I've been working on a tool for finding potential data races and I tried it out on Tomcat. I found a couple of things that look like problems; you might be interested in fixing them, and I would like to know if my results are correct or not.
The first one is minor: org.apache.catalina.connector.http.HttpConnector.start() checks curProcessors without holding the "processors" lock. It looks like this could lead to maxProcessors+1 processors being created if it creates a processor just as a connection arrives on the HttpConnector thread. Perhaps more serious: org.apache.catalina.core.StandardWrapper increments countAllocated in a few places without holding any locks. This could lead to an increment being lost if multiple threads try to increment it simultaneously. Then the unload() method might leave its loop early because countAllocated is smaller than it should be. Conversely, unlocked decrements of countAllocated might be lost, and then the unload() method would spin forever. I'd love to get some feedback on whether my understanding of this code is correct or not. Thanks, Robert O'Callahan IBM TJ Watson Research Staff Member Office: Hawthorne 1N-J20 Phone: (914) 784-7911 "The first to present his case seems right, till another comes forward and questions him." -- Proverbs 18:17 -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
