From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Layton Date: Sun, 24 Aug 2014 06:08:01 -0400 Subject: [Cluster-devel] [PATCH 06/10] locks: plumb an "aux" pointer into the setlease routines In-Reply-To: <20140824013305.GB21609@infradead.org> References: <1408804878-1331-1-git-send-email-jlayton@primarydata.com> <1408804878-1331-7-git-send-email-jlayton@primarydata.com> <20140824013305.GB21609@infradead.org> Message-ID: <20140824060801.5402880c@synchrony.poochiereds.net> List-Id: To: cluster-devel.redhat.com MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit On Sat, 23 Aug 2014 18:33:05 -0700 Christoph Hellwig wrote: > On Sat, Aug 23, 2014 at 10:41:14AM -0400, Jeff Layton wrote: > > In later patches, we're going to add a new lock_manager_operation to > > finish setting up the lease while still holding the i_lock. To do > > this, we'll need to pass a little bit of info in the fcntl setlease > > case (primarily an fasync structure). Plumb the extra pointer into > > there in advance of that. > > > > We declare this pointer as a void ** to make it clear that this is > > auxillary info, and that the caller isn't required to set this unless > > the lm_setup specifically requires it. > > Can you just return -EEXIST if reusing an existing one and make it a > normal private pointer a we use elsewhere? > That sounds a little confusing... We have two pointers we pass down to generic_setlease: the file_lock itself and with this patch, the "aux" pointer. We can end up using either, neither or both during a call to generic_setlease. A simple error code can't properly indicate which of the two pointers got used. It might be clearer to turn the file_lock into a normal pointer and return -EEXIST if we reused it, but leave aux as a double pointer. > Btw, two things I came up with to make the lease filesystem API nicer: > > - bypass vfs_setlease/->setlease for deletes and just call directly > into the lease code. no one does anything special for it (except > cifs, which forgets about it), and then rename the method to > ->add_lease. > - provide a no_add_lease for nfs/gfs to centralize the no-op version > in a single place (similar no no_lseek). > > Otherwise this looks really nice. Both of those sound good. I'll plan to roll those changes in with the next iteration. Thanks! -- Jeff Layton