Re: [Qemu-devel] Forbid to pass lun 0 to iscsi driver

[Peter Lieven]

Am 03.08.2015 um 09:47 schrieb Andrey Korolyov:
> On Mon, Aug 3, 2015 at 9:45 AM, Peter Lieven <address@hidden> wrote:
> > Am 02.08.2015 um 13:42 schrieb Andrey Korolyov:
> > > Hello,
> > >
> > > As we will never pass LUN#0 as a storage lun, it would be better to
> > > prohibit this at least in iscsi.c, otherwise it will result in an FPU
> > > exception and emulator crash:
> > >
> > > traps: qemu-system-x86[32430] trap divide error ip:7f1dab7b5073
> > > sp:7f1d713e4ae0 error:0 in block-iscsi.so[7f1dab7b0000+8000]
> > >
> > >       353 static bool is_request_lun_aligned(int64_t sector_num, int
> > > nb_sectors,
> > >       354                                       IscsiLun *iscsilun)
> > >       355 {
> > >       356     if ((sector_num * BDRV_SECTOR_SIZE) % iscsilun->block_size ||
> > >       357         (nb_sectors * BDRV_SECTOR_SIZE) % iscsilun->block_size) {
> > >
> > > As far as I can see the LUN#0 can be thrown out on a top level, as one
> > > will never use it directly as an iSCSI backend. Please correct me if
> > > I`m wrong in this assumption.
> >
> > Hi Andrey,
> >
> > LUN 0 is quite common on iSCSI targets. I think what causes the problem
> > is not the LUN ID, but the target which returns 0 for the blocksize. Which
> > target are you using?
> >
> > Peter
> Hi Peter,
>
> I`ve mistyped lun for tgtd upon volume hotplug, which resulted in an
> accidental crash, there is nothing but human factor. Until only LUN0
> may possess such unusual properties, I`d vote to explicitly work it
> around instead of adding generic protection from volumes with
> advertized zero block size.

I will work out a patch that forbids to mount a LUN with zero blocksize.

Peter