svn commit: r265691 - head/sys/netinet

Gleb Smirnoff glebius at
Thu May 8 20:23:13 UTC 2014

On Thu, May 08, 2014 at 05:27:46PM +0000, Michael Tuexen wrote:
M> Author: tuexen
M> Date: Thu May  8 17:27:46 2014
M> New Revision: 265691
M> Log:
M>   For some UDP packets (for example with 200 byte payload) and IP options,
M>   the IP header and the UDP header are not in the same mbuf.
M>   Add code to in_delayed_cksum() to deal with this case.
M>   MFC after: 3 days
M> Modified:
M>   head/sys/netinet/ip_output.c
M> Modified: head/sys/netinet/ip_output.c
M> ==============================================================================
M> --- head/sys/netinet/ip_output.c	Thu May  8 17:20:45 2014	(r265690)
M> +++ head/sys/netinet/ip_output.c	Thu May  8 17:27:46 2014	(r265691)
M> @@ -887,15 +887,23 @@ in_delayed_cksum(struct mbuf *m)
M>  		csum = 0xffff;
M>  	offset += m->m_pkthdr.csum_data;	/* checksum offset */
M> +	/* find the mbuf in the chain where the checksum starts*/
M> +	while ((m != NULL) && (offset >= m->m_len)) {
M> +		offset -= m->m_len;
M> +		m = m->m_next;
M> +	}
M> +	if (m == NULL) {
M> +		/* This should not happen. */
M> +		printf("in_delayed_cksum(): checksum outside mbuf chain.\n");
M> +		return;
M> +	}
M>  	if (offset + sizeof(u_short) > m->m_len) {
M> -		printf("delayed m_pullup, m->len: %d  off: %d  p: %d\n",
M> -		    m->m_len, offset, ip->ip_p);
M>  		/*
M>  		 * XXX
M> -		 * this shouldn't happen, but if it does, the
M> -		 * correct behavior may be to insert the checksum
M> -		 * in the appropriate next mbuf in the chain.
M> +		 * This should not happen, but if it does, it might make more
M> +		 * sense to fix the caller than to add code to split it here.
M>  		 */
M> +		printf("in_delayed_cksum(): checksum split between mbufs.\n");
M>  		return;
M>  	}
M>  	*(u_short *)(m->m_data + offset) = csum;

If "this should not happen" really means that we do not expect this issue
at all, assuming we are coding correctly, then all these comments and printfs
should be converted to KASSERTs.

If "this should not happen" means 'such packets shouldn't arrrive from
network", then we need to remove printfs, because having a remotely
triggerable printf(9) is a DDoS vulnerability.

Totus tuus, Glebius.

More information about the svn-src-head mailing list