don't vgone the vnode in vcount if it is ours
[unix-history] / usr / src / sys / kern / vfs_subr.c
index 4d21604..176db6b 100644 (file)
@@ -1,10 +1,15 @@
 /*
 /*
- * Copyright (c) 1989 The Regents of the University of California.
- * All rights reserved.
+ * Copyright (c) 1989, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ * (c) UNIX System Laboratories, Inc.
+ * All or some portions of this file are derived from material licensed
+ * to the University of California by American Telephone and Telegraph
+ * Co. or Unix System Laboratories, Inc. and are reproduced herein with
+ * the permission of UNIX System Laboratories, Inc.
  *
  * %sccs.include.redist.c%
  *
  *
  * %sccs.include.redist.c%
  *
- *     @(#)vfs_subr.c  7.87 (Berkeley) %G%
+ *     @(#)vfs_subr.c  8.12 (Berkeley) %G%
  */
 
 /*
  */
 
 /*
 #include <sys/buf.h>
 #include <sys/errno.h>
 #include <sys/malloc.h>
 #include <sys/buf.h>
 #include <sys/errno.h>
 #include <sys/malloc.h>
+#include <sys/domain.h>
+#include <sys/mbuf.h>
+
+#include <vm/vm.h>
+#include <sys/sysctl.h>
 
 #include <miscfs/specfs/specdev.h>
 
 
 #include <miscfs/specfs/specdev.h>
 
@@ -38,24 +48,23 @@ int vttoif_tab[9] = {
 /*
  * Insq/Remq for the vnode usage lists.
  */
 /*
  * Insq/Remq for the vnode usage lists.
  */
-#define        bufinsvn(bp, dp)        list_enter_head(dp, bp, struct buf *, b_vnbufs)
-#define        bufremvn(bp)            list_remove(bp, struct buf *, b_vnbufs)
+#define        bufinsvn(bp, dp)        LIST_INSERT_HEAD(dp, bp, b_vnbufs)
+#define        bufremvn(bp) {  \
+       LIST_REMOVE(bp, b_vnbufs); \
+       (bp)->b_vnbufs.le_next = NOLIST; \
+}
+
+TAILQ_HEAD(freelst, vnode) vnode_free_list;    /* vnode free list */
+struct mntlist mountlist;                      /* mounted filesystem list */
 
 /*
 
 /*
- * Remove a mount point from the list of mounted filesystems.
- * Unmount of the root is illegal.
+ * Initialize the vnode management data structures.
  */
  */
-void
-vfs_remove(mp)
-       register struct mount *mp;
+vntblinit()
 {
 
 {
 
-       if (mp == rootfs)
-               panic("vfs_remove: unmounting root");
-       mp->mnt_prev->mnt_next = mp->mnt_next;
-       mp->mnt_next->mnt_prev = mp->mnt_prev;
-       mp->mnt_vnodecovered->v_mountedhere = (struct mount *)0;
-       vfs_unlock(mp);
+       TAILQ_INIT(&vnode_free_list);
+       TAILQ_INIT(&mountlist);
 }
 
 /*
 }
 
 /*
@@ -136,14 +145,11 @@ getvfs(fsid)
 {
        register struct mount *mp;
 
 {
        register struct mount *mp;
 
-       mp = rootfs;
-       do {
+       for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) {
                if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
                if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
-                   mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
+                   mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
                        return (mp);
                        return (mp);
-               }
-               mp = mp->mnt_next;
-       } while (mp != rootfs);
+       }
        return ((struct mount *)0);
 }
 
        return ((struct mount *)0);
 }
 
@@ -159,13 +165,13 @@ static u_short xxxfs_mntid;
 
        fsid_t tfsid;
 
 
        fsid_t tfsid;
 
-       mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + 11, 0);  /* XXX */
+       mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0);
        mp->mnt_stat.f_fsid.val[1] = mtype;
        if (xxxfs_mntid == 0)
                ++xxxfs_mntid;
        mp->mnt_stat.f_fsid.val[1] = mtype;
        if (xxxfs_mntid == 0)
                ++xxxfs_mntid;
-       tfsid.val[0] = makedev(nblkdev, xxxfs_mntid);
+       tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid);
        tfsid.val[1] = mtype;
        tfsid.val[1] = mtype;
-       if (rootfs) {
+       if (mountlist.tqh_first != NULL) {
                while (getvfs(&tfsid)) {
                        tfsid.val[0]++;
                        xxxfs_mntid++;
                while (getvfs(&tfsid)) {
                        tfsid.val[0]++;
                        xxxfs_mntid++;
@@ -190,16 +196,18 @@ void vattr_null(vap)
                vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec =
                vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec =
                vap->va_flags = vap->va_gen = VNOVAL;
                vap->va_mtime.ts_sec = vap->va_mtime.ts_nsec =
                vap->va_ctime.ts_sec = vap->va_ctime.ts_nsec =
                vap->va_flags = vap->va_gen = VNOVAL;
+       vap->va_vaflags = 0;
 }
 
 /*
  * Routines having to do with the management of the vnode table.
  */
 }
 
 /*
  * Routines having to do with the management of the vnode table.
  */
-struct vnode *vfreeh, **vfreet = &vfreeh;
 extern int (**dead_vnodeop_p)();
 extern void vclean();
 long numvnodes;
 extern struct vattr va_null;
 extern int (**dead_vnodeop_p)();
 extern void vclean();
 long numvnodes;
 extern struct vattr va_null;
+int newnodes = 0;
+int printcnt = 0;
 
 /*
  * Return the next vnode from the free list.
 
