From: "J. Bruce Fields" <bfields@fieldses.org>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: "J. Bruce Fields" <bfields@redhat.com>,
linux-nfs@vger.kernel.org, linux-fsdevel@vger.kernel.org,
Christoph Hellwig <hch@infradead.org>
Subject: Re: [PATCH 4/4] dcache: don't clear DCACHE_DISCONNECTED too early
Date: Fri, 11 Oct 2013 12:02:37 -0400 [thread overview]
Message-ID: <20131011160237.GA21258@fieldses.org> (raw)
In-Reply-To: <20130909204655.GA10599@fieldses.org>
On Mon, Sep 09, 2013 at 04:46:55PM -0400, J. Bruce Fields wrote:
> On Mon, Sep 09, 2013 at 01:46:47AM +0100, Al Viro wrote:
> > On Sat, Sep 07, 2013 at 02:46:01PM -0400, J. Bruce Fields wrote:
> > > From: "J. Bruce Fields" <bfields@redhat.com>
> > >
> > > DCACHE_DISCONNECTED should not be cleared until we're sure the dentry is
> > > connected all the way up to the root of the filesystem. It *shouldn't*
> > > be cleared as soon as the dentry is connected to a parent. That will
> > > cause bugs at least on exportable filesystems.
> >
> > Then you probably want this
> > if (!IS_ROOT(pd)) {
> > /* must have found a connected parent - great */
> > spin_lock(&pd->d_lock);
> > pd->d_flags &= ~DCACHE_DISCONNECTED;
> > spin_unlock(&pd->d_lock);
> > noprogress = 0;
> > to go through all intermediates, clearing DCACHE_DISCONNECTED on all of
> > them; O(depth^2) can suck when we have a long chain of directories...
>
> I'm not sure what you mean--something like the following?
>
> Not having seen complaints about filehandle lookup performance I'm
> inclined to leave it alone, though.
But just for fun--I did some cleanup and fixed some other quadratic
behavior here and can notice a difference on lookups of very deep
subdirectories.
For example I'm seeing an uncached lookup of an 8000-deep directory
taking about 6 seconds, and can get that down to a tenth of a second.
I'm not sure yet if the difference on less extreme examples is really
significant, I need to experiment some more.
I'll do some more review and post patches and results.
--b.
> diff --git a/fs/exportfs/expfs.c b/fs/exportfs/expfs.c
> index 5816e19..72ed705 100644
> --- a/fs/exportfs/expfs.c
> +++ b/fs/exportfs/expfs.c
> @@ -90,6 +90,22 @@ find_disconnected_root(struct dentry *dentry)
> return dentry;
> }
>
> +static void clear_disconnected(struct dentry *dentry)
> +{
> + dget(dentry);
> + while (dentry->d_flags & DCACHE_DISCONNECTED) {
> + struct dentry *parent = dget_parent(dentry);
> +
> + spin_lock(&dentry->d_lock);
> + dentry->d_flags &= ~DCACHE_DISCONNECTED;
> + spin_unlock(&dentry->d_lock);
> +
> + dput(dentry);
> + dentry = parent;
> + }
> + dput(dentry);
> +}
> +
> /*
> * Make sure target_dir is fully connected to the dentry tree.
> *
> @@ -114,15 +130,11 @@ reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
>
> if (!IS_ROOT(pd)) {
> /* must have found a connected parent - great */
> - spin_lock(&pd->d_lock);
> - pd->d_flags &= ~DCACHE_DISCONNECTED;
> - spin_unlock(&pd->d_lock);
> + clear_disconnected(target_dir);
> noprogress = 0;
> } else if (pd == mnt->mnt_sb->s_root) {
> printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n");
> - spin_lock(&pd->d_lock);
> - pd->d_flags &= ~DCACHE_DISCONNECTED;
> - spin_unlock(&pd->d_lock);
> + clear_disconnected(target_dir);
> noprogress = 0;
> } else {
> /*
next prev parent reply other threads:[~2013-10-11 16:02 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-09-07 18:45 4 DCACHE_DISCONNECTED patches J. Bruce Fields
2013-09-07 18:45 ` [PATCH 1/4] dcache: use IS_ROOT to decide where dentry is hashed J. Bruce Fields
2013-09-07 18:45 ` [PATCH 2/4] dcache: make __d_shrink callers check d_unhashed() J. Bruce Fields
2013-09-07 18:46 ` [PATCH 3/4] dcache: Don't set DISCONNECTED on "pseudo filesystem" dentries J. Bruce Fields
2013-09-07 18:46 ` [PATCH 4/4] dcache: don't clear DCACHE_DISCONNECTED too early J. Bruce Fields
2013-09-09 0:46 ` Al Viro
2013-09-09 7:45 ` Christoph Hellwig
2013-09-09 20:46 ` J. Bruce Fields
2013-10-11 16:02 ` J. Bruce Fields [this message]
2013-10-12 8:42 ` Christoph Hellwig
2013-10-15 19:29 ` J. Bruce Fields
2013-09-08 9:16 ` 4 DCACHE_DISCONNECTED patches Christoph Hellwig
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20131011160237.GA21258@fieldses.org \
--to=bfields@fieldses.org \
--cc=bfields@redhat.com \
--cc=hch@infradead.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-nfs@vger.kernel.org \
--cc=viro@ZenIV.linux.org.uk \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).