RE: [PATCH] aio: fix kernel bug when page is temporally busy

["Ananiev, Leonid I" <leonid.i.ananiev@intel.com>]

But if page is busy as invalidate_inode_pages2_range() says because of
bh_count>0 then aio_complet() is not called from aio_run_iocb() and next
retry() will get the iocb result.

Leonid

-----Original Message-----
From: Ken Chen [mailto:kenchen@google.com] 
Sent: Saturday, February 10, 2007 9:05 PM
To: Ananiev, Leonid I
Cc: suparna@in.ibm.com; Andrew Morton; linux-kernel@vger.kernel.org;
linux-aio; Zach Brown; Chris Mason; Badari Pulavarty
Subject: Re: [PATCH] aio: fix kernel bug when page is temporally busy

On 2/9/07, Ananiev, Leonid I <leonid.i.ananiev@intel.com> wrote:
> I have used EIOCBRETRY in the patch to minimize source code
modification
> only.
> [...]
> A lot of errno's have different meaning in different functions or
> contexts. EAGAIN could be used instated of EIOCBRETRY for irredundant
> set.

I also think the original patch is wrong.  It might shut up kernel
panic by eliminate double calls to aio_complete(), but it will
silently introduce data corruption.

If invalidate_inode_pages2_range() says it can not invalidate pages,
while dio to the same file offset range is in flight, something is
really wrong there.  In generic_file_direct_IO, the function
explicitly flushes all dirty pages and wait on them before submits
DIO.

So any error value returned from invalidate_inode_pages2_range() has
to be taken seriously in the direct IO submit path instead of dropping
it to the floor.

- Ken