[UBI] UFFS : Unified Flash File System

Corentin Chary corentin.chary at gmail.com
Tue Nov 25 08:49:23 EST 2008


On Tue, Nov 25, 2008 at 2:14 PM, Sidney Amani <seed95 at gmail.com> wrote:
> ---------- Forwarded message ----------
> From: Liu Hui <onlyflyer at gmail.com>
> Date: Tue, Nov 25, 2008 at 2:30 AM
> Subject: Re: [UBI] UFFS : Unified Flash File System
> To: Sidney Amani <seed95 at gmail.com>
> Cc: linux-mtd at lists.infradead.org
>

Hope I'm not messing with the mailing list as I reply to a forwarded message.

> This is interesting. I have some questions about UFFS:
> 1)UFFS can work both with bare flash(MTD or UBI) and FTL(as a block
> device), how do you organize the metadata of UFFS(log-structured?).

I think it'll be log-structured, but we haven't yet finalized the
"on-flash" format.
We will work on that for the next few month.
Public website/mailing-list/svn will be created soon.

> 2)As we know, bare flash and FTL provide very different interfaces for
> FS. For MTD, FS should do garbage collection, wear leveling, bad block
> management by itself; For UBI, FS should also do garbage collection
> and wear leveling; and for FTL, FS just treat flash as block device.
> What will UFFS do to treat all of kind of interfaces? Will UFFS
> provide different strategy for different interfaces?

On linux, we will always use UBI for bare flash.
On WinCE or other OS where we can find MTD-like drivers (or just a
stupid simple flash driver) we will implement a UBI-like layer.
On FTL, we will try to implement a layer wich will be a kind of "UBI
for flash device with FTL": no garbage collection, but wear leveling
(for dumb ftl), REQ_FORGET, etc ...



More information about the linux-mtd mailing list