jffs2_get_inode_nodes(): Data CRC failed on NAND device

jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Frédéric Janot]

Hi,

I have a board with samsung K9F5608UOC nand flash and mpc5200 ppc processor.
I run a 2.4.25 kernel (yes I know it's very old) on this board with mtd 
from 20041007 snapshot.

I burn a jffs2 image on the nand with "nandwrite -j" (from nfs). All is 
ok and the filesystem is good but
when I reboot the board or when I remove the power "violently", I get 
somes messages like this :
/Empty flash at 0x009826d8 ends at 0x00982800
jffs2_get_inode_nodes(): Data CRC failed on node at 0x00cc0f90: Read 
0x6e17981a,  calculated 0x4614a389
/And I get one more message each time I reboot the board. But the 
filesystem doesn't seem corrupted.

If I umount / before rebooting, I don't have one more error message at 
the next reboot.

Here is my kernel command line (passed from uboot). Do I forget some 
mount options ?
/Kernel command line: root=/dev/mtdblock3 rw 
ip=172.16.48.19:172.16.16.240:172.16 .254.254:255.255.0.0::eth0:off panic=1
/
Do you have any idea about where the problem could come from ?

Frederic.

PS: sorry for my poor english.

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Estelle HAMMACHE]

Frédéric Janot wrote:
> 
> Hi,
> 
> I have a board with samsung K9F5608UOC nand flash and mpc5200 ppc processor.

JFFS2 on NAND flash uses a page buffer. It is flushed to the flash when
it is full, or when you do fsync, sync or umount.

> when I reboot the board or when I remove the power "violently"

Some data may remain in the page buffer and is never written
to the Flash.

> jffs2_get_inode_nodes(): Data CRC failed on node at 0x00cc0f90: Read
> 0x6e17981a,  calculated 0x4614a389

Well obviously a node was partly written to flash and the end of the
node remained in the cache and was never written. This means you
lost some data.

> /And I get one more message each time I reboot the board. But the
> filesystem doesn't seem corrupted.

It is a warning only. The remainder of the file system will probably
still work. Only the last operation(s) before reboot are lost.

> If I umount / before rebooting, I don't have one more error message at
> the next reboot.

This is correct procedure if you don't want to lose any data.

> Do you have any idea about where the problem could come from ?

There is no problem. This is standard JFFS2 processing. Nodes with
wrong CRCs (partly written for instance) are ignored and not taken 
in account when building the FS tree and files.

Estelle

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Frédéric Janot]

Estelle,

Thanks for this very fast answer !!!

Estelle HAMMACHE wrote:

>JFFS2 on NAND flash uses a page buffer. It is flushed to the flash when
>it is full, or when you do fsync, sync or umount.
>  
>
Does that mean that I have to call fsync each time after fclose a file ?
Even if I do not edit some files, I get this kind of error. Does that 
mean something (jffs2 itself or other) write some data to the flash all 
the time ?

>Some data may remain in the page buffer and is never written
>to the Flash.
>  
>
So I really loose some data :(

>>If I umount / before rebooting, I don't have one more error message at
>>the next reboot.
>>    
>>
>
>This is correct procedure if you don't want to lose any data.
>  
>
It is an embedded board without power supervisor. I cannot detect when 
the customer power down the board :(

Frederic

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Thomas Gleixner]

On Mon, 2005-02-14 at 18:58 +0100, Frédéric Janot wrote:
> >JFFS2 on NAND flash uses a page buffer. It is flushed to the flash when
> >it is full, or when you do fsync, sync or umount.
> >
> Does that mean that I have to call fsync each time after fclose a file ?
> Even if I do not edit some files, I get this kind of error. Does that 
> mean something (jffs2 itself or other) write some data to the flash all 
> the time ?

No, jffs2 does not write data just for fun. There is some application,
script, daemon writing/modifying file(s).

tglx

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Josh Boyer]

On Mon, 2005-02-14 at 18:58 +0100, Frédéric Janot wrote:
> >
> Does that mean that I have to call fsync each time after fclose a file ?
> Even if I do not edit some files, I get this kind of error. Does that 
> mean something (jffs2 itself or other) write some data to the flash all 
> the time ?

Sorta.  See below.

> 
> >Some data may remain in the page buffer and is never written
> >to the Flash.
> >  
> >
> So I really loose some data :(

Maybe not.  It could be a GC node that was cut short.  In that case, you
shouldn't loose any data since the original node should still be in
place.

