You are viewing a plain text version of this content. The canonical link for it is here.
Posted to cvs@httpd.apache.org by tr...@apache.org on 2010/05/10 22:47:59 UTC

svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Author: trawick
Date: Mon May 10 20:47:59 2010
New Revision: 942897

URL: http://svn.apache.org/viewvc?rev=942897&view=rev
Log:
prefork MPM: Run cleanups for final request when process exits gracefully.
PR: 43857
Submitted by: Tom Donovan, simplified slightly by me

Modified:
    httpd/httpd/trunk/CHANGES
    httpd/httpd/trunk/server/mpm/prefork/prefork.c

Modified: httpd/httpd/trunk/CHANGES
URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/CHANGES?rev=942897&r1=942896&r2=942897&view=diff
==============================================================================
--- httpd/httpd/trunk/CHANGES [utf-8] (original)
+++ httpd/httpd/trunk/CHANGES [utf-8] Mon May 10 20:47:59 2010
@@ -28,6 +28,9 @@ Changes with Apache 2.3.7
      processing is completed, avoiding orphaned callback pointers.
      [Brett Gervasoni <brettg senseofsecurity.com>, Jeff Trawick]
 
+  *) prefork MPM: Run cleanups for final request when process exits gracefully.
+     PR 43857.  [Tom Donovan, Jeff Trawick]
+
   *) ab: fix number of requests sent by ab when keepalive is enabled.  PR 48497.
      [Bryn Dole <dole blekko.com>]
 

Modified: httpd/httpd/trunk/server/mpm/prefork/prefork.c
URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/server/mpm/prefork/prefork.c?rev=942897&r1=942896&r2=942897&view=diff
==============================================================================
--- httpd/httpd/trunk/server/mpm/prefork/prefork.c (original)
+++ httpd/httpd/trunk/server/mpm/prefork/prefork.c Mon May 10 20:47:59 2010
@@ -549,12 +549,6 @@ static void child_main(int child_num_arg
         conn_rec *current_conn;
         void *csd;
 
-        /*
-         * (Re)initialize this child to a pre-connection state.
-         */
-
-        apr_pool_clear(ptrans);
-
         if ((ap_max_requests_per_child > 0
              && requests_this_child++ >= ap_max_requests_per_child)) {
             clean_child_exit(0);
@@ -668,6 +662,12 @@ static void child_main(int child_num_arg
             ap_lingering_close(current_conn);
         }
 
+        /*
+         * (Re)initialize this child to a pre-connection state.
+         */
+
+        apr_pool_clear(ptrans);
+
         /* Check the pod and the generation number after processing a
          * connection so that we'll go away if a graceful restart occurred
          * while we were processing the connection or we are the lucky



Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Jeff Trawick <tr...@gmail.com>.
On Thu, May 13, 2010 at 4:43 AM, Ruediger Pluem <rp...@apache.org> wrote:
> On 10.05.2010 22:47, trawick@apache.org wrote:
>> Author: trawick
>> Date: Mon May 10 20:47:59 2010
>> New Revision: 942897
>>
>> URL: http://svn.apache.org/viewvc?rev=942897&view=rev
>> Log:
>> prefork MPM: Run cleanups for final request when process exits gracefully.
>> PR: 43857
>> Submitted by: Tom Donovan, simplified slightly by me
>>
>> Modified:
>>     httpd/httpd/trunk/CHANGES
>>     httpd/httpd/trunk/server/mpm/prefork/prefork.c
>>
>> Modified: httpd/httpd/trunk/CHANGES
>> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/CHANGES?rev=942897&r1=942896&r2=942897&view=diff
>> ==============================================================================
>> --- httpd/httpd/trunk/CHANGES [utf-8] (original)
>> +++ httpd/httpd/trunk/CHANGES [utf-8] Mon May 10 20:47:59 2010
>> @@ -28,6 +28,9 @@ Changes with Apache 2.3.7
>>       processing is completed, avoiding orphaned callback pointers.
>>       [Brett Gervasoni <brettg senseofsecurity.com>, Jeff Trawick]
>>
>> +  *) prefork MPM: Run cleanups for final request when process exits gracefully.
>> +     PR 43857.  [Tom Donovan, Jeff Trawick]
>> +
>>    *) ab: fix number of requests sent by ab when keepalive is enabled.  PR 48497.
>>       [Bryn Dole <dole blekko.com>]
>>
>>
>> Modified: httpd/httpd/trunk/server/mpm/prefork/prefork.c
>> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/server/mpm/prefork/prefork.c?rev=942897&r1=942896&r2=942897&view=diff
>> ==============================================================================
>> --- httpd/httpd/trunk/server/mpm/prefork/prefork.c (original)
>> +++ httpd/httpd/trunk/server/mpm/prefork/prefork.c Mon May 10 20:47:59 2010
>> @@ -549,12 +549,6 @@ static void child_main(int child_num_arg
>>          conn_rec *current_conn;
>>          void *csd;
>>
>> -        /*
>> -         * (Re)initialize this child to a pre-connection state.
>> -         */
>> -
>> -        apr_pool_clear(ptrans);
>> -
>
> Doesn't this create a memory leak? Don't we need to reset this pool for every
> new connection?

