Details
-
Bug
-
Status: Resolved
-
Critical
-
Resolution: Invalid
-
None
-
None
-
None
-
I was running Linux (Ubuntu 8.04), Apache httpd v. 2 2 9, but this probably affects all platforms.
Description
A copy of my original e-mail to the Axis2/C mail list with the description of the problem:
Hi,
I found another serious memory leak in the mod_axis2.
Using the SVN's version of the Axis2/C, in the file
src/core/transport/http/server/apache2/mod_axis2.c
in the function axis2_handler():
at the very beginning of the function there are two function calls:
apr_allocator_create(&local_allocator)
apr_pool_create_ex(&local_pool, NULL, NULL, local_allocator);
(right after the local variable declarations).
These calls apparently allocate memory that is supposed to be released
by the call
axis2_module_pool_allocator_destroy(allocator);
at the end of the function.
Apache calls axis2_handler() for every request it receives, not only
for the Axis requests. So this function checks if this is request for
the Axis:
if (strcmp(req->handler, "axis2_module"))
{ return DECLINED; }...
and returns DECLINED if it it not.
The problem is that this check is done after the calls of
apr_allocator_create() and apr_pool_create_ex(). So in the case the
request is not for Axis, the memory allocated by these functions
will be lost (leaked).
On my machine, I'm developing an Axis service and another Apache
module, completely unrelated to the Axis service. In the Apache config
I'm loading the Axis module mod_axis2 - even when I'm not working on
the Axis service.
Couple of days ago I did a stress-test for this my other module (like,
running 800 client test threads that all are sending requests to the
Apache that runs my module, for sustained period of time) and found
out that the memory in the Apache client processes is exhausting very
quickly. I have spend these couple of days trying to understand where
all this memory goes
Finally, today I removed loading of the mod_axis from the Apache
config - and the memory leaks stopped
So I looked at the Axis2/C source code, and found the problem as
described above.
I did an experiment - I moved those two calls of
apr_allocator_create() and apr_pool_create_ex() after the check of "if
(strcmp(req->handler, "axis2_module"))" and re-build the Axis2/C. This
fixed the memory leak.
So here I submit this message instead of the patch, as I'm not sure,
it looks like there are another couple of returns (in case of some
errors) from the axis2_handler() where the memory allocated by those
two calls is not released, so please, would someone who is more
familiar with the code take a look at this?
Thank you,
alex.