Put more struct vnode fields under splbio().

Buffer cache related struct vnode fields can be accessed in interrupt
context. Be more consistent with the use of splbio().

OK mpi@

9 files changed, 96 insertions, 41 deletions

diff --git a/sys/kern/vfs_bio.c b/sys/kern/vfs_bio.c
index 1e29dfcbce0..7ff46e196b9 100644
--- a/sys/kern/vfs_bio.c
+++ b/sys/kern/vfs_bio.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_bio.c,v 1.208 2021/12/12 09:14:59 visa Exp $	*/
+/*	$OpenBSD: vfs_bio.c,v 1.209 2022/08/12 14:30:52 visa Exp $	*/
 /*	$NetBSD: vfs_bio.c,v 1.44 1996/06/11 11:15:36 pk Exp $	*/
 
 /*
@@ -1554,7 +1554,10 @@ bufcache_getcleanbuf(int cachenum, int discard)
 
 
 void
-discard_buffer(struct buf *bp) {
+discard_buffer(struct buf *bp)
+{
+	splassert(IPL_BIO);
+
 	bufcache_take(bp);
 	if (bp->b_vp) {
 		RBT_REMOVE(buf_rb_bufs,
diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index 13c70890ccc..4ace1adde8d 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vfs_subr.c,v 1.315 2022/03/27 16:19:39 semarie Exp $	*/
+/*	$OpenBSD: vfs_subr.c,v 1.316 2022/08/12 14:30:52 visa Exp $	*/
 /*	$NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $	*/
 
 /*
@@ -662,16 +662,16 @@ vget(struct vnode *vp, int flags)
 	}
 	mtx_leave(&vnode_mtx);
 
+	s = splbio();
 	onfreelist = vp->v_bioflag & VBIOONFREELIST;
 	if (vp->v_usecount == 0 && onfreelist) {
-		s = splbio();
 		if (vp->v_holdcnt > 0)
 			TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
 		else
 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
 		vp->v_bioflag &= ~VBIOONFREELIST;
-		splx(s);
 	}
+	splx(s);
 
 	vp->v_usecount++;
 	if (flags & LK_TYPE_MASK) {
@@ -749,6 +749,7 @@ void
 vput(struct vnode *vp)
 {
 	struct proc *p = curproc;
+	int s;
 
 #ifdef DIAGNOSTIC
 	if (vp == NULL)
@@ -777,8 +778,10 @@ vput(struct vnode *vp)
 
 	VOP_INACTIVE(vp, p);
 
+	s = splbio();
 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
 		vputonfreelist(vp);
+	splx(s);
 }
 
 /*
@@ -790,6 +793,7 @@ int
 vrele(struct vnode *vp)
 {
 	struct proc *p = curproc;
+	int s;
 
 #ifdef DIAGNOSTIC
 	if (vp == NULL)
@@ -822,8 +826,10 @@ vrele(struct vnode *vp)
 
 	VOP_INACTIVE(vp, p);
 
+	s = splbio();
 	if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST))
 		vputonfreelist(vp);
+	splx(s);
 	return (1);
 }
 
@@ -831,6 +837,10 @@ vrele(struct vnode *vp)
 void
 vhold(struct vnode *vp)
 {
+	int s;
+
+	s = splbio();
+
 	/*
 	 * If it is on the freelist and the hold count is currently
 	 * zero, move it to the hold list.
@@ -841,12 +851,18 @@ vhold(struct vnode *vp)
 		TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist);
 	}
 	vp->v_holdcnt++;
+
+	splx(s);
 }
 
 /* Lose interest in a vnode. */
 void
 vdrop(struct vnode *vp)
 {
+	int s;
+
+	s = splbio();
+
 #ifdef DIAGNOSTIC
 	if (vp->v_holdcnt == 0)
 		panic("vdrop: zero holdcnt");
@@ -863,6 +879,8 @@ vdrop(struct vnode *vp)
 		TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist);
 		TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
 	}
+
+	splx(s);
 }
 
 /*
@@ -909,6 +927,7 @@ vflush_vnode(struct vnode *vp, void *arg)
 {
 	struct vflush_args *va = arg;
 	struct proc *p = curproc;
+	int empty, s;
 
 	if (vp == va->skipvp) {
 		return (0);
@@ -958,8 +977,11 @@ vflush_vnode(struct vnode *vp, void *arg)
 	 * XXX Might be nice to check per-fs "inode" flags, but
 	 * generally the filesystem is sync'd already, right?
 	 */
-	if ((va->flags & IGNORECLEAN) &&
-	    LIST_EMPTY(&vp->v_dirtyblkhd))
+	s = splbio();
+	empty = (va->flags & IGNORECLEAN) && LIST_EMPTY(&vp->v_dirtyblkhd);
+	splx(s);
+
+	if (empty)
 		return (0);
 