 /*
  * Return the next vnode from the free list.
@@ -210,36 +218,56 @@ getnewvnode(tag, mp, vops, vpp)
        int (**vops)();
        struct vnode **vpp;
 {
        int (**vops)();
        struct vnode **vpp;
 {
-       register struct vnode *vp, *vq;
+       register struct vnode *vp;
+       int s;
 
 
-       if ((vfreeh == NULL && numvnodes < 2 * desiredvnodes) ||
+newnodes++;
+       if ((vnode_free_list.tqh_first == NULL &&
+            numvnodes < 2 * desiredvnodes) ||
            numvnodes < desiredvnodes) {
                vp = (struct vnode *)malloc((u_long)sizeof *vp,
                    M_VNODE, M_WAITOK);
                bzero((char *)vp, sizeof *vp);
            numvnodes < desiredvnodes) {
                vp = (struct vnode *)malloc((u_long)sizeof *vp,
                    M_VNODE, M_WAITOK);
                bzero((char *)vp, sizeof *vp);
+               vp->v_freelist.tqe_next = (struct vnode *)0xdeadf;
+               vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
+               vp->v_mntvnodes.le_next = (struct vnode *)0xdeadf;
+               vp->v_mntvnodes.le_prev = (struct vnode **)0xdeadb;
                numvnodes++;
                numvnodes++;
+               vp->v_spare[0] = numvnodes;
        } else {
        } else {
-               if ((vp = vfreeh) == NULL) {
+               if ((vp = vnode_free_list.tqh_first) == NULL) {
                        tablefull("vnode");
                        *vpp = 0;
                        return (ENFILE);
                }
                if (vp->v_usecount)
                        panic("free vnode isn't");
                        tablefull("vnode");
                        *vpp = 0;
                        return (ENFILE);
                }
                if (vp->v_usecount)
                        panic("free vnode isn't");
-               if (vq = vp->v_freef)
-                       vq->v_freeb = &vfreeh;
-               else
-                       vfreet = &vfreeh;
-               vfreeh = vq;
-               vp->v_freef = NULL;
-               vp->v_freeb = NULL;
+               if (vp->v_freelist.tqe_next == (struct vnode *)0xdeadf ||
+                   vp->v_freelist.tqe_prev == (struct vnode **)0xdeadb)
+                       panic("getnewvnode: not on queue");
+               TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
+               vp->v_freelist.tqe_next = (struct vnode *)0xdeadf;
+               /* see comment on why 0xdeadb is set at end of vgone (below) */
+               vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
                vp->v_lease = NULL;
                if (vp->v_type != VBAD)
                        vgone(vp);
                vp->v_lease = NULL;
                if (vp->v_type != VBAD)
                        vgone(vp);
+#ifdef DIAGNOSTIC
                if (vp->v_data)
                        panic("cleaned vnode isn't");
                if (vp->v_data)
                        panic("cleaned vnode isn't");
+               s = splbio();
+               if (vp->v_numoutput)
+                       panic("Clean vnode has pending I/O's");
+               splx(s);
+#endif
                vp->v_flag = 0;
                vp->v_lastr = 0;
                vp->v_flag = 0;
                vp->v_lastr = 0;
+               vp->v_ralen = 0;
+               vp->v_maxra = 0;
+               vp->v_lastw = 0;
+               vp->v_lasta = 0;
+               vp->v_cstart = 0;
+               vp->v_clen = 0;
                vp->v_socket = 0;
        }
        vp->v_type = VNON;
                vp->v_socket = 0;
        }
        vp->v_type = VNON;
@@ -247,8 +275,10 @@ getnewvnode(tag, mp, vops, vpp)
        vp->v_tag = tag;
        vp->v_op = vops;
        insmntque(vp, mp);
        vp->v_tag = tag;
        vp->v_op = vops;
        insmntque(vp, mp);
-       VREF(vp);
        *vpp = vp;
        *vpp = vp;
+       vp->v_usecount = 1;
+       vp->v_data = 0;
+       if (printcnt-- > 0) vprint("getnewvnode got", vp);
        return (0);
 }
 
        return (0);
 }
 
@@ -259,30 +289,27 @@ insmntque(vp, mp)
        register struct vnode *vp;
        register struct mount *mp;
 {
        register struct vnode *vp;
        register struct mount *mp;
 {
-       register struct vnode *vq;
 
        /*
         * Delete from old mount point vnode list, if on one.
         */
 
        /*
         * Delete from old mount point vnode list, if on one.
         */
-       if (vp->v_mountb) {
-               if (vq = vp->v_mountf)
-                       vq->v_mountb = vp->v_mountb;
-               *vp->v_mountb = vq;
+       if (vp->v_mount != NULL) {
+               if (vp->v_mntvnodes.le_next == (struct vnode *)0xdeadf ||
+                   vp->v_mntvnodes.le_prev == (struct vnode **)0xdeadb)
+                       panic("insmntque: not on queue");
+               LIST_REMOVE(vp, v_mntvnodes);
+               vp->v_mntvnodes.le_next = (struct vnode *)0xdeadf;
+               vp->v_mntvnodes.le_prev = (struct vnode **)0xdeadb;
        }
        /*
         * Insert into list of vnodes for the new mount point, if available.
         */
        }
        /*
         * Insert into list of vnodes for the new mount point, if available.
         */
-       vp->v_mount = mp;
-       if (mp == NULL) {
-               vp->v_mountf = NULL;
-               vp->v_mountb = NULL;
+       if ((vp->v_mount = mp) == NULL)
                return;
                return;
-       }
-       if (vq = mp->mnt_mounth)
-               vq->v_mountb = &vp->v_mountf;
-       vp->v_mountf = vq;
-       vp->v_mountb = &mp->mnt_mounth;
-       mp->mnt_mounth = vp;
+       if (vp->v_mntvnodes.le_next != (struct vnode *)0xdeadf ||
+           vp->v_mntvnodes.le_prev != (struct vnode **)0xdeadb)
+               panic("insmntque: already on queue");
+       LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes);
 }
 
