netdev - Re: Adding Support for SG,GSO,GRO

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <AANLkTi=7NLkHW6c88gUcyW8i0Wwmf2Cw4NdmRiGci4kE@mail.gmail.com>
Date:	Fri, 10 Dec 2010 15:18:11 +0100
From:	Michał Mirosław <mirqus@...il.com>
To:	David Miller <davem@...emloft.net>
Cc:	bhutchings@...arflare.com, srk@...com, netdev@...r.kernel.org
Subject: Re: Adding Support for SG,GSO,GRO

2010/12/9 David Miller <davem@...emloft.net>:
> From: Michał Mirosław <mirqus@...il.com>
> Date: Thu, 9 Dec 2010 19:47:57 +0100
>> Isn't that condition too broad? If the data could change after packet
>> is submitted to the driver then results would be unpredictable and
>> allow sending wrong data with correct (because hw-calculated)
>> checksum.
> They are intentionally like that, without question.
>
> Otherwise we'd need to interlock with all application mapped,
> filesystem, and other page writes while sending any page over the
> network.
>
> We absolutely do not want to have to freeze every page we try to send
> via sendfile() or similar, the cost is just too high.
>
> If the application or networked filesystem needs such synchronization,
> it provides it for itself.
>
> For example, SAMBA only uses sendfile() when the file has an op-lock
> held on it.
>
> The checksum requirement for using SG is not going away, so continuing
> to discuss along the lines of removing that requirement is not a good
> use of your time I don't think.

I'm trying to understand the dependency because it looks artificial for me.

Unless I totally misunderstood, you say that we accept bogus data to
being sent using sendfile(). If yes, then we might as well allow
broken checksum (CPU calculated, before data changed and then was sent
to network).

If the splice/sendfile is taken out of the picture, are there any
other scenarios, when data pages could be changed after
ndo_start_xmit() entry and before TX DMA completion (between dma_map
.. dma_unmap)? And is it really what can happen with splice/sendfile?

Best Regards,
Michał Mirosław
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html