Use of nameidata in 2.6

Use of nameidata in 2.6

[Ian Kent]

Hi all,

It appears that in 2.6 a nameidata struct is passed to lookup and
revalidate VFS functions. It appears to contain the parent vfsmount and
dentry addresses.

If a mount is carried out in either of these routines should this struct
be updated with pointers to the new parent entries or should this be left
to the caller (say link_path_walk).

Any comments on the intended use and what it should contain during
differnt function calls would be welcome.

Thanks
Ian

-- 

   ,-._|\    Ian Kent
  /      \   Perth, Western Australia
  *_.--._/   E-mail: raven@themaw.net
        v    Web: http://themaw.net/

Re: Use of nameidata in 2.6

[Trond Myklebust]

>>>>> " " == Ian Kent <raven@themaw.net> writes:

     > Any comments on the intended use and what it should contain
     > during differnt function calls would be welcome.

The nameidata+intents are there in order to provide hints to
filesystems like NFS and Lustre as to what is the ultimate purpose of
the call to lookup(). This allows the filesystem to perform
optimizations in certain cases.
It also will tell NFSv4 when to use the OPEN call rather than LOOKUP
on the wire (patches not yet merged).

I suggest you treat the nameidata as a read-only structure in this
case unless you are planning on bypassing the VFS and calling the
newly mounted filesystem directly (but why would you want to do that?).

Cheers,
  Trond

Re: Use of nameidata in 2.6

[Ian Kent]

On 25 Nov 2003, Trond Myklebust wrote:

> >>>>> " " == Ian Kent <raven@themaw.net> writes:
>
>      > Any comments on the intended use and what it should contain
>      > during differnt function calls would be welcome.
>
> The nameidata+intents are there in order to provide hints to
> filesystems like NFS and Lustre as to what is the ultimate purpose of
> the call to lookup(). This allows the filesystem to perform
> optimizations in certain cases.
> It also will tell NFSv4 when to use the OPEN call rather than LOOKUP
> on the wire (patches not yet merged).
>
> I suggest you treat the nameidata as a read-only structure in this
> case unless you are planning on bypassing the VFS and calling the
> newly mounted filesystem directly (but why would you want to do that?).
>

That was what I originally thought.

Possibly when a lookup causes a mount in autofs4 and needs to redo the
lookup (it appears I need new values for dentry and mnt) and return
the new looked up dentry.

That was when I became unsure of it's purpose.

-- 

   ,-._|\    Ian Kent
  /      \   Perth, Western Australia
  *_.--._/   E-mail: raven@themaw.net
        v    Web: http://themaw.net/