Re: [patch 1/6] fs: icache RCU free inodes

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mardi 09 novembre 2010 à 08:02 -0800, Linus Torvalds a écrit :
> On Tue, Nov 9, 2010 at 4:46 AM, Nick Piggin <npiggin@kernel.dk> wrote:
> > So here is the inode RCU code. It's obviously not worth doing until the
> > actual rcu-walk path walking is in, but I'd like to get opinions on it.
> > It would be nice to merge it in Al's tree at some point, though.
> 
> Remind me why it wasn't sufficient to just use SLAB_DESTROY_BY_RCU?
> 
> Especially if we still lock things for the actual (few) inode list
> operations, the added complexity of actually freeing _individual_
> inodes by RCU seems to be a bad thing.
> 
> The only thing we care about is the pathname walk - there are no other
> inode operations that are common enough to worry about. And the only
> thing _that_ needs is the ability to look at the inode under RCU, and
> SLAB_DESTROY_BY_RCU should be entirely sufficient for that.
> 
> But we had some discussion about this long ago, and I may have
> forgotten some of the context.
> 

David Chinner sent a patch using SLAB_DESTROY_BY_RCU

You can see problems using this fancy thing :

- Need to use slab ctor() to not overwrite some sensitive fields of
reused inodes.
  (spinlock, next pointer)

- Fancy algo to detect an inode moved from one chain to another. Lookups
should be able to detect and restart their loop.

- After a match, need to get a stable reference on inode (lock), then
recheck the keys to make sure the target inode is the right one.

http://www.gossamer-threads.com/lists/linux/kernel/1298710

A bit tricky, but doable IMHO, especially if covering only the main
lookup. The seldom used can still get a (spin)lock.