For cluster and distributed file systems we actually need a somewhat
different approach to locking. The best thing to have would be for the FS
to know that a lock will be needed and for what operation at lookup time.
Al and I have come up with a scheme where we want to pass a data structure
to lookup/dentry_revalidate that indicates to the FS, e.g., we are looking
at this directory dentry with the intent of creating a new file in it.
That allows the FS to acquire the locks at lookup time and the VFS
semantic verifications (such as for O_EXCL) to proceed normally. When the
lock is released, the FS needs to be told.
The nameidata Al introduced a few months ago might be exactly what we
need, together with some reorganization of parameters and a lock release
method.
- Peter -
-
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to [EMAIL PROTECTED]