Re: [squid-users] Re: [PATCH] Revised fix for download corruption

I see your point. Your patch does kill the request higher up in the chain, and is therefore preferable -- it avoids the OPEN which my patch does.

Thanks,
Phil Oester

On Tue, Nov 05, 2002 at 12:39:52AM +0100, Henrik Nordstrom wrote:
> Exacly, and is why I did my patch where I did. If we see a
> storeClose() for a object beeing read then immediately close the file
> and abort any pending I/O operations. Only if the file is open for
> writing and there is pending write operations in the queue delay the
> close until the writes have finished.
>
> The two patches are similar, but addresses the read problem at
> different places in time:
>
> * Your patch address the problem upon completetion of a I/O event
> some time after a storeClose() has been requested
>
> * My match address it immediately when storeClose() is called, not
> waiting for the pending I/O event to complete. This also completely
> aborts the I/O event if it hasn't started yet.
>
> As said in the discussion leading up to the patch it is only a
> temporary bandaid. The real problem is that aioRead() cannot verify
> the validity of the buffer before copying the content into it. This
> problem occurs due to a unneeded indirection layer in the aufs
> implementation breaking a cbdata boundary and should be addressed,
> but until then the simple patch should suffice.
>
>
> And no, the patch addresses reads only, not writes like Robert
> claimed..
>
> Regards
> Henrik
Received on Mon Nov 04 2002 - 17:14:50 MST