experience at all.

It's a very good information and from now, it seems to be correct. In fact,=
=20
Linux seems perfect to recycle old VAX program, even written in ADA :-) You=
,=20
and other Decnet developpers, have done a really good job !

> Ok. If you have a way to set up a correctly working connection (without
> Linux DECnet) you should be able to use tcpdump or some similar tool to
> work out whats going on and reverse engineer it. All the packet formats
> (bar one!) are listed in the DECnet documents. The one missing bit relates
> to the formatting of the "end user names" where the formats are shown, but
> the uses are not explained that well - the Linux code knows when to send
> which type.

It's the next step. Try to understand a little more all the packet layer. W=
e=20
have found a tool (unfortunatly under Windows, and not gpl) which is able t=
o=20
dump decnet packet into xml files : guess that it would be much easier to=20
understand packets. Of course, tcpdump is a good candidate.

Thanks for your comments and your support,
Lionel

Re: [Linux-decnet-user] Question about "Rejected by object" (DNSTAT_REJECTED)

[Steven W. <st...@gw...>]

Hi,

>=20
> Le Mercredi 24 Septembre 2003 10:45, Steven Whitehouse a =E9crit :
> > > I have a question about an error message : i have an object (server=
)
> > > running on a VAX machine (written in ADA) that i try to connect by =
a
> > > socket connection (a remote named object) from a LINUX box.
> >
> > Me neither :-)
>=20
> Bad for me :-)
> But thanks for the response !
>=20
> > Did you set the connect data and/or the access data before calling co=
nnect?
> > Not that it should case a crash if it wasn't set, but I would expect =
the
> > "object rejected connect" response if something didn't match there.
>=20
> If you talk about the sockaddr_dn structure, i think it's correctly con=
figured=20
> (it's a named object) :
>=20
>     sockaddr.sdn_family =3D AF_DECnet;
>     sockaddr.sdn_flags  =3D 0x00;
>     sockaddr.sdn_objnum  =3D 0x00; /* for a named object ? */
>     sockaddr.sdn_objname  =3D "TASK";
>     sockaddr.sdn_objnamel  =3D 4;
>     memcpy(sockaddr.sdn_add.a_addr, np->n_addr,2);
>     ...
>=20
> etc ....
>
That looks ok, and you are right that objnum =3D 0 for named objects.
=20
> Even if we configure the accessdata_dn structure, it doesn't works (hop=
e it's=20
> optional ...).
>
It depends on what the other end expects to see. Mostly it appears that
it is not optional from the DEC applications I've seen interacting
with Linux DECnet.
=20
> For the moment, we are going to compile a single/simple server with all=
 the=20
> API used in the real one : TASK_QIOW (which called itself QIO). ADA is =
a very=20
> strange langage (for our point of view) and we need to understand much =
more=20
> how it works. But we hope there are no uncompatibility between both (AD=
A=20
> decnet layer and Linux layer) ...
>
Well ADA has bindings to the POSIX network API afaik, so if you are using
those, all the features of Linux DECnet should be available. As for ADA
on VAXen, I have no experience at all.
=20
> > I think thats the next step. You need to set them before the connect =
call
> > as certain defaults are assumed by the DECnet code if none are set by
> > the user. There are some pointers on the DECnet kernel pages as to wh=
at
> > the parameters are: http://www.chygwyn.com/DECnet/
>=20
> You have trying to change some parameters (with setsockopt), but withou=
t=20
> success. We need to continue :-)
>=20
> > Steve.
>=20
> Thanks,
> Lionel
>=20
Ok. If you have a way to set up a correctly working connection (without L=
inux
DECnet) you should be able to use tcpdump or some similar tool to work
out whats going on and reverse engineer it. All the packet formats (bar
one!) are listed in the DECnet documents. The one missing bit relates to
the formatting of the "end user names" where the formats are shown, but
the uses are not explained that well - the Linux code knows when to send
which type.

Steve.

