In <4.2.0.58.20000114201627.009852a0@pop3.concentric.net> Stephen Satchell (satch@concentric.net) wrote:
> At 04:09 AM 1/14/00 , you wrote:
>>nevertheless, it doesn't look like linux always sends fragments in reverse
>>order, but i'm still trying to familiarize myself with the code. i know
>>that fragmented udp is always sent in reverse order, and i suspected that
>>all fragments were sent in revers order but i couldn't verify that.
> When I was writing "Linux IP Stacks Commentary" for CoriolisOpen books, I
> discovered that in 2.0.34 the kernel TCP/IP code was inconsistent about
> whether to write fragments in reverse order or forward order.
> In the "every other fortnight" department, if you run out of memory during
> a fragmentation procedure, the fragmentation process...stops. You end up
> with orphan fragments out on the net. Once 2.4 is out, I'm going to
> propose some patches to take care of this condition (assuming that someone
> else hasn't beat me to it). The proposed change would be to save the
> fragments until the skbuffs have been built, THEN insert the fragments onto
> the device output queue. That way, if you run out of memory you don't send
> anything at all, and you have less dead traffic.
IMO it's bad change. Since it will make situation slighly better in uncommon
case while slowing down common case. Since it's legal to send orphan fragments
out on the net (IP is NOT reliable anyway) it looks REALLY wrong.
> In the same vein, I would also propose that some SMTP and/or /proc counters
> be implemented to count the total number of memory-out conditions
> encountered by the networking software; right now it happens silently so
> there is no way for the sysadmin to be aware there are any problems. (I
> prefer that to generating a syslog message because the admin can write his
> own userland stuff to track any memory problems.)
This can be good change if it's possible to do without affecting speed in
normal situation (I hope it's possible to do).
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Jan 15 2000 - 21:00:26 EST