On 9/6/07, Iruata Souza <iru.muzgo@gmail.com> wrote:

> your better documentation really is source code and man pages. i'm
> developing a filesystem capable of speaking 9P [1][2], the Plan 9 file
> protocol [3], for OpenBSD and these has shown as a great source of
> information.

that sounds like an interesting project (I've done a little bit of
playing with Plan9 myself) - let me know how it goes!

> so, in my understanding of things...
>
> sys/kern/vfs_cache.c states:
> "Cache is indexed by hash value obtained from (vp, name) where vp
> refers to the directory containing name. "
>
> a name in a directory is hashed with it's parent directory vnode, one
> at a time. it wouldn't make sense to me if the entire directory
> contents where hashed since we're looking up only one entry.

*nod* I was thinking along the lines of some kind of hash database; I
see that this doesn't make much sense.

> the lookup process is a serial one. i think this is why the cache is
> so important. once you call lookup(), in sys/kern/vfs_lookup.c, it's
> guaranteed that the path resolution process is done when it returns -
> the name being found or not.

so I guess now my question is more related to size of that cache,
where it's stored, how it's managed, etc.

> i hope that answers some of your questions.

indeed - got me thinking in a different direction anyway. Thanks!
--
darkuncle@{gmail.com,darkuncle.net} || 0x5537F527
encrypted email to the latter address please
http://darkuncle.net/pubkey.asc for public key