On Wed, 28 Oct 2026 at 05:21, Harsh Prateek Bora <[email protected]> wrote: > > From: Chinmay Rath <[email protected]> > > vscsi_send_adapter_info() used the guest-supplied common.length > field directly as the size argument to spapr_vio_dma_write(). > Since common.length is controlled by the guest, a value larger > than sizeof(struct mad_adapter_info_data) (148 bytes) caused > dma_memory_write() to read past the end of the stack-allocated info > struct, leaking arbitrary host stack contents into guest memory. > > Fix this by clamping the write length to sizeof(info), then using > spapr_vio_dma_set() to zero-fill any remaining bytes the guest > requested beyond that size. This mirrors the existing pattern in > vscsi_send_capabilities(). > > Closes: https://gitlab.com/qemu-project/qemu/-/work_items/4626 > Reported-by: Guthrie McAfee Armstrong <[email protected]> > Signed-off-by: Chinmay Rath <[email protected]> > Reviewed-by: Gautam Menghani <[email protected]> > Tested-by: Gautam Menghani <[email protected]> > Link: > https://lore.kernel.org/qemu-devel/[email protected] > Signed-off-by: Harsh Prateek Bora <[email protected]>
Shouldn't this be Cc: [email protected] ? Given that spapr-vscsi is supposed to be covered by the "virtualization use case", have we deliberately decided we didn't want to give this a CVE, or did that step in the process get missed ? thanks -- PMM
