Hi Marcus, I tagged the bug moreinfo because I suspected HAL was causing the timeout mechanism to fail, and Roman hadn't made a final statement whether stopping HAL fixed the problem for him.
[indexing your links for easy backreference ;) ] > But the original problem has been reported elsewhere: > [1] http://www.mail-archive.com/autofs%40linux.kernel.org/msg03967.html > > There is a patch floating around: > [2] > http://www.kernel.org/pub/linux/daemons/autofs/v4/old/autofs-4.1.3-signal-race-fix.patch > > I noticed that only the first half of it was applied in the current version. > Perhaps the other half should be applied as well? The first part of the patch at [2] is fine - block all relevant signals before modifying the state. But the second half exactly contradicts Ian's patch at [1]... and IMO [1] is the correct order. sigprocmask(... &ready_sigs ...) is actually unblocking SIGCHLD here, which should be done _after_ ap.state has been changed. Could you give a more detailed description of your bug? What version(s) do you use? How do you see those periodic accesses - local log file or network sniffer? Is there some kind of pattern? Regards, Jan
signature.asc
Description: Digital signature
