Fw: 2.6.12-rc4-mm[12] - ULOG problem

Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Andrew Morton]

This is the second report of this happening.


Begin forwarded message:

Date: Wed, 18 May 2005 23:10:49 -0400
From: Nick Orlov <bugfixer@list.ru>
To: linux-kernel@vger.kernel.org
Cc: akpm@osdl.org
Subject: 2.6.12-rc4-mm[12] - ULOG problem


ipt_ULOG fails to load starting from 2.6.12-rc4-mm1 with the following
error message:


FATAL: Error inserting ipt_ULOG (/lib/modules/2.6.12-rc4-mm1-2/kernel/net/ipv4/netfilter/ipt_ULOG.ko): Cannot allocate memory

rc3-mm3 works fine.

Unfortunately, I wont have time to do a binary search till the weekend,
but I hope you'll point suspicious patch(es) without it.

Any additional info available by request.

-- 
With best wishes,
	Nick Orlov.

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Herbert Xu]

Andrew Morton <akpm@osdl.org> wrote:
> 
> This is the second report of this happening.
> 
> FATAL: Error inserting ipt_ULOG (/lib/modules/2.6.12-rc4-mm1-2/kernel/net/ipv4/netfilter/ipt_ULOG.ko): Cannot allocate memory

This probably means that someone has stolen the NETLINK_NFLOG id.

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Andrew Morton]

Herbert Xu <herbert@gondor.apana.org.au> wrote:
>
> Andrew Morton <akpm@osdl.org> wrote:
>  > 
>  > This is the second report of this happening.
>  > 
>  > FATAL: Error inserting ipt_ULOG (/lib/modules/2.6.12-rc4-mm1-2/kernel/net/ipv4/netfilter/ipt_ULOG.ko): Cannot allocate memory

Ah yes, thanks.  netlink_kernel_create() has a dumbish return-value protocol.

>  This probably means that someone has stolen the NETLINK_NFLOG id.

cn_init() uses NETLINK_NFLOG.  Peculiar.

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 810 bytes --]

On Wed, 2005-05-18 at 22:27 -0700, Andrew Morton wrote:
> Herbert Xu <herbert@gondor.apana.org.au> wrote:
> >
> > Andrew Morton <akpm@osdl.org> wrote:
> >  > 
> >  > This is the second report of this happening.
> >  > 
> >  > FATAL: Error inserting ipt_ULOG (/lib/modules/2.6.12-rc4-mm1-2/kernel/net/ipv4/netfilter/ipt_ULOG.ko): Cannot allocate memory
> 
> Ah yes, thanks.  netlink_kernel_create() has a dumbish return-value protocol.
> 
> >  This probably means that someone has stolen the NETLINK_NFLOG id.
> 
> cn_init() uses NETLINK_NFLOG.  Peculiar.

Yep...
One may use "unit" module parameter to specify different socket.
Should I ask David Miller to allocate new one specially for connector?

-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Andrew Morton]

Evgeniy Polyakov <johnpol@2ka.mipt.ru> wrote:
>
> On Wed, 2005-05-18 at 22:27 -0700, Andrew Morton wrote:
> > Herbert Xu <herbert@gondor.apana.org.au> wrote:
> > >
> > > Andrew Morton <akpm@osdl.org> wrote:
> > >  > 
> > >  > This is the second report of this happening.
> > >  > 
> > >  > FATAL: Error inserting ipt_ULOG (/lib/modules/2.6.12-rc4-mm1-2/kernel/net/ipv4/netfilter/ipt_ULOG.ko): Cannot allocate memory
> > 
> > Ah yes, thanks.  netlink_kernel_create() has a dumbish return-value protocol.
> > 
> > >  This probably means that someone has stolen the NETLINK_NFLOG id.
> > 
> > cn_init() uses NETLINK_NFLOG.  Peculiar.
> 
> Yep...

Bah.

> One may use "unit" module parameter to specify different socket.

Is there any reason for being able to do that?

> Should I ask David Miller to allocate new one specially for connector?