 /*
 }
 
 /*
@@ -293,9 +320,11 @@ vwakeup(bp)
 {
        register struct vnode *vp;
 
 {
        register struct vnode *vp;
 
-       bp->b_dirtyoff = bp->b_dirtyend = 0;
+       bp->b_flags &= ~B_WRITEINPROG;
        if (vp = bp->b_vp) {
                vp->v_numoutput--;
        if (vp = bp->b_vp) {
                vp->v_numoutput--;
+               if (vp->v_numoutput < 0)
+                       panic("vwakeup: neg numoutput");
                if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
                        if (vp->v_numoutput < 0)
                                panic("vwakeup: neg numoutput");
                if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
                        if (vp->v_numoutput < 0)
                                panic("vwakeup: neg numoutput");
@@ -310,11 +339,12 @@ vwakeup(bp)
  * Called with the underlying object locked.
  */
 int
  * Called with the underlying object locked.
  */
 int
-vinvalbuf(vp, flags, cred, p)
+vinvalbuf(vp, flags, cred, p, slpflag, slptimeo)
        register struct vnode *vp;
        int flags;
        struct ucred *cred;
        struct proc *p;
        register struct vnode *vp;
        int flags;
        struct ucred *cred;
        struct proc *p;
+       int slpflag, slptimeo;
 {
        register struct buf *bp;
        struct buf *nbp, *blist;
 {
        register struct buf *bp;
        struct buf *nbp, *blist;
@@ -323,40 +353,53 @@ vinvalbuf(vp, flags, cred, p)
        if (flags & V_SAVE) {
                if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p))
                        return (error);
        if (flags & V_SAVE) {
                if (error = VOP_FSYNC(vp, cred, MNT_WAIT, p))
                        return (error);
-               if (vp->v_dirtyblkhd.le_next != NULL)
+               if (vp->v_dirtyblkhd.lh_first != NULL)
                        panic("vinvalbuf: dirty bufs");
        }
        for (;;) {
                        panic("vinvalbuf: dirty bufs");
        }
        for (;;) {
-               if ((blist = vp->v_cleanblkhd.le_next) && flags & V_SAVEMETA)
+               if ((blist = vp->v_cleanblkhd.lh_first) && flags & V_SAVEMETA)
                        while (blist && blist->b_lblkno < 0)
                        while (blist && blist->b_lblkno < 0)
-                               blist = blist->b_vnbufs.qe_next;
-               if (!blist && (blist = vp->v_dirtyblkhd.le_next) && 
+                               blist = blist->b_vnbufs.le_next;
+               if (!blist && (blist = vp->v_dirtyblkhd.lh_first) && 
                    (flags & V_SAVEMETA))
                        while (blist && blist->b_lblkno < 0)
                    (flags & V_SAVEMETA))
                        while (blist && blist->b_lblkno < 0)
-                               blist = blist->b_vnbufs.qe_next;
+                               blist = blist->b_vnbufs.le_next;
                if (!blist)
                        break;
 
                for (bp = blist; bp; bp = nbp) {
                if (!blist)
                        break;
 
                for (bp = blist; bp; bp = nbp) {
-                       nbp = bp->b_vnbufs.qe_next;
+                       nbp = bp->b_vnbufs.le_next;
                        if (flags & V_SAVEMETA && bp->b_lblkno < 0)
                                continue;
                        s = splbio();
                        if (bp->b_flags & B_BUSY) {
                                bp->b_flags |= B_WANTED;
                        if (flags & V_SAVEMETA && bp->b_lblkno < 0)
                                continue;
                        s = splbio();
                        if (bp->b_flags & B_BUSY) {
                                bp->b_flags |= B_WANTED;
-                               sleep((caddr_t)bp, PRIBIO + 1);
+                               error = tsleep((caddr_t)bp,
+                                       slpflag | (PRIBIO + 1), "vinvalbuf",
+                                       slptimeo);
                                splx(s);
                                splx(s);
+                               if (error)
+                                       return (error);
                                break;
                        }
                        bremfree(bp);
                        bp->b_flags |= B_BUSY;
                        splx(s);
                                break;
                        }
                        bremfree(bp);
                        bp->b_flags |= B_BUSY;
                        splx(s);
+                       /*
+                        * XXX Since there are no node locks for NFS, I believe
+                        * there is a slight chance that a delayed write will
+                        * occur while sleeping just above, so check for it.
+                        */
+                       if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) {
+                               (void) VOP_BWRITE(bp);
+                               break;
+                       }
                        bp->b_flags |= B_INVAL;
                        brelse(bp);
                }
        }
        if (!(flags & V_SAVEMETA) &&
                        bp->b_flags |= B_INVAL;
                        brelse(bp);
                }
        }
        if (!(flags & V_SAVEMETA) &&
-           (vp->v_dirtyblkhd.le_next || vp->v_cleanblkhd.le_next))
+           (vp->v_dirtyblkhd.lh_first || vp->v_cleanblkhd.lh_first))
                panic("vinvalbuf: flush failed");
        return (0);
 }
                panic("vinvalbuf: flush failed");
        return (0);
 }
@@ -368,7 +411,6 @@ bgetvp(vp, bp)
        register struct vnode *vp;
        register struct buf *bp;
 {
        register struct vnode *vp;
        register struct buf *bp;
 {
-       register struct vnode *vq;
 
        if (bp->b_vp)
                panic("bgetvp: not free");
 
