Re: [RFC PATCH v3 0/9] Suppress negative dentry

[Miklos Szeredi <miklos@szeredi.hu>]

On Fri, May 15, 2020 at 10:26 AM Chengguang Xu <cgxu519@mykernel.net> wrote:
>
>  ---- 在 星期五, 2020-05-15 15:30:27 Amir Goldstein <amir73il@gmail.com> 撰写 ----
>  > On Fri, May 15, 2020 at 10:21 AM Chengguang Xu <cgxu519@mykernel.net> wrote:
>  > >
>  > > This series adds a new lookup flag LOOKUP_DONTCACHE_NEGATIVE
>  > > to indicate to drop negative dentry in slow path of lookup.
>  > >
>  > > In overlayfs, negative dentries in upper/lower layers are useless
>  > > after construction of overlayfs' own dentry, so in order to
>  > > effectively reclaim those dentries, specify LOOKUP_DONTCACHE_NEGATIVE
>  > > flag when doing lookup in upper/lower layers.
>  > >
>  > > Patch 1 adds flag LOOKUP_DONTCACHE_NEGATIVE and related logic in vfs layer.
>  > > Patch 2 does lookup optimazation for overlayfs.
>  > > Patch 3-9 just adjusts function argument when calling
>  > > lookup_positive_unlocked() and lookup_one_len_unlocked().
>  >
>  > Hmm you cannot do that, build must not be broken mid series.
>  > When Miklos said split he meant to patch 1 and 2.
>  > Patch 1 must convert all callers to the new argument list,
>  > at which point all overlayfs calls are with 0 flags.
>  >
>  > Once that's done, you may add:
>  > Reviewed-by: Amir Goldstein <amir73il@gmail.com>
>  >
>
> OK, I got it, I'll still wait for a while in case of other feedbacks.
>
> Miklos, AI
>
> I'm not sure this series will go into whose tree in the end,
> so I just rebased on current linus-tree, any suggestion for the code base?

Linus' tree is a good base in this case.  I'm happier if VFS changes
go through Al's tree, but simple stuff can go through overlayfs tree
as well.

Thanks,
Miklos