You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Justin Erenkrantz <je...@apache.org> on 2002/06/25 22:21:29 UTC

[PATCH] httpd discards body too soon

On Tue, Jun 25, 2002 at 04:51:06PM -0500, Karl Fogel wrote:
> Justin Erenkrantz <je...@apache.org> writes:
> > Hmph.  I'll try to reproduce here, but without the series of SVN
> > commands, this'll be hard.  Even if you can send me hard repro
> > recipies, I can probably sort it out.  
> > 
> > Looks like it's not eating the last line of the previous
> > request.  Odd.  -- justin
> 
> Well, basically I just ran an svn tests over DAV:
> 
>    $ cd subversion/tests/clients/cmdline
>    $ ./basic_tests.py 1 --url http://localhost
> 
> after having set up my Apache in the way recommended in the README in
> that directory, of course.

Okay, the problem is that ap_discard_request_body now checks the
keepalive status of the connection (correctly defaults to 0 which
means that ap_discard_request_body never discards requests).  That
causes a problem when ap_http_header_filter has not yet run to set
the connection's keepalive state.  mod_dav never sent an EOS, so
ap_http_header_filter never executed.  So, we must call
ap_discard_request_body only when we know the state of the
connection - which means that we must do so after the EOS is sent
to guarantee that the HTTP header filter ran.

This is a temporary fix, as I think we need to rethink checking the
connection status in ap_discard_request_body since the connection
status is only set once ap_http_header_filter is involved.  This
pretty much invalidates calling ap_discard_request_body() anywhere
except once we know the EOS is sent.  -- justin

Index: server/protocol.c
===================================================================
RCS file: /home/cvs/httpd-2.0/server/protocol.c,v
retrieving revision 1.106
diff -u -r1.106 protocol.c
--- server/protocol.c	10 Jun 2002 18:51:38 -0000	1.106
+++ server/protocol.c	25 Jun 2002 22:14:28 -0000
@@ -1046,8 +1046,6 @@
  */
 AP_DECLARE(void) ap_finalize_request_protocol(request_rec *r)
 {
-    (void) ap_discard_request_body(r);
-
     while (r->next) {
         r = r->next;
     }
@@ -1056,6 +1054,12 @@
     if (!r->eos_sent) {
         end_output_stream(r);
     }
+
+    /* Discard the body after we have ended our output stream.
+     * This allows the ap_http_header_filter to set the keepalive
+     * status of the request.
+     */
+    (void) ap_discard_request_body(r);
 }
 
 /*

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH] httpd discards body too soon

Posted by Cliff Woolley <jw...@virginia.edu>.
On Tue, 25 Jun 2002, Aaron Bannert wrote:

> On Tue, Jun 25, 2002 at 04:25:10PM -0700, Greg Stein wrote:
> > No more magic constants!
>
> Here here! +1

++1


Re: [PATCH] httpd discards body too soon

Posted by Aaron Bannert <aa...@clove.org>.
On Tue, Jun 25, 2002 at 04:25:10PM -0700, Greg Stein wrote:
> No more magic constants!

Here here! +1

-a

Re: [PATCH] httpd discards body too soon

Posted by Greg Stein <gs...@lyra.org>.
On Tue, Jun 25, 2002 at 04:14:00PM -0700, Justin Erenkrantz wrote:
> On Tue, Jun 25, 2002 at 03:55:52PM -0700, Greg Stein wrote:
> > What do you think of the tri-state solution?
> > 
> > It *does* mean that we might discard a body, but then later find out we
> > didn't need to. However, it would remove an accidental non-discard.
> 
> That'd work. I believe there are some modules that are expecting
> that 0 is closed.  
> 
> Hmm.  Oh, joy:
> 
> /** Are we going to keep the connection alive for another request?
>  *  -1 fatal error, 0 undecided, 1 yes   */
> signed int keepalive:2;
> 
> Sounds like we just need an explicit no here.  Or, we can use the
> -1 to mean 'no.'  However, a lot of places are using 0 for no -
> that behavior looks incorrect.  -- justin

I say, axe the darned bitfield. That is no 80's. Replace the keepalive flag
with an explicit enumeration.

No more magic constants!

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH] httpd discards body too soon

Posted by Greg Stein <gs...@lyra.org>.
On Tue, Jun 25, 2002 at 04:14:00PM -0700, Justin Erenkrantz wrote:
> On Tue, Jun 25, 2002 at 03:55:52PM -0700, Greg Stein wrote:
> > What do you think of the tri-state solution?
> > 
> > It *does* mean that we might discard a body, but then later find out we
> > didn't need to. However, it would remove an accidental non-discard.
> 
> That'd work. I believe there are some modules that are expecting
> that 0 is closed.  
> 
> Hmm.  Oh, joy:
> 
> /** Are we going to keep the connection alive for another request?
>  *  -1 fatal error, 0 undecided, 1 yes   */
> signed int keepalive:2;
> 
> Sounds like we just need an explicit no here.  Or, we can use the
> -1 to mean 'no.'  However, a lot of places are using 0 for no -
> that behavior looks incorrect.  -- justin

I say, axe the darned bitfield. That is no 80's. Replace the keepalive flag
with an explicit enumeration.

No more magic constants!

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

Re: [PATCH] httpd discards body too soon

Posted by Justin Erenkrantz <je...@apache.org>.
On Tue, Jun 25, 2002 at 03:55:52PM -0700, Greg Stein wrote:
> What do you think of the tri-state solution?
> 
> It *does* mean that we might discard a body, but then later find out we
> didn't need to. However, it would remove an accidental non-discard.

That'd work. I believe there are some modules that are expecting
that 0 is closed.  

Hmm.  Oh, joy:

/** Are we going to keep the connection alive for another request?
 *  -1 fatal error, 0 undecided, 1 yes   */
signed int keepalive:2;

Sounds like we just need an explicit no here.  Or, we can use the
-1 to mean 'no.'  However, a lot of places are using 0 for no -
that behavior looks incorrect.  -- justin

Re: [PATCH] httpd discards body too soon

Posted by Justin Erenkrantz <je...@apache.org>.
On Tue, Jun 25, 2002 at 03:55:52PM -0700, Greg Stein wrote:
> What do you think of the tri-state solution?
> 
> It *does* mean that we might discard a body, but then later find out we
> didn't need to. However, it would remove an accidental non-discard.

That'd work. I believe there are some modules that are expecting
that 0 is closed.  

Hmm.  Oh, joy:

/** Are we going to keep the connection alive for another request?
 *  -1 fatal error, 0 undecided, 1 yes   */
signed int keepalive:2;

Sounds like we just need an explicit no here.  Or, we can use the
-1 to mean 'no.'  However, a lot of places are using 0 for no -
that behavior looks incorrect.  -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org

Re: [PATCH] httpd discards body too soon

Posted by Greg Stein <gs...@lyra.org>.
On Tue, Jun 25, 2002 at 03:21:29PM -0700, Justin Erenkrantz wrote:
>...
> Okay, the problem is that ap_discard_request_body now checks the
> keepalive status of the connection (correctly defaults to 0 which
> means that ap_discard_request_body never discards requests).

I'm not sure that "correctly defaults to 0" is truly correct. It seems that
we would want a tri-state value for that variable: unknown, connection-
closes, and connection-stays-open.

This would have solved the problem at hand because the value would have been
"unknown", so the discard would have happened.

>...
> This is a temporary fix, as I think we need to rethink checking the
> connection status in ap_discard_request_body since the connection
> status is only set once ap_http_header_filter is involved.  This
> pretty much invalidates calling ap_discard_request_body() anywhere
> except once we know the EOS is sent.  -- justin

What do you think of the tri-state solution?

It *does* mean that we might discard a body, but then later find out we
didn't need to. However, it would remove an accidental non-discard.

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

Re: [PATCH] httpd discards body too soon

Posted by Greg Stein <gs...@lyra.org>.
On Tue, Jun 25, 2002 at 03:21:29PM -0700, Justin Erenkrantz wrote:
>...
> Okay, the problem is that ap_discard_request_body now checks the
> keepalive status of the connection (correctly defaults to 0 which
> means that ap_discard_request_body never discards requests).

I'm not sure that "correctly defaults to 0" is truly correct. It seems that
we would want a tri-state value for that variable: unknown, connection-
closes, and connection-stays-open.

This would have solved the problem at hand because the value would have been
"unknown", so the discard would have happened.

>...
> This is a temporary fix, as I think we need to rethink checking the
> connection status in ap_discard_request_body since the connection
> status is only set once ap_http_header_filter is involved.  This
> pretty much invalidates calling ap_discard_request_body() anywhere
> except once we know the EOS is sent.  -- justin

What do you think of the tri-state solution?

It *does* mean that we might discard a body, but then later find out we
didn't need to. However, it would remove an accidental non-discard.

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org