        if (bp->b_vp)
                panic("bgetvp: not free");
@@ -397,7 +439,7 @@ brelvp(bp)
        /*
         * Delete from old vnode list, if on one.
         */
        /*
         * Delete from old vnode list, if on one.
         */
-       if (bp->b_vnbufs.qe_next != NOLIST)
+       if (bp->b_vnbufs.le_next != NOLIST)
                bufremvn(bp);
        vp = bp->b_vp;
        bp->b_vp = (struct vnode *) 0;
                bufremvn(bp);
        vp = bp->b_vp;
        bp->b_vp = (struct vnode *) 0;
@@ -413,7 +455,7 @@ reassignbuf(bp, newvp)
        register struct buf *bp;
        register struct vnode *newvp;
 {
        register struct buf *bp;
        register struct vnode *newvp;
 {
-       register struct list_entry *listheadp;
+       register struct buflists *listheadp;
 
        if (newvp == NULL) {
                printf("reassignbuf: NULL");
 
        if (newvp == NULL) {
                printf("reassignbuf: NULL");
@@ -422,7 +464,7 @@ reassignbuf(bp, newvp)
        /*
         * Delete from old vnode list, if on one.
         */
        /*
         * Delete from old vnode list, if on one.
         */
-       if (bp->b_vnbufs.qe_next != NOLIST)
+       if (bp->b_vnbufs.le_next != NOLIST)
                bufremvn(bp);
        /*
         * If dirty, put on list of dirty buffers;
                bufremvn(bp);
        /*
         * If dirty, put on list of dirty buffers;
@@ -497,7 +539,7 @@ loop:
                        vgone(vp);
                        goto loop;
                }
                        vgone(vp);
                        goto loop;
                }
-               if (vget(vp))
+               if (vget(vp, 1))
                        goto loop;
                break;
        }
                        goto loop;
                break;
        }
@@ -533,10 +575,10 @@ loop:
  * indicate that the vnode is no longer usable (possibly having
  * been changed to a new file system type).
  */
  * indicate that the vnode is no longer usable (possibly having
  * been changed to a new file system type).
  */
-vget(vp)
+vget(vp, lockflag)
        register struct vnode *vp;
        register struct vnode *vp;
+       int lockflag;
 {
 {
-       register struct vnode *vq;
 
        if (vp->v_flag & VXLOCK) {
                vp->v_flag |= VXWANT;
 
        if (vp->v_flag & VXLOCK) {
                vp->v_flag |= VXWANT;
@@ -544,16 +586,18 @@ vget(vp)
                return (1);
        }
        if (vp->v_usecount == 0) {
                return (1);
        }
        if (vp->v_usecount == 0) {
-               if (vq = vp->v_freef)
-                       vq->v_freeb = vp->v_freeb;
-               else
-                       vfreet = vp->v_freeb;
-               *vp->v_freeb = vq;
-               vp->v_freef = NULL;
-               vp->v_freeb = NULL;
+               if (vp->v_freelist.tqe_next == (struct vnode *)0xdeadf ||
+                   vp->v_freelist.tqe_prev == (struct vnode **)0xdeadb)
+                       panic("vget: not on queue");
+               TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
        }
        }
-       VREF(vp);
-       VOP_LOCK(vp);
+               vp->v_freelist.tqe_next = (struct vnode *)0xdeadf;
+               vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb;
+       }
+       vp->v_usecount++;
+       if (lockflag)
+               VOP_LOCK(vp);
+       if (printcnt-- > 0) vprint("vget got", vp);
        return (0);
 }
 
        return (0);
 }
 
@@ -566,7 +610,13 @@ void vref(vp)
        struct vnode *vp;
 {
 
        struct vnode *vp;
 {
 
+       if (vp->v_usecount <= 0)
+               panic("vref used where vget required");
+       if (vp->v_freelist.tqe_next != (struct vnode *)0xdeadf ||
+           vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb)
+               panic("vref: not free");
        vp->v_usecount++;
        vp->v_usecount++;
+       if (printcnt-- > 0) vprint("vref get", vp);
        if (vp->v_type != VBLK && curproc)
                curproc->p_spare[0]++;
        if (bug_refs)
        if (vp->v_type != VBLK && curproc)
                curproc->p_spare[0]++;
        if (bug_refs)
@@ -597,6 +647,7 @@ void vrele(vp)
                panic("vrele: null vp");
 #endif
        vp->v_usecount--;
                panic("vrele: null vp");
 #endif
        vp->v_usecount--;
+       if (printcnt-- > 0) vprint("vrele put", vp);
        if (vp->v_type != VBLK && curproc)
                curproc->p_spare[0]--;
        if (bug_refs)
        if (vp->v_type != VBLK && curproc)
                curproc->p_spare[0]--;
        if (bug_refs)
@@ -612,10 +663,10 @@ void vrele(vp)
        /*
         * insert at tail of LRU list
         */
        /*
         * insert at tail of LRU list
         */
-       *vfreet = vp;
-       vp->v_freeb = vfreet;
-       vp->v_freef = NULL;
-       vfreet = &vp->v_freef;
+       if (vp->v_freelist.tqe_next != (struct vnode *)0xdeadf ||
+           vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb)
+               panic("vrele: not free");
+       TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
        VOP_INACTIVE(vp);
 }
 
