Re: [fuse-devel] which would be the best modules to steal from for this project...

[Jan E. <je...@co...>]

On Aug 14 2007 04:20, Graham Toal wrote:
>On 8/14/07, Jan Engelhardt <je...@co...> wrote:
>> On Aug 13 2007 19:44, Graham Toal wrote:
>> So you could use http://aufs.sf.net/ - it's not FUSE,
>> but does COW in a way.
>
>Not too keen on rebuilding my kernel

It builds as a module.

>just to add a fuse lookalike of unknown quality,

It certainly does not silently lock up on an NFS setup like unionfs.

>when I have a perfectly good implementation of the real fuse already installed
>and working.  Anyway, I want to be able to distribute a working fuse module,
>not a fuse module and instructions to rebuild your kernel :-/
>
>> encfs.
>
>It's nice but you need to enter a password which I don't want.  Would
>prefer public key.

Change it.

>
>> RAIF1.
>
>Would be a *much* simpler implementation if the duplication were done by a
>fuse module (see above).  And I have no idea how to make software raid write
>to a non-blocked device

This is RAIF, not RAID. It works on *f*iles, not *d*evices.

>nor do I want to back up an entire filing system image (using scp, ftp,
>whatever) for a change to a single file.
>
>Anyway, if I do them all as fuse systems, once they're working I can
>consolidate the code into a single fuse module that does everything I want and
>which can be distributed to anyone with fuse installed as a trivial add-on.
>
>> There is not really anything to a nullfs.
>>
>> #define XRET(f) \
>>        return ({ \
>>                int __ret = (v); \
>>                (__ret >= 0) ? __ret : -errno; \
>>        })
>> int root_fd;
>> int nullfs_getattr(const char *path, struct stat *sb)
>> {
>>        XRET(fstatat(root_fd, path+1, sb));
>> }
>> int main(int argc, const char **argv)
>> {
>>        root_fd = open("/underlyin_dir", O_RDONLY | O_DIRECTORY);
>>        return fuse_main(...);
>> }
>
>Thanks for the effort, but I didn't really mean quite such a minimal
>nullfs! - I was looking for something that implemented all the
>filesystem calls so that if I wanted to intercept any particular one,
>the hook would be obvious (and hopefully trivial).

Well, symlink for example is exactly like getattr.

{
	XRET(symlinkat(root_fd, orig_path+1, root_fd, new_path+1);
}

Hence I can see the reason why noone has even bothered to do a nullfs. It's
like saying "oh write me a program which outputs the hard-coded PI constant."



	Jan
--