Re: fcntl method for file_operations

[Jamie Lokier <jamie@shareable.org>]

Matthew Wilcox wrote:
> perhaps the right thing to do is just to add the following ops to
> file_operations:
> 
> 	int (*fcntl_setfl)(int fd, struct file *filp, unsigned long arg);
> 	int (*fcntl_getlease)(struct file *filp);
> 	int (*fcntl_setlease)(int fd, struct file *filp, unsigned long arg);
> 	int (*fcntl_dirnotify(int fd, struct file *filp, unsigned long arg);
> 
> comments?

I agree that a generic fcntl() doesn't make a lot of sense.

However, rather than your suggestion, I'd rather
getlease/setlease/dirnotify have more appropriate structure for those
operations, be called from a generic lease/dnotify core, and be added
as and when they are actually used by a filesystem.

Otherwise I can easily see NFS dirnotify being implemented, and
dnotify.c being enhanced to support some feature like non-signal
results (DN_NOSIGNAL), ->poll() waiting, last writer close, or
whatever, and those new features fine *except* on NFS filesystems
which has it's own signal delivery code etc. copied from dnotify.c.

In other words, for dnotify, the generic dnotify code should always be
used so that the API is consistent among filesystems, but it should
call the filesystem to do the filesystem-specific part namely
registering a callback function to be called when a non-local change
is detected.

Similarly for leases, and any other wonders that are added in future
like mount notifications.

-- Jamie