file corruption with mtd / jffs2

file corruption with mtd / jffs2

[alfred hitch]

Hi,

I would like some inputs, if someone can give some pointer / idea.

We are running jffs2 based partitions on linux 2.4 snapgear distro boards.
partitions were mounted read-write.

We are observing that after running for 3-5 months, now boards have
started giving corruped executables !
We are getting many boards back which have a corrupted binary on flash itself.

How is this possible ?

1) We by mistake were writting also in flash'es for logs /
configurations (dont ask why please ) .. and so is it be possible that
in log rotation etc jffs2 uses these got lost across a bad shutdown ?

Anyone ever observed anything like this ??
Is it safe to use jffs2 and assume all's well across unsafe shutdowns ?
Basic question perhaps, but will appreciate any directions.

Regards,
Nitin

Re: file corruption with mtd / jffs2

[Artem B. Bityutskiy]

alfred hitch wrote:
> I would like some inputs, if someone can give some pointer / idea.
> 
> We are running jffs2 based partitions on linux 2.4 snapgear distro boards.
> partitions were mounted read-write.
> 
> We are observing that after running for 3-5 months, now boards have
> started giving corruped executables !
> We are getting many boards back which have a corrupted binary on flash itself.
> 
Hmm...
Which JFFS2 snapshot and what kernel do you use?

-- 
Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

Re: file corruption with mtd / jffs2

[alfred hitch]

Hi,

we are using snapgear linux distro .. arm linux port kernel 2.4.25-x
jffs2 version I dont know where to look ?

Regards,
Alfred


On 12/4/05, Artem B. Bityutskiy <dedekind@yandex.ru> wrote:
> alfred hitch wrote:
> > I would like some inputs, if someone can give some pointer / idea.
> >
> > We are running jffs2 based partitions on linux 2.4 snapgear distro boards.
> > partitions were mounted read-write.
> >
> > We are observing that after running for 3-5 months, now boards have
> > started giving corruped executables !
> > We are getting many boards back which have a corrupted binary on flash itself.
> >
> Hmm...
> Which JFFS2 snapshot and what kernel do you use?
>
> --
> Best Regards,
> Artem B. Bityutskiy,
> St.-Petersburg, Russia.
>

Re: file corruption with mtd / jffs2

[Artem B. Bityutskiy]

alfred hitch wrote:
> Hi,
> 
> we are using snapgear linux distro .. arm linux port kernel 2.4.25-x
> jffs2 version I dont know where to look ?
Look to $Id$ versions in header of wbuf.c, nodelist.c and readinode.c 
please.

-- 
Best Regards,
Artem B. Bityutskiy,
St.-Petersburg, Russia.

Re: file corruption with mtd / jffs2

[alfred hitch]

Hi Vahid,

yea something similar.
My diff though on binaries doesnt necessarilly shows any specific
location where it is corupted (or which one also for that matter).
I can see some bytes (ranging from 20 byte diff I see or even upto 300 bytes).
But, we are beginning to see this more often in 2 cases:
1) If there is a crash (read seg. fault / kernel panic both  ) .. then
the process in execution last / near about are the victims. This is
happening more across resets also, we left 10 boards on weekend, doing
nothing but reset every 3 minutes and 1 of them went bad.

2) If we made system read only partition mounted, so far we havent
observed any on disk image corruption (obviosuly ?) .. but once I have
personally also faced a image in memory being bad, and I wasnt able to
do anything with that binary till I resetted board and it came / read
back from flash again.

ARTEM:
$Id: wbuf.c,v 1.1 2002/09/12 00:23:22 davidm Exp $


Some questions come to mind:
1)  Files in our case are executables, so how come the code sections,
which are read only I believe marked in page properties get corrupted
?
2) Even if I suspect so many 3rd party code I am running, could be
trashing in kernel space, the read only attribute shouldn't have
allowed it ?
3) Why is the file being written back to disk (executable, read only pages) ?

Regards,
Alfred

On 12/6/05, Vahid Fereydunkolahi <fereydunk@yahoo.com> wrote:
> I believe we have the same issue that you guys have
> regaring the corruption in binary files. The problem
> seems to be related to having some page(s) of all zero
> in the middle of the binary file. I believe the data
> is perfectly fine on the flash, but there is a memory
> corruption (or something similar) that causes the file
> to be costurced wrong in the memory. Most probaly when
> some section of the file is read it thinks that, the
> section is a frag hole, and returns all zero for that
> section.
>
> We are using the 2.4 kernel and I ported the JFFS2
> fron the 2.6 kernel.
>
> Regards,
> --vahid
>
>
>
> __________________________________________
> Yahoo! DSL – Something to write home about.
> Just $16.99/mo. or less.
> dsl.yahoo.com
>
>

file corruption with mtd / jffs2

[Vahid Fereydunkolahi]

In our case the data is perfectly fine on the flash,
but gets corruped in the memory.
I noticed that when I copy the binary files, and I
enable the debugging, sometimes I get the following
message:
"Filling non-frag hole from 7000-8192 ( inod 1323,
size 1192)"

Why should a complete copy of a file has a non-frag
hole?

--vahid

By the way, Is the data corruption in your case caused
on the NAND or in the memory?


		
__________________________________________ 
Yahoo! DSL – Something to write home about. 
Just $16.99/mo. or less. 
dsl.yahoo.com