xfs
[Top] [All Lists]

Re: [PATCH 0/5] Per-superblock shrinkers

To: Dave Chinner <david@xxxxxxxxxxxxx>
Subject: Re: [PATCH 0/5] Per-superblock shrinkers
From: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
Date: Sat, 15 May 2010 02:30:05 +0100
Cc: linux-kernel@xxxxxxxxxxxxxxx, xfs@xxxxxxxxxxx, linux-fsdevel@xxxxxxxxxxxxxxx, linux-mm@xxxxxxxxx
In-reply-to: <1273821863-29524-1-git-send-email-david@xxxxxxxxxxxxx>
References: <1273821863-29524-1-git-send-email-david@xxxxxxxxxxxxx>
Sender: Al Viro <viro@xxxxxxxxxxxxxxxx>
User-agent: Mutt/1.5.20 (2009-08-17)
On Fri, May 14, 2010 at 05:24:18PM +1000, Dave Chinner wrote:
> 
> This series reworks the filesystem shrinkers. We currently have a
> set of issues with the current filesystem shrinkers:
> 
>       1. There is an dependency between dentry and inode cache
>          shrinking that is only implicitly defined by the order of
>          shrinker registration.
>       2. The shrinkers need to walk the superblock list and pin
>          the superblock to avoid unmount races with the sb going
>          away.
>       3. The dentry cache uses per-superblock LRUs and proportions
>          reclaim between all the superblocks which means we are
>          doing breadth based reclaim. This means we touch every
>          superblock for every shrinker call, and may only reclaim
>          a single dentry at a time from a given superblock.
>       4. The inode cache has a global LRU, so it has different
>          reclaim patterns to the dentry cache, despite the fact
>          that the dentry cache is generally the only thing that
>          pins inodes in memory.
>       5. Filesystems need to register their own shrinkers for
>          caches and can't co-ordinate them with the dentry and
>          inode cache shrinkers.

NAK in that form; sb refcounting and iterators had been reworked for .34,
so at least it needs rediff on top of that.  What's more, it's very
obviously broken wrt locking - you are unregistering a shrinker
from __put_super().  I.e. grab rwsem exclusively under a spinlock.

Essentially, you've turned dropping a _passive_ reference to superblock
(currently an operation safe in any context) into an operation allowed
only when no fs or vm locks are held by caller.  Not going to work...

<Prev in Thread] Current Thread [Next in Thread>