[RFD][PATCH] Add JMEMCMP to Berkeley Packet Filters

[RFD][PATCH] Add JMEMCMP to Berkeley Packet Filters

[Ian Molton]

Hi folks,

This patch implements an extension for BPF to allow filter programs to use a
data section, along with a MEMCMP instruction.

There are a few issues noted in the patch itself, which can easily be
addressed, and I would like to check wether sk_run_filter is ever expected to
be called from a context that cannot sleep (I dont think it is).

I think the patch should probably be split into a patch to add data sections
and one adding the JMEMCMP instruction, but that can be done after some review! 

-Ian

Re: [RFD][PATCH] Add JMEMCMP to Berkeley Packet Filters

[Ian Molton]

Shit, sorry about the noise. git-send-email bit me :)

-Ian

Re: [RFD][PATCH] Add JMEMCMP to Berkeley Packet Filters

[Paul Mackerras]

On Thu, Feb 10, 2011 at 12:14:46PM +0000, Ian Molton wrote:

> This patch implements an extension for BPF to allow filter programs to use a
> data section, along with a MEMCMP instruction.
> 
> There are a few issues noted in the patch itself, which can easily be
> addressed, and I would like to check wether sk_run_filter is ever expected to
> be called from a context that cannot sleep (I dont think it is).

The ppp driver (ppp_generic.c) calls sk_run_filter inside a region
protected by spin_lock_bh, so we can't sleep there.  Having sk_run_filter
potentially sleep would make it useless for ppp.

Paul.

Re: [RFD][PATCH] Add JMEMCMP to Berkeley Packet Filters

[Ian Molton]

On 12/02/11 02:14, Paul Mackerras wrote:
> On Thu, Feb 10, 2011 at 12:14:46PM +0000, Ian Molton wrote:
>
>> This patch implements an extension for BPF to allow filter programs to use a
>> data section, along with a MEMCMP instruction.
>>
>> There are a few issues noted in the patch itself, which can easily be
>> addressed, and I would like to check wether sk_run_filter is ever expected to
>> be called from a context that cannot sleep (I dont think it is).
>
> The ppp driver (ppp_generic.c) calls sk_run_filter inside a region
> protected by spin_lock_bh, so we can't sleep there.  Having sk_run_filter
> potentially sleep would make it useless for ppp.

Thanks for the review.

I'll fix it not to sleep :)

Do I take it from the lack of people comlpaining that the concept of 
adding this type of instruction itself is not a problem ?

Thanks!

-Ian

[RFD][PATCH] Add JMEMCMP to Berkeley Packet Filters

[Ian Molton]

 Documentation/networking/filter.txt |    9 ++
 drivers/isdn/i4l/isdn_ppp.c         |    2 
 drivers/net/ppp_generic.c           |    2 
 include/asm-generic/socket.h        |    2 
 include/linux/filter.h              |   17 ++++-
 include/linux/ptp_classify.h        |    2 
 net/core/filter.c                   |  115 ++++++++++++++++++++++++++++++++++--
 net/core/sock.c                     |   14 ++++
 net/core/timestamping.c             |    4 -
 net/packet/af_packet.c              |    3 

This patch adds support for adding a data section to BPF. It is intended to be
used by the JMEMCMP instruction also added in this patch.

There are some issues, mostly noted int he commit message, and I'd like to
check that sk_run_filter() does not get called from a context that cannot sleep
(I dont think so).

Comments welcome!

-Ian