Re: [PATCH] virtio: fix out of range array access

[Amit Shah <amit.shah@redhat.com>]

On (Thu) Feb 25 2010 [19:13:41], Michael S. Tsirkin wrote:
> I have observed the following error on virtio-net module unload:
> 
> ------------[ cut here ]------------
> WARNING: at kernel/irq/manage.c:858 __free_irq+0xa0/0x14c()
> Hardware name: Bochs
> Trying to free already-free IRQ 0
> Modules linked in: virtio_net(-) virtio_blk virtio_pci virtio_ring
> virtio af_packet e1000 shpchp aacraid uhci_hcd ohci_hcd ehci_hcd [last
> unloaded: scsi_wait_scan]
> Pid: 1957, comm: rmmod Not tainted 2.6.33-rc8-vhost #24
> Call Trace:
>  [<ffffffff8103e195>] warn_slowpath_common+0x7c/0x94
>  [<ffffffff8103e204>] warn_slowpath_fmt+0x41/0x43
>  [<ffffffff810a7a36>] ? __free_pages+0x5a/0x70
>  [<ffffffff8107cc00>] __free_irq+0xa0/0x14c
>  [<ffffffff8107cceb>] free_irq+0x3f/0x65
>  [<ffffffffa0081424>] vp_del_vqs+0x81/0xb1 [virtio_pci]
>  [<ffffffffa0091d29>] virtnet_remove+0xda/0x10b [virtio_net]
>  [<ffffffffa0075200>] virtio_dev_remove+0x22/0x4a [virtio]
>  [<ffffffff812709ee>] __device_release_driver+0x66/0xac
>  [<ffffffff81270ab7>] driver_detach+0x83/0xa9
>  [<ffffffff8126fc66>] bus_remove_driver+0x91/0xb4
>  [<ffffffff81270fcf>] driver_unregister+0x6c/0x74
>  [<ffffffffa0075418>] unregister_virtio_driver+0xe/0x10 [virtio]
>  [<ffffffffa0091c4d>] fini+0x15/0x17 [virtio_net]
>  [<ffffffff8106997b>] sys_delete_module+0x1c3/0x230
>  [<ffffffff81007465>] ? old_ich_force_enable_hpet+0x117/0x164
>  [<ffffffff813bb720>] ? do_page_fault+0x29c/0x2cc
>  [<ffffffff81028e58>] sysenter_dispatch+0x7/0x27
> ---[ end trace 15e88e4c576cc62b ]---
> 
> The bug is in virtio-pci: we use msix_vector as array index to get irq
> entry, but some vqs do not have a dedicated vector so this causes an out
> of bounds access.  By chance, we seem to often get 0 value, which
> results in this error.

This is triggered when per-vq MSI vectors are enabled and some vqs do
not have a handler registered, in which case they do not need a vector
at all.

I'm wondering why I didn't hit this with the virtio_console code when I
test module removal, where I do have such a config.

> Fix by verifying that vector is legal before using it as index.
> 
> Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> ---
> Shirley, Amit, with Rusty on vacation, need other reviewers.  Could you
> please review the following patch and ack on list if appropriate?
> 
>  drivers/virtio/virtio_pci.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/virtio/virtio_pci.c b/drivers/virtio/virtio_pci.c
> index 28d9cf7..7127bfe 100644
> --- a/drivers/virtio/virtio_pci.c
> +++ b/drivers/virtio/virtio_pci.c
> @@ -473,7 +473,8 @@ static void vp_del_vqs(struct virtio_device *vdev)
>  
>  	list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
>  		info = vq->priv;
> -		if (vp_dev->per_vq_vectors)
> +		if (vp_dev->per_vq_vectors &&
> +			info->msix_vector != VIRTIO_MSI_NO_VECTOR)
>  			free_irq(vp_dev->msix_entries[info->msix_vector].vector,
>  				 vq);
>  		vp_del_vq(vq);

Acked-by: Amit Shah <amit.shah@redhat.com>

		Amit