ext2 for read-only file system on UBI

ext2 for read-only file system on UBI

[Hamish Moffatt]

My embedded device has a read-only root file system which is only
replaced by writing a whole image (dd, flashcp, ubiupdatevol etc).

Using gluebi and mtdblock, I think I can put a traditional block file
system (eg ext2) on top of NAND flash. What are the disadvantages of
this? (For the read-only application only.)

Background: I've got older hardware which uses ext2 on top of compact
flash in IDE mode, and new hardware which has replaced the compact flash
with NAND. I'd like to share an ext2 image between the two if possible.

The read-write file systems use ubifs. I'm only considering this for the
read-only volumes. One obvious disadvantage is lack of compression. Are
there others? Do I still get the reliability of ubi? 

ie what does "UBI is not an FTL" mean in practice?

thanks,
Hamish
-- 
Hamish Moffatt VK3SB <hamish@debian.org> <hamish@cloud.net.au>

Re: ext2 for read-only file system on UBI

[Artem Bityutskiy]

Hello Hamish,

On Mon, 2008-05-19 at 16:21 +1000, Hamish Moffatt wrote:
> My embedded device has a read-only root file system which is only
> replaced by writing a whole image (dd, flashcp, ubiupdatevol etc).
> 
> Using gluebi and mtdblock, I think I can put a traditional block file
> system (eg ext2) on top of NAND flash. What are the disadvantages of
> this? (For the read-only application only.)

For read-only it should be ok, although I am not sure mtdblock will like
non power of 2 eraseblock sizes. But this should be trivial to fix.

> Background: I've got older hardware which uses ext2 on top of compact
> flash in IDE mode, and new hardware which has replaced the compact flash
> with NAND. I'd like to share an ext2 image between the two if possible.

Should be possible for R/O. Different sizes of eraseblocks may add extra
work though.

> The read-write file systems use ubifs. I'm only considering this for the
> read-only volumes. One obvious disadvantage is lack of compression. Are
> there others? Do I still get the reliability of ubi? 

No, should be fine. Well, you'll still have WL across whole NAND chip,
yes. You'll still have bit-flip handling.

> ie what does "UBI is not an FTL" mean in practice?

In practice it means that /dev/ubiX_Y are not block devices, where you
may directly put ext2 and the like.

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: ext2 for read-only file system on UBI

[Hamish Moffatt]

Hi Artem, thanks for your response.

On Mon, May 19, 2008 at 09:37:53AM +0300, Artem Bityutskiy wrote:
> On Mon, 2008-05-19 at 16:21 +1000, Hamish Moffatt wrote:
> > My embedded device has a read-only root file system which is only
> > replaced by writing a whole image (dd, flashcp, ubiupdatevol etc).
> > 
> > Using gluebi and mtdblock, I think I can put a traditional block file
> > system (eg ext2) on top of NAND flash. What are the disadvantages of
> > this? (For the read-only application only.)
> 
> For read-only it should be ok, although I am not sure mtdblock will like
> non power of 2 eraseblock sizes. But this should be trivial to fix.
> 
> > Background: I've got older hardware which uses ext2 on top of compact
> > flash in IDE mode, and new hardware which has replaced the compact flash
> > with NAND. I'd like to share an ext2 image between the two if possible.
> 
> Should be possible for R/O. Different sizes of eraseblocks may add extra
> work though.
> 
> > The read-write file systems use ubifs. I'm only considering this for the
> > read-only volumes. One obvious disadvantage is lack of compression. Are
> > there others? Do I still get the reliability of ubi? 
> 
> No, should be fine. Well, you'll still have WL across whole NAND chip,
> yes. You'll still have bit-flip handling.