Re: [Linux-decnet-user] Question about "Rejected by object" (DNSTAT_REJECTED)

[Lionel T. <lio...@fr...>]

Le Mercredi 24 Septembre 2003 10:45, Steven Whitehouse a =E9crit :
> > I have a question about an error message : i have an object (server)
> > running on a VAX machine (written in ADA) that i try to connect by a
> > socket connection (a remote named object) from a LINUX box.
>
> Me neither :-)

Bad for me :-)
But thanks for the response !

> Did you set the connect data and/or the access data before calling connec=
t?
> Not that it should case a crash if it wasn't set, but I would expect the
> "object rejected connect" response if something didn't match there.

If you talk about the sockaddr_dn structure, i think it's correctly configu=
red=20
(it's a named object) :

    sockaddr.sdn_family =3D AF_DECnet;
    sockaddr.sdn_flags  =3D 0x00;
    sockaddr.sdn_objnum  =3D 0x00; /* for a named object ? */
    sockaddr.sdn_objname  =3D "TASK";
    sockaddr.sdn_objnamel  =3D 4;
    memcpy(sockaddr.sdn_add.a_addr, np->n_addr,2);
    ...

etc ....

Even if we configure the accessdata_dn structure, it doesn't works (hope it=
's=20
optional ...).

=46or the moment, we are going to compile a single/simple server with all t=
he=20
API used in the real one : TASK_QIOW (which called itself QIO). ADA is a ve=
ry=20
strange langage (for our point of view) and we need to understand much more=
=20
how it works. But we hope there are no uncompatibility between both (ADA=20
decnet layer and Linux layer) ...

> I think thats the next step. You need to set them before the connect call
> as certain defaults are assumed by the DECnet code if none are set by
> the user. There are some pointers on the DECnet kernel pages as to what
> the parameters are: http://www.chygwyn.com/DECnet/

You have trying to change some parameters (with setsockopt), but without=20
success. We need to continue :-)

> Steve.

Thanks,
Lionel

Re: [Linux-decnet-user] Question about "Rejected by object" (DNSTAT_REJECTED)

[Steven W. <st...@gw...>]

Hi,

> 
> Howdy all,
> 
> I have a question about an error message : i have an object (server) running 
> on a VAX machine (written in ADA) that i try to connect by a socket 
> connection (a remote named object) from a LINUX box.  I use the "socket" 
> function (args : AF_DECNET,SOCK_SEQPACKET, DNPROTO_NSP), and after i call the 
> "connect" function with sockadd structure filled with family=AD_DECNET, 
> FLAGS=0x00, OBJNUM=0, OBJNAMEL=4 and OBJNAME="TASK". The remote server 
> receive a connection but crash with an obscur "exception" (i'm not familiar 
> with ADA!).
>
Me neither :-)

Did you set the connect data and/or the access data before calling connect?
Not that it should case a crash if it wasn't set, but I would expect the
"object rejected connect" response if something didn't match there.
 
> On the client side, i try to trap the error by using getsockopt like following 
> : getsockopt(sockfd, DNPROTO_NSP, DSO_DISDATA, &optdata, &len). The 
> opt_status return DNSTAT_REJECTED. I have to say that i have to create a non 
> blocking socket with fcntl(sockfd,F_SETFL, ONONBLOCK) in order to get that 
> kind of information. A blocking socket is failing too but i have no 
> possibility to get an error code.
> 
> So, if someone know something about that kind of error, he would be welcome! I 
> don't understand how networking works under ADA and, for the moment, the 
> source code is really difficult to read. In particular, parts about 
> networking.
> 
> Hope that decnet socket is very close to the linux side under ADA : someone 
> already use that kind of stuff ?
> 
> For the moment, i try to understand the setsockopt() arguments to modify 
> parameters of the socket, but it's not very clear.
> 
> Thanks,
> Lionel Tricon
>
I think thats the next step. You need to set them before the connect call
as certain defaults are assumed by the DECnet code if none are set by
the user. There are some pointers on the DECnet kernel pages as to what
the parameters are: http://www.chygwyn.com/DECnet/

Steve.

