Portability problems with flock on solaris
------------------------------------------
Key: QPID-1158
URL: https://issues.apache.org/jira/browse/QPID-1158
Project: Qpid
Issue Type: Bug
Reporter: Alan Conway
Assignee: Alan Conway
On Thu, 2008-06-26 at 12:12 +0200, Manuel Teira wrote:
Hello.
> After further investigation and tests, related with the change in
> r671604 to drop the file locking strategy in favour of a flock on the
> data dir.
>
> Trying to write a similar code, but using lockf, I hit the issue that
> the file must be opened using O_RDWR or O_RWONLY, and that's not allowed
> for a directory.
> The same happens trying to use a fcntl call.
> And unexpectedly, the same for flock. In the solaris manual page:
>
> <snip>
> Read permission is required on a file to obtain a shared
> lock, and write permission is required to obtain an
> exclusive lock.
> </snip>
>
> But the linux man page claims:
>
> <snip>
> A shared or exclusive lock can be placed on a file regardless of the
> mode in which the file was opened.
> </snip>
>
> I've searched the web for some BSD system pages, but they don't say
> anything about the file mode.
>
>
> On the other way, POSIX fcntl specification says, apropos the failure
> causes:
>
> [EBADF]
> The /fildes/ argument is not a valid open file descriptor, or the
> argument /cmd/ is F_SETLK or F_SETLKW, the type of lock, *l_type*,
> is a shared lock (F_RDLCK), and /fildes/ is not a valid file
> descriptor open for reading, or the type of lock *l_type*, is an
> exclusive lock (F_WRLCK), and /fildes/ is not a valid file
> descriptor open for writing.
>
> Posix specs also forces write permissions for lockf:
> http://www.opengroup.org/onlinepubs/007908799/xsh/lockf.html
>
>
>
> This leads to solaris not being able to lock directly on a directory,
> I'm afraid. Any idea?
>
Yes, we can create (if it doesn't already exist) a lock file in the directory
and then use lockf to lock it. There's already code in Daemon.cpp that does
exactly this for the PID file. The reason I switched to flock was because
crashing or killed brokers were sometimes leaving the lock file behind them,
whereas a flock (or lockf) lock is automatically released when the process
exits.
We need to
- create a qpid::sys::LockFile class that can be re-implemented on different
platforms.
- use the Daemon.cpp code as the posix implementation.
- Replace the locking code in Daemon.cpp and DataDir.cpp with the common
sys::LockFile.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
