ipv4: only create late gso-skb if skb is already set up with CHECKSUM_PARTIAL

Otherwise we break the contract with GSO to only pass CHECKSUM_PARTIAL
skbs down. This can easily happen with UDP+IPv4 sockets with the first
MSG_MORE write smaller than the MTU, second write is a sendfile.

Returning -EOPNOTSUPP lets the callers fall back into normal sendmsg path,
were we calculate the checksum manually during copying.

Commit d749c9cbffd6 ("ipv4: no CHECKSUM_PARTIAL on MSG_MORE corked
sockets") started to exposes this bug.

Fixes: d749c9cbffd6 ("ipv4: no CHECKSUM_PARTIAL on MSG_MORE corked sockets")
Reported-by: Jiri Benc <jbenc@redhat.com>
Cc: Jiri Benc <jbenc@redhat.com>
Reported-by: Wakko Warner <wakko@animx.eu.org>
Cc: Wakko Warner <wakko@animx.eu.org>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c
index 64878efa045c132d1ce511c326c2da04fc77a895..565bf64b2b7d6047a29e69df00fb3b85ec84f06a 100644 (file)
--- a/net/ipv4/ip_output.c
+++ b/net/ipv4/ip_output.c
@@ -1236,13 +1236,16 @@ ssize_t ip_append_page(struct sock *sk, struct flowi4 *fl4, struct page *page,
        if (!skb)
                return -EINVAL;
 
-       cork->length += size;
        if ((size + skb->len > mtu) &&
            (sk->sk_protocol == IPPROTO_UDP) &&
            (rt->dst.dev->features & NETIF_F_UFO)) {
+               if (skb->ip_summed != CHECKSUM_PARTIAL)
+                       return -EOPNOTSUPP;
+
                skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
                skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
        }
+       cork->length += size;
 
        while (size > 0) {
                if (skb_is_gso(skb)) {