Date: Sun, 22 Jun 2014 21:16:19 +0000 (UTC) From: Alexander Motin <mav@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-10@freebsd.org Subject: svn commit: r267749 - stable/10/sys/kern Message-ID: <201406222116.s5MLGJmp074739@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: mav Date: Sun Jun 22 21:16:18 2014 New Revision: 267749 URL: http://svnweb.freebsd.org/changeset/base/267749 Log: MFC r267392: 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. Modified: stable/10/sys/kern/vfs_subr.c Directory Properties: stable/10/ (props changed) Modified: stable/10/sys/kern/vfs_subr.c ============================================================================== --- stable/10/sys/kern/vfs_subr.c Sun Jun 22 21:14:34 2014 (r267748) +++ stable/10/sys/kern/vfs_subr.c Sun Jun 22 21:16:18 2014 (r267749) @@ -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?201406222116.s5MLGJmp074739>