[Linux-decnet-user] Question about "Rejected by object" (DNSTAT_REJECTED)

[Lionel T. <lio...@fr...>]

Howdy all,

I have a question about an error message : i have an object (server) running 
on a VAX machine (written in ADA) that i try to connect by a socket 
connection (a remote named object) from a LINUX box.  I use the "socket" 
function (args : AF_DECNET,SOCK_SEQPACKET, DNPROTO_NSP), and after i call the 
"connect" function with sockadd structure filled with family=AD_DECNET, 
FLAGS=0x00, OBJNUM=0, OBJNAMEL=4 and OBJNAME="TASK". The remote server 
receive a connection but crash with an obscur "exception" (i'm not familiar 
with ADA!).

On the client side, i try to trap the error by using getsockopt like following 
: getsockopt(sockfd, DNPROTO_NSP, DSO_DISDATA, &optdata, &len). The 
opt_status return DNSTAT_REJECTED. I have to say that i have to create a non 
blocking socket with fcntl(sockfd,F_SETFL, ONONBLOCK) in order to get that 
kind of information. A blocking socket is failing too but i have no 
possibility to get an error code.

So, if someone know something about that kind of error, he would be welcome! I 
don't understand how networking works under ADA and, for the moment, the 
source code is really difficult to read. In particular, parts about 
networking.

Hope that decnet socket is very close to the linux side under ADA : someone 
already use that kind of stuff ?

For the moment, i try to understand the setsockopt() arguments to modify 
parameters of the socket, but it's not very clear.

Thanks,
Lionel Tricon

Re: RE : RE : [Linux-decnet-user] X and Decnet Freeze

[Steven W. <st...@gw...>]

Hi,

Its always useful to have this kind of information as even if it doesn't
help us work out what it is right now, it may provide a clue in the
future,

Steve.

> 
> Hi,
> 
> Concerning the Xfree freeze problem, i have do some test. It's appear that the Xserver does not freeze when i use the Xfree 3.3.6 version (Mandrake 9.1) with Fbdev driver.
> If this information can help you ....
> 
> M.Bassir
> 

RE : RE : [Linux-decnet-user] X and Decnet Freeze

[BASSIR M. <Moh...@ax...>]

Hi,

Concerning the Xfree freeze problem, i have do some test. It's appear =
that the Xserver does not freeze when i use the Xfree 3.3.6 version =
(Mandrake 9.1) with Fbdev driver.
If this information can help you ....

M.Bassir

Re: [Linux-decnet-user] Having troubles....

[Steven W. <st...@gw...>]

Hi,

> 
> Hi
> 
> In the past ive had decent working great.....that might have been on
> mandrake 9....anyway, im now using mandrake 9.1 and 9.2 adn i get the same
> erro on both.
> 
> The kernel version of 9.2 is 2.4.21, 9.1 is 2.4.20
> 
> I compiled 2.23.2 from source
> 
> Ive set up the scripts and everything looks ok.  I copied setether.sh into
> /usr/local/sbin and use that from /etc/init.d/decnet to set the macaddress
> of the card, the node address and default_route using the /proc filesystem.
> 
> Ive can see the other machine on the network (vms 5) in
> /proc/net/decnet_neigh.
>
Assuming that your ethernet card was correctly showing in /proc/net/decnet_dev
then:
 
> But when i dnping it i get
> Loopback error
> write: transport endpont is not connected
>
is rather unusual since it normally means a routing error has prevented
a connection being set up.

If you can't solve the problem any other way, do an strace of a broken
binary (something simple like dnping is best) and a tcpdump if any
traffic is exchanged. Also the output of an ifconfig eth0 would be
good too... this is the give away if Mandrake have done the same thing
as Red Hat,

Steve.

Re: [Linux-decnet-user] troubles with kernel 2.4.18 ?

[Patrick C. <pa...@ty...>]

On Fri, Sep 12, 2003 at 08:30:07PM +0000, Lionel Tricon wrote:
> Le Mercredi 10 Septembre 2003 07:04, Patrick Caulfield a écrit :
> > I don't know how distros build their kernels, but I seem to remember
> > hearing that the Red Hat kernels shipped with a broken DECnet simply
> > because they had enabled all the options rather than actually reading what
> > they meant :-)
> 
> One thing interresting is that it was correctly working between two Linux node 
> (SuSE 8.0) ... and the VAX correctly talking to an another real decnet node.
> 

Hmm, which makes me wonder whether this is a compiler problem. I do know that
some versions of GCC did not compile the DECnet code correctly on HP (it just
used to oops). Now that could just be a backend issue of course.

