jonab@sdcrdcf.UUCP (Jonathan Biggar) (05/07/86)
We have a problem with IP fragmentation-reassembly causing bad tcp checksums at the receiving end under Sun 3.0 UNIX. Any pointers to bug fixes related to this for SUN 3.0, SUN 2.0, or 4.2bsd would be appreciated. Jonathan Biggar sdcrdcf!jonab
shinoda@takuya.uucp (Yoichi Shinoda) (05/10/86)
I'm not sure if this helps, but....
When fragmented IP packets are routed with further fragmentation,
they are sent wrong with ip_off field of the first outgoing packet
corresponding to each orginal fragmented text reset to 0, causing
reassembly of the original packet to be impossible.
original fragmented further fragmented
udp packet. ip packet. ip packet.
--------- --------- ---------
| | |off=0 | |off=0 |
| | | | | |
| | | | ---------
| | | | ---------
| | | | |off=50 |
| | | | | |
| | --------- ---------
| | --------- ---------
| | |off=100| |off=0 | (should be
| | | | | | 100)
| | | | ---------
| | | | ---------
| | | | |off=50 | (should be
| | | | | | 150)
--------- --------- ---------
MTU~=~100 MTU~=~50
Note that the bug shows up only if interfaces of different MTU
are employed in the transmission.
-----------------------------------------------------------------------
*** ip_output.c Sat May 10 16:44:16 1986
--- ip_output.c.old Mon Oct 29 02:11:17 1984
***************
*** 155,161
mh->m_len = sizeof (struct ip) + olen;
} else
mh->m_len = sizeof (struct ip);
! mhip->ip_off = (off >> 3) + (ip->ip_off & ~(IP_MF|IP_DF)) ;
if (off + len >= ip->ip_len-hlen)
len = mhip->ip_len = ip->ip_len - hlen - off;
else {
--- 155,161 -----
mh->m_len = sizeof (struct ip) + olen;
} else
mh->m_len = sizeof (struct ip);
! mhip->ip_off = off >> 3;
if (off + len >= ip->ip_len-hlen)
len = mhip->ip_len = ip->ip_len - hlen - off;
else {
-----------------------------------------------------------------------
Yoichi Shinoda
Tokyo Inst. of Tech.
Dept. of CS
UUCP: ...{ihnp4!seismo!mcvax}!kddlab!koudai!shinoda
JUNET: shinoda@cs.titech.junethutch@sdcsvax.UUCP (Jim Hutchison) (05/12/86)
In article <913@takuya.uucp> shinoda@takuya.UUCP (Yoichi Shinoda) writes: > When fragmented IP packets are routed with further fragmentation, >they are sent wrong with ip_off field of the first outgoing packet >corresponding to each orginal fragmented text reset to 0, causing >reassembly of the original packet to be impossible. Does anyone know how this would affect the transmission of large datagrams (UDP) between a vax and a sun network (4.2sunII-to-4.2/3vax780)? -- /* Jim Hutchison UUCP: {dcdwest,ucbvax}!sdcsvax!hutch ARPA: Hutch@sdcsvax.ucsd.edu [ Disclaimer eaten by a passing kiwi ] */