So, is there a benefit to Nancy's proposed ubi block layer as opposed to
gluebi + mtdblock?
( http://lists.infradead.org/pipermail/linux-mtd/2008-May/021609.html )

regards
Hamish
-- 
Hamish Moffatt VK3SB <hamish@debian.org> <hamish@cloud.net.au>

Re: ext2 for read-only file system on UBI

[Artem Bityutskiy]

On Mon, 2008-05-19 at 16:56 +1000, Hamish Moffatt wrote:
> > No, should be fine. Well, you'll still have WL across whole NAND chip,
> > yes. You'll still have bit-flip handling.
> 
> So, is there a benefit to Nancy's proposed ubi block layer as opposed to
> gluebi + mtdblock?
> ( http://lists.infradead.org/pipermail/linux-mtd/2008-May/021609.html )

Yes, her layer is assumed to be R/W block device.

You could want to glance here as well:

http://www.linux-mtd.infradead.org/faq/general.html#L_ext2_mtd
http://www.linux-mtd.infradead.org/faq/ubi.html#L_ext2_over_ubi

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: ext2 for read-only file system on UBI

[Hamish Moffatt]

On Mon, May 19, 2008 at 10:08:24AM +0300, Artem Bityutskiy wrote:
> On Mon, 2008-05-19 at 16:56 +1000, Hamish Moffatt wrote:
> > > No, should be fine. Well, you'll still have WL across whole NAND chip,
> > > yes. You'll still have bit-flip handling.
> > 
> > So, is there a benefit to Nancy's proposed ubi block layer as opposed to
> > gluebi + mtdblock?
> > ( http://lists.infradead.org/pipermail/linux-mtd/2008-May/021609.html )
> 
> Yes, her layer is assumed to be R/W block device.
> 
> You could want to glance here as well:
> 
> http://www.linux-mtd.infradead.org/faq/general.html#L_ext2_mtd
> http://www.linux-mtd.infradead.org/faq/ubi.html#L_ext2_over_ubi

Thanks for the links. So, read-write with mtdblock is dangerous
(regardless of UBI) because it must erase and rewrite a whole block,
right? Proposed ubiblk (linked from above) would use UBI ops to make the
update operation safe.

Hamish
-- 
Hamish Moffatt VK3SB <hamish@debian.org> <hamish@cloud.net.au>

Re: ext2 for read-only file system on UBI

[Artem Bityutskiy]

On Mon, 2008-05-19 at 17:44 +1000, Hamish Moffatt wrote:
> > Yes, her layer is assumed to be R/W block device.
> > 
> > You could want to glance here as well:
> > 
> > http://www.linux-mtd.infradead.org/faq/general.html#L_ext2_mtd
> > http://www.linux-mtd.infradead.org/faq/ubi.html#L_ext2_over_ubi
> 
> Thanks for the links. So, read-write with mtdblock is dangerous
> (regardless of UBI) because it must erase and rewrite a whole block,
> right?

Yes, and you loose plenty of data in case of power-cuts.

>  Proposed ubiblk (linked from above) would use UBI ops to make the
> update operation safe.

I have not looked that close at Nancy's patches because I did not like
her unreasonable copy-paste stuff. But I assume she would do something
like I proposed here:

http://lists.infradead.org/pipermail/linux-mtd/2008-January/020381.html

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)

Re: ext2 for read-only file system on UBI

[Artem Bityutskiy]

On Mon, 2008-05-19 at 10:54 +0300, Artem Bityutskiy wrote:
> > Thanks for the links. So, read-write with mtdblock is dangerous
> > (regardless of UBI) because it must erase and rewrite a whole block,
> > right?
> 
> Yes, and you loose plenty of data in case of power-cuts.
> 
> >  Proposed ubiblk (linked from above) would use UBI ops to make the
> > update operation safe.
> 
> I have not looked that close at Nancy's patches because I did not like
> her unreasonable copy-paste stuff. But I assume she would do something
> like I proposed here:
> 
> http://lists.infradead.org/pipermail/linux-mtd/2008-January/020381.html

Basically the idea is to implement something as straightforward as
mtdblock, but using the "atomic LEB change" [1] operation and do not
loose data in case of power-cuts.

1. http://www.linux-mtd.infradead.org/doc/ubi.html#L_lebchange

-- 
Best regards,
Artem Bityutskiy (Битюцкий Артём)