-- 

patrick

Re: [Linux-decnet-user] troubles with kernel 2.4.18 ?

[Lionel T. <lio...@fr...>]

Le Mercredi 10 Septembre 2003 07:04, Patrick Caulfield a =E9crit :
> I don't know how distros build their kernels, but I seem to remember
> hearing that the Red Hat kernels shipped with a broken DECnet simply
> because they had enabled all the options rather than actually reading what
> they meant :-)

One thing interresting is that it was correctly working between two Linux n=
ode=20
(SuSE 8.0) ... and the VAX correctly talking to an another real decnet node.

Just my two (euro) cents,
Lionel

Re: [Linux-decnet-user] Having troubles....

[Patrick C. <pa...@ty...>]

On Thu, Sep 11, 2003 at 03:43:01PM +0000, ada...@bn... wrote:
> Hi
> 
> In the past ive had decent working great.....that might have been on
> mandrake 9....anyway, im now using mandrake 9.1 and 9.2 adn i get the same
> erro on both.
> 
> The kernel version of 9.2 is 2.4.21, 9.1 is 2.4.20
> 
> I compiled 2.23.2 from source
> 
> Ive set up the scripts and everything looks ok.  I copied setether.sh into
> /usr/local/sbin and use that from /etc/init.d/decnet to set the macaddress
> of the card, the node address and default_route using the /proc filesystem.
> 
> Ive can see the other machine on the network (vms 5) in
> /proc/net/decnet_neigh.
> 
> But when i dnping it i get
> Loopback error
> write: transport endpont is not connected
> 
> When i dndir using a known account and password i get
> read failed: rejected by object
> error in config: read failed: rejected by object
> 
> When i sethost i get
> ct_setup_link: not bind from host
> 
> If i dnping myself i get
> write: transport endpont is not connected
> 
> but sethost and dndir to myself work.
> 
> Any ideas?
> 

I'm wondering if this is related to the kernel problems some other seem to have
seen with Red Hat and SuSE kernels. is it possible to compile a "Linus" kernel
of your own to try? Or maybe just try the Mandrake 9.0 kernel (the one that
worked) to eliminate any other variables.
-- 

patrick

[Linux-decnet-user] Having troubles....

[<ada...@bn...>]

Hi

In the past ive had decent working great.....that might have been on
mandrake 9....anyway, im now using mandrake 9.1 and 9.2 adn i get the same
erro on both.

The kernel version of 9.2 is 2.4.21, 9.1 is 2.4.20

I compiled 2.23.2 from source

Ive set up the scripts and everything looks ok.  I copied setether.sh into
/usr/local/sbin and use that from /etc/init.d/decnet to set the macaddress
of the card, the node address and default_route using the /proc filesystem.

Ive can see the other machine on the network (vms 5) in
/proc/net/decnet_neigh.

But when i dnping it i get
Loopback error
write: transport endpont is not connected

When i dndir using a known account and password i get
read failed: rejected by object
error in config: read failed: rejected by object

When i sethost i get
ct_setup_link: not bind from host

If i dnping myself i get
write: transport endpont is not connected

but sethost and dndir to myself work.

Any ideas?

Cheers

Adam Pigg






The information contained in this email may be commercially sensitive and/or legally privileged. It is intended solely for the person(s) to whom it is addressed. If you are not a named recipient, you are on notice of its status. Please notify the sender immediately by reply e-mail and then delete this message from your system. You must not disclose it to any other person, copy or distribute it or use it for any purpose.

Re: [Linux-decnet-user] troubles with kernel 2.4.18 ?

