freeze_bdev and scrub/re-balance

freeze_bdev and scrub/re-balance

[Wang, Zhiye]

Hi all,


If I understand it correctly, defragment operation is done in user space tools, while scrub/re-balance is done in kernel thread.


So, if my kernel module calls freeze_bdev when scrub/re-balance is in progress, will I still be able to get a consistent file system state?




Thanks
Mike


     

RE: freeze_bdev and scrub/re-balance

[Wang, Zhiye]

Can anyone help me with this question?

Not only the kernel space API freeze_bdev, but also the IO control which can be called from user space.



-----Original Message-----
From: Wang, Zhiye 
Sent: Saturday, December 5, 2015 5:57 PM
To: linux-btrfs@vger.kernel.org
Subject: freeze_bdev and scrub/re-balance

Hi all,


If I understand it correctly, defragment operation is done in user space tools, while scrub/re-balance is done in kernel thread.


So, if my kernel module calls freeze_bdev when scrub/re-balance is in progress, will I still be able to get a consistent file system state?




Thanks
Mike


     

Re: freeze_bdev and scrub/re-balance

[Liu Bo]

On Sat, Dec 05, 2015 at 09:57:18AM +0000, Wang, Zhiye wrote:
> Hi all,
> 
> 
> If I understand it correctly, defragment operation is done in user space tools, while scrub/re-balance is done in kernel thread.

Defragment is done via a IOCTL, so it also works in the kernel.

> 
> 
> So, if my kernel module calls freeze_bdev when scrub/re-balance is in progress, will I still be able to get a consistent file system state?

It should be like that with COW enabled.

Thanks,

-liubo

RE: freeze_bdev and scrub/re-balance

[Wang, Zhiye]

Thank you liubo for your reply.

But I am very clear with your meaning of "It should be like that with COW enabled"

I'd like to confirm, if defragment/scrub/rebalance is in progress, and my code calls "freeze_bdev" (in kernel code, or in user space code via ioctl), I can get a consistent file system state. "consistent file system state" means, if I run a LVM snapshot (or hardware snapshot, or even "dd" if it can do that quickly) after call freeze_bdev, the snapshot is file system consistent.


Thanks
Mike


-----Original Message-----
From: Liu Bo [mailto:bo.li.liu@oracle.com] 
Sent: Thursday, December 10, 2015 1:22 AM
To: Wang, Zhiye
Cc: linux-btrfs@vger.kernel.org
Subject: Re: freeze_bdev and scrub/re-balance

On Sat, Dec 05, 2015 at 09:57:18AM +0000, Wang, Zhiye wrote:
> Hi all,
> 
> 
> If I understand it correctly, defragment operation is done in user space tools, while scrub/re-balance is done in kernel thread.

Defragment is done via a IOCTL, so it also works in the kernel.

> 
> 
> So, if my kernel module calls freeze_bdev when scrub/re-balance is in progress, will I still be able to get a consistent file system state?

It should be like that with COW enabled.

Thanks,

-liubo

Re: freeze_bdev and scrub/re-balance

[Liu Bo]

On Mon, Dec 14, 2015 at 09:27:10AM +0000, Wang, Zhiye wrote:
> Thank you liubo for your reply.
> 
> But I am very clear with your meaning of "It should be like that with COW enabled"
> 
> I'd like to confirm, if defragment/scrub/rebalance is in progress, and my code calls "freeze_bdev" (in kernel code, or in user space code via ioctl), I can get a consistent file system state. "consistent file system state" means, if I run a LVM snapshot (or hardware snapshot, or even "dd" if it can do that quickly) after call freeze_bdev, the snapshot is file system consistent.

I think so, the metadata is always written as a copy-on-write way, so
it's consistent.

Thanks,

-liubo
> 
> 
> Thanks
> Mike
> 
> 
> -----Original Message-----
> From: Liu Bo [mailto:bo.li.liu@oracle.com] 
> Sent: Thursday, December 10, 2015 1:22 AM
> To: Wang, Zhiye
> Cc: linux-btrfs@vger.kernel.org
> Subject: Re: freeze_bdev and scrub/re-balance
> 
> On Sat, Dec 05, 2015 at 09:57:18AM +0000, Wang, Zhiye wrote:
> > Hi all,
> > 
> > 
> > If I understand it correctly, defragment operation is done in user space tools, while scrub/re-balance is done in kernel thread.
> 
> Defragment is done via a IOCTL, so it also works in the kernel.
> 
> > 
> > 
> > So, if my kernel module calls freeze_bdev when scrub/re-balance is in progress, will I still be able to get a consistent file system state?
> 
> It should be like that with COW enabled.
> 
> Thanks,
> 
> -liubo

RE: freeze_bdev and scrub/re-balance

[Wang, Zhiye]

Thank you Bo.

After some testing, I also confirmed that freeze_bdev has no issue with defragment/scrub/rebalance.



-----Original Message-----
From: Liu Bo [mailto:bo.li.liu@oracle.com] 
Sent: Friday, January 8, 2016 6:15 AM
To: Wang, Zhiye
Cc: linux-btrfs@vger.kernel.org
Subject: Re: freeze_bdev and scrub/re-balance

On Mon, Dec 14, 2015 at 09:27:10AM +0000, Wang, Zhiye wrote:
> Thank you liubo for your reply.
> 
> But I am very clear with your meaning of "It should be like that with COW enabled"
> 
> I'd like to confirm, if defragment/scrub/rebalance is in progress, and my code calls "freeze_bdev" (in kernel code, or in user space code via ioctl), I can get a consistent file system state. "consistent file system state" means, if I run a LVM snapshot (or hardware snapshot, or even "dd" if it can do that quickly) after call freeze_bdev, the snapshot is file system consistent.

I think so, the metadata is always written as a copy-on-write way, so it's consistent.

Thanks,

-liubo
> 
> 
> Thanks
> Mike
> 
> 
> -----Original Message-----
> From: Liu Bo [mailto:bo.li.liu@oracle.com]
> Sent: Thursday, December 10, 2015 1:22 AM
> To: Wang, Zhiye
> Cc: linux-btrfs@vger.kernel.org
> Subject: Re: freeze_bdev and scrub/re-balance
> 
> On Sat, Dec 05, 2015 at 09:57:18AM +0000, Wang, Zhiye wrote:
> > Hi all,
> > 
> > 
> > If I understand it correctly, defragment operation is done in user space tools, while scrub/re-balance is done in kernel thread.
> 
> Defragment is done via a IOCTL, so it also works in the kernel.
> 
> > 
> > 
> > So, if my kernel module calls freeze_bdev when scrub/re-balance is in progress, will I still be able to get a consistent file system state?
> 
> It should be like that with COW enabled.
> 
> Thanks,
> 
> -liubo