Dave?


diff -puN include/linux/netlink.h~connector-netlink-id-fix include/linux/netlink.h
--- 25/include/linux/netlink.h~connector-netlink-id-fix	2005-05-18 23:44:43.000000000 -0700
+++ 25-akpm/include/linux/netlink.h	2005-05-18 23:44:43.000000000 -0700
@@ -14,6 +14,7 @@
 #define NETLINK_SELINUX		7	/* SELinux event notifications */
 #define NETLINK_ARPD		8
 #define NETLINK_AUDIT		9	/* auditing */
+#define NETLINK_CONNECTOR	10	/* kernel connector */
 #define NETLINK_ROUTE6		11	/* af_inet6 route comm channel */
 #define NETLINK_IP6_FW		13
 #define NETLINK_DNRTMSG		14	/* DECnet routing messages */
diff -puN drivers/connector/connector.c~connector-netlink-id-fix drivers/connector/connector.c
--- 25/drivers/connector/connector.c~connector-netlink-id-fix	2005-05-18 23:44:43.000000000 -0700
+++ 25-akpm/drivers/connector/connector.c	2005-05-18 23:44:43.000000000 -0700
@@ -33,7 +33,7 @@ MODULE_LICENSE("GPL");
 MODULE_AUTHOR("Evgeniy Polyakov <johnpol@2ka.mipt.ru>");
 MODULE_DESCRIPTION("Generic userspace <-> kernelspace connector.");
 
-static int unit = NETLINK_NFLOG;
+static int unit = NETLINK_CONNECTOR;
 static u32 cn_idx = CN_IDX_CONNECTOR;
 static u32 cn_val = CN_VAL_CONNECTOR;
 
_

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 2514 bytes --]

On Wed, 2005-05-18 at 23:45 -0700, Andrew Morton wrote:
> Evgeniy Polyakov <johnpol@2ka.mipt.ru> wrote:
> >
> > On Wed, 2005-05-18 at 22:27 -0700, Andrew Morton wrote:
> > > Herbert Xu <herbert@gondor.apana.org.au> wrote:
> > > >
> > > > Andrew Morton <akpm@osdl.org> wrote:
> > > >  > 
> > > >  > This is the second report of this happening.
> > > >  > 
> > > >  > FATAL: Error inserting ipt_ULOG (/lib/modules/2.6.12-rc4-mm1-2/kernel/net/ipv4/netfilter/ipt_ULOG.ko): Cannot allocate memory
> > > 
> > > Ah yes, thanks.  netlink_kernel_create() has a dumbish return-value protocol.
> > > 
> > > >  This probably means that someone has stolen the NETLINK_NFLOG id.
> > > 
> > > cn_init() uses NETLINK_NFLOG.  Peculiar.
> > 
> > Yep...
> 
> Bah.
> 
> > One may use "unit" module parameter to specify different socket.
> 
> Is there any reason for being able to do that?

It was created to be able to load several different connector devices 
from different modules [basically it was module1.ko unit=10, module2.ko
unit=11 and so on].

> > Should I ask David Miller to allocate new one specially for connector?
> 
> Dave?
> 
> 
> diff -puN include/linux/netlink.h~connector-netlink-id-fix include/linux/netlink.h
> --- 25/include/linux/netlink.h~connector-netlink-id-fix	2005-05-18 23:44:43.000000000 -0700
> +++ 25-akpm/include/linux/netlink.h	2005-05-18 23:44:43.000000000 -0700
> @@ -14,6 +14,7 @@
>  #define NETLINK_SELINUX		7	/* SELinux event notifications */
>  #define NETLINK_ARPD		8
>  #define NETLINK_AUDIT		9	/* auditing */
> +#define NETLINK_CONNECTOR	10	/* kernel connector */
>  #define NETLINK_ROUTE6		11	/* af_inet6 route comm channel */
>  #define NETLINK_IP6_FW		13
>  #define NETLINK_DNRTMSG		14	/* DECnet routing messages */
> diff -puN drivers/connector/connector.c~connector-netlink-id-fix drivers/connector/connector.c
> --- 25/drivers/connector/connector.c~connector-netlink-id-fix	2005-05-18 23:44:43.000000000 -0700
> +++ 25-akpm/drivers/connector/connector.c	2005-05-18 23:44:43.000000000 -0700
> @@ -33,7 +33,7 @@ MODULE_LICENSE("GPL");
>  MODULE_AUTHOR("Evgeniy Polyakov <johnpol@2ka.mipt.ru>");
>  MODULE_DESCRIPTION("Generic userspace <-> kernelspace connector.");
>  
> -static int unit = NETLINK_NFLOG;
> +static int unit = NETLINK_CONNECTOR;
>  static u32 cn_idx = CN_IDX_CONNECTOR;
>  static u32 cn_val = CN_VAL_CONNECTOR;
>  
> _
-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Andrew Morton]