        VOP_INACTIVE(vp);
 }
 
@@ -649,7 +700,10 @@ void holdrele(vp)
  * system error). If MNT_FORCE is specified, detach any active vnodes
  * that are found.
  */
  * system error). If MNT_FORCE is specified, detach any active vnodes
  * that are found.
  */
-int busyprt = 0;       /* patch to print out busy vnodes */
+#ifdef DIAGNOSTIC
+int busyprt = 0;       /* print out busy vnodes */
+struct ctldebug debug1 = { "busyprt", &busyprt };
+#endif
 
 vflush(mp, skipvp, flags)
        struct mount *mp;
 
 vflush(mp, skipvp, flags)
        struct mount *mp;
@@ -662,10 +716,10 @@ vflush(mp, skipvp, flags)
        if ((mp->mnt_flag & MNT_MPBUSY) == 0)
                panic("vflush: not busy");
 loop:
        if ((mp->mnt_flag & MNT_MPBUSY) == 0)
                panic("vflush: not busy");
 loop:
-       for (vp = mp->mnt_mounth; vp; vp = nvp) {
+       for (vp = mp->mnt_vnodelist.lh_first; vp; vp = nvp) {
                if (vp->v_mount != mp)
                        goto loop;
                if (vp->v_mount != mp)
                        goto loop;
-               nvp = vp->v_mountf;
+               nvp = vp->v_mntvnodes.le_next;
                /*
                 * Skip over a selected vnode.
                 */
                /*
                 * Skip over a selected vnode.
                 */
@@ -676,6 +730,13 @@ loop:
                 */
                if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
                        continue;
                 */
                if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
                        continue;
+               /*
+                * If WRITECLOSE is set, only flush out regular file
+                * vnodes open for writing.
+                */
+               if ((flags & WRITECLOSE) &&
+                   (vp->v_writecount == 0 || vp->v_type != VREG))
+                       continue;
                /*
                 * With v_usecount == 0, all we need to do is clear
                 * out the vnode data structures and we are done.
                /*
                 * With v_usecount == 0, all we need to do is clear
                 * out the vnode data structures and we are done.
@@ -685,6 +746,7 @@ loop:
                        continue;
                }
                /*
                        continue;
                }
                /*
+                * If FORCECLOSE is set, forcibly close the vnode.
                 * For block or character devices, revert to an
                 * anonymous device. For all other files, just kill them.
                 */
                 * For block or character devices, revert to an
                 * anonymous device. For all other files, just kill them.
                 */
@@ -698,8 +760,10 @@ loop:
                        }
                        continue;
                }
                        }
                        continue;
                }
+#ifdef DIAGNOSTIC
                if (busyprt)
                        vprint("vflush: busy vnode", vp);
                if (busyprt)
                        vprint("vflush: busy vnode", vp);
+#endif
                busy++;
        }
        if (busy)
                busy++;
        }
        if (busy)
@@ -715,11 +779,6 @@ vclean(vp, flags)
        register struct vnode *vp;
        int flags;
 {
        register struct vnode *vp;
        int flags;
 {
-       struct vop_inactive_args vop_inactive_a;
-       struct vop_reclaim_args vop_reclaim_a;
-       struct vop_unlock_args vop_unlock_a;
-       struct vop_close_args vop_close_a;
-       int (**origops)();
        int active;
 
        /*
        int active;
 
        /*
@@ -730,6 +789,14 @@ vclean(vp, flags)
         */
        if (active = vp->v_usecount)
                VREF(vp);
         */
        if (active = vp->v_usecount)
                VREF(vp);
+       /*
+        * Even if the count is zero, the VOP_INACTIVE routine may still
+        * have the object locked while it cleans it out. The VOP_LOCK
+        * ensures that the VOP_INACTIVE routine is done with its work.
+        * For active vnodes, it ensures that no other activity can
+        * occur while the underlying object is being cleaned out.
+        */
+       VOP_LOCK(vp);
        /*
         * Prevent the vnode from being recycled or
         * brought into use while we clean it out.
        /*
         * Prevent the vnode from being recycled or
         * brought into use while we clean it out.
@@ -738,63 +805,37 @@ vclean(vp, flags)
                panic("vclean: deadlock");
        vp->v_flag |= VXLOCK;
        /*
                panic("vclean: deadlock");
        vp->v_flag |= VXLOCK;
        /*
-        * Even if the count is zero, the VOP_INACTIVE routine may still
-        * have the object locked while it cleans it out. The VOP_LOCK
-        * ensures that the VOP_INACTIVE routine is done with its work.
-        * For active vnodes, it ensures that no other activity can
-        * occur while the buffer list is being cleaned out.
+        * Clean out any buffers associated with the vnode.
         */
         */
-       VOP_LOCK(vp);
        if (flags & DOCLOSE)
        if (flags & DOCLOSE)
-               vinvalbuf(vp, 1, NOCRED, NULL);
+               vinvalbuf(vp, V_SAVE, NOCRED, NULL, 0, 0);
        /*
        /*
-        * Prevent any further operations on the vnode from
-        * being passed through to the old file system.
+        * Any other processes trying to obtain this lock must first
+        * wait for VXLOCK to clear, then call the new lock operation.
         */
         */
-       origops = vp->v_op;
-       vp->v_op = dead_vnodeop_p;
-       vp->v_tag = VT_NON;
+       VOP_UNLOCK(vp);
        /*
        /*
-        * If purging an active vnode, it must be unlocked, closed,
-        * and deactivated before being reclaimed.
+        * If purging an active vnode, it must be closed and
+        * deactivated before being reclaimed.
         */
         */
-       vop_unlock_a.a_desc = VDESC(vop_unlock);
-       vop_unlock_a.a_vp = vp;
-       VOCALL(origops,VOFFSET(vop_unlock),&vop_unlock_a);
        if (active) {
        if (active) {
-               /*
-                * Note: these next two calls imply
-                * that vop_close and vop_inactive implementations
-                * cannot count on the ops vector being correctly
-                * set.
-                */
-               if (flags & DOCLOSE) {
-                       vop_close_a.a_desc = VDESC(vop_close);
-                       vop_close_a.a_vp = vp;
-                       vop_close_a.a_fflag = IO_NDELAY;
-                       vop_close_a.a_p = NULL;
-                       VOCALL(origops,VOFFSET(vop_close),&vop_close_a);
-               };
-               vop_inactive_a.a_desc = VDESC(vop_inactive);
-               vop_inactive_a.a_vp = vp;
-               VOCALL(origops,VOFFSET(vop_inactive),&vop_inactive_a);
+               if (flags & DOCLOSE)
+                       VOP_CLOSE(vp, IO_NDELAY, NOCRED, NULL);
+               VOP_INACTIVE(vp);
        }
        /*
         * Reclaim the vnode.
         */
        }
        /*
         * Reclaim the vnode.
         */
-       /*
-        * Emulate VOP_RECLAIM.
-        */
-       vop_reclaim_a.a_desc = VDESC(vop_reclaim);
-       vop_reclaim_a.a_vp = vp;
-       if (VOCALL(origops,VOFFSET(vop_reclaim),&vop_reclaim_a))
+       if (VOP_RECLAIM(vp))
                panic("vclean: cannot reclaim");
        if (active)
                vrele(vp);
 
