> For production uses, I'd avoid stacked NFS mounts, instead mount
> (NFS, and some would say, everything) only directly from the root
> filesys, so one hanging cannot prevent an umount of anything else,
> and then either use a symlink, or an overlay type mount (null mount)
> to make it appear to be where you would prefer it to be mounted, if
> not for this kind of issue.
OK, what I really have is:
1. some r/o NFS mount on /var/chroot/tftpd
2. some r/w NFS mount on /var/chroot/tftpd/RW
I could mount 1. on /var/chroot/tftpd/RO, put a symlink RW->../RW in there
and start tftpd with RO as an argument, but I'm unsure whether tftpd wil
follow the symlink.
I could use a null mount instead of the symlink, but then I'd run into the
same problem as now (except that I would probably be able to umount the
r/w NFS mount.
Better ideas?
> But we should also be fixing our code, so that umount -f *never* fails
> (for anything except the root), no matter what else happens (the -R
> just stops the umount program from causing issues).
Yes.
