Re: kern/45441: aio_write and aio_read connected to one pipe don't make progress

[Alexander Nasonov <alnsn%yandex.ru@localhost>]

David Laight wrote:
>  On Wed, Oct 12, 2011 at 09:26:40PM +0100, Alexander Nasonov wrote:
>  > The best fix, I think, is return ESPIPE for non-seekable fildes rather
>  > than silently ignoring aio_offset. This would indirectly solve the issue
>  > described in this PR.
>  
>  IIRC that breaks historic practise.

Do you have any reference or any more details on this?

>  The likely fubar here is that 'aio' is synchronous on a single kernel
>  thread.

Yes, I know. I wrote the test to prove that the current implementation
has problems.

Having separate writer and reader kernel threads would solve the
problem but it's quite a lot of work.

Alex