Mike's questions

I'm writing an internet application and I'm at the stage of crafting my own TCP packets. Except there is a major problem. According to the TCPDUMP utility in linux, every tcp packet that I send out from any local application (like telnet and curl) to any IPV4 address on the network causes TCPDUMP to report checksum errors.

A few sources on the internet claim it is due to checksum offloading being turned on. I tried various network cards and used ethtool -K to disable offloading on each card, and no matter what I try, I still receive the checksum error.

I even went as far as hooking the local computer directly to the target computer with a crossover cable. This means I bypassed routers and wireless connections.

Is there a linux setting I can adjust to fix the checksum issue or a program better than TCPDUMP that can work with all unix systems that can truthfully tell me if the outgoing packets have the correct checksum? because surely not every linux application would be crafting faulty TCP checksums or I would have a problem accessing this site.

Any ideas?

I'm having a strange issue. In linux I have two windows open. one running tcpdump and one running curl. I used curl to try to make a connection to local host and tcpdump logs invalid checksum errors. This is the log from the tcpdump window:

bash-4.2# tcpdump -evvvt -i lo
tcpdump: listening on lo, link-type EN10MB (Ethernet), capture size 65535 bytes
00:00:00:00:00:00 (oui Ethernet) > 00:00:00:00:00:00 (oui Ethernet), ethertype
IPv4 (0x0800), length 74: (tos 0x0, ttl 64, id 18215, offset 0, flags [DF], 
proto TCP (6), length 60)
localhost.37550 > localhost.http: Flags [S], cksum 0xfe30 (incorrect -> 0x5d4b), 
seq 112438199, win 32792, options [mss 16396,sackOK,TS val 4648080 ecr 0,nop,wscale 9], length 0
00:00:00:00:00:00 (oui Ethernet) > 00:00:00:00:00:00 (oui Ethernet), ethertype 
IPv4 (0x0800), length 54: (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto TCP (6), length 40)
localhost.http > localhost.37550: Flags [R.], cksum 0x6c64 (correct), seq 0, ack 112438200, win 0, length 0

And the curl window offers nothing exciting to see. If I used max verbosity, all I get is connecting to 127.0.0.1. then it immediately terminates with couldn't connect to host.

I then use ethtool -k to see what parameters I can adjust to disable offloading since many people claim my problem is because of offloading. These are the parameters after the adjustments I could make:

Offload parameters for lo:                                                                      
rx-checksumming: on                                                                             
tx-checksumming: on                                                                             
scatter-gather: on                                                                              
tcp-segmentation-offload: off                                                                   
udp-fragmentation-offload: off                                                                  
generic-segmentation-offload: off                                                               
generic-receive-offload: off                                                                    
large-receive-offload: off
rx-vlan-offload: off
tx-vlan-offload: off
ntuple-filters: off
receive-hashing: off

I tried turning every setting off, but the ones I can't turn off, I get errors as follows:

bash-4.2# ethtool -K lo rx off
Cannot set device rx csum settings: Operation not supported
bash-4.2# ethtool -K lo tx off
Cannot set device tx csum settings: Operation not supported
bash-4.2# ethtool -K lo sg off
Cannot set device scatter-gather settings: Operation not supported

And the device lo is the local host interface.

How do I fix this so that when I use tcpdump or any other tcp packet tool that I never get checksum errors from utilities that successfully connected to other IP addresses with TCP?