UBIFS recovery fails in barebox while Linux suceeds

[Sascha Hauer]

On Thu, Mar 03, 2016 at 09:38:30AM +0100, Alexander Stein wrote:
> On Thursday 03 March 2016 09:28:56, Sascha Hauer wrote:
> > On Thu, Mar 03, 2016 at 08:16:13AM +0100, Alexander Stein wrote:
> > > On Wednesday 02 March 2016 20:56:13, Sascha Hauer wrote:
> > > > On Wed, Mar 02, 2016 at 05:19:08PM +0100, Alexander Stein wrote:
> > > > > Hi,
> > > > > 
> > > > > in case a UBIFS needs recovery (unclean write or whatever on NOR
> > > > > flash) it is possible that barebox fails to do so while Linux suceeds.
> > > > > The main cause, IMHO, is that Linux takes max_write_size into account
> > > > > (starting with commit 2765df7da540687c4d57ca840182122f074c5b9c "UBIFS:
> > > > > use max_write_size during recovery") while barebox doesn't. Apparently
> > > > > is_last_write (fs/ubifs/recovery.c) results differently due to that
> > > > > fact which explains why recovery progress differently. I don't know
> > > > > which linux version the ubifs code in barebox is taken from but I
> > > > > guess this needs to be updated. Are there any plans?
> > > > 
> > > > The barebox UBIFS code is taken from U-Boot 2013.07 which is taken from
> > > > Linux-2.6.29-rc6, so indeed the code is quite old. U-Boot has updated
> > > > UBIFS support to Linux-4.2. The question is if we update UBIFS from
> > > > U-Boot or directly from the Kernel, I have no idea which way is easier.
> > > 
> > > I still wonder if this problem should have been avoided in the first
> > > place. It seems like the change in the kernel is like chaning the
> > > on-disk-format.
> > 
> > BTW what happens after the failed recovery? Is the filesystem not
> > readable?
> 
> It is not even mounted, neither in barebox in current state, nor e.g. using mtdram in Linux with a different writebuf size.
> Recovery fails so does mounting ubifs. Once recovery suceeded everybody is happy again.
> BTW: mtdram in barebox would suffer the same problem, writebufsize is fixed 64.
> 
> > > 
> > > > Currently there are no plans to update UBIFS, but of course you are
> > > > invited to create them ;)
> > > 
> > > I expected an answer like that ;-) I did a quick compare and there are
> > > a lot of changes in barebox upon the code taken from u-boot. AFAICS
> > > those are not documented :(
> > 
> > I wouldn't say that. I just checked out barebox 551b412 (the initial
> > ubifs commit) and U-Boot 2013.07 and copied over the ubifs files from
> > U-Boot to barebox. The diff doesn't look too bad, it nearly only shows
> > the changes necessary to adopt to the barebox filesystem layer.
> > I assume fs/ubifs/ubifs.c can be left nearly unmodified. The rest is
> > copying over the kernel files and boxing them through the compiler
> > (which indeed might be a significant amount of work given the size of
> > the code)
> 
> Well, linux and u-boot, to some degree at least, IIRC gained support for fastmap, xattr and even optional atime support. Maybe it won't that easy anymore.

Fastmap is a UBI feature, not a UBIFS feature. barebox supports that
aswell.

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |