On Fri, Jan 30, 2009 at 12:35 PM, Ian Kent <<a href="mailto:raven@themaw.net">raven@themaw.net</a>> wrote:<br>><br>[...]<br>> ><br>> > So, following the instructions here:<br>> > <a href="http://wiki.centos.org/HowTos/Custom_Kernel">http://wiki.centos.org/HowTos/Custom_Kernel</a>, I did the following:<br>
> ><br>> > 1. Added the patch ( unionfs-2.4_for_2.6.18-RHEL5.diff ) to<br>> > ~/rpmbuild/SOURCES, and referenced it in<br>> > ~/rpmbuild/SPECS/kernel-2.6.spec .<br>> ><br>> > 2.Turned off the KABI check and provided a new buildid in the<br>
> > kernel-2.6.spec in ~/rpmbuild/SPECS<br>> ><br>> > 3. Build using rpmbuild -bb --target=`uname -m` --with baseonly<br>> > kernel-2.6.spec<br>><br>> Right, but I didn't seem to be able to disable the kabichk either on the<br>
> command line or in the spec file itself, a bit odd.<br>><br><br>For disabling the kabi check, I followed the instructions given in CentOS HowTo verbatim:<br><a href="http://wiki.centos.org/HowTos/Custom_Kernel">http://wiki.centos.org/HowTos/Custom_Kernel</a>, Section 3 (Modifying the kernel spec file) .<br>
<br><br><br>> But, lets not forget that if you use these rpms later some time and you<br>> have another third party module on the machine that relies on this kabi<br>> function then it will break and you will probably have forgotten the<br>
> kabi issue by then, caught!!! It's best to at least maintain the kabi in<br>> the RHEL kernel if possible.<br><br>Correct.<br>Even if I try to spin a custom kernel ( and thereby forgo the kabi ), it may well not work !<br>
<br>Is it possible to apply the unionfs patch to RHEL kernel, without breaking the ABI - I guess not !<br><br>> ><br>> > and the build went fine. I haven't yet tried the new kernel and<br>> > module, but Erez mentioned in one of the messages about XATTR support<br>
> > in unionfs thus:<br>> ><br>> > <a href="http://www.fsl.cs.sunysb.edu/pipermail/unionfs/2008-May/005849.html">http://www.fsl.cs.sunysb.edu/pipermail/unionfs/2008-May/005849.html</a><br>> ><br>> > >Andy, there's no xattr support in unionfs for 2.6.18 (the VFS was<br>
> > lacking<br>> > >some vital support, I don't recommend it). So plz turn off<br>> > >CONFIG_UNION_FS_XATTR.<br>> ><br>> > However, Without the CONFIG_UNION_FS_XATTR=y insertion into spec file,<br>
> > the build fails !!<br>> ><br>> > Hence I'm wondering whether will this workaround work ? And do we have<br>> > a solution to this issue already, which I may have missed ?<br>><br>> Have to wait a while and see if I can build it I guess.<br>
><br>> Ian<br>><br><br>Tried the new kernel... its kernel panic ! <br>I sensed that unionfs design is in conflict with the VFS layer, is that issue resolved ?<br><br><br><br>