        /*
                panic("vclean: cannot reclaim");
        if (active)
                vrele(vp);
 
        /*
-        * Done with purge, notify sleepers in vget of the grim news.
+        * Done with purge, notify sleepers of the grim news.
         */
         */
+       vp->v_op = dead_vnodeop_p;
+       vp->v_tag = VT_NON;
        vp->v_flag &= ~VXLOCK;
        if (vp->v_flag & VXWANT) {
                vp->v_flag &= ~VXWANT;
        vp->v_flag &= ~VXLOCK;
        if (vp->v_flag & VXWANT) {
                vp->v_flag &= ~VXWANT;
@@ -871,12 +912,13 @@ void vgone(vp)
        /*
         * Delete from old mount point vnode list, if on one.
         */
        /*
         * Delete from old mount point vnode list, if on one.
         */
-       if (vp->v_mountb) {
-               if (vq = vp->v_mountf)
-                       vq->v_mountb = vp->v_mountb;
-               *vp->v_mountb = vq;
-               vp->v_mountf = NULL;
-               vp->v_mountb = NULL;
+       if (vp->v_mount != NULL) {
+               if (vp->v_mntvnodes.le_next == (struct vnode *)0xdeadf ||
+                   vp->v_mntvnodes.le_prev == (struct vnode **)0xdeadb)
+                       panic("vgone: not on queue");
+               LIST_REMOVE(vp, v_mntvnodes);
+               vp->v_mntvnodes.le_next = (struct vnode *)0xdeadf;
+               vp->v_mntvnodes.le_prev = (struct vnode **)0xdeadb;
                vp->v_mount = NULL;
        }
        /*
                vp->v_mount = NULL;
        }
        /*
@@ -915,18 +957,25 @@ void vgone(vp)
                vp->v_specinfo = NULL;
        }
        /*
                vp->v_specinfo = NULL;
        }
        /*
-        * If it is on the freelist, move it to the head of the list.
+        * If it is on the freelist and not already at the head,
+        * move it to the head of the list. The test of the back
+        * pointer and the reference count of zero is because
+        * it will be removed from the free list by getnewvnode,
+        * but will not have its reference count incremented until
+        * after calling vgone. If the reference count were
+        * incremented first, vgone would (incorrectly) try to
+        * close the previous instance of the underlying object.
+        * So, the back pointer is explicitly set to `0xdeadb' in
+        * getnewvnode after removing it from the freelist to ensure
+        * that we do not try to move it here.
         */
         */
-       if (vp->v_freeb) {
-               if (vq = vp->v_freef)
-                       vq->v_freeb = vp->v_freeb;
-               else
-                       vfreet = vp->v_freeb;
-               *vp->v_freeb = vq;
-               vp->v_freef = vfreeh;
-               vp->v_freeb = &vfreeh;
-               vfreeh->v_freeb = &vp->v_freef;
-               vfreeh = vp;
+       if (vp->v_usecount == 0 &&
+           vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb &&
+           vnode_free_list.tqh_first != vp) {
+               if (vp->v_freelist.tqe_next == (struct vnode *)0xdeadf)
+                       panic("vgone: use 0, not free");
+               TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
+               TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
        }
        vp->v_type = VBAD;
 }
        }
        vp->v_type = VBAD;
 }
@@ -945,9 +994,9 @@ vfinddev(dev, type, vpp)
                if (dev != vp->v_rdev || type != vp->v_type)
                        continue;
                *vpp = vp;
                if (dev != vp->v_rdev || type != vp->v_type)
                        continue;
                *vpp = vp;
-               return (0);
+               return (1);
        }
        }
-       return (1);
+       return (0);
 }
 
 /*
 }
 
 /*
@@ -956,19 +1005,20 @@ vfinddev(dev, type, vpp)
 vcount(vp)
        register struct vnode *vp;
 {
 vcount(vp)
        register struct vnode *vp;
 {
-       register struct vnode *vq;
+       register struct vnode *vq, *vnext;
        int count;
 
        int count;
 
