Skip site navigation (1)Skip section navigation (2) 
Date:      Thu, 12 Jun 2014 12:43:49 +0000 (UTC)
From:      Alexander Motin <mav@FreeBSD.org>
To:        src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org
Subject:   svn commit: r267392 - head/sys/kern
Message-ID:  <201406121243.s5CChnfg026979@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help 
Author: mav
Date: Thu Jun 12 12:43:48 2014
New Revision: 267392
URL: http://svnweb.freebsd.org/changeset/base/267392

Log:
  Implement simple direct-mapped cache for popular filesystem identifiers to
  avoid congestion on global mountlist_mtx mutex in vfs_busyfs(), while
  traversing through the list of mount points.
  
  This change significantly improves NFS server scalability, since it had
  to do this translation for every request, and the global lock becomes quite
  congested.
  
  This code is more optimized for relatively small number of mount points.
  On systems with hundreds of active mount points this simple cache may have
  many collisions.  But the original traversal code in that case should also
  behave much worse, so we are not loosing much.
  
  Reviewed by:	attilio
  MFC after:	2 weeks
  Sponsored by:	iXsystems, Inc.

Modified:
  head/sys/kern/vfs_subr.c

Modified: head/sys/kern/vfs_subr.c
==============================================================================
--- head/sys/kern/vfs_subr.c	Thu Jun 12 11:57:07 2014	(r267391)
+++ head/sys/kern/vfs_subr.c	Thu Jun 12 12:43:48 2014	(r267392)
@@ -500,23 +500,53 @@ vfs_getvfs(fsid_t *fsid)
 /*
  * Lookup a mount point by filesystem identifier, busying it before
  * returning.
+ *
+ * To avoid congestion on mountlist_mtx, implement simple direct-mapped
+ * cache for popular filesystem identifiers.  The cache is lockess, using
+ * the fact that struct mount's are never freed.  In worst case we may
+ * get pointer to unmounted or even different filesystem, so we have to
+ * check what we got, and go slow way if so.
  */
 struct mount *
 vfs_busyfs(fsid_t *fsid)
 {
+#define	FSID_CACHE_SIZE	256
+	typedef struct mount * volatile vmp_t;
+	static vmp_t cache[FSID_CACHE_SIZE];
 	struct mount *mp;
 	int error;
+	uint32_t hash;
 
 	CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
+	hash = fsid->val[0] ^ fsid->val[1];
+	hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
+	mp = cache[hash];
+	if (mp == NULL ||
+	    mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
+	    mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
+		goto slow;
+	if (vfs_busy(mp, 0) != 0) {
+		cache[hash] = NULL;
+		goto slow;
+	}
+	if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
+	    mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
+		return (mp);
+	else
+	    vfs_unbusy(mp);
+
+slow:
 	mtx_lock(&mountlist_mtx);
 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 		if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
 		    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
 			error = vfs_busy(mp, MBF_MNTLSTLOCK);
 			if (error) {
+				cache[hash] = NULL;
 				mtx_unlock(&mountlist_mtx);
 				return (NULL);
 			}
+			cache[hash] = mp;
 			return (mp);
 		}
 	}

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201406121243.s5CChnfg026979>