netdev - Re: Broken TX checksumming offloads

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 for Android: free password hash cracker in your pocket

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

Message-ID: <fddashfjkasdhklfjashlkjfs@rechot.qmqm.pl>
Date:	Tue, 30 Nov 2010 15:23:52 +0100
From:	MichałMirosław 
	<mirq-linux@...e.qmqm.pl>
To:	netdev@...t.kernel.org
Cc:	Ben Hutchings <bhutchings@...arflare.com>,
	Jesse Gross <jesse@...ira.com>
Subject: Re: Broken TX checksumming offloads

On Mon, Nov 29, 2010 at 06:35:44PM -0800, Jesse Gross wrote:
> 2010/11/29 Ben Hutchings <bhutchings@...arflare.com>:
> > On Mon, 2010-11-29 at 19:17 +0100, Michał Mirosław wrote:
> >> Hi!
> >>
> >> Unless I'm horribly mistaken, generic HW checksumming works as follows:
> >>  - driver sets netdev->features & NETIF_F_HW_CSUM to indicate support
> >>    for generic checksumming; if the flag is not set, networking core
> >>    will checksum skb before calling ndo_start_xmit (let's ignore
> >>    other checksumming options for now) and not pass skb with
> >>    skb->ip_summed == CHECKSUM_PARTIAL
> >>  - ndo_start_xmit() should use skb->csum_start and skb->csum_offset
> >>    (or skb->csum) to update checksum in software or instruct HW to do so
> >>
> >> Looking at pch_gbe_xmit_frame() and its callee - pch_gbe_tx_queue() it
> >> looks like the driver should set NETIF_F_IP_CSUM instead of NETIF_F_HW_CSUM
> >> feature.
> >>
> >> Similar thing happens in ixgbe driver: it sets NETIF_F_HW_CSUM and checks
> >> for skb->ip_summed == CHECKSUM_PARTIAL, but then just warns on protocols
> >> other that TCP and SCTP (see: ixgbe_psum()).
> >
> > AFAIK only {TCP,UDP}/IPv{4,6} use the simple 16-bit checksum algorithm
> > that NETIF_F_HW_CSUM implies, so in practice it is equivalent to
> > NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM even though it doesn't mean the same
> > thing.
> 
> There are a few cases that I can think of where the extra generality
> of NETIF_F_HW_CSUM is useful:
> 
> * The checksum in GRE (maybe other protocols as well) uses the same
> algorithm as TCP/UDP.  In theory this means that we could use hardware
> offloading for GRE on cards that have NETIF_F_HW_CSUM.  The one issue
> is that NETIF_F_IP_CSUM really means TCP/UDP checksumming but we only
> test if the protocol is IP.  This means that other IP based protocols
> (like GRE) can't use checksum offloading since there is no software
> fallback path.
> 
> * TCP/UDP packets that are deeply encapsulated.  Currently we can
> express that checksum offloading is supported in one level of vlan
> encapsulation through vlan_features.  However, a NIC that exposes
> NETIF_F_HW_CSUM should in theory be able to checksum over a packet
> that has an arbitrary number of vlan tags in front of it, a GRE
> header, or any other type of encapsulation.

So the following patches should fix what's in tree regarding
NETIF_F_HW_CSUM interpretation. The patches have no interdependencies,
so can be applied independently. The last one is a cleanup that tagged
along.

Drivers that didn't care about csum_{start,offset}:

benet
bnx2x
jme
pch_gbe
sc92031
stmmac
vxge

Best Regards,
Michał Mirosław
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html