Re: [Qemu-devel] should we try to stop using variable length arrays?

["Philippe Mathieu-Daudé" <philmd@redhat.com>]

On 2/7/19 8:30 PM, Peter Maydell wrote:
> Currently QEMU has 9 uses of variable length arrays
> (found using -Wvla):
> 
> hw/block/dataplane/virtio-blk.c:62:25: warning: variable length array
> used [-Wvla]
>     unsigned long bitmap[BITS_TO_LONGS(nvqs)];
>                         ^
> hw/i386/multiboot.c:364:18: warning: variable length array used [-Wvla]
>     char kcmdline[strlen(kernel_filename) + strlen(kernel_cmdline) + 2];
>                  ^
> hw/i386/xen/xen-hvm.c:618:25: warning: variable length array used [-Wvla]
>     unsigned long bitmap[DIV_ROUND_UP(npages, width)];
>                         ^
> hw/intc/xics.c:604:18: warning: variable length array used [-Wvla]
>     uint8_t flags[ics->nr_irqs];
>                  ^
> hw/net/fsl_etsec/rings.c:383:18: warning: variable length array used [-Wvla]
>     uint8_t  padd[etsec->rx_padding];
>                  ^
> hw/ppc/pnv.c:130:26: warning: variable length array used [-Wvla]
>     uint32_t servers_prop[smt_threads];
>                          ^
> hw/ppc/spapr.c:162:26: warning: variable length array used [-Wvla]
>     uint32_t servers_prop[smt_threads];
>                          ^
> hw/ppc/spapr.c:163:27: warning: variable length array used [-Wvla]
>     uint32_t gservers_prop[smt_threads * 2];
>                           ^
> linux-user/syscall.c:3478:23: warning: variable length array used [-Wvla]
>     struct sembuf sops[nsops];
>                       ^
> 
> Should we be looking to get rid of these and turn on the -Wvla
> warning? I know the Linux kernel has recently decided to do this

Yes please!

> (some rationale at the start of https://lwn.net/Articles/749064/).
> Now that doesn't necessarily apply to us as a userspace program,
> but on the other hand if any of these were allowing the guest to
> determine the size of an on-stack array that would not be great.
> (The linux-user one is bogus in that way, though not a security issue
> as the guest code there has full control anyway.)

The manpage says "[The semaphore] limit should not be raised above 1000".

> 
> Opinions? I admit that to some extent this is just my sense of
> tidiness thinking that if we only have a handful of uses of
> something we should squash that down to zero :-)

None of these case have strong justification to use the stack rather
than the heap (and the rest of the codebase heavily uses the heap).

> 
> thanks
> -- PMM
>