[Patrick C. <pa...@ty...>]

On Tue, Sep 09, 2003 at 10:13:04PM +0000, Lionel Tricon wrote:
> Hi all,
> 
> Intrigued, i installed a new version of the SuSE (8.2) with a 2.4.20 kernel 
> (NET 4: DECnet for Linux: v2.4.20-pre1s), and (thanks Linux hackers!) the 
> communication is now fully correct (the dnping/LOOP commands works greats).
> 
> So, my question is : is there a problem with Linux Decnet on 2.4.18 kernel (or 
> SuSE flavour) or it's an another problem ?

I don't know how distros build their kernels, but I seem to remember hearing
that the Red Hat kernels shipped with a broken DECnet simply because they had
enabled all the options rather than actually reading what they meant :-)

This may have been the case with earlier SuSE kernels but I can't say for sure.

Thanks for letting us know though, it's worth knowing things like that for
future reference.

-- 

patrick

[Linux-decnet-user] troubles with kernel 2.4.18 ?

[Lionel T. <lio...@fr...>]

Hi all,

Last week, i've got troubles with a decnet connection between a VAX 4000-60 
and a Linux 2.4.18 with a recent dnprogs tools installed (version 2.23.2). It 
was in fact impossible to communicate between these two nodes ("dnping" on 
Linux side and "loop node ..." on the VAX side) both connected with a BNC 
cable :

* VAX 4000-60 (OpenVMS 6.1)
* End node : 45.100

and

* LINUX SuSE 8.0 with a 2.4.18 kernel
* Kernel side : NET 4: DECnet for Linux: v2.4.15-pre5s
* End node : 45.200

At the beginning, i guessed that it was a stupid network tuning problem but 
with the tcpdump tool, i realised that a lot of decnet packets (hello 
packets) were in fact generated from the VAX into Linux box without any 
response. On the contrary, no packets were generated from the Linux side 
despite dnping command.

Intrigued, i installed a new version of the SuSE (8.2) with a 2.4.20 kernel 
(NET 4: DECnet for Linux: v2.4.20-pre1s), and (thanks Linux hackers!) the 
communication is now fully correct (the dnping/LOOP commands works greats).

So, my question is : is there a problem with Linux Decnet on 2.4.18 kernel (or 
SuSE flavour) or it's an another problem ?

Thanks by advance,
Lionel Tricon

Re: RE : RE : [Linux-decnet-user] X and Decnet Freeze

[Steven W. <st...@gw...>]

Hi,

Thats the 2.4 version of the patch I was referring to. The problem is tha=
t
for whatever reason, DECnet, probably in its sendmsg() and/or recvmsg()
routines returns an error which is unexpected by X and hence X breaks.

What really needs doing for 2.5, is that both these two routines and the
code paths need rewriting. This would not just solve this problem, but
probably a bunch of others and certainly it would allow support for
zerocopy at the same time (sendpage()).

It is just possible that there is another problem as well, but I think th=
is
is the most likely and Patrick's patches prove where the problem lies to
a certain extent. If you are looking for a good model of what the routine=
s
should look like, look in linux/net/ipv4/tcp.c at the equivalent tcp rout=
ines
taking into account that we need support for SOCK_SEQPACKET as well as
SOCK_STREAM.

I've been working towards this goal for a while and most of the other rec=
ent
patches in 2.5 have been designed to pave the way for these changes. If y=
ou
want to do 2.5 patches, then I suggest starting at the recvmsg() end sinc=
e
that needs to be changed in order to test a new sendmsg() routine! Breaki=
ng
down the changes into small bits is also a good idea where at all possibl=
e,

Steve.

