On Thu, 17 Dec 2009, Davide Libenzi wrote: > On Thu, 17 Dec 2009, Stephan Müller wrote: > > > On 17.12.2009, Davide Libenzi <davi...@xmailserver.org> wrote: > > > On Thu, 17 Dec 2009, Stephan Mueller wrote: > > > > Hi, > > > > > > > > I am starting to use xmail on an embedded system based on ARM. Due to > > > > the > > > > limited space available, I had to make IPv6 and SSL support a compile > > > > time option. > > > > > > > > Essentially, I added a bunch of ifdefs around the problematic code. > > > > There > > > > are not that many though. > > > > > > > > Do you want to have these patches? > > > > > > I will not merge them, but you can send them in if you like. Better yet, > > > is if you post a link, which I can add to the XMail home page. > > > Keep in mind though, that the new random tmp file name generation is based > > > in part upon RAND_pseudo_bytes(), which is part of OpenSSL. > > > > Hm, is it possible to refrain from OpenSSL? > > > > The best solution IMHO (because it uses an atomic operation) is mkdir. > > > > 1. register signal handler for signals 0, 1, 2, 3, 15 which removes > > /tmp/xmail > > > > 2. mkdir(/tmp/xmail) with permissions 755 at the startup of xmail > > > > 3. return /tmp/xmail/<sometmpfile> during the operation of xmail > > It'd be possible something similar, yes. But this will need to be > optional, since existing configs cannot be broken. > So a stronger temp file names generation is still necessary for legacy > systems. > You can patch-out the call in your code if you like, or provide a trivial > rand()-based implementation.
I made the XMail temp directory on Unix configurable via an 'XMAIL_TEMP' environment variable, defaulting to '/tmp'. So the user can set XMAIL_TEMP to whatever they like, and set the owner/permissions accordingly (which should be taken care also when running filters). - Davide
_______________________________________________ xmail mailing list xmail@xmailserver.org http://xmailserver.org/mailman/listinfo/xmail