This original commit had a potential leak when apr_socket_accept()
failed, but it did clear the pool at the end of handling a
successfully-accepted connection.  The final commit should handle
either situation.

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Rainer Jung <ra...@kippdata.de>.
Rüdiger,

Jeff rolled back this change in r943648 after a comment by Joe and 
committed an alternative approach in r943650.

Regards,

Rainer

On 13.05.2010 10:43, Ruediger Pluem wrote:
> On 10.05.2010 22:47, trawick@apache.org wrote:
>> Author: trawick
>> Date: Mon May 10 20:47:59 2010
>> New Revision: 942897
>>
>> URL: http://svn.apache.org/viewvc?rev=942897&view=rev
>> Log:
>> prefork MPM: Run cleanups for final request when process exits gracefully.
>> PR: 43857
>> Submitted by: Tom Donovan, simplified slightly by me
>>
>> Modified:
>>      httpd/httpd/trunk/CHANGES
>>      httpd/httpd/trunk/server/mpm/prefork/prefork.c
>>
>> Modified: httpd/httpd/trunk/CHANGES
>> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/CHANGES?rev=942897&r1=942896&r2=942897&view=diff
>> ==============================================================================
>> --- httpd/httpd/trunk/CHANGES [utf-8] (original)
>> +++ httpd/httpd/trunk/CHANGES [utf-8] Mon May 10 20:47:59 2010
>> @@ -28,6 +28,9 @@ Changes with Apache 2.3.7
>>        processing is completed, avoiding orphaned callback pointers.
>>        [Brett Gervasoni<brettg senseofsecurity.com>, Jeff Trawick]
>>
>> +  *) prefork MPM: Run cleanups for final request when process exits gracefully.
>> +     PR 43857.  [Tom Donovan, Jeff Trawick]
>> +
>>     *) ab: fix number of requests sent by ab when keepalive is enabled.  PR 48497.
>>        [Bryn Dole<dole blekko.com>]
>>
>>
>> Modified: httpd/httpd/trunk/server/mpm/prefork/prefork.c
>> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/server/mpm/prefork/prefork.c?rev=942897&r1=942896&r2=942897&view=diff
>> ==============================================================================
>> --- httpd/httpd/trunk/server/mpm/prefork/prefork.c (original)
>> +++ httpd/httpd/trunk/server/mpm/prefork/prefork.c Mon May 10 20:47:59 2010
>> @@ -549,12 +549,6 @@ static void child_main(int child_num_arg
>>           conn_rec *current_conn;
>>           void *csd;
>>
>> -        /*
>> -         * (Re)initialize this child to a pre-connection state.
>> -         */
>> -
>> -        apr_pool_clear(ptrans);
>> -
>
> Doesn't this create a memory leak? Don't we need to reset this pool for every
> new connection?
>
> Regards
>
> Rüdiger