>=20
> Ok, if i think well, it seems the same freeze problem appear on the 2.5=
. So it is not a good solution for me to increase the 2.4 version to 2.5 =
with all the patchs.
>=20
> At this time i'm in locked state.=20
> Could you tell me the Patrick patch which reduces the number of hangs a=
 lot ?=20
> Is it a patch to add to the last send by Patrick ?
>=20
> **************
> Last Patrick patch:
>=20
> --- net/decnet/af_decnet.c.old  2002-11-04 14:12:26.000000000 -0200
> +++ net/decnet/af_decnet.c      2002-11-04 12:45:24.000000000 -0200
> @@ -1771,7 +1771,7 @@
>                         goto out;
>=20
>                 if (signal_pending(current)) {
> -                       rv =3D -ERESTARTSYS;
> +                       rv =3D -EAGAIN;
>                         goto out;
>                 }
>=20
> @@ -1964,7 +1964,8 @@
>                         goto out;
>=20
>                 if (signal_pending(current)) {
> -                       err =3D -ERESTARTSYS;
> +                       /* VMH - testing X crash - printk("DECnet 6\n")=
; */
> +                       err =3D -EAGAIN;
>                         goto out;
>                 }
> ********************
>=20
> Could you explain to me the problem state and if i can help you to do s=
ome test or anything, i'm ready to do it.
>=20
> Thanks,
> M.Bassir
>=20
>=20
> > -----Message d'origine-----
> > De : Steven Whitehouse [mailto:st...@gw...]=20
> > Envoy=E9 : vendredi 5 septembre 2003 12:07
> > =C0 : BASSIR Mohcine
> > Cc : Patrick Caulfield; DECnet list
> > Objet : Re: RE : [Linux-decnet-user] X and Decnet Freeze
> >=20
> >=20
> > Hi,
> >=20
> > >=20
> > > Hi,
> > >=20
> > > I want to solve my XFree freeze problem. For this i think it was=20
> > > necessary to apply all the kernel patch that was write=20
> > since the the=20
> > > 2.4.18 kernel version. Could any one tell me where i can=20
> > found a list=20
> > > of all the patch since this version ? If it is possible i want to=20
> > > increase the decnet part of the kernel to the same state of=20
> > the 2.5.x=20
> > > version.
> > >=20
> > > Steven,
> > >=20
> > > According to your last answare you tell me to wait about=20
> > the porting=20
> > > kernel. Could you telle me the result ?
> >=20
> > Yes. Sorry its been so long - things have got very busy for=20
> > me recently at work and all the DECnet stuff is really just a=20
> > spare time activity for me at the moment.
> >=20
> > It appears that the 2.5 kernel also suffers from the same=20
> > problems as 2.4 when it comes to XFree freezing as Patrick=20
> > has recently been investigating it. He sent me one patch=20
> > which reduces the number of hangs a lot, but still (as I=20
> > understand) does not cure it entirely.
> >=20
> > There are two places to look for patches to DECnet in 2.5.=20
> > Firstly my patches page http://www.chygwyn.com/~steve/kpatch/=20
> > has most if not all of the ones I've written and=20
> http://linux.bkbits.net:8080/linux-2.5
> can be used to query all patches applied to DECnet, whoever wrote them =
and might be the easier reference for you.
>=20
> By all means feel free to produce patches for DECnet in 2.4 but I suspe=
ct it will be a long job since the two kernels are different enough that =
you won't just be able to copy & paste parts. Also beware that some behav=
iour has changed and that I'd rather not back port anything to 2.4 which =
will change DECnet from the users perspective unless it can be shown to b=
e 100% backwards compatible.
>=20
> When 2.6 proper comes out, I'll probably want to give up maintaining DE=
Cnet in 2.4. I think one stable version and one development version at a =
time are enough and there are times when I think that I'm not really givi=
ng them the attention that they deserve. I'm expecting things at work to =
calm down in a month or two and I hope to do some of the larger tasks I h=
ave on my todo list then,
>=20
> Steve.
>=20

Re: RE : RE : [Linux-decnet-user] X and Decnet Freeze

[Patrick C. <pa...@ty...>]

On Fri, Sep 05, 2003 at 12:35:13PM +0200, BASSIR Mohcine wrote:
> Ok, if i think well, it seems the same freeze problem appear on the 2.5. So it is not a good solution for me to increase the 2.4 version to 2.5 with all the patchs.
> 
> At this time i'm in locked state. 
> Could you tell me the Patrick patch which reduces the number of hangs a lot ? 
> Is it a patch to add to the last send by Patrick ?
> 
> **************
> Last Patrick patch:
> 
> --- net/decnet/af_decnet.c.old  2002-11-04 14:12:26.000000000 -0200
> +++ net/decnet/af_decnet.c      2002-11-04 12:45:24.000000000 -0200
> @@ -1771,7 +1771,7 @@
>                         goto out;
> 
>                 if (signal_pending(current)) {
> -                       rv = -ERESTARTSYS;
> +                       rv = -EAGAIN;
>                         goto out;
>                 }
> 
> @@ -1964,7 +1964,8 @@
>                         goto out;
> 
>                 if (signal_pending(current)) {
> -                       err = -ERESTARTSYS;
> +                       /* VMH - testing X crash - printk("DECnet 6\n"); */
> +                       err = -EAGAIN;
>                         goto out;
>                 }
> ********************

That's the patch I'm using at the moment. The changes for 2.6 are similar (with
an additional one for dnetd).

