Re: [Qemu-devel] [Fwd: [PATCH v4 07/21] iscsi: Handle failure for potentially large allocations]

[Kevin Wolf]

Am 22.08.2014 um 09:40 hat Peter Lieven geschrieben:
> Am 22.08.2014 um 09:35 schrieb Peter Lieven:
> > Some code in the block layer makes potentially huge allocations. Failure
> > is not completely unexpected there, so avoid aborting qemu and handle
> > out-of-memory situations gracefully.
> >
> > This patch addresses the allocations in the iscsi block driver.
> >
> > Signed-off-by: Kevin Wolf <address@hidden>
> > Acked-by: Paolo Bonzini <address@hidden>
> > Reviewed-by: Benoit Canet <address@hidden>
> > Reviewed-by: Eric Blake <address@hidden>
> > ---
> >  block/iscsi.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/block/iscsi.c b/block/iscsi.c
> > index 84aa22a..06afa78 100644
> > --- a/block/iscsi.c
> > +++ b/block/iscsi.c
> > @@ -893,7 +893,10 @@ coroutine_fn iscsi_co_write_zeroes(BlockDriverState
> > *bs, int64_t sector_num,
> >      nb_blocks = sector_qemu2lun(nb_sectors, iscsilun);
> >
> >      if (iscsilun->zeroblock == NULL) {
> > -        iscsilun->zeroblock = g_malloc0(iscsilun->block_size);
> > +        iscsilun->zeroblock = g_try_malloc0(iscsilun->block_size);
> > +        if (iscsilun->zeroblock == NULL) {
> > +            return -ENOMEM;
> > +        }
> >      }
> >
> >      iscsi_co_init_iscsitask(iscsilun, &iTask);
> 
> Unfortunately, I missed that one. The zeroblock is typicalls 512 Byte or 4K 
> depending
> on the blocksize.

I don't remember the details, but I think when I went through all
drivers, I couldn't convince myself that a reasonable block size is
enforced somewhere. So I just went ahead and converted the call to be on
the safe side. It can never hurt anyway.

> What is significantly larger is the allocationmap. It is typically created
> on iscsi_open, but is also recreated on iscsi_truncate. I don't have the 
> context why this
> patch was introduced, but I would vote for introducing a bitmap_try_new and 
> issue
> a warning if the allocation fails. The allocationmap is optional we can work 
> without it.
> If the pointer is NULL its not used.

Right, that one I missed because it doesn't directly use g_malloc().

Your proposal sounds good to me. Are you going to prepare a patch?

Kevin