[LSF/MM TOPIC] Support for case-insensitive lookups

[LSF/MM TOPIC] Support for case-insensitive lookups

[Gabriel Krisman Bertazi]

Case-insensitive lookups for filesystems is a recurrent topic for
filesystems, which got some attention in LSF/MM last year, during the
discussion to get free of the Android wrapfs hackery.  This year, I have
been working on an approach to enable bind mount points supporting
casefolding, whose goal is to allow side-by-side lookups of
case-sensitive and case-insensitive names.  This effort has a direct
benefit to the Android filesystem, which requires some racy code to do
this exactly feature, as well as other use cases for SteamOS.

I published an early version with full UTF-8 (based on the SGI
patches) support for performing case-folding in the EXT4 filesystem, as
a RFC to the fsdev mailing list.  The latest thread is here:

https://www.spinics.net/lists/linux-ext4/msg59577.html

Right now, I have focused on adapting the VFS to support a MS_CASEFOLD
VFS flag, which would allow bind mounts, but I am having concerns about
how to best explore the design of the dentry cache to handle multiple
entries only differing by case.  I've been experimenting with multiple
approaches recently and getting live feedback from the community will be
very helpful.

In addition, other semantics for case-insensitive lookups could be
discussed, like how to deal with invalid utf8 sequences, alternatives to
the VFS approach, should it prove itself unreliable, and how to best
adapt the dentry hashes to explore name sequences that only differ by
case and normalization.

-- 
Gabriel Krisman Bertazi

Re: [LSF/MM TOPIC] Support for case-insensitive lookups

[Steve French]

This is of interest to me (could help Samba performance a lot if file
system optionally supports this), and other filesystems
(cifs.ko) could trivially advertise when the underlying mount (the
target volume) is case insensitive.

On Wed, Jan 31, 2018 at 8:05 PM, Gabriel Krisman Bertazi
<krisman@collabora.co.uk> wrote:
>
> Case-insensitive lookups for filesystems is a recurrent topic for
> filesystems, which got some attention in LSF/MM last year, during the
> discussion to get free of the Android wrapfs hackery.  This year, I have
> been working on an approach to enable bind mount points supporting
> casefolding, whose goal is to allow side-by-side lookups of
> case-sensitive and case-insensitive names.  This effort has a direct
> benefit to the Android filesystem, which requires some racy code to do
> this exactly feature, as well as other use cases for SteamOS.
>
> I published an early version with full UTF-8 (based on the SGI
> patches) support for performing case-folding in the EXT4 filesystem, as
> a RFC to the fsdev mailing list.  The latest thread is here:
>
> https://www.spinics.net/lists/linux-ext4/msg59577.html
>
> Right now, I have focused on adapting the VFS to support a MS_CASEFOLD
> VFS flag, which would allow bind mounts, but I am having concerns about
> how to best explore the design of the dentry cache to handle multiple
> entries only differing by case.  I've been experimenting with multiple
> approaches recently and getting live feedback from the community will be
> very helpful.
>
> In addition, other semantics for case-insensitive lookups could be
> discussed, like how to deal with invalid utf8 sequences, alternatives to
> the VFS approach, should it prove itself unreliable, and how to best
> adapt the dentry hashes to explore name sequences that only differ by
> case and normalization.
>
> --
> Gabriel Krisman Bertazi



-- 
Thanks,

Steve