httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Yann Ylavic <ylavic....@gmail.com>
Subject Re: svn commit: r1741310 - in /httpd/httpd/trunk: modules/http2/ server/ server/mpm/event/ server/mpm/motorz/ server/mpm/simple/ server/mpm/winnt/ server/mpm/worker/
Date Thu, 28 Apr 2016 20:00:09 GMT
On Thu, Apr 28, 2016 at 5:22 PM, William A Rowe Jr <wrowe@rowe-clan.net> wrote:
>
> What I meant is that the request line *MUST* be touched - wiped
> clean - for any status_from_conn(c) request.
>
> When we say we preserve the most recent state of the worker in
> the scoreboard, that old request line is gone.  Every time we are
> doing a conn-based refresh of that worker status, the req-based
> data is no longer relevant and needs to go away.

OK, I'm fine with this, much prefered over mixing (which finally I
introduced in my incomplete revert, understood ;)

>
> See
> http://svn.apache.org/viewvc/httpd/httpd/branches/2.4.x/server/scoreboard.c?r1=1239125&r2=1741240&diff_format=h
>
> for the delta between 2.4.1 and 2.4.broken (current state).
>
>          else if (c) {
> -            apr_cpystrn(ws->client, ap_get_remote_host(c, NULL,
> -                        REMOTE_NOLOOKUP, NULL), sizeof(ws->client));
> -            ws->request[0]='\0';
> -            ws->vhost[0]='\0';

Indeed, I missed that, too much influenced by the original report
about blanks introduced in 2.4.20, but that was only about the client
IP, not the vhost/request-line which have always been blank in
SERVER_BUSY_READ...

Your proposed backport looks good to me now, thanks for your patience :)

Mime
View raw message