+loop:
        if ((vp->v_flag & VALIASED) == 0)
                return (vp->v_usecount);
        if ((vp->v_flag & VALIASED) == 0)
                return (vp->v_usecount);
-loop:
-       for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+       for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) {
+               vnext = vq->v_specnext;
                if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
                        continue;
                /*
                 * Alias, but not in use, so flush it out.
                 */
                if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
                        continue;
                /*
                 * Alias, but not in use, so flush it out.
                 */
-               if (vq->v_usecount == 0) {
+               if (vq->v_usecount == 0 && vq != vp) {
                        vgone(vq);
                        goto loop;
                }
                        vgone(vq);
                        goto loop;
                }
@@ -991,6 +1041,7 @@ vprint(label, vp)
 
        if (label != NULL)
                printf("%s: ", label);
 
        if (label != NULL)
                printf("%s: ", label);
+       printf("num %d ", vp->v_spare[0]);
        printf("type %s, usecount %d, writecount %d, refcount %d,",
                typename[vp->v_type], vp->v_usecount, vp->v_writecount,
                vp->v_holdcnt);
        printf("type %s, usecount %d, writecount %d, refcount %d,",
                typename[vp->v_type], vp->v_usecount, vp->v_writecount,
                vp->v_holdcnt);
@@ -1011,8 +1062,12 @@ vprint(label, vp)
                strcat(buf, "|VALIASED");
        if (buf[0] != '\0')
                printf(" flags (%s)", &buf[1]);
                strcat(buf, "|VALIASED");
        if (buf[0] != '\0')
                printf(" flags (%s)", &buf[1]);
-       printf("\n\t");
-       VOP_PRINT(vp);
+       if (vp->v_data == NULL) {
+               printf("\n");
+       } else {
+               printf("\n\t");
+               VOP_PRINT(vp);
+       }
 }
 
 #ifdef DEBUG
 }
 
 #ifdef DEBUG
@@ -1026,13 +1081,13 @@ printlockedvnodes()
        register struct vnode *vp;
 
        printf("Locked vnodes\n");
        register struct vnode *vp;
 
        printf("Locked vnodes\n");
-       mp = rootfs;
-       do {
-               for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf)
+       for (mp = mountlist.tqh_first; mp != NULL; mp = mp->mnt_list.tqe_next) {
+               for (vp = mp->mnt_vnodelist.lh_first;
+                    vp != NULL;
+                    vp = vp->v_mntvnodes.le_next)
                        if (VOP_ISLOCKED(vp))
                                vprint((char *)0, vp);
                        if (VOP_ISLOCKED(vp))
                                vprint((char *)0, vp);
-               mp = mp->mnt_next;
-       } while (mp != rootfs);
+       }
 }
 #endif
 
 }
 #endif
 
@@ -1040,17 +1095,15 @@ int kinfo_vdebug = 1;
 int kinfo_vgetfailed;
 #define KINFO_VNODESLOP        10
 /*
 int kinfo_vgetfailed;
 #define KINFO_VNODESLOP        10
 /*
- * Dump vnode list (via kinfo).
+ * Dump vnode list (via sysctl).
  * Copyout address of vnode followed by vnode.
  */
 /* ARGSUSED */
  * Copyout address of vnode followed by vnode.
  */
 /* ARGSUSED */
-kinfo_vnode(op, where, acopysize, arg, aneeded)
-       int op;
+sysctl_vnode(where, sizep)
        char *where;
        char *where;
-       int *acopysize, arg, *aneeded;
+       size_t *sizep;
 {
 {
-       register struct mount *mp = rootfs;
-       struct mount *omp;
+       register struct mount *mp, *nmp;
        struct vnode *vp;
        register char *bp = where, *savebp;
        char *ewhere;
        struct vnode *vp;
        register char *bp = where, *savebp;
        char *ewhere;
@@ -1059,19 +1112,20 @@ kinfo_vnode(op, where, acopysize, arg, aneeded)
 #define VPTRSZ sizeof (struct vnode *)
 #define VNODESZ        sizeof (struct vnode)
        if (where == NULL) {
 #define VPTRSZ sizeof (struct vnode *)
 #define VNODESZ        sizeof (struct vnode)
        if (where == NULL) {
-               *aneeded = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
+               *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
                return (0);
        }
                return (0);
        }
-       ewhere = where + *acopysize;
+       ewhere = where + *sizep;
                