With that in place I can run X sessions for several hours at a time but they
still can die. I haven't done any more investigation into this. I suppose what
needs doing is to strace the Xserver for a whole day, or until the problem
happens and then have a look at the last few (hundred!) lines.


patrick

RE : RE : [Linux-decnet-user] X and Decnet Freeze

[BASSIR M. <Moh...@ax...>]

Ok, if i think well, it seems the same freeze problem appear on the 2.5. =
So it is not a good solution for me to increase the 2.4 version to 2.5 =
with all the patchs.

At this time i'm in locked state.=20
Could you tell me the Patrick patch which reduces the number of hangs a =
lot ?=20
Is it a patch to add to the last send by Patrick ?

**************
Last Patrick patch:

--- net/decnet/af_decnet.c.old  2002-11-04 14:12:26.000000000 -0200
+++ net/decnet/af_decnet.c      2002-11-04 12:45:24.000000000 -0200
@@ -1771,7 +1771,7 @@
                        goto out;

                if (signal_pending(current)) {
-                       rv =3D -ERESTARTSYS;
+                       rv =3D -EAGAIN;
                        goto out;
                }

@@ -1964,7 +1964,8 @@
                        goto out;

                if (signal_pending(current)) {
-                       err =3D -ERESTARTSYS;
+                       /* VMH - testing X crash - printk("DECnet 6\n"); =
*/
+                       err =3D -EAGAIN;
                        goto out;
                }
********************

Could you explain to me the problem state and if i can help you to do =
some test or anything, i'm ready to do it.

Thanks,
M.Bassir


> -----Message d'origine-----
> De : Steven Whitehouse [mailto:st...@gw...]=20
> Envoy=E9 : vendredi 5 septembre 2003 12:07
> =C0 : BASSIR Mohcine
> Cc : Patrick Caulfield; DECnet list
> Objet : Re: RE : [Linux-decnet-user] X and Decnet Freeze
>=20
>=20
> Hi,
>=20
> >=20
> > Hi,
> >=20
> > I want to solve my XFree freeze problem. For this i think it was=20
> > necessary to apply all the kernel patch that was write=20
> since the the=20
> > 2.4.18 kernel version. Could any one tell me where i can=20
> found a list=20
> > of all the patch since this version ? If it is possible i want to=20
> > increase the decnet part of the kernel to the same state of=20
> the 2.5.x=20
> > version.
> >=20
> > Steven,
> >=20
> > According to your last answare you tell me to wait about=20
> the porting=20
> > kernel. Could you telle me the result ?
>=20
> Yes. Sorry its been so long - things have got very busy for=20
> me recently at work and all the DECnet stuff is really just a=20
> spare time activity for me at the moment.
>=20
> It appears that the 2.5 kernel also suffers from the same=20
> problems as 2.4 when it comes to XFree freezing as Patrick=20
> has recently been investigating it. He sent me one patch=20
> which reduces the number of hangs a lot, but still (as I=20
> understand) does not cure it entirely.
>=20
> There are two places to look for patches to DECnet in 2.5.=20
> Firstly my patches page http://www.chygwyn.com/~steve/kpatch/=20
> has most if not all of the ones I've written and=20
http://linux.bkbits.net:8080/linux-2.5
can be used to query all patches applied to DECnet, whoever wrote them =
and might be the easier reference for you.

By all means feel free to produce patches for DECnet in 2.4 but I =
suspect it will be a long job since the two kernels are different enough =
that you won't just be able to copy & paste parts. Also beware that some =
behaviour has changed and that I'd rather not back port anything to 2.4 =
which will change DECnet from the users perspective unless it can be =
shown to be 100% backwards compatible.

