public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Ian Kent <raven@themaw.net>
To: NeilBrown <neilb@suse.de>
Cc: autofs@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 5/6] autofs4: avoid taking fs_lock during rcu-walk
Date: Wed, 16 Jul 2014 17:52:30 +0800	[thread overview]
Message-ID: <1405504350.2527.95.camel@perseus.fritz.box> (raw)
In-Reply-To: <20140709234114.4525.24652.stgit@notabene.brown>

On Thu, 2014-07-10 at 09:41 +1000, NeilBrown wrote:
> ->fs_lock protects AUTOFS_INF_EXPIRING.  We need to be sure
> that once the flag is set, no new references beneath the dentry
> are taken.  So rcu-walk currently needs to take fs_lock before
> checking the flag.  This hurts performance.
> 
> Change the expiry to a two-stage process.
> First set AUTHFS_INF_NO_RCU which forces any path walk into
> ref-walk mode, then drop the lock and call synchronize_rcu().
> Once that returns we can be sure no rcu-walk is active beneath
> the dentry and we can check reference counts again.
> 
> Now during an RCU-walk we can test AUTHFS_INF_EXPIRING without
> taking the lock as along as we test AUTHFS_INF_NO_RCU too.

Couple of typos above, eeek!

> If either are set, we must abort the RCU-walk
> If neither are set, we know that refcounts will be tested again
> after we finish the RCU-walk so we are safe to continue.

I believe the idea is sound and the patch looks good.
Nevertheless I think this is probably the tricky bit and if there is a
problem I'm not seeing it's probably in this patch.

The submount-test will probably help with that.

> 
> Signed-off-by: NeilBrown <neilb@suse.de>
> ---
>  fs/autofs4/autofs_i.h |    4 ++++
>  fs/autofs4/expire.c   |   46 ++++++++++++++++++++++++++++++++++++++--------
>  2 files changed, 42 insertions(+), 8 deletions(-)
> 
> diff --git a/fs/autofs4/autofs_i.h b/fs/autofs4/autofs_i.h
> index 99dbb05d6148..469724d7568c 100644
> --- a/fs/autofs4/autofs_i.h
> +++ b/fs/autofs4/autofs_i.h
> @@ -79,6 +79,10 @@ struct autofs_info {
>  };
>  
>  #define AUTOFS_INF_EXPIRING	(1<<0) /* dentry is in the process of expiring */
> +#define AUTOFS_INF_NO_RCU	(1<<1) /* the dentry is being considered
> +					* for expiry, so RCU_walk is
> +					* not permitted
> +					*/
>  #define AUTOFS_INF_PENDING	(1<<2) /* dentry pending mount */
>  
>  struct autofs_wait_queue {
> diff --git a/fs/autofs4/expire.c b/fs/autofs4/expire.c
> index fb0b5003353f..98a6fd4957f8 100644
> --- a/fs/autofs4/expire.c
> +++ b/fs/autofs4/expire.c
> @@ -333,10 +333,19 @@ struct dentry *autofs4_expire_direct(struct super_block *sb,
>  	if (ino->flags & AUTOFS_INF_PENDING)
>  		goto out;
>  	if (!autofs4_direct_busy(mnt, root, timeout, do_now)) {
> -		ino->flags |= AUTOFS_INF_EXPIRING;
> -		init_completion(&ino->expire_complete);
> +		ino->flags |= AUTOFS_INF_NO_RCU;
>  		spin_unlock(&sbi->fs_lock);
> -		return root;
> +		synchronize_rcu();
> +		spin_lock(&sbi->fs_lock);
> +		if (!autofs4_direct_busy(mnt, root, timeout, do_now)) {
> +			ino->flags |= AUTOFS_INF_EXPIRING;
> +			smp_mb()
> +			ino->flags &= ~AUTOFS_INF_NO_RCU;
> +			init_completion(&ino->expire_complete);
> +			spin_unlock(&sbi->fs_lock);
> +			return root;
> +		}
> +		ino->flags &= ~AUTOFS_INF_NO_RCU;
>  	}
>  out:
>  	spin_unlock(&sbi->fs_lock);
> @@ -445,12 +454,29 @@ struct dentry *autofs4_expire_indirect(struct super_block *sb,
>  	dentry = NULL;
>  	while ((dentry = get_next_positive_subdir(dentry, root))) {
>  		spin_lock(&sbi->fs_lock);
> -		expired = should_expire(dentry, mnt, timeout, how);
> -		if (expired) {
> +		ino = autofs4_dentry_ino(dentry);
> +		if (ino->flags & AUTOFS_INF_NO_RCU)
> +			expired = NULL;
> +		else
> +			expired = should_expire(dentry, mnt, timeout, how);
> +		if (!expired) {
> +			spin_unlock(&sbi->fs_lock);
> +			continue;
> +		}
> +		ino = autofs4_dentry_ino(expired);
> +		ino->flags |= AUTOFS_INF_NO_RCU;
> +		spin_unlock(&sbi->fs_lock);
> +		synchronize_rcu();
> +		spin_lock(&sbi->fs_lock);
> +		if (should_expire(expired, mnt, timeout, how)) {
>  			if (expired != dentry)
>  				dput(dentry);
>  			goto found;
>  		}
> +
> +		ino->flags &= ~AUTOFS_INF_NO_RCU;
> +		if (expired != dentry)
> +			dput(expired);
>  		spin_unlock(&sbi->fs_lock);
>  	}
>  	return NULL;
> @@ -458,8 +484,9 @@ struct dentry *autofs4_expire_indirect(struct super_block *sb,
>  found:
>  	DPRINTK("returning %p %.*s",
>  		expired, (int)expired->d_name.len, expired->d_name.name);
> -	ino = autofs4_dentry_ino(expired);
>  	ino->flags |= AUTOFS_INF_EXPIRING;
> +	smp_mb()
> +	ino->flags &= ~AUTOFS_INF_NO_RCU;
>  	init_completion(&ino->expire_complete);
>  	spin_unlock(&sbi->fs_lock);
>  	spin_lock(&sbi->lookup_lock);
> @@ -479,11 +506,14 @@ int autofs4_expire_wait(struct dentry *dentry, int rcu_walk)
>  	int status;
>  
>  	/* Block on any pending expire */
> +	if (!(ino->flags & (AUTOFS_INF_EXPIRING | AUTOFS_INF_NO_RCU)))
> +		return 0;
> +
> +	if (rcu_walk)
> +		return -ECHILD;

