Will xfs_growfs succeed on a full file system?

Will xfs_growfs succeed on a full file system?

[Geir A. Myrestrand]

Does xfs_growfs depend on some space left on the file system in order to 
be able to grow it?

I have a colleague who ran into an issue where a file system resize 
failed. The file system is 100% full.

Aside from analyzing what happened in his case, should XFS be able to 
grow a file system that is 100% full?

The device has already been expanded, it is the XFS file system that 
fails to resize. I just wonder if that is by design, or whether it is an 
issue.

-- 

Geir A. Myrestrand

Re: Will xfs_growfs succeed on a full file system?

[Eric Sandeen]

Geir A. Myrestrand wrote:
> Does xfs_growfs depend on some space left on the file system in order to 
> be able to grow it?
> 
> I have a colleague who ran into an issue where a file system resize 
> failed. The file system is 100% full.
> 
> Aside from analyzing what happened in his case, should XFS be able to 
> grow a file system that is 100% full?
> 
> The device has already been expanded, it is the XFS file system that 
> fails to resize. I just wonder if that is by design, or whether it is an 
> issue.
> 

Off the top of my head, I think it should work ok even if full, although
I could be (and apparently I am) wrong here.  How exactly did the growfs
fail?

I actually wasn't able to completely fill my filesystem, got stuck at
20k left.  :)  but growing that from 50M to 100M worked fine for me.

-Eric

Re: Will xfs_growfs succeed on a full file system?

[Geir A. Myrestrand]

Eric Sandeen wrote:
> Geir A. Myrestrand wrote:
>> Does xfs_growfs depend on some space left on the file system in order to 
>> be able to grow it?
>>
>> I have a colleague who ran into an issue where a file system resize 
>> failed. The file system is 100% full.
>>
>> Aside from analyzing what happened in his case, should XFS be able to 
>> grow a file system that is 100% full?
>>
>> The device has already been expanded, it is the XFS file system that 
>> fails to resize. I just wonder if that is by design, or whether it is an 
>> issue.
>>
> 
> Off the top of my head, I think it should work ok even if full, although
> I could be (and apparently I am) wrong here.  How exactly did the growfs
> fail?
> 
> I actually wasn't able to completely fill my filesystem, got stuck at
> 20k left.  :)  but growing that from 50M to 100M worked fine for me.
> 
> -Eric

The only error I saw in his output was this line:

xfs_growfs:
XFS_IOC_FSGROWFSDATA xfsctl failed: No space left on device

He claims that the file system is actually resized after it has been 
re-mounted. He verifies with df:

After expansion (with xfs_growfs):

# df -k
Filesystem           1K-blocks      Used Available Use% Mounted on
/dev/vbdi6               93504     93504         0 100% /nas/NASDisk-00006

# df -i
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/vbdi6                64       6      58   10% /nas/NASDisk-00006

After re-mount:

# df -k
Filesystem           1K-blocks      Used Available Use% Mounted on
/dev/vbdi6              200000     93516    106484  47% /nas/NASDisk-00006

# df -i
Filesystem            Inodes   IUsed   IFree IUse% Mounted on
/dev/vbdi6            204800       6  204794    1% /nas/NASDisk-00006


-- 

Geir A. Myrestrand

Re: Will xfs_growfs succeed on a full file system?

[David Chinner]

On Tue, Jan 23, 2007 at 10:19:27AM -0500, Geir A. Myrestrand wrote:
> Does xfs_growfs depend on some space left on the file system in order to 
> be able to grow it?

Yes, it does require some space to be left because growing the
filesystem can require extending the original last AG to the full
size and that means there may be btree work to be journalled
and hence we have to reserve blocks for that to succeed in all
cases.

> I have a colleague who ran into an issue where a file system resize 
> failed. The file system is 100% full.

Yup, we hit that in QA recently and have an open bug for it.

> Aside from analyzing what happened in his case, should XFS be able to 
> grow a file system that is 100% full?

Yes, it should.

> The device has already been expanded, it is the XFS file system that 
> fails to resize. I just wonder if that is by design, or whether it is an 
> issue.

It's a bug, really, but one you can easily work around by freeing
up about 50k of space. If you've got a really large filesystem, then
you might need to free more space than that.

At some point in my copious amounts of free time I'll fix it properly...

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group

Re: Will xfs_growfs succeed on a full file system?

[Geir A. Myrestrand]

David Chinner wrote:
> On Tue, Jan 23, 2007 at 10:19:27AM -0500, Geir A. Myrestrand wrote:
>> Does xfs_growfs depend on some space left on the file system in order to 
>> be able to grow it?
> 
> Yes, it does require some space to be left because growing the
> filesystem can require extending the original last AG to the full
> size and that means there may be btree work to be journalled
> and hence we have to reserve blocks for that to succeed in all
> cases.
> 
>> I have a colleague who ran into an issue where a file system resize 
>> failed. The file system is 100% full.
> 
> Yup, we hit that in QA recently and have an open bug for it.
> 
>> Aside from analyzing what happened in his case, should XFS be able to 
>> grow a file system that is 100% full?
> 
> Yes, it should.
> 
>> The device has already been expanded, it is the XFS file system that 
>> fails to resize. I just wonder if that is by design, or whether it is an 
>> issue.
> 
> It's a bug, really, but one you can easily work around by freeing
> up about 50k of space. If you've got a really large filesystem, then
> you might need to free more space than that.
> 
> At some point in my copious amounts of free time I'll fix it properly...

Thanks a lot for your feedback David. We look forward to a fix, but just 
knowing that it is an issue and that we easily can work around it helps 
a lot. Thanks again!


-- 

Geir A. Myrestrand