[GSoC project] Implementation of ISO9660 filesystem

[Christophe Huriaux]

2011/4/4 Sebastien Bourdeauducq <sebastien at milkymist.org>:
> Hi,

Hi,


>
> On Mon, 2011-04-04 at 10:25 +0200, Christophe Huriaux wrote:
>> develop an ISO9660 filesystem implementation into RTEMS.
>
> Have you had a look at the internal RTEMS filesystem APIs yet?
>
> Maybe you'd like to start by improving them, instead of (painfully)
> writing new code for the current APIs which will increase the cost of
> switching to a more sane filesystem API in the future (which I hope will
> happen).
> In particular, the eval_path system should disappear (it means a lot of
> code duplication, pain and C string manipulation woes and bugs which
> could easily be avoided), and the filesystem lacks locking (e.g. opening
> a file then deleting it from another task breaks the FS and may crash
> the complete system).

  In fact while discussing about my project on IRC the eval_path and
filesystem locking problems in the actual filesystem API have been
pointed out, but apparently things to refactor are not easily
listable, that's why I proposed to refactor filesystem helpers while
developing this new filesystem.

  It would provide additional goals to my project (as a "side effect"
of my implementation), but I'm not currently able to determine how
much time I will need to spend on these tasks, since I'm not really
aware of what I'll need to do, that's why API improvement  is part of
the project without being really detailed in my proposal.


>
> A good example to draw ideas from is the FUSE API:
> http://www.ibm.com/developerworks/linux/library/l-fuse/
>
> Regards,
> Sebastien
>
> PS: I'm not participating in GSoC this year. I have simply ported YAFFS2
> to RTEMS and I had a bad experience (in fact I'm still having it, we
> recently found some more bugs which I suspect are eval_path related) due
> to the poorly designed API. Just sharing my thoughts - deciding whether
> to go with the current API or improving it isn't up to me.
>
>

   As far as i know Chris Johns had similiar problems when
implementing RFS and have also a fresh insight of what things should
be fixed. Both of your guidances would be useful for me.


Regards,
Christophe Huriaux