tech-kern archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: ffs_balloc_ufs1 error handling

On Sun, Dec 07, 2008 at 10:34:09PM +0200, Antti Kantee wrote:
> 1) did you even read the block?  Why don't you propose removing the
>    softdep conditionals from inside of it?

It would only clutter up the discussion for no purpose. As this code is
supposed to be removed with the rest of softdep at some point, it would
only create churn as well.

> 2) you do not state why you want to do this

This is doing out-of-band manipulation of cylinder groups. I'm not even
sure if the locking is correct in the current form, but it will bite any
changes to the serialisation model around cylinder groups. But it
doesn't matter for why I want to clean this up.

> 3) you do not explain why the same hypothetical situation would not be
>    valid for ufs2

It can and the very same reasons would apply for UFS2. So if this
doesn't serve a purpose for the much older UFS1 code base, it is useless
for UFS2 as well.

> 4) you do not state where unwinding is done now

Did you read my mail at all? It wasn't done at all before the
introduction of softdep. From reading the code it only forces a lot of
serialisation and manual pruning of the buffer cache for !softdep.

> One situation where block allocation can fail even without the file
> system being full is with B_CONTIG.  Does that case still work?

Is there any way to test B_CONTIG allocations? This looks like dead code
right now...


Home | Main Index | Thread Index | Old Index