 #ifdef DEBUG_SYSCTL
@@ -992,6 +1014,7 @@ void
 vclean(struct vnode *vp, int flags, struct proc *p)
 {
 	int active, do_wakeup = 0;
+	int s;
 
 	/*
 	 * Check to see if the vnode is in use.
@@ -1066,9 +1089,11 @@ vclean(struct vnode *vp, int flags, struct proc *p)
 	if (active) {
 		vp->v_usecount--;
 		if (vp->v_usecount == 0) {
+			s = splbio();
 			if (vp->v_holdcnt > 0)
 				panic("vclean: not clean");
 			vputonfreelist(vp);
+			splx(s);
 		}
 	}
 	cache_purge(vp);
@@ -1125,6 +1150,7 @@ vgonel(struct vnode *vp, struct proc *p)
 {
 	struct vnode *vq;
 	struct vnode *vx;
+	int s;
 
 	KASSERT(vp->v_uvcount == 0);
 
@@ -1192,12 +1218,9 @@ vgonel(struct vnode *vp, struct proc *p)
 	 * Move onto the free list, unless we were called from
 	 * getnewvnode and we're not on any free list
 	 */
+	s = splbio();
 	if (vp->v_usecount == 0 &&
 	    (vp->v_bioflag & VBIOONFREELIST)) {
-		int s;
-
-		s = splbio();
-
 		if (vp->v_holdcnt > 0)
 			panic("vgonel: not clean");
 
@@ -1205,8 +1228,8 @@ vgonel(struct vnode *vp, struct proc *p)
 			TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
 			TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
 		}
-		splx(s);
 	}
+	splx(s);
 }
 
 /*
diff --git a/sys/msdosfs/msdosfs_vfsops.c b/sys/msdosfs/msdosfs_vfsops.c
index 543854c7a7e..0de37665dfd 100644
--- a/sys/msdosfs/msdosfs_vfsops.c
+++ b/sys/msdosfs/msdosfs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: msdosfs_vfsops.c,v 1.95 2021/11/13 18:18:59 kn Exp $	*/
+/*	$OpenBSD: msdosfs_vfsops.c,v 1.96 2022/08/12 14:30:52 visa Exp $	*/
 /*	$NetBSD: msdosfs_vfsops.c,v 1.48 1997/10/18 02:54:57 briggs Exp $	*/
 
 /*-
@@ -640,16 +640,22 @@ int
 msdosfs_sync_vnode(struct vnode *vp, void *arg)
 {
 	struct msdosfs_sync_arg *msa = arg;
-	int error;
 	struct denode *dep;
+	int error;
+	int s, skip = 0;
 
 	dep = VTODE(vp);
+	s = splbio();
 	if (vp->v_type == VNON ||
 	    ((dep->de_flag & (DE_ACCESS | DE_CREATE | DE_UPDATE | DE_MODIFIED)) == 0
 	      && LIST_EMPTY(&vp->v_dirtyblkhd)) ||
 	    msa->waitfor == MNT_LAZY) {
-		return (0);
+		skip = 1;
 	}
+	splx(s);
+
+	if (skip)
+		return (0);
 
 	if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT))
 		return (0);
diff --git a/sys/nfs/nfs_vfsops.c b/sys/nfs/nfs_vfsops.c
index d0a7bcf38bc..5b06158752d 100644
--- a/sys/nfs/nfs_vfsops.c
+++ b/sys/nfs/nfs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_vfsops.c,v 1.126 2021/01/02 02:41:42 cheloha Exp $	*/
+/*	$OpenBSD: nfs_vfsops.c,v 1.127 2022/08/12 14:30:53 visa Exp $	*/
 /*	$NetBSD: nfs_vfsops.c,v 1.46.4.1 1996/05/25 22:40:35 fvdl Exp $	*/
 
 /*
@@ -793,7 +793,8 @@ int
 nfs_sync(struct mount *mp, int waitfor, int stall, struct ucred *cred, struct proc *p)
 {
 	struct vnode *vp;
-	int error, allerror = 0;
+	int allerror = 0;
+	int empty, error, s;
 
