Hi, We were struck by a similar problem as the one Brian had with his Squid-server. Hopefully I can add some more pointers as to what caused this. Sorry for this rather long message ;) I was helping a friend re-install his IRC-server using Linux 2.4.0 on a PII-300 Intel 440-BX chipset. The box has one eepro 100 NIC connected to the outside, and one 3COM 3c905 on an internal LAN. Kernel was compiled using gcc 2.95.2. Shortly after ircd was lanched and we started to see incoming connections the box started to spew out "kernel: reset_xmit_timer et.c et.c". We got about 20 of those. All network-related activity became jumpy and slow :( Doing 'netstat -n|wc -l' showed about 750 connections to the box. Now looking at the "caller=0x....." adress in System.map from the above message, it pointed to the inline function "tcp_ack_packets_out" defined in ipv4/tcp_input.c, this function is in turn called/inlined from function "tcp_clean_rtx_queue" also in tcp_input.c and here the adress matches the error message. I'm no kernelhacker so sorry if I'm pointing you in the wrong direction here! AFAIK, the box never oopsed but my ssh-session became slow and choppy, doing 'ls' would show about 15 lines, output would stop for maybe two seconds and then 10-20 more lines. Finally the box just froze. Yesterday we decided to try 2.4.1 and the results are the same, but with an additional: kernel: TCP: peer xxx.xxx.xxx.xxx:xxx/xx shrinks window xxxxxxxx. Bad, what else can I say? Sorry for all the x:s but I cant log on to the box since it died again :( Regards, Magnus Walldal - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
