Conntrack bridge only tracks untagged and 802.1q.
To make the bridge-fastpath experience more similar to the
forward-fastpath experience, introduce patches for double vlan,
pppoe and pppoe-in-q tagged packets to bridge conntrack.
Changes in v20:
- Moved skb_pull/push for icmpv4/6 checksum calculation correction to
underlying functions, as these underlying functions are also called
directly. Adjusted commit title and message accordingly.
- Altered nf_ct_bridge_pre_inner() so it can also be used when doing
re-fragmentation.
- Added ip-fragmented packet handling for double vlan, pppoe and
pppoe-in-q.
- Renamed nf_ct_bridge_pre_inner() to nf_ct_bridge_inner(), as it is also
used in nf_ct_bridge_post().
- Dropped "netfilter: nft_chain_filter: Add bridge double vlan and pppoe".
- Dropped "netfilter: nft_set_pktinfo_ipv4/6_validate".
(They are replaced by other patches using meta).
- Dropped "Add net: pppoe: avoid zero-length arrays in struct pppoe_hdr"
(It is applied separately)
Changes in v19:
- Add net: pppoe: avoid zero-length arrays in struct pppoe_hdr.
(It was part of other patch-set of mine, moved to this patch-set)
Changes in v18:
- Rebased
- nf_conntrack_bridge: added #include <linux/ppp_defs.h>
- nf_checksum(_partial)(): changed WARN_ON to WARN_ON_ONCE.
- nft_set_bridge_pktinfo(): changed call to pskb_may_pull() to
skb_header_pointer().
Changes in v17:
- Add patch for nft_set_pktinfo_ipv4/6_validate() adding nhoff argument.
- Stopped using skb_set_network_header() in nft_set_bridge_pktinfo,
using the new offset for nft_set_pktinfo_ipv4/6_validate instead.
- When pskb_may_pull() fails in nft_set_bridge_pktinfo() set proto to 0,
resulting in pktinfo unspecified.
Changes in v16:
- Changed nft_chain_filter patch: Only help populating pktinfo offsets,
call nft_do_chain() with original network_offset.
- Changed commit messages.
- Removed kernel-doc comments.
Changes in v15:
- Do not munge skb->protocol.
- Introduce nft_set_bridge_pktinfo() helper.
- Introduce nf_ct_bridge_pre_inner() helper.
- nf_ct_bridge_pre(): Don't trim on ph->hdr.length, only compare to what
ip header claims and return NF_ACCEPT if it does not match.
- nf_ct_bridge_pre(): Renamed u32 data_len to pppoe_len.
- nf_ct_bridge_pre(): Reset network_header only when ret == NF_ACCEPT.
- nf_checksum(_partial)(): Use of skb_network_offset().
- nf_checksum(_partial)(): Use 'if (WARN_ON()) return 0' instead.
- nf_checksum(_partial)(): Added comments
Changes in v14:
- nf_checksum(_patial): Use DEBUG_NET_WARN_ON_ONCE(
!skb_pointer_if_linear()) instead of pskb_may_pull().
- nft_do_chain_bridge: Added default case ph->proto is neither
ipv4 nor ipv6.
- nft_do_chain_bridge: only reset network header when ret == NF_ACCEPT.
Changes in v13:
- Do not use pull/push before/after calling nf_conntrack_in() or
nft_do_chain().
- Add patch to correct calculating checksum when skb->data !=
skb_network_header(skb).
Changes in v12:
- Only allow tracking this traffic when a conntrack zone is set.
- nf_ct_bridge_pre(): skb pull/push without touching the checksum,
because the pull is always restored with push.
- nft_do_chain_bridge(): handle the extra header similar to
nf_ct_bridge_pre(), using pull/push.
Changes in v11:
- nft_do_chain_bridge(): Proper readout of encapsulated proto.
- nft_do_chain_bridge(): Use skb_set_network_header() instead of thoff.
- removed test script, it is now in separate patch.
v10 split from patch-set: bridge-fastpath and related improvements v9
Eric Woudstra (2):
netfilter: utils: nf_ip(6)_checksum(_partial) correct
data!=networkheader
netfilter: bridge: Add conntrack double vlan and pppoe
include/linux/netfilter_bridge.h | 6 +
net/bridge/netfilter/nf_conntrack_bridge.c | 203 ++++++++++++++++++---
net/netfilter/utils.c | 52 +++++-
3 files changed, 228 insertions(+), 33 deletions(-)
--
2.53.0
