netdev - [PATCH bpf-next RFCv3 5/6] veth: add AF

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: <1545856073-8680-6-git-send-email-u9012063@gmail.com>
Date:   Wed, 26 Dec 2018 12:27:52 -0800
From:   William Tu <u9012063@...il.com>
To:     bjorn.topel@...il.com, magnus.karlsson@...il.com, ast@...nel.org,
        daniel@...earbox.net, netdev@...r.kernel.org,
        makita.toshiaki@....ntt.co.jp, yihung.wei@...il.com,
        magnus.karlsson@...el.com
Subject: [PATCH bpf-next RFCv3 5/6] veth: add AF_XDP RX support.

If the receiving veth side has umem rx enabled, the patch
directly copy the packet from the peer side's send buffer
to the umem receive buffer.  This requires running AF_XDP
as both side of the veth peer.  For example:
Receiver:
  # ip netns exec at_ns0 xdpsock -r -N -z -i p0
Sender:
  # xdpsock -i p1 -t -N -z
The performance increases from 1.4Mpps to 2.3Mpps.

Signed-off-by: William Tu <u9012063@...il.com>
---
 drivers/net/veth.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index 551444195398..8aac67554880 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -766,11 +766,28 @@ static int veth_xsk_poll(struct napi_struct *napi, int budget)
 		struct sk_buff *skb;
 		struct page *page;
 		void *vaddr;
+		u64 handle;
 		u32 len;
 
 		if (!xsk_umem_consume_tx_virtual(peer_rq->xsk_umem, &vaddr, &len))
 			break;
 
+		if (rq->xsk_umem && xsk_umem_peek_addr(rq->xsk_umem, &handle)) {
+			char *daddr;
+			u64 hr = 0;
+
+			/* the peer side also has umem enabled,
+			 * copy directly to it.
+			 */
+			handle &= rq->xsk_umem->chunk_mask;
+			hr = rq->xsk_umem->headroom + XDP_PACKET_HEADROOM;
+			daddr = xdp_umem_get_data(rq->xsk_umem, handle);
+			daddr += hr;
+			memcpy((void *)daddr, vaddr, len);
+			xsk_umem_discard_addr(rq->xsk_umem);
+			vaddr = daddr;
+		}
+
 		xdpf.data = vaddr + metasize;
 		xdpf.len = len;
 		xdpf.headroom = 0;
-- 
2.7.4