[Help-gnutls] GnuTLS 2.2.1 memory leaks or just bad usage?

[Nikos Mavrogiannopoulos]

O/H Daniel Stenberg έγραψε:
> First it doesn't build, it needs a #include "gnutls_datum.h" in 
> gnutls_session.c! (If you want me to, I can produce an updated patch 
> that applies cleanly.)
>
> Then, yes it _does_ fix the worst leaks in my test case. The reamining 
> leak now looks like this (with a build using only -g and not -O2 that 
> otherwise confuse symbols like the ones in my initial report):
>
> ==13872== 128 bytes in 1 blocks are definitely lost in loss record 2 of 8
> ==13872== at 0x4022AB8: malloc (vg_replace_malloc.c:207)
> ==13872== by 0x40FBC15: _gnutls_mpi_dprint_lz (gnutls_mpi.c:146)
> ==13872== by 0x410B4C3: _gnutls_dh_set_peer_public (gnutls_state.c:474)
> ==13872== by 0x411354E: _gnutls_proc_dh_common_server_kx 
> (auth_dh_common.c:297)
> ==13872== by 0x4108404: proc_dhe_server_kx (auth_dhe.c:199)
> ==13872== by 0x40F079F: _gnutls_recv_server_kx_message (gnutls_kx.c:401)
> ==13872== by 0x40EB332: _gnutls_handshake_client 
> (gnutls_handshake.c:2342)
> ==13872== by 0x40EADF9: gnutls_handshake (gnutls_handshake.c:2238)
> ==13872== by 0x8083C40: handshake (gtls.c:156)
> ==13872== by 0x80841D5: Curl_gtls_connect (gtls.c:326)
> ==13872== by 0x806169A: Curl_ssl_connect (sslgen.c:215)
> ==13872== by 0x806C605: Curl_http_connect (http.c:1700)
>
What about this patch? It's quite intrusive. My development environment 
is almost inexistent now, thus I'd appreciate if you or any other 
volunteers could test it.

regards,
Nikos

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch.txt
URL: </pipermail/attachments/20080216/05b21555/attachment.txt>