Problems with network fragments under 2.1

Ben Clifford (benc@dass.prestel.co.uk)
Tue, 5 Aug 1997 18:26:29 +0000 (GMT)


For the past few versions of 2.1, I've had a problem with running tin to
my news-server (leafnode), running on the same machine.

It will hang and eventually display the message:

tcp_write_wakeup: fragment failed

I noticed that the "Total failed network buffer allocs" counter displayed
by Alt-Sysrq-M goes up by a few thousand every second while this is
happening.

I can still change VTs and can type text on consoles, such as syslogd's,
which are echoed by the kernel. However, all userland stuff stops working.

If I kill the news server with Alt-Sysrq-N (sends SIGTERM to last
process), then it all gets better, until I try to run tin again.

This appears to occur when there are 0 large free pages (eg 128k, 64k) as
shown by SysRq-M.

When there are some large free pages left, tin works fine.

--
PGP 0xD4C9C0B1 http://www.geocities.com/SiliconValley/Lakes/7456/
1016kb of spam & junk filtered; Linux 2.1.48 on dass up for 2 hours