Is scrubbing md-aware in any way?

Is scrubbing md-aware in any way?

[Mariusz Mazur]

Hi, it's my understanding that when running a scrub on a btrfs raid,
if data corruption is detected, the process will check copies on other
devices and heal the data if possible.

Is any of that functionality available when running on top of an md
raid? When a scrub notices an issue, does it have any mechanism of
telling md "hey, there's a problem with these sectors" and working
with it to do something about that or is it all up to the admin to
deal with the "file corrupted" message?

Re: Is scrubbing md-aware in any way?

[Lukas Straub]

[-- Attachment #1: Type: text/plain, Size: 620 bytes --]

On Tue, 13 Sep 2022 16:15:26 +0200
Mariusz Mazur <mariusz.g.mazur@gmail.com> wrote:

> Hi, it's my understanding that when running a scrub on a btrfs raid,
> if data corruption is detected, the process will check copies on other
> devices and heal the data if possible.
> 
> Is any of that functionality available when running on top of an md
> raid? When a scrub notices an issue, does it have any mechanism of
> telling md "hey, there's a problem with these sectors" and working
> with it to do something about that or is it all up to the admin to
> deal with the "file corrupted" message?

No.

-- 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: Is scrubbing md-aware in any way?

[David Sterba]

On Tue, Sep 13, 2022 at 04:15:26PM +0200, Mariusz Mazur wrote:
> Hi, it's my understanding that when running a scrub on a btrfs raid,
> if data corruption is detected, the process will check copies on other
> devices and heal the data if possible.
> 
> Is any of that functionality available when running on top of an md
> raid?

No, the type of block device under btrfs is considered the same in all
cases, except zoned devices, so any advanced information exchange or
control like devices reporting bad sectors or btrfs asking to repair the
underlying device by its own means.

> When a scrub notices an issue, does it have any mechanism of
> telling md "hey, there's a problem with these sectors" and working
> with it to do something about that or is it all up to the admin to
> deal with the "file corrupted" message?

It's up to the admin. I've looked if there's some API outside of the
md-raid implementation, but there's none so it would have to be created
first in for the btrfs <-> md cooperation.

Re: Is scrubbing md-aware in any way?

[Neal Gompa]

On Wed, Sep 14, 2022 at 10:37 AM David Sterba <dsterba@suse.cz> wrote:
>
> On Tue, Sep 13, 2022 at 04:15:26PM +0200, Mariusz Mazur wrote:
> > Hi, it's my understanding that when running a scrub on a btrfs raid,
> > if data corruption is detected, the process will check copies on other
> > devices and heal the data if possible.
> >
> > Is any of that functionality available when running on top of an md
> > raid?
>
> No, the type of block device under btrfs is considered the same in all
> cases, except zoned devices, so any advanced information exchange or
> control like devices reporting bad sectors or btrfs asking to repair the
> underlying device by its own means.
>
> > When a scrub notices an issue, does it have any mechanism of
> > telling md "hey, there's a problem with these sectors" and working
> > with it to do something about that or is it all up to the admin to
> > deal with the "file corrupted" message?
>
> It's up to the admin. I've looked if there's some API outside of the
> md-raid implementation, but there's none so it would have to be created
> first in for the btrfs <-> md cooperation.

IIRC, the Synology folks created such an API for their own use-case.
Does anyone on-list know the folks at Synology to see if they'd be
interested in working with us and the md folks to make this fully
supported upstream?


-- 
真実はいつも一つ！/ Always, there's only one truth!

Re: Is scrubbing md-aware in any way?

[Mariusz Mazur]

niedz., 25 wrz 2022 o 20:46 Neal Gompa <ngompa13@gmail.com> napisał(a):
>
> On Wed, Sep 14, 2022 at 10:37 AM David Sterba <dsterba@suse.cz> wrote:
> >
> > On Tue, Sep 13, 2022 at 04:15:26PM +0200, Mariusz Mazur wrote:
> > > Hi, it's my understanding that when running a scrub on a btrfs raid,
> > > if data corruption is detected, the process will check copies on other
> > > devices and heal the data if possible.
> > >
> > > Is any of that functionality available when running on top of an md
> > > raid?
> >
> > No, the type of block device under btrfs is considered the same in all
> > cases, except zoned devices, so any advanced information exchange or
> > control like devices reporting bad sectors or btrfs asking to repair the
> > underlying device by its own means.
> >
> > > When a scrub notices an issue, does it have any mechanism of
> > > telling md "hey, there's a problem with these sectors" and working
> > > with it to do something about that or is it all up to the admin to
> > > deal with the "file corrupted" message?
> >
> > It's up to the admin. I've looked if there's some API outside of the
> > md-raid implementation, but there's none so it would have to be created
> > first in for the btrfs <-> md cooperation.
>
> IIRC, the Synology folks created such an API for their own use-case.
> Does anyone on-list know the folks at Synology to see if they'd be
> interested in working with us and the md folks to make this fully
> supported upstream?
>
>
> --
> 真実はいつも一つ！/ Always, there's only one truth!

Yes, wondering how synology uses btrfs on top of md raid effectively
is what prompted my question. I'd like to be able to use btrfs (and
other filesystems) in such a fashion as well.

Re: Is scrubbing md-aware in any way?

[David Sterba]

On Mon, Sep 26, 2022 at 05:36:04PM +0200, Mariusz Mazur wrote:
> niedz., 25 wrz 2022 o 20:46 Neal Gompa <ngompa13@gmail.com> napisał(a):
> >
> > On Wed, Sep 14, 2022 at 10:37 AM David Sterba <dsterba@suse.cz> wrote:
> > >
> > > On Tue, Sep 13, 2022 at 04:15:26PM +0200, Mariusz Mazur wrote:
> > > > Hi, it's my understanding that when running a scrub on a btrfs raid,
> > > > if data corruption is detected, the process will check copies on other
> > > > devices and heal the data if possible.
> > > >
> > > > Is any of that functionality available when running on top of an md
> > > > raid?
> > >
> > > No, the type of block device under btrfs is considered the same in all
> > > cases, except zoned devices, so any advanced information exchange or
> > > control like devices reporting bad sectors or btrfs asking to repair the
> > > underlying device by its own means.
> > >
> > > > When a scrub notices an issue, does it have any mechanism of
> > > > telling md "hey, there's a problem with these sectors" and working
> > > > with it to do something about that or is it all up to the admin to
> > > > deal with the "file corrupted" message?
> > >
> > > It's up to the admin. I've looked if there's some API outside of the
> > > md-raid implementation, but there's none so it would have to be created
> > > first in for the btrfs <-> md cooperation.
> >
> > IIRC, the Synology folks created such an API for their own use-case.
> > Does anyone on-list know the folks at Synology to see if they'd be
> > interested in working with us and the md folks to make this fully
> > supported upstream?
> 
> Yes, wondering how synology uses btrfs on top of md raid effectively
> is what prompted my question. I'd like to be able to use btrfs (and
> other filesystems) in such a fashion as well.

IIRC the synology kernel sources are available as some .tar or .rpm
file, I don't know if there are the separate patches available anywhere.
Also it's up to synology ty upstream the patches. We've got some btrfs
patches from them that were good, but adding the MD API itself seems to
be an extensive change and harder to get accepted upstream. I haven't
seen the code so I can't say how much it would be intrusive to the btrfs
code but with enough cleanups and abstractions everything is possible.