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Ruediger Pluem <rp...@apache.org>.
On 10.05.2010 22:47, trawick@apache.org wrote:
> Author: trawick
> Date: Mon May 10 20:47:59 2010
> New Revision: 942897
> 
> URL: http://svn.apache.org/viewvc?rev=942897&view=rev
> Log:
> prefork MPM: Run cleanups for final request when process exits gracefully.
> PR: 43857
> Submitted by: Tom Donovan, simplified slightly by me
> 
> Modified:
>     httpd/httpd/trunk/CHANGES
>     httpd/httpd/trunk/server/mpm/prefork/prefork.c
> 
> Modified: httpd/httpd/trunk/CHANGES
> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/CHANGES?rev=942897&r1=942896&r2=942897&view=diff
> ==============================================================================
> --- httpd/httpd/trunk/CHANGES [utf-8] (original)
> +++ httpd/httpd/trunk/CHANGES [utf-8] Mon May 10 20:47:59 2010
> @@ -28,6 +28,9 @@ Changes with Apache 2.3.7
>       processing is completed, avoiding orphaned callback pointers.
>       [Brett Gervasoni <brettg senseofsecurity.com>, Jeff Trawick]
>  
> +  *) prefork MPM: Run cleanups for final request when process exits gracefully.
> +     PR 43857.  [Tom Donovan, Jeff Trawick]
> +
>    *) ab: fix number of requests sent by ab when keepalive is enabled.  PR 48497.
>       [Bryn Dole <dole blekko.com>]
>  
> 
> Modified: httpd/httpd/trunk/server/mpm/prefork/prefork.c
> URL: http://svn.apache.org/viewvc/httpd/httpd/trunk/server/mpm/prefork/prefork.c?rev=942897&r1=942896&r2=942897&view=diff
> ==============================================================================
> --- httpd/httpd/trunk/server/mpm/prefork/prefork.c (original)
> +++ httpd/httpd/trunk/server/mpm/prefork/prefork.c Mon May 10 20:47:59 2010
> @@ -549,12 +549,6 @@ static void child_main(int child_num_arg
>          conn_rec *current_conn;
>          void *csd;
>  
> -        /*
> -         * (Re)initialize this child to a pre-connection state.
> -         */
> -
> -        apr_pool_clear(ptrans);
> -

Doesn't this create a memory leak? Don't we need to reset this pool for every
new connection?

Regards

Rüdiger


Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Jeff Trawick <tr...@gmail.com>.
On Wed, May 19, 2010 at 5:26 PM, Joe Orton <jo...@redhat.com> wrote:
> On Wed, May 12, 2010 at 03:30:29PM -0400, Jeff Trawick wrote:
>> The multiple-calls-to-pool_clear solution is definitely safer.
>
> OK, so now I reviewed that too ;) The only difference between:
>
>   apr_pool_clear(ptrans);
>   apr_pool_destroy(pchild);
>
> and simply:
>
>   apr_pool_destroy(pchild);
>
> (given that ptrans is a child of ptrans and _destroy runs cleanups in
> child pools)
>
> is that it hides problems in cleanup ordering/pool nesting.

hmmm, definitely a negative characteristic...  too bad it is
(apparently) rolled up in the following issue

> I think PR 43857 is about the reslist cleanup issue in APR-util, r677505
> and all that.  Not sure what the status of that is.

main, initial discussion ends here:
http://marc.info/?l=apr-dev&m=121728874016894&w=3

meanwhile, unreleased apr trunk and unreleased apr-util 1.4.x have

APU_DECLARE(void) apr_reslist_cleanup_order_set(apr_reslist_t *rl,
                                                apr_uint32_t mode)
{
    apr_pool_cleanup_kill(rl->pool, rl, reslist_cleanup);
    if (mode == APR_RESLIST_CLEANUP_FIRST)
        apr_pool_pre_cleanup_register(rl->pool, rl, reslist_cleanup);
    else
        apr_pool_cleanup_register(rl->pool, rl, reslist_cleanup,
                                  apr_pool_cleanup_null);
}

and apr_memcache uses that

discussion of backporting seems to have stopped at this point:
http://marc.info/?l=apr-dev&m=121661852120171&w=3

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Jeff Trawick <tr...@gmail.com>.
On Wed, May 19, 2010 at 5:26 PM, Joe Orton <jo...@redhat.com> wrote:
> On Wed, May 12, 2010 at 03:30:29PM -0400, Jeff Trawick wrote:
>> The multiple-calls-to-pool_clear solution is definitely safer.
>
> OK, so now I reviewed that too ;) The only difference between:
>
>   apr_pool_clear(ptrans);
>   apr_pool_destroy(pchild);
>
> and simply:
>
>   apr_pool_destroy(pchild);
>
> (given that ptrans is a child of ptrans and _destroy runs cleanups in
> child pools)
>
> is that it hides problems in cleanup ordering/pool nesting.

hmmm, definitely a negative characteristic...  too bad it is
(apparently) rolled up in the following issue

