Re: [Qemu-devel] [PATCH v4] Stop VM on ENOSPC error.

[Ian Jackson]

Gleb Natapov writes ("Re: [Qemu-devel] [PATCH v4] Stop VM on ENOSPC error."):
> On Wed, Jan 21, 2009 at 04:37:08PM +0000, Ian Jackson wrote:
> > But ide.c doesn't call bdrv_pwrite; it calls bdrv_aio_write.
> > bdrv_aio_write would do what your ide.c code currently does.
> 
> ide.c calls bdrv_write.

In the non-DMA case, yes.  That would have to be changed to call
bdrv_aio_write.

> And synchronous calls will behave differently. I am all for doing it in
> one place, but I am against doing it just for part of API calls. So do
> you have better solution for sync calls except don't use them?

Well, stopping the VM can't be done for synchronous calls so we
clearly have three choices
  * write out the complicated asynchronous vm_stop-on-enospc logic
    at least twice, once for ide and scsi
  * do it only for ide
  * do it only for part of the api, and use only that api in ide
    and scsi

I prefer the last of these.  Doing it for only part of the internal
API is less bad than doing it only for some of the actual emulated
hardware.  I also think doing it for only part of the internal API is
less bad than writing the same functionality twice.

Writing the same functionality twice seems to me to be one of the
cardinal sins of programming.

But it would be good to hear what other people think.  If they prefer
the other options then your patch should go in as-is.

> And BTW block.c is compiled also for qemu-{img|mbd} no vmstop there.

We can finesse this with a suitable hook or an #ifdef.

Ian.