A case was reported where s->io_buffer_index can be out of range.
The report skimped on the details but it seems to be triggered
by s->lba == -1 on the READ/READ CD paths (e.g. by sending an
ATAPI command with LBA = 0xFFFFFFFF).  For now paper over it
with assertions.  The first one ensures that there is no overflow
when incrementing s->io_buffer_index, the second checks for the
buffer overrun.

Note that the buffer overrun is only a read, so I am not sure
if the assertion failure is actually less harmful than the overrun.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/ide/atapi.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/hw/ide/atapi.c b/hw/ide/atapi.c
index 14a2b0bb2f..e79157863f 100644
--- a/hw/ide/atapi.c
+++ b/hw/ide/atapi.c
@@ -276,6 +276,8 @@ void ide_atapi_cmd_reply_end(IDEState *s)
         s->packet_transfer_size -= size;
         s->elementary_transfer_size -= size;
         s->io_buffer_index += size;
+        assert(size <= s->io_buffer_total_len);
+        assert(s->io_buffer_index <= s->io_buffer_total_len);
 
         /* Some adapters process PIO data right away.  In that case, we need
          * to avoid mutual recursion between ide_transfer_start
-- 
2.28.0

Am 01.12.2020 um 13:09 hat Paolo Bonzini geschrieben:
> A case was reported where s->io_buffer_index can be out of range.
> The report skimped on the details but it seems to be triggered
> by s->lba == -1 on the READ/READ CD paths (e.g. by sending an
> ATAPI command with LBA = 0xFFFFFFFF).  For now paper over it
> with assertions.  The first one ensures that there is no overflow
> when incrementing s->io_buffer_index, the second checks for the
> buffer overrun.
> 
> Note that the buffer overrun is only a read, so I am not sure
> if the assertion failure is actually less harmful than the overrun.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

I don't think reading LBA 0xFFFFFFFF from a CD image would ever be
valid (or at least I have never seen an 8 TB CD...), so it's probably a
malicious guest. Assertion failure seems okay to me, guests have already
enough ways to kill themselves, so it feels slightly preferable to an
information leak.

Reviewed-by: Kevin Wolf <kwolf@redhat.com>

On Tue, 1 Dec 2020 at 15:17, Kevin Wolf <kwolf@redhat.com> wrote:
>
> Am 01.12.2020 um 13:09 hat Paolo Bonzini geschrieben:
> > A case was reported where s->io_buffer_index can be out of range.
> > The report skimped on the details but it seems to be triggered
> > by s->lba == -1 on the READ/READ CD paths (e.g. by sending an
> > ATAPI command with LBA = 0xFFFFFFFF).  For now paper over it
> > with assertions.  The first one ensures that there is no overflow
> > when incrementing s->io_buffer_index, the second checks for the
> > buffer overrun.
> >
> > Note that the buffer overrun is only a read, so I am not sure
> > if the assertion failure is actually less harmful than the overrun.
> >
> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>
> I don't think reading LBA 0xFFFFFFFF from a CD image would ever be
> valid (or at least I have never seen an 8 TB CD...), so it's probably a
> malicious guest. Assertion failure seems okay to me, guests have already
> enough ways to kill themselves, so it feels slightly preferable to an
> information leak.
>
> Reviewed-by: Kevin Wolf <kwolf@redhat.com>

Thanks; applied to master for 5.2.

-- PMM