2007/416 NFSv4 Mirror-mounts

Calum Mackay Calum.Mackay at sun.com
Wed Jul 18 16:33:02 PDT 2007


thanks much for the comments Nico.

The NFSv4 namespace work that has been considered can be split into
several areas:

o Mirror-mounts

o Referrals

o Replication & Migration

o Automount refactoring to not use MOUNT for NFSv4

This case is about enabling mirror-mounts. Whilst the automount
refactoring is useful work, I feel that it's a separate case. In 
addition, the Referrals work proposed also requires automounter 
refactoring, and if the automount/MOUNT work were to be bundled in with 
something, it would make more sense for it to be with Referrals.

With the proposed functionality, for mirror-mounts to be available
requires at least one NFSv4 filesystem to already be mounted, either
manually, or via the automounter.

Once that first v4 mount has occurred, mirror-mounts will be possible
from that point down.

Some examples:

Assume the server shares these separate filesystems:

     /export
     /export/sparc
     /export/x86

and also has a set of ZFS filesystems under /tank, all separately
shared (but /tank itself perhaps not explicitly shared).

If the client were to manually v4 mount:

<server>:/
mirror-mounts would show and provide access to everything

<server>:/export
mirror-mounts would show and provide access to everything under /export

<server>:/tank
mirror-mounts would show and provide access to everything under /tank

If automounter maps were setup to mount the same directories, the same
behaviour would apply. Regular v4 mounts would be used for these
directories themselves, and mirror-mounts would then be available to
show and provide access to everything under them.

Any new ZFS filesystems created (and shared) under /tank would become 
immediately visible, and accessible, via mirror-mounts.

Regarding /net: because the automounter retrieves the entire export list
from the server, using the MOUNT protocol, and then populates /net with
autofs trigger points for all of them, mirror-mounts do not get a chance
(because we don't override an autofs trigger point).



More information about the opensolaris-arc mailing list