[MPTCP] sock diag uAPI and MPTCP

[MPTCP] sock diag uAPI and MPTCP

[Paolo Abeni]

[-- Attachment #1: Type: text/plain, Size: 722 bytes --]

Hi,

IPPROTO_MPTCP value (0x106) can't be represented using the current sock
diag uAPI, as the 'sdiag_protocol' field is 8 bits wide.

To implement diag support for MPTCP socket, we will likely need a
'inet_diag_req_v3' with a wider 'sdiag_protocol'
field. inet_diag_handler_cmd() could detect the version of
the inet_diag_req_v* provided by user-space checking nlmsg_len() and
convert _v2 reqs to _v3.

This change will be a bit invasive, as all in kernel diag users will
then operate only on 'inet_diag_req_v3' (many functions' signature
change required), but the code-related changes will be encapsulated
by inet_diag_handler_cmd().

Would the above be acceptable?

Thanks for any feedback!

Paolo

sock diag uAPI and MPTCP

[Paolo Abeni]

Hi,

IPPROTO_MPTCP value (0x106) can't be represented using the current sock
diag uAPI, as the 'sdiag_protocol' field is 8 bits wide.

To implement diag support for MPTCP socket, we will likely need a
'inet_diag_req_v3' with a wider 'sdiag_protocol'
field. inet_diag_handler_cmd() could detect the version of
the inet_diag_req_v* provided by user-space checking nlmsg_len() and
convert _v2 reqs to _v3.

This change will be a bit invasive, as all in kernel diag users will
then operate only on 'inet_diag_req_v3' (many functions' signature
change required), but the code-related changes will be encapsulated
by inet_diag_handler_cmd().

Would the above be acceptable?

Thanks for any feedback!

Paolo

[MPTCP] Re: sock diag uAPI and MPTCP

[David Miller]

[-- Attachment #1: Type: text/plain, Size: 1061 bytes --]

From: Paolo Abeni <pabeni(a)redhat.com>
Date: Fri, 19 Jun 2020 12:54:40 +0200

> IPPROTO_MPTCP value (0x106) can't be represented using the current sock
> diag uAPI, as the 'sdiag_protocol' field is 8 bits wide.
> 
> To implement diag support for MPTCP socket, we will likely need a
> 'inet_diag_req_v3' with a wider 'sdiag_protocol'
> field. inet_diag_handler_cmd() could detect the version of
> the inet_diag_req_v* provided by user-space checking nlmsg_len() and
> convert _v2 reqs to _v3.
> 
> This change will be a bit invasive, as all in kernel diag users will
> then operate only on 'inet_diag_req_v3' (many functions' signature
> change required), but the code-related changes will be encapsulated
> by inet_diag_handler_cmd().

Another way to extend the size of a field is to add an attribute which
supercedes the header structure field when present.

We did this when we needed to make the fib rule table ID number larger,
see FRA_TABLE.

You'd only need to specify this when using protocol values larger than
8 bits in size.

Re: sock diag uAPI and MPTCP

[David Miller]

From: Paolo Abeni <pabeni@redhat.com>
Date: Fri, 19 Jun 2020 12:54:40 +0200

> IPPROTO_MPTCP value (0x106) can't be represented using the current sock
> diag uAPI, as the 'sdiag_protocol' field is 8 bits wide.
> 
> To implement diag support for MPTCP socket, we will likely need a
> 'inet_diag_req_v3' with a wider 'sdiag_protocol'
> field. inet_diag_handler_cmd() could detect the version of
> the inet_diag_req_v* provided by user-space checking nlmsg_len() and
> convert _v2 reqs to _v3.
> 
> This change will be a bit invasive, as all in kernel diag users will
> then operate only on 'inet_diag_req_v3' (many functions' signature
> change required), but the code-related changes will be encapsulated
> by inet_diag_handler_cmd().

Another way to extend the size of a field is to add an attribute which
supercedes the header structure field when present.

We did this when we needed to make the fib rule table ID number larger,
see FRA_TABLE.

You'd only need to specify this when using protocol values larger than
8 bits in size.

[MPTCP] Re: sock diag uAPI and MPTCP

[Paolo Abeni]

[-- Attachment #1: Type: text/plain, Size: 1313 bytes --]

On Sat, 2020-06-20 at 17:44 -0700, David Miller wrote:
> From: Paolo Abeni <pabeni(a)redhat.com>
> Date: Fri, 19 Jun 2020 12:54:40 +0200
> 
> > IPPROTO_MPTCP value (0x106) can't be represented using the current sock
> > diag uAPI, as the 'sdiag_protocol' field is 8 bits wide.
> > 
> > To implement diag support for MPTCP socket, we will likely need a
> > 'inet_diag_req_v3' with a wider 'sdiag_protocol'
> > field. inet_diag_handler_cmd() could detect the version of
> > the inet_diag_req_v* provided by user-space checking nlmsg_len() and
> > convert _v2 reqs to _v3.
> > 
> > This change will be a bit invasive, as all in kernel diag users will
> > then operate only on 'inet_diag_req_v3' (many functions' signature
> > change required), but the code-related changes will be encapsulated
> > by inet_diag_handler_cmd().
> 
> Another way to extend the size of a field is to add an attribute which
> supercedes the header structure field when present.
> 
> We did this when we needed to make the fib rule table ID number larger,
> see FRA_TABLE.
> 
> You'd only need to specify this when using protocol values larger than
> 8 bits in size.

Thank you very much for the directions! This looks like a better, more
encapsulated solution. I'll try to give it a shot.

Thanks!

Paolo

Re: sock diag uAPI and MPTCP

[Paolo Abeni]

On Sat, 2020-06-20 at 17:44 -0700, David Miller wrote:
> From: Paolo Abeni <pabeni@redhat.com>
> Date: Fri, 19 Jun 2020 12:54:40 +0200
> 
> > IPPROTO_MPTCP value (0x106) can't be represented using the current sock
> > diag uAPI, as the 'sdiag_protocol' field is 8 bits wide.
> > 
> > To implement diag support for MPTCP socket, we will likely need a
> > 'inet_diag_req_v3' with a wider 'sdiag_protocol'
> > field. inet_diag_handler_cmd() could detect the version of
> > the inet_diag_req_v* provided by user-space checking nlmsg_len() and
> > convert _v2 reqs to _v3.
> > 
> > This change will be a bit invasive, as all in kernel diag users will
> > then operate only on 'inet_diag_req_v3' (many functions' signature
> > change required), but the code-related changes will be encapsulated
> > by inet_diag_handler_cmd().
> 
> Another way to extend the size of a field is to add an attribute which
> supercedes the header structure field when present.
> 
> We did this when we needed to make the fib rule table ID number larger,
> see FRA_TABLE.
> 
> You'd only need to specify this when using protocol values larger than
> 8 bits in size.

Thank you very much for the directions! This looks like a better, more
encapsulated solution. I'll try to give it a shot.

Thanks!

Paolo