On Mon, Jun 06, 2016 at 06:22:34PM +0300, Andrey Ryabinin wrote:
> RHEL7 kernel backported 5a21232983aa ("net: Support for csum_bad in skbuff")
> That's a broken commit because skb_clone() leaves ->csum_bad field
> uninitialized. Later this leads to checksum failures
> (e.g. tcp_v4_rcv() -> skb_checksum_init()) and loosing packets.
>
> The upstream kernel is not affected by this, because the bug was silently
> fixed up by b19372273164 ("net: reorganize sk_buff for faster
> __copy_skb_header()")
> in the same kernel version.
>
> We don't want to backport b19372273164 because it changes KABI and we would
> like to avoid that. Instead, lets directly copy ->csum_bad along with
> ->csum_level and ->remcsum_offload fields which are also copied in the
> upstream kernel after b19372273164 commit.
>
> Note that we already have stepped on this in
> https://jira.sw.ru/browse/PSBM-38924
> but this wasn't properly fixed at the time.
>
> https://jira.sw.ru/browse/PSBM-47290
>
Acked-by: Andrew Vagin <[email protected]>
Thanks
> Signed-off-by: Andrey Ryabinin <[email protected]>
> ---
> net/core/skbuff.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index 686ae59..7334046 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -817,6 +817,9 @@ static void __copy_skb_header(struct sk_buff *new, const
> struct sk_buff *old)
> new->encap_hdr_csum = old->encap_hdr_csum;
> new->csum_valid = old->csum_valid;
> new->csum_complete_sw = old->csum_complete_sw;
> + new->csum_bad = old->csum_bad;
> + new->csum_level = old->csum_level;
> + new->remcsum_offload = old->remcsum_offload;
> #ifdef CONFIG_XFRM
> new->sp = secpath_get(old->sp);
> #endif
> --
> 2.7.3
>
_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel
