added getnewfsid: global fsid allocator.
[unix-history] / usr / src / sys / kern / vfs_subr.c
index 7e20c23..aea0a38 100644 (file)
@@ -2,33 +2,37 @@
  * Copyright (c) 1989 The Regents of the University of California.
  * All rights reserved.
  *
  * Copyright (c) 1989 The Regents of the University of California.
  * All rights reserved.
  *
- * Redistribution and use in source and binary forms are permitted
- * provided that the above copyright notice and this paragraph are
- * duplicated in all such forms and that any documentation,
- * advertising materials, and other materials related to such
- * distribution and use acknowledge that the software was developed
- * by the University of California, Berkeley.  The name of the
- * University may not be used to endorse or promote products derived
- * from this software without specific prior written permission.
- * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
- * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
- * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
+ * %sccs.include.redist.c%
  *
  *
- *     @(#)vfs_subr.c  7.25 (Berkeley) %G%
+ *     @(#)vfs_subr.c  7.78 (Berkeley) %G%
  */
 
 /*
  * External virtual filesystem routines
  */
 
  */
 
 /*
  * External virtual filesystem routines
  */
 
-#include "param.h"
-#include "mount.h"
-#include "time.h"
-#include "vnode.h"
-#include "namei.h"
-#include "ucred.h"
-#include "errno.h"
-#include "malloc.h"
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/proc.h>
+#include <sys/mount.h>
+#include <sys/time.h>
+#include <sys/vnode.h>
+#include <sys/stat.h>
+#include <sys/specdev.h>
+#include <sys/namei.h>
+#include <sys/ucred.h>
+#include <sys/buf.h>
+#include <sys/errno.h>
+#include <sys/malloc.h>
+
+enum vtype iftovt_tab[16] = {
+       VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
+       VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
+};
+int    vttoif_tab[9] = {
+       0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
+       S_IFSOCK, S_IFIFO, S_IFMT,
+};
 
 /*
  * Remove a mount point from the list of mounted filesystems.
 
 /*
  * Remove a mount point from the list of mounted filesystems.
@@ -41,9 +45,9 @@ vfs_remove(mp)
 
        if (mp == rootfs)
                panic("vfs_remove: unmounting root");
 
        if (mp == rootfs)
                panic("vfs_remove: unmounting root");
-       mp->m_prev->m_next = mp->m_next;
-       mp->m_next->m_prev = mp->m_prev;
-       mp->m_vnodecovered->v_mountedhere = (struct mount *)0;
+       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);
 }
 
        vfs_unlock(mp);
 }
 
@@ -55,11 +59,11 @@ vfs_lock(mp)
        register struct mount *mp;
 {
 
        register struct mount *mp;
 {
 
-       while(mp->m_flag & M_MLOCK) {
-               mp->m_flag |= M_MWAIT;
+       while(mp->mnt_flag & MNT_MLOCK) {
+               mp->mnt_flag |= MNT_MWAIT;
                sleep((caddr_t)mp, PVFS);
        }
                sleep((caddr_t)mp, PVFS);
        }
-       mp->m_flag |= M_MLOCK;
+       mp->mnt_flag |= MNT_MLOCK;
        return (0);
 }
 
        return (0);
 }
 
@@ -72,15 +76,50 @@ vfs_unlock(mp)
        register struct mount *mp;
 {
 
        register struct mount *mp;
 {
 
-       if ((mp->m_flag & M_MLOCK) == 0)
-               panic("vfs_unlock: locked fs");
-       mp->m_flag &= ~M_MLOCK;
-       if (mp->m_flag & M_MWAIT) {
-               mp->m_flag &= ~M_MWAIT;
+       if ((mp->mnt_flag & MNT_MLOCK) == 0)
+               panic("vfs_unlock: not locked");
+       mp->mnt_flag &= ~MNT_MLOCK;
+       if (mp->mnt_flag & MNT_MWAIT) {
+               mp->mnt_flag &= ~MNT_MWAIT;
                wakeup((caddr_t)mp);
        }
 }
 
                wakeup((caddr_t)mp);
        }
 }
 
+/*
+ * Mark a mount point as busy.
+ * Used to synchronize access and to delay unmounting.
+ */
+vfs_busy(mp)
+       register struct mount *mp;
+{
+
+       while(mp->mnt_flag & MNT_MPBUSY) {
+               mp->mnt_flag |= MNT_MPWANT;
+               sleep((caddr_t)&mp->mnt_flag, PVFS);
+       }
+       if (mp->mnt_flag & MNT_UNMOUNT)
+               return (1);
+       mp->mnt_flag |= MNT_MPBUSY;
+       return (0);
+}
+
+/*
+ * Free a busy filesystem.
+ * Panic if filesystem is not busy.
+ */
+vfs_unbusy(mp)
+       register struct mount *mp;
+{
+
+       if ((mp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("vfs_unbusy: not busy");
+       mp->mnt_flag &= ~MNT_MPBUSY;
+       if (mp->mnt_flag & MNT_MPWANT) {
+               mp->mnt_flag &= ~MNT_MPWANT;
+               wakeup((caddr_t)&mp->mnt_flag);
+       }
+}
+
 /*
  * Lookup a mount point by filesystem identifier.
  */
 /*
  * Lookup a mount point by filesystem identifier.
  */
@@ -92,15 +131,40 @@ getvfs(fsid)
 
        mp = rootfs;
        do {
 
        mp = rootfs;
        do {
-               if (mp->m_fsid.val[0] == fsid->val[0] &&
-                   mp->m_fsid.val[1] == fsid->val[1]) {
+               if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
+                   mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
                        return (mp);
                }
                        return (mp);
                }
-               mp = mp->m_next;
+               mp = mp->mnt_next;
        } while (mp != rootfs);
        return ((struct mount *)0);
 }
 
        } while (mp != rootfs);
        return ((struct mount *)0);
 }
 
+/*
+ * Get a new unique fsid
+ */
+void
+getnewfsid(mp, mtype)
+       struct mount *mp;
+       int mtype;
+{
+static u_short xxxfs_mntid;
+
+       fsid_t tfsid;
+
+       mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + 11, 0);  /* XXX */
+       mp->mnt_stat.f_fsid.val[1] = mtype;
+       if (xxxfs_mntid == 0)
+               ++xxxfs_mntid;
+       tfsid.val[0] = makedev(nblkdev, xxxfs_mntid);
+       tfsid.val[1] = mtype;
+       while (getvfs(&tfsid)) {
+               tfsid.val[0]++;
+               xxxfs_mntid++;
+       }
+       mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
+}
+
 /*
  * Set vnode attributes to VNOVAL
  */
 /*
  * Set vnode attributes to VNOVAL
  */
@@ -109,114 +173,28 @@ void vattr_null(vap)
 {
 
        vap->va_type = VNON;
 {
 
        vap->va_type = VNON;
+       vap->va_size = vap->va_bytes = VNOVAL;
+#ifdef _NOQUAD
+       vap->va_size_rsv = vap->va_bytes_rsv = VNOVAL;
+#endif
        vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
        vap->va_mode = vap->va_nlink = vap->va_uid = vap->va_gid =
-               vap->va_fsid = vap->va_fileid = vap->va_size =
-               vap->va_size1 = vap->va_blocksize = vap->va_rdev =
-               vap->va_bytes = vap->va_bytes1 =
+               vap->va_fsid = vap->va_fileid =
+               vap->va_blocksize = vap->va_rdev =
                vap->va_atime.tv_sec = vap->va_atime.tv_usec =
                vap->va_mtime.tv_sec = vap->va_mtime.tv_usec =
                vap->va_ctime.tv_sec = vap->va_ctime.tv_usec =
                vap->va_flags = vap->va_gen = VNOVAL;
 }
 
                vap->va_atime.tv_sec = vap->va_atime.tv_usec =
                vap->va_mtime.tv_sec = vap->va_mtime.tv_usec =
                vap->va_ctime.tv_sec = vap->va_ctime.tv_usec =
                vap->va_flags = vap->va_gen = VNOVAL;
 }
 
-/*
- * Initialize a nameidata structure
- */
-ndinit(ndp)
-       register struct nameidata *ndp;
-{
-
-       bzero((caddr_t)ndp, sizeof(struct nameidata));
-       ndp->ni_iov = &ndp->ni_nd.nd_iovec;
-       ndp->ni_iovcnt = 1;
-       ndp->ni_base = (caddr_t)&ndp->ni_dent;
-       ndp->ni_rw = UIO_WRITE;
-       ndp->ni_uioseg = UIO_SYSSPACE;
-}
-
-/*
- * Duplicate a nameidata structure
- */
-nddup(ndp, newndp)
-       register struct nameidata *ndp, *newndp;
-{
-
-       ndinit(newndp);
-       newndp->ni_cdir = ndp->ni_cdir;
-       VREF(newndp->ni_cdir);
-       newndp->ni_rdir = ndp->ni_rdir;
-       if (newndp->ni_rdir)
-               VREF(newndp->ni_rdir);
-       newndp->ni_cred = ndp->ni_cred;
-       crhold(newndp->ni_cred);
-}
-
-/*
- * Release a nameidata structure
- */
-ndrele(ndp)
-       register struct nameidata *ndp;
-{
-
-       vrele(ndp->ni_cdir);
-       if (ndp->ni_rdir)
-               vrele(ndp->ni_rdir);
-       crfree(ndp->ni_cred);
-}
-
 /*
  * 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;
-extern struct vnodeops dead_vnodeops, spec_vnodeops;
+extern struct vnode *vfreeh, **vfreet;
+extern int (**dead_vnodeop_p)();
+extern int (**spec_vnodeop_p)();
 extern void vclean();
 extern void vclean();
-
-#define        SPECHSZ 64
-#if    ((SPECHSZ&(SPECHSZ-1)) == 0)
-#define        SPECHASH(rdev)  (((rdev>>5)+(rdev))&(SPECHSZ-1))
-#else
-#define        SPECHASH(rdev)  (((unsigned)((rdev>>5)+(rdev)))%SPECHSZ)
-#endif
-struct vnode *speclisth[SPECHSZ];
-
-/*
- * Initialize the vnode structures and initialize each file system type.
- */
-vfsinit()
-{
-       register struct vnode *vp = vnode;
-       struct vfsops **vfsp;
-
-       /*
-        * Build vnode free list.
-        */
-       vfreeh = vp;
-       vfreet = &vp->v_freef;
-       vp->v_freeb = &vfreeh;
-       vp->v_op = &dead_vnodeops;
-       vp->v_type = VBAD;
-       for (vp++; vp < vnodeNVNODE; vp++) {
-               *vfreet = vp;
-               vp->v_freeb = vfreet;
-               vfreet = &vp->v_freef;
-               vp->v_op = &dead_vnodeops;
-               vp->v_type = VBAD;
-       }
-       vp--;
-       vp->v_freef = NULL;
-       /*
-        * Initialize the vnode name cache
-        */
-       nchinit();
-       /*
-        * Initialize each file system type.
-        */
-       for (vfsp = &vfssw[0]; vfsp <= &vfssw[MOUNT_MAXTYPE]; vfsp++) {
-               if (*vfsp == NULL)
-                       continue;
-               (*(*vfsp)->vfs_init)();
-       }
-}
+long numvnodes;
+extern struct vattr va_null;
 
 /*
  * Return the next vnode from the free list.
 
 /*
  * Return the next vnode from the free list.
@@ -224,31 +202,41 @@ vfsinit()
 getnewvnode(tag, mp, vops, vpp)
        enum vtagtype tag;
        struct mount *mp;
 getnewvnode(tag, mp, vops, vpp)
        enum vtagtype tag;
        struct mount *mp;
-       struct vnodeops *vops;
+       int (**vops)();
        struct vnode **vpp;
 {
        register struct vnode *vp, *vq;
 
        struct vnode **vpp;
 {
        register struct vnode *vp, *vq;
 
-       if ((vp = vfreeh) == NULL) {
-               tablefull("vnode");
-               *vpp = 0;
-               return (ENFILE);
+       if (numvnodes < desiredvnodes) {
+               vp = (struct vnode *)malloc((u_long)sizeof *vp,
+                   M_VNODE, M_WAITOK);
+               bzero((char *)vp, sizeof *vp);
+               numvnodes++;
+       } else {
+               if ((vp = vfreeh) == NULL) {
+                       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;
+               vp->v_lease = NULL;
+               if (vp->v_type != VBAD)
+                       vgone(vp);
+               if (vp->v_data)
+                       panic("cleaned vnode isn't");
+               vp->v_flag = 0;
+               vp->v_lastr = 0;
+               vp->v_socket = 0;
        }
        }
-       if (vp->v_usecount)
-               panic("free vnode isn't");
-       if (vq = vp->v_freef)
-               vq->v_freeb = &vfreeh;
-       vfreeh = vq;
-       vp->v_freef = NULL;
-       vp->v_freeb = NULL;
-       if (vp->v_type != VBAD)
-               vgone(vp);
        vp->v_type = VNON;
        vp->v_type = VNON;
-       vp->v_flag = 0;
-       vp->v_shlockc = 0;
-       vp->v_exlockc = 0;
-       vp->v_lastr = 0;
-       vp->v_socket = 0;
        cache_purge(vp);
        vp->v_tag = tag;
        vp->v_op = vops;
        cache_purge(vp);
        vp->v_tag = tag;
        vp->v_op = vops;
@@ -265,7 +253,7 @@ insmntque(vp, mp)
        register struct vnode *vp;
        register struct mount *mp;
 {
        register struct vnode *vp;
        register struct mount *mp;
 {
-       struct vnode *vq;
+       register struct vnode *vq;
 
        /*
         * Delete from old mount point vnode list, if on one.
 
        /*
         * Delete from old mount point vnode list, if on one.
@@ -284,16 +272,276 @@ insmntque(vp, mp)
                vp->v_mountb = NULL;
                return;
        }
                vp->v_mountb = NULL;
                return;
        }
-       if (mp->m_mounth) {
-               vp->v_mountf = mp->m_mounth;
-               vp->v_mountb = &mp->m_mounth;
-               mp->m_mounth->v_mountb = &vp->v_mountf;
-               mp->m_mounth = vp;
-       } else {
-               mp->m_mounth = vp;
-               vp->v_mountb = &mp->m_mounth;
-               vp->v_mountf = NULL;
+       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;
+}
+
+/*
+ * Make sure all write-behind blocks associated
+ * with mount point are flushed out (from sync).
+ */
+mntflushbuf(mountp, flags)
+       struct mount *mountp;
+       int flags;
+{
+       USES_VOP_ISLOCKED;
+       register struct vnode *vp;
+
+       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("mntflushbuf: not busy");
+loop:
+       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
+               if (VOP_ISLOCKED(vp))
+                       continue;
+               if (vget(vp))
+                       goto loop;
+               vflushbuf(vp, flags);
+               vput(vp);
+               if (vp->v_mount != mountp)
+                       goto loop;
+       }
+}
+
+/*
+ * Flush all dirty buffers associated with a vnode.
+ */
+vflushbuf(vp, flags)
+       register struct vnode *vp;
+       int flags;
+{
+       register struct buf *bp;
+       struct buf *nbp;
+       int s;
+
+loop:
+       s = splbio();
+       for (bp = vp->v_dirtyblkhd; bp; bp = nbp) {
+               nbp = bp->b_blockf;
+               if ((bp->b_flags & B_BUSY))
+                       continue;
+               if ((bp->b_flags & B_DELWRI) == 0)
+                       panic("vflushbuf: not dirty");
+               bremfree(bp);
+               bp->b_flags |= B_BUSY;
+               splx(s);
+               /*
+                * Wait for I/O associated with indirect blocks to complete,
+                * since there is no way to quickly wait for them below.
+                * NB: This is really specific to ufs, but is done here
+                * as it is easier and quicker.
+                */
+               if (bp->b_vp == vp || (flags & B_SYNC) == 0)
+                       (void) bawrite(bp);
+               else
+                       (void) bwrite(bp);
+               goto loop;
+       }
+       splx(s);
+       if ((flags & B_SYNC) == 0)
+               return;
+       s = splbio();
+       while (vp->v_numoutput) {
+               vp->v_flag |= VBWAIT;
+               sleep((caddr_t)&vp->v_numoutput, PRIBIO + 1);
+       }
+       splx(s);
+       if (vp->v_dirtyblkhd) {
+               vprint("vflushbuf: dirty", vp);
+               goto loop;
+       }
+}
+
+/*
+ * Update outstanding I/O count and do wakeup if requested.
+ */
+vwakeup(bp)
+       register struct buf *bp;
+{
+       register struct vnode *vp;
+
+       bp->b_dirtyoff = bp->b_dirtyend = 0;
+       if (vp = bp->b_vp) {
+               vp->v_numoutput--;
+               if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) {
+                       if (vp->v_numoutput < 0)
+                               panic("vwakeup: neg numoutput");
+                       vp->v_flag &= ~VBWAIT;
+                       wakeup((caddr_t)&vp->v_numoutput);
+               }
+       }
+}
+
+/*
+ * Invalidate in core blocks belonging to closed or umounted filesystem
+ *
+ * Go through the list of vnodes associated with the file system;
+ * for each vnode invalidate any buffers that it holds. Normally
+ * this routine is preceeded by a bflush call, so that on a quiescent
+ * filesystem there will be no dirty buffers when we are done. Binval
+ * returns the count of dirty buffers when it is finished.
+ */
+mntinvalbuf(mountp)
+       struct mount *mountp;
+{
+       register struct vnode *vp;
+       int dirty = 0;
+
+       if ((mountp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("mntinvalbuf: not busy");
+loop:
+       for (vp = mountp->mnt_mounth; vp; vp = vp->v_mountf) {
+               if (vget(vp))
+                       goto loop;
+               dirty += vinvalbuf(vp, 1);
+               vput(vp);
+               if (vp->v_mount != mountp)
+                       goto loop;
+       }
+       return (dirty);
+}
+
+/*
+ * Flush out and invalidate all buffers associated with a vnode.
+ * Called with the underlying object locked.
+ */
+vinvalbuf(vp, save)
+       register struct vnode *vp;
+       int save;
+{
+       USES_VOP_BWRITE;
+       register struct buf *bp;
+       struct buf *nbp, *blist;
+       int s, dirty = 0;
+
+       for (;;) {
+               if (blist = vp->v_dirtyblkhd)
+                       /* void */;
+               else if (blist = vp->v_cleanblkhd)
+                       /* void */;
+               else
+                       break;
+               for (bp = blist; bp; bp = nbp) {
+                       nbp = bp->b_blockf;
+                       s = splbio();
+                       if (bp->b_flags & B_BUSY) {
+                               bp->b_flags |= B_WANTED;
+                               sleep((caddr_t)bp, PRIBIO + 1);
+                               splx(s);
+                               break;
+                       }
+                       bremfree(bp);
+                       bp->b_flags |= B_BUSY;
+                       splx(s);
+                       if (save && (bp->b_flags & B_DELWRI)) {
+                               dirty++;
+                               (void) VOP_BWRITE(bp);
+                               break;
+                       }
+                       if (bp->b_vp != vp)
+                               reassignbuf(bp, bp->b_vp);
+                       else
+                               bp->b_flags |= B_INVAL;
+                       brelse(bp);
+               }
+       }
+       if (vp->v_dirtyblkhd || vp->v_cleanblkhd)
+               panic("vinvalbuf: flush failed");
+       return (dirty);
+}
+
+/*
+ * Associate a buffer with a vnode.
+ */
+bgetvp(vp, bp)
+       register struct vnode *vp;
+       register struct buf *bp;
+{
+       register struct vnode *vq;
+       register struct buf *bq;
+
+       if (bp->b_vp)
+               panic("bgetvp: not free");
+       VHOLD(vp);
+       bp->b_vp = vp;
+       if (vp->v_type == VBLK || vp->v_type == VCHR)
+               bp->b_dev = vp->v_rdev;
+       else
+               bp->b_dev = NODEV;
+       /*
+        * Insert onto list for new vnode.
+        */
+       if (bq = vp->v_cleanblkhd)
+               bq->b_blockb = &bp->b_blockf;
+       bp->b_blockf = bq;
+       bp->b_blockb = &vp->v_cleanblkhd;
+       vp->v_cleanblkhd = bp;
+}
+
+/*
+ * Disassociate a buffer from a vnode.
+ */
+brelvp(bp)
+       register struct buf *bp;
+{
+       struct buf *bq;
+       struct vnode *vp;
+
+       if (bp->b_vp == (struct vnode *) 0)
+               panic("brelvp: NULL");
+       /*
+        * Delete from old vnode list, if on one.
+        */
+       if (bp->b_blockb) {
+               if (bq = bp->b_blockf)
+                       bq->b_blockb = bp->b_blockb;
+               *bp->b_blockb = bq;
+               bp->b_blockf = NULL;
+               bp->b_blockb = NULL;
        }
        }
+       vp = bp->b_vp;
+       bp->b_vp = (struct vnode *) 0;
+       HOLDRELE(vp);
+}
+
+/*
+ * Reassign a buffer from one vnode to another.
+ * Used to assign file specific control information
+ * (indirect blocks) to the vnode to which they belong.
+ */
+reassignbuf(bp, newvp)
+       register struct buf *bp;
+       register struct vnode *newvp;
+{
+       register struct buf *bq, **listheadp;
+
+       if (newvp == NULL) {
+               printf("reassignbuf: NULL");
+               return;
+       }
+       /*
+        * Delete from old vnode list, if on one.
+        */
+       if (bp->b_blockb) {
+               if (bq = bp->b_blockf)
+                       bq->b_blockb = bp->b_blockb;
+               *bp->b_blockb = bq;
+       }
+       /*
+        * If dirty, put on list of dirty buffers;
+        * otherwise insert onto list of clean buffers.
+        */
+       if (bp->b_flags & B_DELWRI)
+               listheadp = &newvp->v_dirtyblkhd;
+       else
+               listheadp = &newvp->v_cleanblkhd;
+       if (bq = *listheadp)
+               bq->b_blockb = &bp->b_blockf;
+       bp->b_blockf = bq;
+       bp->b_blockb = listheadp;
+       *listheadp = bp;
 }
 
 /*
 }
 
 /*
@@ -309,7 +557,9 @@ bdevvp(dev, vpp)
        struct vnode *nvp;
        int error;
 
        struct vnode *nvp;
        int error;
 
-       error = getnewvnode(VT_NON, (struct mount *)0, &spec_vnodeops, &nvp);
+       if (dev == NODEV)
+               return (0);
+       error = getnewvnode(VT_NON, (struct mount *)0, spec_vnodeop_p, &nvp);
        if (error) {
                *vpp = 0;
                return (error);
        if (error) {
                *vpp = 0;
                return (error);
@@ -338,11 +588,12 @@ checkalias(nvp, nvp_rdev, mp)
        dev_t nvp_rdev;
        struct mount *mp;
 {
        dev_t nvp_rdev;
        struct mount *mp;
 {
+       USES_VOP_UNLOCK;
        register struct vnode *vp;
        struct vnode **vpp;
 
        if (nvp->v_type != VBLK && nvp->v_type != VCHR)
        register struct vnode *vp;
        struct vnode **vpp;
 
        if (nvp->v_type != VBLK && nvp->v_type != VCHR)
-               return ((struct vnode *)0);
+               return (NULLVP);
 
        vpp = &speclisth[SPECHASH(nvp_rdev)];
 loop:
 
        vpp = &speclisth[SPECHASH(nvp_rdev)];
 loop:
@@ -361,18 +612,19 @@ loop:
                break;
        }
        if (vp == NULL || vp->v_tag != VT_NON) {
                break;
        }
        if (vp == NULL || vp->v_tag != VT_NON) {
-               if (vp != NULL) {
-                       nvp->v_flag |= VALIASED;
-                       vp->v_flag |= VALIASED;
-                       vput(vp);
-               }
                MALLOC(nvp->v_specinfo, struct specinfo *,
                        sizeof(struct specinfo), M_VNODE, M_WAITOK);
                nvp->v_rdev = nvp_rdev;
                nvp->v_hashchain = vpp;
                nvp->v_specnext = *vpp;
                MALLOC(nvp->v_specinfo, struct specinfo *,
                        sizeof(struct specinfo), M_VNODE, M_WAITOK);
                nvp->v_rdev = nvp_rdev;
                nvp->v_hashchain = vpp;
                nvp->v_specnext = *vpp;
+               nvp->v_specflags = 0;
                *vpp = nvp;
                *vpp = nvp;
-               return ((struct vnode *)0);
+               if (vp != NULL) {
+                       nvp->v_flag |= VALIASED;
+                       vp->v_flag |= VALIASED;
+                       vput(vp);
+               }
+               return (NULLVP);
        }
        VOP_UNLOCK(vp);
        vclean(vp, 0);
        }
        VOP_UNLOCK(vp);
        vclean(vp, 0);
@@ -394,6 +646,7 @@ loop:
 vget(vp)
        register struct vnode *vp;
 {
 vget(vp)
        register struct vnode *vp;
 {
+       USES_VOP_LOCK;
        register struct vnode *vq;
 
        if (vp->v_flag & VXLOCK) {
        register struct vnode *vq;
 
        if (vp->v_flag & VXLOCK) {
@@ -415,6 +668,8 @@ vget(vp)
        return (0);
 }
 
        return (0);
 }
 
+int bug_refs = 0;
+
 /*
  * Vnode reference, just increment the count
  */
 /*
  * Vnode reference, just increment the count
  */
@@ -423,6 +678,10 @@ void vref(vp)
 {
 
        vp->v_usecount++;
 {
 
        vp->v_usecount++;
+       if (vp->v_type != VBLK && curproc)
+               curproc->p_spare[0]++;
+       if (bug_refs)
+               vprint("vref: ");
 }
 
 /*
 }
 
 /*
@@ -431,6 +690,8 @@ void vref(vp)
 void vput(vp)
        register struct vnode *vp;
 {
 void vput(vp)
        register struct vnode *vp;
 {
+       USES_VOP_UNLOCK;
+
        VOP_UNLOCK(vp);
        vrele(vp);
 }
        VOP_UNLOCK(vp);
        vrele(vp);
 }
@@ -442,15 +703,27 @@ void vput(vp)
 void vrele(vp)
        register struct vnode *vp;
 {
 void vrele(vp)
        register struct vnode *vp;
 {
+       USES_VOP_INACTIVE;
+       struct proc *p = curproc;               /* XXX */
 
 
+#ifdef DIAGNOSTIC
        if (vp == NULL)
                panic("vrele: null vp");
        if (vp == NULL)
                panic("vrele: null vp");
+#endif
        vp->v_usecount--;
        vp->v_usecount--;
-       if (vp->v_usecount < 0)
-               vprint("vrele: bad ref count", vp);
+       if (vp->v_type != VBLK && curproc)
+               curproc->p_spare[0]--;
+       if (bug_refs)
+               vprint("vref: ");
        if (vp->v_usecount > 0)
                return;
        if (vp->v_usecount > 0)
                return;
-       if (vfreeh == (struct vnode *)0) {
+#ifdef DIAGNOSTIC
+       if (vp->v_usecount != 0 || vp->v_writecount != 0) {
+               vprint("vrele: bad ref count", vp);
+               panic("vrele: ref cnt");
+       }
+#endif
+       if (vfreeh == NULLVP) {
                /*
                 * insert into empty list
                 */
                /*
                 * insert into empty list
                 */
@@ -465,13 +738,13 @@ void vrele(vp)
        }
        vp->v_freef = NULL;
        vfreet = &vp->v_freef;
        }
        vp->v_freef = NULL;
        vfreet = &vp->v_freef;
-       VOP_INACTIVE(vp);
+       VOP_INACTIVE(vp, p);
 }
 
 /*
  * Page or buffer structure gets a reference.
  */
 }
 
 /*
  * Page or buffer structure gets a reference.
  */
-vhold(vp)
+void vhold(vp)
        register struct vnode *vp;
 {
 
        register struct vnode *vp;
 {
 
@@ -481,7 +754,7 @@ vhold(vp)
 /*
  * Page or buffer structure frees a reference.
  */
 /*
  * Page or buffer structure frees a reference.
  */
-holdrele(vp)
+void holdrele(vp)
        register struct vnode *vp;
 {
 
        register struct vnode *vp;
 {
 
@@ -508,14 +781,23 @@ vflush(mp, skipvp, flags)
        register struct vnode *vp, *nvp;
        int busy = 0;
 
        register struct vnode *vp, *nvp;
        int busy = 0;
 
-       for (vp = mp->m_mounth; vp; vp = nvp) {
+       if ((mp->mnt_flag & MNT_MPBUSY) == 0)
+               panic("vflush: not busy");
+loop:
+       for (vp = mp->mnt_mounth; vp; vp = nvp) {
+               if (vp->v_mount != mp)
+                       goto loop;
                nvp = vp->v_mountf;
                /*
                 * Skip over a selected vnode.
                nvp = vp->v_mountf;
                /*
                 * Skip over a selected vnode.
-                * Used by ufs to skip over the quota structure inode.
                 */
                if (vp == skipvp)
                        continue;
                 */
                if (vp == skipvp)
                        continue;
+               /*
+                * Skip over a vnodes marked VSYSTEM.
+                */
+               if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM))
+                       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.
@@ -528,12 +810,12 @@ vflush(mp, skipvp, flags)
                 * 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.
                 */
-               if (flags & MNT_FORCE) {
+               if (flags & FORCECLOSE) {
                        if (vp->v_type != VBLK && vp->v_type != VCHR) {
                                vgone(vp);
                        } else {
                                vclean(vp, 0);
                        if (vp->v_type != VBLK && vp->v_type != VCHR) {
                                vgone(vp);
                        } else {
                                vclean(vp, 0);
-                               vp->v_op = &spec_vnodeops;
+                               vp->v_op = spec_vnodeop_p;
                                insmntque(vp, (struct mount *)0);
                        }
                        continue;
                                insmntque(vp, (struct mount *)0);
                        }
                        continue;
@@ -550,12 +832,18 @@ vflush(mp, skipvp, flags)
 /*
  * Disassociate the underlying file system from a vnode.
  */
 /*
  * Disassociate the underlying file system from a vnode.
  */
-void vclean(vp, doclose)
+void vclean(vp, flags)
        register struct vnode *vp;
        register struct vnode *vp;
-       long doclose;
+       int flags;
 {
 {
-       struct vnodeops *origops;
+       USES_VOP_LOCK;
+       USES_VOP_UNLOCK;
+       USES_VOP_CLOSE;
+       USES_VOP_INACTIVE;
+       USES_VOP_RECLAIM;
+       int (**origops)();
        int active;
        int active;
+       struct proc *p = curproc;       /* XXX */
 
        /*
         * Check to see if the vnode is in use.
 
        /*
         * Check to see if the vnode is in use.
@@ -580,32 +868,54 @@ void vclean(vp, doclose)
         * occur while the buffer list is being cleaned out.
         */
        VOP_LOCK(vp);
         * occur while the buffer list is being cleaned out.
         */
        VOP_LOCK(vp);
-       if (doclose)
+       if (flags & DOCLOSE)
                vinvalbuf(vp, 1);
        /*
         * Prevent any further operations on the vnode from
         * being passed through to the old file system.
         */
        origops = vp->v_op;
                vinvalbuf(vp, 1);
        /*
         * Prevent any further operations on the vnode from
         * being passed through to the old file system.
         */
        origops = vp->v_op;
-       vp->v_op = &dead_vnodeops;
+       vp->v_op = dead_vnodeop_p;
        vp->v_tag = VT_NON;
        /*
         * If purging an active vnode, it must be unlocked, closed,
         * and deactivated before being reclaimed.
         */
        vp->v_tag = VT_NON;
        /*
         * If purging an active vnode, it must be unlocked, closed,
         * and deactivated before being reclaimed.
         */
-       (*(origops->vn_unlock))(vp);
+       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) {
-               if (doclose)
-                       (*(origops->vn_close))(vp, 0, NOCRED);
-               (*(origops->vn_inactive))(vp);
+               /*
+                * 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 = p;
+                       VOCALL(origops,VOFFSET(vop_close),&vop_close_a);
+               };
+               vop_inactive_a.a_desc = VDESC(vop_inactive);
+               vop_inactive_a.a_vp = vp;
+               vop_inactive_a.a_p = p;
+               VOCALL(origops,VOFFSET(vop_inactive),&vop_inactive_a);
        }
        /*
         * Reclaim the vnode.
         */
        }
        /*
         * Reclaim the vnode.
         */
-       if ((*(origops->vn_reclaim))(vp))
+       /*
+        * 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))
                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 in vget of the grim news.
         */
@@ -625,13 +935,36 @@ void vgoneall(vp)
 {
        register struct vnode *vq;
 
 {
        register struct vnode *vq;
 
-       while (vp->v_flag & VALIASED) {
-               for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
-                       if (vq->v_rdev != vp->v_rdev || vp == vq)
-                               continue;
-                       vgone(vq);
-                       break;
+       if (vp->v_flag & VALIASED) {
+               /*
+                * If a vgone (or vclean) is already in progress,
+                * wait until it is done and return.
+                */
+               if (vp->v_flag & VXLOCK) {
+                       vp->v_flag |= VXWANT;
+                       sleep((caddr_t)vp, PINOD);
+                       return;
+               }
+               /*
+                * Ensure that vp will not be vgone'd while we
+                * are eliminating its aliases.
+                */
+               vp->v_flag |= VXLOCK;
+               while (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 || vp == vq)
+                                       continue;
+                               vgone(vq);
+                               break;
+                       }
                }
                }
+               /*
+                * Remove the lock so that vgone below will
+                * really eliminate the vnode after which time
+                * vgone will awaken any sleepers.
+                */
+               vp->v_flag &= ~VXLOCK;
        }
        vgone(vp);
 }
        }
        vgone(vp);
 }
@@ -645,12 +978,20 @@ void vgone(vp)
 {
        register struct vnode *vq;
        struct vnode *vx;
 {
        register struct vnode *vq;
        struct vnode *vx;
-       long count;
 
 
+       /*
+        * If a vgone (or vclean) is already in progress,
+        * wait until it is done and return.
+        */
+       if (vp->v_flag & VXLOCK) {
+               vp->v_flag |= VXWANT;
+               sleep((caddr_t)vp, PINOD);
+               return;
+       }
        /*
         * Clean out the filesystem specific data.
         */
        /*
         * Clean out the filesystem specific data.
         */
-       vclean(vp, 1);
+       vclean(vp, DOCLOSE);
        /*
         * Delete from old mount point vnode list, if on one.
         */
        /*
         * Delete from old mount point vnode list, if on one.
         */
@@ -660,6 +1001,7 @@ void vgone(vp)
                *vp->v_mountb = vq;
                vp->v_mountf = NULL;
                vp->v_mountb = NULL;
                *vp->v_mountb = vq;
                vp->v_mountf = NULL;
                vp->v_mountb = NULL;
+               vp->v_mount = NULL;
        }
        /*
         * If special device, remove it from special device alias list.
        }
        /*
         * If special device, remove it from special device alias list.
@@ -678,16 +1020,18 @@ void vgone(vp)
                                panic("missing bdev");
                }
                if (vp->v_flag & VALIASED) {
                                panic("missing bdev");
                }
                if (vp->v_flag & VALIASED) {
-                       count = 0;
+                       vx = NULL;
                        for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
                        for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
-                               if (vq->v_rdev != vp->v_rdev)
+                               if (vq->v_rdev != vp->v_rdev ||
+                                   vq->v_type != vp->v_type)
                                        continue;
                                        continue;
-                               count++;
+                               if (vx)
+                                       break;
                                vx = vq;
                        }
                                vx = vq;
                        }
-                       if (count == 0)
+                       if (vx == NULL)
                                panic("missing alias");
                                panic("missing alias");
-                       if (count == 1)
+                       if (vq == NULL)
                                vx->v_flag &= ~VALIASED;
                        vp->v_flag &= ~VALIASED;
                }
                                vx->v_flag &= ~VALIASED;
                        vp->v_flag &= ~VALIASED;
                }
@@ -743,7 +1087,7 @@ vcount(vp)
                return (vp->v_usecount);
 loop:
        for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
                return (vp->v_usecount);
 loop:
        for (count = 0, vq = *vp->v_hashchain; vq; vq = vq->v_specnext) {
-               if (vq->v_rdev != vp->v_rdev)
+               if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type)
                        continue;
                /*
                 * Alias, but not in use, so flush it out.
                        continue;
                /*
                 * Alias, but not in use, so flush it out.
@@ -761,49 +1105,126 @@ loop:
  * Print out a description of a vnode.
  */
 static char *typename[] =
  * Print out a description of a vnode.
  */
 static char *typename[] =
-       { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VBAD" };
+   { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" };
 
 vprint(label, vp)
        char *label;
        register struct vnode *vp;
 {
 
 vprint(label, vp)
        char *label;
        register struct vnode *vp;
 {
+       USES_VOP_PRINT;
        char buf[64];
 
        if (label != NULL)
                printf("%s: ", label);
        char buf[64];
 
        if (label != NULL)
                printf("%s: ", label);
-       printf("type %s, usecount %d, refcount %d,", typename[vp->v_type],
-               vp->v_usecount, 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);
        buf[0] = '\0';
        if (vp->v_flag & VROOT)
                strcat(buf, "|VROOT");
        if (vp->v_flag & VTEXT)
                strcat(buf, "|VTEXT");
        buf[0] = '\0';
        if (vp->v_flag & VROOT)
                strcat(buf, "|VROOT");
        if (vp->v_flag & VTEXT)
                strcat(buf, "|VTEXT");
+       if (vp->v_flag & VSYSTEM)
+               strcat(buf, "|VSYSTEM");
        if (vp->v_flag & VXLOCK)
                strcat(buf, "|VXLOCK");
        if (vp->v_flag & VXWANT)
                strcat(buf, "|VXWANT");
        if (vp->v_flag & VXLOCK)
                strcat(buf, "|VXLOCK");
        if (vp->v_flag & VXWANT)
                strcat(buf, "|VXWANT");
-       if (vp->v_flag & VEXLOCK)
-               strcat(buf, "|VEXLOCK");
-       if (vp->v_flag & VSHLOCK)
-               strcat(buf, "|VSHLOCK");
-       if (vp->v_flag & VLWAIT)
-               strcat(buf, "|VLWAIT");
-       if (vp->v_flag & VALIASED)
-               strcat(buf, "|VALIASED");
        if (vp->v_flag & VBWAIT)
                strcat(buf, "|VBWAIT");
        if (vp->v_flag & VBWAIT)
                strcat(buf, "|VBWAIT");
+       if (vp->v_flag & VALIASED)
+               strcat(buf, "|VALIASED");
        if (buf[0] != '\0')
                printf(" flags (%s)", &buf[1]);
        printf("\n\t");
        VOP_PRINT(vp);
 }
 
        if (buf[0] != '\0')
                printf(" flags (%s)", &buf[1]);
        printf("\n\t");
        VOP_PRINT(vp);
 }
 
-strcat(src, append)
-       register char *src, *append;
+#ifdef DEBUG
+/*
+ * List all of the locked vnodes in the system.
+ * Called when debugging the kernel.
+ */
+printlockedvnodes()
 {
 {
+       USES_VOP_ISLOCKED;
+       register struct mount *mp;
+       register struct vnode *vp;
+
+       printf("Locked vnodes\n");
+       mp = rootfs;
+       do {
+               for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf)
+                       if (VOP_ISLOCKED(vp))
+                               vprint((char *)0, vp);
+               mp = mp->mnt_next;
+       } while (mp != rootfs);
+}
+#endif
 
 
-       for (; *src; ++src)
-               /* void */;
-       while (*src++ = *append++)
-               /* void */;
+int kinfo_vdebug = 1;
+int kinfo_vgetfailed;
+#define KINFO_VNODESLOP        10
+/*
+ * Dump vnode list (via kinfo).
+ * Copyout address of vnode followed by vnode.
+ */
+/* ARGSUSED */
+kinfo_vnode(op, where, acopysize, arg, aneeded)
+       int op;
+       char *where;
+       int *acopysize, arg, *aneeded;
+{
+       register struct mount *mp = rootfs;
+       struct mount *omp;
+       struct vnode *vp;
+       register char *bp = where, *savebp;
+       char *ewhere;
+       int error;
+
+#define VPTRSZ sizeof (struct vnode *)
+#define VNODESZ        sizeof (struct vnode)
+       if (where == NULL) {
+               *aneeded = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ);
+               return (0);
+       }
+       ewhere = where + *acopysize;
+               
+       do {
+               if (vfs_busy(mp)) {
+                       mp = mp->mnt_next;
+                       continue;
+               }
+               savebp = bp;
+again:
+               for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) {
+                       /*
+                        * Check that the vp is still associated with
+                        * this filesystem.  RACE: could have been
+                        * recycled onto the same filesystem.
+                        */
+                       if (vp->v_mount != mp) {
+                               if (kinfo_vdebug)
+                                       printf("kinfo: vp changed\n");
+                               bp = savebp;
+                               goto again;
+                       }
+                       if ((bp + VPTRSZ + VNODESZ <= ewhere) && 
+                           ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) ||
+                            (error = copyout((caddr_t)vp, 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;
+       return (0);
 }
 }