From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.3.1) Gecko/20030618 Debian/1.3.1-3 Description of problem: clntudp_call passes the udp retransmit in to a call to select directly, which causes it to get overwritten with {0, 0}. This breaks the time_waited semantics and causes the main retransmit loop to spin fast, causing a udp packet storm on the network. This is seen in Anaconda when the first MOUNTPROC_MNT rpc call is made to the NFS server as mentioned in bug 103952 's attachment 'detailed description and patch'. I've created a patch against dietlibc 0.19 (should apply cleanly even against today's release of 0.23). I have just sent this upstream, but have not yet received a reply. Version-Release number of selected component (if applicable): 0.21-4 How reproducible: Sometimes Steps to Reproduce: 1. Make a udp rpc call 2. Drop the response Actual Results: The client will storm the network with RPC CALLS as it does'nt what for a response (timeout = near nil to the select call). Expected Results: Retransmits should occur as set in clntudp_create or as set in clntudp_control. Total timeout should work as defined in the call to clntudp_call. Additional info:
Created attachment 94467 [details] patch
Added for -7.