why UBI static volumes are flagged as DEVFS_IS_CHARACTER_DEV

[Teresa Remmet]

Hello Sascha,

Am Freitag, den 07.10.2016, 09:27 +0200 schrieb Sascha Hauer:
> On Thu, Oct 06, 2016 at 10:17:27AM +0200, Teresa Remmet wrote:
> > Hello Giorgio,
> > 
> > Am Mittwoch, den 05.10.2016, 15:23 +0200 schrieb iw3gtf at arcor.de:
> > > Hi,
> > > 
> > > I noticed that the commit id c087e0804f0290e9886899e8a3cccb07c4ce088b flagged static
> > > UBI volumes as DEVFS_IS_CHARACTER_DEV.
> > > 
> > > A consequence of this flag is that commands like:
> > > 
> > > # cp /dev/nand0.ubi_volumes.ubi.my_static_vol file
> > > 
> > > will not work because the cp command will see a src file (the static UBI volume) with a size
> > > of -1 (FILE_SIZE_STREAM) and keep on reading from the volume until a flood of
> > > "UBI assert failed in ubi_eba_read_leb at 359" asserts comes out of the console.
> > > 
> > > I tried to comment out the flag assignment, just to see what happen:
> > > 
> > > int ubi_volume_cdev_add(struct ubi_device *ubi, struct ubi_volume *vol)
> > > {
> > > ...
> > > 	cdev->size = vol->used_bytes;
> > > 
> > > //	if (vol->vol_type == UBI_STATIC_VOLUME)
> > > //		cdev->flags = DEVFS_IS_CHARACTER_DEV;
> > > 
> > > 	cdev->dev = &vol->dev;
> > > ...
> > > 
> > > and then the cp command worked than as expected.
> > > 
> > > Could someone shortly confirm that the DEVFS_IS_CHARACTER_DEV flag for static UBI volumes
> > > is really needed (to avoid some other problems that my superficial test does not triggers) ?
> > 
> > the size of a static ubi volume device is equal to the image size you
> > flashed. When you create a new static ubi volume the size is 0, as it is
> > empty.
> > We need the chardev flag to be able to update the static ubi volume or
> > barebox will complain that there is not enough space.
> 
> Then I think we need another solution for that. UBI volumes are clearly
> not character devices, so we shouldn't set this flag to work around
> other issues we have.

Yes, you are write using the character dev flag is more a work around.
Which now brings up other problems like dumping the data back from
flash.

> 
> Currently devfs_truncate is implemented like:
> 
> static int devfs_truncate(struct device_d *dev, FILE *f, ulong size)
> {
>         if (f->fsdev->dev.num_resources < 1)
>                 return -ENOSPC;
>         if (size > resource_size(&f->fsdev->dev.resource[0]))
>                 return -ENOSPC;
>         return 0;
> }
> 
> Maybe we need an optional cdev->ops->truncate() to let the UBI volumes be
> able to overwrite this behaviour?

Truncate sounds like a better way. But UBI itself is not a filesystem.
So I do not see a good place to overwrite it on a first look.

I will try to look into this deeper next week. 

Regards,
Teresa

> 
> Sascha
>