From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: From: David Presotto To: 9fans@cse.psu.edu MIME-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Subject: [9fans] bugs, races, more bugs Date: Mon, 17 Mar 2003 21:33:59 -0500 Topicbox-Message-UUID: 81598526-eacb-11e9-9e20-41e7f4b1d025 Timing out arp entries uncovered some bugs that we hadn't noticed before plus I added one with the timeouts. The symptom was quiescent connections timing out. Fixed in ip/arp.c. We also found a few races between packets being rerouted and IP stacks being torn down. The symptom was a null dereference in ip.c crashing the kernel. Fixed in ip/ip.c, ip/ipifc.c, ip/iproute.c, ip/ipv6.c. I left an assert in ip.c to make the race more obvious in case we haven't stomped on it yet. Finally, I updated port/proc.c and port/devproc.c with suggestions on scheduler changes from the list. Processes trapped at low priority levels now float up. Also, sleep(0) now drops a process' priority one level each call so that processes doing something like: while(condition not true) sleep(0); should dip down below whatever its waiting for. I've been testing it on all our systems for a few days and haven't seen a problem. However, I never trust this stuff till a wider audience tries it. Could people that didn't do well with the last version try this out soon?