Evgeniy Polyakov <johnpol@2ka.mipt.ru> wrote:
>
>  > Is there any reason for being able to do that?
> 
>  It was created to be able to load several different connector devices 
>  from different modules [basically it was module1.ko unit=10, module2.ko
>  unit=11 and so on].

Loading multiple instances of the same module so that we get multiple
instances of its data structures?   That's fairly ghastly.

We should remove the feature.  If there is demonstrated need for multiple
instances then it should be done by more conventional means - syscall,
allocation of a device major then use the minor as a selector or whatever.

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 1151 bytes --]

On Thu, 2005-05-19 at 00:08 -0700, Andrew Morton wrote:
> Evgeniy Polyakov <johnpol@2ka.mipt.ru> wrote:
> >
> >  > Is there any reason for being able to do that?
> > 
> >  It was created to be able to load several different connector devices 
> >  from different modules [basically it was module1.ko unit=10, module2.ko
> >  unit=11 and so on].
> 
> Loading multiple instances of the same module so that we get multiple
> instances of its data structures?   That's fairly ghastly.

It was not the same module.
it was quite different module which implemented similar functionality
and used different connector device for different purposes.

> We should remove the feature.  If there is demonstrated need for multiple
> instances then it should be done by more conventional means - syscall,
> allocation of a device major then use the minor as a selector or whatever.

It is easy to use feature which prevents problems
with situation when the same socket number is used, for example, 
in 2.6 kernel while connector uses it in -mm tree.

-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[James Morris]

On Wed, 18 May 2005, Andrew Morton wrote:

> cn_init() uses NETLINK_NFLOG.  Peculiar.

I raised this some time ago, and it was never fixed by the author.


- James
-- 
James Morris
<jmorris@redhat.com>

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[James Morris]

On Thu, 19 May 2005, Andrew Morton wrote:

> We should remove the feature.  If there is demonstrated need for multiple
> instances then it should be done by more conventional means - syscall,
> allocation of a device major then use the minor as a selector or whatever.

I'd suggest removing the connector code completely.


- James
-- 
James Morris
<jmorris@redhat.com>

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[jamal]

On Thu, 2005-19-05 at 09:16 -0400, James Morris wrote:
> On Wed, 18 May 2005, Andrew Morton wrote:
> 
> > cn_init() uses NETLINK_NFLOG.  Peculiar.
> 
> I raised this some time ago, and it was never fixed by the author.
> 

Sharing as was done here is not proper; perhaps the suggestion from
Andrew is the way to go.

The other thing we could do is allocate a set of IDs for private use
(as in private IPs).
However, something like connector should only get one ID and it can do
its own multiplexing based on "units".

cheers,
jamal

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[Andrew Morton]

James Morris <jmorris@redhat.com> wrote:
>
> On Thu, 19 May 2005, Andrew Morton wrote:
> 
> > We should remove the feature.  If there is demonstrated need for multiple
> > instances then it should be done by more conventional means - syscall,
> > allocation of a device major then use the minor as a selector or whatever.
> 
> I'd suggest removing the connector code completely.
> 

How come?  It seems sensible to have library code which consoldates the
common netlink setup and teardown stuff and which offers a higher-level
messaging API, although connector may be a bit overdesigned for that.  The
fork connector guys seemed to find it useful?

Re: Fw: 2.6.12-rc4-mm[12] - ULOG problem

[James Morris]

On Thu, 19 May 2005, Andrew Morton wrote:

> How come?  It seems sensible to have library code which consoldates the
> common netlink setup and teardown stuff and which offers a higher-level
> messaging API, although connector may be a bit overdesigned for that.  The
> fork connector guys seemed to find it useful?

