netdev - [PATCH] vfs: avoid call to inode_lru_list

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <1311828913.2697.33.camel@edumazet-laptop>
Date:	Thu, 28 Jul 2011 06:55:13 +0200
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Christoph Hellwig <hch@...radead.org>
Cc:	Tim Chen <tim.c.chen@...ux.intel.com>,
	Al Viro <viro@...IV.linux.org.uk>,
	David Miller <davem@...emloft.net>,
	Andi Kleen <andi@...stfloor.org>,
	Matthew Wilcox <matthew@....cx>,
	Anton Blanchard <anton@...ba.org>, npiggin@...nel.dk,
	linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
	netdev <netdev@...r.kernel.org>
Subject: [PATCH] vfs: avoid call to inode_lru_list_del() if possible

Le mercredi 27 juillet 2011 à 16:44 -0400, Christoph Hellwig a écrit :
> On Wed, Jul 27, 2011 at 05:21:05PM +0200, Eric Dumazet wrote:
> > If I am not mistaken, we can add unlocked checks on the three hot spots.
> > 
> > After following patch, a close(socket(PF_INET, SOCK_DGRAM, 0)) pair on
> > my dev machine takes ~3us instead of ~9us.
> > 
> > Maybe its better to split it in three patches, just let me know.
> 
> I think three patches would be a lot cleaner.
> 
> As for safety of the unlocked checks:
> 
>  - inode are either hashed when created or never, so that one looks
>    fine.
>  - same for the sb list.
>  - the writeback list is a bit more dynamic as we move things around
>    quite a bit.  But in additon to the inode_wb_list_del call from
>    evict() it only ever gets remove in writeback_single_inode, which
>    for a freeing inode can only be called from the callers of evict().
> 
> Btw, I wonder if you should micro-optimize things a bit further by
> moving the unhashed checks from the deletion functions into the callers
> and thus save a function call for each of them.
> 

Here is the last patch, addressing inode_lru_list_del() call.

Only the call done from iput_final() can obviously benefit from checking
i_lru being empty or not, so it makes sense to perform the check at
caller site instead of doing it in inode_lru_list_del()

[PATCH] vfs: avoid call to inode_lru_list_del() if possible

inode_lru_list_del() is expensive because of per superblock lru locking,
while some inodes are not in lru list.

Adding a check in iput_final() can speedup pipe/sockets workloads on
SMP.

Signed-off-by: Eric Dumazet <eric.dumazet@...il.com>
---
 fs/inode.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/inode.c b/fs/inode.c
index d0c72ff..b8b8939 100644
--- a/fs/inode.c
+++ b/fs/inode.c
@@ -1328,7 +1328,8 @@ static void iput_final(struct inode *inode)
 	}
 
 	inode->i_state |= I_FREEING;
-	inode_lru_list_del(inode);
+	if (!list_empty(&inode->i_lru))
+		inode_lru_list_del(inode);
 	spin_unlock(&inode->i_lock);
 
 	evict(inode);


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html