[Xorp-cvs] XORP cvs commit: xorp/fea

[Pavlin Radoslavov]

> Pavlin Radoslavov wrote:
> > 	  The reason for this change is that in case of Linux the IP_HDRINCL
> > 	  IP packets are not fragmented and are limited to the interface MTU.
> > 	  The raw(7) Linux manual is wrong by saying it is a limitation only
> > 	  in Linux 2.2. It is a limitation in 2.4 and 2.6, and there is no
> > 	  indication this is going to be fixed in the future.
> 
> This sentence will be gone in next manpages version. In case of such 
> trivial and obvious bugs in manpages I recommend to report it to manpages 
> maintainer - "Michael Kerrisk" <mtk-manpages@gmx.net>. I already reported 
> this problem myself.
> 
> Linux have been quite long time without manpages maintainer and there is a 
> lot of such stupid things. This particular manpage was just probably 
> written when 2.4 didn't exist yet and this sentence in BUGS section made 
> sense.

Thank you for the info and for sending the correction.

Myself, I'd prefer to have the Linux kernel being fixed so it does
the fragmentation for you even for IP_HDRINCL packets.
The argument I presume I will get that "this is a raw IP packet so
the kernel shouldn't touch it or fragment it" is not really valid,
because the current behavior (getting error messages when the packet
is larger than the MTU) is useful only to discover the MTU and even
this is not practical. Furthermore, the kernel actually DOES
touch some of the IP_HDRINCL IP header as indicated (see raw(7)).
In other words, the Linux IP_HDRINCL support is useful only if you
are sending packets smaller than the MTU. Otherwise, you have to use
a number of system calls to achieve similar behavior.

Anyway, this is not the right email list for such discussion.
If I want to bring it to the attention of the Linux folks, do you
know which is the right email address?
The archive for the old address for Linux network development
"netdev@oss.sgi.com" seems to be spammed (see
http://oss.sgi.com/archives/netdev/), so I don't know whether this
address is still valid. Other alternative addresses I came across
are linux-net@vger.kernel.org and netdev@vger.kernel.org

If you agree with my arguments above and if you are already
subscriber to the appropriate list, feel free to raise the issue
there :)

Pavlin