Re: Weird UBIFS error to cause connmand crashed in UBIFS volume root

[Daniel Wagner <wagi@xxxxxxxxx>]

Hi,

On Sun, Feb 16, 2020 at 08:24:04PM +1100, JH wrote:
> I have been playing kernel mont to UBIFS rootfs volume for several
> days, the kernel can now mount to UBIFS volume root, the system
> services were running fine except the connection manager connmand
> service failed, I restarted it again and again, the connman was always
> crashed and ubifs_read_node: bad node type. Despite the UBIFS errors,
> the system could be still running, I could type shell comands in
> console when connman was stopped.
> 
> I am not clear if UBIFS errors caused connman failure or connman
> caused UBIFS error, and why it always failed in connman, was it
> coincident or was some special in connman man link to other things
> such as udev to cause the issues?

Have you enabled the stats file feature in ConnMan? It's enabled on
default.  Try again with '--disable-stats' and check if it makes a
difference.

Background: ConnMan save networking stats into a ring buffer file. The
file write operations are done via a mmap'ed memory. To avoid data
loss a double buffering approach with an 'atomic' swap is used:

        stats_file_unmap(history_file);
        stats_file_unmap(temp_file);

        close(temp_file->fd);

        unlink(history_file->name);

        err = link(temp_file->name, history_file->name);

        unlink(temp_file->name);

        close(history_file->fd);

        stats_file_cleanup(history_file);
        stats_file_cleanup(temp_file);

IIRC, f2fs is known not to work. Maybe there is also some limits on
ubifs as well.

Thanks,
Daniel

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