A potentially serious bug in KIO/kio_http and a possible patch ??

[Maksim Orlovich]

This patch looks wrong to me, as even in the limited case, you could
potentially have the data that's already buffered fulfill the request
entirely, can you not?

Will look at other stuff later in the evening... Though I must say I
am impressed you could trace down anything gmail-related to such a low
level.


On 7/17/10, Dawit A <adawit at kde.org> wrote:
> Hi,
>
> The attached patch makes it possible for HTTPProtocol::readBuffered function
> to
> behaves differently when called from readLimited. This patch is intended to
> address the issue of the Gmail chat functionality no longer working in
> kdewebkit based browsers.
>
> Though the patch fully addresses the aforementioned bug, it really does not
> fix
> the actual cause of this bug. So what is the actual cause ?  Well when you
> log
> in to Gmail, a check is preformed to see if the chat functionality is
> enabled
> or not.  This is done through a simple GET request that returns a 20 byte
> response. At that point kio_http has read the header and few bytes of the
> content as well, 3 bytes to be exact. kio_http stores these few bytes of
> data
> that are not part of the header in a buffer.
>
> It then attempts to read the content and that is where the fun starts...
> Because kio_http has received the actual size of the content it is supposed
> to
> read, it will use readLimited to get the content. readLimited in turn calls
> readBuffered which, as a result of a fix for bug 180631, always sends
> whatever
> data was read ahead while retrieving the header above. This means the 3
> bytes
> above will be sent to the client first in our case. After that kio_http
> retrieves and attempts to send the remaining 17 bytes. However, this last
> portion never ever makes it to the client application. It seems to disappear
> into the ether! Actually it never gets past KIO::Connection::send and at
> this
> point I cannot figure out why!
>
> So why this patch then ?  Simple. If you followed what I attempted to
> explain
> above, you would see that it is very inefficient for kio_http to split a 20
> byte
> data especially when it already knows the final size of the content it is
> supposed to retrieve from the server ; so addressing that through this patch
> should be no brainer. The fact that the Gmail chat issue gets fixed as a
> result
> is only a bonus at this point.
>
> And hence my request...
>
> #1. Can someone using KHTML verify whether or not the Gmail chat
> functionality
> works for them ? I cannot get Gmail to work with KHTML here ; so I am unable
> to do this myself.
>
> #2. Can someone tell me if there is a scenario where sending two MSG_DATA
> requests in rapid succession might result in one of them being dropped ?
>
> #3. Any objections, concerns, suggestions etc about the patch are welcome.
>
> Regards,
> Dawit A.
>