Be nice to add a blank line here.

>  	spin_lock(&sbi->fs_lock);
>  	if (ino->flags & AUTOFS_INF_EXPIRING) {
>  		spin_unlock(&sbi->fs_lock);
> -		if (rcu_walk)
> -			return -ECHILD;
>  
>  		DPRINTK("waiting for expire %p name=%.*s",
>  			 dentry, dentry->d_name.len, dentry->d_name.name);
> 
> 



  reply	other threads:[~2014-07-16  9:52 UTC|newest]

Thread overview: 33+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-07-09 23:41 [PATCH 0/6] autofs4: support RCU-walk NeilBrown
2014-07-09 23:41 ` [PATCH 6/6] autofs4: don't take spinlock when not needed in autofs4_lookup_expiring NeilBrown
2014-07-16  3:42   ` Ian Kent
2014-07-16  6:10     ` NeilBrown
2014-07-09 23:41 ` [PATCH 5/6] autofs4: avoid taking fs_lock during rcu-walk NeilBrown
2014-07-16  9:52   ` Ian Kent [this message]
2014-07-09 23:41 ` [PATCH 2/6] autofs4: remove a redundant assignment NeilBrown
2014-07-16  3:27   ` Ian Kent
2014-07-09 23:41 ` [PATCH 1/6] autofs4: remove unused autofs4_ispending() NeilBrown
2014-07-16  3:26   ` Ian Kent
2014-07-09 23:41 ` [PATCH 4/6] autofs4: factor should_expire() out of autofs4_expire_indirect NeilBrown
2014-07-14  0:53   ` [PATCH 4/6 v2] " NeilBrown
2014-07-15  3:48     ` Ian Kent
2014-07-15  4:05       ` NeilBrown
2014-07-15  7:44         ` Ian Kent
2014-07-16  7:50     ` Ian Kent
2014-07-17  4:34       ` NeilBrown
2014-07-09 23:41 ` [PATCH 3/6] autofs4: allow RCU-walk to walk through autofs4 NeilBrown
2014-07-16  4:44   ` Ian Kent
2014-07-16  5:51     ` NeilBrown
2014-07-16  6:56       ` Ian Kent
2014-07-17  5:00         ` Ian Kent
2014-07-17  8:04           ` NeilBrown
2014-07-17 10:17             ` Ian Kent
2014-07-29  1:51               ` NeilBrown
2014-07-29  6:37                 ` Ian Kent
2014-07-10  7:43 ` [PATCH 0/6] autofs4: support RCU-walk Ian Kent
2014-07-10  7:45   ` Ian Kent
2014-07-10  8:25   ` NeilBrown
2014-07-11  2:49     ` Ian Kent
2014-07-16  3:24 ` Ian Kent
2014-07-16  6:00   ` NeilBrown
2014-07-16  7:21     ` Ian Kent

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=1405504350.2527.95.camel@perseus.fritz.box \
    --to=raven@themaw.net \
    --cc=autofs@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=neilb@suse.de \
    /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