Re: update for 2.1.56 smbfs

Linus Torvalds (torvalds@transmeta.com)
Sat, 20 Sep 1997 10:45:01 -0700 (PDT)


On Sat, 20 Sep 1997, Bill Hawes wrote:
>
> The patch includes a couple of proposed additions to fs/dcache,
> shrink_dcache_sb() and a new unhash_dentries() routine. The latter does
> a search of the dentry hash table to unhash all dentries using an inode
> that has become invalid (e.g. becasue of a mode change.) Because it's
> possible that several dentries could be using a single inode, we want to
> drop all of them when the inode goes bad, rather than waiting for the
> next use. Any suggestions on this approach?

Duh. This sounds like there would finally have been a valid use of the
i_dentry list that I removed.

I suspect that it might be more straightforward to just re-instate the
i_dentry list. The reason I removed it was not because the i_dentry list
was evil in itself, it was that nobody used it and I prefer trying to
remove features that nobody uses because otherwise they tend to become
barnacles over time.

(Yes, technically smbfs actually _did_ use it, but that particular use was
more of a "mis-use" than anything else - because it didn't handle the
aliasing problem. Your suggested use is exactly _because_ of the aliasing
problem, not in spite of it).

Let nobody call me inflexible: I was probably wrong, and if you want to
re-instate i_dentry you should go right ahead, Bill..

Linus