Re: [fuse-devel] Re: '

[Chris F. <ch...@fr...>]

On Thu, May 04, 2006 at 01:58:46PM +0200, Miklos Szeredi wrote:
> In theory there shouldn't be any probelm.  I just can't imagine any
> useful application for multiple filesystems per process.  What are you
> using it for?

A fwiw, we run our fuse-based fileserver with multiple filesystems per
process. We run in fuse (vs in the kernel) for safer testing and
easier debugging and profiling. We run in a single process because our
server's mounts must all share the same virtual address space and we
found telling (lowlevel) fuse about each mount to be cleaner than
exporting a single mount and then mapping to internal mounts, which
would require combining all mounts' inode spaces and special support
for a few operations (eg rename). (Inode-space combining and other
issues do not arise when exporting each mount since the mounts are
kept separate all the way into the kernel.)

Our fuse-based fileserver process exports multiple mounts through two
changes (both local to our program; nicely, there is no need to modify
the fuse library). (1) We rewrote a local version of lib/fuse_loop.c
to support multiple mounts. (2) We run mounts and unmounts in a second
thread so that when fuse_mount() and fuse_unmount() block (for fork,
exec, and wait) to get information about the mount's parent directory
(which might be served by the same fileserver) there is no deadlock.

A downside to this local modification approach is that we of course
have to watch for changes to lib/fuse_loop.c and to the general
internal fuse architecture.

-- 
Chris Frost  |  <http://www.frostnet.net/chris/>
-------------+----------------------------------
Public PGP Key:
   Email ch...@fr... with the subject "retrieve pgp key"
   or visit <http://www.frostnet.net/chris/about/pgp_key.phtml>