[fuse-devel] [RFC] Case Insensitive on Purpose File System

[Marc A. T. <ma...@br...>]

Hi,

I have been working on a case insensitive fuse file system and it's now
at a point where it mostly works. As i am new to fuse it would be nice
if one of the more experienced list members could take a quick look and
give me some hints on what could be improved. The code is located in a
git repository and can be fetched from:

 http://repo.or.cz/w/ciopfs.git?a=snapshot

The fs works by translating every path element to lower case before
further operations take place. On file creation the original file name
is stored in an extended attribute, this value is later returned upon
request. In order for this to work you will need libattr[1] and an
underlying file system which supports extended attribtues (for example
for ext{2,3} you will need a kernel with CONFIG_EXT{2,3}_FS_XATTR
enabled). You probably also want to mount the underlying filesystem 
with the user_xattr option which allows non root users to create
extended attributes.

The fs can be compiled with support for unicode characters in file 
names. In this case the libraries from the icu project[2] are used to
perform case folding. (Are there any unicode experts here? Should
there be some kind of normalization in place or is the case folding 
enough?). If you don't have this library at hand, disable it in
config.mk. The fs will use standard C library function which are only
defined for [a-zA-Z]. 

Here is a little demonstration to let you see how it should work:

  mkdir .data case-insensitiv
  ./ciopfs .data case-insensitiv
  mkdir -p case-insensitive/DeMo/SubFolder
  touch case-insensitive/DemO/subFolder/MyFile

At this point the fs should look like this:

  case-insensitive
  `-- DeMo
      `-- SubFolder
          `-- MyFile
  .data
  `-- demo
      `-- subfolder
          `-- myfile

This all works as long as you don't start to mess around with the
data directory directly. File names which aren't all lowercase in the
data directory are ignored. So you should start with an empty data
directory and copy your content over via the mountpoint. All further
file creation or renames should also be done over the mountpoint.

I don't think there is a way to avoid this restriction, or am i missing
something?

Also the current implementation uses malloc/free quite extensively,
should this be converted to a stack based solution?

Any comments are appreciated.

Thanks,
Marc

[1] http://acl.bestbits.at/download.html#Attr
[2] http://www.icu-project.org/

-- 
 Marc Andre Tanner >< http://www.brain-dump.org/ >< GPG key: CF7D56C0