I don't think it's the right implementation, and providing feedback on it
has proven frustrating and unfruitful.


- James
-- 
James Morris
<jmorris@redhat.com>

Re: 2.6.12-rc4-mm[12] - ULOG problem

[David S. Miller]

From: Evgeniy Polyakov <johnpol@2ka.mipt.ru>
Date: Thu, 19 May 2005 11:06:25 +0400

> It was created to be able to load several different connector devices 
> from different modules [basically it was module1.ko unit=10, module2.ko
> unit=11 and so on].

That's bogus, each connector will need to allocate it's own
specific define in the netlink header, and we're about to
run out of numbers.

Please redesign the connector so that it doesn't eat more than
one netlink number, we're running out of them soon.

Re: 2.6.12-rc4-mm[12] - ULOG problem

[David S. Miller]

From: Andrew Morton <akpm@osdl.org>
Date: Wed, 18 May 2005 23:45:52 -0700

> > Should I ask David Miller to allocate new one specially for connector?
> 
> Dave?

Giving the connector a single value 10 is OK, the rest of
what the connector is doing in this area is not.

I wish you hadn't put the connector into -mm, it's really
in need of several fixups still.  The bogus reuse of the
ULOG netlink number is simply indicative of that.

Re: 2.6.12-rc4-mm[12] - ULOG problem

[David S. Miller]

From: James Morris <jmorris@redhat.com>
Date: Thu, 19 May 2005 09:19:33 -0400 (EDT)

> On Thu, 19 May 2005, Andrew Morton wrote:
> 
> > We should remove the feature.  If there is demonstrated need for multiple
> > instances then it should be done by more conventional means - syscall,
> > allocation of a device major then use the minor as a selector or whatever.
> 
> I'd suggest removing the connector code completely.

I totally agree, it's crap at the current time and the author doesn't
fix the problems being shown in it, in fact he defends some of them
even this one.  It needs lots of work still.

Re: 2.6.12-rc4-mm[12] - ULOG problem

[David S. Miller]

From: jamal <hadi@cyberus.ca>
Date: Thu, 19 May 2005 09:46:50 -0400

> However, something like connector should only get one ID and it can do
> its own multiplexing based on "units".

Yes.  There is no reason connector needs more than one
netlink ID.

Re: 2.6.12-rc4-mm[12] - ULOG problem

[David S. Miller]

From: James Morris <jmorris@redhat.com>
Date: Thu, 19 May 2005 13:09:44 -0400 (EDT)

> I don't think it's the right implementation, and providing feedback on it
> has proven frustrating and unfruitful.

Yes, Evgeni is a complete control freak.

Re: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 463 bytes --]

On Thu, 2005-05-19 at 11:49 -0700, David S. Miller wrote:
> From: jamal <hadi@cyberus.ca>
> Date: Thu, 19 May 2005 09:46:50 -0400
> 
> > However, something like connector should only get one ID and it can do
> > its own multiplexing based on "units".
> 
> Yes.  There is no reason connector needs more than one
> netlink ID.

It does not require several IDs.

-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 983 bytes --]

On Thu, 2005-05-19 at 11:48 -0700, David S. Miller wrote:
> From: James Morris <jmorris@redhat.com>
> Date: Thu, 19 May 2005 09:19:33 -0400 (EDT)
> 
> > On Thu, 19 May 2005, Andrew Morton wrote:
> > 
> > > We should remove the feature.  If there is demonstrated need for multiple
> > > instances then it should be done by more conventional means - syscall,
> > > allocation of a device major then use the minor as a selector or whatever.
> > 
> > I'd suggest removing the connector code completely.
> 
> I totally agree, it's crap at the current time and the author doesn't
> fix the problems being shown in it, in fact he defends some of them
> even this one.  It needs lots of work still.

There were no problems reported.
Completely.

What I defend is ability to function correctly when some changes 
may be introduced by providing some extensions that may be used.