> It is an embedded board without power supervisor. I cannot detect when 
> the customer power down the board :(

If you are worried about data loss, then fsync/sync is probably what you
should do.

josh

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[David Woodhouse]

On Mon, 2005-02-14 at 19:04 +0100, Thomas Gleixner wrote:
> On Mon, 2005-02-14 at 18:58 +0100, Frédéric Janot wrote:
> > >JFFS2 on NAND flash uses a page buffer. It is flushed to the flash when
> > >it is full, or when you do fsync, sync or umount.
> > >
> > Does that mean that I have to call fsync each time after fclose a file ?

Yes. It's precisely the same as with other file systems. If you write to
a file, close it and then lose power immediately, your data may be lost.
It's not guaranteed to be on the medium until you sync.

> > Even if I do not edit some files, I get this kind of error. Does that 
> > mean something (jffs2 itself or other) write some data to the flash all 
> > the time ?
> 
> No, jffs2 does not write data just for fun. There is some application,
> script, daemon writing/modifying file(s).

Or garbage collection, but in that case no data are lost. It was just
copying from one part of the flash to another, and the old copy will
still be intact.

-- 
dwmw2

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Martin Egholm Nielsen]

Hi,

>>>>JFFS2 on NAND flash uses a page buffer. It is flushed to the flash when
>>>>it is full, or when you do fsync, sync or umount.
>>>Does that mean that I have to call fsync each time after fclose a file ?
> Yes. It's precisely the same as with other file systems. If you write to
> a file, close it and then lose power immediately, your data may be lost.
> It's not guaranteed to be on the medium until you sync.
Well, as someone told me the other day on the IRC-channel, the 
writebuffer is flushed automatically if the buffer has not been accessed 
for X (2?) seconds (controlled by the VFS, right?)...

Correct me if I'm wrong...

// Martin

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Josh Boyer]

On Tue, 2005-02-15 at 20:22 +0100, Martin Egholm Nielsen wrote:
> Well, as someone told me the other day on the IRC-channel, the 
> writebuffer is flushed automatically if the buffer has not been accessed 
> for X (2?) seconds (controlled by the VFS, right?)...
> 
> Correct me if I'm wrong...

Not quite.  The write buffer is flushed every 5 seconds, unless you tune
it using the /proc entry.  It's controlled by kupdated, which syncs the
supers, which flushes the write buffer in JFFS2.

If the write buffer contains GC data, that isn't flushed at all.

josh

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Josh Boyer]

On Tue, 2005-02-15 at 13:46 -0600, Josh Boyer wrote:
> On Tue, 2005-02-15 at 20:22 +0100, Martin Egholm Nielsen wrote:
> > Well, as someone told me the other day on the IRC-channel, the 
> > writebuffer is flushed automatically if the buffer has not been accessed 
> > for X (2?) seconds (controlled by the VFS, right?)...
> > 
> > Correct me if I'm wrong...
> 
> Not quite.  The write buffer is flushed every 5 seconds, unless you tune
> it using the /proc entry.  It's controlled by kupdated, which syncs the
> supers, which flushes the write buffer in JFFS2.
> 
> If the write buffer contains GC data, that isn't flushed at all.

Just to clarify that, if the buffer contains _only_ GC data it isn't
flushed.  If it contains GC + inode data, all of that data is.

josh

Re: jffs2_get_inode_nodes(): Data CRC failed on NAND device

[Martin Egholm Nielsen]

>>>Well, as someone told me the other day on the IRC-channel, the 
>>>writebuffer is flushed automatically if the buffer has not been accessed 
>>>for X (2?) seconds (controlled by the VFS, right?)...
>>>Correct me if I'm wrong...
>>Not quite.  The write buffer is flushed every 5 seconds, unless you tune
>>it using the /proc entry.  It's controlled by kupdated, which syncs the
>>supers, which flushes the write buffer in JFFS2.
But this might be the better solution for Frederic, instead of poluting 
his code with fsync's. That is, just tune the /proc entry (which one?) 
to a value suitable for your needs - e.g. 2 secs.

>>If the write buffer contains GC data, that isn't flushed at all.
> Just to clarify that, if the buffer contains _only_ GC data it isn't
> flushed.  If it contains GC + inode data, all of that data is.
:o)

// Martin