On Wed, 2002-05-01 at 21:55, Thor Lancelot Simon wrote:
>
> This is quite a common problem with XFS filesystems in academic
> fileserver applications. What's probably going on here is that
> you have enough fragmentation that XFS cannot find a 64K extent to
> allocate for inodes. Back when it wasn't safe to use xfs_fsr on
> Irix (it had a nasty habit of eating files with lseek holes, among
> other ugliness) we used to have to regularly dump, recreate, and
> restore the student home directories at the university where I
> worked, in order to get around this.
>
> Have a look at xfs_db -c freesp and see if you've got any extents
> 64K or larger left free. If you don't, you won't be able to create
> files once you've used up all the inodes in the extents already thus
> allocated. However, if you can find a little bit of stuff to delete
> so fsr can do its thing, it will probably solve your problem for you.
>
> Thor
Just to correct this - XFS does not allocate inodes in 64K chunks,
for filesystems with block sizes less than 8K it uses 8K for each
inode cluster, for larger block sizes it uses one filesystem block.
On ia32 linux we are currently limited to 4K filesystem blocks, so
inodes are allocated in 8K chunks.
We have seen filesystems get into the state where there is no room
for another inode cluster, but it is not something we in the SGI
filesystem group have seen very often.
Steve
|