-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 795 bytes --]

On Thu, 2005-05-19 at 11:42 -0700, David S. Miller wrote:
> From: Evgeniy Polyakov <johnpol@2ka.mipt.ru>
> Date: Thu, 19 May 2005 11:06:25 +0400
> 
> > It was created to be able to load several different connector devices 
> > from different modules [basically it was module1.ko unit=10, module2.ko
> > unit=11 and so on].
> 
> That's bogus, each connector will need to allocate it's own
> specific define in the netlink header, and we're about to
> run out of numbers.
> 
> Please redesign the connector so that it doesn't eat more than
> one netlink number, we're running out of them soon.

It does not require more than one socket number.
It just can use any other number provided.

-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 1992 bytes --]

On Thu, 2005-05-19 at 11:44 -0700, David S. Miller wrote:
> From: Andrew Morton <akpm@osdl.org>
> Date: Wed, 18 May 2005 23:45:52 -0700
> 
> > > Should I ask David Miller to allocate new one specially for connector?
> > 
> > Dave?
> 
> Giving the connector a single value 10 is OK, the rest of
> what the connector is doing in this area is not.
> 
> I wish you hadn't put the connector into -mm, it's really
> in need of several fixups still.  The bogus reuse of the
> ULOG netlink number is simply indicative of that.

And how does it indicate that it is bogus? :)
It is one netlink user that can use any socket number - and now
it is called bogus.

What other people suggest is just create some wrapper over 
netlink_broadcast() and alloc_skb().
if you do think it will solve all problems, then I even will not
say anything. Point.
I just want to note, that if you want message bus, you require
at least oppsite direction, and thus input callback, and
either several socket number for each user [kobject, audit, iscsi, 
xfrm - all they implement what connector already does] or
some header and thus parser in input callback, and thus
some registration mechanism. Magically it is what connector does.
And nothing more, btw.
It only allows not to rewrite the whole driver just to move
to new transport layer, only register new cn_dev device
with different method.

Ok, Dave, we all see your point, feel free to remove connector,
I will not bother people to include it yet another time...
But I believe in a month or two someone will reinvent the wheel
and create somthing similar. If it will be simple wrapper, then
it just can not solve message bus problem, only notification, 
and thus someone will create it again.

Thank you.

P.S. smtp server from my old university years seems to still drop
your messages, sorry for that, I will kick admins again.

-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: 2.6.12-rc4-mm[12] - ULOG problem

[jamal]

On Fri, 2005-20-05 at 10:39 +0400, Evgeniy Polyakov wrote:

> I just want to note, that if you want message bus, you require
> at least oppsite direction, and thus input callback, and
> either several socket number for each user [kobject, audit, iscsi, 
> xfrm - all they implement what connector already does] or
> some header and thus parser in input callback, and thus
> some registration mechanism. 


But if you are given your own ID, there should no issues, correct?
i.e it will be no different than say rtnetlink. Users would still have
to subscribe to topics etc, but thats something you already handle.
IOW, Iam hoping you are not resorting to a single socket with some user
space mux to do filtering.

cheers,
jamal

Re: 2.6.12-rc4-mm[12] - ULOG problem

[Evgeniy Polyakov]

[-- Attachment #1: Type: text/plain, Size: 1079 bytes --]

On Fri, 2005-05-20 at 07:42 -0400, jamal wrote:
> On Fri, 2005-20-05 at 10:39 +0400, Evgeniy Polyakov wrote:
> 
> > I just want to note, that if you want message bus, you require
> > at least oppsite direction, and thus input callback, and
> > either several socket number for each user [kobject, audit, iscsi, 
> > xfrm - all they implement what connector already does] or
> > some header and thus parser in input callback, and thus
> > some registration mechanism. 
> 
> 
> But if you are given your own ID, there should no issues, correct?

Yes.

> i.e it will be no different than say rtnetlink. Users would still have
> to subscribe to topics etc, but thats something you already handle.
> IOW, Iam hoping you are not resorting to a single socket with some user
> space mux to do filtering.

Not following, sorry.
Userspace should use the same socket as kernel - my application
previously
obtained it from dmesg and later used ULOG.

> cheers,
> jamal
> 
-- 
        Evgeniy Polyakov

Crash is better than data corruption -- Arthur Grabowski

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]