> I think PR 43857 is about the reslist cleanup issue in APR-util, r677505
> and all that.  Not sure what the status of that is.

main, initial discussion ends here:
http://marc.info/?l=apr-dev&m=121728874016894&w=3

meanwhile, unreleased apr trunk and unreleased apr-util 1.4.x have

APU_DECLARE(void) apr_reslist_cleanup_order_set(apr_reslist_t *rl,
                                                apr_uint32_t mode)
{
    apr_pool_cleanup_kill(rl->pool, rl, reslist_cleanup);
    if (mode == APR_RESLIST_CLEANUP_FIRST)
        apr_pool_pre_cleanup_register(rl->pool, rl, reslist_cleanup);
    else
        apr_pool_cleanup_register(rl->pool, rl, reslist_cleanup,
                                  apr_pool_cleanup_null);
}

and apr_memcache uses that

discussion of backporting seems to have stopped at this point:
http://marc.info/?l=apr-dev&m=121661852120171&w=3

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Joe Orton <jo...@redhat.com>.
On Wed, May 12, 2010 at 03:30:29PM -0400, Jeff Trawick wrote:
> The multiple-calls-to-pool_clear solution is definitely safer.

OK, so now I reviewed that too ;) The only difference between:

   apr_pool_clear(ptrans);
   apr_pool_destroy(pchild);

and simply:

   apr_pool_destroy(pchild);

(given that ptrans is a child of ptrans and _destroy runs cleanups in 
child pools)

is that it hides problems in cleanup ordering/pool nesting.

I think PR 43857 is about the reslist cleanup issue in APR-util, r677505 
and all that.  Not sure what the status of that is.

Regards, Joe

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Joe Orton <jo...@redhat.com>.
On Wed, May 12, 2010 at 03:30:29PM -0400, Jeff Trawick wrote:
> The multiple-calls-to-pool_clear solution is definitely safer.

OK, so now I reviewed that too ;) The only difference between:

   apr_pool_clear(ptrans);
   apr_pool_destroy(pchild);

and simply:

   apr_pool_destroy(pchild);

(given that ptrans is a child of ptrans and _destroy runs cleanups in 
child pools)

is that it hides problems in cleanup ordering/pool nesting.

I think PR 43857 is about the reslist cleanup issue in APR-util, r677505 
and all that.  Not sure what the status of that is.

Regards, Joe

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Jeff Trawick <tr...@gmail.com>.
On Wed, May 12, 2010 at 4:24 AM, Joe Orton <jo...@redhat.com> wrote:
> On Mon, May 10, 2010 at 08:47:59PM -0000, Jeff Trawick wrote:
>> --- httpd/httpd/trunk/server/mpm/prefork/prefork.c (original)
>> +++ httpd/httpd/trunk/server/mpm/prefork/prefork.c Mon May 10 20:47:59 2010
>> @@ -549,12 +549,6 @@ static void child_main(int child_num_arg
>>          conn_rec *current_conn;
>>          void *csd;
>>
>> -        /*
>> -         * (Re)initialize this child to a pre-connection state.
>> -         */
>> -
>> -        apr_pool_clear(ptrans);
>> -
>
> ptrans will not be cleared between loop iterations in the case where
> accept fails (the "continue" path) - isn't that a potential memory leak?

If the accept() call itself fails, I don't think it will leak with
current code; if something in APR after accept() fails (probably
should-not-occur), then yes.

The multiple-calls-to-pool_clear solution is definitely safer.

(Thanks!)

Re: svn commit: r942897 - in /httpd/httpd/trunk: CHANGES server/mpm/prefork/prefork.c

Posted by Joe Orton <jo...@redhat.com>.
On Mon, May 10, 2010 at 08:47:59PM -0000, Jeff Trawick wrote:
> --- httpd/httpd/trunk/server/mpm/prefork/prefork.c (original)
> +++ httpd/httpd/trunk/server/mpm/prefork/prefork.c Mon May 10 20:47:59 2010
> @@ -549,12 +549,6 @@ static void child_main(int child_num_arg
>          conn_rec *current_conn;
>          void *csd;
>  
> -        /*
> -         * (Re)initialize this child to a pre-connection state.
> -         */
> -
> -        apr_pool_clear(ptrans);
> -

ptrans will not be cleared between loop iterations in the case where 
accept fails (the "continue" path) - isn't that a potential memory leak?

Regards, Joe