Re: ubifs_check_node: bad CRC

[Richard Weinberger <richard.weinberger@xxxxxxxxx>]

Mickael,

On Mon, Jul 29, 2019 at 3:28 PM Mickael Chazaux
<mickael.chazaux@xxxxxxxxxxxxxxx> wrote:
>
> Hello,
>
> I have been using UBIFS on embedded devices for several years now.
> We never had a corrupted filesystem. Recently, I used it on a
> board design based on the previous ones. On this new board I
> had several filesystem corruptions like the one below.
>
> I checked the CRC of the corrupted UBIFS node, and it is incorrect.
> However, the CRC of the UBI LEB and the ECC of the NAND page _are_
> correct.
>
> This node is on a PEB that was copied (copy_flag = 1 in the header)

Side note, you are only allowed to check the CRC of a LEB iff copy_flag is set
and there is a second LEB with the same volume id and LEB number present.
Then you can use the CRC of the LEB with the higher sqnum

Otherwise the CRC can be outdated.

> I have been working on this a lot now, and I would like to hear the
> advice of the experts on how to find the cause of this corruption.
>
> A bit of background :
>         The volume "userdata" is used for logging.
>         Approx. 1 message per second is logged.
>         It is mounted with option sync.
>         Its size is 10MiB, the ubi0 device itself is 491MiB.
>         There are other volumes on ubi0, that are not
>         corrupted (rootfs, etc.) . They are less busy.
>
> I imaged the NAND and can reproduce the error like this :
>
> flash_eraseall /dev/mtd8
> nandwrite -no /dev/mtd8 nand-image.img
> ubiattach /dev/ubi_ctrl -m 8
> mount -t ubifs ubi0:userdata /mnt/
> echo "azer" > /mnt/azer

That easy?
Hmm, sounds a bit like a bug in your NAND driver. Did you test it?

Please also disable compression in UBIFS, then the read back data maybe gives
you a hint what went wrong.

Thanks,
//richard

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/