When 2.6 proper comes out, I'll probably want to give up maintaining =
DECnet in 2.4. I think one stable version and one development version at =
a time are enough and there are times when I think that I'm not really =
giving them the attention that they deserve. I'm expecting things at =
work to calm down in a month or two and I hope to do some of the larger =
tasks I have on my todo list then,

Steve.

Re: RE : [Linux-decnet-user] X and Decnet Freeze

[Steven W. <st...@gw...>]

Hi,

> 
> Hi,
> 
> I want to solve my XFree freeze problem. For this i think it was necessary to apply all the kernel patch that was write since the the 2.4.18 kernel version.
> Could any one tell me where i can found a list of all the patch since this version ?
> If it is possible i want to increase the decnet part of the kernel to the same state of the 2.5.x version.
> 
> Steven,
> 
> According to your last answare you tell me to wait about the porting kernel. Could you telle me the result ?

Yes. Sorry its been so long - things have got very busy for me recently at
work and all the DECnet stuff is really just a spare time activity for me
at the moment.

It appears that the 2.5 kernel also suffers from the same problems as 2.4
when it comes to XFree freezing as Patrick has recently been investigating
it. He sent me one patch which reduces the number of hangs a lot, but still
(as I understand) does not cure it entirely.

There are two places to look for patches to DECnet in 2.5. Firstly my
patches page http://www.chygwyn.com/~steve/kpatch/ has most if not all
of the ones I've written and http://linux.bkbits.net:8080/linux-2.5
can be used to query all patches applied to DECnet, whoever wrote them
and might be the easier reference for you.

By all means feel free to produce patches for DECnet in 2.4 but I suspect
it will be a long job since the two kernels are different enough that
you won't just be able to copy & paste parts. Also beware that some
behaviour has changed and that I'd rather not back port anything to 2.4 which
will change DECnet from the users perspective unless it can be shown to be
100% backwards compatible.

When 2.6 proper comes out, I'll probably want to give up maintaining DECnet
in 2.4. I think one stable version and one development version at a time
are enough and there are times when I think that I'm not really giving them
the attention that they deserve. I'm expecting things at work to calm down
in a month or two and I hope to do some of the larger tasks I have on
my todo list then,

Steve.

RE : [Linux-decnet-user] X and Decnet Freeze

[BASSIR M. <Moh...@ax...>]

Hi,

I want to solve my XFree freeze problem. For this i think it was =
necessary to apply all the kernel patch that was write since the the =
2.4.18 kernel version.
Could any one tell me where i can found a list of all the patch since =
this version ?
If it is possible i want to increase the decnet part of the kernel to =
the same state of the 2.5.x version.

Steven,

According to your last answare you tell me to wait about the porting =
kernel. Could you telle me the result ?
*************
Steven last answare:
	Hi,

	If you wait for a bit I might get around to porting it. At the moment =
I've got a patch (posted a few 	days ago on this list) which fixed some =
of the issues, though probably not the one you mentioned.

	I want to be sure that when I back port things, that they really do =
work so the best way to speed up 	the process is to test 2.5 out and let =
me know which bits work for you and which bits break.

	Currently there is still part of my recent 2.5 patch thats outstanding. =
I'm going to send it to 	davem as soon as the "part 1" patch has made it =
into the -bk snapshots so that I can send him an 	uptodate patch. I'm =
afraid that I still use patch/diff and that I haven't made it as far as =
using bk 	yet.

	Alternatively, since all the patches are available, you could do the =
port yourself if you are 	feeling enthusiastic :-)

	Steve.
************

Thanks for your answare.

M.Bassir

[Linux-decnet-user] Kernel with dapfs

[<ce...@pa...>]

Hi,
tell me please why if I want to build dnmount from dnprogs-2.23.2 then dn=
mount is not created ?
Before make dnprogs I patched the kernel with dapfs.diff. I hope that cor=
rectly :-).
How can i identify support for dapfs on my kernel ? 

Thanks.
Josef