keychain locking problem (Was test -f occasionally fails on sym links)

[Karl M]

Hi All...

The reason I ask is that keychain uses

    if tl_error=`ln -s $$ "$lockf" 2>&1`; then

inside its takelock function as an atomic operation for creating a lock. It 
then uses

    if [ -f "$lockf" ]; then

to test for an old style lock file, and this sometimes fails (incorrectly 
succeeds) and generates an error message.

>From what I can find, this is expected to be an atomic operation and one of 
the ways programs do file locking.

I ran into this when I did high load testing for the ssh-add hang problem.

In preparation for a keychain-service release, I provide an example, where I 
only run keychain on the first launched (login) shell (which also has the 
advantage of speeding things up substantially). To do this, I launch 
keychain as follows

    if (umask 0377; : > ~/.keychain/${HOSTNAME}-keys) 2>/dev/null; then
      keychain --quiet ~/.ssh/identity ~/.ssh/id_dsa ~/.ssh/id_rsa
    fi

and the lock file is cleaned up by the keychain-service. This does seem to 
be safe (only verified by load testing under Cygwin). I found this method in 
the UNIX CD Bookshelf.

So...my questions are

1) Is this a safe method?

2) Is there a better method for me to use in Cygwin?

3) Should this problem be fed to the upstream keychain maintainer?

Thanks,

...Karl

>From: Corinna Vinschen
>Subject: Re: test -f occasionally fails on sym links (keychain related)
>Date: Tue, 29 Mar 2005 10:40:59 +0200
>
>On Mar 25 09:37, Karl M wrote:
> > Sometimes doing a [ -f foo ]; will show a false true while the symlink 
>is
> > being created. You can see this by opening two bash shells and executing
>
>Creating symlinks is not an atomic process in Cygwin.  It requires
>several OS calls to create a symlink.
>
>
>Corinna



--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/