IP packets lost between ip_output and driver's xmit handler

IP packets lost between ip_output and driver's xmit handler

Post by ore » Fri, 08 Mar 2002 09:41:13



User apps user select and send/sendto on UDP socket. It works for a
while, afterwards select/sendto pair gives infinite rate of outgoing
udp traffic.

I tested some counters to localize buggy area and currently I stick
with area between ip_output (ip_output.c) and my own device's xmit
handler.

Counter IpOutRequests machs with my apps counter and amount of packets
which came to xmit function significally smaller.

Does anyone have a glue, and fixed this bug already.

 
 
 

IP packets lost between ip_output and driver's xmit handler

Post by ore » Fri, 08 Mar 2002 23:52:46



> User apps user select and send/sendto on UDP socket. It works for a
> while, afterwards select/sendto pair gives infinite rate of outgoing
> udp traffic.

> I tested some counters to localize buggy area and currently I stick
> with area between ip_output (ip_output.c) and my own device's xmit
> handler.

> Counter IpOutRequests machs with my apps counter and amount of packets
> which came to xmit function significally smaller.

> Does anyone have a glue, and fixed this bug already.

Additional information: when this application works, the bandwidth
over ethernet device close to maximum interface's ability. I beleive
that maybe because special packets (I beleive ARP) coulnd't went out
of interface and comeback again in defined somethere time, IP stack
blocks interface as dead one, at this time it gives infinite traffic
for applications, packets are dropped, interface isn't busy anymore,
ARPing resumes, IP stack switch state of the link bank to normal,
application obtain back his legs as select(2) and send(2).

 
 
 

1. ? attempt to xmit but no xmit i/o buffer

I'm getting a message in /var/adm/messages:
        "attempt to xmit but no xmit i/o buffers"
Environment:
        server: sparc 10 96mb, solaris 2.3 - standard patches,
                sun's fddi, nfs
        client: sun ipx, ethernet, nfs mounted fs from the server,
                running an intense read test (100MB/10Krecs).
I'm testing concurrent reads from 12 ipx clients attached to the
fddi ring via 12 ethernets. the clients start reading and the
messages come out in the servers message file... and all slows
to a crawl.

Any ideas on what parameter(s) to change to allow more xmit buffers?

2. windows nt network

3. token ring message - tr0: xmit ret_code: 22 xmit error code: 00

4. c control character question

5. Token Ring error xmit ret_code: 22 xmit error code: 88

6. ptrace

7. bf0: attempt to xmit but no xmit i/o

8. Apache content type negotiation

9. incoming IP packet with local interface's IP as origin?

10. Handlers, Handlers, Handlers

11. ISDN: "isdn, xmit: Packet blocked: 0, 0"

12. 'xmit on stopped card' problem

13. dynamic IP, first TCP-packet lost