On Wed, Jan 30, 2002 at 01:12:11PM +0100, Oskar Sandberg wrote:
> On Tue, Jan 29, 2002 at 07:34:54PM -0500, Tavin Cole wrote:
> <>
> > What happens is that the node gets locked in a loop where it keeps
> > reading the same request message from a socket over and over again.
> > Since it has already handled that chain, it issues a QueryRejected.
> > This loop is on a 3 second interval, so the unfortunate node that first
> > sent that request to the evil-1.1 node starts receiving an identical
> > QueryRejected every 3 seconds ad infinitum.
>
> I know why this happens. Their implementation of BufferedInputStream
> doesn't consider that there are IOExceptions that are non-terminal for
> the InputStream and doesn't watch where the buffer pointer goes after an
> exception is thrown. Consider something like:
>
> int pointer;
> int length;
> byte[] buffer;
>
> private int read() throws IOException {
> while (pointer >= length) {
> pointer = 0;
> length = in.read(buffer);
> }
> return buffer[pointer];
> }
>
> compared to if you did in.read() first, and the then set pointer to
> zero. The result is that when it comes back from an
> InterruptedIOException while waiting on in.read() it thinks the pointer
> is already reset - and reads the last message off the buffer.
>
> This bug is quite common - I believe it was the reason for the problems
> with Microsoft's 1.1 jvm ("jview") and I know I had to fix it for
> Kaffe's BufferedReader (they accepted the patch).
>
> We could use our own BufferedInputStream (grab it from kaffe) and see if
> that helps.
OTOH, we could take the opportunity to abandon these buggy 1.1 JVMs
altogether and start using 1.2+ features if they are supported by the
leading free JVMs (only Kaffe atm).
-tc
_______________________________________________
Devl mailing list
Devl at freenetproject.org
http://lists.freenetproject.org/mailman/listinfo/devl
