On Tuesday 02 December 2008 21:16:13 Daniel Phillips wrote: > I only had to make one hack to core vfs to make this work. Essentially, > we are changing the dentry cache from a volatile, read only cache to a > writeback cache which is entirely consistent on its own, as opposed to > just being an accelerator. We always knew this was possible, because > that is how RAMFS works. But I needed to make it work when the cache > is backed by a filesystem. All I had to do here is add one new method, > to prevent the dentry cache from forcibly removing a dentry from the > cache that has been unlinked, but which is still in used for some hard > to understand reason. With my new hook (d_ops->hide) the dentry just > stays in cache as a negative dentry and we take a reference count on it > to make sure it does not disappear before our code has a chance to find > it and remove the underlying directory entry.
Hi Daniel, While I'm certainly not qualified to comment on the _content_ of the patch, I have been watching Linux filesystem flamewars for a while... I _strongly_ suggest that you write to [EMAIL PROTECTED] about these VFS changes, and sooner rather than later. (It probably would be a good idea to directly CC at least Al Viro and Christoph Hellwig; Ted Ts'o, Christoph Lameter, Linus, and Andrew Morton are also good candidates.) Tux3 looks very promising, and I would hate to see it get tied up in avoidable VFS battles when it's otherwise ready for merging. Thanks, -- BKS _______________________________________________ Tux3 mailing list [email protected] http://tux3.org/cgi-bin/mailman/listinfo/tux3
