On 07/22/2013 12:50 PM, J. R. Okajima wrote: > I think I could see the scenario. > > - on wheezy, /etc/mtab becomes a symlink to /proc/mounts. > - mount.nfs writes the given mount options to /etc/mtab, (but not to > /proc/mounts.) > - I found the problematic option is "sec=sys". > - when /etc/mtab is a regular file, "sec=sys" is not written since it > was not specified explicitly at the mount-time. > - but /proc/mounts always shows "sec=". > - when /etc/mtab is a symlink, mount(8) retrives the option strings from > /proc/mounts including "sec=sys" and passes it to mount.nfs. > - NFS in kernel-space parses the given options, compares them with the > internally stored flags, and found that > + "sec=sys" is given explicitly. > + but the internal status doesn't have "sec=sys" flag set. > - finally this un-matching is the cause of "mount.nfs: mount(2): Invalid > argument." > > In order to fix the problem, there may exist several ways. But I am not > sure which is best. > - stop /etc/mtab symlink-ing /proc/mounts (by Debian developers). > - at the mount-time of NFS, set "sec=sys" (NFS_MOUNT_SECFLAVOUR) by > default (by NFS developers). > - mount.nfs always writes the "sec=sys" string (by mount.nfs > developers). > > I suppose the best way should be decided after the discussion between > NFS developers, mount.nfs developers and Debian developers.
Interesting. I can confirm that the problem does happen on a test
wheezy instance using sec=sys with /etc/mtab symlinked to /proc/mounts,
and that it does *not* happen on a wheezy instance (also with the
/etc/mtab→/proc/mounts symlink) using sec=krb5p.
This suggests to me that it might be a bug in the way that sec=sys in
particular is handled by the NFS tools.
--dkg
signature.asc
Description: OpenPGP digital signature
