determining if a block erasure failed

determining if a block erasure failed

[Joel Reardon]

I want the caller of erase_worker() to be aware if the erase block became
bad. Here is the relevant code at the end of erase_worker():

/* at this time, the erase has failed. err == -EIO */
        ubi_msg("mark PEB %d as bad", pnum);
        err = ubi_io_mark_bad(ubi, pnum);
        if (err)
                goto out_ro;

/* does other stuff, not touching err */

/* err == 0 here, because of above goto */
        return err;

out_ro:
        ubi_ro_mode(ubi);
        return err;


So, if this function should return zero when the erase block fails, then I
will add a "result" field to ubi_work that will store the result of the
work function. But I figured I'd check first if the return err that always
returns 0 is not an oversight and this function should return -EIO if it
fails.

cheers,
Joel Reardon

Re: determining if a block erasure failed

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 988 bytes --]

On Tue, 2012-05-15 at 09:53 +0200, Joel Reardon wrote:
> I want the caller of erase_worker() to be aware if the erase block became
> bad. Here is the relevant code at the end of erase_worker():
> 
> /* at this time, the erase has failed. err == -EIO */
>         ubi_msg("mark PEB %d as bad", pnum);
>         err = ubi_io_mark_bad(ubi, pnum);
>         if (err)
>                 goto out_ro;
> 
> /* does other stuff, not touching err */
> 
> /* err == 0 here, because of above goto */
>         return err;
> 
> out_ro:
>         ubi_ro_mode(ubi);
>         return err;
> 
> 
> So, if this function should return zero when the erase block fails, then I
> will add a "result" field to ubi_work that will store the result of the
> work function. But I figured I'd check first if the return err that always
> returns 0 is not an oversight and this function should return -EIO if it
> fails.

Why do you need the results?

-- 
Best Regards,
Artem Bityutskiy

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: determining if a block erasure failed

[Joel Reardon]

Ah yes, if wl_wrk is freed in schedule_erase, theres no way to pass it
back that way.

The reason is that, if erasing a block of keys fails, then those keys may
remain on the flash device for eternity. While a rare occurance, it may be
nice to have ubifs perform GC on all main LEBs storing the data nodes that
can no longer be deleted through key erasure / reencrypt the valid data
nodes.

Cheers,
Joel Reardon
On Tue, 15 May 2012, Artem Bityutskiy wrote:

> On Tue, 2012-05-15 at 09:53 +0200, Joel Reardon wrote:
> > I want the caller of erase_worker() to be aware if the erase block became
> > bad. Here is the relevant code at the end of erase_worker():
> >
> > /* at this time, the erase has failed. err == -EIO */
> >         ubi_msg("mark PEB %d as bad", pnum);
> >         err = ubi_io_mark_bad(ubi, pnum);
> >         if (err)
> >                 goto out_ro;
> >
> > /* does other stuff, not touching err */
> >
> > /* err == 0 here, because of above goto */
> >         return err;
> >
> > out_ro:
> >         ubi_ro_mode(ubi);
> >         return err;
> >
> >
> > So, if this function should return zero when the erase block fails, then I
> > will add a "result" field to ubi_work that will store the result of the
> > work function. But I figured I'd check first if the return err that always
> > returns 0 is not an oversight and this function should return -EIO if it
> > fails.
>
> Why do you need the results?
>
> --
> Best Regards,
> Artem Bityutskiy
>

Re: determining if a block erasure failed

[Artem Bityutskiy]

[-- Attachment #1: Type: text/plain, Size: 616 bytes --]

On Tue, 2012-05-15 at 13:56 +0200, Joel Reardon wrote:
> Ah yes, if wl_wrk is freed in schedule_erase, theres no way to pass it
> back that way.
> 
> The reason is that, if erasing a block of keys fails, then those keys may
> remain on the flash device for eternity. While a rare occurance, it may be
> nice to have ubifs perform GC on all main LEBs storing the data nodes that
> can no longer be deleted through key erasure / reencrypt the valid data
> nodes.

Erasure is asynchronous and I do not see an good way to implement your
idea. Do you have a good plan?

-- 
Best Regards,
Artem Bityutskiy

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]