]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
esp6: Fix RX checksum after header pull
authorYossi Kuperman <yossiku@mellanox.com>
Tue, 1 Aug 2017 09:49:07 +0000 (12:49 +0300)
committerSteffen Klassert <steffen.klassert@secunet.com>
Wed, 2 Aug 2017 09:00:15 +0000 (11:00 +0200)
Both ip6_input_finish (non-GRO) and esp6_gro_receive (GRO) strip
the IPv6 header without adjusting skb->csum accordingly. As a
result CHECKSUM_COMPLETE breaks and "hw csum failure" is written
to the kernel log by netdev_rx_csum_fault (dev.c).

Fix skb->csum by substracting the checksum value of the pulled IPv6
header using a call to skb_postpull_rcsum.

This affects both transport and tunnel modes.

Note that the fix occurs far from the place that the header was
pulled. This is based on existing code, see:
ipv6_srh_rcv() in exthdrs.c and rawv6_rcv() in raw.c

Signed-off-by: Yossi Kuperman <yossiku@mellanox.com>
Signed-off-by: Ilan Tayari <ilant@mellanox.com>
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
net/ipv6/esp6.c

index 0ca1db62e3815a559cd26e037287107e649318ba..74bde202eb9a28c42b3b2875a2444acf04aba30d 100644 (file)
@@ -495,6 +495,8 @@ int esp6_input_done2(struct sk_buff *skb, int err)
 
        trimlen = alen + padlen + 2;
        if (skb->ip_summed == CHECKSUM_COMPLETE) {
+               skb_postpull_rcsum(skb, skb_network_header(skb),
+                                  skb_network_header_len(skb));
                csumdiff = skb_checksum(skb, skb->len - trimlen, trimlen, 0);
                skb->csum = csum_block_sub(skb->csum, csumdiff,
                                           skb->len - trimlen);