From becc575eec74ac5ab5b5c8e2033984ef817278cd Mon Sep 17 00:00:00 2001 From: kib Date: Wed, 17 Jun 2015 04:46:58 +0000 Subject: vfs_msync(), called from syncer vnode fsync VOP, only iterates over the active vnode list for the given mount point, with the assumption that vnodes with dirty pages are active. This is enforced by vinactive() doing vm_object_page_clean() pass over the vnode pages. The issue is, if vinactive() cannot be called during vput() due to the vnode being only shared-locked, we might end up with the dirty pages for the vnode on the free list. Such vnode is invisible to syncer, and pages are only cleaned on the vnode reactivation. In other words, the race results in the broken guarantee that user data, written through the mmap(2), is written to the disk not later than in 30 seconds after the write. Fix this by keeping the vnode which is freed but still owing inactivation, on the active list. When syncer loops find such vnode, it is deactivated and cleaned by the final vput() call. Tested by: pho Sponsored by: The FreeBSD Foundation MFC after: 2 weeks --- sys/kern/vfs_subr.c | 46 +++++++++++++++++++++++++++------------------- 1 file changed, 27 insertions(+), 19 deletions(-) (limited to 'sys/kern/vfs_subr.c') diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 625d193..ce194f9 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -173,6 +173,11 @@ static int reassignbufcalls; SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "Number of calls to reassignbuf"); +static u_long free_owe_inact; +SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0, + "Number of times free vnodes kept on active list due to VFS " + "owing inactivation"); + /* * Cache for the mount type id assigned to NFS. This is used for * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. @@ -2368,11 +2373,8 @@ vholdl(struct vnode *vp) CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS /* getnewvnode() calls v_incr_usecount() without holding interlock. */ - if (vp->v_type != VNON || vp->v_data != NULL) { + if (vp->v_type != VNON || vp->v_data != NULL) ASSERT_VI_LOCKED(vp, "vholdl"); - VNASSERT(vp->v_holdcnt > 0 || (vp->v_iflag & VI_FREE) != 0, - vp, ("vholdl: free vnode is held")); - } #endif vp->v_holdcnt++; if ((vp->v_iflag & VI_FREE) == 0) @@ -2443,23 +2445,29 @@ vdropl(struct vnode *vp) VNASSERT(vp->v_holdcnt == 0, vp, ("vdropl: freeing when we shouldn't")); active = vp->v_iflag & VI_ACTIVE; - vp->v_iflag &= ~VI_ACTIVE; - mp = vp->v_mount; - mtx_lock(&vnode_free_list_mtx); - if (active) { - TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, - v_actfreelist); - mp->mnt_activevnodelistsize--; - } - if (vp->v_iflag & VI_AGE) { - TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist); + if ((vp->v_iflag & VI_OWEINACT) == 0) { + vp->v_iflag &= ~VI_ACTIVE; + mp = vp->v_mount; + mtx_lock(&vnode_free_list_mtx); + if (active) { + TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, + v_actfreelist); + mp->mnt_activevnodelistsize--; + } + if (vp->v_iflag & VI_AGE) { + TAILQ_INSERT_HEAD(&vnode_free_list, vp, + v_actfreelist); + } else { + TAILQ_INSERT_TAIL(&vnode_free_list, vp, + v_actfreelist); + } + freevnodes++; + vp->v_iflag &= ~VI_AGE; + vp->v_iflag |= VI_FREE; + mtx_unlock(&vnode_free_list_mtx); } else { - TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); + atomic_add_long(&free_owe_inact, 1); } - freevnodes++; - vp->v_iflag &= ~VI_AGE; - vp->v_iflag |= VI_FREE; - mtx_unlock(&vnode_free_list_mtx); VI_UNLOCK(vp); return; } -- cgit v1.1