Graham Leggett <[EMAIL PROTECTED]> writes: > Peter Van Biesen wrote: > > > Does anybody have another idea for me to try ? > > Have you tried the latest fix for the client_block stuff, I think I saw > a very recent CVS checkin...? > > There could of course be more than one leak, and we'll only fix the > problem once all of them are found...
There's also a refcount problem in http_protocol.c wrt chunked transfer codings. The problem is that the bucket holding the chunk size isn't ever freed, so the corresponding data block is overcounted. Here's a diff for http_protocol.c against current anon-cvs (which doesn't seem to have any of the newer changes to ap_get_client_block). Light testing seems to indicate that this fixes the refcount problem. diff -u -r1.454 http_protocol.c --- http_protocol.c 13 Aug 2002 14:27:39 -0000 1.454 +++ http_protocol.c 5 Sep 2002 17:15:12 -0000 @@ -901,6 +901,7 @@ if (rv == APR_SUCCESS) { rv = apr_brigade_flatten(bb, line, &len); if (rv == APR_SUCCESS) { + apr_brigade_cleanup(bb); ctx->remaining = get_chunk_size(line); } } @@ -966,6 +967,7 @@ if (rv == APR_SUCCESS) { rv = apr_brigade_flatten(bb, line, &len); if (rv == APR_SUCCESS) { + apr_brigade_cleanup(bb); ctx->remaining = get_chunk_size(line); } }