Re: [syzbot] [virtualization?] KMSAN: uninit-value in virtqueue_add (4)

[Alexander Potapenko]

On Thu, Jan 4, 2024 at 9:45 PM Stefan Hajnoczi <stefanha@xxxxxxxxxx> wrote:
>
> On Tue, Jan 02, 2024 at 08:03:46AM -0500, Michael S. Tsirkin wrote:
> > On Mon, Jan 01, 2024 at 05:38:24AM -0800, syzbot wrote:
> > > Hello,
> > >
> > > syzbot found the following issue on:
> > >
> > > HEAD commit:    fbafc3e621c3 Merge tag 'for_linus' of git://git.kernel.org..
> > > git tree:       upstream
> > > console+strace: https://syzkaller.appspot.com/x/log.txt?x=173df3e9e80000
> > > kernel config:  https://syzkaller.appspot.com/x/.config?x=e0c7078a6b901aa3
> > > dashboard link: https://syzkaller.appspot.com/bug?extid=d7521c1e3841ed075a42
> > > compiler:       Debian clang version 15.0.6, GNU ld (GNU Binutils for Debian) 2.40
> > > syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=1300b4a1e80000
> > > C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=130b0379e80000
> > >
> > > Downloadable assets:
> > > disk image: https://storage.googleapis.com/syzbot-assets/1520f7b6daa4/disk-fbafc3e6.raw.xz
> > > vmlinux: https://storage.googleapis.com/syzbot-assets/8b490af009d5/vmlinux-fbafc3e6.xz
> > > kernel image: https://storage.googleapis.com/syzbot-assets/202ca200f4a4/bzImage-fbafc3e6.xz
> > >
> > > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > > Reported-by: syzbot+d7521c1e3841ed075a42@xxxxxxxxxxxxxxxxxxxxxxxxx
> > >
> > > =====================================================
>
> Hi Alexander,
> Please take a look at this KMSAN failure. The uninitialized memory was
> created for the purpose of writing a coredump. vring_map_one_sg() should
> have direction=DMA_TO_DEVICE.
>
Hi Stefan,

I took a closer look, and am pretty confident this is a false positive.
I tried adding memset(..., 0xab, PAGE_SIZE << order) to alloc_pages()
and never saw
the 0xab pattern in the buffers for which KMSAN reported an error.

This probably isn't an error in 88938359e2df ("virtio: kmsan:
check/unpoison scatterlist in
vring_map_one_sg()"), which by itself should be doing a sane thing:
report an error if an
uninitialized buffer is passed to it. It is more likely that we're
missing some initialization that
happens in coredump.c

Does anyone have an idea where coredump.c is supposed to be
initializing these pages?
Maybe there are some inline assembly functions involved in copying the data?