netdev - Re: [PATCH net 1/2] sfc: handle nonlinear SKBs in efx_filter

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1464281796.5939.96.camel@edumazet-glaptop3.roam.corp.google.com>
Date:	Thu, 26 May 2016 09:56:36 -0700
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Edward Cree <ecree@...arflare.com>
Cc:	linux-net-drivers@...arflare.com, davem@...emloft.net,
	netdev@...r.kernel.org
Subject: Re: [PATCH net 1/2] sfc: handle nonlinear SKBs in efx_filter_rfs()

On Thu, 2016-05-26 at 17:01 +0100, Edward Cree wrote:
> Previously efx_filter_rfs() assumed that the headers it needed (802.1Q, IP)
>  would be present in the linear data area of the SKB.
> When running with debugging I found that this is not always the case and
>  that in fact the data may all be paged.
> So now use skb_header_pointer() to extract the data.
> 
> Also replace EFX_BUG_ON_PARANOID checks for insufficient data with checks
>  that return -EINVAL, as this case is possible if the received packet was
>  too short.
> 
> Signed-off-by: Edward Cree <ecree@...arflare.com>
> ---
>  drivers/net/ethernet/sfc/rx.c | 39 +++++++++++++++++++++++----------------
>  1 file changed, 23 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/net/ethernet/sfc/rx.c b/drivers/net/ethernet/sfc/rx.c
> index 8956995..52790f0 100644
> --- a/drivers/net/ethernet/sfc/rx.c
> +++ b/drivers/net/ethernet/sfc/rx.c
> @@ -842,25 +842,32 @@ int efx_filter_rfs(struct net_device *net_dev, const struct sk_buff *skb,
>  	struct efx_nic *efx = netdev_priv(net_dev);
>  	struct efx_channel *channel;
>  	struct efx_filter_spec spec;
> +	/* 60 octets is the maximum length of an IPv4 header (all IPv6 headers
> +	 * are 40 octets), and we pull 4 more to get the port numbers
> +	 */
> +	#define EFX_RFS_HEADER_LENGTH	(sizeof(struct vlan_hdr) + 60 + 4)

Lot of magic here. Yet another flow dissection.

Seems to be a good place to use net/core/flow_dissector.c helpers.

I truly believe that every time a driver has a private flow dissector,
it always have at least one bug.