File Locking

Nitzan Shaked · *nix forums beginner Joined: 27 Jun 2005 Posts: 8

(Hope this is not considered off-topic)

I wish to perform a seemingly simple task, but all the FAQ reading and
example browsing don't seem to offer a solution.

I wish to create mutual exclusion on a resource I have which several
processes might want to change, let's assume for now that all are on
the same machine.

I do *not* wish to leave my lock files hanging around the file-system
any longer than necessary (I have hundreds of thousands, if not more,
such resouces).

The "text book" solution might be:
1) fd = open( "my.lock", O_RDONLY | O_CREAT );
2) flock( fd, LOCK_EX );
3) do_something_with_the_resource();
4) unlink( "my.lock" );
5) close( fd );

This FAILS for a subtle reason: after having done unlink(), other
processes trying to run the above code will create a new file by the
same new and will lock *that* file (which has a different file-table
entry, of course). ANOTHER process, which trying to get flock() before
the unlink, will also succeed right after the close(), so now 2
processes think they have exclusive lock.

Removing the unlink() call solves this, but leaves my lock files in the
file-system for all eternity.

SysV IPC, multiple lock files, etc -- all don't seem to offer solace :)

What's the right way to do this?

tia,
Nitzan

Barry Margolin · *nix forums Guru Joined: 24 Feb 2005 Posts: 323

In article <1120292556.251595.173690@o13g2000cwo.googlegroups.com>,
nitzan.shaked@gmail.com wrote:

Nitzan Shaked · *nix forums beginner Joined: 27 Jun 2005 Posts: 8

Thanks for the reply, but this simply causes another race: if I close()
first, then I am releasing the lock. By the time I will unlink()
another process has already got the lock (and long since has opened the
file, btw, since this is done prior to locking) and I will be
unlink()-ing its file.

Barry Margolin · *nix forums Guru Joined: 24 Feb 2005 Posts: 323

In article <1120333718.686226.150700@o13g2000cwo.googlegroups.com>,
"Nitzan Shaked" <nitzan.shaked@gmail.com> wrote:

Andrew Smallshaw · *nix forums beginner Joined: 23 Feb 2005 Posts: 26

On 2005-07-02, Nitzan Shaked <nitzan.shaked@gmail.com> wrote:

Nitzan Shaked · *nix forums beginner Joined: 27 Jun 2005 Posts: 8

I've already almost "given in" to the not-feeling-bad strategy,
though I think that this morning on my way to work I've found a
solution. If it works out I'll post it.
In any case, I have hundreds of thousands of files, if not more. At the
very least this is a small pinch in my tender heart Smile

Nitzan Shaked · *nix forums beginner Joined: 27 Jun 2005 Posts: 8

As I said before, O_EXCL is not good because I am using the file for
actually locking something else. Thus I want to blocked (sleep while
waiting), and O_EXCL does not give me that. It will force polling.

Andrew Smallshaw · *nix forums beginner Joined: 23 Feb 2005 Posts: 26

On 2005-07-03, Nitzan Shaked <nitzan.shaked@gmail.com> wrote:

venku · *nix forums beginner Joined: 07 Jul 2005 Posts: 5

Hi,
I feel that on the following steps ..
1) after having done unlink()
2) other processes trying to run the above code will create a new file
by the same new and will lock *that* file (which has a different
file-table entry, of course)
3)ANOTHER process, which trying to get flock() before the unlink, will
also succeed right after the close(), so now 2 processes think they
have exclusive lock

2 & 3 would not happen both at same time because
if 3) i.e if another process waiting (which should increment file ref
count of my.lock), there no chance of having new file creation which
is 2.
Am i misunderstood ???

Google