First I thought that this was because of my hardware timer problem,
as the problems seemed to go away after patching kernel to kluge around that.
The real reason for was that I tried ftp:ing to a host as slow as mine, and
the problem doesn't occur with those, only ones with faster net
connection. (weird huh?) Or that what it seems to be, of course it might
be random, I haven't really tested with many machines.
When I strace the ncftp process, I see that it writes the data and begins
read():ing, and never receives anything back. The socket is in FIN_WAIT1
state, so I think that the data is never delivered. I can't check for
sure because my tcpdump says "tcpdump: can't resolve symbol 'ether_ntohost'"
:)
I have tried the pre-patch-2.0.31 (actually I'm running it now), but it
doesn't help.
Relevant info from ifconfig (after testing, no errors as you see):
ppp0 Link encap:Point-Point Protocol
inet addr:195.74.5.136 P-t-P:195.74.1.18 Mask:255.255.255.0
UP POINTOPOINT RUNNING MTU:1500 Metric:1
RX packets:25 errors:0 dropped:0 overruns:0
TX packets:51 errors:0 dropped:0 overruns:0
And pppd:
May 11 00:16:35 fishy pppd[2011]: pppd 2.2.0 started by root, uid 0
May 11 00:17:08 fishy pppd[2011]: Serial connection established.
May 11 00:17:09 fishy pppd[2011]: Using interface ppp0
May 11 00:17:09 fishy pppd[2011]: Connect: ppp0 <--> /dev/modem
May 11 00:17:12 fishy pppd[2011]: local IP address 195.74.5.136
May 11 00:17:12 fishy pppd[2011]: remote IP address 195.74.1.18
fishy:/home/raulir>rpm -qa|grep ppp
ppp-2.2.0f-3
fishy:/home/raulir>
The only thing I (_definitely_ not a kernel expert) can think for the
reason is broken nagle algorithm or TCP/IP stack..