On Fri, 2008-08-15 at 12:10 +0200, Ondrej Valousek wrote: > >> We had a related issue quite a bit and /etc/mtab and > >> /proc/mounts went out of sync. Right now we are symlinking > >> /etc/mtab to /proc/mounts. It works, but I don't know if > >> that's the best solution for it. > >> > > > > Yes, we could do that and that seems to be a canonical solution that is > > even talked about in the 'man' pages. But, there may be a problem with > > that. > > > > Have you seen any cases where different processes are specifically > > writing either to /etc/mtab _or_ /proc/mounts in an asynchronous manner? > > Since it would now be the same file due to it being symlinked, what > > would that sort of write access do to the integrity of the mount tables? > > Would it even matter? > > > > > 1. You can not write to /proc/mounts as it is read only > 2. mount.nfs is broken (bug already filed) the way that it can not > handle symlink /etc/mtab -> /proc/mounts properly. > 3. I believe content handling of /proc/mounts is the kernel's job. If it > does not do it well, it should be patched -> we should not patch > automounter as it is not its responsibility. The similar with /etc/mtab > - it is the mount responsibility to handle it.
That's right. We do have a patch to check if /etc/mtab is a symlink to /proc/mounts and use the "-n" option with mount(8) if it is as mount(8) returns a fail (error is failed to update mtab, surprise, surprise) if we don't. Ian _______________________________________________ autofs mailing list [email protected] http://linux.kernel.org/mailman/listinfo/autofs
