httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rainer Jung <rainer.j...@kippdata.de>
Subject Re: [PATCH] proxy/balancer: fix PR 45434 regression
Date Tue, 24 Jul 2012 18:37:02 GMT
On 24.07.2012 19:40, Joe Orton wrote:
> The test case for PR 45434 seems to have regressed across 2.2->2.4.
>
> https://issues.apache.org/bugzilla/show_bug.cgi?id=45434
>
> I have not tried to understand the mechanics here, but a dumb
> side-by-side analysis found a missing piece, below.  2.2 hardcodes this
> as "real + 11" but 2.4 uses the constant elsewhere.  Any reason why this
> would be wrong?  It fixes the test case I added to t/modules/proxy.t.
>
> 2.2.x code for reference:
>
> http://svn.apache.org/viewvc/httpd/httpd/branches/2.2.x/modules/proxy/proxy_util.c?annotate=1058624#l1090
>
> Index: modules/proxy/proxy_util.c
> ===================================================================
> --- modules/proxy/proxy_util.c	(revision 1365029)
> +++ modules/proxy/proxy_util.c	(working copy)
> @@ -860,7 +860,7 @@
>               (balancer = ap_proxy_get_balancer(r->pool, sconf, real, 1))) {
>               int n, l3 = 0;
>               proxy_worker **worker = (proxy_worker **)balancer->workers->elts;
> -            const char *urlpart = ap_strchr_c(real, '/');
> +            const char *urlpart = ap_strchr_c(real + sizeof(BALANCER_PREFIX) - 1, '/');
>               if (urlpart) {
>                   if (!urlpart[1])
>                       urlpart = NULL;
>

The code including the offset was introduced in trunk in

http://svn.apache.org/viewvc?view=revision&amp;revision=771587

and then ported back to 2.2 in

http://svn.apache.org/viewvc?view=revision&revision=777191

Later in first the offset got replaced by using a separate function call 
first which returned a new variable "bname" which was already advanced 
by the correct offset and then put that one into ap_strchr_c:

http://svn.apache.org/viewvc/httpd/httpd/trunk/modules/proxy/proxy_util.c?r1=1058621&r2=1058622&diff_format=h

A bit later the function was rearranged to not return a char * but a 
boolean and the new variable was removed again. The old variable "real" 
was now used without the previous offset. That seems to be the culprit 
and a bug. There's no indication, that this was intentional.

So: Your change seems good to me.

Regards,

Rainer

Mime
View raw message