On Sun, Sep 22, 2013 at 03:28:58AM +0000, David Holland wrote: > So, here's some thinking out loud on the subject of fixing it. > : > (I also have no real idea yet how to get to where I'm describing from > where we are in a decently incremental fashion.)
I think the first step forward on that is to move the ufs_icache code to kern/ and generalize it where needed (not much I think) so other fses can use it, then convert the other fses one by one. Another model entirely, on the other hand, is to get rid of the vfs-level vnode cache and make each fs fully responsible for its own vnodes. This can also get rid of a lot of the mess; the problem then is that there's no obvious way to control how many inodes each fs is allowed to keep live. (And also it promotes C&P code, replicated bugs, and so on, which is why I don't think it's a good idea.) -- David A. Holland [email protected]
