Re: autofs stops working after a while with USE_MISC_DEVICE

[Ian Kent <raven@themaw.net>]

Thomas Chan wrote:
> 
> I am using autofs 5.0.4 with the latest patches up to 12-Jun-2009 with
> the 2.6.29.1 kernel.
> 
> I noticed that after a while (overnight) new automounts would no longer
> work.
> 
> An strace shows:
> rt_sigtimedwait(~[CHLD CONT RTMIN RT_1],
> 
> [pid 26289] 11:22:57 gettimeofday({1246558977, 61322}, NULL) = 0
> [pid 26289] 11:22:57 futex(0x7f6f5db43160, FUTEX_WAKE_PRIVATE, 1) = 0
> [pid 26289] 11:22:57 clock_gettime(CLOCK_REALTIME, {1246558977,
> 61421426}) = 0
> [pid 26289] 11:22:57 futex(0x7f6f5db43124, FUTEX_WAIT_PRIVATE, 150501,
> {0, 999900574}) = -1 ETIMEDOUT (Connection timed out)
> [pid 26289] 11:22:58 gettimeofday({1246558978, 61986}, NULL) = 0
> [pid 26289] 11:22:58 futex(0x7f6f5db43160, FUTEX_WAKE_PRIVATE, 1) = 0
> [pid 26289] 11:22:58 clock_gettime(CLOCK_REALTIME, {1246558978,
> 62086850}) = 0
> [pid 26289] 11:22:58 futex(0x7f6f5db43124, FUTEX_WAIT_PRIVATE, 150503,
> {0, 999899150}) = -1 ETIMEDOUT (Connection timed out)
> 
> lsof shows that automount has 10288 files open, of which most of these
> are like:
> 
> automount 15539      root *217u     sock                0,4 595374 can't
> identify protocol
> automount 15539      root *218u     sock                0,4 595376 can't
> identify protocol
> automount 15539      root *219u     sock                0,4 595378 can't
> identify protocol
> automount 15539      root *220u     sock                0,4 595380 can't
> identify protocol
> 
> Upon restarting autofs, these open files are cleared out and automount
> works, but the open files gradually build up again.
> 
> After setting USE_MISC_DEVICE=no, I no longer get the open file issue,
> but we are trying to remedy the autofs lazy unmount / cwd issue.

Mmm .. that's not good.
I'll have a look and see what I can see.

Ian