                
-       do {
-               if (vfs_busy(mp)) {
-                       mp = mp->mnt_next;
+       for (mp = mountlist.tqh_first; mp != NULL; mp = nmp) {
+               nmp = mp->mnt_list.tqe_next;
+               if (vfs_busy(mp))
                        continue;
                        continue;
-               }
                savebp = bp;
 again:
                savebp = bp;
 again:
-               for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) {
+               for (vp = mp->mnt_vnodelist.lh_first;
+                    vp != NULL;
+                    vp = vp->v_mntvnodes.le_next) {
                        /*
                         * Check that the vp is still associated with
                         * this filesystem.  RACE: could have been
                        /*
                         * Check that the vp is still associated with
                         * this filesystem.  RACE: could have been
@@ -1083,22 +1137,204 @@ again:
                                bp = savebp;
                                goto again;
                        }
                                bp = savebp;
                                goto again;
                        }
-                       if ((bp + VPTRSZ + VNODESZ <= ewhere) && 
-                           ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
-                            (error = copyout((caddr_t)vp, bp + VPTRSZ, 
-                             VNODESZ))))
+                       if (bp + VPTRSZ + VNODESZ > ewhere) {
+                               *sizep = bp - where;
+                               return (ENOMEM);
+                       }
+                       if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
+                          (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ)))
                                return (error);
                        bp += VPTRSZ + VNODESZ;
                }
                                return (error);
                        bp += VPTRSZ + VNODESZ;
                }
-               omp = mp;
-               mp = mp->mnt_next;
-               vfs_unbusy(omp);
-       } while (mp != rootfs);
-
-       *aneeded = bp - where;
-       if (bp > ewhere)
-               *acopysize = ewhere - where;
-       else
-               *acopysize = bp - where;
+               vfs_unbusy(mp);
+       }
+
+       *sizep = bp - where;
        return (0);
 }
        return (0);
 }
+
+/*
+ * Check to see if a filesystem is mounted on a block device.
+ */
+int
+vfs_mountedon(vp)
+       register struct vnode *vp;
+{
+       register struct vnode *vq;
+
+       if (vp->v_specflags & SI_MOUNTEDON)
+               return (EBUSY);
+       if (vp->v_flag & VALIASED) {
+               for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
+                       if (vq->v_rdev != vp->v_rdev ||
+                           vq->v_type != vp->v_type)
+                               continue;
+                       if (vq->v_specflags & SI_MOUNTEDON)
+                               return (EBUSY);
+               }
+       }
+       return (0);
+}
+
+/*
+ * Build hash lists of net addresses and hang them off the mount point.
+ * Called by ufs_mount() to set up the lists of export addresses.
+ */
+static int
+vfs_hang_addrlist(mp, nep, argp)
+       struct mount *mp;
+       struct netexport *nep;
+       struct export_args *argp;
+{
+       register struct netcred *np;
+       register struct radix_node_head *rnh;
+       register int i;
+       struct radix_node *rn;
+       struct sockaddr *saddr, *smask = 0;
+       struct domain *dom;
+       int error;
+
+       if (argp->ex_addrlen == 0) {
+               if (mp->mnt_flag & MNT_DEFEXPORTED)
+                       return (EPERM);
+               np = &nep->ne_defexported;
+               np->netc_exflags = argp->ex_flags;
+               np->netc_anon = argp->ex_anon;
+               np->netc_anon.cr_ref = 1;
+               mp->mnt_flag |= MNT_DEFEXPORTED;
+               return (0);
+       }
+       i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
+       np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK);
+       bzero((caddr_t)np, i);
+       saddr = (struct sockaddr *)(np + 1);
+       if (error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen))
+               goto out;
+       if (saddr->sa_len > argp->ex_addrlen)
+               saddr->sa_len = argp->ex_addrlen;
+       if (argp->ex_masklen) {
+               smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
+               error = copyin(argp->ex_addr, (caddr_t)smask, argp->ex_masklen);
+               if (error)
+                       goto out;
+               if (smask->sa_len > argp->ex_masklen)
+                       smask->sa_len = argp->ex_masklen;
+       }
+       i = saddr->sa_family;
+       if ((rnh = nep->ne_rtable[i]) == 0) {
+               /*
+                * Seems silly to initialize every AF when most are not
+                * used, do so on demand here
+                */
+               for (dom = domains; dom; dom = dom->dom_next)
+                       if (dom->dom_family == i && dom->dom_rtattach) {
+                               dom->dom_rtattach((void **)&nep->ne_rtable[i],
+                                       dom->dom_rtoffset);
+                               break;
+                       }
+               if ((rnh = nep->ne_rtable[i]) == 0) {
+                       error = ENOBUFS;
+                       goto out;
+               }
+       }
+       rn = (*rnh->rnh_addaddr)((caddr_t)saddr, (caddr_t)smask, rnh,
+               np->netc_rnodes);
+       if (rn == 0 || np != (struct netcred *)rn) { /* already exists */
+               error = EPERM;
+               goto out;
+       }
+       np->netc_exflags = argp->ex_flags;
+       np->netc_anon = argp->ex_anon;
+       np->netc_anon.cr_ref = 1;
+       return (0);
+out:
+       free(np, M_NETADDR);
+       return (error);
+}
+
+/* ARGSUSED */
+static int
+vfs_free_netcred(rn, w)
+       struct radix_node *rn;
+       caddr_t w;
+{
+       register struct radix_node_head *rnh = (struct radix_node_head *)w;
+
+       (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh);
+       free((caddr_t)rn, M_NETADDR);
+       return (0);
+}
+       
+/*
+ * Free the net address hash lists that are hanging off the mount points.
+ */
+static void
+vfs_free_addrlist(nep)
+       struct netexport *nep;
+{
+       register int i;
+       register struct radix_node_head *rnh;
+
+       for (i = 0; i <= AF_MAX; i++)
+               if (rnh = nep->ne_rtable[i]) {
+                       (*rnh->rnh_walktree)(rnh, vfs_free_netcred,
+                           (caddr_t)rnh);
+                       free((caddr_t)rnh, M_RTABLE);
+                       nep->ne_rtable[i] = 0;
+               }
+}
+
+int
+vfs_export(mp, nep, argp)
+       struct mount *mp;
+       struct netexport *nep;
+       struct export_args *argp;
+{
+       int error;
+
+       if (argp->ex_flags & MNT_DELEXPORT) {
+               vfs_free_addrlist(nep);
+               mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
+       }
+       if (argp->ex_flags & MNT_EXPORTED) {
+               if (error = vfs_hang_addrlist(mp, nep, argp))
+                       return (error);
+               mp->mnt_flag |= MNT_EXPORTED;
+       }
+       return (0);
+}
+
+struct netcred *
+vfs_export_lookup(mp, nep, nam)
+       register struct mount *mp;
+       struct netexport *nep;
+       struct mbuf *nam;
+{
+       register struct netcred *np;
+       register struct radix_node_head *rnh;
+       struct sockaddr *saddr;
+
+       np = NULL;
+       if (mp->mnt_flag & MNT_EXPORTED) {
+               /*
+                * Lookup in the export list first.
+                */
+               if (nam != NULL) {
+                       saddr = mtod(nam, struct sockaddr *);
+                       rnh = nep->ne_rtable[saddr->sa_family];
+                       if (rnh != NULL) {
+                               np = (struct netcred *)
+                                       (*rnh->rnh_matchaddr)((caddr_t)saddr,
+                                                             rnh);
+                               if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
+                                       np = NULL;
+                       }
+               }
+               /*
+                * If no address match, use the default if it exists.
+                */
+               if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
+                       np = &nep->ne_defexported;
+       }
+       return (np);
+}