 	/*
 	 * Don't traverse the vnode list if we want to skip all of them.
@@ -812,7 +813,12 @@ loop:
 		 */
 		if (vp->v_mount != mp)
 			goto loop;
-		if (VOP_ISLOCKED(vp) || LIST_EMPTY(&vp->v_dirtyblkhd))
+		if (VOP_ISLOCKED(vp))
+			continue;
+		s = splbio();
+		empty = LIST_EMPTY(&vp->v_dirtyblkhd);
+		splx(s);
+		if (empty)
 			continue;
 		if (vget(vp, LK_EXCLUSIVE))
 			goto loop;
diff --git a/sys/nfs/nfs_vnops.c b/sys/nfs/nfs_vnops.c
index 3cc87e3bc52..8e10c520ef8 100644
--- a/sys/nfs/nfs_vnops.c
+++ b/sys/nfs/nfs_vnops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: nfs_vnops.c,v 1.189 2022/06/26 05:20:42 visa Exp $	*/
+/*	$OpenBSD: nfs_vnops.c,v 1.190 2022/08/12 14:30:53 visa Exp $	*/
 /*	$NetBSD: nfs_vnops.c,v 1.62.4.1 1996/07/08 20:26:52 jtc Exp $	*/
 
 /*
@@ -2849,7 +2849,7 @@ nfs_flush(struct vnode *vp, struct ucred *cred, int waitfor, struct proc *p,
 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
 	uint64_t slptimeo = INFSLP;
 	int s, error = 0, slpflag = 0, retv, bvecpos;
-	int passone = 1;
+	int dirty, passone = 1;
 	u_quad_t off = (u_quad_t)-1, endoff = 0, toff;
 #ifndef NFS_COMMITBVECSIZ
 #define NFS_COMMITBVECSIZ	20
@@ -2982,8 +2982,8 @@ loop:
  loop2:
 		s = splbio();
 		error = vwaitforio(vp, slpflag, "nfs_fsync", slptimeo);
-		splx(s);
 		if (error) {
+			splx(s);
 			if (nfs_sigintr(nmp, NULL, p))
 				return (EINTR);
 			if (slpflag == PCATCH) {
@@ -2992,8 +2992,9 @@ loop:
 			}
 			goto loop2;
 		}
-
-		if (!LIST_EMPTY(&vp->v_dirtyblkhd) && commit) {
+		dirty = (!LIST_EMPTY(&vp->v_dirtyblkhd) && commit);
+		splx(s);
+		if (dirty) {
 #if 0
 			vprint("nfs_fsync: dirty", vp);
 #endif
diff --git a/sys/sys/vnode.h b/sys/sys/vnode.h
index a5adb99a033..c9bb267e78e 100644
--- a/sys/sys/vnode.h
+++ b/sys/sys/vnode.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vnode.h,v 1.166 2022/06/26 05:20:42 visa Exp $	*/
+/*	$OpenBSD: vnode.h,v 1.167 2022/08/12 14:30:53 visa Exp $	*/
 /*	$NetBSD: vnode.h,v 1.38 1996/02/29 20:59:05 cgd Exp $	*/
 
 /*
@@ -104,18 +104,17 @@ struct vnode {
 	u_int   v_writecount;		/* reference count of writers */
 	u_int	v_lockcount;		/* [V] # threads waiting on lock */
 
-	/* Flags that can be read/written in interrupts */
-	u_int   v_bioflag;
-	u_int   v_holdcnt;			/* buffer references */
+	u_int   v_bioflag;		/* [B] flags accessed in interrupts */
+	u_int   v_holdcnt;		/* [B] buffer references */
 	u_int   v_id;				/* capability identifier */
 	struct	mount *v_mount;			/* ptr to vfs we are in */
-	TAILQ_ENTRY(vnode) v_freelist;		/* vnode freelist */
+	TAILQ_ENTRY(vnode) v_freelist;	/* [B] vnode freelist */
 	TAILQ_ENTRY(vnode) v_mntvnodes;		/* vnodes for mount point */
-	struct	buf_rb_bufs v_bufs_tree;	/* lookup of all bufs */
-	struct	buflists v_cleanblkhd;		/* clean blocklist head */
-	struct	buflists v_dirtyblkhd;		/* dirty blocklist head */
+	struct	buf_rb_bufs v_bufs_tree;/* [B] lookup of all bufs */
+	struct	buflists v_cleanblkhd;	/* [B] clean blocklist head */
+	struct	buflists v_dirtyblkhd;	/* [B] dirty blocklist head */
 	u_int   v_numoutput;		/* [B] num of writes in progress */
-	LIST_ENTRY(vnode) v_synclist;		/* vnode with dirty buffers */
+	LIST_ENTRY(vnode) v_synclist;	/* [B] vnode with dirty buffers */
 	union {
 		struct mount	*vu_mountedhere;/* ptr to mounted vfs (VDIR) */
 		struct socket	*vu_socket;	/* unix ipc (VSOCK) */
diff --git a/sys/ufs/ext2fs/ext2fs_inode.c b/sys/ufs/ext2fs/ext2fs_inode.c
index 5c76a5956b3..2962628f865 100644
--- a/sys/ufs/ext2fs/ext2fs_inode.c
+++ b/sys/ufs/ext2fs/ext2fs_inode.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ext2fs_inode.c,v 1.65 2021/12/12 09:14:59 visa Exp $	*/
+/*	$OpenBSD: ext2fs_inode.c,v 1.66 2022/08/12 14:30:53 visa Exp $	*/
 /*	$NetBSD: ext2fs_inode.c,v 1.24 2001/06/19 12:59:18 wiz Exp $	*/
 
