Re: squid 2.6-rproxy: locrewrite.c bug

For this kind of situations there is the cbdataReferenceDoneValid (or
is it cbdataReferenceValidDone, I never remeber which). See the
programmers guide for relatively detailed instructions and a couple
of examples. There is also numerous examples in other parts of the
code.

Regards
Henrik

On Wednesday 19 June 2002 00.52, Jeffrey D. Wheelhouse wrote:
> I've run into a snag with the new-style cbdata handling in
> locrewrite.c.
>
> The pre-patch locationRewriteHandleReply:
> 65: valid = cbdataValid(r->data);
> 66: cbdataUnlock(r->data);
> 67: if (valid)
> 68: r->handler(r->data, reply);
> 69: locationRewriteStateFree(r);
>
> The updated code:
>
> 65: valid = cbdataReferenceValid(r->data);
> 66: cbdataReferenceDone(r->data);
> 67: if (valid)
> 68: r->handler(r->data, reply);
> 69: locationRewriteStateFree(r);
>
> This promptly leads to cores, because cbdataReferenceDone wipes out
> r->data. Duh. I feel very silly for not realizing this the first
> time.
>
> I am testing a patch (attached) that yields:
>
> 64: if (cbdataReferenceValid(r->data))
> 65: r->handler(r->data, reply);
> 66: cbdataReferenceDone(r->data);
> 67: locationRewriteStateFree(r);
>
> However, the whole point of "valid" seems to be to do the
> unreference before calling r->handler, and I'm not sure why. So
> this may not be the correct fix.
>
> This patch should also be tabbed correctly (finally).
>
> I also have the read_ahead_gap config patch done, but nontrivial
> testing of that has been gated by this issue until now. If this is
> a good fix, the read_ahead_gap patch will be done testing Thursday.
>
> Thanks,
> Jeff
Received on Tue Jun 18 2002 - 17:39:42 MDT