Re: Data corruption issue with splice() on 2.6.27.10

[Ben Mansell <ben@zeus.com>]

Ingo Molnar wrote:
> * Willy Tarreau <w@1wt.eu> wrote:
> 
>> On Thu, Jan 08, 2009 at 07:16:51AM +0000, Jarek Poplawski wrote:
>>> On 06-01-2009 19:15, Willy Tarreau wrote:
>>> ...
>>>> Ah, so you might also have discovered a few annoyances with the API, eg
>>>> the fact that splice() returns after the first read in non-blocking mode,
>>>> as well as the fact that it never returns zero on close, but -EAGAIN,
>>>> which requires an additional recv(MSG_PEEK) to distinguish between a
>>>> close and a lack of data. But I leave that for a later discussion, let's
>>>> address the corruption issue first.
>>> FYI, this should be just fixed:
>>> http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=4f7d54f59bc470f0aaa932f747a95232d7ebf8b1
>>>
>> Ah cool, thanks Jarek for notifying us. Indeed, it's the exact same patch
>> I had pending here ;-)
>>
>> I'll ping Greg for a backport into -stable, as applications relying on 
>> this will clearly not work without that fix.
>>
>> The other one I had consists in removing "|| !timeo" at the end of the 
>> loop, because otherwise splice() returns very small chunks (typically 
>> 1448 or 1460 bytes), leading to disastrous performance on high bandwidth 
>> links. At 10 Gbps, this means about 800000 calls to splice() per second!
> 
> looks interesting - would you mind to submit it?

FWIW, I've also tested this change with some splice() benchmarks. I can 
confirm that removing the "|| !timeo" works well and improves 
performance significantly.


Ben