Control: forwarded -1 https://github.com/dinhviethoa/libetpan/issues/145
Hi Steven, On Fri, Jun 13, 2014 at 12:11:06AM +1000, Steven McDonald wrote: > Package: libetpan17 > Version: 1.5-1 > Severity: important > Tags: upstream patch > > Hi there, > > Claws Mail (all versions since 3.9.3-2+b3, inclusive) reliably > segfaults seconds after startup on my system. I have attached a full > backtrace taken after such an event. > > I have tracked the problem down to a corner case in the interaction > between claws-mail and libetpan17, which seems simplest to fix in > libetpan. The trigger for the bug is that claws-mail links directly > against libgnutls26, but also against libetpan17, which links against > libgnutls-deb0-28. This has been corrected in claws-mail/3.10.1-2 and libetpan/1.5-1 and should fix the issue without patching. Can you upgrade and check? > This problem does not occur with versions of claws-mail prior to > 3.9.3-2+b3, which all link against libetpan16 (which links against > libgnutls26). > > claws-mail first calls gnutls_global_init, which initialises > libgnutls26, and then calls mailstream_gnutls_init_not_required in > libetpan17, which causes libetpan not to initialise libgnutls-deb0-28. > The result of this is that libetpan calls gnutls with an uninitialised > session handle, resulting in a segfault. > > However, since gnutls_global_init is reentrant, it is perfectly safe to > ignore any calls to mailstream_gnutls_init_not_required and initialise > gnutls anyway. In the case where gnutls has already been initialised, > doing so will simply increment a counter; and in the case where this > bug is triggered, gnutls will be properly initialised before use. > > OpenSSL's SSL_library_init is not reentrant, and therefore this change > can only apply to gnutls. > > I am attaching a patch which implements this change, and I have > confirmed that it fixes the segfaulting behaviour described above. > Please let me know if you think it needs to be revised, or if you need > any further information. Thanks for the detailed analysis! I'm forwarding this upstream to see if it's worth to add it, although seems good to me :) best regards, -- Ricardo Mones ~ Never send a human to do a machine's job. Agent Smith
signature.asc
Description: Digital signature