 /*
@@ -387,10 +387,15 @@ done:
 	for (i = 0; i < NDADDR; i++)
 		if (newblks[i] != oip->i_e2fs_blocks[i])
 			panic("ext2fs_truncate2");
-	if (length == 0 &&
-	    (!LIST_EMPTY(&ovp->v_cleanblkhd) ||
-	     !LIST_EMPTY(&ovp->v_dirtyblkhd)))
-		panic("ext2fs_truncate3");
+	if (length == 0) {
+		int s;
+
+		s = splbio();
+		if (!LIST_EMPTY(&ovp->v_cleanblkhd) ||
+		    !LIST_EMPTY(&ovp->v_dirtyblkhd))
+			panic("ext2fs_truncate3");
+		splx(s);
+	}
 #endif /* DIAGNOSTIC */
 	/*
 	 * Put back the real size.
diff --git a/sys/ufs/ext2fs/ext2fs_vfsops.c b/sys/ufs/ext2fs/ext2fs_vfsops.c
index 371ca53f096..af3f0a9ab33 100644
--- a/sys/ufs/ext2fs/ext2fs_vfsops.c
+++ b/sys/ufs/ext2fs/ext2fs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ext2fs_vfsops.c,v 1.116 2021/10/04 08:11:02 claudio Exp $	*/
+/*	$OpenBSD: ext2fs_vfsops.c,v 1.117 2022/08/12 14:30:53 visa Exp $	*/
 /*	$NetBSD: ext2fs_vfsops.c,v 1.1 1997/06/11 09:34:07 bouyer Exp $	*/
 
 /*
@@ -713,6 +713,7 @@ ext2fs_sync_vnode(struct vnode *vp, void *args)
 	struct ext2fs_sync_args *esa = args;
 	struct inode *ip;
 	int error, nlink0 = 0;
+	int s, skip = 0;
 
 	if (vp->v_type == VNON)
 		return (0);
@@ -722,10 +723,15 @@ ext2fs_sync_vnode(struct vnode *vp, void *args)
 	if (ip->i_e2fs_nlink == 0)
 		nlink0 = 1;
 
+	s = splbio();
 	if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
 	    LIST_EMPTY(&vp->v_dirtyblkhd)) {
-		goto end;
+		skip = 1;
 	}
+	splx(s);
+
+	if (skip)
+		goto end;
 
 	if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT)) {
 		esa->inflight = MIN(esa->inflight+1, 65536);
diff --git a/sys/ufs/ffs/ffs_vfsops.c b/sys/ufs/ffs/ffs_vfsops.c
index b513a271877..730a36b0fab 100644
--- a/sys/ufs/ffs/ffs_vfsops.c
+++ b/sys/ufs/ffs/ffs_vfsops.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ffs_vfsops.c,v 1.192 2021/10/20 06:35:39 semarie Exp $	*/
+/*	$OpenBSD: ffs_vfsops.c,v 1.193 2022/08/12 14:30:53 visa Exp $	*/
 /*	$NetBSD: ffs_vfsops.c,v 1.19 1996/02/09 22:22:26 christos Exp $	*/
 
 /*
@@ -1159,6 +1159,7 @@ ffs_sync_vnode(struct vnode *vp, void *arg)
 	struct ffs_sync_args *fsa = arg;
 	struct inode *ip;
 	int error, nlink0 = 0;
+	int s, skip = 0;
 
 	if (vp->v_type == VNON)
 		return (0);
@@ -1177,11 +1178,16 @@ ffs_sync_vnode(struct vnode *vp, void *arg)
 	if (ip->i_effnlink == 0)
 		nlink0 = 1;
 
+	s = splbio();
 	if ((ip->i_flag &
 	    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
 	    LIST_EMPTY(&vp->v_dirtyblkhd)) {
-		goto end;
+		skip = 1;
 	}
+	splx(s);
+
+	if (skip)
+		goto end;
 
 	if (vget(vp, LK_EXCLUSIVE | LK_NOWAIT)) {
 		fsa->inflight = MIN(fsa->inflight+1, 65536);