On 25.08.22 09:38, Jan Beulich wrote:
On 24.08.2022 16:26, Juergen Gross wrote:The error exit of privcmd_ioctl_dm_op() is calling unlock_pages() potentially with pages being NULL, leading to a NULL dereference. Fix that by calling unlock_pages only if lock_pages() was at least partially successful. Cc: <stable@xxxxxxxxxxxxxxx> Fixes: ab520be8cd5d ("xen/privcmd: Add IOCTL_PRIVCMD_DM_OP") Reported-by: Rustam Subkhankulov <subkhankulov@xxxxxxxxx> Signed-off-by: Juergen Gross <jgross@xxxxxxxx>Reviewed-by: Jan Beulich <jbeulich@xxxxxxxx> albeit I wonder whether you did consider the variant actually reducing code size (and avoiding the need for yet another label), ...--- a/drivers/xen/privcmd.c +++ b/drivers/xen/privcmd.c @@ -679,7 +679,7 @@ static long privcmd_ioctl_dm_op(struct file *file, void __user *udata) rc = lock_pages(kbufs, kdata.num, pages, nr_pages, &pinned); if (rc < 0) { nr_pages = pinned;... dropping this line and ...- goto out; + goto unlock; }for (i = 0; i < kdata.num; i++) {@@ -691,8 +691,9 @@ static long privcmd_ioctl_dm_op(struct file *file, void __user *udata) rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs); xen_preemptible_hcall_end();-out:+ unlock: unlock_pages(pages, nr_pages);... passing "pinned" here.
Looking into this I found another problem: NOT using pinned is wrong, as lock_pages() doesn't guarantee that all pages were really locked. I think lock_pages() should return an error, in case pin_user_pages_fast() didn't lock as many pages es expected. Juergen
Attachment:
OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key
Attachment:
OpenPGP_signature
Description: OpenPGP digital signature
