From owner-svn-src-all@freebsd.org Sun Sep 6 05:50:53 2015 Return-Path: Delivered-To: svn-src-all@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id 2220C9CB5EA; Sun, 6 Sep 2015 05:50:53 +0000 (UTC) (envelope-from mckusick@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2001:1900:2254:2068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id 1358112EA; Sun, 6 Sep 2015 05:50:53 +0000 (UTC) (envelope-from mckusick@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.70]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id t865oqNJ068448; Sun, 6 Sep 2015 05:50:52 GMT (envelope-from mckusick@FreeBSD.org) Received: (from mckusick@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id t865oqkB068441; Sun, 6 Sep 2015 05:50:52 GMT (envelope-from mckusick@FreeBSD.org) Message-Id: <201509060550.t865oqkB068441@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: mckusick set sender to mckusick@FreeBSD.org using -f From: Kirk McKusick Date: Sun, 6 Sep 2015 05:50:52 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r287497 - in head/sys: kern sys X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Sun, 06 Sep 2015 05:50:53 -0000 Author: mckusick Date: Sun Sep 6 05:50:51 2015 New Revision: 287497 URL: https://svnweb.freebsd.org/changeset/base/287497 Log: Track changes to kern.maxvnodes and appropriately increase or decrease the size of the name cache hash table (mapping file names to vnodes) and the vnode hash table (mapping mount point and inode number to vnode). An appropriate locking strategy is the key to changing hash table sizes while they are in active use. Reviewed by: kib Tested by: Peter Holm Differential Revision: https://reviews.freebsd.org/D2265 MFC after: 2 weeks Modified: head/sys/kern/vfs_cache.c head/sys/kern/vfs_hash.c head/sys/kern/vfs_subr.c head/sys/sys/vnode.h Modified: head/sys/kern/vfs_cache.c ============================================================================== --- head/sys/kern/vfs_cache.c Sat Sep 5 23:22:59 2015 (r287496) +++ head/sys/kern/vfs_cache.c Sun Sep 6 05:50:51 2015 (r287497) @@ -327,11 +327,17 @@ sysctl_debug_hashstat_rawnchash(SYSCTL_H struct namecache *ncp; int i, error, n_nchash, *cntbuf; +retry: n_nchash = nchash + 1; /* nchash is max index, not count */ if (req->oldptr == NULL) return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); CACHE_RLOCK(); + if (n_nchash != nchash + 1) { + CACHE_RUNLOCK(); + free(cntbuf, M_TEMP); + goto retry; + } /* Scan hash tables counting entries */ for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) LIST_FOREACH(ncp, ncpp, nc_hash) @@ -930,6 +936,44 @@ nchinit(void *dummy __unused) } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); +void +cache_changesize(int newmaxvnodes) +{ + struct nchashhead *new_nchashtbl, *old_nchashtbl; + u_long new_nchash, old_nchash; + struct namecache *ncp; + uint32_t hash; + int i; + + new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash); + /* If same hash table size, nothing to do */ + if (nchash == new_nchash) { + free(new_nchashtbl, M_VFSCACHE); + return; + } + /* + * Move everything from the old hash table to the new table. + * None of the namecache entries in the table can be removed + * because to do so, they have to be removed from the hash table. + */ + CACHE_WLOCK(); + old_nchashtbl = nchashtbl; + old_nchash = nchash; + nchashtbl = new_nchashtbl; + nchash = new_nchash; + for (i = 0; i <= old_nchash; i++) { + while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { + hash = fnv_32_buf(nc_get_name(ncp), ncp->nc_nlen, + FNV1_32_INIT); + hash = fnv_32_buf(&ncp->nc_dvp, sizeof(ncp->nc_dvp), + hash); + LIST_REMOVE(ncp, nc_hash); + LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); + } + } + CACHE_WUNLOCK(); + free(old_nchashtbl, M_VFSCACHE); +} /* * Invalidate all entries to a particular vnode. Modified: head/sys/kern/vfs_hash.c ============================================================================== --- head/sys/kern/vfs_hash.c Sat Sep 5 23:22:59 2015 (r287496) +++ head/sys/kern/vfs_hash.c Sun Sep 6 05:50:51 2015 (r287497) @@ -161,3 +161,40 @@ vfs_hash_rehash(struct vnode *vp, u_int vp->v_hash = hash; rw_wunlock(&vfs_hash_lock); } + +void +vfs_hash_changesize(int newmaxvnodes) +{ + struct vfs_hash_head *vfs_hash_newtbl, *vfs_hash_oldtbl; + u_long vfs_hash_newmask, vfs_hash_oldmask; + struct vnode *vp; + int i; + + vfs_hash_newtbl = hashinit(newmaxvnodes, M_VFS_HASH, + &vfs_hash_newmask); + /* If same hash table size, nothing to do */ + if (vfs_hash_mask == vfs_hash_newmask) { + free(vfs_hash_newtbl, M_VFS_HASH); + return; + } + /* + * Move everything from the old hash table to the new table. + * None of the vnodes in the table can be recycled because to + * do so, they have to be removed from the hash table. + */ + rw_wlock(&vfs_hash_lock); + vfs_hash_oldtbl = vfs_hash_tbl; + vfs_hash_oldmask = vfs_hash_mask; + vfs_hash_tbl = vfs_hash_newtbl; + vfs_hash_mask = vfs_hash_newmask; + for (i = 0; i <= vfs_hash_oldmask; i++) { + while ((vp = LIST_FIRST(&vfs_hash_oldtbl[i])) != NULL) { + LIST_REMOVE(vp, v_hashlist); + LIST_INSERT_HEAD( + vfs_hash_bucket(vp->v_mount, vp->v_hash), + vp, v_hashlist); + } + } + rw_wunlock(&vfs_hash_lock); + free(vfs_hash_oldtbl, M_VFS_HASH); +} Modified: head/sys/kern/vfs_subr.c ============================================================================== --- head/sys/kern/vfs_subr.c Sat Sep 5 23:22:59 2015 (r287496) +++ head/sys/kern/vfs_subr.c Sun Sep 6 05:50:51 2015 (r287497) @@ -281,8 +281,25 @@ static enum { SYNCER_RUNNING, SYNCER_SHU * XXX desiredvnodes is historical cruft and should not exist. */ int desiredvnodes; -SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, - &desiredvnodes, 0, "Maximum number of vnodes"); + +static int +sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS) +{ + int error, old_desiredvnodes; + + old_desiredvnodes = desiredvnodes; + if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0) + return (error); + if (old_desiredvnodes != desiredvnodes) { + vfs_hash_changesize(desiredvnodes); + cache_changesize(desiredvnodes); + } + return (0); +} + +SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, + CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0, + sysctl_update_desiredvnodes, "I", "Maximum number of vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); static int vnlru_nowhere; Modified: head/sys/sys/vnode.h ============================================================================== --- head/sys/sys/vnode.h Sat Sep 5 23:22:59 2015 (r287496) +++ head/sys/sys/vnode.h Sun Sep 6 05:50:51 2015 (r287497) @@ -607,6 +607,7 @@ struct vnode; typedef int (*vn_get_ino_t)(struct mount *, void *, int, struct vnode **); /* cache_* may belong in namei.h. */ +void cache_changesize(int newhashsize); #define cache_enter(dvp, vp, cnp) \ cache_enter_time(dvp, vp, cnp, NULL, NULL) void cache_enter_time(struct vnode *dvp, struct vnode *vp, @@ -843,6 +844,7 @@ int fifo_printinfo(struct vnode *); /* vfs_hash.c */ typedef int vfs_hash_cmp_t(struct vnode *vp, void *arg); +void vfs_hash_changesize(int newhashsize); int vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); u_int vfs_hash_index(struct vnode *vp); int vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg);