All processes opening UDP sockets in order to send data without prior binding to a speciffic address get the same port number. It goes on until an error occures during packet reception. After that, an ICMP message "port xxx unreachable" is send, the random port number is incremeted by 1, and the situation starts from the beginning. I am not sure if it is a bug, or an implementation or design feature. However, servers that receive these ICMP packets treat them as communication errors an fill error logs with "connection refused" messages.
BTW, kernel version numbers I have tesed are 2.2.16-3 and 2.2.16-4.lfs
O.K, so maybe there is a bug in an application. The application that suffers is "kprop" found in "krb5-server-1.1.1-21" componnet. It opens a random UDP port and sends a packet to a "kpropd" daemon listening on the well-known "krb5_prop" port on another machine. It usualy happens that the answer received from that machine is rejected, and ICMP mesage "port xxx unreachable" is sent. After that, "kprop" tries again, this time selecting a random port that's number is higher by 1 than before, and it succeedes. However, netowork badwidth is unnecessarily consumed, as well as disk space on the machine where kpropd error messages are logged. Could you please reopen this bug, changing the component name to "krb5-server"? Janusz