> Well, looking at the code I don't readily see how this works either, > but note that tcp_sendmsg has the same check also. If your analysis is > correct then checksum for IPv6 would commonly be broken (i.e. several > devices support IPv4 checksum but not IPv6)-- so I find that hard to > believe so I'm probably missing something too! (at least this > ambiguity is one more reason why we need to get rid of NETIF_F_IP_CSUM > and NETIF_F_V6_CSUM!).
I think the checksum-building is not broken because of this on itself. The problem i see is for example in case pages are sent from a block-device over the net. Depending on the block-device this pages might be changed at any time. Because of the described issue the page-data wouldn't be copied as a call to sock_no_sendpage would do it. So the page-data might change at any time during or after the checksum-building. From my understanding this would just cause a crc-error anytime after sending and would only lead to a resent of the data, probably with a new checksum which will then be correct So there wouldn't be a obvious impact except some bad tcp packages. > Can you try to verify this a bug with some testing? Will do. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html
